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.
35 * Copyright (c) 2021, Colm Buckley <colm@tuatha.org>
39 * SPA: Storage Pool Allocator
41 * This file contains all the routines used when modifying on-disk SPA state.
42 * This includes opening, importing, destroying, exporting a pool, and syncing a
46 #include <sys/zfs_context.h>
47 #include <sys/fm/fs/zfs.h>
48 #include <sys/spa_impl.h>
50 #include <sys/zio_checksum.h>
52 #include <sys/dmu_tx.h>
56 #include <sys/vdev_impl.h>
57 #include <sys/vdev_removal.h>
58 #include <sys/vdev_indirect_mapping.h>
59 #include <sys/vdev_indirect_births.h>
60 #include <sys/vdev_initialize.h>
61 #include <sys/vdev_rebuild.h>
62 #include <sys/vdev_trim.h>
63 #include <sys/vdev_disk.h>
64 #include <sys/vdev_draid.h>
65 #include <sys/metaslab.h>
66 #include <sys/metaslab_impl.h>
68 #include <sys/uberblock_impl.h>
71 #include <sys/bpobj.h>
72 #include <sys/dmu_traverse.h>
73 #include <sys/dmu_objset.h>
74 #include <sys/unique.h>
75 #include <sys/dsl_pool.h>
76 #include <sys/dsl_dataset.h>
77 #include <sys/dsl_dir.h>
78 #include <sys/dsl_prop.h>
79 #include <sys/dsl_synctask.h>
80 #include <sys/fs/zfs.h>
82 #include <sys/callb.h>
83 #include <sys/systeminfo.h>
84 #include <sys/spa_boot.h>
85 #include <sys/zfs_ioctl.h>
86 #include <sys/dsl_scan.h>
87 #include <sys/zfeature.h>
88 #include <sys/dsl_destroy.h>
92 #include <sys/fm/protocol.h>
93 #include <sys/fm/util.h>
94 #include <sys/callb.h>
96 #include <sys/vmsystm.h>
100 #include "zfs_comutil.h"
103 * The interval, in seconds, at which failed configuration cache file writes
106 int zfs_ccw_retry_interval = 300;
108 typedef enum zti_modes {
109 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
110 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
111 ZTI_MODE_SCALE, /* Taskqs scale with CPUs. */
112 ZTI_MODE_NULL, /* don't create a taskq */
116 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
117 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
118 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
119 #define ZTI_SCALE { ZTI_MODE_SCALE, 0, 1 }
120 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
122 #define ZTI_N(n) ZTI_P(n, 1)
123 #define ZTI_ONE ZTI_N(1)
125 typedef struct zio_taskq_info {
126 zti_modes_t zti_mode;
131 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
132 "iss", "iss_h", "int", "int_h"
136 * This table defines the taskq settings for each ZFS I/O type. When
137 * initializing a pool, we use this table to create an appropriately sized
138 * taskq. Some operations are low volume and therefore have a small, static
139 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
140 * macros. Other operations process a large amount of data; the ZTI_BATCH
141 * macro causes us to create a taskq oriented for throughput. Some operations
142 * are so high frequency and short-lived that the taskq itself can become a
143 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
144 * additional degree of parallelism specified by the number of threads per-
145 * taskq and the number of taskqs; when dispatching an event in this case, the
146 * particular taskq is chosen at random. ZTI_SCALE is similar to ZTI_BATCH,
147 * but with number of taskqs also scaling with number of CPUs.
149 * The different taskq priorities are to handle the different contexts (issue
150 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
151 * need to be handled with minimum delay.
153 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
154 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
155 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
156 { ZTI_N(8), ZTI_NULL, ZTI_SCALE, ZTI_NULL }, /* READ */
157 { ZTI_BATCH, ZTI_N(5), ZTI_SCALE, ZTI_N(5) }, /* WRITE */
158 { ZTI_SCALE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
159 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
160 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
161 { ZTI_N(4), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* TRIM */
164 static void spa_sync_version(void *arg, dmu_tx_t *tx);
165 static void spa_sync_props(void *arg, dmu_tx_t *tx);
166 static boolean_t spa_has_active_shared_spare(spa_t *spa);
167 static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
168 static void spa_vdev_resilver_done(spa_t *spa);
170 uint_t zio_taskq_batch_pct = 80; /* 1 thread per cpu in pset */
171 uint_t zio_taskq_batch_tpq; /* threads per taskq */
172 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
173 uint_t zio_taskq_basedc = 80; /* base duty cycle */
175 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
178 * Report any spa_load_verify errors found, but do not fail spa_load.
179 * This is used by zdb to analyze non-idle pools.
181 boolean_t spa_load_verify_dryrun = B_FALSE;
184 * Allow read spacemaps in case of readonly import (spa_mode == SPA_MODE_READ).
185 * This is used by zdb for spacemaps verification.
187 boolean_t spa_mode_readable_spacemaps = B_FALSE;
190 * This (illegal) pool name is used when temporarily importing a spa_t in order
191 * to get the vdev stats associated with the imported devices.
193 #define TRYIMPORT_NAME "$import"
196 * For debugging purposes: print out vdev tree during pool import.
198 int spa_load_print_vdev_tree = B_FALSE;
201 * A non-zero value for zfs_max_missing_tvds means that we allow importing
202 * pools with missing top-level vdevs. This is strictly intended for advanced
203 * pool recovery cases since missing data is almost inevitable. Pools with
204 * missing devices can only be imported read-only for safety reasons, and their
205 * fail-mode will be automatically set to "continue".
207 * With 1 missing vdev we should be able to import the pool and mount all
208 * datasets. User data that was not modified after the missing device has been
209 * added should be recoverable. This means that snapshots created prior to the
210 * addition of that device should be completely intact.
212 * With 2 missing vdevs, some datasets may fail to mount since there are
213 * dataset statistics that are stored as regular metadata. Some data might be
214 * recoverable if those vdevs were added recently.
216 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
217 * may be missing entirely. Chances of data recovery are very low. Note that
218 * there are also risks of performing an inadvertent rewind as we might be
219 * missing all the vdevs with the latest uberblocks.
221 unsigned long zfs_max_missing_tvds = 0;
224 * The parameters below are similar to zfs_max_missing_tvds but are only
225 * intended for a preliminary open of the pool with an untrusted config which
226 * might be incomplete or out-dated.
228 * We are more tolerant for pools opened from a cachefile since we could have
229 * an out-dated cachefile where a device removal was not registered.
230 * We could have set the limit arbitrarily high but in the case where devices
231 * are really missing we would want to return the proper error codes; we chose
232 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
233 * and we get a chance to retrieve the trusted config.
235 uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
238 * In the case where config was assembled by scanning device paths (/dev/dsks
239 * by default) we are less tolerant since all the existing devices should have
240 * been detected and we want spa_load to return the right error codes.
242 uint64_t zfs_max_missing_tvds_scan = 0;
245 * Debugging aid that pauses spa_sync() towards the end.
247 boolean_t zfs_pause_spa_sync = B_FALSE;
250 * Variables to indicate the livelist condense zthr func should wait at certain
251 * points for the livelist to be removed - used to test condense/destroy races
253 int zfs_livelist_condense_zthr_pause = 0;
254 int zfs_livelist_condense_sync_pause = 0;
257 * Variables to track whether or not condense cancellation has been
258 * triggered in testing.
260 int zfs_livelist_condense_sync_cancel = 0;
261 int zfs_livelist_condense_zthr_cancel = 0;
264 * Variable to track whether or not extra ALLOC blkptrs were added to a
265 * livelist entry while it was being condensed (caused by the way we track
266 * remapped blkptrs in dbuf_remap_impl)
268 int zfs_livelist_condense_new_alloc = 0;
271 * ==========================================================================
272 * SPA properties routines
273 * ==========================================================================
277 * Add a (source=src, propname=propval) list to an nvlist.
280 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
281 uint64_t intval, zprop_source_t src)
283 const char *propname = zpool_prop_to_name(prop);
286 propval = fnvlist_alloc();
287 fnvlist_add_uint64(propval, ZPROP_SOURCE, src);
290 fnvlist_add_string(propval, ZPROP_VALUE, strval);
292 fnvlist_add_uint64(propval, ZPROP_VALUE, intval);
294 fnvlist_add_nvlist(nvl, propname, propval);
295 nvlist_free(propval);
299 * Get property values from the spa configuration.
302 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
304 vdev_t *rvd = spa->spa_root_vdev;
305 dsl_pool_t *pool = spa->spa_dsl_pool;
306 uint64_t size, alloc, cap, version;
307 const zprop_source_t src = ZPROP_SRC_NONE;
308 spa_config_dirent_t *dp;
309 metaslab_class_t *mc = spa_normal_class(spa);
311 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
314 alloc = metaslab_class_get_alloc(mc);
315 alloc += metaslab_class_get_alloc(spa_special_class(spa));
316 alloc += metaslab_class_get_alloc(spa_dedup_class(spa));
317 alloc += metaslab_class_get_alloc(spa_embedded_log_class(spa));
319 size = metaslab_class_get_space(mc);
320 size += metaslab_class_get_space(spa_special_class(spa));
321 size += metaslab_class_get_space(spa_dedup_class(spa));
322 size += metaslab_class_get_space(spa_embedded_log_class(spa));
324 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
325 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
326 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
327 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
329 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
330 spa->spa_checkpoint_info.sci_dspace, src);
332 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
333 metaslab_class_fragmentation(mc), src);
334 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
335 metaslab_class_expandable_space(mc), src);
336 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
337 (spa_mode(spa) == SPA_MODE_READ), src);
339 cap = (size == 0) ? 0 : (alloc * 100 / size);
340 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
342 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
343 ddt_get_pool_dedup_ratio(spa), src);
345 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
346 rvd->vdev_state, src);
348 version = spa_version(spa);
349 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) {
350 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL,
351 version, ZPROP_SRC_DEFAULT);
353 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL,
354 version, ZPROP_SRC_LOCAL);
356 spa_prop_add_list(*nvp, ZPOOL_PROP_LOAD_GUID,
357 NULL, spa_load_guid(spa), src);
362 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
363 * when opening pools before this version freedir will be NULL.
365 if (pool->dp_free_dir != NULL) {
366 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
367 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
370 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
374 if (pool->dp_leak_dir != NULL) {
375 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
376 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
379 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
384 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
386 if (spa->spa_comment != NULL) {
387 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
391 if (spa->spa_compatibility != NULL) {
392 spa_prop_add_list(*nvp, ZPOOL_PROP_COMPATIBILITY,
393 spa->spa_compatibility, 0, ZPROP_SRC_LOCAL);
396 if (spa->spa_root != NULL)
397 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
400 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
401 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
402 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
404 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
405 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
408 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
409 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
410 DNODE_MAX_SIZE, ZPROP_SRC_NONE);
412 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
413 DNODE_MIN_SIZE, ZPROP_SRC_NONE);
416 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
417 if (dp->scd_path == NULL) {
418 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
419 "none", 0, ZPROP_SRC_LOCAL);
420 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
421 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
422 dp->scd_path, 0, ZPROP_SRC_LOCAL);
428 * Get zpool property values.
431 spa_prop_get(spa_t *spa, nvlist_t **nvp)
433 objset_t *mos = spa->spa_meta_objset;
439 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP);
443 dp = spa_get_dsl(spa);
444 dsl_pool_config_enter(dp, FTAG);
445 mutex_enter(&spa->spa_props_lock);
448 * Get properties from the spa config.
450 spa_prop_get_config(spa, nvp);
452 /* If no pool property object, no more prop to get. */
453 if (mos == NULL || spa->spa_pool_props_object == 0)
457 * Get properties from the MOS pool property object.
459 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
460 (err = zap_cursor_retrieve(&zc, &za)) == 0;
461 zap_cursor_advance(&zc)) {
464 zprop_source_t src = ZPROP_SRC_DEFAULT;
467 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
470 switch (za.za_integer_length) {
472 /* integer property */
473 if (za.za_first_integer !=
474 zpool_prop_default_numeric(prop))
475 src = ZPROP_SRC_LOCAL;
477 if (prop == ZPOOL_PROP_BOOTFS) {
478 dsl_dataset_t *ds = NULL;
480 err = dsl_dataset_hold_obj(dp,
481 za.za_first_integer, FTAG, &ds);
485 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
487 dsl_dataset_name(ds, strval);
488 dsl_dataset_rele(ds, FTAG);
491 intval = za.za_first_integer;
494 spa_prop_add_list(*nvp, prop, strval, intval, src);
497 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
502 /* string property */
503 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
504 err = zap_lookup(mos, spa->spa_pool_props_object,
505 za.za_name, 1, za.za_num_integers, strval);
507 kmem_free(strval, za.za_num_integers);
510 spa_prop_add_list(*nvp, prop, strval, 0, src);
511 kmem_free(strval, za.za_num_integers);
518 zap_cursor_fini(&zc);
520 mutex_exit(&spa->spa_props_lock);
521 dsl_pool_config_exit(dp, FTAG);
522 if (err && err != ENOENT) {
532 * Validate the given pool properties nvlist and modify the list
533 * for the property values to be set.
536 spa_prop_validate(spa_t *spa, nvlist_t *props)
539 int error = 0, reset_bootfs = 0;
541 boolean_t has_feature = B_FALSE;
544 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
546 char *strval, *slash, *check, *fname;
547 const char *propname = nvpair_name(elem);
548 zpool_prop_t prop = zpool_name_to_prop(propname);
551 case ZPOOL_PROP_INVAL:
552 if (!zpool_prop_feature(propname)) {
553 error = SET_ERROR(EINVAL);
558 * Sanitize the input.
560 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
561 error = SET_ERROR(EINVAL);
565 if (nvpair_value_uint64(elem, &intval) != 0) {
566 error = SET_ERROR(EINVAL);
571 error = SET_ERROR(EINVAL);
575 fname = strchr(propname, '@') + 1;
576 if (zfeature_lookup_name(fname, NULL) != 0) {
577 error = SET_ERROR(EINVAL);
581 has_feature = B_TRUE;
584 case ZPOOL_PROP_VERSION:
585 error = nvpair_value_uint64(elem, &intval);
587 (intval < spa_version(spa) ||
588 intval > SPA_VERSION_BEFORE_FEATURES ||
590 error = SET_ERROR(EINVAL);
593 case ZPOOL_PROP_DELEGATION:
594 case ZPOOL_PROP_AUTOREPLACE:
595 case ZPOOL_PROP_LISTSNAPS:
596 case ZPOOL_PROP_AUTOEXPAND:
597 case ZPOOL_PROP_AUTOTRIM:
598 error = nvpair_value_uint64(elem, &intval);
599 if (!error && intval > 1)
600 error = SET_ERROR(EINVAL);
603 case ZPOOL_PROP_MULTIHOST:
604 error = nvpair_value_uint64(elem, &intval);
605 if (!error && intval > 1)
606 error = SET_ERROR(EINVAL);
609 uint32_t hostid = zone_get_hostid(NULL);
611 spa->spa_hostid = hostid;
613 error = SET_ERROR(ENOTSUP);
618 case ZPOOL_PROP_BOOTFS:
620 * If the pool version is less than SPA_VERSION_BOOTFS,
621 * or the pool is still being created (version == 0),
622 * the bootfs property cannot be set.
624 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
625 error = SET_ERROR(ENOTSUP);
630 * Make sure the vdev config is bootable
632 if (!vdev_is_bootable(spa->spa_root_vdev)) {
633 error = SET_ERROR(ENOTSUP);
639 error = nvpair_value_string(elem, &strval);
644 if (strval == NULL || strval[0] == '\0') {
645 objnum = zpool_prop_default_numeric(
650 error = dmu_objset_hold(strval, FTAG, &os);
655 if (dmu_objset_type(os) != DMU_OST_ZFS) {
656 error = SET_ERROR(ENOTSUP);
658 objnum = dmu_objset_id(os);
660 dmu_objset_rele(os, FTAG);
664 case ZPOOL_PROP_FAILUREMODE:
665 error = nvpair_value_uint64(elem, &intval);
666 if (!error && intval > ZIO_FAILURE_MODE_PANIC)
667 error = SET_ERROR(EINVAL);
670 * This is a special case which only occurs when
671 * the pool has completely failed. This allows
672 * the user to change the in-core failmode property
673 * without syncing it out to disk (I/Os might
674 * currently be blocked). We do this by returning
675 * EIO to the caller (spa_prop_set) to trick it
676 * into thinking we encountered a property validation
679 if (!error && spa_suspended(spa)) {
680 spa->spa_failmode = intval;
681 error = SET_ERROR(EIO);
685 case ZPOOL_PROP_CACHEFILE:
686 if ((error = nvpair_value_string(elem, &strval)) != 0)
689 if (strval[0] == '\0')
692 if (strcmp(strval, "none") == 0)
695 if (strval[0] != '/') {
696 error = SET_ERROR(EINVAL);
700 slash = strrchr(strval, '/');
701 ASSERT(slash != NULL);
703 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
704 strcmp(slash, "/..") == 0)
705 error = SET_ERROR(EINVAL);
708 case ZPOOL_PROP_COMMENT:
709 if ((error = nvpair_value_string(elem, &strval)) != 0)
711 for (check = strval; *check != '\0'; check++) {
712 if (!isprint(*check)) {
713 error = SET_ERROR(EINVAL);
717 if (strlen(strval) > ZPROP_MAX_COMMENT)
718 error = SET_ERROR(E2BIG);
729 (void) nvlist_remove_all(props,
730 zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO));
732 if (!error && reset_bootfs) {
733 error = nvlist_remove(props,
734 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
737 error = nvlist_add_uint64(props,
738 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
746 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
749 spa_config_dirent_t *dp;
751 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
755 dp = kmem_alloc(sizeof (spa_config_dirent_t),
758 if (cachefile[0] == '\0')
759 dp->scd_path = spa_strdup(spa_config_path);
760 else if (strcmp(cachefile, "none") == 0)
763 dp->scd_path = spa_strdup(cachefile);
765 list_insert_head(&spa->spa_config_list, dp);
767 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
771 spa_prop_set(spa_t *spa, nvlist_t *nvp)
774 nvpair_t *elem = NULL;
775 boolean_t need_sync = B_FALSE;
777 if ((error = spa_prop_validate(spa, nvp)) != 0)
780 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
781 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
783 if (prop == ZPOOL_PROP_CACHEFILE ||
784 prop == ZPOOL_PROP_ALTROOT ||
785 prop == ZPOOL_PROP_READONLY)
788 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
791 if (prop == ZPOOL_PROP_VERSION) {
792 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
794 ASSERT(zpool_prop_feature(nvpair_name(elem)));
795 ver = SPA_VERSION_FEATURES;
799 /* Save time if the version is already set. */
800 if (ver == spa_version(spa))
804 * In addition to the pool directory object, we might
805 * create the pool properties object, the features for
806 * read object, the features for write object, or the
807 * feature descriptions object.
809 error = dsl_sync_task(spa->spa_name, NULL,
810 spa_sync_version, &ver,
811 6, ZFS_SPACE_CHECK_RESERVED);
822 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
823 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
830 * If the bootfs property value is dsobj, clear it.
833 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
835 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
836 VERIFY(zap_remove(spa->spa_meta_objset,
837 spa->spa_pool_props_object,
838 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
845 spa_change_guid_check(void *arg, dmu_tx_t *tx)
847 uint64_t *newguid __maybe_unused = arg;
848 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
849 vdev_t *rvd = spa->spa_root_vdev;
852 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
853 int error = (spa_has_checkpoint(spa)) ?
854 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
855 return (SET_ERROR(error));
858 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
859 vdev_state = rvd->vdev_state;
860 spa_config_exit(spa, SCL_STATE, FTAG);
862 if (vdev_state != VDEV_STATE_HEALTHY)
863 return (SET_ERROR(ENXIO));
865 ASSERT3U(spa_guid(spa), !=, *newguid);
871 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
873 uint64_t *newguid = arg;
874 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
876 vdev_t *rvd = spa->spa_root_vdev;
878 oldguid = spa_guid(spa);
880 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
881 rvd->vdev_guid = *newguid;
882 rvd->vdev_guid_sum += (*newguid - oldguid);
883 vdev_config_dirty(rvd);
884 spa_config_exit(spa, SCL_STATE, FTAG);
886 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
887 (u_longlong_t)oldguid, (u_longlong_t)*newguid);
891 * Change the GUID for the pool. This is done so that we can later
892 * re-import a pool built from a clone of our own vdevs. We will modify
893 * the root vdev's guid, our own pool guid, and then mark all of our
894 * vdevs dirty. Note that we must make sure that all our vdevs are
895 * online when we do this, or else any vdevs that weren't present
896 * would be orphaned from our pool. We are also going to issue a
897 * sysevent to update any watchers.
900 spa_change_guid(spa_t *spa)
905 mutex_enter(&spa->spa_vdev_top_lock);
906 mutex_enter(&spa_namespace_lock);
907 guid = spa_generate_guid(NULL);
909 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
910 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
913 spa_write_cachefile(spa, B_FALSE, B_TRUE);
914 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
917 mutex_exit(&spa_namespace_lock);
918 mutex_exit(&spa->spa_vdev_top_lock);
924 * ==========================================================================
925 * SPA state manipulation (open/create/destroy/import/export)
926 * ==========================================================================
930 spa_error_entry_compare(const void *a, const void *b)
932 const spa_error_entry_t *sa = (const spa_error_entry_t *)a;
933 const spa_error_entry_t *sb = (const spa_error_entry_t *)b;
936 ret = memcmp(&sa->se_bookmark, &sb->se_bookmark,
937 sizeof (zbookmark_phys_t));
939 return (TREE_ISIGN(ret));
943 * Utility function which retrieves copies of the current logs and
944 * re-initializes them in the process.
947 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
949 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
951 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
952 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
954 avl_create(&spa->spa_errlist_scrub,
955 spa_error_entry_compare, sizeof (spa_error_entry_t),
956 offsetof(spa_error_entry_t, se_avl));
957 avl_create(&spa->spa_errlist_last,
958 spa_error_entry_compare, sizeof (spa_error_entry_t),
959 offsetof(spa_error_entry_t, se_avl));
963 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
965 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
966 enum zti_modes mode = ztip->zti_mode;
967 uint_t value = ztip->zti_value;
968 uint_t count = ztip->zti_count;
969 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
970 uint_t cpus, flags = TASKQ_DYNAMIC;
971 boolean_t batch = B_FALSE;
975 ASSERT3U(value, >, 0);
980 flags |= TASKQ_THREADS_CPU_PCT;
981 value = MIN(zio_taskq_batch_pct, 100);
985 flags |= TASKQ_THREADS_CPU_PCT;
987 * We want more taskqs to reduce lock contention, but we want
988 * less for better request ordering and CPU utilization.
990 cpus = MAX(1, boot_ncpus * zio_taskq_batch_pct / 100);
991 if (zio_taskq_batch_tpq > 0) {
992 count = MAX(1, (cpus + zio_taskq_batch_tpq / 2) /
993 zio_taskq_batch_tpq);
996 * Prefer 6 threads per taskq, but no more taskqs
997 * than threads in them on large systems. For 80%:
1000 * cpus taskqs percent threads threads
1001 * ------- ------- ------- ------- -------
1012 count = 1 + cpus / 6;
1013 while (count * count > cpus)
1016 /* Limit each taskq within 100% to not trigger assertion. */
1017 count = MAX(count, (zio_taskq_batch_pct + 99) / 100);
1018 value = (zio_taskq_batch_pct + count / 2) / count;
1022 tqs->stqs_count = 0;
1023 tqs->stqs_taskq = NULL;
1027 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
1029 zio_type_name[t], zio_taskq_types[q], mode, value);
1033 ASSERT3U(count, >, 0);
1034 tqs->stqs_count = count;
1035 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
1037 for (uint_t i = 0; i < count; i++) {
1042 (void) snprintf(name, sizeof (name), "%s_%s_%u",
1043 zio_type_name[t], zio_taskq_types[q], i);
1045 (void) snprintf(name, sizeof (name), "%s_%s",
1046 zio_type_name[t], zio_taskq_types[q]);
1048 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
1050 flags |= TASKQ_DC_BATCH;
1052 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
1053 spa->spa_proc, zio_taskq_basedc, flags);
1055 pri_t pri = maxclsyspri;
1057 * The write issue taskq can be extremely CPU
1058 * intensive. Run it at slightly less important
1059 * priority than the other taskqs.
1061 * Under Linux and FreeBSD this means incrementing
1062 * the priority value as opposed to platforms like
1063 * illumos where it should be decremented.
1065 * On FreeBSD, if priorities divided by four (RQ_PPQ)
1066 * are equal then a difference between them is
1069 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) {
1070 #if defined(__linux__)
1072 #elif defined(__FreeBSD__)
1078 tq = taskq_create_proc(name, value, pri, 50,
1079 INT_MAX, spa->spa_proc, flags);
1082 tqs->stqs_taskq[i] = tq;
1087 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1089 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1091 if (tqs->stqs_taskq == NULL) {
1092 ASSERT3U(tqs->stqs_count, ==, 0);
1096 for (uint_t i = 0; i < tqs->stqs_count; i++) {
1097 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1098 taskq_destroy(tqs->stqs_taskq[i]);
1101 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1102 tqs->stqs_taskq = NULL;
1106 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1107 * Note that a type may have multiple discrete taskqs to avoid lock contention
1108 * on the taskq itself. In that case we choose which taskq at random by using
1109 * the low bits of gethrtime().
1112 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1113 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1115 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1118 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1119 ASSERT3U(tqs->stqs_count, !=, 0);
1121 if (tqs->stqs_count == 1) {
1122 tq = tqs->stqs_taskq[0];
1124 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
1127 taskq_dispatch_ent(tq, func, arg, flags, ent);
1131 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1134 spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1135 task_func_t *func, void *arg, uint_t flags)
1137 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1141 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1142 ASSERT3U(tqs->stqs_count, !=, 0);
1144 if (tqs->stqs_count == 1) {
1145 tq = tqs->stqs_taskq[0];
1147 tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
1150 id = taskq_dispatch(tq, func, arg, flags);
1152 taskq_wait_id(tq, id);
1156 spa_create_zio_taskqs(spa_t *spa)
1158 for (int t = 0; t < ZIO_TYPES; t++) {
1159 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1160 spa_taskqs_init(spa, t, q);
1166 * Disabled until spa_thread() can be adapted for Linux.
1168 #undef HAVE_SPA_THREAD
1170 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1172 spa_thread(void *arg)
1174 psetid_t zio_taskq_psrset_bind = PS_NONE;
1175 callb_cpr_t cprinfo;
1178 user_t *pu = PTOU(curproc);
1180 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1183 ASSERT(curproc != &p0);
1184 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1185 "zpool-%s", spa->spa_name);
1186 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1188 /* bind this thread to the requested psrset */
1189 if (zio_taskq_psrset_bind != PS_NONE) {
1191 mutex_enter(&cpu_lock);
1192 mutex_enter(&pidlock);
1193 mutex_enter(&curproc->p_lock);
1195 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1196 0, NULL, NULL) == 0) {
1197 curthread->t_bind_pset = zio_taskq_psrset_bind;
1200 "Couldn't bind process for zfs pool \"%s\" to "
1201 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1204 mutex_exit(&curproc->p_lock);
1205 mutex_exit(&pidlock);
1206 mutex_exit(&cpu_lock);
1210 if (zio_taskq_sysdc) {
1211 sysdc_thread_enter(curthread, 100, 0);
1214 spa->spa_proc = curproc;
1215 spa->spa_did = curthread->t_did;
1217 spa_create_zio_taskqs(spa);
1219 mutex_enter(&spa->spa_proc_lock);
1220 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1222 spa->spa_proc_state = SPA_PROC_ACTIVE;
1223 cv_broadcast(&spa->spa_proc_cv);
1225 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1226 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1227 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1228 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1230 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1231 spa->spa_proc_state = SPA_PROC_GONE;
1232 spa->spa_proc = &p0;
1233 cv_broadcast(&spa->spa_proc_cv);
1234 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1236 mutex_enter(&curproc->p_lock);
1242 * Activate an uninitialized pool.
1245 spa_activate(spa_t *spa, spa_mode_t mode)
1247 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1249 spa->spa_state = POOL_STATE_ACTIVE;
1250 spa->spa_mode = mode;
1251 spa->spa_read_spacemaps = spa_mode_readable_spacemaps;
1253 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1254 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1255 spa->spa_embedded_log_class =
1256 metaslab_class_create(spa, zfs_metaslab_ops);
1257 spa->spa_special_class = metaslab_class_create(spa, zfs_metaslab_ops);
1258 spa->spa_dedup_class = metaslab_class_create(spa, zfs_metaslab_ops);
1260 /* Try to create a covering process */
1261 mutex_enter(&spa->spa_proc_lock);
1262 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1263 ASSERT(spa->spa_proc == &p0);
1266 #ifdef HAVE_SPA_THREAD
1267 /* Only create a process if we're going to be around a while. */
1268 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1269 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1271 spa->spa_proc_state = SPA_PROC_CREATED;
1272 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1273 cv_wait(&spa->spa_proc_cv,
1274 &spa->spa_proc_lock);
1276 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1277 ASSERT(spa->spa_proc != &p0);
1278 ASSERT(spa->spa_did != 0);
1282 "Couldn't create process for zfs pool \"%s\"\n",
1287 #endif /* HAVE_SPA_THREAD */
1288 mutex_exit(&spa->spa_proc_lock);
1290 /* If we didn't create a process, we need to create our taskqs. */
1291 if (spa->spa_proc == &p0) {
1292 spa_create_zio_taskqs(spa);
1295 for (size_t i = 0; i < TXG_SIZE; i++) {
1296 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1300 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1301 offsetof(vdev_t, vdev_config_dirty_node));
1302 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1303 offsetof(objset_t, os_evicting_node));
1304 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1305 offsetof(vdev_t, vdev_state_dirty_node));
1307 txg_list_create(&spa->spa_vdev_txg_list, spa,
1308 offsetof(struct vdev, vdev_txg_node));
1310 avl_create(&spa->spa_errlist_scrub,
1311 spa_error_entry_compare, sizeof (spa_error_entry_t),
1312 offsetof(spa_error_entry_t, se_avl));
1313 avl_create(&spa->spa_errlist_last,
1314 spa_error_entry_compare, sizeof (spa_error_entry_t),
1315 offsetof(spa_error_entry_t, se_avl));
1317 spa_keystore_init(&spa->spa_keystore);
1320 * This taskq is used to perform zvol-minor-related tasks
1321 * asynchronously. This has several advantages, including easy
1322 * resolution of various deadlocks.
1324 * The taskq must be single threaded to ensure tasks are always
1325 * processed in the order in which they were dispatched.
1327 * A taskq per pool allows one to keep the pools independent.
1328 * This way if one pool is suspended, it will not impact another.
1330 * The preferred location to dispatch a zvol minor task is a sync
1331 * task. In this context, there is easy access to the spa_t and minimal
1332 * error handling is required because the sync task must succeed.
1334 spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri,
1338 * Taskq dedicated to prefetcher threads: this is used to prevent the
1339 * pool traverse code from monopolizing the global (and limited)
1340 * system_taskq by inappropriately scheduling long running tasks on it.
1342 spa->spa_prefetch_taskq = taskq_create("z_prefetch", 100,
1343 defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT);
1346 * The taskq to upgrade datasets in this pool. Currently used by
1347 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1349 spa->spa_upgrade_taskq = taskq_create("z_upgrade", 100,
1350 defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT);
1354 * Opposite of spa_activate().
1357 spa_deactivate(spa_t *spa)
1359 ASSERT(spa->spa_sync_on == B_FALSE);
1360 ASSERT(spa->spa_dsl_pool == NULL);
1361 ASSERT(spa->spa_root_vdev == NULL);
1362 ASSERT(spa->spa_async_zio_root == NULL);
1363 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1365 spa_evicting_os_wait(spa);
1367 if (spa->spa_zvol_taskq) {
1368 taskq_destroy(spa->spa_zvol_taskq);
1369 spa->spa_zvol_taskq = NULL;
1372 if (spa->spa_prefetch_taskq) {
1373 taskq_destroy(spa->spa_prefetch_taskq);
1374 spa->spa_prefetch_taskq = NULL;
1377 if (spa->spa_upgrade_taskq) {
1378 taskq_destroy(spa->spa_upgrade_taskq);
1379 spa->spa_upgrade_taskq = NULL;
1382 txg_list_destroy(&spa->spa_vdev_txg_list);
1384 list_destroy(&spa->spa_config_dirty_list);
1385 list_destroy(&spa->spa_evicting_os_list);
1386 list_destroy(&spa->spa_state_dirty_list);
1388 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
1390 for (int t = 0; t < ZIO_TYPES; t++) {
1391 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1392 spa_taskqs_fini(spa, t, q);
1396 for (size_t i = 0; i < TXG_SIZE; i++) {
1397 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1398 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1399 spa->spa_txg_zio[i] = NULL;
1402 metaslab_class_destroy(spa->spa_normal_class);
1403 spa->spa_normal_class = NULL;
1405 metaslab_class_destroy(spa->spa_log_class);
1406 spa->spa_log_class = NULL;
1408 metaslab_class_destroy(spa->spa_embedded_log_class);
1409 spa->spa_embedded_log_class = NULL;
1411 metaslab_class_destroy(spa->spa_special_class);
1412 spa->spa_special_class = NULL;
1414 metaslab_class_destroy(spa->spa_dedup_class);
1415 spa->spa_dedup_class = NULL;
1418 * If this was part of an import or the open otherwise failed, we may
1419 * still have errors left in the queues. Empty them just in case.
1421 spa_errlog_drain(spa);
1422 avl_destroy(&spa->spa_errlist_scrub);
1423 avl_destroy(&spa->spa_errlist_last);
1425 spa_keystore_fini(&spa->spa_keystore);
1427 spa->spa_state = POOL_STATE_UNINITIALIZED;
1429 mutex_enter(&spa->spa_proc_lock);
1430 if (spa->spa_proc_state != SPA_PROC_NONE) {
1431 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1432 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1433 cv_broadcast(&spa->spa_proc_cv);
1434 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1435 ASSERT(spa->spa_proc != &p0);
1436 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1438 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1439 spa->spa_proc_state = SPA_PROC_NONE;
1441 ASSERT(spa->spa_proc == &p0);
1442 mutex_exit(&spa->spa_proc_lock);
1445 * We want to make sure spa_thread() has actually exited the ZFS
1446 * module, so that the module can't be unloaded out from underneath
1449 if (spa->spa_did != 0) {
1450 thread_join(spa->spa_did);
1456 * Verify a pool configuration, and construct the vdev tree appropriately. This
1457 * will create all the necessary vdevs in the appropriate layout, with each vdev
1458 * in the CLOSED state. This will prep the pool before open/creation/import.
1459 * All vdev validation is done by the vdev_alloc() routine.
1462 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1463 uint_t id, int atype)
1469 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1472 if ((*vdp)->vdev_ops->vdev_op_leaf)
1475 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1478 if (error == ENOENT)
1484 return (SET_ERROR(EINVAL));
1487 for (int c = 0; c < children; c++) {
1489 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1497 ASSERT(*vdp != NULL);
1503 spa_should_flush_logs_on_unload(spa_t *spa)
1505 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1508 if (!spa_writeable(spa))
1511 if (!spa->spa_sync_on)
1514 if (spa_state(spa) != POOL_STATE_EXPORTED)
1517 if (zfs_keep_log_spacemaps_at_export)
1524 * Opens a transaction that will set the flag that will instruct
1525 * spa_sync to attempt to flush all the metaslabs for that txg.
1528 spa_unload_log_sm_flush_all(spa_t *spa)
1530 dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
1531 VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
1533 ASSERT3U(spa->spa_log_flushall_txg, ==, 0);
1534 spa->spa_log_flushall_txg = dmu_tx_get_txg(tx);
1537 txg_wait_synced(spa_get_dsl(spa), spa->spa_log_flushall_txg);
1541 spa_unload_log_sm_metadata(spa_t *spa)
1543 void *cookie = NULL;
1545 while ((sls = avl_destroy_nodes(&spa->spa_sm_logs_by_txg,
1546 &cookie)) != NULL) {
1547 VERIFY0(sls->sls_mscount);
1548 kmem_free(sls, sizeof (spa_log_sm_t));
1551 for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
1552 e != NULL; e = list_head(&spa->spa_log_summary)) {
1553 VERIFY0(e->lse_mscount);
1554 list_remove(&spa->spa_log_summary, e);
1555 kmem_free(e, sizeof (log_summary_entry_t));
1558 spa->spa_unflushed_stats.sus_nblocks = 0;
1559 spa->spa_unflushed_stats.sus_memused = 0;
1560 spa->spa_unflushed_stats.sus_blocklimit = 0;
1564 spa_destroy_aux_threads(spa_t *spa)
1566 if (spa->spa_condense_zthr != NULL) {
1567 zthr_destroy(spa->spa_condense_zthr);
1568 spa->spa_condense_zthr = NULL;
1570 if (spa->spa_checkpoint_discard_zthr != NULL) {
1571 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1572 spa->spa_checkpoint_discard_zthr = NULL;
1574 if (spa->spa_livelist_delete_zthr != NULL) {
1575 zthr_destroy(spa->spa_livelist_delete_zthr);
1576 spa->spa_livelist_delete_zthr = NULL;
1578 if (spa->spa_livelist_condense_zthr != NULL) {
1579 zthr_destroy(spa->spa_livelist_condense_zthr);
1580 spa->spa_livelist_condense_zthr = NULL;
1585 * Opposite of spa_load().
1588 spa_unload(spa_t *spa)
1590 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1591 ASSERT(spa_state(spa) != POOL_STATE_UNINITIALIZED);
1593 spa_import_progress_remove(spa_guid(spa));
1594 spa_load_note(spa, "UNLOADING");
1596 spa_wake_waiters(spa);
1599 * If the log space map feature is enabled and the pool is getting
1600 * exported (but not destroyed), we want to spend some time flushing
1601 * as many metaslabs as we can in an attempt to destroy log space
1602 * maps and save import time.
1604 if (spa_should_flush_logs_on_unload(spa))
1605 spa_unload_log_sm_flush_all(spa);
1610 spa_async_suspend(spa);
1612 if (spa->spa_root_vdev) {
1613 vdev_t *root_vdev = spa->spa_root_vdev;
1614 vdev_initialize_stop_all(root_vdev, VDEV_INITIALIZE_ACTIVE);
1615 vdev_trim_stop_all(root_vdev, VDEV_TRIM_ACTIVE);
1616 vdev_autotrim_stop_all(spa);
1617 vdev_rebuild_stop_all(spa);
1623 if (spa->spa_sync_on) {
1624 txg_sync_stop(spa->spa_dsl_pool);
1625 spa->spa_sync_on = B_FALSE;
1629 * This ensures that there is no async metaslab prefetching
1630 * while we attempt to unload the spa.
1632 if (spa->spa_root_vdev != NULL) {
1633 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) {
1634 vdev_t *vc = spa->spa_root_vdev->vdev_child[c];
1635 if (vc->vdev_mg != NULL)
1636 taskq_wait(vc->vdev_mg->mg_taskq);
1640 if (spa->spa_mmp.mmp_thread)
1641 mmp_thread_stop(spa);
1644 * Wait for any outstanding async I/O to complete.
1646 if (spa->spa_async_zio_root != NULL) {
1647 for (int i = 0; i < max_ncpus; i++)
1648 (void) zio_wait(spa->spa_async_zio_root[i]);
1649 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1650 spa->spa_async_zio_root = NULL;
1653 if (spa->spa_vdev_removal != NULL) {
1654 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1655 spa->spa_vdev_removal = NULL;
1658 spa_destroy_aux_threads(spa);
1660 spa_condense_fini(spa);
1662 bpobj_close(&spa->spa_deferred_bpobj);
1664 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1669 if (spa->spa_root_vdev)
1670 vdev_free(spa->spa_root_vdev);
1671 ASSERT(spa->spa_root_vdev == NULL);
1674 * Close the dsl pool.
1676 if (spa->spa_dsl_pool) {
1677 dsl_pool_close(spa->spa_dsl_pool);
1678 spa->spa_dsl_pool = NULL;
1679 spa->spa_meta_objset = NULL;
1683 spa_unload_log_sm_metadata(spa);
1686 * Drop and purge level 2 cache
1688 spa_l2cache_drop(spa);
1690 for (int i = 0; i < spa->spa_spares.sav_count; i++)
1691 vdev_free(spa->spa_spares.sav_vdevs[i]);
1692 if (spa->spa_spares.sav_vdevs) {
1693 kmem_free(spa->spa_spares.sav_vdevs,
1694 spa->spa_spares.sav_count * sizeof (void *));
1695 spa->spa_spares.sav_vdevs = NULL;
1697 if (spa->spa_spares.sav_config) {
1698 nvlist_free(spa->spa_spares.sav_config);
1699 spa->spa_spares.sav_config = NULL;
1701 spa->spa_spares.sav_count = 0;
1703 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) {
1704 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1705 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1707 if (spa->spa_l2cache.sav_vdevs) {
1708 kmem_free(spa->spa_l2cache.sav_vdevs,
1709 spa->spa_l2cache.sav_count * sizeof (void *));
1710 spa->spa_l2cache.sav_vdevs = NULL;
1712 if (spa->spa_l2cache.sav_config) {
1713 nvlist_free(spa->spa_l2cache.sav_config);
1714 spa->spa_l2cache.sav_config = NULL;
1716 spa->spa_l2cache.sav_count = 0;
1718 spa->spa_async_suspended = 0;
1720 spa->spa_indirect_vdevs_loaded = B_FALSE;
1722 if (spa->spa_comment != NULL) {
1723 spa_strfree(spa->spa_comment);
1724 spa->spa_comment = NULL;
1726 if (spa->spa_compatibility != NULL) {
1727 spa_strfree(spa->spa_compatibility);
1728 spa->spa_compatibility = NULL;
1731 spa_config_exit(spa, SCL_ALL, spa);
1735 * Load (or re-load) the current list of vdevs describing the active spares for
1736 * this pool. When this is called, we have some form of basic information in
1737 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1738 * then re-generate a more complete list including status information.
1741 spa_load_spares(spa_t *spa)
1750 * zdb opens both the current state of the pool and the
1751 * checkpointed state (if present), with a different spa_t.
1753 * As spare vdevs are shared among open pools, we skip loading
1754 * them when we load the checkpointed state of the pool.
1756 if (!spa_writeable(spa))
1760 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1763 * First, close and free any existing spare vdevs.
1765 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1766 vd = spa->spa_spares.sav_vdevs[i];
1768 /* Undo the call to spa_activate() below */
1769 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1770 B_FALSE)) != NULL && tvd->vdev_isspare)
1771 spa_spare_remove(tvd);
1776 if (spa->spa_spares.sav_vdevs)
1777 kmem_free(spa->spa_spares.sav_vdevs,
1778 spa->spa_spares.sav_count * sizeof (void *));
1780 if (spa->spa_spares.sav_config == NULL)
1783 VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1784 ZPOOL_CONFIG_SPARES, &spares, &nspares));
1786 spa->spa_spares.sav_count = (int)nspares;
1787 spa->spa_spares.sav_vdevs = NULL;
1793 * Construct the array of vdevs, opening them to get status in the
1794 * process. For each spare, there is potentially two different vdev_t
1795 * structures associated with it: one in the list of spares (used only
1796 * for basic validation purposes) and one in the active vdev
1797 * configuration (if it's spared in). During this phase we open and
1798 * validate each vdev on the spare list. If the vdev also exists in the
1799 * active configuration, then we also mark this vdev as an active spare.
1801 spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *),
1803 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1804 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1805 VDEV_ALLOC_SPARE) == 0);
1808 spa->spa_spares.sav_vdevs[i] = vd;
1810 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1811 B_FALSE)) != NULL) {
1812 if (!tvd->vdev_isspare)
1816 * We only mark the spare active if we were successfully
1817 * able to load the vdev. Otherwise, importing a pool
1818 * with a bad active spare would result in strange
1819 * behavior, because multiple pool would think the spare
1820 * is actively in use.
1822 * There is a vulnerability here to an equally bizarre
1823 * circumstance, where a dead active spare is later
1824 * brought back to life (onlined or otherwise). Given
1825 * the rarity of this scenario, and the extra complexity
1826 * it adds, we ignore the possibility.
1828 if (!vdev_is_dead(tvd))
1829 spa_spare_activate(tvd);
1833 vd->vdev_aux = &spa->spa_spares;
1835 if (vdev_open(vd) != 0)
1838 if (vdev_validate_aux(vd) == 0)
1843 * Recompute the stashed list of spares, with status information
1846 fnvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES);
1848 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1850 for (i = 0; i < spa->spa_spares.sav_count; i++)
1851 spares[i] = vdev_config_generate(spa,
1852 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1853 fnvlist_add_nvlist_array(spa->spa_spares.sav_config,
1854 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count);
1855 for (i = 0; i < spa->spa_spares.sav_count; i++)
1856 nvlist_free(spares[i]);
1857 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1861 * Load (or re-load) the current list of vdevs describing the active l2cache for
1862 * this pool. When this is called, we have some form of basic information in
1863 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1864 * then re-generate a more complete list including status information.
1865 * Devices which are already active have their details maintained, and are
1869 spa_load_l2cache(spa_t *spa)
1871 nvlist_t **l2cache = NULL;
1873 int i, j, oldnvdevs;
1875 vdev_t *vd, **oldvdevs, **newvdevs;
1876 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1880 * zdb opens both the current state of the pool and the
1881 * checkpointed state (if present), with a different spa_t.
1883 * As L2 caches are part of the ARC which is shared among open
1884 * pools, we skip loading them when we load the checkpointed
1885 * state of the pool.
1887 if (!spa_writeable(spa))
1891 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1893 oldvdevs = sav->sav_vdevs;
1894 oldnvdevs = sav->sav_count;
1895 sav->sav_vdevs = NULL;
1898 if (sav->sav_config == NULL) {
1904 VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config,
1905 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache));
1906 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1909 * Process new nvlist of vdevs.
1911 for (i = 0; i < nl2cache; i++) {
1912 guid = fnvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID);
1915 for (j = 0; j < oldnvdevs; j++) {
1917 if (vd != NULL && guid == vd->vdev_guid) {
1919 * Retain previous vdev for add/remove ops.
1927 if (newvdevs[i] == NULL) {
1931 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1932 VDEV_ALLOC_L2CACHE) == 0);
1937 * Commit this vdev as an l2cache device,
1938 * even if it fails to open.
1940 spa_l2cache_add(vd);
1945 spa_l2cache_activate(vd);
1947 if (vdev_open(vd) != 0)
1950 (void) vdev_validate_aux(vd);
1952 if (!vdev_is_dead(vd))
1953 l2arc_add_vdev(spa, vd);
1956 * Upon cache device addition to a pool or pool
1957 * creation with a cache device or if the header
1958 * of the device is invalid we issue an async
1959 * TRIM command for the whole device which will
1960 * execute if l2arc_trim_ahead > 0.
1962 spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM);
1966 sav->sav_vdevs = newvdevs;
1967 sav->sav_count = (int)nl2cache;
1970 * Recompute the stashed list of l2cache devices, with status
1971 * information this time.
1973 fnvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE);
1975 if (sav->sav_count > 0)
1976 l2cache = kmem_alloc(sav->sav_count * sizeof (void *),
1978 for (i = 0; i < sav->sav_count; i++)
1979 l2cache[i] = vdev_config_generate(spa,
1980 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1981 fnvlist_add_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, l2cache,
1986 * Purge vdevs that were dropped
1988 for (i = 0; i < oldnvdevs; i++) {
1993 ASSERT(vd->vdev_isl2cache);
1995 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1996 pool != 0ULL && l2arc_vdev_present(vd))
1997 l2arc_remove_vdev(vd);
1998 vdev_clear_stats(vd);
2004 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
2006 for (i = 0; i < sav->sav_count; i++)
2007 nvlist_free(l2cache[i]);
2009 kmem_free(l2cache, sav->sav_count * sizeof (void *));
2013 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
2016 char *packed = NULL;
2021 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
2025 nvsize = *(uint64_t *)db->db_data;
2026 dmu_buf_rele(db, FTAG);
2028 packed = vmem_alloc(nvsize, KM_SLEEP);
2029 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
2032 error = nvlist_unpack(packed, nvsize, value, 0);
2033 vmem_free(packed, nvsize);
2039 * Concrete top-level vdevs that are not missing and are not logs. At every
2040 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
2043 spa_healthy_core_tvds(spa_t *spa)
2045 vdev_t *rvd = spa->spa_root_vdev;
2048 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
2049 vdev_t *vd = rvd->vdev_child[i];
2052 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
2060 * Checks to see if the given vdev could not be opened, in which case we post a
2061 * sysevent to notify the autoreplace code that the device has been removed.
2064 spa_check_removed(vdev_t *vd)
2066 for (uint64_t c = 0; c < vd->vdev_children; c++)
2067 spa_check_removed(vd->vdev_child[c]);
2069 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
2070 vdev_is_concrete(vd)) {
2071 zfs_post_autoreplace(vd->vdev_spa, vd);
2072 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
2077 spa_check_for_missing_logs(spa_t *spa)
2079 vdev_t *rvd = spa->spa_root_vdev;
2082 * If we're doing a normal import, then build up any additional
2083 * diagnostic information about missing log devices.
2084 * We'll pass this up to the user for further processing.
2086 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
2087 nvlist_t **child, *nv;
2090 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *),
2092 nv = fnvlist_alloc();
2094 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2095 vdev_t *tvd = rvd->vdev_child[c];
2098 * We consider a device as missing only if it failed
2099 * to open (i.e. offline or faulted is not considered
2102 if (tvd->vdev_islog &&
2103 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
2104 child[idx++] = vdev_config_generate(spa, tvd,
2105 B_FALSE, VDEV_CONFIG_MISSING);
2110 fnvlist_add_nvlist_array(nv,
2111 ZPOOL_CONFIG_CHILDREN, child, idx);
2112 fnvlist_add_nvlist(spa->spa_load_info,
2113 ZPOOL_CONFIG_MISSING_DEVICES, nv);
2115 for (uint64_t i = 0; i < idx; i++)
2116 nvlist_free(child[i]);
2119 kmem_free(child, rvd->vdev_children * sizeof (char **));
2122 spa_load_failed(spa, "some log devices are missing");
2123 vdev_dbgmsg_print_tree(rvd, 2);
2124 return (SET_ERROR(ENXIO));
2127 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2128 vdev_t *tvd = rvd->vdev_child[c];
2130 if (tvd->vdev_islog &&
2131 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
2132 spa_set_log_state(spa, SPA_LOG_CLEAR);
2133 spa_load_note(spa, "some log devices are "
2134 "missing, ZIL is dropped.");
2135 vdev_dbgmsg_print_tree(rvd, 2);
2145 * Check for missing log devices
2148 spa_check_logs(spa_t *spa)
2150 boolean_t rv = B_FALSE;
2151 dsl_pool_t *dp = spa_get_dsl(spa);
2153 switch (spa->spa_log_state) {
2156 case SPA_LOG_MISSING:
2157 /* need to recheck in case slog has been restored */
2158 case SPA_LOG_UNKNOWN:
2159 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2160 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
2162 spa_set_log_state(spa, SPA_LOG_MISSING);
2169 * Passivate any log vdevs (note, does not apply to embedded log metaslabs).
2172 spa_passivate_log(spa_t *spa)
2174 vdev_t *rvd = spa->spa_root_vdev;
2175 boolean_t slog_found = B_FALSE;
2177 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2179 for (int c = 0; c < rvd->vdev_children; c++) {
2180 vdev_t *tvd = rvd->vdev_child[c];
2182 if (tvd->vdev_islog) {
2183 ASSERT3P(tvd->vdev_log_mg, ==, NULL);
2184 metaslab_group_passivate(tvd->vdev_mg);
2185 slog_found = B_TRUE;
2189 return (slog_found);
2193 * Activate any log vdevs (note, does not apply to embedded log metaslabs).
2196 spa_activate_log(spa_t *spa)
2198 vdev_t *rvd = spa->spa_root_vdev;
2200 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2202 for (int c = 0; c < rvd->vdev_children; c++) {
2203 vdev_t *tvd = rvd->vdev_child[c];
2205 if (tvd->vdev_islog) {
2206 ASSERT3P(tvd->vdev_log_mg, ==, NULL);
2207 metaslab_group_activate(tvd->vdev_mg);
2213 spa_reset_logs(spa_t *spa)
2217 error = dmu_objset_find(spa_name(spa), zil_reset,
2218 NULL, DS_FIND_CHILDREN);
2221 * We successfully offlined the log device, sync out the
2222 * current txg so that the "stubby" block can be removed
2225 txg_wait_synced(spa->spa_dsl_pool, 0);
2231 spa_aux_check_removed(spa_aux_vdev_t *sav)
2233 for (int i = 0; i < sav->sav_count; i++)
2234 spa_check_removed(sav->sav_vdevs[i]);
2238 spa_claim_notify(zio_t *zio)
2240 spa_t *spa = zio->io_spa;
2245 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
2246 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2247 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2248 mutex_exit(&spa->spa_props_lock);
2251 typedef struct spa_load_error {
2252 uint64_t sle_meta_count;
2253 uint64_t sle_data_count;
2257 spa_load_verify_done(zio_t *zio)
2259 blkptr_t *bp = zio->io_bp;
2260 spa_load_error_t *sle = zio->io_private;
2261 dmu_object_type_t type = BP_GET_TYPE(bp);
2262 int error = zio->io_error;
2263 spa_t *spa = zio->io_spa;
2265 abd_free(zio->io_abd);
2267 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2268 type != DMU_OT_INTENT_LOG)
2269 atomic_inc_64(&sle->sle_meta_count);
2271 atomic_inc_64(&sle->sle_data_count);
2274 mutex_enter(&spa->spa_scrub_lock);
2275 spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
2276 cv_broadcast(&spa->spa_scrub_io_cv);
2277 mutex_exit(&spa->spa_scrub_lock);
2281 * Maximum number of inflight bytes is the log2 fraction of the arc size.
2282 * By default, we set it to 1/16th of the arc.
2284 int spa_load_verify_shift = 4;
2285 int spa_load_verify_metadata = B_TRUE;
2286 int spa_load_verify_data = B_TRUE;
2290 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2291 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2293 if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
2294 BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp))
2297 * Note: normally this routine will not be called if
2298 * spa_load_verify_metadata is not set. However, it may be useful
2299 * to manually set the flag after the traversal has begun.
2301 if (!spa_load_verify_metadata)
2303 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2306 uint64_t maxinflight_bytes =
2307 arc_target_bytes() >> spa_load_verify_shift;
2309 size_t size = BP_GET_PSIZE(bp);
2311 mutex_enter(&spa->spa_scrub_lock);
2312 while (spa->spa_load_verify_bytes >= maxinflight_bytes)
2313 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2314 spa->spa_load_verify_bytes += size;
2315 mutex_exit(&spa->spa_scrub_lock);
2317 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2318 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2319 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2320 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2326 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2328 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2329 return (SET_ERROR(ENAMETOOLONG));
2335 spa_load_verify(spa_t *spa)
2338 spa_load_error_t sle = { 0 };
2339 zpool_load_policy_t policy;
2340 boolean_t verify_ok = B_FALSE;
2343 zpool_get_load_policy(spa->spa_config, &policy);
2345 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2348 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2349 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2350 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2352 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2356 rio = zio_root(spa, NULL, &sle,
2357 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2359 if (spa_load_verify_metadata) {
2360 if (spa->spa_extreme_rewind) {
2361 spa_load_note(spa, "performing a complete scan of the "
2362 "pool since extreme rewind is on. This may take "
2363 "a very long time.\n (spa_load_verify_data=%u, "
2364 "spa_load_verify_metadata=%u)",
2365 spa_load_verify_data, spa_load_verify_metadata);
2368 error = traverse_pool(spa, spa->spa_verify_min_txg,
2369 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
2370 TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio);
2373 (void) zio_wait(rio);
2374 ASSERT0(spa->spa_load_verify_bytes);
2376 spa->spa_load_meta_errors = sle.sle_meta_count;
2377 spa->spa_load_data_errors = sle.sle_data_count;
2379 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2380 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2381 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2382 (u_longlong_t)sle.sle_data_count);
2385 if (spa_load_verify_dryrun ||
2386 (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2387 sle.sle_data_count <= policy.zlp_maxdata)) {
2391 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2392 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2394 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2395 fnvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_LOAD_TIME,
2396 spa->spa_load_txg_ts);
2397 fnvlist_add_int64(spa->spa_load_info, ZPOOL_CONFIG_REWIND_TIME,
2399 fnvlist_add_uint64(spa->spa_load_info,
2400 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count);
2402 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2405 if (spa_load_verify_dryrun)
2409 if (error != ENXIO && error != EIO)
2410 error = SET_ERROR(EIO);
2414 return (verify_ok ? 0 : EIO);
2418 * Find a value in the pool props object.
2421 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2423 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2424 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2428 * Find a value in the pool directory object.
2431 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2433 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2434 name, sizeof (uint64_t), 1, val);
2436 if (error != 0 && (error != ENOENT || log_enoent)) {
2437 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2438 "[error=%d]", name, error);
2445 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2447 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2448 return (SET_ERROR(err));
2452 spa_livelist_delete_check(spa_t *spa)
2454 return (spa->spa_livelists_to_delete != 0);
2459 spa_livelist_delete_cb_check(void *arg, zthr_t *z)
2462 return (spa_livelist_delete_check(spa));
2466 delete_blkptr_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2469 zio_free(spa, tx->tx_txg, bp);
2470 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD,
2471 -bp_get_dsize_sync(spa, bp),
2472 -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx);
2477 dsl_get_next_livelist_obj(objset_t *os, uint64_t zap_obj, uint64_t *llp)
2482 zap_cursor_init(&zc, os, zap_obj);
2483 err = zap_cursor_retrieve(&zc, &za);
2484 zap_cursor_fini(&zc);
2486 *llp = za.za_first_integer;
2491 * Components of livelist deletion that must be performed in syncing
2492 * context: freeing block pointers and updating the pool-wide data
2493 * structures to indicate how much work is left to do
2495 typedef struct sublist_delete_arg {
2500 } sublist_delete_arg_t;
2503 sublist_delete_sync(void *arg, dmu_tx_t *tx)
2505 sublist_delete_arg_t *sda = arg;
2506 spa_t *spa = sda->spa;
2507 dsl_deadlist_t *ll = sda->ll;
2508 uint64_t key = sda->key;
2509 bplist_t *to_free = sda->to_free;
2511 bplist_iterate(to_free, delete_blkptr_cb, spa, tx);
2512 dsl_deadlist_remove_entry(ll, key, tx);
2515 typedef struct livelist_delete_arg {
2519 } livelist_delete_arg_t;
2522 livelist_delete_sync(void *arg, dmu_tx_t *tx)
2524 livelist_delete_arg_t *lda = arg;
2525 spa_t *spa = lda->spa;
2526 uint64_t ll_obj = lda->ll_obj;
2527 uint64_t zap_obj = lda->zap_obj;
2528 objset_t *mos = spa->spa_meta_objset;
2531 /* free the livelist and decrement the feature count */
2532 VERIFY0(zap_remove_int(mos, zap_obj, ll_obj, tx));
2533 dsl_deadlist_free(mos, ll_obj, tx);
2534 spa_feature_decr(spa, SPA_FEATURE_LIVELIST, tx);
2535 VERIFY0(zap_count(mos, zap_obj, &count));
2537 /* no more livelists to delete */
2538 VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT,
2539 DMU_POOL_DELETED_CLONES, tx));
2540 VERIFY0(zap_destroy(mos, zap_obj, tx));
2541 spa->spa_livelists_to_delete = 0;
2542 spa_notify_waiters(spa);
2547 * Load in the value for the livelist to be removed and open it. Then,
2548 * load its first sublist and determine which block pointers should actually
2549 * be freed. Then, call a synctask which performs the actual frees and updates
2550 * the pool-wide livelist data.
2554 spa_livelist_delete_cb(void *arg, zthr_t *z)
2557 uint64_t ll_obj = 0, count;
2558 objset_t *mos = spa->spa_meta_objset;
2559 uint64_t zap_obj = spa->spa_livelists_to_delete;
2561 * Determine the next livelist to delete. This function should only
2562 * be called if there is at least one deleted clone.
2564 VERIFY0(dsl_get_next_livelist_obj(mos, zap_obj, &ll_obj));
2565 VERIFY0(zap_count(mos, ll_obj, &count));
2568 dsl_deadlist_entry_t *dle;
2570 ll = kmem_zalloc(sizeof (dsl_deadlist_t), KM_SLEEP);
2571 dsl_deadlist_open(ll, mos, ll_obj);
2572 dle = dsl_deadlist_first(ll);
2573 ASSERT3P(dle, !=, NULL);
2574 bplist_create(&to_free);
2575 int err = dsl_process_sub_livelist(&dle->dle_bpobj, &to_free,
2578 sublist_delete_arg_t sync_arg = {
2581 .key = dle->dle_mintxg,
2584 zfs_dbgmsg("deleting sublist (id %llu) from"
2585 " livelist %llu, %lld remaining",
2586 (u_longlong_t)dle->dle_bpobj.bpo_object,
2587 (u_longlong_t)ll_obj, (longlong_t)count - 1);
2588 VERIFY0(dsl_sync_task(spa_name(spa), NULL,
2589 sublist_delete_sync, &sync_arg, 0,
2590 ZFS_SPACE_CHECK_DESTROY));
2592 VERIFY3U(err, ==, EINTR);
2594 bplist_clear(&to_free);
2595 bplist_destroy(&to_free);
2596 dsl_deadlist_close(ll);
2597 kmem_free(ll, sizeof (dsl_deadlist_t));
2599 livelist_delete_arg_t sync_arg = {
2604 zfs_dbgmsg("deletion of livelist %llu completed",
2605 (u_longlong_t)ll_obj);
2606 VERIFY0(dsl_sync_task(spa_name(spa), NULL, livelist_delete_sync,
2607 &sync_arg, 0, ZFS_SPACE_CHECK_DESTROY));
2612 spa_start_livelist_destroy_thread(spa_t *spa)
2614 ASSERT3P(spa->spa_livelist_delete_zthr, ==, NULL);
2615 spa->spa_livelist_delete_zthr =
2616 zthr_create("z_livelist_destroy",
2617 spa_livelist_delete_cb_check, spa_livelist_delete_cb, spa,
2621 typedef struct livelist_new_arg {
2624 } livelist_new_arg_t;
2627 livelist_track_new_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
2631 livelist_new_arg_t *lna = arg;
2633 bplist_append(lna->frees, bp);
2635 bplist_append(lna->allocs, bp);
2636 zfs_livelist_condense_new_alloc++;
2641 typedef struct livelist_condense_arg {
2644 uint64_t first_size;
2646 } livelist_condense_arg_t;
2649 spa_livelist_condense_sync(void *arg, dmu_tx_t *tx)
2651 livelist_condense_arg_t *lca = arg;
2652 spa_t *spa = lca->spa;
2654 dsl_dataset_t *ds = spa->spa_to_condense.ds;
2656 /* Have we been cancelled? */
2657 if (spa->spa_to_condense.cancelled) {
2658 zfs_livelist_condense_sync_cancel++;
2662 dsl_deadlist_entry_t *first = spa->spa_to_condense.first;
2663 dsl_deadlist_entry_t *next = spa->spa_to_condense.next;
2664 dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist;
2667 * It's possible that the livelist was changed while the zthr was
2668 * running. Therefore, we need to check for new blkptrs in the two
2669 * entries being condensed and continue to track them in the livelist.
2670 * Because of the way we handle remapped blkptrs (see dbuf_remap_impl),
2671 * it's possible that the newly added blkptrs are FREEs or ALLOCs so
2672 * we need to sort them into two different bplists.
2674 uint64_t first_obj = first->dle_bpobj.bpo_object;
2675 uint64_t next_obj = next->dle_bpobj.bpo_object;
2676 uint64_t cur_first_size = first->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2677 uint64_t cur_next_size = next->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2679 bplist_create(&new_frees);
2680 livelist_new_arg_t new_bps = {
2681 .allocs = &lca->to_keep,
2682 .frees = &new_frees,
2685 if (cur_first_size > lca->first_size) {
2686 VERIFY0(livelist_bpobj_iterate_from_nofree(&first->dle_bpobj,
2687 livelist_track_new_cb, &new_bps, lca->first_size));
2689 if (cur_next_size > lca->next_size) {
2690 VERIFY0(livelist_bpobj_iterate_from_nofree(&next->dle_bpobj,
2691 livelist_track_new_cb, &new_bps, lca->next_size));
2694 dsl_deadlist_clear_entry(first, ll, tx);
2695 ASSERT(bpobj_is_empty(&first->dle_bpobj));
2696 dsl_deadlist_remove_entry(ll, next->dle_mintxg, tx);
2698 bplist_iterate(&lca->to_keep, dsl_deadlist_insert_alloc_cb, ll, tx);
2699 bplist_iterate(&new_frees, dsl_deadlist_insert_free_cb, ll, tx);
2700 bplist_destroy(&new_frees);
2702 char dsname[ZFS_MAX_DATASET_NAME_LEN];
2703 dsl_dataset_name(ds, dsname);
2704 zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu "
2705 "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu "
2706 "(%llu blkptrs)", (u_longlong_t)tx->tx_txg, dsname,
2707 (u_longlong_t)ds->ds_object, (u_longlong_t)first_obj,
2708 (u_longlong_t)cur_first_size, (u_longlong_t)next_obj,
2709 (u_longlong_t)cur_next_size,
2710 (u_longlong_t)first->dle_bpobj.bpo_object,
2711 (u_longlong_t)first->dle_bpobj.bpo_phys->bpo_num_blkptrs);
2713 dmu_buf_rele(ds->ds_dbuf, spa);
2714 spa->spa_to_condense.ds = NULL;
2715 bplist_clear(&lca->to_keep);
2716 bplist_destroy(&lca->to_keep);
2717 kmem_free(lca, sizeof (livelist_condense_arg_t));
2718 spa->spa_to_condense.syncing = B_FALSE;
2722 spa_livelist_condense_cb(void *arg, zthr_t *t)
2724 while (zfs_livelist_condense_zthr_pause &&
2725 !(zthr_has_waiters(t) || zthr_iscancelled(t)))
2729 dsl_deadlist_entry_t *first = spa->spa_to_condense.first;
2730 dsl_deadlist_entry_t *next = spa->spa_to_condense.next;
2731 uint64_t first_size, next_size;
2733 livelist_condense_arg_t *lca =
2734 kmem_alloc(sizeof (livelist_condense_arg_t), KM_SLEEP);
2735 bplist_create(&lca->to_keep);
2738 * Process the livelists (matching FREEs and ALLOCs) in open context
2739 * so we have minimal work in syncing context to condense.
2741 * We save bpobj sizes (first_size and next_size) to use later in
2742 * syncing context to determine if entries were added to these sublists
2743 * while in open context. This is possible because the clone is still
2744 * active and open for normal writes and we want to make sure the new,
2745 * unprocessed blockpointers are inserted into the livelist normally.
2747 * Note that dsl_process_sub_livelist() both stores the size number of
2748 * blockpointers and iterates over them while the bpobj's lock held, so
2749 * the sizes returned to us are consistent which what was actually
2752 int err = dsl_process_sub_livelist(&first->dle_bpobj, &lca->to_keep, t,
2755 err = dsl_process_sub_livelist(&next->dle_bpobj, &lca->to_keep,
2759 while (zfs_livelist_condense_sync_pause &&
2760 !(zthr_has_waiters(t) || zthr_iscancelled(t)))
2763 dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
2764 dmu_tx_mark_netfree(tx);
2765 dmu_tx_hold_space(tx, 1);
2766 err = dmu_tx_assign(tx, TXG_NOWAIT | TXG_NOTHROTTLE);
2769 * Prevent the condense zthr restarting before
2770 * the synctask completes.
2772 spa->spa_to_condense.syncing = B_TRUE;
2774 lca->first_size = first_size;
2775 lca->next_size = next_size;
2776 dsl_sync_task_nowait(spa_get_dsl(spa),
2777 spa_livelist_condense_sync, lca, tx);
2783 * Condensing can not continue: either it was externally stopped or
2784 * we were unable to assign to a tx because the pool has run out of
2785 * space. In the second case, we'll just end up trying to condense
2786 * again in a later txg.
2789 bplist_clear(&lca->to_keep);
2790 bplist_destroy(&lca->to_keep);
2791 kmem_free(lca, sizeof (livelist_condense_arg_t));
2792 dmu_buf_rele(spa->spa_to_condense.ds->ds_dbuf, spa);
2793 spa->spa_to_condense.ds = NULL;
2795 zfs_livelist_condense_zthr_cancel++;
2800 * Check that there is something to condense but that a condense is not
2801 * already in progress and that condensing has not been cancelled.
2804 spa_livelist_condense_cb_check(void *arg, zthr_t *z)
2807 if ((spa->spa_to_condense.ds != NULL) &&
2808 (spa->spa_to_condense.syncing == B_FALSE) &&
2809 (spa->spa_to_condense.cancelled == B_FALSE)) {
2816 spa_start_livelist_condensing_thread(spa_t *spa)
2818 spa->spa_to_condense.ds = NULL;
2819 spa->spa_to_condense.first = NULL;
2820 spa->spa_to_condense.next = NULL;
2821 spa->spa_to_condense.syncing = B_FALSE;
2822 spa->spa_to_condense.cancelled = B_FALSE;
2824 ASSERT3P(spa->spa_livelist_condense_zthr, ==, NULL);
2825 spa->spa_livelist_condense_zthr =
2826 zthr_create("z_livelist_condense",
2827 spa_livelist_condense_cb_check,
2828 spa_livelist_condense_cb, spa, minclsyspri);
2832 spa_spawn_aux_threads(spa_t *spa)
2834 ASSERT(spa_writeable(spa));
2836 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2838 spa_start_indirect_condensing_thread(spa);
2839 spa_start_livelist_destroy_thread(spa);
2840 spa_start_livelist_condensing_thread(spa);
2842 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2843 spa->spa_checkpoint_discard_zthr =
2844 zthr_create("z_checkpoint_discard",
2845 spa_checkpoint_discard_thread_check,
2846 spa_checkpoint_discard_thread, spa, minclsyspri);
2850 * Fix up config after a partly-completed split. This is done with the
2851 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2852 * pool have that entry in their config, but only the splitting one contains
2853 * a list of all the guids of the vdevs that are being split off.
2855 * This function determines what to do with that list: either rejoin
2856 * all the disks to the pool, or complete the splitting process. To attempt
2857 * the rejoin, each disk that is offlined is marked online again, and
2858 * we do a reopen() call. If the vdev label for every disk that was
2859 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2860 * then we call vdev_split() on each disk, and complete the split.
2862 * Otherwise we leave the config alone, with all the vdevs in place in
2863 * the original pool.
2866 spa_try_repair(spa_t *spa, nvlist_t *config)
2873 boolean_t attempt_reopen;
2875 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2878 /* check that the config is complete */
2879 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2880 &glist, &gcount) != 0)
2883 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2885 /* attempt to online all the vdevs & validate */
2886 attempt_reopen = B_TRUE;
2887 for (i = 0; i < gcount; i++) {
2888 if (glist[i] == 0) /* vdev is hole */
2891 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2892 if (vd[i] == NULL) {
2894 * Don't bother attempting to reopen the disks;
2895 * just do the split.
2897 attempt_reopen = B_FALSE;
2899 /* attempt to re-online it */
2900 vd[i]->vdev_offline = B_FALSE;
2904 if (attempt_reopen) {
2905 vdev_reopen(spa->spa_root_vdev);
2907 /* check each device to see what state it's in */
2908 for (extracted = 0, i = 0; i < gcount; i++) {
2909 if (vd[i] != NULL &&
2910 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2917 * If every disk has been moved to the new pool, or if we never
2918 * even attempted to look at them, then we split them off for
2921 if (!attempt_reopen || gcount == extracted) {
2922 for (i = 0; i < gcount; i++)
2925 vdev_reopen(spa->spa_root_vdev);
2928 kmem_free(vd, gcount * sizeof (vdev_t *));
2932 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2934 char *ereport = FM_EREPORT_ZFS_POOL;
2937 spa->spa_load_state = state;
2938 (void) spa_import_progress_set_state(spa_guid(spa),
2939 spa_load_state(spa));
2941 gethrestime(&spa->spa_loaded_ts);
2942 error = spa_load_impl(spa, type, &ereport);
2945 * Don't count references from objsets that are already closed
2946 * and are making their way through the eviction process.
2948 spa_evicting_os_wait(spa);
2949 spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
2951 if (error != EEXIST) {
2952 spa->spa_loaded_ts.tv_sec = 0;
2953 spa->spa_loaded_ts.tv_nsec = 0;
2955 if (error != EBADF) {
2956 (void) zfs_ereport_post(ereport, spa,
2957 NULL, NULL, NULL, 0);
2960 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2963 (void) spa_import_progress_set_state(spa_guid(spa),
2964 spa_load_state(spa));
2971 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2972 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2973 * spa's per-vdev ZAP list.
2976 vdev_count_verify_zaps(vdev_t *vd)
2978 spa_t *spa = vd->vdev_spa;
2981 if (vd->vdev_top_zap != 0) {
2983 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2984 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2986 if (vd->vdev_leaf_zap != 0) {
2988 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2989 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2992 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2993 total += vdev_count_verify_zaps(vd->vdev_child[i]);
3001 * Determine whether the activity check is required.
3004 spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label,
3008 uint64_t hostid = 0;
3009 uint64_t tryconfig_txg = 0;
3010 uint64_t tryconfig_timestamp = 0;
3011 uint16_t tryconfig_mmp_seq = 0;
3014 if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
3015 nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
3016 (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG,
3018 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3019 &tryconfig_timestamp);
3020 (void) nvlist_lookup_uint16(nvinfo, ZPOOL_CONFIG_MMP_SEQ,
3021 &tryconfig_mmp_seq);
3024 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state);
3027 * Disable the MMP activity check - This is used by zdb which
3028 * is intended to be used on potentially active pools.
3030 if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP)
3034 * Skip the activity check when the MMP feature is disabled.
3036 if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0)
3040 * If the tryconfig_ values are nonzero, they are the results of an
3041 * earlier tryimport. If they all match the uberblock we just found,
3042 * then the pool has not changed and we return false so we do not test
3045 if (tryconfig_txg && tryconfig_txg == ub->ub_txg &&
3046 tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp &&
3047 tryconfig_mmp_seq && tryconfig_mmp_seq ==
3048 (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0))
3052 * Allow the activity check to be skipped when importing the pool
3053 * on the same host which last imported it. Since the hostid from
3054 * configuration may be stale use the one read from the label.
3056 if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID))
3057 hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID);
3059 if (hostid == spa_get_hostid(spa))
3063 * Skip the activity test when the pool was cleanly exported.
3065 if (state != POOL_STATE_ACTIVE)
3072 * Nanoseconds the activity check must watch for changes on-disk.
3075 spa_activity_check_duration(spa_t *spa, uberblock_t *ub)
3077 uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1);
3078 uint64_t multihost_interval = MSEC2NSEC(
3079 MMP_INTERVAL_OK(zfs_multihost_interval));
3080 uint64_t import_delay = MAX(NANOSEC, import_intervals *
3081 multihost_interval);
3084 * Local tunables determine a minimum duration except for the case
3085 * where we know when the remote host will suspend the pool if MMP
3086 * writes do not land.
3088 * See Big Theory comment at the top of mmp.c for the reasoning behind
3089 * these cases and times.
3092 ASSERT(MMP_IMPORT_SAFETY_FACTOR >= 100);
3094 if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
3095 MMP_FAIL_INT(ub) > 0) {
3097 /* MMP on remote host will suspend pool after failed writes */
3098 import_delay = MMP_FAIL_INT(ub) * MSEC2NSEC(MMP_INTERVAL(ub)) *
3099 MMP_IMPORT_SAFETY_FACTOR / 100;
3101 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3102 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3103 "import_intervals=%llu", (u_longlong_t)import_delay,
3104 (u_longlong_t)MMP_FAIL_INT(ub),
3105 (u_longlong_t)MMP_INTERVAL(ub),
3106 (u_longlong_t)import_intervals);
3108 } else if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
3109 MMP_FAIL_INT(ub) == 0) {
3111 /* MMP on remote host will never suspend pool */
3112 import_delay = MAX(import_delay, (MSEC2NSEC(MMP_INTERVAL(ub)) +
3113 ub->ub_mmp_delay) * import_intervals);
3115 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3116 "mmp_interval=%llu ub_mmp_delay=%llu "
3117 "import_intervals=%llu", (u_longlong_t)import_delay,
3118 (u_longlong_t)MMP_INTERVAL(ub),
3119 (u_longlong_t)ub->ub_mmp_delay,
3120 (u_longlong_t)import_intervals);
3122 } else if (MMP_VALID(ub)) {
3124 * zfs-0.7 compatibility case
3127 import_delay = MAX(import_delay, (multihost_interval +
3128 ub->ub_mmp_delay) * import_intervals);
3130 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3131 "import_intervals=%llu leaves=%u",
3132 (u_longlong_t)import_delay,
3133 (u_longlong_t)ub->ub_mmp_delay,
3134 (u_longlong_t)import_intervals,
3135 vdev_count_leaves(spa));
3137 /* Using local tunings is the only reasonable option */
3138 zfs_dbgmsg("pool last imported on non-MMP aware "
3139 "host using import_delay=%llu multihost_interval=%llu "
3140 "import_intervals=%llu", (u_longlong_t)import_delay,
3141 (u_longlong_t)multihost_interval,
3142 (u_longlong_t)import_intervals);
3145 return (import_delay);
3149 * Perform the import activity check. If the user canceled the import or
3150 * we detected activity then fail.
3153 spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config)
3155 uint64_t txg = ub->ub_txg;
3156 uint64_t timestamp = ub->ub_timestamp;
3157 uint64_t mmp_config = ub->ub_mmp_config;
3158 uint16_t mmp_seq = MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0;
3159 uint64_t import_delay;
3160 hrtime_t import_expire;
3161 nvlist_t *mmp_label = NULL;
3162 vdev_t *rvd = spa->spa_root_vdev;
3167 cv_init(&cv, NULL, CV_DEFAULT, NULL);
3168 mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL);
3172 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3173 * during the earlier tryimport. If the txg recorded there is 0 then
3174 * the pool is known to be active on another host.
3176 * Otherwise, the pool might be in use on another host. Check for
3177 * changes in the uberblocks on disk if necessary.
3179 if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
3180 nvlist_t *nvinfo = fnvlist_lookup_nvlist(config,
3181 ZPOOL_CONFIG_LOAD_INFO);
3183 if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) &&
3184 fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) {
3185 vdev_uberblock_load(rvd, ub, &mmp_label);
3186 error = SET_ERROR(EREMOTEIO);
3191 import_delay = spa_activity_check_duration(spa, ub);
3193 /* Add a small random factor in case of simultaneous imports (0-25%) */
3194 import_delay += import_delay * random_in_range(250) / 1000;
3196 import_expire = gethrtime() + import_delay;
3198 while (gethrtime() < import_expire) {
3199 (void) spa_import_progress_set_mmp_check(spa_guid(spa),
3200 NSEC2SEC(import_expire - gethrtime()));
3202 vdev_uberblock_load(rvd, ub, &mmp_label);
3204 if (txg != ub->ub_txg || timestamp != ub->ub_timestamp ||
3205 mmp_seq != (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) {
3206 zfs_dbgmsg("multihost activity detected "
3207 "txg %llu ub_txg %llu "
3208 "timestamp %llu ub_timestamp %llu "
3209 "mmp_config %#llx ub_mmp_config %#llx",
3210 (u_longlong_t)txg, (u_longlong_t)ub->ub_txg,
3211 (u_longlong_t)timestamp,
3212 (u_longlong_t)ub->ub_timestamp,
3213 (u_longlong_t)mmp_config,
3214 (u_longlong_t)ub->ub_mmp_config);
3216 error = SET_ERROR(EREMOTEIO);
3221 nvlist_free(mmp_label);
3225 error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz);
3227 error = SET_ERROR(EINTR);
3235 mutex_destroy(&mtx);
3239 * If the pool is determined to be active store the status in the
3240 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3241 * available from configuration read from disk store them as well.
3242 * This allows 'zpool import' to generate a more useful message.
3244 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3245 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3246 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3248 if (error == EREMOTEIO) {
3249 char *hostname = "<unknown>";
3250 uint64_t hostid = 0;
3253 if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) {
3254 hostname = fnvlist_lookup_string(mmp_label,
3255 ZPOOL_CONFIG_HOSTNAME);
3256 fnvlist_add_string(spa->spa_load_info,
3257 ZPOOL_CONFIG_MMP_HOSTNAME, hostname);
3260 if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) {
3261 hostid = fnvlist_lookup_uint64(mmp_label,
3262 ZPOOL_CONFIG_HOSTID);
3263 fnvlist_add_uint64(spa->spa_load_info,
3264 ZPOOL_CONFIG_MMP_HOSTID, hostid);
3268 fnvlist_add_uint64(spa->spa_load_info,
3269 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE);
3270 fnvlist_add_uint64(spa->spa_load_info,
3271 ZPOOL_CONFIG_MMP_TXG, 0);
3273 error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO);
3277 nvlist_free(mmp_label);
3283 spa_verify_host(spa_t *spa, nvlist_t *mos_config)
3287 uint64_t myhostid = 0;
3289 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
3290 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
3291 hostname = fnvlist_lookup_string(mos_config,
3292 ZPOOL_CONFIG_HOSTNAME);
3294 myhostid = zone_get_hostid(NULL);
3296 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
3297 cmn_err(CE_WARN, "pool '%s' could not be "
3298 "loaded as it was last accessed by "
3299 "another system (host: %s hostid: 0x%llx). "
3300 "See: https://openzfs.github.io/openzfs-docs/msg/"
3302 spa_name(spa), hostname, (u_longlong_t)hostid);
3303 spa_load_failed(spa, "hostid verification failed: pool "
3304 "last accessed by host: %s (hostid: 0x%llx)",
3305 hostname, (u_longlong_t)hostid);
3306 return (SET_ERROR(EBADF));
3314 spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
3317 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
3322 char *compatibility;
3325 * Versioning wasn't explicitly added to the label until later, so if
3326 * it's not present treat it as the initial version.
3328 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
3329 &spa->spa_ubsync.ub_version) != 0)
3330 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
3332 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
3333 spa_load_failed(spa, "invalid config provided: '%s' missing",
3334 ZPOOL_CONFIG_POOL_GUID);
3335 return (SET_ERROR(EINVAL));
3339 * If we are doing an import, ensure that the pool is not already
3340 * imported by checking if its pool guid already exists in the
3343 * The only case that we allow an already imported pool to be
3344 * imported again, is when the pool is checkpointed and we want to
3345 * look at its checkpointed state from userland tools like zdb.
3348 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
3349 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
3350 spa_guid_exists(pool_guid, 0)) {
3352 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
3353 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
3354 spa_guid_exists(pool_guid, 0) &&
3355 !spa_importing_readonly_checkpoint(spa)) {
3357 spa_load_failed(spa, "a pool with guid %llu is already open",
3358 (u_longlong_t)pool_guid);
3359 return (SET_ERROR(EEXIST));
3362 spa->spa_config_guid = pool_guid;
3364 nvlist_free(spa->spa_load_info);
3365 spa->spa_load_info = fnvlist_alloc();
3367 ASSERT(spa->spa_comment == NULL);
3368 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
3369 spa->spa_comment = spa_strdup(comment);
3371 ASSERT(spa->spa_compatibility == NULL);
3372 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMPATIBILITY,
3373 &compatibility) == 0)
3374 spa->spa_compatibility = spa_strdup(compatibility);
3376 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
3377 &spa->spa_config_txg);
3379 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
3380 spa->spa_config_splitting = fnvlist_dup(nvl);
3382 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
3383 spa_load_failed(spa, "invalid config provided: '%s' missing",
3384 ZPOOL_CONFIG_VDEV_TREE);
3385 return (SET_ERROR(EINVAL));
3389 * Create "The Godfather" zio to hold all async IOs
3391 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3393 for (int i = 0; i < max_ncpus; i++) {
3394 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3395 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3396 ZIO_FLAG_GODFATHER);
3400 * Parse the configuration into a vdev tree. We explicitly set the
3401 * value that will be returned by spa_version() since parsing the
3402 * configuration requires knowing the version number.
3404 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3405 parse = (type == SPA_IMPORT_EXISTING ?
3406 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
3407 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
3408 spa_config_exit(spa, SCL_ALL, FTAG);
3411 spa_load_failed(spa, "unable to parse config [error=%d]",
3416 ASSERT(spa->spa_root_vdev == rvd);
3417 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
3418 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
3420 if (type != SPA_IMPORT_ASSEMBLE) {
3421 ASSERT(spa_guid(spa) == pool_guid);
3428 * Recursively open all vdevs in the vdev tree. This function is called twice:
3429 * first with the untrusted config, then with the trusted config.
3432 spa_ld_open_vdevs(spa_t *spa)
3437 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3438 * missing/unopenable for the root vdev to be still considered openable.
3440 if (spa->spa_trust_config) {
3441 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
3442 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
3443 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
3444 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
3445 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
3447 spa->spa_missing_tvds_allowed = 0;
3450 spa->spa_missing_tvds_allowed =
3451 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
3453 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3454 error = vdev_open(spa->spa_root_vdev);
3455 spa_config_exit(spa, SCL_ALL, FTAG);
3457 if (spa->spa_missing_tvds != 0) {
3458 spa_load_note(spa, "vdev tree has %lld missing top-level "
3459 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
3460 if (spa->spa_trust_config && (spa->spa_mode & SPA_MODE_WRITE)) {
3462 * Although theoretically we could allow users to open
3463 * incomplete pools in RW mode, we'd need to add a lot
3464 * of extra logic (e.g. adjust pool space to account
3465 * for missing vdevs).
3466 * This limitation also prevents users from accidentally
3467 * opening the pool in RW mode during data recovery and
3468 * damaging it further.
3470 spa_load_note(spa, "pools with missing top-level "
3471 "vdevs can only be opened in read-only mode.");
3472 error = SET_ERROR(ENXIO);
3474 spa_load_note(spa, "current settings allow for maximum "
3475 "%lld missing top-level vdevs at this stage.",
3476 (u_longlong_t)spa->spa_missing_tvds_allowed);
3480 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
3483 if (spa->spa_missing_tvds != 0 || error != 0)
3484 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
3490 * We need to validate the vdev labels against the configuration that
3491 * we have in hand. This function is called twice: first with an untrusted
3492 * config, then with a trusted config. The validation is more strict when the
3493 * config is trusted.
3496 spa_ld_validate_vdevs(spa_t *spa)
3499 vdev_t *rvd = spa->spa_root_vdev;
3501 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3502 error = vdev_validate(rvd);
3503 spa_config_exit(spa, SCL_ALL, FTAG);
3506 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
3510 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
3511 spa_load_failed(spa, "cannot open vdev tree after invalidating "
3513 vdev_dbgmsg_print_tree(rvd, 2);
3514 return (SET_ERROR(ENXIO));
3521 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
3523 spa->spa_state = POOL_STATE_ACTIVE;
3524 spa->spa_ubsync = spa->spa_uberblock;
3525 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
3526 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
3527 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
3528 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
3529 spa->spa_claim_max_txg = spa->spa_first_txg;
3530 spa->spa_prev_software_version = ub->ub_software_version;
3534 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
3536 vdev_t *rvd = spa->spa_root_vdev;
3538 uberblock_t *ub = &spa->spa_uberblock;
3539 boolean_t activity_check = B_FALSE;
3542 * If we are opening the checkpointed state of the pool by
3543 * rewinding to it, at this point we will have written the
3544 * checkpointed uberblock to the vdev labels, so searching
3545 * the labels will find the right uberblock. However, if
3546 * we are opening the checkpointed state read-only, we have
3547 * not modified the labels. Therefore, we must ignore the
3548 * labels and continue using the spa_uberblock that was set
3549 * by spa_ld_checkpoint_rewind.
3551 * Note that it would be fine to ignore the labels when
3552 * rewinding (opening writeable) as well. However, if we
3553 * crash just after writing the labels, we will end up
3554 * searching the labels. Doing so in the common case means
3555 * that this code path gets exercised normally, rather than
3556 * just in the edge case.
3558 if (ub->ub_checkpoint_txg != 0 &&
3559 spa_importing_readonly_checkpoint(spa)) {
3560 spa_ld_select_uberblock_done(spa, ub);
3565 * Find the best uberblock.
3567 vdev_uberblock_load(rvd, ub, &label);
3570 * If we weren't able to find a single valid uberblock, return failure.
3572 if (ub->ub_txg == 0) {
3574 spa_load_failed(spa, "no valid uberblock found");
3575 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
3578 if (spa->spa_load_max_txg != UINT64_MAX) {
3579 (void) spa_import_progress_set_max_txg(spa_guid(spa),
3580 (u_longlong_t)spa->spa_load_max_txg);
3582 spa_load_note(spa, "using uberblock with txg=%llu",
3583 (u_longlong_t)ub->ub_txg);
3587 * For pools which have the multihost property on determine if the
3588 * pool is truly inactive and can be safely imported. Prevent
3589 * hosts which don't have a hostid set from importing the pool.
3591 activity_check = spa_activity_check_required(spa, ub, label,
3593 if (activity_check) {
3594 if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay &&
3595 spa_get_hostid(spa) == 0) {
3597 fnvlist_add_uint64(spa->spa_load_info,
3598 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
3599 return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
3602 int error = spa_activity_check(spa, ub, spa->spa_config);
3608 fnvlist_add_uint64(spa->spa_load_info,
3609 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE);
3610 fnvlist_add_uint64(spa->spa_load_info,
3611 ZPOOL_CONFIG_MMP_TXG, ub->ub_txg);
3612 fnvlist_add_uint16(spa->spa_load_info,
3613 ZPOOL_CONFIG_MMP_SEQ,
3614 (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0));
3618 * If the pool has an unsupported version we can't open it.
3620 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
3622 spa_load_failed(spa, "version %llu is not supported",
3623 (u_longlong_t)ub->ub_version);
3624 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
3627 if (ub->ub_version >= SPA_VERSION_FEATURES) {
3631 * If we weren't able to find what's necessary for reading the
3632 * MOS in the label, return failure.
3634 if (label == NULL) {
3635 spa_load_failed(spa, "label config unavailable");
3636 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3640 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
3643 spa_load_failed(spa, "invalid label: '%s' missing",
3644 ZPOOL_CONFIG_FEATURES_FOR_READ);
3645 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3650 * Update our in-core representation with the definitive values
3653 nvlist_free(spa->spa_label_features);
3654 spa->spa_label_features = fnvlist_dup(features);
3660 * Look through entries in the label nvlist's features_for_read. If
3661 * there is a feature listed there which we don't understand then we
3662 * cannot open a pool.
3664 if (ub->ub_version >= SPA_VERSION_FEATURES) {
3665 nvlist_t *unsup_feat;
3667 unsup_feat = fnvlist_alloc();
3669 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
3671 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
3672 if (!zfeature_is_supported(nvpair_name(nvp))) {
3673 fnvlist_add_string(unsup_feat,
3674 nvpair_name(nvp), "");
3678 if (!nvlist_empty(unsup_feat)) {
3679 fnvlist_add_nvlist(spa->spa_load_info,
3680 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3681 nvlist_free(unsup_feat);
3682 spa_load_failed(spa, "some features are unsupported");
3683 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3687 nvlist_free(unsup_feat);
3690 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
3691 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3692 spa_try_repair(spa, spa->spa_config);
3693 spa_config_exit(spa, SCL_ALL, FTAG);
3694 nvlist_free(spa->spa_config_splitting);
3695 spa->spa_config_splitting = NULL;
3699 * Initialize internal SPA structures.
3701 spa_ld_select_uberblock_done(spa, ub);
3707 spa_ld_open_rootbp(spa_t *spa)
3710 vdev_t *rvd = spa->spa_root_vdev;
3712 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
3714 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
3715 "[error=%d]", error);
3716 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3718 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
3724 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
3725 boolean_t reloading)
3727 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
3728 nvlist_t *nv, *mos_config, *policy;
3729 int error = 0, copy_error;
3730 uint64_t healthy_tvds, healthy_tvds_mos;
3731 uint64_t mos_config_txg;
3733 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
3735 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3738 * If we're assembling a pool from a split, the config provided is
3739 * already trusted so there is nothing to do.
3741 if (type == SPA_IMPORT_ASSEMBLE)
3744 healthy_tvds = spa_healthy_core_tvds(spa);
3746 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
3748 spa_load_failed(spa, "unable to retrieve MOS config");
3749 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3753 * If we are doing an open, pool owner wasn't verified yet, thus do
3754 * the verification here.
3756 if (spa->spa_load_state == SPA_LOAD_OPEN) {
3757 error = spa_verify_host(spa, mos_config);
3759 nvlist_free(mos_config);
3764 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
3766 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3769 * Build a new vdev tree from the trusted config
3771 error = spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD);
3773 nvlist_free(mos_config);
3774 spa_config_exit(spa, SCL_ALL, FTAG);
3775 spa_load_failed(spa, "spa_config_parse failed [error=%d]",
3777 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3781 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3782 * obtained by scanning /dev/dsk, then it will have the right vdev
3783 * paths. We update the trusted MOS config with this information.
3784 * We first try to copy the paths with vdev_copy_path_strict, which
3785 * succeeds only when both configs have exactly the same vdev tree.
3786 * If that fails, we fall back to a more flexible method that has a
3787 * best effort policy.
3789 copy_error = vdev_copy_path_strict(rvd, mrvd);
3790 if (copy_error != 0 || spa_load_print_vdev_tree) {
3791 spa_load_note(spa, "provided vdev tree:");
3792 vdev_dbgmsg_print_tree(rvd, 2);
3793 spa_load_note(spa, "MOS vdev tree:");
3794 vdev_dbgmsg_print_tree(mrvd, 2);
3796 if (copy_error != 0) {
3797 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
3798 "back to vdev_copy_path_relaxed");
3799 vdev_copy_path_relaxed(rvd, mrvd);
3804 spa->spa_root_vdev = mrvd;
3806 spa_config_exit(spa, SCL_ALL, FTAG);
3809 * We will use spa_config if we decide to reload the spa or if spa_load
3810 * fails and we rewind. We must thus regenerate the config using the
3811 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3812 * pass settings on how to load the pool and is not stored in the MOS.
3813 * We copy it over to our new, trusted config.
3815 mos_config_txg = fnvlist_lookup_uint64(mos_config,
3816 ZPOOL_CONFIG_POOL_TXG);
3817 nvlist_free(mos_config);
3818 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
3819 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
3821 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
3822 spa_config_set(spa, mos_config);
3823 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
3826 * Now that we got the config from the MOS, we should be more strict
3827 * in checking blkptrs and can make assumptions about the consistency
3828 * of the vdev tree. spa_trust_config must be set to true before opening
3829 * vdevs in order for them to be writeable.
3831 spa->spa_trust_config = B_TRUE;
3834 * Open and validate the new vdev tree
3836 error = spa_ld_open_vdevs(spa);
3840 error = spa_ld_validate_vdevs(spa);
3844 if (copy_error != 0 || spa_load_print_vdev_tree) {
3845 spa_load_note(spa, "final vdev tree:");
3846 vdev_dbgmsg_print_tree(rvd, 2);
3849 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
3850 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
3852 * Sanity check to make sure that we are indeed loading the
3853 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3854 * in the config provided and they happened to be the only ones
3855 * to have the latest uberblock, we could involuntarily perform
3856 * an extreme rewind.
3858 healthy_tvds_mos = spa_healthy_core_tvds(spa);
3859 if (healthy_tvds_mos - healthy_tvds >=
3860 SPA_SYNC_MIN_VDEVS) {
3861 spa_load_note(spa, "config provided misses too many "
3862 "top-level vdevs compared to MOS (%lld vs %lld). ",
3863 (u_longlong_t)healthy_tvds,
3864 (u_longlong_t)healthy_tvds_mos);
3865 spa_load_note(spa, "vdev tree:");
3866 vdev_dbgmsg_print_tree(rvd, 2);
3868 spa_load_failed(spa, "config was already "
3869 "provided from MOS. Aborting.");
3870 return (spa_vdev_err(rvd,
3871 VDEV_AUX_CORRUPT_DATA, EIO));
3873 spa_load_note(spa, "spa must be reloaded using MOS "
3875 return (SET_ERROR(EAGAIN));
3879 error = spa_check_for_missing_logs(spa);
3881 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
3883 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
3884 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
3885 "guid sum (%llu != %llu)",
3886 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
3887 (u_longlong_t)rvd->vdev_guid_sum);
3888 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
3896 spa_ld_open_indirect_vdev_metadata(spa_t *spa)
3899 vdev_t *rvd = spa->spa_root_vdev;
3902 * Everything that we read before spa_remove_init() must be stored
3903 * on concreted vdevs. Therefore we do this as early as possible.
3905 error = spa_remove_init(spa);
3907 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
3909 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3913 * Retrieve information needed to condense indirect vdev mappings.
3915 error = spa_condense_init(spa);
3917 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
3919 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3926 spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
3929 vdev_t *rvd = spa->spa_root_vdev;
3931 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
3932 boolean_t missing_feat_read = B_FALSE;
3933 nvlist_t *unsup_feat, *enabled_feat;
3935 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
3936 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
3937 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3940 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
3941 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
3942 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3945 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
3946 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
3947 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3950 enabled_feat = fnvlist_alloc();
3951 unsup_feat = fnvlist_alloc();
3953 if (!spa_features_check(spa, B_FALSE,
3954 unsup_feat, enabled_feat))
3955 missing_feat_read = B_TRUE;
3957 if (spa_writeable(spa) ||
3958 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3959 if (!spa_features_check(spa, B_TRUE,
3960 unsup_feat, enabled_feat)) {
3961 *missing_feat_writep = B_TRUE;
3965 fnvlist_add_nvlist(spa->spa_load_info,
3966 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3968 if (!nvlist_empty(unsup_feat)) {
3969 fnvlist_add_nvlist(spa->spa_load_info,
3970 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3973 fnvlist_free(enabled_feat);
3974 fnvlist_free(unsup_feat);
3976 if (!missing_feat_read) {
3977 fnvlist_add_boolean(spa->spa_load_info,
3978 ZPOOL_CONFIG_CAN_RDONLY);
3982 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3983 * twofold: to determine whether the pool is available for
3984 * import in read-write mode and (if it is not) whether the
3985 * pool is available for import in read-only mode. If the pool
3986 * is available for import in read-write mode, it is displayed
3987 * as available in userland; if it is not available for import
3988 * in read-only mode, it is displayed as unavailable in
3989 * userland. If the pool is available for import in read-only
3990 * mode but not read-write mode, it is displayed as unavailable
3991 * in userland with a special note that the pool is actually
3992 * available for open in read-only mode.
3994 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3995 * missing a feature for write, we must first determine whether
3996 * the pool can be opened read-only before returning to
3997 * userland in order to know whether to display the
3998 * abovementioned note.
4000 if (missing_feat_read || (*missing_feat_writep &&
4001 spa_writeable(spa))) {
4002 spa_load_failed(spa, "pool uses unsupported features");
4003 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
4008 * Load refcounts for ZFS features from disk into an in-memory
4009 * cache during SPA initialization.
4011 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
4014 error = feature_get_refcount_from_disk(spa,
4015 &spa_feature_table[i], &refcount);
4017 spa->spa_feat_refcount_cache[i] = refcount;
4018 } else if (error == ENOTSUP) {
4019 spa->spa_feat_refcount_cache[i] =
4020 SPA_FEATURE_DISABLED;
4022 spa_load_failed(spa, "error getting refcount "
4023 "for feature %s [error=%d]",
4024 spa_feature_table[i].fi_guid, error);
4025 return (spa_vdev_err(rvd,
4026 VDEV_AUX_CORRUPT_DATA, EIO));
4031 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
4032 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
4033 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
4034 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4038 * Encryption was added before bookmark_v2, even though bookmark_v2
4039 * is now a dependency. If this pool has encryption enabled without
4040 * bookmark_v2, trigger an errata message.
4042 if (spa_feature_is_enabled(spa, SPA_FEATURE_ENCRYPTION) &&
4043 !spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) {
4044 spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION;
4051 spa_ld_load_special_directories(spa_t *spa)
4054 vdev_t *rvd = spa->spa_root_vdev;
4056 spa->spa_is_initializing = B_TRUE;
4057 error = dsl_pool_open(spa->spa_dsl_pool);
4058 spa->spa_is_initializing = B_FALSE;
4060 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
4061 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4068 spa_ld_get_props(spa_t *spa)
4072 vdev_t *rvd = spa->spa_root_vdev;
4074 /* Grab the checksum salt from the MOS. */
4075 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
4076 DMU_POOL_CHECKSUM_SALT, 1,
4077 sizeof (spa->spa_cksum_salt.zcs_bytes),
4078 spa->spa_cksum_salt.zcs_bytes);
4079 if (error == ENOENT) {
4080 /* Generate a new salt for subsequent use */
4081 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
4082 sizeof (spa->spa_cksum_salt.zcs_bytes));
4083 } else if (error != 0) {
4084 spa_load_failed(spa, "unable to retrieve checksum salt from "
4085 "MOS [error=%d]", error);
4086 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4089 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
4090 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4091 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
4093 spa_load_failed(spa, "error opening deferred-frees bpobj "
4094 "[error=%d]", error);
4095 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4099 * Load the bit that tells us to use the new accounting function
4100 * (raid-z deflation). If we have an older pool, this will not
4103 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
4104 if (error != 0 && error != ENOENT)
4105 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4107 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
4108 &spa->spa_creation_version, B_FALSE);
4109 if (error != 0 && error != ENOENT)
4110 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4113 * Load the persistent error log. If we have an older pool, this will
4116 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
4118 if (error != 0 && error != ENOENT)
4119 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4121 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
4122 &spa->spa_errlog_scrub, B_FALSE);
4123 if (error != 0 && error != ENOENT)
4124 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4127 * Load the livelist deletion field. If a livelist is queued for
4128 * deletion, indicate that in the spa
4130 error = spa_dir_prop(spa, DMU_POOL_DELETED_CLONES,
4131 &spa->spa_livelists_to_delete, B_FALSE);
4132 if (error != 0 && error != ENOENT)
4133 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4136 * Load the history object. If we have an older pool, this
4137 * will not be present.
4139 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
4140 if (error != 0 && error != ENOENT)
4141 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4144 * Load the per-vdev ZAP map. If we have an older pool, this will not
4145 * be present; in this case, defer its creation to a later time to
4146 * avoid dirtying the MOS this early / out of sync context. See
4147 * spa_sync_config_object.
4150 /* The sentinel is only available in the MOS config. */
4151 nvlist_t *mos_config;
4152 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
4153 spa_load_failed(spa, "unable to retrieve MOS config");
4154 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4157 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
4158 &spa->spa_all_vdev_zaps, B_FALSE);
4160 if (error == ENOENT) {
4161 VERIFY(!nvlist_exists(mos_config,
4162 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
4163 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
4164 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
4165 } else if (error != 0) {
4166 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4167 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
4169 * An older version of ZFS overwrote the sentinel value, so
4170 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4171 * destruction to later; see spa_sync_config_object.
4173 spa->spa_avz_action = AVZ_ACTION_DESTROY;
4175 * We're assuming that no vdevs have had their ZAPs created
4176 * before this. Better be sure of it.
4178 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
4180 nvlist_free(mos_config);
4182 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4184 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
4186 if (error && error != ENOENT)
4187 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4190 uint64_t autoreplace = 0;
4192 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
4193 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
4194 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
4195 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
4196 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
4197 spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost);
4198 spa_prop_find(spa, ZPOOL_PROP_AUTOTRIM, &spa->spa_autotrim);
4199 spa->spa_autoreplace = (autoreplace != 0);
4203 * If we are importing a pool with missing top-level vdevs,
4204 * we enforce that the pool doesn't panic or get suspended on
4205 * error since the likelihood of missing data is extremely high.
4207 if (spa->spa_missing_tvds > 0 &&
4208 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
4209 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4210 spa_load_note(spa, "forcing failmode to 'continue' "
4211 "as some top level vdevs are missing");
4212 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
4219 spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
4222 vdev_t *rvd = spa->spa_root_vdev;
4225 * If we're assembling the pool from the split-off vdevs of
4226 * an existing pool, we don't want to attach the spares & cache
4231 * Load any hot spares for this pool.
4233 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
4235 if (error != 0 && error != ENOENT)
4236 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4237 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
4238 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
4239 if (load_nvlist(spa, spa->spa_spares.sav_object,
4240 &spa->spa_spares.sav_config) != 0) {
4241 spa_load_failed(spa, "error loading spares nvlist");
4242 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4245 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4246 spa_load_spares(spa);
4247 spa_config_exit(spa, SCL_ALL, FTAG);
4248 } else if (error == 0) {
4249 spa->spa_spares.sav_sync = B_TRUE;
4253 * Load any level 2 ARC devices for this pool.
4255 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
4256 &spa->spa_l2cache.sav_object, B_FALSE);
4257 if (error != 0 && error != ENOENT)
4258 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4259 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
4260 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
4261 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
4262 &spa->spa_l2cache.sav_config) != 0) {
4263 spa_load_failed(spa, "error loading l2cache nvlist");
4264 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4267 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4268 spa_load_l2cache(spa);
4269 spa_config_exit(spa, SCL_ALL, FTAG);
4270 } else if (error == 0) {
4271 spa->spa_l2cache.sav_sync = B_TRUE;
4278 spa_ld_load_vdev_metadata(spa_t *spa)
4281 vdev_t *rvd = spa->spa_root_vdev;
4284 * If the 'multihost' property is set, then never allow a pool to
4285 * be imported when the system hostid is zero. The exception to
4286 * this rule is zdb which is always allowed to access pools.
4288 if (spa_multihost(spa) && spa_get_hostid(spa) == 0 &&
4289 (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) {
4290 fnvlist_add_uint64(spa->spa_load_info,
4291 ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
4292 return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
4296 * If the 'autoreplace' property is set, then post a resource notifying
4297 * the ZFS DE that it should not issue any faults for unopenable
4298 * devices. We also iterate over the vdevs, and post a sysevent for any
4299 * unopenable vdevs so that the normal autoreplace handler can take
4302 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4303 spa_check_removed(spa->spa_root_vdev);
4305 * For the import case, this is done in spa_import(), because
4306 * at this point we're using the spare definitions from
4307 * the MOS config, not necessarily from the userland config.
4309 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
4310 spa_aux_check_removed(&spa->spa_spares);
4311 spa_aux_check_removed(&spa->spa_l2cache);
4316 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4318 error = vdev_load(rvd);
4320 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
4321 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
4324 error = spa_ld_log_spacemaps(spa);
4326 spa_load_failed(spa, "spa_ld_log_sm_data failed [error=%d]",
4328 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
4332 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4334 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4335 vdev_dtl_reassess(rvd, 0, 0, B_FALSE, B_FALSE);
4336 spa_config_exit(spa, SCL_ALL, FTAG);
4342 spa_ld_load_dedup_tables(spa_t *spa)
4345 vdev_t *rvd = spa->spa_root_vdev;
4347 error = ddt_load(spa);
4349 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
4350 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
4357 spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
4359 vdev_t *rvd = spa->spa_root_vdev;
4361 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
4362 boolean_t missing = spa_check_logs(spa);
4364 if (spa->spa_missing_tvds != 0) {
4365 spa_load_note(spa, "spa_check_logs failed "
4366 "so dropping the logs");
4368 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
4369 spa_load_failed(spa, "spa_check_logs failed");
4370 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
4380 spa_ld_verify_pool_data(spa_t *spa)
4383 vdev_t *rvd = spa->spa_root_vdev;
4386 * We've successfully opened the pool, verify that we're ready
4387 * to start pushing transactions.
4389 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
4390 error = spa_load_verify(spa);
4392 spa_load_failed(spa, "spa_load_verify failed "
4393 "[error=%d]", error);
4394 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
4403 spa_ld_claim_log_blocks(spa_t *spa)
4406 dsl_pool_t *dp = spa_get_dsl(spa);
4409 * Claim log blocks that haven't been committed yet.
4410 * This must all happen in a single txg.
4411 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4412 * invoked from zil_claim_log_block()'s i/o done callback.
4413 * Price of rollback is that we abandon the log.
4415 spa->spa_claiming = B_TRUE;
4417 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
4418 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
4419 zil_claim, tx, DS_FIND_CHILDREN);
4422 spa->spa_claiming = B_FALSE;
4424 spa_set_log_state(spa, SPA_LOG_GOOD);
4428 spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
4429 boolean_t update_config_cache)
4431 vdev_t *rvd = spa->spa_root_vdev;
4432 int need_update = B_FALSE;
4435 * If the config cache is stale, or we have uninitialized
4436 * metaslabs (see spa_vdev_add()), then update the config.
4438 * If this is a verbatim import, trust the current
4439 * in-core spa_config and update the disk labels.
4441 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
4442 spa->spa_load_state == SPA_LOAD_IMPORT ||
4443 spa->spa_load_state == SPA_LOAD_RECOVER ||
4444 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
4445 need_update = B_TRUE;
4447 for (int c = 0; c < rvd->vdev_children; c++)
4448 if (rvd->vdev_child[c]->vdev_ms_array == 0)
4449 need_update = B_TRUE;
4452 * Update the config cache asynchronously in case we're the
4453 * root pool, in which case the config cache isn't writable yet.
4456 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
4460 spa_ld_prepare_for_reload(spa_t *spa)
4462 spa_mode_t mode = spa->spa_mode;
4463 int async_suspended = spa->spa_async_suspended;
4466 spa_deactivate(spa);
4467 spa_activate(spa, mode);
4470 * We save the value of spa_async_suspended as it gets reset to 0 by
4471 * spa_unload(). We want to restore it back to the original value before
4472 * returning as we might be calling spa_async_resume() later.
4474 spa->spa_async_suspended = async_suspended;
4478 spa_ld_read_checkpoint_txg(spa_t *spa)
4480 uberblock_t checkpoint;
4483 ASSERT0(spa->spa_checkpoint_txg);
4484 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4486 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
4487 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
4488 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
4490 if (error == ENOENT)
4496 ASSERT3U(checkpoint.ub_txg, !=, 0);
4497 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
4498 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
4499 spa->spa_checkpoint_txg = checkpoint.ub_txg;
4500 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
4506 spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
4510 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4511 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
4514 * Never trust the config that is provided unless we are assembling
4515 * a pool following a split.
4516 * This means don't trust blkptrs and the vdev tree in general. This
4517 * also effectively puts the spa in read-only mode since
4518 * spa_writeable() checks for spa_trust_config to be true.
4519 * We will later load a trusted config from the MOS.
4521 if (type != SPA_IMPORT_ASSEMBLE)
4522 spa->spa_trust_config = B_FALSE;
4525 * Parse the config provided to create a vdev tree.
4527 error = spa_ld_parse_config(spa, type);
4531 spa_import_progress_add(spa);
4534 * Now that we have the vdev tree, try to open each vdev. This involves
4535 * opening the underlying physical device, retrieving its geometry and
4536 * probing the vdev with a dummy I/O. The state of each vdev will be set
4537 * based on the success of those operations. After this we'll be ready
4538 * to read from the vdevs.
4540 error = spa_ld_open_vdevs(spa);
4545 * Read the label of each vdev and make sure that the GUIDs stored
4546 * there match the GUIDs in the config provided.
4547 * If we're assembling a new pool that's been split off from an
4548 * existing pool, the labels haven't yet been updated so we skip
4549 * validation for now.
4551 if (type != SPA_IMPORT_ASSEMBLE) {
4552 error = spa_ld_validate_vdevs(spa);
4558 * Read all vdev labels to find the best uberblock (i.e. latest,
4559 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
4560 * get the list of features required to read blkptrs in the MOS from
4561 * the vdev label with the best uberblock and verify that our version
4562 * of zfs supports them all.
4564 error = spa_ld_select_uberblock(spa, type);
4569 * Pass that uberblock to the dsl_pool layer which will open the root
4570 * blkptr. This blkptr points to the latest version of the MOS and will
4571 * allow us to read its contents.
4573 error = spa_ld_open_rootbp(spa);
4581 spa_ld_checkpoint_rewind(spa_t *spa)
4583 uberblock_t checkpoint;
4586 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4587 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4589 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
4590 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
4591 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
4594 spa_load_failed(spa, "unable to retrieve checkpointed "
4595 "uberblock from the MOS config [error=%d]", error);
4597 if (error == ENOENT)
4598 error = ZFS_ERR_NO_CHECKPOINT;
4603 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
4604 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
4607 * We need to update the txg and timestamp of the checkpointed
4608 * uberblock to be higher than the latest one. This ensures that
4609 * the checkpointed uberblock is selected if we were to close and
4610 * reopen the pool right after we've written it in the vdev labels.
4611 * (also see block comment in vdev_uberblock_compare)
4613 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
4614 checkpoint.ub_timestamp = gethrestime_sec();
4617 * Set current uberblock to be the checkpointed uberblock.
4619 spa->spa_uberblock = checkpoint;
4622 * If we are doing a normal rewind, then the pool is open for
4623 * writing and we sync the "updated" checkpointed uberblock to
4624 * disk. Once this is done, we've basically rewound the whole
4625 * pool and there is no way back.
4627 * There are cases when we don't want to attempt and sync the
4628 * checkpointed uberblock to disk because we are opening a
4629 * pool as read-only. Specifically, verifying the checkpointed
4630 * state with zdb, and importing the checkpointed state to get
4631 * a "preview" of its content.
4633 if (spa_writeable(spa)) {
4634 vdev_t *rvd = spa->spa_root_vdev;
4636 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4637 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
4639 int children = rvd->vdev_children;
4640 int c0 = random_in_range(children);
4642 for (int c = 0; c < children; c++) {
4643 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
4645 /* Stop when revisiting the first vdev */
4646 if (c > 0 && svd[0] == vd)
4649 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
4650 !vdev_is_concrete(vd))
4653 svd[svdcount++] = vd;
4654 if (svdcount == SPA_SYNC_MIN_VDEVS)
4657 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
4659 spa->spa_last_synced_guid = rvd->vdev_guid;
4660 spa_config_exit(spa, SCL_ALL, FTAG);
4663 spa_load_failed(spa, "failed to write checkpointed "
4664 "uberblock to the vdev labels [error=%d]", error);
4673 spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
4674 boolean_t *update_config_cache)
4679 * Parse the config for pool, open and validate vdevs,
4680 * select an uberblock, and use that uberblock to open
4683 error = spa_ld_mos_init(spa, type);
4688 * Retrieve the trusted config stored in the MOS and use it to create
4689 * a new, exact version of the vdev tree, then reopen all vdevs.
4691 error = spa_ld_trusted_config(spa, type, B_FALSE);
4692 if (error == EAGAIN) {
4693 if (update_config_cache != NULL)
4694 *update_config_cache = B_TRUE;
4697 * Redo the loading process with the trusted config if it is
4698 * too different from the untrusted config.
4700 spa_ld_prepare_for_reload(spa);
4701 spa_load_note(spa, "RELOADING");
4702 error = spa_ld_mos_init(spa, type);
4706 error = spa_ld_trusted_config(spa, type, B_TRUE);
4710 } else if (error != 0) {
4718 * Load an existing storage pool, using the config provided. This config
4719 * describes which vdevs are part of the pool and is later validated against
4720 * partial configs present in each vdev's label and an entire copy of the
4721 * config stored in the MOS.
4724 spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
4727 boolean_t missing_feat_write = B_FALSE;
4728 boolean_t checkpoint_rewind =
4729 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4730 boolean_t update_config_cache = B_FALSE;
4732 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4733 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
4735 spa_load_note(spa, "LOADING");
4737 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
4742 * If we are rewinding to the checkpoint then we need to repeat
4743 * everything we've done so far in this function but this time
4744 * selecting the checkpointed uberblock and using that to open
4747 if (checkpoint_rewind) {
4749 * If we are rewinding to the checkpoint update config cache
4752 update_config_cache = B_TRUE;
4755 * Extract the checkpointed uberblock from the current MOS
4756 * and use this as the pool's uberblock from now on. If the
4757 * pool is imported as writeable we also write the checkpoint
4758 * uberblock to the labels, making the rewind permanent.
4760 error = spa_ld_checkpoint_rewind(spa);
4765 * Redo the loading process again with the
4766 * checkpointed uberblock.
4768 spa_ld_prepare_for_reload(spa);
4769 spa_load_note(spa, "LOADING checkpointed uberblock");
4770 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
4776 * Retrieve the checkpoint txg if the pool has a checkpoint.
4778 error = spa_ld_read_checkpoint_txg(spa);
4783 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
4784 * from the pool and their contents were re-mapped to other vdevs. Note
4785 * that everything that we read before this step must have been
4786 * rewritten on concrete vdevs after the last device removal was
4787 * initiated. Otherwise we could be reading from indirect vdevs before
4788 * we have loaded their mappings.
4790 error = spa_ld_open_indirect_vdev_metadata(spa);
4795 * Retrieve the full list of active features from the MOS and check if
4796 * they are all supported.
4798 error = spa_ld_check_features(spa, &missing_feat_write);
4803 * Load several special directories from the MOS needed by the dsl_pool
4806 error = spa_ld_load_special_directories(spa);
4811 * Retrieve pool properties from the MOS.
4813 error = spa_ld_get_props(spa);
4818 * Retrieve the list of auxiliary devices - cache devices and spares -
4821 error = spa_ld_open_aux_vdevs(spa, type);
4826 * Load the metadata for all vdevs. Also check if unopenable devices
4827 * should be autoreplaced.
4829 error = spa_ld_load_vdev_metadata(spa);
4833 error = spa_ld_load_dedup_tables(spa);
4838 * Verify the logs now to make sure we don't have any unexpected errors
4839 * when we claim log blocks later.
4841 error = spa_ld_verify_logs(spa, type, ereport);
4845 if (missing_feat_write) {
4846 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
4849 * At this point, we know that we can open the pool in
4850 * read-only mode but not read-write mode. We now have enough
4851 * information and can return to userland.
4853 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
4858 * Traverse the last txgs to make sure the pool was left off in a safe
4859 * state. When performing an extreme rewind, we verify the whole pool,
4860 * which can take a very long time.
4862 error = spa_ld_verify_pool_data(spa);
4867 * Calculate the deflated space for the pool. This must be done before
4868 * we write anything to the pool because we'd need to update the space
4869 * accounting using the deflated sizes.
4871 spa_update_dspace(spa);
4874 * We have now retrieved all the information we needed to open the
4875 * pool. If we are importing the pool in read-write mode, a few
4876 * additional steps must be performed to finish the import.
4878 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
4879 spa->spa_load_max_txg == UINT64_MAX)) {
4880 uint64_t config_cache_txg = spa->spa_config_txg;
4882 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
4885 * In case of a checkpoint rewind, log the original txg
4886 * of the checkpointed uberblock.
4888 if (checkpoint_rewind) {
4889 spa_history_log_internal(spa, "checkpoint rewind",
4890 NULL, "rewound state to txg=%llu",
4891 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
4895 * Traverse the ZIL and claim all blocks.
4897 spa_ld_claim_log_blocks(spa);
4900 * Kick-off the syncing thread.
4902 spa->spa_sync_on = B_TRUE;
4903 txg_sync_start(spa->spa_dsl_pool);
4904 mmp_thread_start(spa);
4907 * Wait for all claims to sync. We sync up to the highest
4908 * claimed log block birth time so that claimed log blocks
4909 * don't appear to be from the future. spa_claim_max_txg
4910 * will have been set for us by ZIL traversal operations
4913 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
4916 * Check if we need to request an update of the config. On the
4917 * next sync, we would update the config stored in vdev labels
4918 * and the cachefile (by default /etc/zfs/zpool.cache).
4920 spa_ld_check_for_config_update(spa, config_cache_txg,
4921 update_config_cache);
4924 * Check if a rebuild was in progress and if so resume it.
4925 * Then check all DTLs to see if anything needs resilvering.
4926 * The resilver will be deferred if a rebuild was started.
4928 if (vdev_rebuild_active(spa->spa_root_vdev)) {
4929 vdev_rebuild_restart(spa);
4930 } else if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
4931 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4932 spa_async_request(spa, SPA_ASYNC_RESILVER);
4936 * Log the fact that we booted up (so that we can detect if
4937 * we rebooted in the middle of an operation).
4939 spa_history_log_version(spa, "open", NULL);
4941 spa_restart_removal(spa);
4942 spa_spawn_aux_threads(spa);
4945 * Delete any inconsistent datasets.
4948 * Since we may be issuing deletes for clones here,
4949 * we make sure to do so after we've spawned all the
4950 * auxiliary threads above (from which the livelist
4951 * deletion zthr is part of).
4953 (void) dmu_objset_find(spa_name(spa),
4954 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
4957 * Clean up any stale temporary dataset userrefs.
4959 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
4961 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4962 vdev_initialize_restart(spa->spa_root_vdev);
4963 vdev_trim_restart(spa->spa_root_vdev);
4964 vdev_autotrim_restart(spa);
4965 spa_config_exit(spa, SCL_CONFIG, FTAG);
4968 spa_import_progress_remove(spa_guid(spa));
4969 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD);
4971 spa_load_note(spa, "LOADED");
4977 spa_load_retry(spa_t *spa, spa_load_state_t state)
4979 spa_mode_t mode = spa->spa_mode;
4982 spa_deactivate(spa);
4984 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
4986 spa_activate(spa, mode);
4987 spa_async_suspend(spa);
4989 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
4990 (u_longlong_t)spa->spa_load_max_txg);
4992 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
4996 * If spa_load() fails this function will try loading prior txg's. If
4997 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
4998 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
4999 * function will not rewind the pool and will return the same error as
5003 spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
5006 nvlist_t *loadinfo = NULL;
5007 nvlist_t *config = NULL;
5008 int load_error, rewind_error;
5009 uint64_t safe_rewind_txg;
5012 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
5013 spa->spa_load_max_txg = spa->spa_load_txg;
5014 spa_set_log_state(spa, SPA_LOG_CLEAR);
5016 spa->spa_load_max_txg = max_request;
5017 if (max_request != UINT64_MAX)
5018 spa->spa_extreme_rewind = B_TRUE;
5021 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
5022 if (load_error == 0)
5024 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
5026 * When attempting checkpoint-rewind on a pool with no
5027 * checkpoint, we should not attempt to load uberblocks
5028 * from previous txgs when spa_load fails.
5030 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
5031 spa_import_progress_remove(spa_guid(spa));
5032 return (load_error);
5035 if (spa->spa_root_vdev != NULL)
5036 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5038 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
5039 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
5041 if (rewind_flags & ZPOOL_NEVER_REWIND) {
5042 nvlist_free(config);
5043 spa_import_progress_remove(spa_guid(spa));
5044 return (load_error);
5047 if (state == SPA_LOAD_RECOVER) {
5048 /* Price of rolling back is discarding txgs, including log */
5049 spa_set_log_state(spa, SPA_LOG_CLEAR);
5052 * If we aren't rolling back save the load info from our first
5053 * import attempt so that we can restore it after attempting
5056 loadinfo = spa->spa_load_info;
5057 spa->spa_load_info = fnvlist_alloc();
5060 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
5061 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
5062 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
5063 TXG_INITIAL : safe_rewind_txg;
5066 * Continue as long as we're finding errors, we're still within
5067 * the acceptable rewind range, and we're still finding uberblocks
5069 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
5070 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
5071 if (spa->spa_load_max_txg < safe_rewind_txg)
5072 spa->spa_extreme_rewind = B_TRUE;
5073 rewind_error = spa_load_retry(spa, state);
5076 spa->spa_extreme_rewind = B_FALSE;
5077 spa->spa_load_max_txg = UINT64_MAX;
5079 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
5080 spa_config_set(spa, config);
5082 nvlist_free(config);
5084 if (state == SPA_LOAD_RECOVER) {
5085 ASSERT3P(loadinfo, ==, NULL);
5086 spa_import_progress_remove(spa_guid(spa));
5087 return (rewind_error);
5089 /* Store the rewind info as part of the initial load info */
5090 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
5091 spa->spa_load_info);
5093 /* Restore the initial load info */
5094 fnvlist_free(spa->spa_load_info);
5095 spa->spa_load_info = loadinfo;
5097 spa_import_progress_remove(spa_guid(spa));
5098 return (load_error);
5105 * The import case is identical to an open except that the configuration is sent
5106 * down from userland, instead of grabbed from the configuration cache. For the
5107 * case of an open, the pool configuration will exist in the
5108 * POOL_STATE_UNINITIALIZED state.
5110 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5111 * the same time open the pool, without having to keep around the spa_t in some
5115 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
5119 spa_load_state_t state = SPA_LOAD_OPEN;
5121 int locked = B_FALSE;
5122 int firstopen = B_FALSE;
5127 * As disgusting as this is, we need to support recursive calls to this
5128 * function because dsl_dir_open() is called during spa_load(), and ends
5129 * up calling spa_open() again. The real fix is to figure out how to
5130 * avoid dsl_dir_open() calling this in the first place.
5132 if (MUTEX_NOT_HELD(&spa_namespace_lock)) {
5133 mutex_enter(&spa_namespace_lock);
5137 if ((spa = spa_lookup(pool)) == NULL) {
5139 mutex_exit(&spa_namespace_lock);
5140 return (SET_ERROR(ENOENT));
5143 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
5144 zpool_load_policy_t policy;
5148 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
5150 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5151 state = SPA_LOAD_RECOVER;
5153 spa_activate(spa, spa_mode_global);
5155 if (state != SPA_LOAD_RECOVER)
5156 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5157 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5159 zfs_dbgmsg("spa_open_common: opening %s", pool);
5160 error = spa_load_best(spa, state, policy.zlp_txg,
5163 if (error == EBADF) {
5165 * If vdev_validate() returns failure (indicated by
5166 * EBADF), it indicates that one of the vdevs indicates
5167 * that the pool has been exported or destroyed. If
5168 * this is the case, the config cache is out of sync and
5169 * we should remove the pool from the namespace.
5172 spa_deactivate(spa);
5173 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5176 mutex_exit(&spa_namespace_lock);
5177 return (SET_ERROR(ENOENT));
5182 * We can't open the pool, but we still have useful
5183 * information: the state of each vdev after the
5184 * attempted vdev_open(). Return this to the user.
5186 if (config != NULL && spa->spa_config) {
5187 *config = fnvlist_dup(spa->spa_config);
5188 fnvlist_add_nvlist(*config,
5189 ZPOOL_CONFIG_LOAD_INFO,
5190 spa->spa_load_info);
5193 spa_deactivate(spa);
5194 spa->spa_last_open_failed = error;
5196 mutex_exit(&spa_namespace_lock);
5202 spa_open_ref(spa, tag);
5205 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5208 * If we've recovered the pool, pass back any information we
5209 * gathered while doing the load.
5211 if (state == SPA_LOAD_RECOVER) {
5212 fnvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
5213 spa->spa_load_info);
5217 spa->spa_last_open_failed = 0;
5218 spa->spa_last_ubsync_txg = 0;
5219 spa->spa_load_txg = 0;
5220 mutex_exit(&spa_namespace_lock);
5224 zvol_create_minors_recursive(spa_name(spa));
5232 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
5235 return (spa_open_common(name, spapp, tag, policy, config));
5239 spa_open(const char *name, spa_t **spapp, void *tag)
5241 return (spa_open_common(name, spapp, tag, NULL, NULL));
5245 * Lookup the given spa_t, incrementing the inject count in the process,
5246 * preventing it from being exported or destroyed.
5249 spa_inject_addref(char *name)
5253 mutex_enter(&spa_namespace_lock);
5254 if ((spa = spa_lookup(name)) == NULL) {
5255 mutex_exit(&spa_namespace_lock);
5258 spa->spa_inject_ref++;
5259 mutex_exit(&spa_namespace_lock);
5265 spa_inject_delref(spa_t *spa)
5267 mutex_enter(&spa_namespace_lock);
5268 spa->spa_inject_ref--;
5269 mutex_exit(&spa_namespace_lock);
5273 * Add spares device information to the nvlist.
5276 spa_add_spares(spa_t *spa, nvlist_t *config)
5286 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5288 if (spa->spa_spares.sav_count == 0)
5291 nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE);
5292 VERIFY0(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5293 ZPOOL_CONFIG_SPARES, &spares, &nspares));
5295 fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, spares,
5297 VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5298 &spares, &nspares));
5301 * Go through and find any spares which have since been
5302 * repurposed as an active spare. If this is the case, update
5303 * their status appropriately.
5305 for (i = 0; i < nspares; i++) {
5306 guid = fnvlist_lookup_uint64(spares[i],
5308 if (spa_spare_exists(guid, &pool, NULL) &&
5310 VERIFY0(nvlist_lookup_uint64_array(spares[i],
5311 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs,
5313 vs->vs_state = VDEV_STATE_CANT_OPEN;
5314 vs->vs_aux = VDEV_AUX_SPARED;
5321 * Add l2cache device information to the nvlist, including vdev stats.
5324 spa_add_l2cache(spa_t *spa, nvlist_t *config)
5327 uint_t i, j, nl2cache;
5334 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5336 if (spa->spa_l2cache.sav_count == 0)
5339 nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE);
5340 VERIFY0(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5341 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache));
5342 if (nl2cache != 0) {
5343 fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, l2cache,
5345 VERIFY0(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5346 &l2cache, &nl2cache));
5349 * Update level 2 cache device stats.
5352 for (i = 0; i < nl2cache; i++) {
5353 guid = fnvlist_lookup_uint64(l2cache[i],
5357 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
5359 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
5360 vd = spa->spa_l2cache.sav_vdevs[j];
5366 VERIFY0(nvlist_lookup_uint64_array(l2cache[i],
5367 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc));
5368 vdev_get_stats(vd, vs);
5369 vdev_config_generate_stats(vd, l2cache[i]);
5376 spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features)
5381 if (spa->spa_feat_for_read_obj != 0) {
5382 for (zap_cursor_init(&zc, spa->spa_meta_objset,
5383 spa->spa_feat_for_read_obj);
5384 zap_cursor_retrieve(&zc, &za) == 0;
5385 zap_cursor_advance(&zc)) {
5386 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
5387 za.za_num_integers == 1);
5388 VERIFY0(nvlist_add_uint64(features, za.za_name,
5389 za.za_first_integer));
5391 zap_cursor_fini(&zc);
5394 if (spa->spa_feat_for_write_obj != 0) {
5395 for (zap_cursor_init(&zc, spa->spa_meta_objset,
5396 spa->spa_feat_for_write_obj);
5397 zap_cursor_retrieve(&zc, &za) == 0;
5398 zap_cursor_advance(&zc)) {
5399 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
5400 za.za_num_integers == 1);
5401 VERIFY0(nvlist_add_uint64(features, za.za_name,
5402 za.za_first_integer));
5404 zap_cursor_fini(&zc);
5409 spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features)
5413 for (i = 0; i < SPA_FEATURES; i++) {
5414 zfeature_info_t feature = spa_feature_table[i];
5417 if (feature_get_refcount(spa, &feature, &refcount) != 0)
5420 VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount));
5425 * Store a list of pool features and their reference counts in the
5428 * The first time this is called on a spa, allocate a new nvlist, fetch
5429 * the pool features and reference counts from disk, then save the list
5430 * in the spa. In subsequent calls on the same spa use the saved nvlist
5431 * and refresh its values from the cached reference counts. This
5432 * ensures we don't block here on I/O on a suspended pool so 'zpool
5433 * clear' can resume the pool.
5436 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
5440 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
5442 mutex_enter(&spa->spa_feat_stats_lock);
5443 features = spa->spa_feat_stats;
5445 if (features != NULL) {
5446 spa_feature_stats_from_cache(spa, features);
5448 VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP));
5449 spa->spa_feat_stats = features;
5450 spa_feature_stats_from_disk(spa, features);
5453 VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
5456 mutex_exit(&spa->spa_feat_stats_lock);
5460 spa_get_stats(const char *name, nvlist_t **config,
5461 char *altroot, size_t buflen)
5467 error = spa_open_common(name, &spa, FTAG, NULL, config);
5471 * This still leaves a window of inconsistency where the spares
5472 * or l2cache devices could change and the config would be
5473 * self-inconsistent.
5475 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5477 if (*config != NULL) {
5478 uint64_t loadtimes[2];
5480 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
5481 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
5482 fnvlist_add_uint64_array(*config,
5483 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2);
5485 fnvlist_add_uint64(*config,
5486 ZPOOL_CONFIG_ERRCOUNT,
5487 spa_get_errlog_size(spa));
5489 if (spa_suspended(spa)) {
5490 fnvlist_add_uint64(*config,
5491 ZPOOL_CONFIG_SUSPENDED,
5493 fnvlist_add_uint64(*config,
5494 ZPOOL_CONFIG_SUSPENDED_REASON,
5495 spa->spa_suspended);
5498 spa_add_spares(spa, *config);
5499 spa_add_l2cache(spa, *config);
5500 spa_add_feature_stats(spa, *config);
5505 * We want to get the alternate root even for faulted pools, so we cheat
5506 * and call spa_lookup() directly.
5510 mutex_enter(&spa_namespace_lock);
5511 spa = spa_lookup(name);
5513 spa_altroot(spa, altroot, buflen);
5517 mutex_exit(&spa_namespace_lock);
5519 spa_altroot(spa, altroot, buflen);
5524 spa_config_exit(spa, SCL_CONFIG, FTAG);
5525 spa_close(spa, FTAG);
5532 * Validate that the auxiliary device array is well formed. We must have an
5533 * array of nvlists, each which describes a valid leaf vdev. If this is an
5534 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
5535 * specified, as long as they are well-formed.
5538 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
5539 spa_aux_vdev_t *sav, const char *config, uint64_t version,
5540 vdev_labeltype_t label)
5547 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5550 * It's acceptable to have no devs specified.
5552 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
5556 return (SET_ERROR(EINVAL));
5559 * Make sure the pool is formatted with a version that supports this
5562 if (spa_version(spa) < version)
5563 return (SET_ERROR(ENOTSUP));
5566 * Set the pending device list so we correctly handle device in-use
5569 sav->sav_pending = dev;
5570 sav->sav_npending = ndev;
5572 for (i = 0; i < ndev; i++) {
5573 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
5577 if (!vd->vdev_ops->vdev_op_leaf) {
5579 error = SET_ERROR(EINVAL);
5585 if ((error = vdev_open(vd)) == 0 &&
5586 (error = vdev_label_init(vd, crtxg, label)) == 0) {
5587 fnvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
5594 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
5601 sav->sav_pending = NULL;
5602 sav->sav_npending = 0;
5607 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
5611 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5613 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
5614 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
5615 VDEV_LABEL_SPARE)) != 0) {
5619 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
5620 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
5621 VDEV_LABEL_L2CACHE));
5625 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
5630 if (sav->sav_config != NULL) {
5636 * Generate new dev list by concatenating with the
5639 VERIFY0(nvlist_lookup_nvlist_array(sav->sav_config, config,
5640 &olddevs, &oldndevs));
5642 newdevs = kmem_alloc(sizeof (void *) *
5643 (ndevs + oldndevs), KM_SLEEP);
5644 for (i = 0; i < oldndevs; i++)
5645 newdevs[i] = fnvlist_dup(olddevs[i]);
5646 for (i = 0; i < ndevs; i++)
5647 newdevs[i + oldndevs] = fnvlist_dup(devs[i]);
5649 fnvlist_remove(sav->sav_config, config);
5651 fnvlist_add_nvlist_array(sav->sav_config, config, newdevs,
5653 for (i = 0; i < oldndevs + ndevs; i++)
5654 nvlist_free(newdevs[i]);
5655 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
5658 * Generate a new dev list.
5660 sav->sav_config = fnvlist_alloc();
5661 fnvlist_add_nvlist_array(sav->sav_config, config, devs, ndevs);
5666 * Stop and drop level 2 ARC devices
5669 spa_l2cache_drop(spa_t *spa)
5673 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5675 for (i = 0; i < sav->sav_count; i++) {
5678 vd = sav->sav_vdevs[i];
5681 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
5682 pool != 0ULL && l2arc_vdev_present(vd))
5683 l2arc_remove_vdev(vd);
5688 * Verify encryption parameters for spa creation. If we are encrypting, we must
5689 * have the encryption feature flag enabled.
5692 spa_create_check_encryption_params(dsl_crypto_params_t *dcp,
5693 boolean_t has_encryption)
5695 if (dcp->cp_crypt != ZIO_CRYPT_OFF &&
5696 dcp->cp_crypt != ZIO_CRYPT_INHERIT &&
5698 return (SET_ERROR(ENOTSUP));
5700 return (dmu_objset_create_crypt_check(NULL, dcp, NULL));
5707 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
5708 nvlist_t *zplprops, dsl_crypto_params_t *dcp)
5711 char *altroot = NULL;
5716 uint64_t txg = TXG_INITIAL;
5717 nvlist_t **spares, **l2cache;
5718 uint_t nspares, nl2cache;
5719 uint64_t version, obj, ndraid = 0;
5720 boolean_t has_features;
5721 boolean_t has_encryption;
5722 boolean_t has_allocclass;
5728 if (props == NULL ||
5729 nvlist_lookup_string(props, "tname", &poolname) != 0)
5730 poolname = (char *)pool;
5733 * If this pool already exists, return failure.
5735 mutex_enter(&spa_namespace_lock);
5736 if (spa_lookup(poolname) != NULL) {
5737 mutex_exit(&spa_namespace_lock);
5738 return (SET_ERROR(EEXIST));
5742 * Allocate a new spa_t structure.
5744 nvl = fnvlist_alloc();
5745 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
5746 (void) nvlist_lookup_string(props,
5747 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5748 spa = spa_add(poolname, nvl, altroot);
5750 spa_activate(spa, spa_mode_global);
5752 if (props && (error = spa_prop_validate(spa, props))) {
5753 spa_deactivate(spa);
5755 mutex_exit(&spa_namespace_lock);
5760 * Temporary pool names should never be written to disk.
5762 if (poolname != pool)
5763 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
5765 has_features = B_FALSE;
5766 has_encryption = B_FALSE;
5767 has_allocclass = B_FALSE;
5768 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
5769 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
5770 if (zpool_prop_feature(nvpair_name(elem))) {
5771 has_features = B_TRUE;
5773 feat_name = strchr(nvpair_name(elem), '@') + 1;
5774 VERIFY0(zfeature_lookup_name(feat_name, &feat));
5775 if (feat == SPA_FEATURE_ENCRYPTION)
5776 has_encryption = B_TRUE;
5777 if (feat == SPA_FEATURE_ALLOCATION_CLASSES)
5778 has_allocclass = B_TRUE;
5782 /* verify encryption params, if they were provided */
5784 error = spa_create_check_encryption_params(dcp, has_encryption);
5786 spa_deactivate(spa);
5788 mutex_exit(&spa_namespace_lock);
5792 if (!has_allocclass && zfs_special_devs(nvroot, NULL)) {
5793 spa_deactivate(spa);
5795 mutex_exit(&spa_namespace_lock);
5799 if (has_features || nvlist_lookup_uint64(props,
5800 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
5801 version = SPA_VERSION;
5803 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5805 spa->spa_first_txg = txg;
5806 spa->spa_uberblock.ub_txg = txg - 1;
5807 spa->spa_uberblock.ub_version = version;
5808 spa->spa_ubsync = spa->spa_uberblock;
5809 spa->spa_load_state = SPA_LOAD_CREATE;
5810 spa->spa_removing_phys.sr_state = DSS_NONE;
5811 spa->spa_removing_phys.sr_removing_vdev = -1;
5812 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
5813 spa->spa_indirect_vdevs_loaded = B_TRUE;
5816 * Create "The Godfather" zio to hold all async IOs
5818 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
5820 for (int i = 0; i < max_ncpus; i++) {
5821 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
5822 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
5823 ZIO_FLAG_GODFATHER);
5827 * Create the root vdev.
5829 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5831 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
5833 ASSERT(error != 0 || rvd != NULL);
5834 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
5836 if (error == 0 && !zfs_allocatable_devs(nvroot))
5837 error = SET_ERROR(EINVAL);
5840 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
5841 (error = vdev_draid_spare_create(nvroot, rvd, &ndraid, 0)) == 0 &&
5842 (error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) == 0) {
5844 * instantiate the metaslab groups (this will dirty the vdevs)
5845 * we can no longer error exit past this point
5847 for (int c = 0; error == 0 && c < rvd->vdev_children; c++) {
5848 vdev_t *vd = rvd->vdev_child[c];
5850 vdev_metaslab_set_size(vd);
5851 vdev_expand(vd, txg);
5855 spa_config_exit(spa, SCL_ALL, FTAG);
5859 spa_deactivate(spa);
5861 mutex_exit(&spa_namespace_lock);
5866 * Get the list of spares, if specified.
5868 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5869 &spares, &nspares) == 0) {
5870 spa->spa_spares.sav_config = fnvlist_alloc();
5871 fnvlist_add_nvlist_array(spa->spa_spares.sav_config,
5872 ZPOOL_CONFIG_SPARES, spares, nspares);
5873 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5874 spa_load_spares(spa);
5875 spa_config_exit(spa, SCL_ALL, FTAG);
5876 spa->spa_spares.sav_sync = B_TRUE;
5880 * Get the list of level 2 cache devices, if specified.
5882 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5883 &l2cache, &nl2cache) == 0) {
5884 spa->spa_l2cache.sav_config = fnvlist_alloc();
5885 fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5886 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache);
5887 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5888 spa_load_l2cache(spa);
5889 spa_config_exit(spa, SCL_ALL, FTAG);
5890 spa->spa_l2cache.sav_sync = B_TRUE;
5893 spa->spa_is_initializing = B_TRUE;
5894 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg);
5895 spa->spa_is_initializing = B_FALSE;
5898 * Create DDTs (dedup tables).
5902 spa_update_dspace(spa);
5904 tx = dmu_tx_create_assigned(dp, txg);
5907 * Create the pool's history object.
5909 if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history)
5910 spa_history_create_obj(spa, tx);
5912 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
5913 spa_history_log_version(spa, "create", tx);
5916 * Create the pool config object.
5918 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
5919 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
5920 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
5922 if (zap_add(spa->spa_meta_objset,
5923 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
5924 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
5925 cmn_err(CE_PANIC, "failed to add pool config");
5928 if (zap_add(spa->spa_meta_objset,
5929 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
5930 sizeof (uint64_t), 1, &version, tx) != 0) {
5931 cmn_err(CE_PANIC, "failed to add pool version");
5934 /* Newly created pools with the right version are always deflated. */
5935 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
5936 spa->spa_deflate = TRUE;
5937 if (zap_add(spa->spa_meta_objset,
5938 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5939 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
5940 cmn_err(CE_PANIC, "failed to add deflate");
5945 * Create the deferred-free bpobj. Turn off compression
5946 * because sync-to-convergence takes longer if the blocksize
5949 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
5950 dmu_object_set_compress(spa->spa_meta_objset, obj,
5951 ZIO_COMPRESS_OFF, tx);
5952 if (zap_add(spa->spa_meta_objset,
5953 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
5954 sizeof (uint64_t), 1, &obj, tx) != 0) {
5955 cmn_err(CE_PANIC, "failed to add bpobj");
5957 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
5958 spa->spa_meta_objset, obj));
5961 * Generate some random noise for salted checksums to operate on.
5963 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
5964 sizeof (spa->spa_cksum_salt.zcs_bytes));
5967 * Set pool properties.
5969 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
5970 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
5971 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
5972 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
5973 spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST);
5974 spa->spa_autotrim = zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM);
5976 if (props != NULL) {
5977 spa_configfile_set(spa, props, B_FALSE);
5978 spa_sync_props(props, tx);
5981 for (int i = 0; i < ndraid; i++)
5982 spa_feature_incr(spa, SPA_FEATURE_DRAID, tx);
5986 spa->spa_sync_on = B_TRUE;
5988 mmp_thread_start(spa);
5989 txg_wait_synced(dp, txg);
5991 spa_spawn_aux_threads(spa);
5993 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5996 * Don't count references from objsets that are already closed
5997 * and are making their way through the eviction process.
5999 spa_evicting_os_wait(spa);
6000 spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
6001 spa->spa_load_state = SPA_LOAD_NONE;
6003 mutex_exit(&spa_namespace_lock);
6009 * Import a non-root pool into the system.
6012 spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
6015 char *altroot = NULL;
6016 spa_load_state_t state = SPA_LOAD_IMPORT;
6017 zpool_load_policy_t policy;
6018 spa_mode_t mode = spa_mode_global;
6019 uint64_t readonly = B_FALSE;
6022 nvlist_t **spares, **l2cache;
6023 uint_t nspares, nl2cache;
6026 * If a pool with this name exists, return failure.
6028 mutex_enter(&spa_namespace_lock);
6029 if (spa_lookup(pool) != NULL) {
6030 mutex_exit(&spa_namespace_lock);
6031 return (SET_ERROR(EEXIST));
6035 * Create and initialize the spa structure.
6037 (void) nvlist_lookup_string(props,
6038 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
6039 (void) nvlist_lookup_uint64(props,
6040 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
6042 mode = SPA_MODE_READ;
6043 spa = spa_add(pool, config, altroot);
6044 spa->spa_import_flags = flags;
6047 * Verbatim import - Take a pool and insert it into the namespace
6048 * as if it had been loaded at boot.
6050 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
6052 spa_configfile_set(spa, props, B_FALSE);
6054 spa_write_cachefile(spa, B_FALSE, B_TRUE);
6055 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
6056 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
6057 mutex_exit(&spa_namespace_lock);
6061 spa_activate(spa, mode);
6064 * Don't start async tasks until we know everything is healthy.
6066 spa_async_suspend(spa);
6068 zpool_get_load_policy(config, &policy);
6069 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
6070 state = SPA_LOAD_RECOVER;
6072 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
6074 if (state != SPA_LOAD_RECOVER) {
6075 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
6076 zfs_dbgmsg("spa_import: importing %s", pool);
6078 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
6079 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
6081 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
6084 * Propagate anything learned while loading the pool and pass it
6085 * back to caller (i.e. rewind info, missing devices, etc).
6087 fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info);
6089 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6091 * Toss any existing sparelist, as it doesn't have any validity
6092 * anymore, and conflicts with spa_has_spare().
6094 if (spa->spa_spares.sav_config) {
6095 nvlist_free(spa->spa_spares.sav_config);
6096 spa->spa_spares.sav_config = NULL;
6097 spa_load_spares(spa);
6099 if (spa->spa_l2cache.sav_config) {
6100 nvlist_free(spa->spa_l2cache.sav_config);
6101 spa->spa_l2cache.sav_config = NULL;
6102 spa_load_l2cache(spa);
6105 nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE);
6106 spa_config_exit(spa, SCL_ALL, FTAG);
6109 spa_configfile_set(spa, props, B_FALSE);
6111 if (error != 0 || (props && spa_writeable(spa) &&
6112 (error = spa_prop_set(spa, props)))) {
6114 spa_deactivate(spa);
6116 mutex_exit(&spa_namespace_lock);
6120 spa_async_resume(spa);
6123 * Override any spares and level 2 cache devices as specified by
6124 * the user, as these may have correct device names/devids, etc.
6126 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
6127 &spares, &nspares) == 0) {
6128 if (spa->spa_spares.sav_config)
6129 fnvlist_remove(spa->spa_spares.sav_config,
6130 ZPOOL_CONFIG_SPARES);
6132 spa->spa_spares.sav_config = fnvlist_alloc();
6133 fnvlist_add_nvlist_array(spa->spa_spares.sav_config,
6134 ZPOOL_CONFIG_SPARES, spares, nspares);
6135 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6136 spa_load_spares(spa);
6137 spa_config_exit(spa, SCL_ALL, FTAG);
6138 spa->spa_spares.sav_sync = B_TRUE;
6140 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
6141 &l2cache, &nl2cache) == 0) {
6142 if (spa->spa_l2cache.sav_config)
6143 fnvlist_remove(spa->spa_l2cache.sav_config,
6144 ZPOOL_CONFIG_L2CACHE);
6146 spa->spa_l2cache.sav_config = fnvlist_alloc();
6147 fnvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
6148 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache);
6149 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6150 spa_load_l2cache(spa);
6151 spa_config_exit(spa, SCL_ALL, FTAG);
6152 spa->spa_l2cache.sav_sync = B_TRUE;
6156 * Check for any removed devices.
6158 if (spa->spa_autoreplace) {
6159 spa_aux_check_removed(&spa->spa_spares);
6160 spa_aux_check_removed(&spa->spa_l2cache);
6163 if (spa_writeable(spa)) {
6165 * Update the config cache to include the newly-imported pool.
6167 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6171 * It's possible that the pool was expanded while it was exported.
6172 * We kick off an async task to handle this for us.
6174 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
6176 spa_history_log_version(spa, "import", NULL);
6178 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
6180 mutex_exit(&spa_namespace_lock);
6182 zvol_create_minors_recursive(pool);
6188 spa_tryimport(nvlist_t *tryconfig)
6190 nvlist_t *config = NULL;
6191 char *poolname, *cachefile;
6195 zpool_load_policy_t policy;
6197 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
6200 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
6204 * Create and initialize the spa structure.
6206 mutex_enter(&spa_namespace_lock);
6207 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
6208 spa_activate(spa, SPA_MODE_READ);
6211 * Rewind pool if a max txg was provided.
6213 zpool_get_load_policy(spa->spa_config, &policy);
6214 if (policy.zlp_txg != UINT64_MAX) {
6215 spa->spa_load_max_txg = policy.zlp_txg;
6216 spa->spa_extreme_rewind = B_TRUE;
6217 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6218 poolname, (longlong_t)policy.zlp_txg);
6220 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
6223 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
6225 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
6226 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
6228 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
6231 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
6234 * If 'tryconfig' was at least parsable, return the current config.
6236 if (spa->spa_root_vdev != NULL) {
6237 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
6238 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, poolname);
6239 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, state);
6240 fnvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
6241 spa->spa_uberblock.ub_timestamp);
6242 fnvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
6243 spa->spa_load_info);
6244 fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
6248 * If the bootfs property exists on this pool then we
6249 * copy it out so that external consumers can tell which
6250 * pools are bootable.
6252 if ((!error || error == EEXIST) && spa->spa_bootfs) {
6253 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6256 * We have to play games with the name since the
6257 * pool was opened as TRYIMPORT_NAME.
6259 if (dsl_dsobj_to_dsname(spa_name(spa),
6260 spa->spa_bootfs, tmpname) == 0) {
6264 dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6266 cp = strchr(tmpname, '/');
6268 (void) strlcpy(dsname, tmpname,
6271 (void) snprintf(dsname, MAXPATHLEN,
6272 "%s/%s", poolname, ++cp);
6274 fnvlist_add_string(config, ZPOOL_CONFIG_BOOTFS,
6276 kmem_free(dsname, MAXPATHLEN);
6278 kmem_free(tmpname, MAXPATHLEN);
6282 * Add the list of hot spares and level 2 cache devices.
6284 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6285 spa_add_spares(spa, config);
6286 spa_add_l2cache(spa, config);
6287 spa_config_exit(spa, SCL_CONFIG, FTAG);
6291 spa_deactivate(spa);
6293 mutex_exit(&spa_namespace_lock);
6299 * Pool export/destroy
6301 * The act of destroying or exporting a pool is very simple. We make sure there
6302 * is no more pending I/O and any references to the pool are gone. Then, we
6303 * update the pool state and sync all the labels to disk, removing the
6304 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6305 * we don't sync the labels or remove the configuration cache.
6308 spa_export_common(const char *pool, int new_state, nvlist_t **oldconfig,
6309 boolean_t force, boolean_t hardforce)
6317 if (!(spa_mode_global & SPA_MODE_WRITE))
6318 return (SET_ERROR(EROFS));
6320 mutex_enter(&spa_namespace_lock);
6321 if ((spa = spa_lookup(pool)) == NULL) {
6322 mutex_exit(&spa_namespace_lock);
6323 return (SET_ERROR(ENOENT));
6326 if (spa->spa_is_exporting) {
6327 /* the pool is being exported by another thread */
6328 mutex_exit(&spa_namespace_lock);
6329 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS));
6331 spa->spa_is_exporting = B_TRUE;
6334 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6335 * reacquire the namespace lock, and see if we can export.
6337 spa_open_ref(spa, FTAG);
6338 mutex_exit(&spa_namespace_lock);
6339 spa_async_suspend(spa);
6340 if (spa->spa_zvol_taskq) {
6341 zvol_remove_minors(spa, spa_name(spa), B_TRUE);
6342 taskq_wait(spa->spa_zvol_taskq);
6344 mutex_enter(&spa_namespace_lock);
6345 spa_close(spa, FTAG);
6347 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
6350 * The pool will be in core if it's openable, in which case we can
6351 * modify its state. Objsets may be open only because they're dirty,
6352 * so we have to force it to sync before checking spa_refcnt.
6354 if (spa->spa_sync_on) {
6355 txg_wait_synced(spa->spa_dsl_pool, 0);
6356 spa_evicting_os_wait(spa);
6360 * A pool cannot be exported or destroyed if there are active
6361 * references. If we are resetting a pool, allow references by
6362 * fault injection handlers.
6364 if (!spa_refcount_zero(spa) || (spa->spa_inject_ref != 0)) {
6365 error = SET_ERROR(EBUSY);
6369 if (spa->spa_sync_on) {
6371 * A pool cannot be exported if it has an active shared spare.
6372 * This is to prevent other pools stealing the active spare
6373 * from an exported pool. At user's own will, such pool can
6374 * be forcedly exported.
6376 if (!force && new_state == POOL_STATE_EXPORTED &&
6377 spa_has_active_shared_spare(spa)) {
6378 error = SET_ERROR(EXDEV);
6383 * We're about to export or destroy this pool. Make sure
6384 * we stop all initialization and trim activity here before
6385 * we set the spa_final_txg. This will ensure that all
6386 * dirty data resulting from the initialization is
6387 * committed to disk before we unload the pool.
6389 if (spa->spa_root_vdev != NULL) {
6390 vdev_t *rvd = spa->spa_root_vdev;
6391 vdev_initialize_stop_all(rvd, VDEV_INITIALIZE_ACTIVE);
6392 vdev_trim_stop_all(rvd, VDEV_TRIM_ACTIVE);
6393 vdev_autotrim_stop_all(spa);
6394 vdev_rebuild_stop_all(spa);
6398 * We want this to be reflected on every label,
6399 * so mark them all dirty. spa_unload() will do the
6400 * final sync that pushes these changes out.
6402 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
6403 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6404 spa->spa_state = new_state;
6405 spa->spa_final_txg = spa_last_synced_txg(spa) +
6407 vdev_config_dirty(spa->spa_root_vdev);
6408 spa_config_exit(spa, SCL_ALL, FTAG);
6413 if (new_state == POOL_STATE_DESTROYED)
6414 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
6415 else if (new_state == POOL_STATE_EXPORTED)
6416 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT);
6418 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6420 spa_deactivate(spa);
6423 if (oldconfig && spa->spa_config)
6424 *oldconfig = fnvlist_dup(spa->spa_config);
6426 if (new_state != POOL_STATE_UNINITIALIZED) {
6428 spa_write_cachefile(spa, B_TRUE, B_TRUE);
6432 * If spa_remove() is not called for this spa_t and
6433 * there is any possibility that it can be reused,
6434 * we make sure to reset the exporting flag.
6436 spa->spa_is_exporting = B_FALSE;
6439 mutex_exit(&spa_namespace_lock);
6443 spa->spa_is_exporting = B_FALSE;
6444 spa_async_resume(spa);
6445 mutex_exit(&spa_namespace_lock);
6450 * Destroy a storage pool.
6453 spa_destroy(const char *pool)
6455 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
6460 * Export a storage pool.
6463 spa_export(const char *pool, nvlist_t **oldconfig, boolean_t force,
6464 boolean_t hardforce)
6466 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
6471 * Similar to spa_export(), this unloads the spa_t without actually removing it
6472 * from the namespace in any way.
6475 spa_reset(const char *pool)
6477 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
6482 * ==========================================================================
6483 * Device manipulation
6484 * ==========================================================================
6488 * This is called as a synctask to increment the draid feature flag
6491 spa_draid_feature_incr(void *arg, dmu_tx_t *tx)
6493 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6494 int draid = (int)(uintptr_t)arg;
6496 for (int c = 0; c < draid; c++)
6497 spa_feature_incr(spa, SPA_FEATURE_DRAID, tx);
6501 * Add a device to a storage pool.
6504 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
6506 uint64_t txg, ndraid = 0;
6508 vdev_t *rvd = spa->spa_root_vdev;
6510 nvlist_t **spares, **l2cache;
6511 uint_t nspares, nl2cache;
6513 ASSERT(spa_writeable(spa));
6515 txg = spa_vdev_enter(spa);
6517 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
6518 VDEV_ALLOC_ADD)) != 0)
6519 return (spa_vdev_exit(spa, NULL, txg, error));
6521 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
6523 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
6527 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
6531 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
6532 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6534 if (vd->vdev_children != 0 &&
6535 (error = vdev_create(vd, txg, B_FALSE)) != 0) {
6536 return (spa_vdev_exit(spa, vd, txg, error));
6540 * The virtual dRAID spares must be added after vdev tree is created
6541 * and the vdev guids are generated. The guid of their associated
6542 * dRAID is stored in the config and used when opening the spare.
6544 if ((error = vdev_draid_spare_create(nvroot, vd, &ndraid,
6545 rvd->vdev_children)) == 0) {
6546 if (ndraid > 0 && nvlist_lookup_nvlist_array(nvroot,
6547 ZPOOL_CONFIG_SPARES, &spares, &nspares) != 0)
6550 return (spa_vdev_exit(spa, vd, txg, error));
6554 * We must validate the spares and l2cache devices after checking the
6555 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
6557 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
6558 return (spa_vdev_exit(spa, vd, txg, error));
6561 * If we are in the middle of a device removal, we can only add
6562 * devices which match the existing devices in the pool.
6563 * If we are in the middle of a removal, or have some indirect
6564 * vdevs, we can not add raidz or dRAID top levels.
6566 if (spa->spa_vdev_removal != NULL ||
6567 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
6568 for (int c = 0; c < vd->vdev_children; c++) {
6569 tvd = vd->vdev_child[c];
6570 if (spa->spa_vdev_removal != NULL &&
6571 tvd->vdev_ashift != spa->spa_max_ashift) {
6572 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6574 /* Fail if top level vdev is raidz or a dRAID */
6575 if (vdev_get_nparity(tvd) != 0)
6576 return (spa_vdev_exit(spa, vd, txg, EINVAL));
6579 * Need the top level mirror to be
6580 * a mirror of leaf vdevs only
6582 if (tvd->vdev_ops == &vdev_mirror_ops) {
6583 for (uint64_t cid = 0;
6584 cid < tvd->vdev_children; cid++) {
6585 vdev_t *cvd = tvd->vdev_child[cid];
6586 if (!cvd->vdev_ops->vdev_op_leaf) {
6587 return (spa_vdev_exit(spa, vd,
6595 for (int c = 0; c < vd->vdev_children; c++) {
6596 tvd = vd->vdev_child[c];
6597 vdev_remove_child(vd, tvd);
6598 tvd->vdev_id = rvd->vdev_children;
6599 vdev_add_child(rvd, tvd);
6600 vdev_config_dirty(tvd);
6604 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
6605 ZPOOL_CONFIG_SPARES);
6606 spa_load_spares(spa);
6607 spa->spa_spares.sav_sync = B_TRUE;
6610 if (nl2cache != 0) {
6611 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
6612 ZPOOL_CONFIG_L2CACHE);
6613 spa_load_l2cache(spa);
6614 spa->spa_l2cache.sav_sync = B_TRUE;
6618 * We can't increment a feature while holding spa_vdev so we
6619 * have to do it in a synctask.
6624 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
6625 dsl_sync_task_nowait(spa->spa_dsl_pool, spa_draid_feature_incr,
6626 (void *)(uintptr_t)ndraid, tx);
6631 * We have to be careful when adding new vdevs to an existing pool.
6632 * If other threads start allocating from these vdevs before we
6633 * sync the config cache, and we lose power, then upon reboot we may
6634 * fail to open the pool because there are DVAs that the config cache
6635 * can't translate. Therefore, we first add the vdevs without
6636 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
6637 * and then let spa_config_update() initialize the new metaslabs.
6639 * spa_load() checks for added-but-not-initialized vdevs, so that
6640 * if we lose power at any point in this sequence, the remaining
6641 * steps will be completed the next time we load the pool.
6643 (void) spa_vdev_exit(spa, vd, txg, 0);
6645 mutex_enter(&spa_namespace_lock);
6646 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
6647 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
6648 mutex_exit(&spa_namespace_lock);
6654 * Attach a device to a mirror. The arguments are the path to any device
6655 * in the mirror, and the nvroot for the new device. If the path specifies
6656 * a device that is not mirrored, we automatically insert the mirror vdev.
6658 * If 'replacing' is specified, the new device is intended to replace the
6659 * existing device; in this case the two devices are made into their own
6660 * mirror using the 'replacing' vdev, which is functionally identical to
6661 * the mirror vdev (it actually reuses all the same ops) but has a few
6662 * extra rules: you can't attach to it after it's been created, and upon
6663 * completion of resilvering, the first disk (the one being replaced)
6664 * is automatically detached.
6666 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
6667 * should be performed instead of traditional healing reconstruction. From
6668 * an administrators perspective these are both resilver operations.
6671 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing,
6674 uint64_t txg, dtl_max_txg;
6675 vdev_t *rvd = spa->spa_root_vdev;
6676 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
6678 char *oldvdpath, *newvdpath;
6682 ASSERT(spa_writeable(spa));
6684 txg = spa_vdev_enter(spa);
6686 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
6688 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6689 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6690 error = (spa_has_checkpoint(spa)) ?
6691 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6692 return (spa_vdev_exit(spa, NULL, txg, error));
6696 if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REBUILD))
6697 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6699 if (dsl_scan_resilvering(spa_get_dsl(spa)))
6700 return (spa_vdev_exit(spa, NULL, txg,
6701 ZFS_ERR_RESILVER_IN_PROGRESS));
6703 if (vdev_rebuild_active(rvd))
6704 return (spa_vdev_exit(spa, NULL, txg,
6705 ZFS_ERR_REBUILD_IN_PROGRESS));
6708 if (spa->spa_vdev_removal != NULL)
6709 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6712 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6714 if (!oldvd->vdev_ops->vdev_op_leaf)
6715 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6717 pvd = oldvd->vdev_parent;
6719 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
6720 VDEV_ALLOC_ATTACH)) != 0)
6721 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6723 if (newrootvd->vdev_children != 1)
6724 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6726 newvd = newrootvd->vdev_child[0];
6728 if (!newvd->vdev_ops->vdev_op_leaf)
6729 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6731 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
6732 return (spa_vdev_exit(spa, newrootvd, txg, error));
6735 * Spares can't replace logs
6737 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
6738 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6741 * A dRAID spare can only replace a child of its parent dRAID vdev.
6743 if (newvd->vdev_ops == &vdev_draid_spare_ops &&
6744 oldvd->vdev_top != vdev_draid_spare_get_parent(newvd)) {
6745 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6750 * For rebuilds, the top vdev must support reconstruction
6751 * using only space maps. This means the only allowable
6752 * vdevs types are the root vdev, a mirror, or dRAID.
6755 if (pvd->vdev_top != NULL)
6756 tvd = pvd->vdev_top;
6758 if (tvd->vdev_ops != &vdev_mirror_ops &&
6759 tvd->vdev_ops != &vdev_root_ops &&
6760 tvd->vdev_ops != &vdev_draid_ops) {
6761 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6767 * For attach, the only allowable parent is a mirror or the root
6770 if (pvd->vdev_ops != &vdev_mirror_ops &&
6771 pvd->vdev_ops != &vdev_root_ops)
6772 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6774 pvops = &vdev_mirror_ops;
6777 * Active hot spares can only be replaced by inactive hot
6780 if (pvd->vdev_ops == &vdev_spare_ops &&
6781 oldvd->vdev_isspare &&
6782 !spa_has_spare(spa, newvd->vdev_guid))
6783 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6786 * If the source is a hot spare, and the parent isn't already a
6787 * spare, then we want to create a new hot spare. Otherwise, we
6788 * want to create a replacing vdev. The user is not allowed to
6789 * attach to a spared vdev child unless the 'isspare' state is
6790 * the same (spare replaces spare, non-spare replaces
6793 if (pvd->vdev_ops == &vdev_replacing_ops &&
6794 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6795 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6796 } else if (pvd->vdev_ops == &vdev_spare_ops &&
6797 newvd->vdev_isspare != oldvd->vdev_isspare) {
6798 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6801 if (newvd->vdev_isspare)
6802 pvops = &vdev_spare_ops;
6804 pvops = &vdev_replacing_ops;
6808 * Make sure the new device is big enough.
6810 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6811 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6814 * The new device cannot have a higher alignment requirement
6815 * than the top-level vdev.
6817 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6818 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6821 * If this is an in-place replacement, update oldvd's path and devid
6822 * to make it distinguishable from newvd, and unopenable from now on.
6824 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6825 spa_strfree(oldvd->vdev_path);
6826 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6828 (void) snprintf(oldvd->vdev_path, strlen(newvd->vdev_path) + 5,
6829 "%s/%s", newvd->vdev_path, "old");
6830 if (oldvd->vdev_devid != NULL) {
6831 spa_strfree(oldvd->vdev_devid);
6832 oldvd->vdev_devid = NULL;
6837 * If the parent is not a mirror, or if we're replacing, insert the new
6838 * mirror/replacing/spare vdev above oldvd.
6840 if (pvd->vdev_ops != pvops)
6841 pvd = vdev_add_parent(oldvd, pvops);
6843 ASSERT(pvd->vdev_top->vdev_parent == rvd);
6844 ASSERT(pvd->vdev_ops == pvops);
6845 ASSERT(oldvd->vdev_parent == pvd);
6848 * Extract the new device from its root and add it to pvd.
6850 vdev_remove_child(newrootvd, newvd);
6851 newvd->vdev_id = pvd->vdev_children;
6852 newvd->vdev_crtxg = oldvd->vdev_crtxg;
6853 vdev_add_child(pvd, newvd);
6856 * Reevaluate the parent vdev state.
6858 vdev_propagate_state(pvd);
6860 tvd = newvd->vdev_top;
6861 ASSERT(pvd->vdev_top == tvd);
6862 ASSERT(tvd->vdev_parent == rvd);
6864 vdev_config_dirty(tvd);
6867 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6868 * for any dmu_sync-ed blocks. It will propagate upward when
6869 * spa_vdev_exit() calls vdev_dtl_reassess().
6871 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6873 vdev_dtl_dirty(newvd, DTL_MISSING,
6874 TXG_INITIAL, dtl_max_txg - TXG_INITIAL);
6876 if (newvd->vdev_isspare) {
6877 spa_spare_activate(newvd);
6878 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6881 oldvdpath = spa_strdup(oldvd->vdev_path);
6882 newvdpath = spa_strdup(newvd->vdev_path);
6883 newvd_isspare = newvd->vdev_isspare;
6886 * Mark newvd's DTL dirty in this txg.
6888 vdev_dirty(tvd, VDD_DTL, newvd, txg);
6891 * Schedule the resilver or rebuild to restart in the future. We do
6892 * this to ensure that dmu_sync-ed blocks have been stitched into the
6893 * respective datasets.
6896 newvd->vdev_rebuild_txg = txg;
6900 newvd->vdev_resilver_txg = txg;
6902 if (dsl_scan_resilvering(spa_get_dsl(spa)) &&
6903 spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER)) {
6904 vdev_defer_resilver(newvd);
6906 dsl_scan_restart_resilver(spa->spa_dsl_pool,
6911 if (spa->spa_bootfs)
6912 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6914 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6919 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6921 spa_history_log_internal(spa, "vdev attach", NULL,
6922 "%s vdev=%s %s vdev=%s",
6923 replacing && newvd_isspare ? "spare in" :
6924 replacing ? "replace" : "attach", newvdpath,
6925 replacing ? "for" : "to", oldvdpath);
6927 spa_strfree(oldvdpath);
6928 spa_strfree(newvdpath);
6934 * Detach a device from a mirror or replacing vdev.
6936 * If 'replace_done' is specified, only detach if the parent
6937 * is a replacing vdev.
6940 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6944 vdev_t *rvd __maybe_unused = spa->spa_root_vdev;
6945 vdev_t *vd, *pvd, *cvd, *tvd;
6946 boolean_t unspare = B_FALSE;
6947 uint64_t unspare_guid = 0;
6950 ASSERT(spa_writeable(spa));
6952 txg = spa_vdev_detach_enter(spa, guid);
6954 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6957 * Besides being called directly from the userland through the
6958 * ioctl interface, spa_vdev_detach() can be potentially called
6959 * at the end of spa_vdev_resilver_done().
6961 * In the regular case, when we have a checkpoint this shouldn't
6962 * happen as we never empty the DTLs of a vdev during the scrub
6963 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6964 * should never get here when we have a checkpoint.
6966 * That said, even in a case when we checkpoint the pool exactly
6967 * as spa_vdev_resilver_done() calls this function everything
6968 * should be fine as the resilver will return right away.
6970 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6971 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6972 error = (spa_has_checkpoint(spa)) ?
6973 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6974 return (spa_vdev_exit(spa, NULL, txg, error));
6978 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6980 if (!vd->vdev_ops->vdev_op_leaf)
6981 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6983 pvd = vd->vdev_parent;
6986 * If the parent/child relationship is not as expected, don't do it.
6987 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6988 * vdev that's replacing B with C. The user's intent in replacing
6989 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6990 * the replace by detaching C, the expected behavior is to end up
6991 * M(A,B). But suppose that right after deciding to detach C,
6992 * the replacement of B completes. We would have M(A,C), and then
6993 * ask to detach C, which would leave us with just A -- not what
6994 * the user wanted. To prevent this, we make sure that the
6995 * parent/child relationship hasn't changed -- in this example,
6996 * that C's parent is still the replacing vdev R.
6998 if (pvd->vdev_guid != pguid && pguid != 0)
6999 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
7002 * Only 'replacing' or 'spare' vdevs can be replaced.
7004 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
7005 pvd->vdev_ops != &vdev_spare_ops)
7006 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
7008 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
7009 spa_version(spa) >= SPA_VERSION_SPARES);
7012 * Only mirror, replacing, and spare vdevs support detach.
7014 if (pvd->vdev_ops != &vdev_replacing_ops &&
7015 pvd->vdev_ops != &vdev_mirror_ops &&
7016 pvd->vdev_ops != &vdev_spare_ops)
7017 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
7020 * If this device has the only valid copy of some data,
7021 * we cannot safely detach it.
7023 if (vdev_dtl_required(vd))
7024 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
7026 ASSERT(pvd->vdev_children >= 2);
7029 * If we are detaching the second disk from a replacing vdev, then
7030 * check to see if we changed the original vdev's path to have "/old"
7031 * at the end in spa_vdev_attach(). If so, undo that change now.
7033 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
7034 vd->vdev_path != NULL) {
7035 size_t len = strlen(vd->vdev_path);
7037 for (int c = 0; c < pvd->vdev_children; c++) {
7038 cvd = pvd->vdev_child[c];
7040 if (cvd == vd || cvd->vdev_path == NULL)
7043 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
7044 strcmp(cvd->vdev_path + len, "/old") == 0) {
7045 spa_strfree(cvd->vdev_path);
7046 cvd->vdev_path = spa_strdup(vd->vdev_path);
7053 * If we are detaching the original disk from a normal spare, then it
7054 * implies that the spare should become a real disk, and be removed
7055 * from the active spare list for the pool. dRAID spares on the
7056 * other hand are coupled to the pool and thus should never be removed
7057 * from the spares list.
7059 if (pvd->vdev_ops == &vdev_spare_ops && vd->vdev_id == 0) {
7060 vdev_t *last_cvd = pvd->vdev_child[pvd->vdev_children - 1];
7062 if (last_cvd->vdev_isspare &&
7063 last_cvd->vdev_ops != &vdev_draid_spare_ops) {
7069 * Erase the disk labels so the disk can be used for other things.
7070 * This must be done after all other error cases are handled,
7071 * but before we disembowel vd (so we can still do I/O to it).
7072 * But if we can't do it, don't treat the error as fatal --
7073 * it may be that the unwritability of the disk is the reason
7074 * it's being detached!
7076 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
7079 * Remove vd from its parent and compact the parent's children.
7081 vdev_remove_child(pvd, vd);
7082 vdev_compact_children(pvd);
7085 * Remember one of the remaining children so we can get tvd below.
7087 cvd = pvd->vdev_child[pvd->vdev_children - 1];
7090 * If we need to remove the remaining child from the list of hot spares,
7091 * do it now, marking the vdev as no longer a spare in the process.
7092 * We must do this before vdev_remove_parent(), because that can
7093 * change the GUID if it creates a new toplevel GUID. For a similar
7094 * reason, we must remove the spare now, in the same txg as the detach;
7095 * otherwise someone could attach a new sibling, change the GUID, and
7096 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
7099 ASSERT(cvd->vdev_isspare);
7100 spa_spare_remove(cvd);
7101 unspare_guid = cvd->vdev_guid;
7102 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
7103 cvd->vdev_unspare = B_TRUE;
7107 * If the parent mirror/replacing vdev only has one child,
7108 * the parent is no longer needed. Remove it from the tree.
7110 if (pvd->vdev_children == 1) {
7111 if (pvd->vdev_ops == &vdev_spare_ops)
7112 cvd->vdev_unspare = B_FALSE;
7113 vdev_remove_parent(cvd);
7117 * We don't set tvd until now because the parent we just removed
7118 * may have been the previous top-level vdev.
7120 tvd = cvd->vdev_top;
7121 ASSERT(tvd->vdev_parent == rvd);
7124 * Reevaluate the parent vdev state.
7126 vdev_propagate_state(cvd);
7129 * If the 'autoexpand' property is set on the pool then automatically
7130 * try to expand the size of the pool. For example if the device we
7131 * just detached was smaller than the others, it may be possible to
7132 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
7133 * first so that we can obtain the updated sizes of the leaf vdevs.
7135 if (spa->spa_autoexpand) {
7137 vdev_expand(tvd, txg);
7140 vdev_config_dirty(tvd);
7143 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
7144 * vd->vdev_detached is set and free vd's DTL object in syncing context.
7145 * But first make sure we're not on any *other* txg's DTL list, to
7146 * prevent vd from being accessed after it's freed.
7148 vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none");
7149 for (int t = 0; t < TXG_SIZE; t++)
7150 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
7151 vd->vdev_detached = B_TRUE;
7152 vdev_dirty(tvd, VDD_DTL, vd, txg);
7154 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
7155 spa_notify_waiters(spa);
7157 /* hang on to the spa before we release the lock */
7158 spa_open_ref(spa, FTAG);
7160 error = spa_vdev_exit(spa, vd, txg, 0);
7162 spa_history_log_internal(spa, "detach", NULL,
7164 spa_strfree(vdpath);
7167 * If this was the removal of the original device in a hot spare vdev,
7168 * then we want to go through and remove the device from the hot spare
7169 * list of every other pool.
7172 spa_t *altspa = NULL;
7174 mutex_enter(&spa_namespace_lock);
7175 while ((altspa = spa_next(altspa)) != NULL) {
7176 if (altspa->spa_state != POOL_STATE_ACTIVE ||
7180 spa_open_ref(altspa, FTAG);
7181 mutex_exit(&spa_namespace_lock);
7182 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
7183 mutex_enter(&spa_namespace_lock);
7184 spa_close(altspa, FTAG);
7186 mutex_exit(&spa_namespace_lock);
7188 /* search the rest of the vdevs for spares to remove */
7189 spa_vdev_resilver_done(spa);
7192 /* all done with the spa; OK to release */
7193 mutex_enter(&spa_namespace_lock);
7194 spa_close(spa, FTAG);
7195 mutex_exit(&spa_namespace_lock);
7201 spa_vdev_initialize_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type,
7204 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7206 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7208 /* Look up vdev and ensure it's a leaf. */
7209 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
7210 if (vd == NULL || vd->vdev_detached) {
7211 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7212 return (SET_ERROR(ENODEV));
7213 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
7214 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7215 return (SET_ERROR(EINVAL));
7216 } else if (!vdev_writeable(vd)) {
7217 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7218 return (SET_ERROR(EROFS));
7220 mutex_enter(&vd->vdev_initialize_lock);
7221 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7224 * When we activate an initialize action we check to see
7225 * if the vdev_initialize_thread is NULL. We do this instead
7226 * of using the vdev_initialize_state since there might be
7227 * a previous initialization process which has completed but
7228 * the thread is not exited.
7230 if (cmd_type == POOL_INITIALIZE_START &&
7231 (vd->vdev_initialize_thread != NULL ||
7232 vd->vdev_top->vdev_removing)) {
7233 mutex_exit(&vd->vdev_initialize_lock);
7234 return (SET_ERROR(EBUSY));
7235 } else if (cmd_type == POOL_INITIALIZE_CANCEL &&
7236 (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE &&
7237 vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) {
7238 mutex_exit(&vd->vdev_initialize_lock);
7239 return (SET_ERROR(ESRCH));
7240 } else if (cmd_type == POOL_INITIALIZE_SUSPEND &&
7241 vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) {
7242 mutex_exit(&vd->vdev_initialize_lock);
7243 return (SET_ERROR(ESRCH));
7247 case POOL_INITIALIZE_START:
7248 vdev_initialize(vd);
7250 case POOL_INITIALIZE_CANCEL:
7251 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED, vd_list);
7253 case POOL_INITIALIZE_SUSPEND:
7254 vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED, vd_list);
7257 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
7259 mutex_exit(&vd->vdev_initialize_lock);
7265 spa_vdev_initialize(spa_t *spa, nvlist_t *nv, uint64_t cmd_type,
7266 nvlist_t *vdev_errlist)
7268 int total_errors = 0;
7271 list_create(&vd_list, sizeof (vdev_t),
7272 offsetof(vdev_t, vdev_initialize_node));
7275 * We hold the namespace lock through the whole function
7276 * to prevent any changes to the pool while we're starting or
7277 * stopping initialization. The config and state locks are held so that
7278 * we can properly assess the vdev state before we commit to
7279 * the initializing operation.
7281 mutex_enter(&spa_namespace_lock);
7283 for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
7284 pair != NULL; pair = nvlist_next_nvpair(nv, pair)) {
7285 uint64_t vdev_guid = fnvpair_value_uint64(pair);
7287 int error = spa_vdev_initialize_impl(spa, vdev_guid, cmd_type,
7290 char guid_as_str[MAXNAMELEN];
7292 (void) snprintf(guid_as_str, sizeof (guid_as_str),
7293 "%llu", (unsigned long long)vdev_guid);
7294 fnvlist_add_int64(vdev_errlist, guid_as_str, error);
7299 /* Wait for all initialize threads to stop. */
7300 vdev_initialize_stop_wait(spa, &vd_list);
7302 /* Sync out the initializing state */
7303 txg_wait_synced(spa->spa_dsl_pool, 0);
7304 mutex_exit(&spa_namespace_lock);
7306 list_destroy(&vd_list);
7308 return (total_errors);
7312 spa_vdev_trim_impl(spa_t *spa, uint64_t guid, uint64_t cmd_type,
7313 uint64_t rate, boolean_t partial, boolean_t secure, list_t *vd_list)
7315 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7317 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7319 /* Look up vdev and ensure it's a leaf. */
7320 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
7321 if (vd == NULL || vd->vdev_detached) {
7322 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7323 return (SET_ERROR(ENODEV));
7324 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
7325 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7326 return (SET_ERROR(EINVAL));
7327 } else if (!vdev_writeable(vd)) {
7328 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7329 return (SET_ERROR(EROFS));
7330 } else if (!vd->vdev_has_trim) {
7331 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7332 return (SET_ERROR(EOPNOTSUPP));
7333 } else if (secure && !vd->vdev_has_securetrim) {
7334 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7335 return (SET_ERROR(EOPNOTSUPP));
7337 mutex_enter(&vd->vdev_trim_lock);
7338 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7341 * When we activate a TRIM action we check to see if the
7342 * vdev_trim_thread is NULL. We do this instead of using the
7343 * vdev_trim_state since there might be a previous TRIM process
7344 * which has completed but the thread is not exited.
7346 if (cmd_type == POOL_TRIM_START &&
7347 (vd->vdev_trim_thread != NULL || vd->vdev_top->vdev_removing)) {
7348 mutex_exit(&vd->vdev_trim_lock);
7349 return (SET_ERROR(EBUSY));
7350 } else if (cmd_type == POOL_TRIM_CANCEL &&
7351 (vd->vdev_trim_state != VDEV_TRIM_ACTIVE &&
7352 vd->vdev_trim_state != VDEV_TRIM_SUSPENDED)) {
7353 mutex_exit(&vd->vdev_trim_lock);
7354 return (SET_ERROR(ESRCH));
7355 } else if (cmd_type == POOL_TRIM_SUSPEND &&
7356 vd->vdev_trim_state != VDEV_TRIM_ACTIVE) {
7357 mutex_exit(&vd->vdev_trim_lock);
7358 return (SET_ERROR(ESRCH));
7362 case POOL_TRIM_START:
7363 vdev_trim(vd, rate, partial, secure);
7365 case POOL_TRIM_CANCEL:
7366 vdev_trim_stop(vd, VDEV_TRIM_CANCELED, vd_list);
7368 case POOL_TRIM_SUSPEND:
7369 vdev_trim_stop(vd, VDEV_TRIM_SUSPENDED, vd_list);
7372 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
7374 mutex_exit(&vd->vdev_trim_lock);
7380 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
7381 * TRIM threads for each child vdev. These threads pass over all of the free
7382 * space in the vdev's metaslabs and issues TRIM commands for that space.
7385 spa_vdev_trim(spa_t *spa, nvlist_t *nv, uint64_t cmd_type, uint64_t rate,
7386 boolean_t partial, boolean_t secure, nvlist_t *vdev_errlist)
7388 int total_errors = 0;
7391 list_create(&vd_list, sizeof (vdev_t),
7392 offsetof(vdev_t, vdev_trim_node));
7395 * We hold the namespace lock through the whole function
7396 * to prevent any changes to the pool while we're starting or
7397 * stopping TRIM. The config and state locks are held so that
7398 * we can properly assess the vdev state before we commit to
7399 * the TRIM operation.
7401 mutex_enter(&spa_namespace_lock);
7403 for (nvpair_t *pair = nvlist_next_nvpair(nv, NULL);
7404 pair != NULL; pair = nvlist_next_nvpair(nv, pair)) {
7405 uint64_t vdev_guid = fnvpair_value_uint64(pair);
7407 int error = spa_vdev_trim_impl(spa, vdev_guid, cmd_type,
7408 rate, partial, secure, &vd_list);
7410 char guid_as_str[MAXNAMELEN];
7412 (void) snprintf(guid_as_str, sizeof (guid_as_str),
7413 "%llu", (unsigned long long)vdev_guid);
7414 fnvlist_add_int64(vdev_errlist, guid_as_str, error);
7419 /* Wait for all TRIM threads to stop. */
7420 vdev_trim_stop_wait(spa, &vd_list);
7422 /* Sync out the TRIM state */
7423 txg_wait_synced(spa->spa_dsl_pool, 0);
7424 mutex_exit(&spa_namespace_lock);
7426 list_destroy(&vd_list);
7428 return (total_errors);
7432 * Split a set of devices from their mirrors, and create a new pool from them.
7435 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
7436 nvlist_t *props, boolean_t exp)
7439 uint64_t txg, *glist;
7441 uint_t c, children, lastlog;
7442 nvlist_t **child, *nvl, *tmp;
7444 char *altroot = NULL;
7445 vdev_t *rvd, **vml = NULL; /* vdev modify list */
7446 boolean_t activate_slog;
7448 ASSERT(spa_writeable(spa));
7450 txg = spa_vdev_enter(spa);
7452 ASSERT(MUTEX_HELD(&spa_namespace_lock));
7453 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
7454 error = (spa_has_checkpoint(spa)) ?
7455 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
7456 return (spa_vdev_exit(spa, NULL, txg, error));
7459 /* clear the log and flush everything up to now */
7460 activate_slog = spa_passivate_log(spa);
7461 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
7462 error = spa_reset_logs(spa);
7463 txg = spa_vdev_config_enter(spa);
7466 spa_activate_log(spa);
7469 return (spa_vdev_exit(spa, NULL, txg, error));
7471 /* check new spa name before going any further */
7472 if (spa_lookup(newname) != NULL)
7473 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
7476 * scan through all the children to ensure they're all mirrors
7478 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
7479 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
7481 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7483 /* first, check to ensure we've got the right child count */
7484 rvd = spa->spa_root_vdev;
7486 for (c = 0; c < rvd->vdev_children; c++) {
7487 vdev_t *vd = rvd->vdev_child[c];
7489 /* don't count the holes & logs as children */
7490 if (vd->vdev_islog || (vd->vdev_ops != &vdev_indirect_ops &&
7491 !vdev_is_concrete(vd))) {
7499 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
7500 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7502 /* next, ensure no spare or cache devices are part of the split */
7503 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
7504 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
7505 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
7507 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
7508 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
7510 /* then, loop over each vdev and validate it */
7511 for (c = 0; c < children; c++) {
7512 uint64_t is_hole = 0;
7514 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
7518 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
7519 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
7522 error = SET_ERROR(EINVAL);
7527 /* deal with indirect vdevs */
7528 if (spa->spa_root_vdev->vdev_child[c]->vdev_ops ==
7532 /* which disk is going to be split? */
7533 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
7535 error = SET_ERROR(EINVAL);
7539 /* look it up in the spa */
7540 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
7541 if (vml[c] == NULL) {
7542 error = SET_ERROR(ENODEV);
7546 /* make sure there's nothing stopping the split */
7547 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
7548 vml[c]->vdev_islog ||
7549 !vdev_is_concrete(vml[c]) ||
7550 vml[c]->vdev_isspare ||
7551 vml[c]->vdev_isl2cache ||
7552 !vdev_writeable(vml[c]) ||
7553 vml[c]->vdev_children != 0 ||
7554 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
7555 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
7556 error = SET_ERROR(EINVAL);
7560 if (vdev_dtl_required(vml[c]) ||
7561 vdev_resilver_needed(vml[c], NULL, NULL)) {
7562 error = SET_ERROR(EBUSY);
7566 /* we need certain info from the top level */
7567 fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
7568 vml[c]->vdev_top->vdev_ms_array);
7569 fnvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
7570 vml[c]->vdev_top->vdev_ms_shift);
7571 fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
7572 vml[c]->vdev_top->vdev_asize);
7573 fnvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
7574 vml[c]->vdev_top->vdev_ashift);
7576 /* transfer per-vdev ZAPs */
7577 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
7578 VERIFY0(nvlist_add_uint64(child[c],
7579 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
7581 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
7582 VERIFY0(nvlist_add_uint64(child[c],
7583 ZPOOL_CONFIG_VDEV_TOP_ZAP,
7584 vml[c]->vdev_parent->vdev_top_zap));
7588 kmem_free(vml, children * sizeof (vdev_t *));
7589 kmem_free(glist, children * sizeof (uint64_t));
7590 return (spa_vdev_exit(spa, NULL, txg, error));
7593 /* stop writers from using the disks */
7594 for (c = 0; c < children; c++) {
7596 vml[c]->vdev_offline = B_TRUE;
7598 vdev_reopen(spa->spa_root_vdev);
7601 * Temporarily record the splitting vdevs in the spa config. This
7602 * will disappear once the config is regenerated.
7604 nvl = fnvlist_alloc();
7605 fnvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, glist, children);
7606 kmem_free(glist, children * sizeof (uint64_t));
7608 mutex_enter(&spa->spa_props_lock);
7609 fnvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, nvl);
7610 mutex_exit(&spa->spa_props_lock);
7611 spa->spa_config_splitting = nvl;
7612 vdev_config_dirty(spa->spa_root_vdev);
7614 /* configure and create the new pool */
7615 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname);
7616 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
7617 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE);
7618 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
7619 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, spa->spa_config_txg);
7620 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
7621 spa_generate_guid(NULL));
7622 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
7623 (void) nvlist_lookup_string(props,
7624 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
7626 /* add the new pool to the namespace */
7627 newspa = spa_add(newname, config, altroot);
7628 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
7629 newspa->spa_config_txg = spa->spa_config_txg;
7630 spa_set_log_state(newspa, SPA_LOG_CLEAR);
7632 /* release the spa config lock, retaining the namespace lock */
7633 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
7635 if (zio_injection_enabled)
7636 zio_handle_panic_injection(spa, FTAG, 1);
7638 spa_activate(newspa, spa_mode_global);
7639 spa_async_suspend(newspa);
7642 * Temporarily stop the initializing and TRIM activity. We set the
7643 * state to ACTIVE so that we know to resume initializing or TRIM
7644 * once the split has completed.
7646 list_t vd_initialize_list;
7647 list_create(&vd_initialize_list, sizeof (vdev_t),
7648 offsetof(vdev_t, vdev_initialize_node));
7650 list_t vd_trim_list;
7651 list_create(&vd_trim_list, sizeof (vdev_t),
7652 offsetof(vdev_t, vdev_trim_node));
7654 for (c = 0; c < children; c++) {
7655 if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) {
7656 mutex_enter(&vml[c]->vdev_initialize_lock);
7657 vdev_initialize_stop(vml[c],
7658 VDEV_INITIALIZE_ACTIVE, &vd_initialize_list);
7659 mutex_exit(&vml[c]->vdev_initialize_lock);
7661 mutex_enter(&vml[c]->vdev_trim_lock);
7662 vdev_trim_stop(vml[c], VDEV_TRIM_ACTIVE, &vd_trim_list);
7663 mutex_exit(&vml[c]->vdev_trim_lock);
7667 vdev_initialize_stop_wait(spa, &vd_initialize_list);
7668 vdev_trim_stop_wait(spa, &vd_trim_list);
7670 list_destroy(&vd_initialize_list);
7671 list_destroy(&vd_trim_list);
7673 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
7674 newspa->spa_is_splitting = B_TRUE;
7676 /* create the new pool from the disks of the original pool */
7677 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
7681 /* if that worked, generate a real config for the new pool */
7682 if (newspa->spa_root_vdev != NULL) {
7683 newspa->spa_config_splitting = fnvlist_alloc();
7684 fnvlist_add_uint64(newspa->spa_config_splitting,
7685 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa));
7686 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
7691 if (props != NULL) {
7692 spa_configfile_set(newspa, props, B_FALSE);
7693 error = spa_prop_set(newspa, props);
7698 /* flush everything */
7699 txg = spa_vdev_config_enter(newspa);
7700 vdev_config_dirty(newspa->spa_root_vdev);
7701 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
7703 if (zio_injection_enabled)
7704 zio_handle_panic_injection(spa, FTAG, 2);
7706 spa_async_resume(newspa);
7708 /* finally, update the original pool's config */
7709 txg = spa_vdev_config_enter(spa);
7710 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
7711 error = dmu_tx_assign(tx, TXG_WAIT);
7714 for (c = 0; c < children; c++) {
7715 if (vml[c] != NULL && vml[c]->vdev_ops != &vdev_indirect_ops) {
7716 vdev_t *tvd = vml[c]->vdev_top;
7719 * Need to be sure the detachable VDEV is not
7720 * on any *other* txg's DTL list to prevent it
7721 * from being accessed after it's freed.
7723 for (int t = 0; t < TXG_SIZE; t++) {
7724 (void) txg_list_remove_this(
7725 &tvd->vdev_dtl_list, vml[c], t);
7730 spa_history_log_internal(spa, "detach", tx,
7731 "vdev=%s", vml[c]->vdev_path);
7736 spa->spa_avz_action = AVZ_ACTION_REBUILD;
7737 vdev_config_dirty(spa->spa_root_vdev);
7738 spa->spa_config_splitting = NULL;
7742 (void) spa_vdev_exit(spa, NULL, txg, 0);
7744 if (zio_injection_enabled)
7745 zio_handle_panic_injection(spa, FTAG, 3);
7747 /* split is complete; log a history record */
7748 spa_history_log_internal(newspa, "split", NULL,
7749 "from pool %s", spa_name(spa));
7751 newspa->spa_is_splitting = B_FALSE;
7752 kmem_free(vml, children * sizeof (vdev_t *));
7754 /* if we're not going to mount the filesystems in userland, export */
7756 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
7763 spa_deactivate(newspa);
7766 txg = spa_vdev_config_enter(spa);
7768 /* re-online all offlined disks */
7769 for (c = 0; c < children; c++) {
7771 vml[c]->vdev_offline = B_FALSE;
7774 /* restart initializing or trimming disks as necessary */
7775 spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART);
7776 spa_async_request(spa, SPA_ASYNC_TRIM_RESTART);
7777 spa_async_request(spa, SPA_ASYNC_AUTOTRIM_RESTART);
7779 vdev_reopen(spa->spa_root_vdev);
7781 nvlist_free(spa->spa_config_splitting);
7782 spa->spa_config_splitting = NULL;
7783 (void) spa_vdev_exit(spa, NULL, txg, error);
7785 kmem_free(vml, children * sizeof (vdev_t *));
7790 * Find any device that's done replacing, or a vdev marked 'unspare' that's
7791 * currently spared, so we can detach it.
7794 spa_vdev_resilver_done_hunt(vdev_t *vd)
7796 vdev_t *newvd, *oldvd;
7798 for (int c = 0; c < vd->vdev_children; c++) {
7799 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
7805 * Check for a completed replacement. We always consider the first
7806 * vdev in the list to be the oldest vdev, and the last one to be
7807 * the newest (see spa_vdev_attach() for how that works). In
7808 * the case where the newest vdev is faulted, we will not automatically
7809 * remove it after a resilver completes. This is OK as it will require
7810 * user intervention to determine which disk the admin wishes to keep.
7812 if (vd->vdev_ops == &vdev_replacing_ops) {
7813 ASSERT(vd->vdev_children > 1);
7815 newvd = vd->vdev_child[vd->vdev_children - 1];
7816 oldvd = vd->vdev_child[0];
7818 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
7819 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
7820 !vdev_dtl_required(oldvd))
7825 * Check for a completed resilver with the 'unspare' flag set.
7826 * Also potentially update faulted state.
7828 if (vd->vdev_ops == &vdev_spare_ops) {
7829 vdev_t *first = vd->vdev_child[0];
7830 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
7832 if (last->vdev_unspare) {
7835 } else if (first->vdev_unspare) {
7842 if (oldvd != NULL &&
7843 vdev_dtl_empty(newvd, DTL_MISSING) &&
7844 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
7845 !vdev_dtl_required(oldvd))
7848 vdev_propagate_state(vd);
7851 * If there are more than two spares attached to a disk,
7852 * and those spares are not required, then we want to
7853 * attempt to free them up now so that they can be used
7854 * by other pools. Once we're back down to a single
7855 * disk+spare, we stop removing them.
7857 if (vd->vdev_children > 2) {
7858 newvd = vd->vdev_child[1];
7860 if (newvd->vdev_isspare && last->vdev_isspare &&
7861 vdev_dtl_empty(last, DTL_MISSING) &&
7862 vdev_dtl_empty(last, DTL_OUTAGE) &&
7863 !vdev_dtl_required(newvd))
7872 spa_vdev_resilver_done(spa_t *spa)
7874 vdev_t *vd, *pvd, *ppvd;
7875 uint64_t guid, sguid, pguid, ppguid;
7877 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7879 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
7880 pvd = vd->vdev_parent;
7881 ppvd = pvd->vdev_parent;
7882 guid = vd->vdev_guid;
7883 pguid = pvd->vdev_guid;
7884 ppguid = ppvd->vdev_guid;
7887 * If we have just finished replacing a hot spared device, then
7888 * we need to detach the parent's first child (the original hot
7891 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
7892 ppvd->vdev_children == 2) {
7893 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
7894 sguid = ppvd->vdev_child[1]->vdev_guid;
7896 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
7898 spa_config_exit(spa, SCL_ALL, FTAG);
7899 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
7901 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
7903 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
7906 spa_config_exit(spa, SCL_ALL, FTAG);
7909 * If a detach was not performed above replace waiters will not have
7910 * been notified. In which case we must do so now.
7912 spa_notify_waiters(spa);
7916 * Update the stored path or FRU for this vdev.
7919 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
7923 boolean_t sync = B_FALSE;
7925 ASSERT(spa_writeable(spa));
7927 spa_vdev_state_enter(spa, SCL_ALL);
7929 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
7930 return (spa_vdev_state_exit(spa, NULL, ENOENT));
7932 if (!vd->vdev_ops->vdev_op_leaf)
7933 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
7936 if (strcmp(value, vd->vdev_path) != 0) {
7937 spa_strfree(vd->vdev_path);
7938 vd->vdev_path = spa_strdup(value);
7942 if (vd->vdev_fru == NULL) {
7943 vd->vdev_fru = spa_strdup(value);
7945 } else if (strcmp(value, vd->vdev_fru) != 0) {
7946 spa_strfree(vd->vdev_fru);
7947 vd->vdev_fru = spa_strdup(value);
7952 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
7956 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
7958 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
7962 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
7964 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
7968 * ==========================================================================
7970 * ==========================================================================
7973 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
7975 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7977 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7978 return (SET_ERROR(EBUSY));
7980 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
7984 spa_scan_stop(spa_t *spa)
7986 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7987 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7988 return (SET_ERROR(EBUSY));
7989 return (dsl_scan_cancel(spa->spa_dsl_pool));
7993 spa_scan(spa_t *spa, pool_scan_func_t func)
7995 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7997 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
7998 return (SET_ERROR(ENOTSUP));
8000 if (func == POOL_SCAN_RESILVER &&
8001 !spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER))
8002 return (SET_ERROR(ENOTSUP));
8005 * If a resilver was requested, but there is no DTL on a
8006 * writeable leaf device, we have nothing to do.
8008 if (func == POOL_SCAN_RESILVER &&
8009 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
8010 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
8014 return (dsl_scan(spa->spa_dsl_pool, func));
8018 * ==========================================================================
8019 * SPA async task processing
8020 * ==========================================================================
8024 spa_async_remove(spa_t *spa, vdev_t *vd)
8026 if (vd->vdev_remove_wanted) {
8027 vd->vdev_remove_wanted = B_FALSE;
8028 vd->vdev_delayed_close = B_FALSE;
8029 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
8032 * We want to clear the stats, but we don't want to do a full
8033 * vdev_clear() as that will cause us to throw away
8034 * degraded/faulted state as well as attempt to reopen the
8035 * device, all of which is a waste.
8037 vd->vdev_stat.vs_read_errors = 0;
8038 vd->vdev_stat.vs_write_errors = 0;
8039 vd->vdev_stat.vs_checksum_errors = 0;
8041 vdev_state_dirty(vd->vdev_top);
8043 /* Tell userspace that the vdev is gone. */
8044 zfs_post_remove(spa, vd);
8047 for (int c = 0; c < vd->vdev_children; c++)
8048 spa_async_remove(spa, vd->vdev_child[c]);
8052 spa_async_probe(spa_t *spa, vdev_t *vd)
8054 if (vd->vdev_probe_wanted) {
8055 vd->vdev_probe_wanted = B_FALSE;
8056 vdev_reopen(vd); /* vdev_open() does the actual probe */
8059 for (int c = 0; c < vd->vdev_children; c++)
8060 spa_async_probe(spa, vd->vdev_child[c]);
8064 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
8066 if (!spa->spa_autoexpand)
8069 for (int c = 0; c < vd->vdev_children; c++) {
8070 vdev_t *cvd = vd->vdev_child[c];
8071 spa_async_autoexpand(spa, cvd);
8074 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
8077 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND);
8081 spa_async_thread(void *arg)
8083 spa_t *spa = (spa_t *)arg;
8084 dsl_pool_t *dp = spa->spa_dsl_pool;
8087 ASSERT(spa->spa_sync_on);
8089 mutex_enter(&spa->spa_async_lock);
8090 tasks = spa->spa_async_tasks;
8091 spa->spa_async_tasks = 0;
8092 mutex_exit(&spa->spa_async_lock);
8095 * See if the config needs to be updated.
8097 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
8098 uint64_t old_space, new_space;
8100 mutex_enter(&spa_namespace_lock);
8101 old_space = metaslab_class_get_space(spa_normal_class(spa));
8102 old_space += metaslab_class_get_space(spa_special_class(spa));
8103 old_space += metaslab_class_get_space(spa_dedup_class(spa));
8104 old_space += metaslab_class_get_space(
8105 spa_embedded_log_class(spa));
8107 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
8109 new_space = metaslab_class_get_space(spa_normal_class(spa));
8110 new_space += metaslab_class_get_space(spa_special_class(spa));
8111 new_space += metaslab_class_get_space(spa_dedup_class(spa));
8112 new_space += metaslab_class_get_space(
8113 spa_embedded_log_class(spa));
8114 mutex_exit(&spa_namespace_lock);
8117 * If the pool grew as a result of the config update,
8118 * then log an internal history event.
8120 if (new_space != old_space) {
8121 spa_history_log_internal(spa, "vdev online", NULL,
8122 "pool '%s' size: %llu(+%llu)",
8123 spa_name(spa), (u_longlong_t)new_space,
8124 (u_longlong_t)(new_space - old_space));
8129 * See if any devices need to be marked REMOVED.
8131 if (tasks & SPA_ASYNC_REMOVE) {
8132 spa_vdev_state_enter(spa, SCL_NONE);
8133 spa_async_remove(spa, spa->spa_root_vdev);
8134 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
8135 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
8136 for (int i = 0; i < spa->spa_spares.sav_count; i++)
8137 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
8138 (void) spa_vdev_state_exit(spa, NULL, 0);
8141 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
8142 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8143 spa_async_autoexpand(spa, spa->spa_root_vdev);
8144 spa_config_exit(spa, SCL_CONFIG, FTAG);
8148 * See if any devices need to be probed.
8150 if (tasks & SPA_ASYNC_PROBE) {
8151 spa_vdev_state_enter(spa, SCL_NONE);
8152 spa_async_probe(spa, spa->spa_root_vdev);
8153 (void) spa_vdev_state_exit(spa, NULL, 0);
8157 * If any devices are done replacing, detach them.
8159 if (tasks & SPA_ASYNC_RESILVER_DONE ||
8160 tasks & SPA_ASYNC_REBUILD_DONE) {
8161 spa_vdev_resilver_done(spa);
8165 * Kick off a resilver.
8167 if (tasks & SPA_ASYNC_RESILVER &&
8168 !vdev_rebuild_active(spa->spa_root_vdev) &&
8169 (!dsl_scan_resilvering(dp) ||
8170 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER)))
8171 dsl_scan_restart_resilver(dp, 0);
8173 if (tasks & SPA_ASYNC_INITIALIZE_RESTART) {
8174 mutex_enter(&spa_namespace_lock);
8175 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8176 vdev_initialize_restart(spa->spa_root_vdev);
8177 spa_config_exit(spa, SCL_CONFIG, FTAG);
8178 mutex_exit(&spa_namespace_lock);
8181 if (tasks & SPA_ASYNC_TRIM_RESTART) {
8182 mutex_enter(&spa_namespace_lock);
8183 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8184 vdev_trim_restart(spa->spa_root_vdev);
8185 spa_config_exit(spa, SCL_CONFIG, FTAG);
8186 mutex_exit(&spa_namespace_lock);
8189 if (tasks & SPA_ASYNC_AUTOTRIM_RESTART) {
8190 mutex_enter(&spa_namespace_lock);
8191 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8192 vdev_autotrim_restart(spa);
8193 spa_config_exit(spa, SCL_CONFIG, FTAG);
8194 mutex_exit(&spa_namespace_lock);
8198 * Kick off L2 cache whole device TRIM.
8200 if (tasks & SPA_ASYNC_L2CACHE_TRIM) {
8201 mutex_enter(&spa_namespace_lock);
8202 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8203 vdev_trim_l2arc(spa);
8204 spa_config_exit(spa, SCL_CONFIG, FTAG);
8205 mutex_exit(&spa_namespace_lock);
8209 * Kick off L2 cache rebuilding.
8211 if (tasks & SPA_ASYNC_L2CACHE_REBUILD) {
8212 mutex_enter(&spa_namespace_lock);
8213 spa_config_enter(spa, SCL_L2ARC, FTAG, RW_READER);
8214 l2arc_spa_rebuild_start(spa);
8215 spa_config_exit(spa, SCL_L2ARC, FTAG);
8216 mutex_exit(&spa_namespace_lock);
8220 * Let the world know that we're done.
8222 mutex_enter(&spa->spa_async_lock);
8223 spa->spa_async_thread = NULL;
8224 cv_broadcast(&spa->spa_async_cv);
8225 mutex_exit(&spa->spa_async_lock);
8230 spa_async_suspend(spa_t *spa)
8232 mutex_enter(&spa->spa_async_lock);
8233 spa->spa_async_suspended++;
8234 while (spa->spa_async_thread != NULL)
8235 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
8236 mutex_exit(&spa->spa_async_lock);
8238 spa_vdev_remove_suspend(spa);
8240 zthr_t *condense_thread = spa->spa_condense_zthr;
8241 if (condense_thread != NULL)
8242 zthr_cancel(condense_thread);
8244 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
8245 if (discard_thread != NULL)
8246 zthr_cancel(discard_thread);
8248 zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr;
8249 if (ll_delete_thread != NULL)
8250 zthr_cancel(ll_delete_thread);
8252 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
8253 if (ll_condense_thread != NULL)
8254 zthr_cancel(ll_condense_thread);
8258 spa_async_resume(spa_t *spa)
8260 mutex_enter(&spa->spa_async_lock);
8261 ASSERT(spa->spa_async_suspended != 0);
8262 spa->spa_async_suspended--;
8263 mutex_exit(&spa->spa_async_lock);
8264 spa_restart_removal(spa);
8266 zthr_t *condense_thread = spa->spa_condense_zthr;
8267 if (condense_thread != NULL)
8268 zthr_resume(condense_thread);
8270 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
8271 if (discard_thread != NULL)
8272 zthr_resume(discard_thread);
8274 zthr_t *ll_delete_thread = spa->spa_livelist_delete_zthr;
8275 if (ll_delete_thread != NULL)
8276 zthr_resume(ll_delete_thread);
8278 zthr_t *ll_condense_thread = spa->spa_livelist_condense_zthr;
8279 if (ll_condense_thread != NULL)
8280 zthr_resume(ll_condense_thread);
8284 spa_async_tasks_pending(spa_t *spa)
8286 uint_t non_config_tasks;
8288 boolean_t config_task_suspended;
8290 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
8291 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
8292 if (spa->spa_ccw_fail_time == 0) {
8293 config_task_suspended = B_FALSE;
8295 config_task_suspended =
8296 (gethrtime() - spa->spa_ccw_fail_time) <
8297 ((hrtime_t)zfs_ccw_retry_interval * NANOSEC);
8300 return (non_config_tasks || (config_task && !config_task_suspended));
8304 spa_async_dispatch(spa_t *spa)
8306 mutex_enter(&spa->spa_async_lock);
8307 if (spa_async_tasks_pending(spa) &&
8308 !spa->spa_async_suspended &&
8309 spa->spa_async_thread == NULL)
8310 spa->spa_async_thread = thread_create(NULL, 0,
8311 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
8312 mutex_exit(&spa->spa_async_lock);
8316 spa_async_request(spa_t *spa, int task)
8318 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
8319 mutex_enter(&spa->spa_async_lock);
8320 spa->spa_async_tasks |= task;
8321 mutex_exit(&spa->spa_async_lock);
8325 spa_async_tasks(spa_t *spa)
8327 return (spa->spa_async_tasks);
8331 * ==========================================================================
8332 * SPA syncing routines
8333 * ==========================================================================
8338 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
8342 bpobj_enqueue(bpo, bp, bp_freed, tx);
8347 bpobj_enqueue_alloc_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8349 return (bpobj_enqueue_cb(arg, bp, B_FALSE, tx));
8353 bpobj_enqueue_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8355 return (bpobj_enqueue_cb(arg, bp, B_TRUE, tx));
8359 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
8363 zio_nowait(zio_free_sync(pio, pio->io_spa, dmu_tx_get_txg(tx), bp,
8369 bpobj_spa_free_sync_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
8373 return (spa_free_sync_cb(arg, bp, tx));
8377 * Note: this simple function is not inlined to make it easier to dtrace the
8378 * amount of time spent syncing frees.
8381 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
8383 zio_t *zio = zio_root(spa, NULL, NULL, 0);
8384 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
8385 VERIFY(zio_wait(zio) == 0);
8389 * Note: this simple function is not inlined to make it easier to dtrace the
8390 * amount of time spent syncing deferred frees.
8393 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
8395 if (spa_sync_pass(spa) != 1)
8400 * If the log space map feature is active, we stop deferring
8401 * frees to the next TXG and therefore running this function
8402 * would be considered a no-op as spa_deferred_bpobj should
8403 * not have any entries.
8405 * That said we run this function anyway (instead of returning
8406 * immediately) for the edge-case scenario where we just
8407 * activated the log space map feature in this TXG but we have
8408 * deferred frees from the previous TXG.
8410 zio_t *zio = zio_root(spa, NULL, NULL, 0);
8411 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
8412 bpobj_spa_free_sync_cb, zio, tx), ==, 0);
8413 VERIFY0(zio_wait(zio));
8417 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
8419 char *packed = NULL;
8424 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
8427 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
8428 * information. This avoids the dmu_buf_will_dirty() path and
8429 * saves us a pre-read to get data we don't actually care about.
8431 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
8432 packed = vmem_alloc(bufsize, KM_SLEEP);
8434 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
8436 bzero(packed + nvsize, bufsize - nvsize);
8438 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
8440 vmem_free(packed, bufsize);
8442 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
8443 dmu_buf_will_dirty(db, tx);
8444 *(uint64_t *)db->db_data = nvsize;
8445 dmu_buf_rele(db, FTAG);
8449 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
8450 const char *config, const char *entry)
8460 * Update the MOS nvlist describing the list of available devices.
8461 * spa_validate_aux() will have already made sure this nvlist is
8462 * valid and the vdevs are labeled appropriately.
8464 if (sav->sav_object == 0) {
8465 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
8466 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
8467 sizeof (uint64_t), tx);
8468 VERIFY(zap_update(spa->spa_meta_objset,
8469 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
8470 &sav->sav_object, tx) == 0);
8473 nvroot = fnvlist_alloc();
8474 if (sav->sav_count == 0) {
8475 fnvlist_add_nvlist_array(nvroot, config, NULL, 0);
8477 list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP);
8478 for (i = 0; i < sav->sav_count; i++)
8479 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
8480 B_FALSE, VDEV_CONFIG_L2CACHE);
8481 fnvlist_add_nvlist_array(nvroot, config, list, sav->sav_count);
8482 for (i = 0; i < sav->sav_count; i++)
8483 nvlist_free(list[i]);
8484 kmem_free(list, sav->sav_count * sizeof (void *));
8487 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
8488 nvlist_free(nvroot);
8490 sav->sav_sync = B_FALSE;
8494 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
8495 * The all-vdev ZAP must be empty.
8498 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
8500 spa_t *spa = vd->vdev_spa;
8502 if (vd->vdev_top_zap != 0) {
8503 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
8504 vd->vdev_top_zap, tx));
8506 if (vd->vdev_leaf_zap != 0) {
8507 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
8508 vd->vdev_leaf_zap, tx));
8510 for (uint64_t i = 0; i < vd->vdev_children; i++) {
8511 spa_avz_build(vd->vdev_child[i], avz, tx);
8516 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
8521 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
8522 * its config may not be dirty but we still need to build per-vdev ZAPs.
8523 * Similarly, if the pool is being assembled (e.g. after a split), we
8524 * need to rebuild the AVZ although the config may not be dirty.
8526 if (list_is_empty(&spa->spa_config_dirty_list) &&
8527 spa->spa_avz_action == AVZ_ACTION_NONE)
8530 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8532 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
8533 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
8534 spa->spa_all_vdev_zaps != 0);
8536 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
8537 /* Make and build the new AVZ */
8538 uint64_t new_avz = zap_create(spa->spa_meta_objset,
8539 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
8540 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
8542 /* Diff old AVZ with new one */
8546 for (zap_cursor_init(&zc, spa->spa_meta_objset,
8547 spa->spa_all_vdev_zaps);
8548 zap_cursor_retrieve(&zc, &za) == 0;
8549 zap_cursor_advance(&zc)) {
8550 uint64_t vdzap = za.za_first_integer;
8551 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
8554 * ZAP is listed in old AVZ but not in new one;
8557 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
8562 zap_cursor_fini(&zc);
8564 /* Destroy the old AVZ */
8565 VERIFY0(zap_destroy(spa->spa_meta_objset,
8566 spa->spa_all_vdev_zaps, tx));
8568 /* Replace the old AVZ in the dir obj with the new one */
8569 VERIFY0(zap_update(spa->spa_meta_objset,
8570 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
8571 sizeof (new_avz), 1, &new_avz, tx));
8573 spa->spa_all_vdev_zaps = new_avz;
8574 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
8578 /* Walk through the AVZ and destroy all listed ZAPs */
8579 for (zap_cursor_init(&zc, spa->spa_meta_objset,
8580 spa->spa_all_vdev_zaps);
8581 zap_cursor_retrieve(&zc, &za) == 0;
8582 zap_cursor_advance(&zc)) {
8583 uint64_t zap = za.za_first_integer;
8584 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
8587 zap_cursor_fini(&zc);
8589 /* Destroy and unlink the AVZ itself */
8590 VERIFY0(zap_destroy(spa->spa_meta_objset,
8591 spa->spa_all_vdev_zaps, tx));
8592 VERIFY0(zap_remove(spa->spa_meta_objset,
8593 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
8594 spa->spa_all_vdev_zaps = 0;
8597 if (spa->spa_all_vdev_zaps == 0) {
8598 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
8599 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
8600 DMU_POOL_VDEV_ZAP_MAP, tx);
8602 spa->spa_avz_action = AVZ_ACTION_NONE;
8604 /* Create ZAPs for vdevs that don't have them. */
8605 vdev_construct_zaps(spa->spa_root_vdev, tx);
8607 config = spa_config_generate(spa, spa->spa_root_vdev,
8608 dmu_tx_get_txg(tx), B_FALSE);
8611 * If we're upgrading the spa version then make sure that
8612 * the config object gets updated with the correct version.
8614 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
8615 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
8616 spa->spa_uberblock.ub_version);
8618 spa_config_exit(spa, SCL_STATE, FTAG);
8620 nvlist_free(spa->spa_config_syncing);
8621 spa->spa_config_syncing = config;
8623 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
8627 spa_sync_version(void *arg, dmu_tx_t *tx)
8629 uint64_t *versionp = arg;
8630 uint64_t version = *versionp;
8631 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
8634 * Setting the version is special cased when first creating the pool.
8636 ASSERT(tx->tx_txg != TXG_INITIAL);
8638 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
8639 ASSERT(version >= spa_version(spa));
8641 spa->spa_uberblock.ub_version = version;
8642 vdev_config_dirty(spa->spa_root_vdev);
8643 spa_history_log_internal(spa, "set", tx, "version=%lld",
8644 (longlong_t)version);
8648 * Set zpool properties.
8651 spa_sync_props(void *arg, dmu_tx_t *tx)
8653 nvlist_t *nvp = arg;
8654 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
8655 objset_t *mos = spa->spa_meta_objset;
8656 nvpair_t *elem = NULL;
8658 mutex_enter(&spa->spa_props_lock);
8660 while ((elem = nvlist_next_nvpair(nvp, elem))) {
8662 char *strval, *fname;
8664 const char *propname;
8665 zprop_type_t proptype;
8668 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
8669 case ZPOOL_PROP_INVAL:
8671 * We checked this earlier in spa_prop_validate().
8673 ASSERT(zpool_prop_feature(nvpair_name(elem)));
8675 fname = strchr(nvpair_name(elem), '@') + 1;
8676 VERIFY0(zfeature_lookup_name(fname, &fid));
8678 spa_feature_enable(spa, fid, tx);
8679 spa_history_log_internal(spa, "set", tx,
8680 "%s=enabled", nvpair_name(elem));
8683 case ZPOOL_PROP_VERSION:
8684 intval = fnvpair_value_uint64(elem);
8686 * The version is synced separately before other
8687 * properties and should be correct by now.
8689 ASSERT3U(spa_version(spa), >=, intval);
8692 case ZPOOL_PROP_ALTROOT:
8694 * 'altroot' is a non-persistent property. It should
8695 * have been set temporarily at creation or import time.
8697 ASSERT(spa->spa_root != NULL);
8700 case ZPOOL_PROP_READONLY:
8701 case ZPOOL_PROP_CACHEFILE:
8703 * 'readonly' and 'cachefile' are also non-persistent
8707 case ZPOOL_PROP_COMMENT:
8708 strval = fnvpair_value_string(elem);
8709 if (spa->spa_comment != NULL)
8710 spa_strfree(spa->spa_comment);
8711 spa->spa_comment = spa_strdup(strval);
8713 * We need to dirty the configuration on all the vdevs
8714 * so that their labels get updated. We also need to
8715 * update the cache file to keep it in sync with the
8716 * MOS version. It's unnecessary to do this for pool
8717 * creation since the vdev's configuration has already
8720 if (tx->tx_txg != TXG_INITIAL) {
8721 vdev_config_dirty(spa->spa_root_vdev);
8722 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
8724 spa_history_log_internal(spa, "set", tx,
8725 "%s=%s", nvpair_name(elem), strval);
8727 case ZPOOL_PROP_COMPATIBILITY:
8728 strval = fnvpair_value_string(elem);
8729 if (spa->spa_compatibility != NULL)
8730 spa_strfree(spa->spa_compatibility);
8731 spa->spa_compatibility = spa_strdup(strval);
8733 * Dirty the configuration on vdevs as above.
8735 if (tx->tx_txg != TXG_INITIAL) {
8736 vdev_config_dirty(spa->spa_root_vdev);
8737 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
8740 spa_history_log_internal(spa, "set", tx,
8741 "%s=%s", nvpair_name(elem), strval);
8746 * Set pool property values in the poolprops mos object.
8748 if (spa->spa_pool_props_object == 0) {
8749 spa->spa_pool_props_object =
8750 zap_create_link(mos, DMU_OT_POOL_PROPS,
8751 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
8755 /* normalize the property name */
8756 propname = zpool_prop_to_name(prop);
8757 proptype = zpool_prop_get_type(prop);
8759 if (nvpair_type(elem) == DATA_TYPE_STRING) {
8760 ASSERT(proptype == PROP_TYPE_STRING);
8761 strval = fnvpair_value_string(elem);
8762 VERIFY0(zap_update(mos,
8763 spa->spa_pool_props_object, propname,
8764 1, strlen(strval) + 1, strval, tx));
8765 spa_history_log_internal(spa, "set", tx,
8766 "%s=%s", nvpair_name(elem), strval);
8767 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
8768 intval = fnvpair_value_uint64(elem);
8770 if (proptype == PROP_TYPE_INDEX) {
8772 VERIFY0(zpool_prop_index_to_string(
8773 prop, intval, &unused));
8775 VERIFY0(zap_update(mos,
8776 spa->spa_pool_props_object, propname,
8777 8, 1, &intval, tx));
8778 spa_history_log_internal(spa, "set", tx,
8779 "%s=%lld", nvpair_name(elem),
8780 (longlong_t)intval);
8782 ASSERT(0); /* not allowed */
8786 case ZPOOL_PROP_DELEGATION:
8787 spa->spa_delegation = intval;
8789 case ZPOOL_PROP_BOOTFS:
8790 spa->spa_bootfs = intval;
8792 case ZPOOL_PROP_FAILUREMODE:
8793 spa->spa_failmode = intval;
8795 case ZPOOL_PROP_AUTOTRIM:
8796 spa->spa_autotrim = intval;
8797 spa_async_request(spa,
8798 SPA_ASYNC_AUTOTRIM_RESTART);
8800 case ZPOOL_PROP_AUTOEXPAND:
8801 spa->spa_autoexpand = intval;
8802 if (tx->tx_txg != TXG_INITIAL)
8803 spa_async_request(spa,
8804 SPA_ASYNC_AUTOEXPAND);
8806 case ZPOOL_PROP_MULTIHOST:
8807 spa->spa_multihost = intval;
8816 mutex_exit(&spa->spa_props_lock);
8820 * Perform one-time upgrade on-disk changes. spa_version() does not
8821 * reflect the new version this txg, so there must be no changes this
8822 * txg to anything that the upgrade code depends on after it executes.
8823 * Therefore this must be called after dsl_pool_sync() does the sync
8827 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
8829 if (spa_sync_pass(spa) != 1)
8832 dsl_pool_t *dp = spa->spa_dsl_pool;
8833 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
8835 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
8836 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
8837 dsl_pool_create_origin(dp, tx);
8839 /* Keeping the origin open increases spa_minref */
8840 spa->spa_minref += 3;
8843 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
8844 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
8845 dsl_pool_upgrade_clones(dp, tx);
8848 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
8849 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
8850 dsl_pool_upgrade_dir_clones(dp, tx);
8852 /* Keeping the freedir open increases spa_minref */
8853 spa->spa_minref += 3;
8856 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
8857 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
8858 spa_feature_create_zap_objects(spa, tx);
8862 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
8863 * when possibility to use lz4 compression for metadata was added
8864 * Old pools that have this feature enabled must be upgraded to have
8865 * this feature active
8867 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
8868 boolean_t lz4_en = spa_feature_is_enabled(spa,
8869 SPA_FEATURE_LZ4_COMPRESS);
8870 boolean_t lz4_ac = spa_feature_is_active(spa,
8871 SPA_FEATURE_LZ4_COMPRESS);
8873 if (lz4_en && !lz4_ac)
8874 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
8878 * If we haven't written the salt, do so now. Note that the
8879 * feature may not be activated yet, but that's fine since
8880 * the presence of this ZAP entry is backwards compatible.
8882 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8883 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
8884 VERIFY0(zap_add(spa->spa_meta_objset,
8885 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
8886 sizeof (spa->spa_cksum_salt.zcs_bytes),
8887 spa->spa_cksum_salt.zcs_bytes, tx));
8890 rrw_exit(&dp->dp_config_rwlock, FTAG);
8894 vdev_indirect_state_sync_verify(vdev_t *vd)
8896 vdev_indirect_mapping_t *vim __maybe_unused = vd->vdev_indirect_mapping;
8897 vdev_indirect_births_t *vib __maybe_unused = vd->vdev_indirect_births;
8899 if (vd->vdev_ops == &vdev_indirect_ops) {
8900 ASSERT(vim != NULL);
8901 ASSERT(vib != NULL);
8904 uint64_t obsolete_sm_object = 0;
8905 ASSERT0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
8906 if (obsolete_sm_object != 0) {
8907 ASSERT(vd->vdev_obsolete_sm != NULL);
8908 ASSERT(vd->vdev_removing ||
8909 vd->vdev_ops == &vdev_indirect_ops);
8910 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
8911 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
8912 ASSERT3U(obsolete_sm_object, ==,
8913 space_map_object(vd->vdev_obsolete_sm));
8914 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
8915 space_map_allocated(vd->vdev_obsolete_sm));
8917 ASSERT(vd->vdev_obsolete_segments != NULL);
8920 * Since frees / remaps to an indirect vdev can only
8921 * happen in syncing context, the obsolete segments
8922 * tree must be empty when we start syncing.
8924 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
8928 * Set the top-level vdev's max queue depth. Evaluate each top-level's
8929 * async write queue depth in case it changed. The max queue depth will
8930 * not change in the middle of syncing out this txg.
8933 spa_sync_adjust_vdev_max_queue_depth(spa_t *spa)
8935 ASSERT(spa_writeable(spa));
8937 vdev_t *rvd = spa->spa_root_vdev;
8938 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
8939 zfs_vdev_queue_depth_pct / 100;
8940 metaslab_class_t *normal = spa_normal_class(spa);
8941 metaslab_class_t *special = spa_special_class(spa);
8942 metaslab_class_t *dedup = spa_dedup_class(spa);
8944 uint64_t slots_per_allocator = 0;
8945 for (int c = 0; c < rvd->vdev_children; c++) {
8946 vdev_t *tvd = rvd->vdev_child[c];
8948 metaslab_group_t *mg = tvd->vdev_mg;
8949 if (mg == NULL || !metaslab_group_initialized(mg))
8952 metaslab_class_t *mc = mg->mg_class;
8953 if (mc != normal && mc != special && mc != dedup)
8957 * It is safe to do a lock-free check here because only async
8958 * allocations look at mg_max_alloc_queue_depth, and async
8959 * allocations all happen from spa_sync().
8961 for (int i = 0; i < mg->mg_allocators; i++) {
8962 ASSERT0(zfs_refcount_count(
8963 &(mg->mg_allocator[i].mga_alloc_queue_depth)));
8965 mg->mg_max_alloc_queue_depth = max_queue_depth;
8967 for (int i = 0; i < mg->mg_allocators; i++) {
8968 mg->mg_allocator[i].mga_cur_max_alloc_queue_depth =
8969 zfs_vdev_def_queue_depth;
8971 slots_per_allocator += zfs_vdev_def_queue_depth;
8974 for (int i = 0; i < spa->spa_alloc_count; i++) {
8975 ASSERT0(zfs_refcount_count(&normal->mc_allocator[i].
8977 ASSERT0(zfs_refcount_count(&special->mc_allocator[i].
8979 ASSERT0(zfs_refcount_count(&dedup->mc_allocator[i].
8981 normal->mc_allocator[i].mca_alloc_max_slots =
8982 slots_per_allocator;
8983 special->mc_allocator[i].mca_alloc_max_slots =
8984 slots_per_allocator;
8985 dedup->mc_allocator[i].mca_alloc_max_slots =
8986 slots_per_allocator;
8988 normal->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8989 special->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8990 dedup->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8994 spa_sync_condense_indirect(spa_t *spa, dmu_tx_t *tx)
8996 ASSERT(spa_writeable(spa));
8998 vdev_t *rvd = spa->spa_root_vdev;
8999 for (int c = 0; c < rvd->vdev_children; c++) {
9000 vdev_t *vd = rvd->vdev_child[c];
9001 vdev_indirect_state_sync_verify(vd);
9003 if (vdev_indirect_should_condense(vd)) {
9004 spa_condense_indirect_start_sync(vd, tx);
9011 spa_sync_iterate_to_convergence(spa_t *spa, dmu_tx_t *tx)
9013 objset_t *mos = spa->spa_meta_objset;
9014 dsl_pool_t *dp = spa->spa_dsl_pool;
9015 uint64_t txg = tx->tx_txg;
9016 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
9019 int pass = ++spa->spa_sync_pass;
9021 spa_sync_config_object(spa, tx);
9022 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
9023 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
9024 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
9025 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
9026 spa_errlog_sync(spa, txg);
9027 dsl_pool_sync(dp, txg);
9029 if (pass < zfs_sync_pass_deferred_free ||
9030 spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
9032 * If the log space map feature is active we don't
9033 * care about deferred frees and the deferred bpobj
9034 * as the log space map should effectively have the
9035 * same results (i.e. appending only to one object).
9037 spa_sync_frees(spa, free_bpl, tx);
9040 * We can not defer frees in pass 1, because
9041 * we sync the deferred frees later in pass 1.
9043 ASSERT3U(pass, >, 1);
9044 bplist_iterate(free_bpl, bpobj_enqueue_alloc_cb,
9045 &spa->spa_deferred_bpobj, tx);
9049 dsl_scan_sync(dp, tx);
9051 spa_sync_upgrades(spa, tx);
9053 spa_flush_metaslabs(spa, tx);
9056 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
9061 * Note: We need to check if the MOS is dirty because we could
9062 * have marked the MOS dirty without updating the uberblock
9063 * (e.g. if we have sync tasks but no dirty user data). We need
9064 * to check the uberblock's rootbp because it is updated if we
9065 * have synced out dirty data (though in this case the MOS will
9066 * most likely also be dirty due to second order effects, we
9067 * don't want to rely on that here).
9070 spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
9071 !dmu_objset_is_dirty(mos, txg)) {
9073 * Nothing changed on the first pass, therefore this
9074 * TXG is a no-op. Avoid syncing deferred frees, so
9075 * that we can keep this TXG as a no-op.
9077 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
9078 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
9079 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
9080 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, txg));
9084 spa_sync_deferred_frees(spa, tx);
9085 } while (dmu_objset_is_dirty(mos, txg));
9089 * Rewrite the vdev configuration (which includes the uberblock) to
9090 * commit the transaction group.
9092 * If there are no dirty vdevs, we sync the uberblock to a few random
9093 * top-level vdevs that are known to be visible in the config cache
9094 * (see spa_vdev_add() for a complete description). If there *are* dirty
9095 * vdevs, sync the uberblock to all vdevs.
9098 spa_sync_rewrite_vdev_config(spa_t *spa, dmu_tx_t *tx)
9100 vdev_t *rvd = spa->spa_root_vdev;
9101 uint64_t txg = tx->tx_txg;
9107 * We hold SCL_STATE to prevent vdev open/close/etc.
9108 * while we're attempting to write the vdev labels.
9110 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9112 if (list_is_empty(&spa->spa_config_dirty_list)) {
9113 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
9115 int children = rvd->vdev_children;
9116 int c0 = random_in_range(children);
9118 for (int c = 0; c < children; c++) {
9120 rvd->vdev_child[(c0 + c) % children];
9122 /* Stop when revisiting the first vdev */
9123 if (c > 0 && svd[0] == vd)
9126 if (vd->vdev_ms_array == 0 ||
9128 !vdev_is_concrete(vd))
9131 svd[svdcount++] = vd;
9132 if (svdcount == SPA_SYNC_MIN_VDEVS)
9135 error = vdev_config_sync(svd, svdcount, txg);
9137 error = vdev_config_sync(rvd->vdev_child,
9138 rvd->vdev_children, txg);
9142 spa->spa_last_synced_guid = rvd->vdev_guid;
9144 spa_config_exit(spa, SCL_STATE, FTAG);
9148 zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR);
9149 zio_resume_wait(spa);
9154 * Sync the specified transaction group. New blocks may be dirtied as
9155 * part of the process, so we iterate until it converges.
9158 spa_sync(spa_t *spa, uint64_t txg)
9162 VERIFY(spa_writeable(spa));
9165 * Wait for i/os issued in open context that need to complete
9166 * before this txg syncs.
9168 (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]);
9169 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL,
9173 * Lock out configuration changes.
9175 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
9177 spa->spa_syncing_txg = txg;
9178 spa->spa_sync_pass = 0;
9180 for (int i = 0; i < spa->spa_alloc_count; i++) {
9181 mutex_enter(&spa->spa_allocs[i].spaa_lock);
9182 VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree));
9183 mutex_exit(&spa->spa_allocs[i].spaa_lock);
9187 * If there are any pending vdev state changes, convert them
9188 * into config changes that go out with this transaction group.
9190 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9191 while (list_head(&spa->spa_state_dirty_list) != NULL) {
9193 * We need the write lock here because, for aux vdevs,
9194 * calling vdev_config_dirty() modifies sav_config.
9195 * This is ugly and will become unnecessary when we
9196 * eliminate the aux vdev wart by integrating all vdevs
9197 * into the root vdev tree.
9199 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9200 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
9201 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
9202 vdev_state_clean(vd);
9203 vdev_config_dirty(vd);
9205 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9206 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9208 spa_config_exit(spa, SCL_STATE, FTAG);
9210 dsl_pool_t *dp = spa->spa_dsl_pool;
9211 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
9213 spa->spa_sync_starttime = gethrtime();
9214 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
9215 spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
9216 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
9217 NSEC_TO_TICK(spa->spa_deadman_synctime));
9220 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9221 * set spa_deflate if we have no raid-z vdevs.
9223 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
9224 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
9225 vdev_t *rvd = spa->spa_root_vdev;
9228 for (i = 0; i < rvd->vdev_children; i++) {
9229 vd = rvd->vdev_child[i];
9230 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
9233 if (i == rvd->vdev_children) {
9234 spa->spa_deflate = TRUE;
9235 VERIFY0(zap_add(spa->spa_meta_objset,
9236 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
9237 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
9241 spa_sync_adjust_vdev_max_queue_depth(spa);
9243 spa_sync_condense_indirect(spa, tx);
9245 spa_sync_iterate_to_convergence(spa, tx);
9248 if (!list_is_empty(&spa->spa_config_dirty_list)) {
9250 * Make sure that the number of ZAPs for all the vdevs matches
9251 * the number of ZAPs in the per-vdev ZAP list. This only gets
9252 * called if the config is dirty; otherwise there may be
9253 * outstanding AVZ operations that weren't completed in
9254 * spa_sync_config_object.
9256 uint64_t all_vdev_zap_entry_count;
9257 ASSERT0(zap_count(spa->spa_meta_objset,
9258 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
9259 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
9260 all_vdev_zap_entry_count);
9264 if (spa->spa_vdev_removal != NULL) {
9265 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
9268 spa_sync_rewrite_vdev_config(spa, tx);
9271 taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
9272 spa->spa_deadman_tqid = 0;
9275 * Clear the dirty config list.
9277 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
9278 vdev_config_clean(vd);
9281 * Now that the new config has synced transactionally,
9282 * let it become visible to the config cache.
9284 if (spa->spa_config_syncing != NULL) {
9285 spa_config_set(spa, spa->spa_config_syncing);
9286 spa->spa_config_txg = txg;
9287 spa->spa_config_syncing = NULL;
9290 dsl_pool_sync_done(dp, txg);
9292 for (int i = 0; i < spa->spa_alloc_count; i++) {
9293 mutex_enter(&spa->spa_allocs[i].spaa_lock);
9294 VERIFY0(avl_numnodes(&spa->spa_allocs[i].spaa_tree));
9295 mutex_exit(&spa->spa_allocs[i].spaa_lock);
9299 * Update usable space statistics.
9301 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
9303 vdev_sync_done(vd, txg);
9305 metaslab_class_evict_old(spa->spa_normal_class, txg);
9306 metaslab_class_evict_old(spa->spa_log_class, txg);
9308 spa_sync_close_syncing_log_sm(spa);
9310 spa_update_dspace(spa);
9313 * It had better be the case that we didn't dirty anything
9314 * since vdev_config_sync().
9316 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
9317 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
9318 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
9320 while (zfs_pause_spa_sync)
9323 spa->spa_sync_pass = 0;
9326 * Update the last synced uberblock here. We want to do this at
9327 * the end of spa_sync() so that consumers of spa_last_synced_txg()
9328 * will be guaranteed that all the processing associated with
9329 * that txg has been completed.
9331 spa->spa_ubsync = spa->spa_uberblock;
9332 spa_config_exit(spa, SCL_CONFIG, FTAG);
9334 spa_handle_ignored_writes(spa);
9337 * If any async tasks have been requested, kick them off.
9339 spa_async_dispatch(spa);
9343 * Sync all pools. We don't want to hold the namespace lock across these
9344 * operations, so we take a reference on the spa_t and drop the lock during the
9348 spa_sync_allpools(void)
9351 mutex_enter(&spa_namespace_lock);
9352 while ((spa = spa_next(spa)) != NULL) {
9353 if (spa_state(spa) != POOL_STATE_ACTIVE ||
9354 !spa_writeable(spa) || spa_suspended(spa))
9356 spa_open_ref(spa, FTAG);
9357 mutex_exit(&spa_namespace_lock);
9358 txg_wait_synced(spa_get_dsl(spa), 0);
9359 mutex_enter(&spa_namespace_lock);
9360 spa_close(spa, FTAG);
9362 mutex_exit(&spa_namespace_lock);
9366 * ==========================================================================
9367 * Miscellaneous routines
9368 * ==========================================================================
9372 * Remove all pools in the system.
9380 * Remove all cached state. All pools should be closed now,
9381 * so every spa in the AVL tree should be unreferenced.
9383 mutex_enter(&spa_namespace_lock);
9384 while ((spa = spa_next(NULL)) != NULL) {
9386 * Stop async tasks. The async thread may need to detach
9387 * a device that's been replaced, which requires grabbing
9388 * spa_namespace_lock, so we must drop it here.
9390 spa_open_ref(spa, FTAG);
9391 mutex_exit(&spa_namespace_lock);
9392 spa_async_suspend(spa);
9393 mutex_enter(&spa_namespace_lock);
9394 spa_close(spa, FTAG);
9396 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
9398 spa_deactivate(spa);
9402 mutex_exit(&spa_namespace_lock);
9406 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
9411 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
9415 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
9416 vd = spa->spa_l2cache.sav_vdevs[i];
9417 if (vd->vdev_guid == guid)
9421 for (i = 0; i < spa->spa_spares.sav_count; i++) {
9422 vd = spa->spa_spares.sav_vdevs[i];
9423 if (vd->vdev_guid == guid)
9432 spa_upgrade(spa_t *spa, uint64_t version)
9434 ASSERT(spa_writeable(spa));
9436 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
9439 * This should only be called for a non-faulted pool, and since a
9440 * future version would result in an unopenable pool, this shouldn't be
9443 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
9444 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
9446 spa->spa_uberblock.ub_version = version;
9447 vdev_config_dirty(spa->spa_root_vdev);
9449 spa_config_exit(spa, SCL_ALL, FTAG);
9451 txg_wait_synced(spa_get_dsl(spa), 0);
9455 spa_has_aux_vdev(spa_t *spa, uint64_t guid, spa_aux_vdev_t *sav)
9460 for (i = 0; i < sav->sav_count; i++)
9461 if (sav->sav_vdevs[i]->vdev_guid == guid)
9464 for (i = 0; i < sav->sav_npending; i++) {
9465 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
9466 &vdev_guid) == 0 && vdev_guid == guid)
9474 spa_has_l2cache(spa_t *spa, uint64_t guid)
9476 return (spa_has_aux_vdev(spa, guid, &spa->spa_l2cache));
9480 spa_has_spare(spa_t *spa, uint64_t guid)
9482 return (spa_has_aux_vdev(spa, guid, &spa->spa_spares));
9486 * Check if a pool has an active shared spare device.
9487 * Note: reference count of an active spare is 2, as a spare and as a replace
9490 spa_has_active_shared_spare(spa_t *spa)
9494 spa_aux_vdev_t *sav = &spa->spa_spares;
9496 for (i = 0; i < sav->sav_count; i++) {
9497 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
9498 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
9507 spa_total_metaslabs(spa_t *spa)
9509 vdev_t *rvd = spa->spa_root_vdev;
9512 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
9513 vdev_t *vd = rvd->vdev_child[c];
9514 if (!vdev_is_concrete(vd))
9516 m += vd->vdev_ms_count;
9522 * Notify any waiting threads that some activity has switched from being in-
9523 * progress to not-in-progress so that the thread can wake up and determine
9524 * whether it is finished waiting.
9527 spa_notify_waiters(spa_t *spa)
9530 * Acquiring spa_activities_lock here prevents the cv_broadcast from
9531 * happening between the waiting thread's check and cv_wait.
9533 mutex_enter(&spa->spa_activities_lock);
9534 cv_broadcast(&spa->spa_activities_cv);
9535 mutex_exit(&spa->spa_activities_lock);
9539 * Notify any waiting threads that the pool is exporting, and then block until
9540 * they are finished using the spa_t.
9543 spa_wake_waiters(spa_t *spa)
9545 mutex_enter(&spa->spa_activities_lock);
9546 spa->spa_waiters_cancel = B_TRUE;
9547 cv_broadcast(&spa->spa_activities_cv);
9548 while (spa->spa_waiters != 0)
9549 cv_wait(&spa->spa_waiters_cv, &spa->spa_activities_lock);
9550 spa->spa_waiters_cancel = B_FALSE;
9551 mutex_exit(&spa->spa_activities_lock);
9554 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
9556 spa_vdev_activity_in_progress_impl(vdev_t *vd, zpool_wait_activity_t activity)
9558 spa_t *spa = vd->vdev_spa;
9560 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER));
9561 ASSERT(MUTEX_HELD(&spa->spa_activities_lock));
9562 ASSERT(activity == ZPOOL_WAIT_INITIALIZE ||
9563 activity == ZPOOL_WAIT_TRIM);
9565 kmutex_t *lock = activity == ZPOOL_WAIT_INITIALIZE ?
9566 &vd->vdev_initialize_lock : &vd->vdev_trim_lock;
9568 mutex_exit(&spa->spa_activities_lock);
9570 mutex_enter(&spa->spa_activities_lock);
9572 boolean_t in_progress = (activity == ZPOOL_WAIT_INITIALIZE) ?
9573 (vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) :
9574 (vd->vdev_trim_state == VDEV_TRIM_ACTIVE);
9580 for (int i = 0; i < vd->vdev_children; i++) {
9581 if (spa_vdev_activity_in_progress_impl(vd->vdev_child[i],
9590 * If use_guid is true, this checks whether the vdev specified by guid is
9591 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
9592 * is being initialized/trimmed. The caller must hold the config lock and
9593 * spa_activities_lock.
9596 spa_vdev_activity_in_progress(spa_t *spa, boolean_t use_guid, uint64_t guid,
9597 zpool_wait_activity_t activity, boolean_t *in_progress)
9599 mutex_exit(&spa->spa_activities_lock);
9600 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9601 mutex_enter(&spa->spa_activities_lock);
9605 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
9606 if (vd == NULL || !vd->vdev_ops->vdev_op_leaf) {
9607 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9611 vd = spa->spa_root_vdev;
9614 *in_progress = spa_vdev_activity_in_progress_impl(vd, activity);
9616 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9621 * Locking for waiting threads
9622 * ---------------------------
9624 * Waiting threads need a way to check whether a given activity is in progress,
9625 * and then, if it is, wait for it to complete. Each activity will have some
9626 * in-memory representation of the relevant on-disk state which can be used to
9627 * determine whether or not the activity is in progress. The in-memory state and
9628 * the locking used to protect it will be different for each activity, and may
9629 * not be suitable for use with a cvar (e.g., some state is protected by the
9630 * config lock). To allow waiting threads to wait without any races, another
9631 * lock, spa_activities_lock, is used.
9633 * When the state is checked, both the activity-specific lock (if there is one)
9634 * and spa_activities_lock are held. In some cases, the activity-specific lock
9635 * is acquired explicitly (e.g. the config lock). In others, the locking is
9636 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
9637 * thread releases the activity-specific lock and, if the activity is in
9638 * progress, then cv_waits using spa_activities_lock.
9640 * The waiting thread is woken when another thread, one completing some
9641 * activity, updates the state of the activity and then calls
9642 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
9643 * needs to hold its activity-specific lock when updating the state, and this
9644 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
9646 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
9647 * and because it is held when the waiting thread checks the state of the
9648 * activity, it can never be the case that the completing thread both updates
9649 * the activity state and cv_broadcasts in between the waiting thread's check
9650 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
9652 * In order to prevent deadlock, when the waiting thread does its check, in some
9653 * cases it will temporarily drop spa_activities_lock in order to acquire the
9654 * activity-specific lock. The order in which spa_activities_lock and the
9655 * activity specific lock are acquired in the waiting thread is determined by
9656 * the order in which they are acquired in the completing thread; if the
9657 * completing thread calls spa_notify_waiters with the activity-specific lock
9658 * held, then the waiting thread must also acquire the activity-specific lock
9663 spa_activity_in_progress(spa_t *spa, zpool_wait_activity_t activity,
9664 boolean_t use_tag, uint64_t tag, boolean_t *in_progress)
9668 ASSERT(MUTEX_HELD(&spa->spa_activities_lock));
9671 case ZPOOL_WAIT_CKPT_DISCARD:
9673 (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT) &&
9674 zap_contains(spa_meta_objset(spa),
9675 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT) ==
9678 case ZPOOL_WAIT_FREE:
9679 *in_progress = ((spa_version(spa) >= SPA_VERSION_DEADLISTS &&
9680 !bpobj_is_empty(&spa->spa_dsl_pool->dp_free_bpobj)) ||
9681 spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY) ||
9682 spa_livelist_delete_check(spa));
9684 case ZPOOL_WAIT_INITIALIZE:
9685 case ZPOOL_WAIT_TRIM:
9686 error = spa_vdev_activity_in_progress(spa, use_tag, tag,
9687 activity, in_progress);
9689 case ZPOOL_WAIT_REPLACE:
9690 mutex_exit(&spa->spa_activities_lock);
9691 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
9692 mutex_enter(&spa->spa_activities_lock);
9694 *in_progress = vdev_replace_in_progress(spa->spa_root_vdev);
9695 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
9697 case ZPOOL_WAIT_REMOVE:
9698 *in_progress = (spa->spa_removing_phys.sr_state ==
9701 case ZPOOL_WAIT_RESILVER:
9702 if ((*in_progress = vdev_rebuild_active(spa->spa_root_vdev)))
9705 case ZPOOL_WAIT_SCRUB:
9707 boolean_t scanning, paused, is_scrub;
9708 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
9710 is_scrub = (scn->scn_phys.scn_func == POOL_SCAN_SCRUB);
9711 scanning = (scn->scn_phys.scn_state == DSS_SCANNING);
9712 paused = dsl_scan_is_paused_scrub(scn);
9713 *in_progress = (scanning && !paused &&
9714 is_scrub == (activity == ZPOOL_WAIT_SCRUB));
9718 panic("unrecognized value for activity %d", activity);
9725 spa_wait_common(const char *pool, zpool_wait_activity_t activity,
9726 boolean_t use_tag, uint64_t tag, boolean_t *waited)
9729 * The tag is used to distinguish between instances of an activity.
9730 * 'initialize' and 'trim' are the only activities that we use this for.
9731 * The other activities can only have a single instance in progress in a
9732 * pool at one time, making the tag unnecessary.
9734 * There can be multiple devices being replaced at once, but since they
9735 * all finish once resilvering finishes, we don't bother keeping track
9736 * of them individually, we just wait for them all to finish.
9738 if (use_tag && activity != ZPOOL_WAIT_INITIALIZE &&
9739 activity != ZPOOL_WAIT_TRIM)
9742 if (activity < 0 || activity >= ZPOOL_WAIT_NUM_ACTIVITIES)
9746 int error = spa_open(pool, &spa, FTAG);
9751 * Increment the spa's waiter count so that we can call spa_close and
9752 * still ensure that the spa_t doesn't get freed before this thread is
9753 * finished with it when the pool is exported. We want to call spa_close
9754 * before we start waiting because otherwise the additional ref would
9755 * prevent the pool from being exported or destroyed throughout the
9756 * potentially long wait.
9758 mutex_enter(&spa->spa_activities_lock);
9760 spa_close(spa, FTAG);
9764 boolean_t in_progress;
9765 error = spa_activity_in_progress(spa, activity, use_tag, tag,
9768 if (error || !in_progress || spa->spa_waiters_cancel)
9773 if (cv_wait_sig(&spa->spa_activities_cv,
9774 &spa->spa_activities_lock) == 0) {
9781 cv_signal(&spa->spa_waiters_cv);
9782 mutex_exit(&spa->spa_activities_lock);
9788 * Wait for a particular instance of the specified activity to complete, where
9789 * the instance is identified by 'tag'
9792 spa_wait_tag(const char *pool, zpool_wait_activity_t activity, uint64_t tag,
9795 return (spa_wait_common(pool, activity, B_TRUE, tag, waited));
9799 * Wait for all instances of the specified activity complete
9802 spa_wait(const char *pool, zpool_wait_activity_t activity, boolean_t *waited)
9805 return (spa_wait_common(pool, activity, B_FALSE, 0, waited));
9809 spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
9811 sysevent_t *ev = NULL;
9815 resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl);
9817 ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP);
9818 ev->resource = resource;
9825 spa_event_post(sysevent_t *ev)
9829 zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb);
9830 kmem_free(ev, sizeof (*ev));
9836 * Post a zevent corresponding to the given sysevent. The 'name' must be one
9837 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
9838 * filled in from the spa and (optionally) the vdev. This doesn't do anything
9839 * in the userland libzpool, as we don't want consumers to misinterpret ztest
9840 * or zdb as real changes.
9843 spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
9845 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));
9848 /* state manipulation functions */
9849 EXPORT_SYMBOL(spa_open);
9850 EXPORT_SYMBOL(spa_open_rewind);
9851 EXPORT_SYMBOL(spa_get_stats);
9852 EXPORT_SYMBOL(spa_create);
9853 EXPORT_SYMBOL(spa_import);
9854 EXPORT_SYMBOL(spa_tryimport);
9855 EXPORT_SYMBOL(spa_destroy);
9856 EXPORT_SYMBOL(spa_export);
9857 EXPORT_SYMBOL(spa_reset);
9858 EXPORT_SYMBOL(spa_async_request);
9859 EXPORT_SYMBOL(spa_async_suspend);
9860 EXPORT_SYMBOL(spa_async_resume);
9861 EXPORT_SYMBOL(spa_inject_addref);
9862 EXPORT_SYMBOL(spa_inject_delref);
9863 EXPORT_SYMBOL(spa_scan_stat_init);
9864 EXPORT_SYMBOL(spa_scan_get_stats);
9866 /* device manipulation */
9867 EXPORT_SYMBOL(spa_vdev_add);
9868 EXPORT_SYMBOL(spa_vdev_attach);
9869 EXPORT_SYMBOL(spa_vdev_detach);
9870 EXPORT_SYMBOL(spa_vdev_setpath);
9871 EXPORT_SYMBOL(spa_vdev_setfru);
9872 EXPORT_SYMBOL(spa_vdev_split_mirror);
9874 /* spare statech is global across all pools) */
9875 EXPORT_SYMBOL(spa_spare_add);
9876 EXPORT_SYMBOL(spa_spare_remove);
9877 EXPORT_SYMBOL(spa_spare_exists);
9878 EXPORT_SYMBOL(spa_spare_activate);
9880 /* L2ARC statech is global across all pools) */
9881 EXPORT_SYMBOL(spa_l2cache_add);
9882 EXPORT_SYMBOL(spa_l2cache_remove);
9883 EXPORT_SYMBOL(spa_l2cache_exists);
9884 EXPORT_SYMBOL(spa_l2cache_activate);
9885 EXPORT_SYMBOL(spa_l2cache_drop);
9888 EXPORT_SYMBOL(spa_scan);
9889 EXPORT_SYMBOL(spa_scan_stop);
9892 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
9893 EXPORT_SYMBOL(spa_sync_allpools);
9896 EXPORT_SYMBOL(spa_prop_set);
9897 EXPORT_SYMBOL(spa_prop_get);
9898 EXPORT_SYMBOL(spa_prop_clear_bootfs);
9900 /* asynchronous event notification */
9901 EXPORT_SYMBOL(spa_event_notify);
9904 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_shift, INT, ZMOD_RW,
9905 "log2 fraction of arc that can be used by inflight I/Os when "
9906 "verifying pool during import");
9908 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_metadata, INT, ZMOD_RW,
9909 "Set to traverse metadata on pool import");
9911 ZFS_MODULE_PARAM(zfs_spa, spa_, load_verify_data, INT, ZMOD_RW,
9912 "Set to traverse data on pool import");
9914 ZFS_MODULE_PARAM(zfs_spa, spa_, load_print_vdev_tree, INT, ZMOD_RW,
9915 "Print vdev tree to zfs_dbgmsg during pool import");
9917 ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_pct, UINT, ZMOD_RD,
9918 "Percentage of CPUs to run an IO worker thread");
9920 ZFS_MODULE_PARAM(zfs_zio, zio_, taskq_batch_tpq, UINT, ZMOD_RD,
9921 "Number of threads per IO worker taskqueue");
9923 ZFS_MODULE_PARAM(zfs, zfs_, max_missing_tvds, ULONG, ZMOD_RW,
9924 "Allow importing pool with up to this number of missing top-level "
9925 "vdevs (in read-only mode)");
9927 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_pause, INT, ZMOD_RW,
9928 "Set the livelist condense zthr to pause");
9930 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_pause, INT, ZMOD_RW,
9931 "Set the livelist condense synctask to pause");
9933 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, sync_cancel, INT, ZMOD_RW,
9934 "Whether livelist condensing was canceled in the synctask");
9936 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, zthr_cancel, INT, ZMOD_RW,
9937 "Whether livelist condensing was canceled in the zthr function");
9939 ZFS_MODULE_PARAM(zfs_livelist_condense, zfs_livelist_condense_, new_alloc, INT, ZMOD_RW,
9940 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
9941 "was being condensed");