4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2018, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2016 Toomas Soome <tsoome@me.com>
30 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
31 * Copyright 2018 Joyent, Inc.
32 * Copyright (c) 2017, 2019, Datto Inc. All rights reserved.
33 * Copyright 2017 Joyent, Inc.
34 * Copyright (c) 2017, Intel Corporation.
38 * SPA: Storage Pool Allocator
40 * This file contains all the routines used when modifying on-disk SPA state.
41 * This includes opening, importing, destroying, exporting a pool, and syncing a
45 #include <sys/zfs_context.h>
46 #include <sys/fm/fs/zfs.h>
47 #include <sys/spa_impl.h>
49 #include <sys/zio_checksum.h>
51 #include <sys/dmu_tx.h>
55 #include <sys/vdev_impl.h>
56 #include <sys/vdev_removal.h>
57 #include <sys/vdev_indirect_mapping.h>
58 #include <sys/vdev_indirect_births.h>
59 #include <sys/vdev_initialize.h>
60 #include <sys/vdev_rebuild.h>
61 #include <sys/vdev_trim.h>
62 #include <sys/vdev_disk.h>
63 #include <sys/metaslab.h>
64 #include <sys/metaslab_impl.h>
66 #include <sys/uberblock_impl.h>
69 #include <sys/bpobj.h>
70 #include <sys/dmu_traverse.h>
71 #include <sys/dmu_objset.h>
72 #include <sys/unique.h>
73 #include <sys/dsl_pool.h>
74 #include <sys/dsl_dataset.h>
75 #include <sys/dsl_dir.h>
76 #include <sys/dsl_prop.h>
77 #include <sys/dsl_synctask.h>
78 #include <sys/fs/zfs.h>
80 #include <sys/callb.h>
81 #include <sys/systeminfo.h>
82 #include <sys/spa_boot.h>
83 #include <sys/zfs_ioctl.h>
84 #include <sys/dsl_scan.h>
85 #include <sys/zfeature.h>
86 #include <sys/dsl_destroy.h>
90 #include <sys/fm/protocol.h>
91 #include <sys/fm/util.h>
92 #include <sys/callb.h>
94 #include <sys/vmsystm.h>
98 #include "zfs_comutil.h"
101 * The interval, in seconds, at which failed configuration cache file writes
104 int zfs_ccw_retry_interval
= 300;
106 typedef enum zti_modes
{
107 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
108 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
109 ZTI_MODE_NULL
, /* don't create a taskq */
113 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
114 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
115 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
116 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
118 #define ZTI_N(n) ZTI_P(n, 1)
119 #define ZTI_ONE ZTI_N(1)
121 typedef struct zio_taskq_info
{
122 zti_modes_t zti_mode
;
127 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
128 "iss", "iss_h", "int", "int_h"
132 * This table defines the taskq settings for each ZFS I/O type. When
133 * initializing a pool, we use this table to create an appropriately sized
134 * taskq. Some operations are low volume and therefore have a small, static
135 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
136 * macros. Other operations process a large amount of data; the ZTI_BATCH
137 * macro causes us to create a taskq oriented for throughput. Some operations
138 * are so high frequency and short-lived that the taskq itself can become a
139 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
140 * additional degree of parallelism specified by the number of threads per-
141 * taskq and the number of taskqs; when dispatching an event in this case, the
142 * particular taskq is chosen at random.
144 * The different taskq priorities are to handle the different contexts (issue
145 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
146 * need to be handled with minimum delay.
148 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
149 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
150 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
151 { ZTI_N(8), ZTI_NULL
, ZTI_P(12, 8), ZTI_NULL
}, /* READ */
152 { ZTI_BATCH
, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */
153 { ZTI_P(12, 8), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
154 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
155 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
156 { ZTI_N(4), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* TRIM */
159 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
160 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
161 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
162 static int spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
);
163 static void spa_vdev_resilver_done(spa_t
*spa
);
165 uint_t zio_taskq_batch_pct
= 75; /* 1 thread per cpu in pset */
166 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
167 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
169 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
172 * Report any spa_load_verify errors found, but do not fail spa_load.
173 * This is used by zdb to analyze non-idle pools.
175 boolean_t spa_load_verify_dryrun
= B_FALSE
;
178 * This (illegal) pool name is used when temporarily importing a spa_t in order
179 * to get the vdev stats associated with the imported devices.
181 #define TRYIMPORT_NAME "$import"
184 * For debugging purposes: print out vdev tree during pool import.
186 int spa_load_print_vdev_tree
= B_FALSE
;
189 * A non-zero value for zfs_max_missing_tvds means that we allow importing
190 * pools with missing top-level vdevs. This is strictly intended for advanced
191 * pool recovery cases since missing data is almost inevitable. Pools with
192 * missing devices can only be imported read-only for safety reasons, and their
193 * fail-mode will be automatically set to "continue".
195 * With 1 missing vdev we should be able to import the pool and mount all
196 * datasets. User data that was not modified after the missing device has been
197 * added should be recoverable. This means that snapshots created prior to the
198 * addition of that device should be completely intact.
200 * With 2 missing vdevs, some datasets may fail to mount since there are
201 * dataset statistics that are stored as regular metadata. Some data might be
202 * recoverable if those vdevs were added recently.
204 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
205 * may be missing entirely. Chances of data recovery are very low. Note that
206 * there are also risks of performing an inadvertent rewind as we might be
207 * missing all the vdevs with the latest uberblocks.
209 unsigned long zfs_max_missing_tvds
= 0;
212 * The parameters below are similar to zfs_max_missing_tvds but are only
213 * intended for a preliminary open of the pool with an untrusted config which
214 * might be incomplete or out-dated.
216 * We are more tolerant for pools opened from a cachefile since we could have
217 * an out-dated cachefile where a device removal was not registered.
218 * We could have set the limit arbitrarily high but in the case where devices
219 * are really missing we would want to return the proper error codes; we chose
220 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
221 * and we get a chance to retrieve the trusted config.
223 uint64_t zfs_max_missing_tvds_cachefile
= SPA_DVAS_PER_BP
- 1;
226 * In the case where config was assembled by scanning device paths (/dev/dsks
227 * by default) we are less tolerant since all the existing devices should have
228 * been detected and we want spa_load to return the right error codes.
230 uint64_t zfs_max_missing_tvds_scan
= 0;
233 * Debugging aid that pauses spa_sync() towards the end.
235 boolean_t zfs_pause_spa_sync
= B_FALSE
;
238 * Variables to indicate the livelist condense zthr func should wait at certain
239 * points for the livelist to be removed - used to test condense/destroy races
241 int zfs_livelist_condense_zthr_pause
= 0;
242 int zfs_livelist_condense_sync_pause
= 0;
245 * Variables to track whether or not condense cancellation has been
246 * triggered in testing.
248 int zfs_livelist_condense_sync_cancel
= 0;
249 int zfs_livelist_condense_zthr_cancel
= 0;
252 * Variable to track whether or not extra ALLOC blkptrs were added to a
253 * livelist entry while it was being condensed (caused by the way we track
254 * remapped blkptrs in dbuf_remap_impl)
256 int zfs_livelist_condense_new_alloc
= 0;
259 * ==========================================================================
260 * SPA properties routines
261 * ==========================================================================
265 * Add a (source=src, propname=propval) list to an nvlist.
268 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
269 uint64_t intval
, zprop_source_t src
)
271 const char *propname
= zpool_prop_to_name(prop
);
274 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
275 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
278 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
280 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
282 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
283 nvlist_free(propval
);
287 * Get property values from the spa configuration.
290 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
292 vdev_t
*rvd
= spa
->spa_root_vdev
;
293 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
294 uint64_t size
, alloc
, cap
, version
;
295 const zprop_source_t src
= ZPROP_SRC_NONE
;
296 spa_config_dirent_t
*dp
;
297 metaslab_class_t
*mc
= spa_normal_class(spa
);
299 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
302 alloc
= metaslab_class_get_alloc(mc
);
303 alloc
+= metaslab_class_get_alloc(spa_special_class(spa
));
304 alloc
+= metaslab_class_get_alloc(spa_dedup_class(spa
));
306 size
= metaslab_class_get_space(mc
);
307 size
+= metaslab_class_get_space(spa_special_class(spa
));
308 size
+= metaslab_class_get_space(spa_dedup_class(spa
));
310 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
311 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
312 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
313 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
315 spa_prop_add_list(*nvp
, ZPOOL_PROP_CHECKPOINT
, NULL
,
316 spa
->spa_checkpoint_info
.sci_dspace
, src
);
318 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
319 metaslab_class_fragmentation(mc
), src
);
320 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
321 metaslab_class_expandable_space(mc
), src
);
322 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
323 (spa_mode(spa
) == SPA_MODE_READ
), src
);
325 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
326 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
328 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
329 ddt_get_pool_dedup_ratio(spa
), src
);
331 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
332 rvd
->vdev_state
, src
);
334 version
= spa_version(spa
);
335 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
)) {
336 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
337 version
, ZPROP_SRC_DEFAULT
);
339 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
340 version
, ZPROP_SRC_LOCAL
);
342 spa_prop_add_list(*nvp
, ZPOOL_PROP_LOAD_GUID
,
343 NULL
, spa_load_guid(spa
), src
);
348 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
349 * when opening pools before this version freedir will be NULL.
351 if (pool
->dp_free_dir
!= NULL
) {
352 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
353 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
356 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
360 if (pool
->dp_leak_dir
!= NULL
) {
361 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
362 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
365 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
370 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
372 if (spa
->spa_comment
!= NULL
) {
373 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
377 if (spa
->spa_root
!= NULL
)
378 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
381 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
382 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
383 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
385 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
386 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
389 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
)) {
390 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
391 DNODE_MAX_SIZE
, ZPROP_SRC_NONE
);
393 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
394 DNODE_MIN_SIZE
, ZPROP_SRC_NONE
);
397 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
398 if (dp
->scd_path
== NULL
) {
399 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
400 "none", 0, ZPROP_SRC_LOCAL
);
401 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
402 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
403 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
409 * Get zpool property values.
412 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
414 objset_t
*mos
= spa
->spa_meta_objset
;
420 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
424 dp
= spa_get_dsl(spa
);
425 dsl_pool_config_enter(dp
, FTAG
);
426 mutex_enter(&spa
->spa_props_lock
);
429 * Get properties from the spa config.
431 spa_prop_get_config(spa
, nvp
);
433 /* If no pool property object, no more prop to get. */
434 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0)
438 * Get properties from the MOS pool property object.
440 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
441 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
442 zap_cursor_advance(&zc
)) {
445 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
448 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPOOL_PROP_INVAL
)
451 switch (za
.za_integer_length
) {
453 /* integer property */
454 if (za
.za_first_integer
!=
455 zpool_prop_default_numeric(prop
))
456 src
= ZPROP_SRC_LOCAL
;
458 if (prop
== ZPOOL_PROP_BOOTFS
) {
459 dsl_dataset_t
*ds
= NULL
;
461 err
= dsl_dataset_hold_obj(dp
,
462 za
.za_first_integer
, FTAG
, &ds
);
466 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
468 dsl_dataset_name(ds
, strval
);
469 dsl_dataset_rele(ds
, FTAG
);
472 intval
= za
.za_first_integer
;
475 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
478 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
483 /* string property */
484 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
485 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
486 za
.za_name
, 1, za
.za_num_integers
, strval
);
488 kmem_free(strval
, za
.za_num_integers
);
491 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
492 kmem_free(strval
, za
.za_num_integers
);
499 zap_cursor_fini(&zc
);
501 mutex_exit(&spa
->spa_props_lock
);
502 dsl_pool_config_exit(dp
, FTAG
);
503 if (err
&& err
!= ENOENT
) {
513 * Validate the given pool properties nvlist and modify the list
514 * for the property values to be set.
517 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
520 int error
= 0, reset_bootfs
= 0;
522 boolean_t has_feature
= B_FALSE
;
525 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
527 char *strval
, *slash
, *check
, *fname
;
528 const char *propname
= nvpair_name(elem
);
529 zpool_prop_t prop
= zpool_name_to_prop(propname
);
532 case ZPOOL_PROP_INVAL
:
533 if (!zpool_prop_feature(propname
)) {
534 error
= SET_ERROR(EINVAL
);
539 * Sanitize the input.
541 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
542 error
= SET_ERROR(EINVAL
);
546 if (nvpair_value_uint64(elem
, &intval
) != 0) {
547 error
= SET_ERROR(EINVAL
);
552 error
= SET_ERROR(EINVAL
);
556 fname
= strchr(propname
, '@') + 1;
557 if (zfeature_lookup_name(fname
, NULL
) != 0) {
558 error
= SET_ERROR(EINVAL
);
562 has_feature
= B_TRUE
;
565 case ZPOOL_PROP_VERSION
:
566 error
= nvpair_value_uint64(elem
, &intval
);
568 (intval
< spa_version(spa
) ||
569 intval
> SPA_VERSION_BEFORE_FEATURES
||
571 error
= SET_ERROR(EINVAL
);
574 case ZPOOL_PROP_DELEGATION
:
575 case ZPOOL_PROP_AUTOREPLACE
:
576 case ZPOOL_PROP_LISTSNAPS
:
577 case ZPOOL_PROP_AUTOEXPAND
:
578 case ZPOOL_PROP_AUTOTRIM
:
579 error
= nvpair_value_uint64(elem
, &intval
);
580 if (!error
&& intval
> 1)
581 error
= SET_ERROR(EINVAL
);
584 case ZPOOL_PROP_MULTIHOST
:
585 error
= nvpair_value_uint64(elem
, &intval
);
586 if (!error
&& intval
> 1)
587 error
= SET_ERROR(EINVAL
);
590 uint32_t hostid
= zone_get_hostid(NULL
);
592 spa
->spa_hostid
= hostid
;
594 error
= SET_ERROR(ENOTSUP
);
599 case ZPOOL_PROP_BOOTFS
:
601 * If the pool version is less than SPA_VERSION_BOOTFS,
602 * or the pool is still being created (version == 0),
603 * the bootfs property cannot be set.
605 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
606 error
= SET_ERROR(ENOTSUP
);
611 * Make sure the vdev config is bootable
613 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
614 error
= SET_ERROR(ENOTSUP
);
620 error
= nvpair_value_string(elem
, &strval
);
625 if (strval
== NULL
|| strval
[0] == '\0') {
626 objnum
= zpool_prop_default_numeric(
631 error
= dmu_objset_hold(strval
, FTAG
, &os
);
636 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
637 error
= SET_ERROR(ENOTSUP
);
639 objnum
= dmu_objset_id(os
);
641 dmu_objset_rele(os
, FTAG
);
645 case ZPOOL_PROP_FAILUREMODE
:
646 error
= nvpair_value_uint64(elem
, &intval
);
647 if (!error
&& intval
> ZIO_FAILURE_MODE_PANIC
)
648 error
= SET_ERROR(EINVAL
);
651 * This is a special case which only occurs when
652 * the pool has completely failed. This allows
653 * the user to change the in-core failmode property
654 * without syncing it out to disk (I/Os might
655 * currently be blocked). We do this by returning
656 * EIO to the caller (spa_prop_set) to trick it
657 * into thinking we encountered a property validation
660 if (!error
&& spa_suspended(spa
)) {
661 spa
->spa_failmode
= intval
;
662 error
= SET_ERROR(EIO
);
666 case ZPOOL_PROP_CACHEFILE
:
667 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
670 if (strval
[0] == '\0')
673 if (strcmp(strval
, "none") == 0)
676 if (strval
[0] != '/') {
677 error
= SET_ERROR(EINVAL
);
681 slash
= strrchr(strval
, '/');
682 ASSERT(slash
!= NULL
);
684 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
685 strcmp(slash
, "/..") == 0)
686 error
= SET_ERROR(EINVAL
);
689 case ZPOOL_PROP_COMMENT
:
690 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
692 for (check
= strval
; *check
!= '\0'; check
++) {
693 if (!isprint(*check
)) {
694 error
= SET_ERROR(EINVAL
);
698 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
699 error
= SET_ERROR(E2BIG
);
710 (void) nvlist_remove_all(props
,
711 zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO
));
713 if (!error
&& reset_bootfs
) {
714 error
= nvlist_remove(props
,
715 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
718 error
= nvlist_add_uint64(props
,
719 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
727 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
730 spa_config_dirent_t
*dp
;
732 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
736 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
739 if (cachefile
[0] == '\0')
740 dp
->scd_path
= spa_strdup(spa_config_path
);
741 else if (strcmp(cachefile
, "none") == 0)
744 dp
->scd_path
= spa_strdup(cachefile
);
746 list_insert_head(&spa
->spa_config_list
, dp
);
748 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
752 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
755 nvpair_t
*elem
= NULL
;
756 boolean_t need_sync
= B_FALSE
;
758 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
761 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
762 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
764 if (prop
== ZPOOL_PROP_CACHEFILE
||
765 prop
== ZPOOL_PROP_ALTROOT
||
766 prop
== ZPOOL_PROP_READONLY
)
769 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPOOL_PROP_INVAL
) {
772 if (prop
== ZPOOL_PROP_VERSION
) {
773 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
775 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
776 ver
= SPA_VERSION_FEATURES
;
780 /* Save time if the version is already set. */
781 if (ver
== spa_version(spa
))
785 * In addition to the pool directory object, we might
786 * create the pool properties object, the features for
787 * read object, the features for write object, or the
788 * feature descriptions object.
790 error
= dsl_sync_task(spa
->spa_name
, NULL
,
791 spa_sync_version
, &ver
,
792 6, ZFS_SPACE_CHECK_RESERVED
);
803 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
804 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
811 * If the bootfs property value is dsobj, clear it.
814 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
816 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
817 VERIFY(zap_remove(spa
->spa_meta_objset
,
818 spa
->spa_pool_props_object
,
819 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
826 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
828 uint64_t *newguid __maybe_unused
= arg
;
829 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
830 vdev_t
*rvd
= spa
->spa_root_vdev
;
833 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
834 int error
= (spa_has_checkpoint(spa
)) ?
835 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
836 return (SET_ERROR(error
));
839 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
840 vdev_state
= rvd
->vdev_state
;
841 spa_config_exit(spa
, SCL_STATE
, FTAG
);
843 if (vdev_state
!= VDEV_STATE_HEALTHY
)
844 return (SET_ERROR(ENXIO
));
846 ASSERT3U(spa_guid(spa
), !=, *newguid
);
852 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
854 uint64_t *newguid
= arg
;
855 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
857 vdev_t
*rvd
= spa
->spa_root_vdev
;
859 oldguid
= spa_guid(spa
);
861 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
862 rvd
->vdev_guid
= *newguid
;
863 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
864 vdev_config_dirty(rvd
);
865 spa_config_exit(spa
, SCL_STATE
, FTAG
);
867 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
868 (u_longlong_t
)oldguid
, (u_longlong_t
)*newguid
);
872 * Change the GUID for the pool. This is done so that we can later
873 * re-import a pool built from a clone of our own vdevs. We will modify
874 * the root vdev's guid, our own pool guid, and then mark all of our
875 * vdevs dirty. Note that we must make sure that all our vdevs are
876 * online when we do this, or else any vdevs that weren't present
877 * would be orphaned from our pool. We are also going to issue a
878 * sysevent to update any watchers.
881 spa_change_guid(spa_t
*spa
)
886 mutex_enter(&spa
->spa_vdev_top_lock
);
887 mutex_enter(&spa_namespace_lock
);
888 guid
= spa_generate_guid(NULL
);
890 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
891 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
894 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
895 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
898 mutex_exit(&spa_namespace_lock
);
899 mutex_exit(&spa
->spa_vdev_top_lock
);
905 * ==========================================================================
906 * SPA state manipulation (open/create/destroy/import/export)
907 * ==========================================================================
911 spa_error_entry_compare(const void *a
, const void *b
)
913 const spa_error_entry_t
*sa
= (const spa_error_entry_t
*)a
;
914 const spa_error_entry_t
*sb
= (const spa_error_entry_t
*)b
;
917 ret
= memcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
918 sizeof (zbookmark_phys_t
));
920 return (TREE_ISIGN(ret
));
924 * Utility function which retrieves copies of the current logs and
925 * re-initializes them in the process.
928 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
930 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
932 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
933 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
935 avl_create(&spa
->spa_errlist_scrub
,
936 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
937 offsetof(spa_error_entry_t
, se_avl
));
938 avl_create(&spa
->spa_errlist_last
,
939 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
940 offsetof(spa_error_entry_t
, se_avl
));
944 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
946 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
947 enum zti_modes mode
= ztip
->zti_mode
;
948 uint_t value
= ztip
->zti_value
;
949 uint_t count
= ztip
->zti_count
;
950 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
952 boolean_t batch
= B_FALSE
;
954 if (mode
== ZTI_MODE_NULL
) {
956 tqs
->stqs_taskq
= NULL
;
960 ASSERT3U(count
, >, 0);
962 tqs
->stqs_count
= count
;
963 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
967 ASSERT3U(value
, >=, 1);
968 value
= MAX(value
, 1);
969 flags
|= TASKQ_DYNAMIC
;
974 flags
|= TASKQ_THREADS_CPU_PCT
;
975 value
= MIN(zio_taskq_batch_pct
, 100);
979 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
981 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
985 for (uint_t i
= 0; i
< count
; i
++) {
989 (void) snprintf(name
, sizeof (name
), "%s_%s",
990 zio_type_name
[t
], zio_taskq_types
[q
]);
992 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
994 flags
|= TASKQ_DC_BATCH
;
996 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
997 spa
->spa_proc
, zio_taskq_basedc
, flags
);
999 pri_t pri
= maxclsyspri
;
1001 * The write issue taskq can be extremely CPU
1002 * intensive. Run it at slightly less important
1003 * priority than the other taskqs. Under Linux this
1004 * means incrementing the priority value on platforms
1005 * like illumos it should be decremented.
1007 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
1010 tq
= taskq_create_proc(name
, value
, pri
, 50,
1011 INT_MAX
, spa
->spa_proc
, flags
);
1014 tqs
->stqs_taskq
[i
] = tq
;
1019 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
1021 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1023 if (tqs
->stqs_taskq
== NULL
) {
1024 ASSERT3U(tqs
->stqs_count
, ==, 0);
1028 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
1029 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
1030 taskq_destroy(tqs
->stqs_taskq
[i
]);
1033 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
1034 tqs
->stqs_taskq
= NULL
;
1038 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1039 * Note that a type may have multiple discrete taskqs to avoid lock contention
1040 * on the taskq itself. In that case we choose which taskq at random by using
1041 * the low bits of gethrtime().
1044 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1045 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
1047 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1050 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1051 ASSERT3U(tqs
->stqs_count
, !=, 0);
1053 if (tqs
->stqs_count
== 1) {
1054 tq
= tqs
->stqs_taskq
[0];
1056 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1059 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
1063 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1066 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1067 task_func_t
*func
, void *arg
, uint_t flags
)
1069 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1073 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1074 ASSERT3U(tqs
->stqs_count
, !=, 0);
1076 if (tqs
->stqs_count
== 1) {
1077 tq
= tqs
->stqs_taskq
[0];
1079 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1082 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1084 taskq_wait_id(tq
, id
);
1088 spa_create_zio_taskqs(spa_t
*spa
)
1090 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1091 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1092 spa_taskqs_init(spa
, t
, q
);
1098 * Disabled until spa_thread() can be adapted for Linux.
1100 #undef HAVE_SPA_THREAD
1102 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1104 spa_thread(void *arg
)
1106 psetid_t zio_taskq_psrset_bind
= PS_NONE
;
1107 callb_cpr_t cprinfo
;
1110 user_t
*pu
= PTOU(curproc
);
1112 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1115 ASSERT(curproc
!= &p0
);
1116 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1117 "zpool-%s", spa
->spa_name
);
1118 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1120 /* bind this thread to the requested psrset */
1121 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1123 mutex_enter(&cpu_lock
);
1124 mutex_enter(&pidlock
);
1125 mutex_enter(&curproc
->p_lock
);
1127 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1128 0, NULL
, NULL
) == 0) {
1129 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1132 "Couldn't bind process for zfs pool \"%s\" to "
1133 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1136 mutex_exit(&curproc
->p_lock
);
1137 mutex_exit(&pidlock
);
1138 mutex_exit(&cpu_lock
);
1142 if (zio_taskq_sysdc
) {
1143 sysdc_thread_enter(curthread
, 100, 0);
1146 spa
->spa_proc
= curproc
;
1147 spa
->spa_did
= curthread
->t_did
;
1149 spa_create_zio_taskqs(spa
);
1151 mutex_enter(&spa
->spa_proc_lock
);
1152 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1154 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1155 cv_broadcast(&spa
->spa_proc_cv
);
1157 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1158 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1159 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1160 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1162 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1163 spa
->spa_proc_state
= SPA_PROC_GONE
;
1164 spa
->spa_proc
= &p0
;
1165 cv_broadcast(&spa
->spa_proc_cv
);
1166 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1168 mutex_enter(&curproc
->p_lock
);
1174 * Activate an uninitialized pool.
1177 spa_activate(spa_t
*spa
, spa_mode_t mode
)
1179 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1181 spa
->spa_state
= POOL_STATE_ACTIVE
;
1182 spa
->spa_mode
= mode
;
1184 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1185 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1186 spa
->spa_special_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1187 spa
->spa_dedup_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1189 /* Try to create a covering process */
1190 mutex_enter(&spa
->spa_proc_lock
);
1191 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1192 ASSERT(spa
->spa_proc
== &p0
);
1195 #ifdef HAVE_SPA_THREAD
1196 /* Only create a process if we're going to be around a while. */
1197 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1198 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1200 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1201 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1202 cv_wait(&spa
->spa_proc_cv
,
1203 &spa
->spa_proc_lock
);
1205 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1206 ASSERT(spa
->spa_proc
!= &p0
);
1207 ASSERT(spa
->spa_did
!= 0);
1211 "Couldn't create process for zfs pool \"%s\"\n",
1216 #endif /* HAVE_SPA_THREAD */
1217 mutex_exit(&spa
->spa_proc_lock
);
1219 /* If we didn't create a process, we need to create our taskqs. */
1220 if (spa
->spa_proc
== &p0
) {
1221 spa_create_zio_taskqs(spa
);
1224 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1225 spa
->spa_txg_zio
[i
] = zio_root(spa
, NULL
, NULL
,
1229 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1230 offsetof(vdev_t
, vdev_config_dirty_node
));
1231 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1232 offsetof(objset_t
, os_evicting_node
));
1233 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1234 offsetof(vdev_t
, vdev_state_dirty_node
));
1236 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1237 offsetof(struct vdev
, vdev_txg_node
));
1239 avl_create(&spa
->spa_errlist_scrub
,
1240 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1241 offsetof(spa_error_entry_t
, se_avl
));
1242 avl_create(&spa
->spa_errlist_last
,
1243 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1244 offsetof(spa_error_entry_t
, se_avl
));
1246 spa_keystore_init(&spa
->spa_keystore
);
1249 * This taskq is used to perform zvol-minor-related tasks
1250 * asynchronously. This has several advantages, including easy
1251 * resolution of various deadlocks (zfsonlinux bug #3681).
1253 * The taskq must be single threaded to ensure tasks are always
1254 * processed in the order in which they were dispatched.
1256 * A taskq per pool allows one to keep the pools independent.
1257 * This way if one pool is suspended, it will not impact another.
1259 * The preferred location to dispatch a zvol minor task is a sync
1260 * task. In this context, there is easy access to the spa_t and minimal
1261 * error handling is required because the sync task must succeed.
1263 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1267 * Taskq dedicated to prefetcher threads: this is used to prevent the
1268 * pool traverse code from monopolizing the global (and limited)
1269 * system_taskq by inappropriately scheduling long running tasks on it.
1271 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", boot_ncpus
,
1272 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1275 * The taskq to upgrade datasets in this pool. Currently used by
1276 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1278 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", boot_ncpus
,
1279 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1283 * Opposite of spa_activate().
1286 spa_deactivate(spa_t
*spa
)
1288 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1289 ASSERT(spa
->spa_dsl_pool
== NULL
);
1290 ASSERT(spa
->spa_root_vdev
== NULL
);
1291 ASSERT(spa
->spa_async_zio_root
== NULL
);
1292 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1294 spa_evicting_os_wait(spa
);
1296 if (spa
->spa_zvol_taskq
) {
1297 taskq_destroy(spa
->spa_zvol_taskq
);
1298 spa
->spa_zvol_taskq
= NULL
;
1301 if (spa
->spa_prefetch_taskq
) {
1302 taskq_destroy(spa
->spa_prefetch_taskq
);
1303 spa
->spa_prefetch_taskq
= NULL
;
1306 if (spa
->spa_upgrade_taskq
) {
1307 taskq_destroy(spa
->spa_upgrade_taskq
);
1308 spa
->spa_upgrade_taskq
= NULL
;
1311 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1313 list_destroy(&spa
->spa_config_dirty_list
);
1314 list_destroy(&spa
->spa_evicting_os_list
);
1315 list_destroy(&spa
->spa_state_dirty_list
);
1317 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1319 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1320 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1321 spa_taskqs_fini(spa
, t
, q
);
1325 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1326 ASSERT3P(spa
->spa_txg_zio
[i
], !=, NULL
);
1327 VERIFY0(zio_wait(spa
->spa_txg_zio
[i
]));
1328 spa
->spa_txg_zio
[i
] = NULL
;
1331 metaslab_class_destroy(spa
->spa_normal_class
);
1332 spa
->spa_normal_class
= NULL
;
1334 metaslab_class_destroy(spa
->spa_log_class
);
1335 spa
->spa_log_class
= NULL
;
1337 metaslab_class_destroy(spa
->spa_special_class
);
1338 spa
->spa_special_class
= NULL
;
1340 metaslab_class_destroy(spa
->spa_dedup_class
);
1341 spa
->spa_dedup_class
= NULL
;
1344 * If this was part of an import or the open otherwise failed, we may
1345 * still have errors left in the queues. Empty them just in case.
1347 spa_errlog_drain(spa
);
1348 avl_destroy(&spa
->spa_errlist_scrub
);
1349 avl_destroy(&spa
->spa_errlist_last
);
1351 spa_keystore_fini(&spa
->spa_keystore
);
1353 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1355 mutex_enter(&spa
->spa_proc_lock
);
1356 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1357 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1358 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1359 cv_broadcast(&spa
->spa_proc_cv
);
1360 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1361 ASSERT(spa
->spa_proc
!= &p0
);
1362 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1364 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1365 spa
->spa_proc_state
= SPA_PROC_NONE
;
1367 ASSERT(spa
->spa_proc
== &p0
);
1368 mutex_exit(&spa
->spa_proc_lock
);
1371 * We want to make sure spa_thread() has actually exited the ZFS
1372 * module, so that the module can't be unloaded out from underneath
1375 if (spa
->spa_did
!= 0) {
1376 thread_join(spa
->spa_did
);
1382 * Verify a pool configuration, and construct the vdev tree appropriately. This
1383 * will create all the necessary vdevs in the appropriate layout, with each vdev
1384 * in the CLOSED state. This will prep the pool before open/creation/import.
1385 * All vdev validation is done by the vdev_alloc() routine.
1388 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1389 uint_t id
, int atype
)
1395 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1398 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1401 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1404 if (error
== ENOENT
)
1410 return (SET_ERROR(EINVAL
));
1413 for (int c
= 0; c
< children
; c
++) {
1415 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1423 ASSERT(*vdp
!= NULL
);
1429 spa_should_flush_logs_on_unload(spa_t
*spa
)
1431 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
1434 if (!spa_writeable(spa
))
1437 if (!spa
->spa_sync_on
)
1440 if (spa_state(spa
) != POOL_STATE_EXPORTED
)
1443 if (zfs_keep_log_spacemaps_at_export
)
1450 * Opens a transaction that will set the flag that will instruct
1451 * spa_sync to attempt to flush all the metaslabs for that txg.
1454 spa_unload_log_sm_flush_all(spa_t
*spa
)
1456 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
1457 VERIFY0(dmu_tx_assign(tx
, TXG_WAIT
));
1459 ASSERT3U(spa
->spa_log_flushall_txg
, ==, 0);
1460 spa
->spa_log_flushall_txg
= dmu_tx_get_txg(tx
);
1463 txg_wait_synced(spa_get_dsl(spa
), spa
->spa_log_flushall_txg
);
1467 spa_unload_log_sm_metadata(spa_t
*spa
)
1469 void *cookie
= NULL
;
1471 while ((sls
= avl_destroy_nodes(&spa
->spa_sm_logs_by_txg
,
1472 &cookie
)) != NULL
) {
1473 VERIFY0(sls
->sls_mscount
);
1474 kmem_free(sls
, sizeof (spa_log_sm_t
));
1477 for (log_summary_entry_t
*e
= list_head(&spa
->spa_log_summary
);
1478 e
!= NULL
; e
= list_head(&spa
->spa_log_summary
)) {
1479 VERIFY0(e
->lse_mscount
);
1480 list_remove(&spa
->spa_log_summary
, e
);
1481 kmem_free(e
, sizeof (log_summary_entry_t
));
1484 spa
->spa_unflushed_stats
.sus_nblocks
= 0;
1485 spa
->spa_unflushed_stats
.sus_memused
= 0;
1486 spa
->spa_unflushed_stats
.sus_blocklimit
= 0;
1490 spa_destroy_aux_threads(spa_t
*spa
)
1492 if (spa
->spa_condense_zthr
!= NULL
) {
1493 zthr_destroy(spa
->spa_condense_zthr
);
1494 spa
->spa_condense_zthr
= NULL
;
1496 if (spa
->spa_checkpoint_discard_zthr
!= NULL
) {
1497 zthr_destroy(spa
->spa_checkpoint_discard_zthr
);
1498 spa
->spa_checkpoint_discard_zthr
= NULL
;
1500 if (spa
->spa_livelist_delete_zthr
!= NULL
) {
1501 zthr_destroy(spa
->spa_livelist_delete_zthr
);
1502 spa
->spa_livelist_delete_zthr
= NULL
;
1504 if (spa
->spa_livelist_condense_zthr
!= NULL
) {
1505 zthr_destroy(spa
->spa_livelist_condense_zthr
);
1506 spa
->spa_livelist_condense_zthr
= NULL
;
1511 * Opposite of spa_load().
1514 spa_unload(spa_t
*spa
)
1516 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1517 ASSERT(spa_state(spa
) != POOL_STATE_UNINITIALIZED
);
1519 spa_import_progress_remove(spa_guid(spa
));
1520 spa_load_note(spa
, "UNLOADING");
1522 spa_wake_waiters(spa
);
1525 * If the log space map feature is enabled and the pool is getting
1526 * exported (but not destroyed), we want to spend some time flushing
1527 * as many metaslabs as we can in an attempt to destroy log space
1528 * maps and save import time.
1530 if (spa_should_flush_logs_on_unload(spa
))
1531 spa_unload_log_sm_flush_all(spa
);
1536 spa_async_suspend(spa
);
1538 if (spa
->spa_root_vdev
) {
1539 vdev_t
*root_vdev
= spa
->spa_root_vdev
;
1540 vdev_initialize_stop_all(root_vdev
, VDEV_INITIALIZE_ACTIVE
);
1541 vdev_trim_stop_all(root_vdev
, VDEV_TRIM_ACTIVE
);
1542 vdev_autotrim_stop_all(spa
);
1543 vdev_rebuild_stop_all(spa
);
1549 if (spa
->spa_sync_on
) {
1550 txg_sync_stop(spa
->spa_dsl_pool
);
1551 spa
->spa_sync_on
= B_FALSE
;
1555 * This ensures that there is no async metaslab prefetching
1556 * while we attempt to unload the spa.
1558 if (spa
->spa_root_vdev
!= NULL
) {
1559 for (int c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++) {
1560 vdev_t
*vc
= spa
->spa_root_vdev
->vdev_child
[c
];
1561 if (vc
->vdev_mg
!= NULL
)
1562 taskq_wait(vc
->vdev_mg
->mg_taskq
);
1566 if (spa
->spa_mmp
.mmp_thread
)
1567 mmp_thread_stop(spa
);
1570 * Wait for any outstanding async I/O to complete.
1572 if (spa
->spa_async_zio_root
!= NULL
) {
1573 for (int i
= 0; i
< max_ncpus
; i
++)
1574 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1575 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1576 spa
->spa_async_zio_root
= NULL
;
1579 if (spa
->spa_vdev_removal
!= NULL
) {
1580 spa_vdev_removal_destroy(spa
->spa_vdev_removal
);
1581 spa
->spa_vdev_removal
= NULL
;
1584 spa_destroy_aux_threads(spa
);
1586 spa_condense_fini(spa
);
1588 bpobj_close(&spa
->spa_deferred_bpobj
);
1590 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1595 if (spa
->spa_root_vdev
)
1596 vdev_free(spa
->spa_root_vdev
);
1597 ASSERT(spa
->spa_root_vdev
== NULL
);
1600 * Close the dsl pool.
1602 if (spa
->spa_dsl_pool
) {
1603 dsl_pool_close(spa
->spa_dsl_pool
);
1604 spa
->spa_dsl_pool
= NULL
;
1605 spa
->spa_meta_objset
= NULL
;
1609 spa_unload_log_sm_metadata(spa
);
1612 * Drop and purge level 2 cache
1614 spa_l2cache_drop(spa
);
1616 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1617 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1618 if (spa
->spa_spares
.sav_vdevs
) {
1619 kmem_free(spa
->spa_spares
.sav_vdevs
,
1620 spa
->spa_spares
.sav_count
* sizeof (void *));
1621 spa
->spa_spares
.sav_vdevs
= NULL
;
1623 if (spa
->spa_spares
.sav_config
) {
1624 nvlist_free(spa
->spa_spares
.sav_config
);
1625 spa
->spa_spares
.sav_config
= NULL
;
1627 spa
->spa_spares
.sav_count
= 0;
1629 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1630 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1631 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1633 if (spa
->spa_l2cache
.sav_vdevs
) {
1634 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1635 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1636 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1638 if (spa
->spa_l2cache
.sav_config
) {
1639 nvlist_free(spa
->spa_l2cache
.sav_config
);
1640 spa
->spa_l2cache
.sav_config
= NULL
;
1642 spa
->spa_l2cache
.sav_count
= 0;
1644 spa
->spa_async_suspended
= 0;
1646 spa
->spa_indirect_vdevs_loaded
= B_FALSE
;
1648 if (spa
->spa_comment
!= NULL
) {
1649 spa_strfree(spa
->spa_comment
);
1650 spa
->spa_comment
= NULL
;
1653 spa_config_exit(spa
, SCL_ALL
, spa
);
1657 * Load (or re-load) the current list of vdevs describing the active spares for
1658 * this pool. When this is called, we have some form of basic information in
1659 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1660 * then re-generate a more complete list including status information.
1663 spa_load_spares(spa_t
*spa
)
1672 * zdb opens both the current state of the pool and the
1673 * checkpointed state (if present), with a different spa_t.
1675 * As spare vdevs are shared among open pools, we skip loading
1676 * them when we load the checkpointed state of the pool.
1678 if (!spa_writeable(spa
))
1682 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1685 * First, close and free any existing spare vdevs.
1687 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1688 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1690 /* Undo the call to spa_activate() below */
1691 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1692 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1693 spa_spare_remove(tvd
);
1698 if (spa
->spa_spares
.sav_vdevs
)
1699 kmem_free(spa
->spa_spares
.sav_vdevs
,
1700 spa
->spa_spares
.sav_count
* sizeof (void *));
1702 if (spa
->spa_spares
.sav_config
== NULL
)
1705 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1706 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1708 spa
->spa_spares
.sav_count
= (int)nspares
;
1709 spa
->spa_spares
.sav_vdevs
= NULL
;
1715 * Construct the array of vdevs, opening them to get status in the
1716 * process. For each spare, there is potentially two different vdev_t
1717 * structures associated with it: one in the list of spares (used only
1718 * for basic validation purposes) and one in the active vdev
1719 * configuration (if it's spared in). During this phase we open and
1720 * validate each vdev on the spare list. If the vdev also exists in the
1721 * active configuration, then we also mark this vdev as an active spare.
1723 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1725 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1726 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1727 VDEV_ALLOC_SPARE
) == 0);
1730 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1732 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1733 B_FALSE
)) != NULL
) {
1734 if (!tvd
->vdev_isspare
)
1738 * We only mark the spare active if we were successfully
1739 * able to load the vdev. Otherwise, importing a pool
1740 * with a bad active spare would result in strange
1741 * behavior, because multiple pool would think the spare
1742 * is actively in use.
1744 * There is a vulnerability here to an equally bizarre
1745 * circumstance, where a dead active spare is later
1746 * brought back to life (onlined or otherwise). Given
1747 * the rarity of this scenario, and the extra complexity
1748 * it adds, we ignore the possibility.
1750 if (!vdev_is_dead(tvd
))
1751 spa_spare_activate(tvd
);
1755 vd
->vdev_aux
= &spa
->spa_spares
;
1757 if (vdev_open(vd
) != 0)
1760 if (vdev_validate_aux(vd
) == 0)
1765 * Recompute the stashed list of spares, with status information
1768 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1769 DATA_TYPE_NVLIST_ARRAY
) == 0);
1771 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1773 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1774 spares
[i
] = vdev_config_generate(spa
,
1775 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1776 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1777 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1778 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1779 nvlist_free(spares
[i
]);
1780 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1784 * Load (or re-load) the current list of vdevs describing the active l2cache for
1785 * this pool. When this is called, we have some form of basic information in
1786 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1787 * then re-generate a more complete list including status information.
1788 * Devices which are already active have their details maintained, and are
1792 spa_load_l2cache(spa_t
*spa
)
1794 nvlist_t
**l2cache
= NULL
;
1796 int i
, j
, oldnvdevs
;
1798 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1799 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1803 * zdb opens both the current state of the pool and the
1804 * checkpointed state (if present), with a different spa_t.
1806 * As L2 caches are part of the ARC which is shared among open
1807 * pools, we skip loading them when we load the checkpointed
1808 * state of the pool.
1810 if (!spa_writeable(spa
))
1814 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1816 oldvdevs
= sav
->sav_vdevs
;
1817 oldnvdevs
= sav
->sav_count
;
1818 sav
->sav_vdevs
= NULL
;
1821 if (sav
->sav_config
== NULL
) {
1827 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1828 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1829 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1832 * Process new nvlist of vdevs.
1834 for (i
= 0; i
< nl2cache
; i
++) {
1835 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1839 for (j
= 0; j
< oldnvdevs
; j
++) {
1841 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1843 * Retain previous vdev for add/remove ops.
1851 if (newvdevs
[i
] == NULL
) {
1855 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1856 VDEV_ALLOC_L2CACHE
) == 0);
1861 * Commit this vdev as an l2cache device,
1862 * even if it fails to open.
1864 spa_l2cache_add(vd
);
1869 spa_l2cache_activate(vd
);
1871 if (vdev_open(vd
) != 0)
1874 (void) vdev_validate_aux(vd
);
1876 if (!vdev_is_dead(vd
))
1877 l2arc_add_vdev(spa
, vd
);
1880 * Upon cache device addition to a pool or pool
1881 * creation with a cache device or if the header
1882 * of the device is invalid we issue an async
1883 * TRIM command for the whole device which will
1884 * execute if l2arc_trim_ahead > 0.
1886 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
1890 sav
->sav_vdevs
= newvdevs
;
1891 sav
->sav_count
= (int)nl2cache
;
1894 * Recompute the stashed list of l2cache devices, with status
1895 * information this time.
1897 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1898 DATA_TYPE_NVLIST_ARRAY
) == 0);
1900 if (sav
->sav_count
> 0)
1901 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *),
1903 for (i
= 0; i
< sav
->sav_count
; i
++)
1904 l2cache
[i
] = vdev_config_generate(spa
,
1905 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1906 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1907 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1911 * Purge vdevs that were dropped
1913 for (i
= 0; i
< oldnvdevs
; i
++) {
1918 ASSERT(vd
->vdev_isl2cache
);
1920 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1921 pool
!= 0ULL && l2arc_vdev_present(vd
))
1922 l2arc_remove_vdev(vd
);
1923 vdev_clear_stats(vd
);
1929 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1931 for (i
= 0; i
< sav
->sav_count
; i
++)
1932 nvlist_free(l2cache
[i
]);
1934 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1938 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1941 char *packed
= NULL
;
1946 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1950 nvsize
= *(uint64_t *)db
->db_data
;
1951 dmu_buf_rele(db
, FTAG
);
1953 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1954 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1957 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1958 vmem_free(packed
, nvsize
);
1964 * Concrete top-level vdevs that are not missing and are not logs. At every
1965 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1968 spa_healthy_core_tvds(spa_t
*spa
)
1970 vdev_t
*rvd
= spa
->spa_root_vdev
;
1973 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
1974 vdev_t
*vd
= rvd
->vdev_child
[i
];
1977 if (vdev_is_concrete(vd
) && !vdev_is_dead(vd
))
1985 * Checks to see if the given vdev could not be opened, in which case we post a
1986 * sysevent to notify the autoreplace code that the device has been removed.
1989 spa_check_removed(vdev_t
*vd
)
1991 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1992 spa_check_removed(vd
->vdev_child
[c
]);
1994 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1995 vdev_is_concrete(vd
)) {
1996 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1997 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
2002 spa_check_for_missing_logs(spa_t
*spa
)
2004 vdev_t
*rvd
= spa
->spa_root_vdev
;
2007 * If we're doing a normal import, then build up any additional
2008 * diagnostic information about missing log devices.
2009 * We'll pass this up to the user for further processing.
2011 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
2012 nvlist_t
**child
, *nv
;
2015 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
2017 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2019 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2020 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2023 * We consider a device as missing only if it failed
2024 * to open (i.e. offline or faulted is not considered
2027 if (tvd
->vdev_islog
&&
2028 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2029 child
[idx
++] = vdev_config_generate(spa
, tvd
,
2030 B_FALSE
, VDEV_CONFIG_MISSING
);
2035 fnvlist_add_nvlist_array(nv
,
2036 ZPOOL_CONFIG_CHILDREN
, child
, idx
);
2037 fnvlist_add_nvlist(spa
->spa_load_info
,
2038 ZPOOL_CONFIG_MISSING_DEVICES
, nv
);
2040 for (uint64_t i
= 0; i
< idx
; i
++)
2041 nvlist_free(child
[i
]);
2044 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
2047 spa_load_failed(spa
, "some log devices are missing");
2048 vdev_dbgmsg_print_tree(rvd
, 2);
2049 return (SET_ERROR(ENXIO
));
2052 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2053 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2055 if (tvd
->vdev_islog
&&
2056 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2057 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2058 spa_load_note(spa
, "some log devices are "
2059 "missing, ZIL is dropped.");
2060 vdev_dbgmsg_print_tree(rvd
, 2);
2070 * Check for missing log devices
2073 spa_check_logs(spa_t
*spa
)
2075 boolean_t rv
= B_FALSE
;
2076 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2078 switch (spa
->spa_log_state
) {
2081 case SPA_LOG_MISSING
:
2082 /* need to recheck in case slog has been restored */
2083 case SPA_LOG_UNKNOWN
:
2084 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
2085 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
2087 spa_set_log_state(spa
, SPA_LOG_MISSING
);
2094 spa_passivate_log(spa_t
*spa
)
2096 vdev_t
*rvd
= spa
->spa_root_vdev
;
2097 boolean_t slog_found
= B_FALSE
;
2099 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2101 if (!spa_has_slogs(spa
))
2104 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2105 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2106 metaslab_group_t
*mg
= tvd
->vdev_mg
;
2108 if (tvd
->vdev_islog
) {
2109 metaslab_group_passivate(mg
);
2110 slog_found
= B_TRUE
;
2114 return (slog_found
);
2118 spa_activate_log(spa_t
*spa
)
2120 vdev_t
*rvd
= spa
->spa_root_vdev
;
2122 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2124 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2125 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2126 metaslab_group_t
*mg
= tvd
->vdev_mg
;
2128 if (tvd
->vdev_islog
)
2129 metaslab_group_activate(mg
);
2134 spa_reset_logs(spa_t
*spa
)
2138 error
= dmu_objset_find(spa_name(spa
), zil_reset
,
2139 NULL
, DS_FIND_CHILDREN
);
2142 * We successfully offlined the log device, sync out the
2143 * current txg so that the "stubby" block can be removed
2146 txg_wait_synced(spa
->spa_dsl_pool
, 0);
2152 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
2154 for (int i
= 0; i
< sav
->sav_count
; i
++)
2155 spa_check_removed(sav
->sav_vdevs
[i
]);
2159 spa_claim_notify(zio_t
*zio
)
2161 spa_t
*spa
= zio
->io_spa
;
2166 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
2167 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
2168 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
2169 mutex_exit(&spa
->spa_props_lock
);
2172 typedef struct spa_load_error
{
2173 uint64_t sle_meta_count
;
2174 uint64_t sle_data_count
;
2178 spa_load_verify_done(zio_t
*zio
)
2180 blkptr_t
*bp
= zio
->io_bp
;
2181 spa_load_error_t
*sle
= zio
->io_private
;
2182 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2183 int error
= zio
->io_error
;
2184 spa_t
*spa
= zio
->io_spa
;
2186 abd_free(zio
->io_abd
);
2188 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2189 type
!= DMU_OT_INTENT_LOG
)
2190 atomic_inc_64(&sle
->sle_meta_count
);
2192 atomic_inc_64(&sle
->sle_data_count
);
2195 mutex_enter(&spa
->spa_scrub_lock
);
2196 spa
->spa_load_verify_bytes
-= BP_GET_PSIZE(bp
);
2197 cv_broadcast(&spa
->spa_scrub_io_cv
);
2198 mutex_exit(&spa
->spa_scrub_lock
);
2202 * Maximum number of inflight bytes is the log2 fraction of the arc size.
2203 * By default, we set it to 1/16th of the arc.
2205 int spa_load_verify_shift
= 4;
2206 int spa_load_verify_metadata
= B_TRUE
;
2207 int spa_load_verify_data
= B_TRUE
;
2211 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2212 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2214 if (zb
->zb_level
== ZB_DNODE_LEVEL
|| BP_IS_HOLE(bp
) ||
2215 BP_IS_EMBEDDED(bp
) || BP_IS_REDACTED(bp
))
2218 * Note: normally this routine will not be called if
2219 * spa_load_verify_metadata is not set. However, it may be useful
2220 * to manually set the flag after the traversal has begun.
2222 if (!spa_load_verify_metadata
)
2224 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2227 uint64_t maxinflight_bytes
=
2228 arc_target_bytes() >> spa_load_verify_shift
;
2230 size_t size
= BP_GET_PSIZE(bp
);
2232 mutex_enter(&spa
->spa_scrub_lock
);
2233 while (spa
->spa_load_verify_bytes
>= maxinflight_bytes
)
2234 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2235 spa
->spa_load_verify_bytes
+= size
;
2236 mutex_exit(&spa
->spa_scrub_lock
);
2238 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2239 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2240 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2241 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2247 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2249 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2250 return (SET_ERROR(ENAMETOOLONG
));
2256 spa_load_verify(spa_t
*spa
)
2259 spa_load_error_t sle
= { 0 };
2260 zpool_load_policy_t policy
;
2261 boolean_t verify_ok
= B_FALSE
;
2264 zpool_get_load_policy(spa
->spa_config
, &policy
);
2266 if (policy
.zlp_rewind
& ZPOOL_NEVER_REWIND
)
2269 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2270 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2271 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2273 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2277 rio
= zio_root(spa
, NULL
, &sle
,
2278 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2280 if (spa_load_verify_metadata
) {
2281 if (spa
->spa_extreme_rewind
) {
2282 spa_load_note(spa
, "performing a complete scan of the "
2283 "pool since extreme rewind is on. This may take "
2284 "a very long time.\n (spa_load_verify_data=%u, "
2285 "spa_load_verify_metadata=%u)",
2286 spa_load_verify_data
, spa_load_verify_metadata
);
2289 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2290 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2291 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2294 (void) zio_wait(rio
);
2295 ASSERT0(spa
->spa_load_verify_bytes
);
2297 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2298 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2300 if (sle
.sle_meta_count
!= 0 || sle
.sle_data_count
!= 0) {
2301 spa_load_note(spa
, "spa_load_verify found %llu metadata errors "
2302 "and %llu data errors", (u_longlong_t
)sle
.sle_meta_count
,
2303 (u_longlong_t
)sle
.sle_data_count
);
2306 if (spa_load_verify_dryrun
||
2307 (!error
&& sle
.sle_meta_count
<= policy
.zlp_maxmeta
&&
2308 sle
.sle_data_count
<= policy
.zlp_maxdata
)) {
2312 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2313 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2315 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2316 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2317 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2318 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2319 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2320 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2321 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2323 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2326 if (spa_load_verify_dryrun
)
2330 if (error
!= ENXIO
&& error
!= EIO
)
2331 error
= SET_ERROR(EIO
);
2335 return (verify_ok
? 0 : EIO
);
2339 * Find a value in the pool props object.
2342 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2344 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2345 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2349 * Find a value in the pool directory object.
2352 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
, boolean_t log_enoent
)
2354 int error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2355 name
, sizeof (uint64_t), 1, val
);
2357 if (error
!= 0 && (error
!= ENOENT
|| log_enoent
)) {
2358 spa_load_failed(spa
, "couldn't get '%s' value in MOS directory "
2359 "[error=%d]", name
, error
);
2366 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2368 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2369 return (SET_ERROR(err
));
2373 spa_livelist_delete_check(spa_t
*spa
)
2375 return (spa
->spa_livelists_to_delete
!= 0);
2380 spa_livelist_delete_cb_check(void *arg
, zthr_t
*z
)
2383 return (spa_livelist_delete_check(spa
));
2387 delete_blkptr_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
2390 zio_free(spa
, tx
->tx_txg
, bp
);
2391 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
, DD_USED_HEAD
,
2392 -bp_get_dsize_sync(spa
, bp
),
2393 -BP_GET_PSIZE(bp
), -BP_GET_UCSIZE(bp
), tx
);
2398 dsl_get_next_livelist_obj(objset_t
*os
, uint64_t zap_obj
, uint64_t *llp
)
2403 zap_cursor_init(&zc
, os
, zap_obj
);
2404 err
= zap_cursor_retrieve(&zc
, &za
);
2405 zap_cursor_fini(&zc
);
2407 *llp
= za
.za_first_integer
;
2412 * Components of livelist deletion that must be performed in syncing
2413 * context: freeing block pointers and updating the pool-wide data
2414 * structures to indicate how much work is left to do
2416 typedef struct sublist_delete_arg
{
2421 } sublist_delete_arg_t
;
2424 sublist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2426 sublist_delete_arg_t
*sda
= arg
;
2427 spa_t
*spa
= sda
->spa
;
2428 dsl_deadlist_t
*ll
= sda
->ll
;
2429 uint64_t key
= sda
->key
;
2430 bplist_t
*to_free
= sda
->to_free
;
2432 bplist_iterate(to_free
, delete_blkptr_cb
, spa
, tx
);
2433 dsl_deadlist_remove_entry(ll
, key
, tx
);
2436 typedef struct livelist_delete_arg
{
2440 } livelist_delete_arg_t
;
2443 livelist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2445 livelist_delete_arg_t
*lda
= arg
;
2446 spa_t
*spa
= lda
->spa
;
2447 uint64_t ll_obj
= lda
->ll_obj
;
2448 uint64_t zap_obj
= lda
->zap_obj
;
2449 objset_t
*mos
= spa
->spa_meta_objset
;
2452 /* free the livelist and decrement the feature count */
2453 VERIFY0(zap_remove_int(mos
, zap_obj
, ll_obj
, tx
));
2454 dsl_deadlist_free(mos
, ll_obj
, tx
);
2455 spa_feature_decr(spa
, SPA_FEATURE_LIVELIST
, tx
);
2456 VERIFY0(zap_count(mos
, zap_obj
, &count
));
2458 /* no more livelists to delete */
2459 VERIFY0(zap_remove(mos
, DMU_POOL_DIRECTORY_OBJECT
,
2460 DMU_POOL_DELETED_CLONES
, tx
));
2461 VERIFY0(zap_destroy(mos
, zap_obj
, tx
));
2462 spa
->spa_livelists_to_delete
= 0;
2463 spa_notify_waiters(spa
);
2468 * Load in the value for the livelist to be removed and open it. Then,
2469 * load its first sublist and determine which block pointers should actually
2470 * be freed. Then, call a synctask which performs the actual frees and updates
2471 * the pool-wide livelist data.
2475 spa_livelist_delete_cb(void *arg
, zthr_t
*z
)
2478 uint64_t ll_obj
= 0, count
;
2479 objset_t
*mos
= spa
->spa_meta_objset
;
2480 uint64_t zap_obj
= spa
->spa_livelists_to_delete
;
2482 * Determine the next livelist to delete. This function should only
2483 * be called if there is at least one deleted clone.
2485 VERIFY0(dsl_get_next_livelist_obj(mos
, zap_obj
, &ll_obj
));
2486 VERIFY0(zap_count(mos
, ll_obj
, &count
));
2488 dsl_deadlist_t ll
= { 0 };
2489 dsl_deadlist_entry_t
*dle
;
2491 dsl_deadlist_open(&ll
, mos
, ll_obj
);
2492 dle
= dsl_deadlist_first(&ll
);
2493 ASSERT3P(dle
, !=, NULL
);
2494 bplist_create(&to_free
);
2495 int err
= dsl_process_sub_livelist(&dle
->dle_bpobj
, &to_free
,
2498 sublist_delete_arg_t sync_arg
= {
2501 .key
= dle
->dle_mintxg
,
2504 zfs_dbgmsg("deleting sublist (id %llu) from"
2505 " livelist %llu, %d remaining",
2506 dle
->dle_bpobj
.bpo_object
, ll_obj
, count
- 1);
2507 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
,
2508 sublist_delete_sync
, &sync_arg
, 0,
2509 ZFS_SPACE_CHECK_DESTROY
));
2511 VERIFY3U(err
, ==, EINTR
);
2513 bplist_clear(&to_free
);
2514 bplist_destroy(&to_free
);
2515 dsl_deadlist_close(&ll
);
2517 livelist_delete_arg_t sync_arg
= {
2522 zfs_dbgmsg("deletion of livelist %llu completed", ll_obj
);
2523 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
, livelist_delete_sync
,
2524 &sync_arg
, 0, ZFS_SPACE_CHECK_DESTROY
));
2529 spa_start_livelist_destroy_thread(spa_t
*spa
)
2531 ASSERT3P(spa
->spa_livelist_delete_zthr
, ==, NULL
);
2532 spa
->spa_livelist_delete_zthr
=
2533 zthr_create("z_livelist_destroy",
2534 spa_livelist_delete_cb_check
, spa_livelist_delete_cb
, spa
);
2537 typedef struct livelist_new_arg
{
2540 } livelist_new_arg_t
;
2543 livelist_track_new_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
2547 livelist_new_arg_t
*lna
= arg
;
2549 bplist_append(lna
->frees
, bp
);
2551 bplist_append(lna
->allocs
, bp
);
2552 zfs_livelist_condense_new_alloc
++;
2557 typedef struct livelist_condense_arg
{
2560 uint64_t first_size
;
2562 } livelist_condense_arg_t
;
2565 spa_livelist_condense_sync(void *arg
, dmu_tx_t
*tx
)
2567 livelist_condense_arg_t
*lca
= arg
;
2568 spa_t
*spa
= lca
->spa
;
2570 dsl_dataset_t
*ds
= spa
->spa_to_condense
.ds
;
2572 /* Have we been cancelled? */
2573 if (spa
->spa_to_condense
.cancelled
) {
2574 zfs_livelist_condense_sync_cancel
++;
2578 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
2579 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
2580 dsl_deadlist_t
*ll
= &ds
->ds_dir
->dd_livelist
;
2583 * It's possible that the livelist was changed while the zthr was
2584 * running. Therefore, we need to check for new blkptrs in the two
2585 * entries being condensed and continue to track them in the livelist.
2586 * Because of the way we handle remapped blkptrs (see dbuf_remap_impl),
2587 * it's possible that the newly added blkptrs are FREEs or ALLOCs so
2588 * we need to sort them into two different bplists.
2590 uint64_t first_obj
= first
->dle_bpobj
.bpo_object
;
2591 uint64_t next_obj
= next
->dle_bpobj
.bpo_object
;
2592 uint64_t cur_first_size
= first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
2593 uint64_t cur_next_size
= next
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
2595 bplist_create(&new_frees
);
2596 livelist_new_arg_t new_bps
= {
2597 .allocs
= &lca
->to_keep
,
2598 .frees
= &new_frees
,
2601 if (cur_first_size
> lca
->first_size
) {
2602 VERIFY0(livelist_bpobj_iterate_from_nofree(&first
->dle_bpobj
,
2603 livelist_track_new_cb
, &new_bps
, lca
->first_size
));
2605 if (cur_next_size
> lca
->next_size
) {
2606 VERIFY0(livelist_bpobj_iterate_from_nofree(&next
->dle_bpobj
,
2607 livelist_track_new_cb
, &new_bps
, lca
->next_size
));
2610 dsl_deadlist_clear_entry(first
, ll
, tx
);
2611 ASSERT(bpobj_is_empty(&first
->dle_bpobj
));
2612 dsl_deadlist_remove_entry(ll
, next
->dle_mintxg
, tx
);
2614 bplist_iterate(&lca
->to_keep
, dsl_deadlist_insert_alloc_cb
, ll
, tx
);
2615 bplist_iterate(&new_frees
, dsl_deadlist_insert_free_cb
, ll
, tx
);
2616 bplist_destroy(&new_frees
);
2618 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
2619 dsl_dataset_name(ds
, dsname
);
2620 zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu "
2621 "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu "
2622 "(%llu blkptrs)", tx
->tx_txg
, dsname
, ds
->ds_object
, first_obj
,
2623 cur_first_size
, next_obj
, cur_next_size
,
2624 first
->dle_bpobj
.bpo_object
,
2625 first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
);
2627 dmu_buf_rele(ds
->ds_dbuf
, spa
);
2628 spa
->spa_to_condense
.ds
= NULL
;
2629 bplist_clear(&lca
->to_keep
);
2630 bplist_destroy(&lca
->to_keep
);
2631 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
2632 spa
->spa_to_condense
.syncing
= B_FALSE
;
2636 spa_livelist_condense_cb(void *arg
, zthr_t
*t
)
2638 while (zfs_livelist_condense_zthr_pause
&&
2639 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
2643 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
2644 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
2645 uint64_t first_size
, next_size
;
2647 livelist_condense_arg_t
*lca
=
2648 kmem_alloc(sizeof (livelist_condense_arg_t
), KM_SLEEP
);
2649 bplist_create(&lca
->to_keep
);
2652 * Process the livelists (matching FREEs and ALLOCs) in open context
2653 * so we have minimal work in syncing context to condense.
2655 * We save bpobj sizes (first_size and next_size) to use later in
2656 * syncing context to determine if entries were added to these sublists
2657 * while in open context. This is possible because the clone is still
2658 * active and open for normal writes and we want to make sure the new,
2659 * unprocessed blockpointers are inserted into the livelist normally.
2661 * Note that dsl_process_sub_livelist() both stores the size number of
2662 * blockpointers and iterates over them while the bpobj's lock held, so
2663 * the sizes returned to us are consistent which what was actually
2666 int err
= dsl_process_sub_livelist(&first
->dle_bpobj
, &lca
->to_keep
, t
,
2669 err
= dsl_process_sub_livelist(&next
->dle_bpobj
, &lca
->to_keep
,
2673 while (zfs_livelist_condense_sync_pause
&&
2674 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
2677 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
2678 dmu_tx_mark_netfree(tx
);
2679 dmu_tx_hold_space(tx
, 1);
2680 err
= dmu_tx_assign(tx
, TXG_NOWAIT
| TXG_NOTHROTTLE
);
2683 * Prevent the condense zthr restarting before
2684 * the synctask completes.
2686 spa
->spa_to_condense
.syncing
= B_TRUE
;
2688 lca
->first_size
= first_size
;
2689 lca
->next_size
= next_size
;
2690 dsl_sync_task_nowait(spa_get_dsl(spa
),
2691 spa_livelist_condense_sync
, lca
, tx
);
2697 * Condensing can not continue: either it was externally stopped or
2698 * we were unable to assign to a tx because the pool has run out of
2699 * space. In the second case, we'll just end up trying to condense
2700 * again in a later txg.
2703 bplist_clear(&lca
->to_keep
);
2704 bplist_destroy(&lca
->to_keep
);
2705 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
2706 dmu_buf_rele(spa
->spa_to_condense
.ds
->ds_dbuf
, spa
);
2707 spa
->spa_to_condense
.ds
= NULL
;
2709 zfs_livelist_condense_zthr_cancel
++;
2714 * Check that there is something to condense but that a condense is not
2715 * already in progress and that condensing has not been cancelled.
2718 spa_livelist_condense_cb_check(void *arg
, zthr_t
*z
)
2721 if ((spa
->spa_to_condense
.ds
!= NULL
) &&
2722 (spa
->spa_to_condense
.syncing
== B_FALSE
) &&
2723 (spa
->spa_to_condense
.cancelled
== B_FALSE
)) {
2730 spa_start_livelist_condensing_thread(spa_t
*spa
)
2732 spa
->spa_to_condense
.ds
= NULL
;
2733 spa
->spa_to_condense
.first
= NULL
;
2734 spa
->spa_to_condense
.next
= NULL
;
2735 spa
->spa_to_condense
.syncing
= B_FALSE
;
2736 spa
->spa_to_condense
.cancelled
= B_FALSE
;
2738 ASSERT3P(spa
->spa_livelist_condense_zthr
, ==, NULL
);
2739 spa
->spa_livelist_condense_zthr
=
2740 zthr_create("z_livelist_condense",
2741 spa_livelist_condense_cb_check
,
2742 spa_livelist_condense_cb
, spa
);
2746 spa_spawn_aux_threads(spa_t
*spa
)
2748 ASSERT(spa_writeable(spa
));
2750 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2752 spa_start_indirect_condensing_thread(spa
);
2753 spa_start_livelist_destroy_thread(spa
);
2754 spa_start_livelist_condensing_thread(spa
);
2756 ASSERT3P(spa
->spa_checkpoint_discard_zthr
, ==, NULL
);
2757 spa
->spa_checkpoint_discard_zthr
=
2758 zthr_create("z_checkpoint_discard",
2759 spa_checkpoint_discard_thread_check
,
2760 spa_checkpoint_discard_thread
, spa
);
2764 * Fix up config after a partly-completed split. This is done with the
2765 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2766 * pool have that entry in their config, but only the splitting one contains
2767 * a list of all the guids of the vdevs that are being split off.
2769 * This function determines what to do with that list: either rejoin
2770 * all the disks to the pool, or complete the splitting process. To attempt
2771 * the rejoin, each disk that is offlined is marked online again, and
2772 * we do a reopen() call. If the vdev label for every disk that was
2773 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2774 * then we call vdev_split() on each disk, and complete the split.
2776 * Otherwise we leave the config alone, with all the vdevs in place in
2777 * the original pool.
2780 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2787 boolean_t attempt_reopen
;
2789 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2792 /* check that the config is complete */
2793 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2794 &glist
, &gcount
) != 0)
2797 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2799 /* attempt to online all the vdevs & validate */
2800 attempt_reopen
= B_TRUE
;
2801 for (i
= 0; i
< gcount
; i
++) {
2802 if (glist
[i
] == 0) /* vdev is hole */
2805 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2806 if (vd
[i
] == NULL
) {
2808 * Don't bother attempting to reopen the disks;
2809 * just do the split.
2811 attempt_reopen
= B_FALSE
;
2813 /* attempt to re-online it */
2814 vd
[i
]->vdev_offline
= B_FALSE
;
2818 if (attempt_reopen
) {
2819 vdev_reopen(spa
->spa_root_vdev
);
2821 /* check each device to see what state it's in */
2822 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2823 if (vd
[i
] != NULL
&&
2824 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2831 * If every disk has been moved to the new pool, or if we never
2832 * even attempted to look at them, then we split them off for
2835 if (!attempt_reopen
|| gcount
== extracted
) {
2836 for (i
= 0; i
< gcount
; i
++)
2839 vdev_reopen(spa
->spa_root_vdev
);
2842 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2846 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
)
2848 char *ereport
= FM_EREPORT_ZFS_POOL
;
2851 spa
->spa_load_state
= state
;
2852 (void) spa_import_progress_set_state(spa_guid(spa
),
2853 spa_load_state(spa
));
2855 gethrestime(&spa
->spa_loaded_ts
);
2856 error
= spa_load_impl(spa
, type
, &ereport
);
2859 * Don't count references from objsets that are already closed
2860 * and are making their way through the eviction process.
2862 spa_evicting_os_wait(spa
);
2863 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
2865 if (error
!= EEXIST
) {
2866 spa
->spa_loaded_ts
.tv_sec
= 0;
2867 spa
->spa_loaded_ts
.tv_nsec
= 0;
2869 if (error
!= EBADF
) {
2870 (void) zfs_ereport_post(ereport
, spa
,
2871 NULL
, NULL
, NULL
, 0);
2874 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2877 (void) spa_import_progress_set_state(spa_guid(spa
),
2878 spa_load_state(spa
));
2885 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2886 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2887 * spa's per-vdev ZAP list.
2890 vdev_count_verify_zaps(vdev_t
*vd
)
2892 spa_t
*spa
= vd
->vdev_spa
;
2895 if (vd
->vdev_top_zap
!= 0) {
2897 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2898 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2900 if (vd
->vdev_leaf_zap
!= 0) {
2902 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2903 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2906 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2907 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2915 * Determine whether the activity check is required.
2918 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
2922 uint64_t hostid
= 0;
2923 uint64_t tryconfig_txg
= 0;
2924 uint64_t tryconfig_timestamp
= 0;
2925 uint16_t tryconfig_mmp_seq
= 0;
2928 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2929 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2930 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2932 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2933 &tryconfig_timestamp
);
2934 (void) nvlist_lookup_uint16(nvinfo
, ZPOOL_CONFIG_MMP_SEQ
,
2935 &tryconfig_mmp_seq
);
2938 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2941 * Disable the MMP activity check - This is used by zdb which
2942 * is intended to be used on potentially active pools.
2944 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
2948 * Skip the activity check when the MMP feature is disabled.
2950 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
2954 * If the tryconfig_ values are nonzero, they are the results of an
2955 * earlier tryimport. If they all match the uberblock we just found,
2956 * then the pool has not changed and we return false so we do not test
2959 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
2960 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
&&
2961 tryconfig_mmp_seq
&& tryconfig_mmp_seq
==
2962 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0))
2966 * Allow the activity check to be skipped when importing the pool
2967 * on the same host which last imported it. Since the hostid from
2968 * configuration may be stale use the one read from the label.
2970 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
2971 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
2973 if (hostid
== spa_get_hostid(spa
))
2977 * Skip the activity test when the pool was cleanly exported.
2979 if (state
!= POOL_STATE_ACTIVE
)
2986 * Nanoseconds the activity check must watch for changes on-disk.
2989 spa_activity_check_duration(spa_t
*spa
, uberblock_t
*ub
)
2991 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2992 uint64_t multihost_interval
= MSEC2NSEC(
2993 MMP_INTERVAL_OK(zfs_multihost_interval
));
2994 uint64_t import_delay
= MAX(NANOSEC
, import_intervals
*
2995 multihost_interval
);
2998 * Local tunables determine a minimum duration except for the case
2999 * where we know when the remote host will suspend the pool if MMP
3000 * writes do not land.
3002 * See Big Theory comment at the top of mmp.c for the reasoning behind
3003 * these cases and times.
3006 ASSERT(MMP_IMPORT_SAFETY_FACTOR
>= 100);
3008 if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3009 MMP_FAIL_INT(ub
) > 0) {
3011 /* MMP on remote host will suspend pool after failed writes */
3012 import_delay
= MMP_FAIL_INT(ub
) * MSEC2NSEC(MMP_INTERVAL(ub
)) *
3013 MMP_IMPORT_SAFETY_FACTOR
/ 100;
3015 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3016 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3017 "import_intervals=%u", import_delay
, MMP_FAIL_INT(ub
),
3018 MMP_INTERVAL(ub
), import_intervals
);
3020 } else if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3021 MMP_FAIL_INT(ub
) == 0) {
3023 /* MMP on remote host will never suspend pool */
3024 import_delay
= MAX(import_delay
, (MSEC2NSEC(MMP_INTERVAL(ub
)) +
3025 ub
->ub_mmp_delay
) * import_intervals
);
3027 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3028 "mmp_interval=%llu ub_mmp_delay=%llu "
3029 "import_intervals=%u", import_delay
, MMP_INTERVAL(ub
),
3030 ub
->ub_mmp_delay
, import_intervals
);
3032 } else if (MMP_VALID(ub
)) {
3034 * zfs-0.7 compatibility case
3037 import_delay
= MAX(import_delay
, (multihost_interval
+
3038 ub
->ub_mmp_delay
) * import_intervals
);
3040 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3041 "import_intervals=%u leaves=%u", import_delay
,
3042 ub
->ub_mmp_delay
, import_intervals
,
3043 vdev_count_leaves(spa
));
3045 /* Using local tunings is the only reasonable option */
3046 zfs_dbgmsg("pool last imported on non-MMP aware "
3047 "host using import_delay=%llu multihost_interval=%llu "
3048 "import_intervals=%u", import_delay
, multihost_interval
,
3052 return (import_delay
);
3056 * Perform the import activity check. If the user canceled the import or
3057 * we detected activity then fail.
3060 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
3062 uint64_t txg
= ub
->ub_txg
;
3063 uint64_t timestamp
= ub
->ub_timestamp
;
3064 uint64_t mmp_config
= ub
->ub_mmp_config
;
3065 uint16_t mmp_seq
= MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0;
3066 uint64_t import_delay
;
3067 hrtime_t import_expire
;
3068 nvlist_t
*mmp_label
= NULL
;
3069 vdev_t
*rvd
= spa
->spa_root_vdev
;
3074 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
3075 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3079 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3080 * during the earlier tryimport. If the txg recorded there is 0 then
3081 * the pool is known to be active on another host.
3083 * Otherwise, the pool might be in use on another host. Check for
3084 * changes in the uberblocks on disk if necessary.
3086 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3087 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
3088 ZPOOL_CONFIG_LOAD_INFO
);
3090 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
3091 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
3092 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3093 error
= SET_ERROR(EREMOTEIO
);
3098 import_delay
= spa_activity_check_duration(spa
, ub
);
3100 /* Add a small random factor in case of simultaneous imports (0-25%) */
3101 import_delay
+= import_delay
* spa_get_random(250) / 1000;
3103 import_expire
= gethrtime() + import_delay
;
3105 while (gethrtime() < import_expire
) {
3106 (void) spa_import_progress_set_mmp_check(spa_guid(spa
),
3107 NSEC2SEC(import_expire
- gethrtime()));
3109 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3111 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
||
3112 mmp_seq
!= (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0)) {
3113 zfs_dbgmsg("multihost activity detected "
3114 "txg %llu ub_txg %llu "
3115 "timestamp %llu ub_timestamp %llu "
3116 "mmp_config %#llx ub_mmp_config %#llx",
3117 txg
, ub
->ub_txg
, timestamp
, ub
->ub_timestamp
,
3118 mmp_config
, ub
->ub_mmp_config
);
3120 error
= SET_ERROR(EREMOTEIO
);
3125 nvlist_free(mmp_label
);
3129 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
3131 error
= SET_ERROR(EINTR
);
3139 mutex_destroy(&mtx
);
3143 * If the pool is determined to be active store the status in the
3144 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3145 * available from configuration read from disk store them as well.
3146 * This allows 'zpool import' to generate a more useful message.
3148 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3149 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3150 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3152 if (error
== EREMOTEIO
) {
3153 char *hostname
= "<unknown>";
3154 uint64_t hostid
= 0;
3157 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
3158 hostname
= fnvlist_lookup_string(mmp_label
,
3159 ZPOOL_CONFIG_HOSTNAME
);
3160 fnvlist_add_string(spa
->spa_load_info
,
3161 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
3164 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
3165 hostid
= fnvlist_lookup_uint64(mmp_label
,
3166 ZPOOL_CONFIG_HOSTID
);
3167 fnvlist_add_uint64(spa
->spa_load_info
,
3168 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
3172 fnvlist_add_uint64(spa
->spa_load_info
,
3173 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
3174 fnvlist_add_uint64(spa
->spa_load_info
,
3175 ZPOOL_CONFIG_MMP_TXG
, 0);
3177 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
3181 nvlist_free(mmp_label
);
3187 spa_verify_host(spa_t
*spa
, nvlist_t
*mos_config
)
3191 uint64_t myhostid
= 0;
3193 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
3194 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
3195 hostname
= fnvlist_lookup_string(mos_config
,
3196 ZPOOL_CONFIG_HOSTNAME
);
3198 myhostid
= zone_get_hostid(NULL
);
3200 if (hostid
!= 0 && myhostid
!= 0 && hostid
!= myhostid
) {
3201 cmn_err(CE_WARN
, "pool '%s' could not be "
3202 "loaded as it was last accessed by "
3203 "another system (host: %s hostid: 0x%llx). "
3204 "See: https://openzfs.github.io/openzfs-docs/msg/"
3206 spa_name(spa
), hostname
, (u_longlong_t
)hostid
);
3207 spa_load_failed(spa
, "hostid verification failed: pool "
3208 "last accessed by host: %s (hostid: 0x%llx)",
3209 hostname
, (u_longlong_t
)hostid
);
3210 return (SET_ERROR(EBADF
));
3218 spa_ld_parse_config(spa_t
*spa
, spa_import_type_t type
)
3221 nvlist_t
*nvtree
, *nvl
, *config
= spa
->spa_config
;
3228 * Versioning wasn't explicitly added to the label until later, so if
3229 * it's not present treat it as the initial version.
3231 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
3232 &spa
->spa_ubsync
.ub_version
) != 0)
3233 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
3235 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
3236 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3237 ZPOOL_CONFIG_POOL_GUID
);
3238 return (SET_ERROR(EINVAL
));
3242 * If we are doing an import, ensure that the pool is not already
3243 * imported by checking if its pool guid already exists in the
3246 * The only case that we allow an already imported pool to be
3247 * imported again, is when the pool is checkpointed and we want to
3248 * look at its checkpointed state from userland tools like zdb.
3251 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3252 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3253 spa_guid_exists(pool_guid
, 0)) {
3255 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3256 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3257 spa_guid_exists(pool_guid
, 0) &&
3258 !spa_importing_readonly_checkpoint(spa
)) {
3260 spa_load_failed(spa
, "a pool with guid %llu is already open",
3261 (u_longlong_t
)pool_guid
);
3262 return (SET_ERROR(EEXIST
));
3265 spa
->spa_config_guid
= pool_guid
;
3267 nvlist_free(spa
->spa_load_info
);
3268 spa
->spa_load_info
= fnvlist_alloc();
3270 ASSERT(spa
->spa_comment
== NULL
);
3271 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
3272 spa
->spa_comment
= spa_strdup(comment
);
3274 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
3275 &spa
->spa_config_txg
);
3277 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) == 0)
3278 spa
->spa_config_splitting
= fnvlist_dup(nvl
);
3280 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
)) {
3281 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3282 ZPOOL_CONFIG_VDEV_TREE
);
3283 return (SET_ERROR(EINVAL
));
3287 * Create "The Godfather" zio to hold all async IOs
3289 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3291 for (int i
= 0; i
< max_ncpus
; i
++) {
3292 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3293 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3294 ZIO_FLAG_GODFATHER
);
3298 * Parse the configuration into a vdev tree. We explicitly set the
3299 * value that will be returned by spa_version() since parsing the
3300 * configuration requires knowing the version number.
3302 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3303 parse
= (type
== SPA_IMPORT_EXISTING
?
3304 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
3305 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
3306 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3309 spa_load_failed(spa
, "unable to parse config [error=%d]",
3314 ASSERT(spa
->spa_root_vdev
== rvd
);
3315 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
3316 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
3318 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3319 ASSERT(spa_guid(spa
) == pool_guid
);
3326 * Recursively open all vdevs in the vdev tree. This function is called twice:
3327 * first with the untrusted config, then with the trusted config.
3330 spa_ld_open_vdevs(spa_t
*spa
)
3335 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3336 * missing/unopenable for the root vdev to be still considered openable.
3338 if (spa
->spa_trust_config
) {
3339 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds
;
3340 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_CACHEFILE
) {
3341 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_cachefile
;
3342 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_SCAN
) {
3343 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_scan
;
3345 spa
->spa_missing_tvds_allowed
= 0;
3348 spa
->spa_missing_tvds_allowed
=
3349 MAX(zfs_max_missing_tvds
, spa
->spa_missing_tvds_allowed
);
3351 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3352 error
= vdev_open(spa
->spa_root_vdev
);
3353 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3355 if (spa
->spa_missing_tvds
!= 0) {
3356 spa_load_note(spa
, "vdev tree has %lld missing top-level "
3357 "vdevs.", (u_longlong_t
)spa
->spa_missing_tvds
);
3358 if (spa
->spa_trust_config
&& (spa
->spa_mode
& SPA_MODE_WRITE
)) {
3360 * Although theoretically we could allow users to open
3361 * incomplete pools in RW mode, we'd need to add a lot
3362 * of extra logic (e.g. adjust pool space to account
3363 * for missing vdevs).
3364 * This limitation also prevents users from accidentally
3365 * opening the pool in RW mode during data recovery and
3366 * damaging it further.
3368 spa_load_note(spa
, "pools with missing top-level "
3369 "vdevs can only be opened in read-only mode.");
3370 error
= SET_ERROR(ENXIO
);
3372 spa_load_note(spa
, "current settings allow for maximum "
3373 "%lld missing top-level vdevs at this stage.",
3374 (u_longlong_t
)spa
->spa_missing_tvds_allowed
);
3378 spa_load_failed(spa
, "unable to open vdev tree [error=%d]",
3381 if (spa
->spa_missing_tvds
!= 0 || error
!= 0)
3382 vdev_dbgmsg_print_tree(spa
->spa_root_vdev
, 2);
3388 * We need to validate the vdev labels against the configuration that
3389 * we have in hand. This function is called twice: first with an untrusted
3390 * config, then with a trusted config. The validation is more strict when the
3391 * config is trusted.
3394 spa_ld_validate_vdevs(spa_t
*spa
)
3397 vdev_t
*rvd
= spa
->spa_root_vdev
;
3399 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3400 error
= vdev_validate(rvd
);
3401 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3404 spa_load_failed(spa
, "vdev_validate failed [error=%d]", error
);
3408 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
3409 spa_load_failed(spa
, "cannot open vdev tree after invalidating "
3411 vdev_dbgmsg_print_tree(rvd
, 2);
3412 return (SET_ERROR(ENXIO
));
3419 spa_ld_select_uberblock_done(spa_t
*spa
, uberblock_t
*ub
)
3421 spa
->spa_state
= POOL_STATE_ACTIVE
;
3422 spa
->spa_ubsync
= spa
->spa_uberblock
;
3423 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
3424 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
3425 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
3426 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
3427 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
3428 spa
->spa_prev_software_version
= ub
->ub_software_version
;
3432 spa_ld_select_uberblock(spa_t
*spa
, spa_import_type_t type
)
3434 vdev_t
*rvd
= spa
->spa_root_vdev
;
3436 uberblock_t
*ub
= &spa
->spa_uberblock
;
3437 boolean_t activity_check
= B_FALSE
;
3440 * If we are opening the checkpointed state of the pool by
3441 * rewinding to it, at this point we will have written the
3442 * checkpointed uberblock to the vdev labels, so searching
3443 * the labels will find the right uberblock. However, if
3444 * we are opening the checkpointed state read-only, we have
3445 * not modified the labels. Therefore, we must ignore the
3446 * labels and continue using the spa_uberblock that was set
3447 * by spa_ld_checkpoint_rewind.
3449 * Note that it would be fine to ignore the labels when
3450 * rewinding (opening writeable) as well. However, if we
3451 * crash just after writing the labels, we will end up
3452 * searching the labels. Doing so in the common case means
3453 * that this code path gets exercised normally, rather than
3454 * just in the edge case.
3456 if (ub
->ub_checkpoint_txg
!= 0 &&
3457 spa_importing_readonly_checkpoint(spa
)) {
3458 spa_ld_select_uberblock_done(spa
, ub
);
3463 * Find the best uberblock.
3465 vdev_uberblock_load(rvd
, ub
, &label
);
3468 * If we weren't able to find a single valid uberblock, return failure.
3470 if (ub
->ub_txg
== 0) {
3472 spa_load_failed(spa
, "no valid uberblock found");
3473 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
3476 if (spa
->spa_load_max_txg
!= UINT64_MAX
) {
3477 (void) spa_import_progress_set_max_txg(spa_guid(spa
),
3478 (u_longlong_t
)spa
->spa_load_max_txg
);
3480 spa_load_note(spa
, "using uberblock with txg=%llu",
3481 (u_longlong_t
)ub
->ub_txg
);
3485 * For pools which have the multihost property on determine if the
3486 * pool is truly inactive and can be safely imported. Prevent
3487 * hosts which don't have a hostid set from importing the pool.
3489 activity_check
= spa_activity_check_required(spa
, ub
, label
,
3491 if (activity_check
) {
3492 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
3493 spa_get_hostid(spa
) == 0) {
3495 fnvlist_add_uint64(spa
->spa_load_info
,
3496 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3497 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3500 int error
= spa_activity_check(spa
, ub
, spa
->spa_config
);
3506 fnvlist_add_uint64(spa
->spa_load_info
,
3507 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
3508 fnvlist_add_uint64(spa
->spa_load_info
,
3509 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
3510 fnvlist_add_uint16(spa
->spa_load_info
,
3511 ZPOOL_CONFIG_MMP_SEQ
,
3512 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0));
3516 * If the pool has an unsupported version we can't open it.
3518 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
3520 spa_load_failed(spa
, "version %llu is not supported",
3521 (u_longlong_t
)ub
->ub_version
);
3522 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
3525 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3529 * If we weren't able to find what's necessary for reading the
3530 * MOS in the label, return failure.
3532 if (label
== NULL
) {
3533 spa_load_failed(spa
, "label config unavailable");
3534 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3538 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
3541 spa_load_failed(spa
, "invalid label: '%s' missing",
3542 ZPOOL_CONFIG_FEATURES_FOR_READ
);
3543 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3548 * Update our in-core representation with the definitive values
3551 nvlist_free(spa
->spa_label_features
);
3552 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
3558 * Look through entries in the label nvlist's features_for_read. If
3559 * there is a feature listed there which we don't understand then we
3560 * cannot open a pool.
3562 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3563 nvlist_t
*unsup_feat
;
3565 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
3568 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
3570 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
3571 if (!zfeature_is_supported(nvpair_name(nvp
))) {
3572 VERIFY(nvlist_add_string(unsup_feat
,
3573 nvpair_name(nvp
), "") == 0);
3577 if (!nvlist_empty(unsup_feat
)) {
3578 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
3579 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
3580 nvlist_free(unsup_feat
);
3581 spa_load_failed(spa
, "some features are unsupported");
3582 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
3586 nvlist_free(unsup_feat
);
3589 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
3590 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3591 spa_try_repair(spa
, spa
->spa_config
);
3592 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3593 nvlist_free(spa
->spa_config_splitting
);
3594 spa
->spa_config_splitting
= NULL
;
3598 * Initialize internal SPA structures.
3600 spa_ld_select_uberblock_done(spa
, ub
);
3606 spa_ld_open_rootbp(spa_t
*spa
)
3609 vdev_t
*rvd
= spa
->spa_root_vdev
;
3611 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
3613 spa_load_failed(spa
, "unable to open rootbp in dsl_pool_init "
3614 "[error=%d]", error
);
3615 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3617 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
3623 spa_ld_trusted_config(spa_t
*spa
, spa_import_type_t type
,
3624 boolean_t reloading
)
3626 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
3627 nvlist_t
*nv
, *mos_config
, *policy
;
3628 int error
= 0, copy_error
;
3629 uint64_t healthy_tvds
, healthy_tvds_mos
;
3630 uint64_t mos_config_txg
;
3632 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
, B_TRUE
)
3634 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3637 * If we're assembling a pool from a split, the config provided is
3638 * already trusted so there is nothing to do.
3640 if (type
== SPA_IMPORT_ASSEMBLE
)
3643 healthy_tvds
= spa_healthy_core_tvds(spa
);
3645 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
)
3647 spa_load_failed(spa
, "unable to retrieve MOS config");
3648 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3652 * If we are doing an open, pool owner wasn't verified yet, thus do
3653 * the verification here.
3655 if (spa
->spa_load_state
== SPA_LOAD_OPEN
) {
3656 error
= spa_verify_host(spa
, mos_config
);
3658 nvlist_free(mos_config
);
3663 nv
= fnvlist_lookup_nvlist(mos_config
, ZPOOL_CONFIG_VDEV_TREE
);
3665 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3668 * Build a new vdev tree from the trusted config
3670 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
3673 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3674 * obtained by scanning /dev/dsk, then it will have the right vdev
3675 * paths. We update the trusted MOS config with this information.
3676 * We first try to copy the paths with vdev_copy_path_strict, which
3677 * succeeds only when both configs have exactly the same vdev tree.
3678 * If that fails, we fall back to a more flexible method that has a
3679 * best effort policy.
3681 copy_error
= vdev_copy_path_strict(rvd
, mrvd
);
3682 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3683 spa_load_note(spa
, "provided vdev tree:");
3684 vdev_dbgmsg_print_tree(rvd
, 2);
3685 spa_load_note(spa
, "MOS vdev tree:");
3686 vdev_dbgmsg_print_tree(mrvd
, 2);
3688 if (copy_error
!= 0) {
3689 spa_load_note(spa
, "vdev_copy_path_strict failed, falling "
3690 "back to vdev_copy_path_relaxed");
3691 vdev_copy_path_relaxed(rvd
, mrvd
);
3696 spa
->spa_root_vdev
= mrvd
;
3698 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3701 * We will use spa_config if we decide to reload the spa or if spa_load
3702 * fails and we rewind. We must thus regenerate the config using the
3703 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3704 * pass settings on how to load the pool and is not stored in the MOS.
3705 * We copy it over to our new, trusted config.
3707 mos_config_txg
= fnvlist_lookup_uint64(mos_config
,
3708 ZPOOL_CONFIG_POOL_TXG
);
3709 nvlist_free(mos_config
);
3710 mos_config
= spa_config_generate(spa
, NULL
, mos_config_txg
, B_FALSE
);
3711 if (nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_LOAD_POLICY
,
3713 fnvlist_add_nvlist(mos_config
, ZPOOL_LOAD_POLICY
, policy
);
3714 spa_config_set(spa
, mos_config
);
3715 spa
->spa_config_source
= SPA_CONFIG_SRC_MOS
;
3718 * Now that we got the config from the MOS, we should be more strict
3719 * in checking blkptrs and can make assumptions about the consistency
3720 * of the vdev tree. spa_trust_config must be set to true before opening
3721 * vdevs in order for them to be writeable.
3723 spa
->spa_trust_config
= B_TRUE
;
3726 * Open and validate the new vdev tree
3728 error
= spa_ld_open_vdevs(spa
);
3732 error
= spa_ld_validate_vdevs(spa
);
3736 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3737 spa_load_note(spa
, "final vdev tree:");
3738 vdev_dbgmsg_print_tree(rvd
, 2);
3741 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
&&
3742 !spa
->spa_extreme_rewind
&& zfs_max_missing_tvds
== 0) {
3744 * Sanity check to make sure that we are indeed loading the
3745 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3746 * in the config provided and they happened to be the only ones
3747 * to have the latest uberblock, we could involuntarily perform
3748 * an extreme rewind.
3750 healthy_tvds_mos
= spa_healthy_core_tvds(spa
);
3751 if (healthy_tvds_mos
- healthy_tvds
>=
3752 SPA_SYNC_MIN_VDEVS
) {
3753 spa_load_note(spa
, "config provided misses too many "
3754 "top-level vdevs compared to MOS (%lld vs %lld). ",
3755 (u_longlong_t
)healthy_tvds
,
3756 (u_longlong_t
)healthy_tvds_mos
);
3757 spa_load_note(spa
, "vdev tree:");
3758 vdev_dbgmsg_print_tree(rvd
, 2);
3760 spa_load_failed(spa
, "config was already "
3761 "provided from MOS. Aborting.");
3762 return (spa_vdev_err(rvd
,
3763 VDEV_AUX_CORRUPT_DATA
, EIO
));
3765 spa_load_note(spa
, "spa must be reloaded using MOS "
3767 return (SET_ERROR(EAGAIN
));
3771 error
= spa_check_for_missing_logs(spa
);
3773 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
3775 if (rvd
->vdev_guid_sum
!= spa
->spa_uberblock
.ub_guid_sum
) {
3776 spa_load_failed(spa
, "uberblock guid sum doesn't match MOS "
3777 "guid sum (%llu != %llu)",
3778 (u_longlong_t
)spa
->spa_uberblock
.ub_guid_sum
,
3779 (u_longlong_t
)rvd
->vdev_guid_sum
);
3780 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3788 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
3791 vdev_t
*rvd
= spa
->spa_root_vdev
;
3794 * Everything that we read before spa_remove_init() must be stored
3795 * on concreted vdevs. Therefore we do this as early as possible.
3797 error
= spa_remove_init(spa
);
3799 spa_load_failed(spa
, "spa_remove_init failed [error=%d]",
3801 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3805 * Retrieve information needed to condense indirect vdev mappings.
3807 error
= spa_condense_init(spa
);
3809 spa_load_failed(spa
, "spa_condense_init failed [error=%d]",
3811 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3818 spa_ld_check_features(spa_t
*spa
, boolean_t
*missing_feat_writep
)
3821 vdev_t
*rvd
= spa
->spa_root_vdev
;
3823 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
3824 boolean_t missing_feat_read
= B_FALSE
;
3825 nvlist_t
*unsup_feat
, *enabled_feat
;
3827 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
3828 &spa
->spa_feat_for_read_obj
, B_TRUE
) != 0) {
3829 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3832 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
3833 &spa
->spa_feat_for_write_obj
, B_TRUE
) != 0) {
3834 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3837 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
3838 &spa
->spa_feat_desc_obj
, B_TRUE
) != 0) {
3839 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3842 enabled_feat
= fnvlist_alloc();
3843 unsup_feat
= fnvlist_alloc();
3845 if (!spa_features_check(spa
, B_FALSE
,
3846 unsup_feat
, enabled_feat
))
3847 missing_feat_read
= B_TRUE
;
3849 if (spa_writeable(spa
) ||
3850 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) {
3851 if (!spa_features_check(spa
, B_TRUE
,
3852 unsup_feat
, enabled_feat
)) {
3853 *missing_feat_writep
= B_TRUE
;
3857 fnvlist_add_nvlist(spa
->spa_load_info
,
3858 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
3860 if (!nvlist_empty(unsup_feat
)) {
3861 fnvlist_add_nvlist(spa
->spa_load_info
,
3862 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
3865 fnvlist_free(enabled_feat
);
3866 fnvlist_free(unsup_feat
);
3868 if (!missing_feat_read
) {
3869 fnvlist_add_boolean(spa
->spa_load_info
,
3870 ZPOOL_CONFIG_CAN_RDONLY
);
3874 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3875 * twofold: to determine whether the pool is available for
3876 * import in read-write mode and (if it is not) whether the
3877 * pool is available for import in read-only mode. If the pool
3878 * is available for import in read-write mode, it is displayed
3879 * as available in userland; if it is not available for import
3880 * in read-only mode, it is displayed as unavailable in
3881 * userland. If the pool is available for import in read-only
3882 * mode but not read-write mode, it is displayed as unavailable
3883 * in userland with a special note that the pool is actually
3884 * available for open in read-only mode.
3886 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3887 * missing a feature for write, we must first determine whether
3888 * the pool can be opened read-only before returning to
3889 * userland in order to know whether to display the
3890 * abovementioned note.
3892 if (missing_feat_read
|| (*missing_feat_writep
&&
3893 spa_writeable(spa
))) {
3894 spa_load_failed(spa
, "pool uses unsupported features");
3895 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
3900 * Load refcounts for ZFS features from disk into an in-memory
3901 * cache during SPA initialization.
3903 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
3906 error
= feature_get_refcount_from_disk(spa
,
3907 &spa_feature_table
[i
], &refcount
);
3909 spa
->spa_feat_refcount_cache
[i
] = refcount
;
3910 } else if (error
== ENOTSUP
) {
3911 spa
->spa_feat_refcount_cache
[i
] =
3912 SPA_FEATURE_DISABLED
;
3914 spa_load_failed(spa
, "error getting refcount "
3915 "for feature %s [error=%d]",
3916 spa_feature_table
[i
].fi_guid
, error
);
3917 return (spa_vdev_err(rvd
,
3918 VDEV_AUX_CORRUPT_DATA
, EIO
));
3923 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
3924 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
3925 &spa
->spa_feat_enabled_txg_obj
, B_TRUE
) != 0)
3926 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3930 * Encryption was added before bookmark_v2, even though bookmark_v2
3931 * is now a dependency. If this pool has encryption enabled without
3932 * bookmark_v2, trigger an errata message.
3934 if (spa_feature_is_enabled(spa
, SPA_FEATURE_ENCRYPTION
) &&
3935 !spa_feature_is_enabled(spa
, SPA_FEATURE_BOOKMARK_V2
)) {
3936 spa
->spa_errata
= ZPOOL_ERRATA_ZOL_8308_ENCRYPTION
;
3943 spa_ld_load_special_directories(spa_t
*spa
)
3946 vdev_t
*rvd
= spa
->spa_root_vdev
;
3948 spa
->spa_is_initializing
= B_TRUE
;
3949 error
= dsl_pool_open(spa
->spa_dsl_pool
);
3950 spa
->spa_is_initializing
= B_FALSE
;
3952 spa_load_failed(spa
, "dsl_pool_open failed [error=%d]", error
);
3953 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3960 spa_ld_get_props(spa_t
*spa
)
3964 vdev_t
*rvd
= spa
->spa_root_vdev
;
3966 /* Grab the checksum salt from the MOS. */
3967 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
3968 DMU_POOL_CHECKSUM_SALT
, 1,
3969 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
3970 spa
->spa_cksum_salt
.zcs_bytes
);
3971 if (error
== ENOENT
) {
3972 /* Generate a new salt for subsequent use */
3973 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
3974 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
3975 } else if (error
!= 0) {
3976 spa_load_failed(spa
, "unable to retrieve checksum salt from "
3977 "MOS [error=%d]", error
);
3978 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3981 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
, B_TRUE
) != 0)
3982 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3983 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
3985 spa_load_failed(spa
, "error opening deferred-frees bpobj "
3986 "[error=%d]", error
);
3987 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3991 * Load the bit that tells us to use the new accounting function
3992 * (raid-z deflation). If we have an older pool, this will not
3995 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
, B_FALSE
);
3996 if (error
!= 0 && error
!= ENOENT
)
3997 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3999 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
4000 &spa
->spa_creation_version
, B_FALSE
);
4001 if (error
!= 0 && error
!= ENOENT
)
4002 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4005 * Load the persistent error log. If we have an older pool, this will
4008 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
,
4010 if (error
!= 0 && error
!= ENOENT
)
4011 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4013 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
4014 &spa
->spa_errlog_scrub
, B_FALSE
);
4015 if (error
!= 0 && error
!= ENOENT
)
4016 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4019 * Load the livelist deletion field. If a livelist is queued for
4020 * deletion, indicate that in the spa
4022 error
= spa_dir_prop(spa
, DMU_POOL_DELETED_CLONES
,
4023 &spa
->spa_livelists_to_delete
, B_FALSE
);
4024 if (error
!= 0 && error
!= ENOENT
)
4025 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4028 * Load the history object. If we have an older pool, this
4029 * will not be present.
4031 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
, B_FALSE
);
4032 if (error
!= 0 && error
!= ENOENT
)
4033 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4036 * Load the per-vdev ZAP map. If we have an older pool, this will not
4037 * be present; in this case, defer its creation to a later time to
4038 * avoid dirtying the MOS this early / out of sync context. See
4039 * spa_sync_config_object.
4042 /* The sentinel is only available in the MOS config. */
4043 nvlist_t
*mos_config
;
4044 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0) {
4045 spa_load_failed(spa
, "unable to retrieve MOS config");
4046 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4049 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
4050 &spa
->spa_all_vdev_zaps
, B_FALSE
);
4052 if (error
== ENOENT
) {
4053 VERIFY(!nvlist_exists(mos_config
,
4054 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
4055 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
4056 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4057 } else if (error
!= 0) {
4058 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4059 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
4061 * An older version of ZFS overwrote the sentinel value, so
4062 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4063 * destruction to later; see spa_sync_config_object.
4065 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
4067 * We're assuming that no vdevs have had their ZAPs created
4068 * before this. Better be sure of it.
4070 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4072 nvlist_free(mos_config
);
4074 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4076 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
,
4078 if (error
&& error
!= ENOENT
)
4079 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4082 uint64_t autoreplace
;
4084 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
4085 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
4086 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
4087 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
4088 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
4089 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
4090 spa_prop_find(spa
, ZPOOL_PROP_AUTOTRIM
, &spa
->spa_autotrim
);
4091 spa
->spa_autoreplace
= (autoreplace
!= 0);
4095 * If we are importing a pool with missing top-level vdevs,
4096 * we enforce that the pool doesn't panic or get suspended on
4097 * error since the likelihood of missing data is extremely high.
4099 if (spa
->spa_missing_tvds
> 0 &&
4100 spa
->spa_failmode
!= ZIO_FAILURE_MODE_CONTINUE
&&
4101 spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4102 spa_load_note(spa
, "forcing failmode to 'continue' "
4103 "as some top level vdevs are missing");
4104 spa
->spa_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
4111 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
4114 vdev_t
*rvd
= spa
->spa_root_vdev
;
4117 * If we're assembling the pool from the split-off vdevs of
4118 * an existing pool, we don't want to attach the spares & cache
4123 * Load any hot spares for this pool.
4125 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
,
4127 if (error
!= 0 && error
!= ENOENT
)
4128 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4129 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4130 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
4131 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
4132 &spa
->spa_spares
.sav_config
) != 0) {
4133 spa_load_failed(spa
, "error loading spares nvlist");
4134 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4137 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4138 spa_load_spares(spa
);
4139 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4140 } else if (error
== 0) {
4141 spa
->spa_spares
.sav_sync
= B_TRUE
;
4145 * Load any level 2 ARC devices for this pool.
4147 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
4148 &spa
->spa_l2cache
.sav_object
, B_FALSE
);
4149 if (error
!= 0 && error
!= ENOENT
)
4150 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4151 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4152 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
4153 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
4154 &spa
->spa_l2cache
.sav_config
) != 0) {
4155 spa_load_failed(spa
, "error loading l2cache nvlist");
4156 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4159 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4160 spa_load_l2cache(spa
);
4161 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4162 } else if (error
== 0) {
4163 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4170 spa_ld_load_vdev_metadata(spa_t
*spa
)
4173 vdev_t
*rvd
= spa
->spa_root_vdev
;
4176 * If the 'multihost' property is set, then never allow a pool to
4177 * be imported when the system hostid is zero. The exception to
4178 * this rule is zdb which is always allowed to access pools.
4180 if (spa_multihost(spa
) && spa_get_hostid(spa
) == 0 &&
4181 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
4182 fnvlist_add_uint64(spa
->spa_load_info
,
4183 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
4184 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
4188 * If the 'autoreplace' property is set, then post a resource notifying
4189 * the ZFS DE that it should not issue any faults for unopenable
4190 * devices. We also iterate over the vdevs, and post a sysevent for any
4191 * unopenable vdevs so that the normal autoreplace handler can take
4194 if (spa
->spa_autoreplace
&& spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4195 spa_check_removed(spa
->spa_root_vdev
);
4197 * For the import case, this is done in spa_import(), because
4198 * at this point we're using the spare definitions from
4199 * the MOS config, not necessarily from the userland config.
4201 if (spa
->spa_load_state
!= SPA_LOAD_IMPORT
) {
4202 spa_aux_check_removed(&spa
->spa_spares
);
4203 spa_aux_check_removed(&spa
->spa_l2cache
);
4208 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4210 error
= vdev_load(rvd
);
4212 spa_load_failed(spa
, "vdev_load failed [error=%d]", error
);
4213 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4216 error
= spa_ld_log_spacemaps(spa
);
4218 spa_load_failed(spa
, "spa_ld_log_sm_data failed [error=%d]",
4220 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4224 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4226 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4227 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
, B_FALSE
);
4228 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4234 spa_ld_load_dedup_tables(spa_t
*spa
)
4237 vdev_t
*rvd
= spa
->spa_root_vdev
;
4239 error
= ddt_load(spa
);
4241 spa_load_failed(spa
, "ddt_load failed [error=%d]", error
);
4242 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4249 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
4251 vdev_t
*rvd
= spa
->spa_root_vdev
;
4253 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
)) {
4254 boolean_t missing
= spa_check_logs(spa
);
4256 if (spa
->spa_missing_tvds
!= 0) {
4257 spa_load_note(spa
, "spa_check_logs failed "
4258 "so dropping the logs");
4260 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
4261 spa_load_failed(spa
, "spa_check_logs failed");
4262 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
,
4272 spa_ld_verify_pool_data(spa_t
*spa
)
4275 vdev_t
*rvd
= spa
->spa_root_vdev
;
4278 * We've successfully opened the pool, verify that we're ready
4279 * to start pushing transactions.
4281 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4282 error
= spa_load_verify(spa
);
4284 spa_load_failed(spa
, "spa_load_verify failed "
4285 "[error=%d]", error
);
4286 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
4295 spa_ld_claim_log_blocks(spa_t
*spa
)
4298 dsl_pool_t
*dp
= spa_get_dsl(spa
);
4301 * Claim log blocks that haven't been committed yet.
4302 * This must all happen in a single txg.
4303 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4304 * invoked from zil_claim_log_block()'s i/o done callback.
4305 * Price of rollback is that we abandon the log.
4307 spa
->spa_claiming
= B_TRUE
;
4309 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
4310 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
4311 zil_claim
, tx
, DS_FIND_CHILDREN
);
4314 spa
->spa_claiming
= B_FALSE
;
4316 spa_set_log_state(spa
, SPA_LOG_GOOD
);
4320 spa_ld_check_for_config_update(spa_t
*spa
, uint64_t config_cache_txg
,
4321 boolean_t update_config_cache
)
4323 vdev_t
*rvd
= spa
->spa_root_vdev
;
4324 int need_update
= B_FALSE
;
4327 * If the config cache is stale, or we have uninitialized
4328 * metaslabs (see spa_vdev_add()), then update the config.
4330 * If this is a verbatim import, trust the current
4331 * in-core spa_config and update the disk labels.
4333 if (update_config_cache
|| config_cache_txg
!= spa
->spa_config_txg
||
4334 spa
->spa_load_state
== SPA_LOAD_IMPORT
||
4335 spa
->spa_load_state
== SPA_LOAD_RECOVER
||
4336 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
4337 need_update
= B_TRUE
;
4339 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
4340 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
4341 need_update
= B_TRUE
;
4344 * Update the config cache asynchronously in case we're the
4345 * root pool, in which case the config cache isn't writable yet.
4348 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4352 spa_ld_prepare_for_reload(spa_t
*spa
)
4354 spa_mode_t mode
= spa
->spa_mode
;
4355 int async_suspended
= spa
->spa_async_suspended
;
4358 spa_deactivate(spa
);
4359 spa_activate(spa
, mode
);
4362 * We save the value of spa_async_suspended as it gets reset to 0 by
4363 * spa_unload(). We want to restore it back to the original value before
4364 * returning as we might be calling spa_async_resume() later.
4366 spa
->spa_async_suspended
= async_suspended
;
4370 spa_ld_read_checkpoint_txg(spa_t
*spa
)
4372 uberblock_t checkpoint
;
4375 ASSERT0(spa
->spa_checkpoint_txg
);
4376 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4378 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4379 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4380 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4382 if (error
== ENOENT
)
4388 ASSERT3U(checkpoint
.ub_txg
, !=, 0);
4389 ASSERT3U(checkpoint
.ub_checkpoint_txg
, !=, 0);
4390 ASSERT3U(checkpoint
.ub_timestamp
, !=, 0);
4391 spa
->spa_checkpoint_txg
= checkpoint
.ub_txg
;
4392 spa
->spa_checkpoint_info
.sci_timestamp
= checkpoint
.ub_timestamp
;
4398 spa_ld_mos_init(spa_t
*spa
, spa_import_type_t type
)
4402 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4403 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4406 * Never trust the config that is provided unless we are assembling
4407 * a pool following a split.
4408 * This means don't trust blkptrs and the vdev tree in general. This
4409 * also effectively puts the spa in read-only mode since
4410 * spa_writeable() checks for spa_trust_config to be true.
4411 * We will later load a trusted config from the MOS.
4413 if (type
!= SPA_IMPORT_ASSEMBLE
)
4414 spa
->spa_trust_config
= B_FALSE
;
4417 * Parse the config provided to create a vdev tree.
4419 error
= spa_ld_parse_config(spa
, type
);
4423 spa_import_progress_add(spa
);
4426 * Now that we have the vdev tree, try to open each vdev. This involves
4427 * opening the underlying physical device, retrieving its geometry and
4428 * probing the vdev with a dummy I/O. The state of each vdev will be set
4429 * based on the success of those operations. After this we'll be ready
4430 * to read from the vdevs.
4432 error
= spa_ld_open_vdevs(spa
);
4437 * Read the label of each vdev and make sure that the GUIDs stored
4438 * there match the GUIDs in the config provided.
4439 * If we're assembling a new pool that's been split off from an
4440 * existing pool, the labels haven't yet been updated so we skip
4441 * validation for now.
4443 if (type
!= SPA_IMPORT_ASSEMBLE
) {
4444 error
= spa_ld_validate_vdevs(spa
);
4450 * Read all vdev labels to find the best uberblock (i.e. latest,
4451 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
4452 * get the list of features required to read blkptrs in the MOS from
4453 * the vdev label with the best uberblock and verify that our version
4454 * of zfs supports them all.
4456 error
= spa_ld_select_uberblock(spa
, type
);
4461 * Pass that uberblock to the dsl_pool layer which will open the root
4462 * blkptr. This blkptr points to the latest version of the MOS and will
4463 * allow us to read its contents.
4465 error
= spa_ld_open_rootbp(spa
);
4473 spa_ld_checkpoint_rewind(spa_t
*spa
)
4475 uberblock_t checkpoint
;
4478 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4479 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4481 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4482 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4483 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4486 spa_load_failed(spa
, "unable to retrieve checkpointed "
4487 "uberblock from the MOS config [error=%d]", error
);
4489 if (error
== ENOENT
)
4490 error
= ZFS_ERR_NO_CHECKPOINT
;
4495 ASSERT3U(checkpoint
.ub_txg
, <, spa
->spa_uberblock
.ub_txg
);
4496 ASSERT3U(checkpoint
.ub_txg
, ==, checkpoint
.ub_checkpoint_txg
);
4499 * We need to update the txg and timestamp of the checkpointed
4500 * uberblock to be higher than the latest one. This ensures that
4501 * the checkpointed uberblock is selected if we were to close and
4502 * reopen the pool right after we've written it in the vdev labels.
4503 * (also see block comment in vdev_uberblock_compare)
4505 checkpoint
.ub_txg
= spa
->spa_uberblock
.ub_txg
+ 1;
4506 checkpoint
.ub_timestamp
= gethrestime_sec();
4509 * Set current uberblock to be the checkpointed uberblock.
4511 spa
->spa_uberblock
= checkpoint
;
4514 * If we are doing a normal rewind, then the pool is open for
4515 * writing and we sync the "updated" checkpointed uberblock to
4516 * disk. Once this is done, we've basically rewound the whole
4517 * pool and there is no way back.
4519 * There are cases when we don't want to attempt and sync the
4520 * checkpointed uberblock to disk because we are opening a
4521 * pool as read-only. Specifically, verifying the checkpointed
4522 * state with zdb, and importing the checkpointed state to get
4523 * a "preview" of its content.
4525 if (spa_writeable(spa
)) {
4526 vdev_t
*rvd
= spa
->spa_root_vdev
;
4528 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4529 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
4531 int children
= rvd
->vdev_children
;
4532 int c0
= spa_get_random(children
);
4534 for (int c
= 0; c
< children
; c
++) {
4535 vdev_t
*vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
4537 /* Stop when revisiting the first vdev */
4538 if (c
> 0 && svd
[0] == vd
)
4541 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
4542 !vdev_is_concrete(vd
))
4545 svd
[svdcount
++] = vd
;
4546 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
4549 error
= vdev_config_sync(svd
, svdcount
, spa
->spa_first_txg
);
4551 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
4552 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4555 spa_load_failed(spa
, "failed to write checkpointed "
4556 "uberblock to the vdev labels [error=%d]", error
);
4565 spa_ld_mos_with_trusted_config(spa_t
*spa
, spa_import_type_t type
,
4566 boolean_t
*update_config_cache
)
4571 * Parse the config for pool, open and validate vdevs,
4572 * select an uberblock, and use that uberblock to open
4575 error
= spa_ld_mos_init(spa
, type
);
4580 * Retrieve the trusted config stored in the MOS and use it to create
4581 * a new, exact version of the vdev tree, then reopen all vdevs.
4583 error
= spa_ld_trusted_config(spa
, type
, B_FALSE
);
4584 if (error
== EAGAIN
) {
4585 if (update_config_cache
!= NULL
)
4586 *update_config_cache
= B_TRUE
;
4589 * Redo the loading process with the trusted config if it is
4590 * too different from the untrusted config.
4592 spa_ld_prepare_for_reload(spa
);
4593 spa_load_note(spa
, "RELOADING");
4594 error
= spa_ld_mos_init(spa
, type
);
4598 error
= spa_ld_trusted_config(spa
, type
, B_TRUE
);
4602 } else if (error
!= 0) {
4610 * Load an existing storage pool, using the config provided. This config
4611 * describes which vdevs are part of the pool and is later validated against
4612 * partial configs present in each vdev's label and an entire copy of the
4613 * config stored in the MOS.
4616 spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
4619 boolean_t missing_feat_write
= B_FALSE
;
4620 boolean_t checkpoint_rewind
=
4621 (spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4622 boolean_t update_config_cache
= B_FALSE
;
4624 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4625 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4627 spa_load_note(spa
, "LOADING");
4629 error
= spa_ld_mos_with_trusted_config(spa
, type
, &update_config_cache
);
4634 * If we are rewinding to the checkpoint then we need to repeat
4635 * everything we've done so far in this function but this time
4636 * selecting the checkpointed uberblock and using that to open
4639 if (checkpoint_rewind
) {
4641 * If we are rewinding to the checkpoint update config cache
4644 update_config_cache
= B_TRUE
;
4647 * Extract the checkpointed uberblock from the current MOS
4648 * and use this as the pool's uberblock from now on. If the
4649 * pool is imported as writeable we also write the checkpoint
4650 * uberblock to the labels, making the rewind permanent.
4652 error
= spa_ld_checkpoint_rewind(spa
);
4657 * Redo the loading process again with the
4658 * checkpointed uberblock.
4660 spa_ld_prepare_for_reload(spa
);
4661 spa_load_note(spa
, "LOADING checkpointed uberblock");
4662 error
= spa_ld_mos_with_trusted_config(spa
, type
, NULL
);
4668 * Retrieve the checkpoint txg if the pool has a checkpoint.
4670 error
= spa_ld_read_checkpoint_txg(spa
);
4675 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
4676 * from the pool and their contents were re-mapped to other vdevs. Note
4677 * that everything that we read before this step must have been
4678 * rewritten on concrete vdevs after the last device removal was
4679 * initiated. Otherwise we could be reading from indirect vdevs before
4680 * we have loaded their mappings.
4682 error
= spa_ld_open_indirect_vdev_metadata(spa
);
4687 * Retrieve the full list of active features from the MOS and check if
4688 * they are all supported.
4690 error
= spa_ld_check_features(spa
, &missing_feat_write
);
4695 * Load several special directories from the MOS needed by the dsl_pool
4698 error
= spa_ld_load_special_directories(spa
);
4703 * Retrieve pool properties from the MOS.
4705 error
= spa_ld_get_props(spa
);
4710 * Retrieve the list of auxiliary devices - cache devices and spares -
4713 error
= spa_ld_open_aux_vdevs(spa
, type
);
4718 * Load the metadata for all vdevs. Also check if unopenable devices
4719 * should be autoreplaced.
4721 error
= spa_ld_load_vdev_metadata(spa
);
4725 error
= spa_ld_load_dedup_tables(spa
);
4730 * Verify the logs now to make sure we don't have any unexpected errors
4731 * when we claim log blocks later.
4733 error
= spa_ld_verify_logs(spa
, type
, ereport
);
4737 if (missing_feat_write
) {
4738 ASSERT(spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
);
4741 * At this point, we know that we can open the pool in
4742 * read-only mode but not read-write mode. We now have enough
4743 * information and can return to userland.
4745 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
4750 * Traverse the last txgs to make sure the pool was left off in a safe
4751 * state. When performing an extreme rewind, we verify the whole pool,
4752 * which can take a very long time.
4754 error
= spa_ld_verify_pool_data(spa
);
4759 * Calculate the deflated space for the pool. This must be done before
4760 * we write anything to the pool because we'd need to update the space
4761 * accounting using the deflated sizes.
4763 spa_update_dspace(spa
);
4766 * We have now retrieved all the information we needed to open the
4767 * pool. If we are importing the pool in read-write mode, a few
4768 * additional steps must be performed to finish the import.
4770 if (spa_writeable(spa
) && (spa
->spa_load_state
== SPA_LOAD_RECOVER
||
4771 spa
->spa_load_max_txg
== UINT64_MAX
)) {
4772 uint64_t config_cache_txg
= spa
->spa_config_txg
;
4774 ASSERT(spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
);
4777 * In case of a checkpoint rewind, log the original txg
4778 * of the checkpointed uberblock.
4780 if (checkpoint_rewind
) {
4781 spa_history_log_internal(spa
, "checkpoint rewind",
4782 NULL
, "rewound state to txg=%llu",
4783 (u_longlong_t
)spa
->spa_uberblock
.ub_checkpoint_txg
);
4787 * Traverse the ZIL and claim all blocks.
4789 spa_ld_claim_log_blocks(spa
);
4792 * Kick-off the syncing thread.
4794 spa
->spa_sync_on
= B_TRUE
;
4795 txg_sync_start(spa
->spa_dsl_pool
);
4796 mmp_thread_start(spa
);
4799 * Wait for all claims to sync. We sync up to the highest
4800 * claimed log block birth time so that claimed log blocks
4801 * don't appear to be from the future. spa_claim_max_txg
4802 * will have been set for us by ZIL traversal operations
4805 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
4808 * Check if we need to request an update of the config. On the
4809 * next sync, we would update the config stored in vdev labels
4810 * and the cachefile (by default /etc/zfs/zpool.cache).
4812 spa_ld_check_for_config_update(spa
, config_cache_txg
,
4813 update_config_cache
);
4816 * Check if a rebuild was in progress and if so resume it.
4817 * Then check all DTLs to see if anything needs resilvering.
4818 * The resilver will be deferred if a rebuild was started.
4820 if (vdev_rebuild_active(spa
->spa_root_vdev
)) {
4821 vdev_rebuild_restart(spa
);
4822 } else if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4823 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
4824 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
4828 * Log the fact that we booted up (so that we can detect if
4829 * we rebooted in the middle of an operation).
4831 spa_history_log_version(spa
, "open", NULL
);
4833 spa_restart_removal(spa
);
4834 spa_spawn_aux_threads(spa
);
4837 * Delete any inconsistent datasets.
4840 * Since we may be issuing deletes for clones here,
4841 * we make sure to do so after we've spawned all the
4842 * auxiliary threads above (from which the livelist
4843 * deletion zthr is part of).
4845 (void) dmu_objset_find(spa_name(spa
),
4846 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
4849 * Clean up any stale temporary dataset userrefs.
4851 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
4853 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4854 vdev_initialize_restart(spa
->spa_root_vdev
);
4855 vdev_trim_restart(spa
->spa_root_vdev
);
4856 vdev_autotrim_restart(spa
);
4857 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4860 spa_import_progress_remove(spa_guid(spa
));
4861 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
4863 spa_load_note(spa
, "LOADED");
4869 spa_load_retry(spa_t
*spa
, spa_load_state_t state
)
4871 spa_mode_t mode
= spa
->spa_mode
;
4874 spa_deactivate(spa
);
4876 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
4878 spa_activate(spa
, mode
);
4879 spa_async_suspend(spa
);
4881 spa_load_note(spa
, "spa_load_retry: rewind, max txg: %llu",
4882 (u_longlong_t
)spa
->spa_load_max_txg
);
4884 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
));
4888 * If spa_load() fails this function will try loading prior txg's. If
4889 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
4890 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
4891 * function will not rewind the pool and will return the same error as
4895 spa_load_best(spa_t
*spa
, spa_load_state_t state
, uint64_t max_request
,
4898 nvlist_t
*loadinfo
= NULL
;
4899 nvlist_t
*config
= NULL
;
4900 int load_error
, rewind_error
;
4901 uint64_t safe_rewind_txg
;
4904 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
4905 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
4906 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
4908 spa
->spa_load_max_txg
= max_request
;
4909 if (max_request
!= UINT64_MAX
)
4910 spa
->spa_extreme_rewind
= B_TRUE
;
4913 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
);
4914 if (load_error
== 0)
4916 if (load_error
== ZFS_ERR_NO_CHECKPOINT
) {
4918 * When attempting checkpoint-rewind on a pool with no
4919 * checkpoint, we should not attempt to load uberblocks
4920 * from previous txgs when spa_load fails.
4922 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4923 spa_import_progress_remove(spa_guid(spa
));
4924 return (load_error
);
4927 if (spa
->spa_root_vdev
!= NULL
)
4928 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4930 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
4931 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
4933 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
4934 nvlist_free(config
);
4935 spa_import_progress_remove(spa_guid(spa
));
4936 return (load_error
);
4939 if (state
== SPA_LOAD_RECOVER
) {
4940 /* Price of rolling back is discarding txgs, including log */
4941 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
4944 * If we aren't rolling back save the load info from our first
4945 * import attempt so that we can restore it after attempting
4948 loadinfo
= spa
->spa_load_info
;
4949 spa
->spa_load_info
= fnvlist_alloc();
4952 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
4953 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
4954 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
4955 TXG_INITIAL
: safe_rewind_txg
;
4958 * Continue as long as we're finding errors, we're still within
4959 * the acceptable rewind range, and we're still finding uberblocks
4961 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
4962 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
4963 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
4964 spa
->spa_extreme_rewind
= B_TRUE
;
4965 rewind_error
= spa_load_retry(spa
, state
);
4968 spa
->spa_extreme_rewind
= B_FALSE
;
4969 spa
->spa_load_max_txg
= UINT64_MAX
;
4971 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
4972 spa_config_set(spa
, config
);
4974 nvlist_free(config
);
4976 if (state
== SPA_LOAD_RECOVER
) {
4977 ASSERT3P(loadinfo
, ==, NULL
);
4978 spa_import_progress_remove(spa_guid(spa
));
4979 return (rewind_error
);
4981 /* Store the rewind info as part of the initial load info */
4982 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
4983 spa
->spa_load_info
);
4985 /* Restore the initial load info */
4986 fnvlist_free(spa
->spa_load_info
);
4987 spa
->spa_load_info
= loadinfo
;
4989 spa_import_progress_remove(spa_guid(spa
));
4990 return (load_error
);
4997 * The import case is identical to an open except that the configuration is sent
4998 * down from userland, instead of grabbed from the configuration cache. For the
4999 * case of an open, the pool configuration will exist in the
5000 * POOL_STATE_UNINITIALIZED state.
5002 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5003 * the same time open the pool, without having to keep around the spa_t in some
5007 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
5011 spa_load_state_t state
= SPA_LOAD_OPEN
;
5013 int locked
= B_FALSE
;
5014 int firstopen
= B_FALSE
;
5019 * As disgusting as this is, we need to support recursive calls to this
5020 * function because dsl_dir_open() is called during spa_load(), and ends
5021 * up calling spa_open() again. The real fix is to figure out how to
5022 * avoid dsl_dir_open() calling this in the first place.
5024 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
5025 mutex_enter(&spa_namespace_lock
);
5029 if ((spa
= spa_lookup(pool
)) == NULL
) {
5031 mutex_exit(&spa_namespace_lock
);
5032 return (SET_ERROR(ENOENT
));
5035 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
5036 zpool_load_policy_t policy
;
5040 zpool_get_load_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
5042 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
5043 state
= SPA_LOAD_RECOVER
;
5045 spa_activate(spa
, spa_mode_global
);
5047 if (state
!= SPA_LOAD_RECOVER
)
5048 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
5049 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
5051 zfs_dbgmsg("spa_open_common: opening %s", pool
);
5052 error
= spa_load_best(spa
, state
, policy
.zlp_txg
,
5055 if (error
== EBADF
) {
5057 * If vdev_validate() returns failure (indicated by
5058 * EBADF), it indicates that one of the vdevs indicates
5059 * that the pool has been exported or destroyed. If
5060 * this is the case, the config cache is out of sync and
5061 * we should remove the pool from the namespace.
5064 spa_deactivate(spa
);
5065 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
5068 mutex_exit(&spa_namespace_lock
);
5069 return (SET_ERROR(ENOENT
));
5074 * We can't open the pool, but we still have useful
5075 * information: the state of each vdev after the
5076 * attempted vdev_open(). Return this to the user.
5078 if (config
!= NULL
&& spa
->spa_config
) {
5079 VERIFY(nvlist_dup(spa
->spa_config
, config
,
5081 VERIFY(nvlist_add_nvlist(*config
,
5082 ZPOOL_CONFIG_LOAD_INFO
,
5083 spa
->spa_load_info
) == 0);
5086 spa_deactivate(spa
);
5087 spa
->spa_last_open_failed
= error
;
5089 mutex_exit(&spa_namespace_lock
);
5095 spa_open_ref(spa
, tag
);
5098 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5101 * If we've recovered the pool, pass back any information we
5102 * gathered while doing the load.
5104 if (state
== SPA_LOAD_RECOVER
) {
5105 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
5106 spa
->spa_load_info
) == 0);
5110 spa
->spa_last_open_failed
= 0;
5111 spa
->spa_last_ubsync_txg
= 0;
5112 spa
->spa_load_txg
= 0;
5113 mutex_exit(&spa_namespace_lock
);
5117 zvol_create_minors_recursive(spa_name(spa
));
5125 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
5128 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
5132 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
5134 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
5138 * Lookup the given spa_t, incrementing the inject count in the process,
5139 * preventing it from being exported or destroyed.
5142 spa_inject_addref(char *name
)
5146 mutex_enter(&spa_namespace_lock
);
5147 if ((spa
= spa_lookup(name
)) == NULL
) {
5148 mutex_exit(&spa_namespace_lock
);
5151 spa
->spa_inject_ref
++;
5152 mutex_exit(&spa_namespace_lock
);
5158 spa_inject_delref(spa_t
*spa
)
5160 mutex_enter(&spa_namespace_lock
);
5161 spa
->spa_inject_ref
--;
5162 mutex_exit(&spa_namespace_lock
);
5166 * Add spares device information to the nvlist.
5169 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
5179 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5181 if (spa
->spa_spares
.sav_count
== 0)
5184 VERIFY(nvlist_lookup_nvlist(config
,
5185 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
5186 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5187 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
5189 VERIFY(nvlist_add_nvlist_array(nvroot
,
5190 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
5191 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
5192 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
5195 * Go through and find any spares which have since been
5196 * repurposed as an active spare. If this is the case, update
5197 * their status appropriately.
5199 for (i
= 0; i
< nspares
; i
++) {
5200 VERIFY(nvlist_lookup_uint64(spares
[i
],
5201 ZPOOL_CONFIG_GUID
, &guid
) == 0);
5202 if (spa_spare_exists(guid
, &pool
, NULL
) &&
5204 VERIFY(nvlist_lookup_uint64_array(
5205 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
5206 (uint64_t **)&vs
, &vsc
) == 0);
5207 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
5208 vs
->vs_aux
= VDEV_AUX_SPARED
;
5215 * Add l2cache device information to the nvlist, including vdev stats.
5218 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
5221 uint_t i
, j
, nl2cache
;
5228 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5230 if (spa
->spa_l2cache
.sav_count
== 0)
5233 VERIFY(nvlist_lookup_nvlist(config
,
5234 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
5235 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5236 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
5237 if (nl2cache
!= 0) {
5238 VERIFY(nvlist_add_nvlist_array(nvroot
,
5239 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
5240 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
5241 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
5244 * Update level 2 cache device stats.
5247 for (i
= 0; i
< nl2cache
; i
++) {
5248 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
5249 ZPOOL_CONFIG_GUID
, &guid
) == 0);
5252 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
5254 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
5255 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
5261 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
5262 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
5264 vdev_get_stats(vd
, vs
);
5265 vdev_config_generate_stats(vd
, l2cache
[i
]);
5272 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
5277 if (spa
->spa_feat_for_read_obj
!= 0) {
5278 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5279 spa
->spa_feat_for_read_obj
);
5280 zap_cursor_retrieve(&zc
, &za
) == 0;
5281 zap_cursor_advance(&zc
)) {
5282 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5283 za
.za_num_integers
== 1);
5284 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5285 za
.za_first_integer
));
5287 zap_cursor_fini(&zc
);
5290 if (spa
->spa_feat_for_write_obj
!= 0) {
5291 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5292 spa
->spa_feat_for_write_obj
);
5293 zap_cursor_retrieve(&zc
, &za
) == 0;
5294 zap_cursor_advance(&zc
)) {
5295 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5296 za
.za_num_integers
== 1);
5297 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5298 za
.za_first_integer
));
5300 zap_cursor_fini(&zc
);
5305 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
5309 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5310 zfeature_info_t feature
= spa_feature_table
[i
];
5313 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
5316 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
5321 * Store a list of pool features and their reference counts in the
5324 * The first time this is called on a spa, allocate a new nvlist, fetch
5325 * the pool features and reference counts from disk, then save the list
5326 * in the spa. In subsequent calls on the same spa use the saved nvlist
5327 * and refresh its values from the cached reference counts. This
5328 * ensures we don't block here on I/O on a suspended pool so 'zpool
5329 * clear' can resume the pool.
5332 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
5336 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5338 mutex_enter(&spa
->spa_feat_stats_lock
);
5339 features
= spa
->spa_feat_stats
;
5341 if (features
!= NULL
) {
5342 spa_feature_stats_from_cache(spa
, features
);
5344 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
5345 spa
->spa_feat_stats
= features
;
5346 spa_feature_stats_from_disk(spa
, features
);
5349 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
5352 mutex_exit(&spa
->spa_feat_stats_lock
);
5356 spa_get_stats(const char *name
, nvlist_t
**config
,
5357 char *altroot
, size_t buflen
)
5363 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
5367 * This still leaves a window of inconsistency where the spares
5368 * or l2cache devices could change and the config would be
5369 * self-inconsistent.
5371 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5373 if (*config
!= NULL
) {
5374 uint64_t loadtimes
[2];
5376 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
5377 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
5378 VERIFY(nvlist_add_uint64_array(*config
,
5379 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
5381 VERIFY(nvlist_add_uint64(*config
,
5382 ZPOOL_CONFIG_ERRCOUNT
,
5383 spa_get_errlog_size(spa
)) == 0);
5385 if (spa_suspended(spa
)) {
5386 VERIFY(nvlist_add_uint64(*config
,
5387 ZPOOL_CONFIG_SUSPENDED
,
5388 spa
->spa_failmode
) == 0);
5389 VERIFY(nvlist_add_uint64(*config
,
5390 ZPOOL_CONFIG_SUSPENDED_REASON
,
5391 spa
->spa_suspended
) == 0);
5394 spa_add_spares(spa
, *config
);
5395 spa_add_l2cache(spa
, *config
);
5396 spa_add_feature_stats(spa
, *config
);
5401 * We want to get the alternate root even for faulted pools, so we cheat
5402 * and call spa_lookup() directly.
5406 mutex_enter(&spa_namespace_lock
);
5407 spa
= spa_lookup(name
);
5409 spa_altroot(spa
, altroot
, buflen
);
5413 mutex_exit(&spa_namespace_lock
);
5415 spa_altroot(spa
, altroot
, buflen
);
5420 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5421 spa_close(spa
, FTAG
);
5428 * Validate that the auxiliary device array is well formed. We must have an
5429 * array of nvlists, each which describes a valid leaf vdev. If this is an
5430 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
5431 * specified, as long as they are well-formed.
5434 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
5435 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
5436 vdev_labeltype_t label
)
5443 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5446 * It's acceptable to have no devs specified.
5448 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
5452 return (SET_ERROR(EINVAL
));
5455 * Make sure the pool is formatted with a version that supports this
5458 if (spa_version(spa
) < version
)
5459 return (SET_ERROR(ENOTSUP
));
5462 * Set the pending device list so we correctly handle device in-use
5465 sav
->sav_pending
= dev
;
5466 sav
->sav_npending
= ndev
;
5468 for (i
= 0; i
< ndev
; i
++) {
5469 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
5473 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5475 error
= SET_ERROR(EINVAL
);
5481 if ((error
= vdev_open(vd
)) == 0 &&
5482 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
5483 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
5484 vd
->vdev_guid
) == 0);
5490 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
5497 sav
->sav_pending
= NULL
;
5498 sav
->sav_npending
= 0;
5503 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
5507 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5509 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5510 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
5511 VDEV_LABEL_SPARE
)) != 0) {
5515 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5516 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
5517 VDEV_LABEL_L2CACHE
));
5521 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
5526 if (sav
->sav_config
!= NULL
) {
5532 * Generate new dev list by concatenating with the
5535 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
5536 &olddevs
, &oldndevs
) == 0);
5538 newdevs
= kmem_alloc(sizeof (void *) *
5539 (ndevs
+ oldndevs
), KM_SLEEP
);
5540 for (i
= 0; i
< oldndevs
; i
++)
5541 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
5543 for (i
= 0; i
< ndevs
; i
++)
5544 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
5547 VERIFY(nvlist_remove(sav
->sav_config
, config
,
5548 DATA_TYPE_NVLIST_ARRAY
) == 0);
5550 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
5551 config
, newdevs
, ndevs
+ oldndevs
) == 0);
5552 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
5553 nvlist_free(newdevs
[i
]);
5554 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
5557 * Generate a new dev list.
5559 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
5561 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
5567 * Stop and drop level 2 ARC devices
5570 spa_l2cache_drop(spa_t
*spa
)
5574 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
5576 for (i
= 0; i
< sav
->sav_count
; i
++) {
5579 vd
= sav
->sav_vdevs
[i
];
5582 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
5583 pool
!= 0ULL && l2arc_vdev_present(vd
))
5584 l2arc_remove_vdev(vd
);
5589 * Verify encryption parameters for spa creation. If we are encrypting, we must
5590 * have the encryption feature flag enabled.
5593 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
5594 boolean_t has_encryption
)
5596 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
5597 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
5599 return (SET_ERROR(ENOTSUP
));
5601 return (dmu_objset_create_crypt_check(NULL
, dcp
, NULL
));
5608 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
5609 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
5612 char *altroot
= NULL
;
5617 uint64_t txg
= TXG_INITIAL
;
5618 nvlist_t
**spares
, **l2cache
;
5619 uint_t nspares
, nl2cache
;
5620 uint64_t version
, obj
;
5621 boolean_t has_features
;
5622 boolean_t has_encryption
;
5623 boolean_t has_allocclass
;
5629 if (props
== NULL
||
5630 nvlist_lookup_string(props
, "tname", &poolname
) != 0)
5631 poolname
= (char *)pool
;
5634 * If this pool already exists, return failure.
5636 mutex_enter(&spa_namespace_lock
);
5637 if (spa_lookup(poolname
) != NULL
) {
5638 mutex_exit(&spa_namespace_lock
);
5639 return (SET_ERROR(EEXIST
));
5643 * Allocate a new spa_t structure.
5645 nvl
= fnvlist_alloc();
5646 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
5647 (void) nvlist_lookup_string(props
,
5648 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5649 spa
= spa_add(poolname
, nvl
, altroot
);
5651 spa_activate(spa
, spa_mode_global
);
5653 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
5654 spa_deactivate(spa
);
5656 mutex_exit(&spa_namespace_lock
);
5661 * Temporary pool names should never be written to disk.
5663 if (poolname
!= pool
)
5664 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
5666 has_features
= B_FALSE
;
5667 has_encryption
= B_FALSE
;
5668 has_allocclass
= B_FALSE
;
5669 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
5670 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
5671 if (zpool_prop_feature(nvpair_name(elem
))) {
5672 has_features
= B_TRUE
;
5674 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
5675 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
5676 if (feat
== SPA_FEATURE_ENCRYPTION
)
5677 has_encryption
= B_TRUE
;
5678 if (feat
== SPA_FEATURE_ALLOCATION_CLASSES
)
5679 has_allocclass
= B_TRUE
;
5683 /* verify encryption params, if they were provided */
5685 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
5687 spa_deactivate(spa
);
5689 mutex_exit(&spa_namespace_lock
);
5693 if (!has_allocclass
&& zfs_special_devs(nvroot
, NULL
)) {
5694 spa_deactivate(spa
);
5696 mutex_exit(&spa_namespace_lock
);
5700 if (has_features
|| nvlist_lookup_uint64(props
,
5701 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
5702 version
= SPA_VERSION
;
5704 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
5706 spa
->spa_first_txg
= txg
;
5707 spa
->spa_uberblock
.ub_txg
= txg
- 1;
5708 spa
->spa_uberblock
.ub_version
= version
;
5709 spa
->spa_ubsync
= spa
->spa_uberblock
;
5710 spa
->spa_load_state
= SPA_LOAD_CREATE
;
5711 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
5712 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
5713 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
5714 spa
->spa_indirect_vdevs_loaded
= B_TRUE
;
5717 * Create "The Godfather" zio to hold all async IOs
5719 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
5721 for (int i
= 0; i
< max_ncpus
; i
++) {
5722 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
5723 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
5724 ZIO_FLAG_GODFATHER
);
5728 * Create the root vdev.
5730 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5732 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
5734 ASSERT(error
!= 0 || rvd
!= NULL
);
5735 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
5737 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
5738 error
= SET_ERROR(EINVAL
);
5741 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
5742 (error
= spa_validate_aux(spa
, nvroot
, txg
,
5743 VDEV_ALLOC_ADD
)) == 0) {
5745 * instantiate the metaslab groups (this will dirty the vdevs)
5746 * we can no longer error exit past this point
5748 for (int c
= 0; error
== 0 && c
< rvd
->vdev_children
; c
++) {
5749 vdev_t
*vd
= rvd
->vdev_child
[c
];
5751 vdev_ashift_optimize(vd
);
5752 vdev_metaslab_set_size(vd
);
5753 vdev_expand(vd
, txg
);
5757 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5761 spa_deactivate(spa
);
5763 mutex_exit(&spa_namespace_lock
);
5768 * Get the list of spares, if specified.
5770 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5771 &spares
, &nspares
) == 0) {
5772 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
5774 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
5775 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
5776 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5777 spa_load_spares(spa
);
5778 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5779 spa
->spa_spares
.sav_sync
= B_TRUE
;
5783 * Get the list of level 2 cache devices, if specified.
5785 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5786 &l2cache
, &nl2cache
) == 0) {
5787 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
5788 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5789 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
5790 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
5791 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5792 spa_load_l2cache(spa
);
5793 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5794 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5797 spa
->spa_is_initializing
= B_TRUE
;
5798 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
5799 spa
->spa_is_initializing
= B_FALSE
;
5802 * Create DDTs (dedup tables).
5806 spa_update_dspace(spa
);
5808 tx
= dmu_tx_create_assigned(dp
, txg
);
5811 * Create the pool's history object.
5813 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
5814 spa_history_create_obj(spa
, tx
);
5816 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
5817 spa_history_log_version(spa
, "create", tx
);
5820 * Create the pool config object.
5822 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
5823 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
5824 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
5826 if (zap_add(spa
->spa_meta_objset
,
5827 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
5828 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
5829 cmn_err(CE_PANIC
, "failed to add pool config");
5832 if (zap_add(spa
->spa_meta_objset
,
5833 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
5834 sizeof (uint64_t), 1, &version
, tx
) != 0) {
5835 cmn_err(CE_PANIC
, "failed to add pool version");
5838 /* Newly created pools with the right version are always deflated. */
5839 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
5840 spa
->spa_deflate
= TRUE
;
5841 if (zap_add(spa
->spa_meta_objset
,
5842 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
5843 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
5844 cmn_err(CE_PANIC
, "failed to add deflate");
5849 * Create the deferred-free bpobj. Turn off compression
5850 * because sync-to-convergence takes longer if the blocksize
5853 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
5854 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
5855 ZIO_COMPRESS_OFF
, tx
);
5856 if (zap_add(spa
->spa_meta_objset
,
5857 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
5858 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
5859 cmn_err(CE_PANIC
, "failed to add bpobj");
5861 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
5862 spa
->spa_meta_objset
, obj
));
5865 * Generate some random noise for salted checksums to operate on.
5867 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
5868 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
5871 * Set pool properties.
5873 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
5874 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
5875 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
5876 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
5877 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
5878 spa
->spa_autotrim
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM
);
5880 if (props
!= NULL
) {
5881 spa_configfile_set(spa
, props
, B_FALSE
);
5882 spa_sync_props(props
, tx
);
5887 spa
->spa_sync_on
= B_TRUE
;
5889 mmp_thread_start(spa
);
5890 txg_wait_synced(dp
, txg
);
5892 spa_spawn_aux_threads(spa
);
5894 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
5897 * Don't count references from objsets that are already closed
5898 * and are making their way through the eviction process.
5900 spa_evicting_os_wait(spa
);
5901 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
5902 spa
->spa_load_state
= SPA_LOAD_NONE
;
5904 mutex_exit(&spa_namespace_lock
);
5910 * Import a non-root pool into the system.
5913 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
5916 char *altroot
= NULL
;
5917 spa_load_state_t state
= SPA_LOAD_IMPORT
;
5918 zpool_load_policy_t policy
;
5919 spa_mode_t mode
= spa_mode_global
;
5920 uint64_t readonly
= B_FALSE
;
5923 nvlist_t
**spares
, **l2cache
;
5924 uint_t nspares
, nl2cache
;
5927 * If a pool with this name exists, return failure.
5929 mutex_enter(&spa_namespace_lock
);
5930 if (spa_lookup(pool
) != NULL
) {
5931 mutex_exit(&spa_namespace_lock
);
5932 return (SET_ERROR(EEXIST
));
5936 * Create and initialize the spa structure.
5938 (void) nvlist_lookup_string(props
,
5939 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5940 (void) nvlist_lookup_uint64(props
,
5941 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
5943 mode
= SPA_MODE_READ
;
5944 spa
= spa_add(pool
, config
, altroot
);
5945 spa
->spa_import_flags
= flags
;
5948 * Verbatim import - Take a pool and insert it into the namespace
5949 * as if it had been loaded at boot.
5951 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
5953 spa_configfile_set(spa
, props
, B_FALSE
);
5955 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
5956 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
5957 zfs_dbgmsg("spa_import: verbatim import of %s", pool
);
5958 mutex_exit(&spa_namespace_lock
);
5962 spa_activate(spa
, mode
);
5965 * Don't start async tasks until we know everything is healthy.
5967 spa_async_suspend(spa
);
5969 zpool_get_load_policy(config
, &policy
);
5970 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
5971 state
= SPA_LOAD_RECOVER
;
5973 spa
->spa_config_source
= SPA_CONFIG_SRC_TRYIMPORT
;
5975 if (state
!= SPA_LOAD_RECOVER
) {
5976 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
5977 zfs_dbgmsg("spa_import: importing %s", pool
);
5979 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5980 "(RECOVERY MODE)", pool
, (longlong_t
)policy
.zlp_txg
);
5982 error
= spa_load_best(spa
, state
, policy
.zlp_txg
, policy
.zlp_rewind
);
5985 * Propagate anything learned while loading the pool and pass it
5986 * back to caller (i.e. rewind info, missing devices, etc).
5988 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
5989 spa
->spa_load_info
) == 0);
5991 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5993 * Toss any existing sparelist, as it doesn't have any validity
5994 * anymore, and conflicts with spa_has_spare().
5996 if (spa
->spa_spares
.sav_config
) {
5997 nvlist_free(spa
->spa_spares
.sav_config
);
5998 spa
->spa_spares
.sav_config
= NULL
;
5999 spa_load_spares(spa
);
6001 if (spa
->spa_l2cache
.sav_config
) {
6002 nvlist_free(spa
->spa_l2cache
.sav_config
);
6003 spa
->spa_l2cache
.sav_config
= NULL
;
6004 spa_load_l2cache(spa
);
6007 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
6009 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6012 spa_configfile_set(spa
, props
, B_FALSE
);
6014 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
6015 (error
= spa_prop_set(spa
, props
)))) {
6017 spa_deactivate(spa
);
6019 mutex_exit(&spa_namespace_lock
);
6023 spa_async_resume(spa
);
6026 * Override any spares and level 2 cache devices as specified by
6027 * the user, as these may have correct device names/devids, etc.
6029 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
6030 &spares
, &nspares
) == 0) {
6031 if (spa
->spa_spares
.sav_config
)
6032 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
6033 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
6035 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
6036 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6037 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
6038 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
6039 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6040 spa_load_spares(spa
);
6041 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6042 spa
->spa_spares
.sav_sync
= B_TRUE
;
6044 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
6045 &l2cache
, &nl2cache
) == 0) {
6046 if (spa
->spa_l2cache
.sav_config
)
6047 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
6048 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
6050 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
6051 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6052 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
6053 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
6054 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6055 spa_load_l2cache(spa
);
6056 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6057 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6061 * Check for any removed devices.
6063 if (spa
->spa_autoreplace
) {
6064 spa_aux_check_removed(&spa
->spa_spares
);
6065 spa_aux_check_removed(&spa
->spa_l2cache
);
6068 if (spa_writeable(spa
)) {
6070 * Update the config cache to include the newly-imported pool.
6072 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6076 * It's possible that the pool was expanded while it was exported.
6077 * We kick off an async task to handle this for us.
6079 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
6081 spa_history_log_version(spa
, "import", NULL
);
6083 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6085 mutex_exit(&spa_namespace_lock
);
6087 zvol_create_minors_recursive(pool
);
6093 spa_tryimport(nvlist_t
*tryconfig
)
6095 nvlist_t
*config
= NULL
;
6096 char *poolname
, *cachefile
;
6100 zpool_load_policy_t policy
;
6102 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
6105 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
6109 * Create and initialize the spa structure.
6111 mutex_enter(&spa_namespace_lock
);
6112 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
6113 spa_activate(spa
, SPA_MODE_READ
);
6116 * Rewind pool if a max txg was provided.
6118 zpool_get_load_policy(spa
->spa_config
, &policy
);
6119 if (policy
.zlp_txg
!= UINT64_MAX
) {
6120 spa
->spa_load_max_txg
= policy
.zlp_txg
;
6121 spa
->spa_extreme_rewind
= B_TRUE
;
6122 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6123 poolname
, (longlong_t
)policy
.zlp_txg
);
6125 zfs_dbgmsg("spa_tryimport: importing %s", poolname
);
6128 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_CACHEFILE
, &cachefile
)
6130 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile
);
6131 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
6133 spa
->spa_config_source
= SPA_CONFIG_SRC_SCAN
;
6136 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
);
6139 * If 'tryconfig' was at least parsable, return the current config.
6141 if (spa
->spa_root_vdev
!= NULL
) {
6142 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
6143 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
6145 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
6147 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
6148 spa
->spa_uberblock
.ub_timestamp
) == 0);
6149 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
6150 spa
->spa_load_info
) == 0);
6151 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
6152 spa
->spa_errata
) == 0);
6155 * If the bootfs property exists on this pool then we
6156 * copy it out so that external consumers can tell which
6157 * pools are bootable.
6159 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
6160 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6163 * We have to play games with the name since the
6164 * pool was opened as TRYIMPORT_NAME.
6166 if (dsl_dsobj_to_dsname(spa_name(spa
),
6167 spa
->spa_bootfs
, tmpname
) == 0) {
6171 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6173 cp
= strchr(tmpname
, '/');
6175 (void) strlcpy(dsname
, tmpname
,
6178 (void) snprintf(dsname
, MAXPATHLEN
,
6179 "%s/%s", poolname
, ++cp
);
6181 VERIFY(nvlist_add_string(config
,
6182 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
6183 kmem_free(dsname
, MAXPATHLEN
);
6185 kmem_free(tmpname
, MAXPATHLEN
);
6189 * Add the list of hot spares and level 2 cache devices.
6191 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6192 spa_add_spares(spa
, config
);
6193 spa_add_l2cache(spa
, config
);
6194 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6198 spa_deactivate(spa
);
6200 mutex_exit(&spa_namespace_lock
);
6206 * Pool export/destroy
6208 * The act of destroying or exporting a pool is very simple. We make sure there
6209 * is no more pending I/O and any references to the pool are gone. Then, we
6210 * update the pool state and sync all the labels to disk, removing the
6211 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6212 * we don't sync the labels or remove the configuration cache.
6215 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
6216 boolean_t force
, boolean_t hardforce
)
6223 if (!(spa_mode_global
& SPA_MODE_WRITE
))
6224 return (SET_ERROR(EROFS
));
6226 mutex_enter(&spa_namespace_lock
);
6227 if ((spa
= spa_lookup(pool
)) == NULL
) {
6228 mutex_exit(&spa_namespace_lock
);
6229 return (SET_ERROR(ENOENT
));
6232 if (spa
->spa_is_exporting
) {
6233 /* the pool is being exported by another thread */
6234 mutex_exit(&spa_namespace_lock
);
6235 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS
));
6237 spa
->spa_is_exporting
= B_TRUE
;
6240 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6241 * reacquire the namespace lock, and see if we can export.
6243 spa_open_ref(spa
, FTAG
);
6244 mutex_exit(&spa_namespace_lock
);
6245 spa_async_suspend(spa
);
6246 if (spa
->spa_zvol_taskq
) {
6247 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
6248 taskq_wait(spa
->spa_zvol_taskq
);
6250 mutex_enter(&spa_namespace_lock
);
6251 spa_close(spa
, FTAG
);
6253 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
6256 * The pool will be in core if it's openable, in which case we can
6257 * modify its state. Objsets may be open only because they're dirty,
6258 * so we have to force it to sync before checking spa_refcnt.
6260 if (spa
->spa_sync_on
) {
6261 txg_wait_synced(spa
->spa_dsl_pool
, 0);
6262 spa_evicting_os_wait(spa
);
6266 * A pool cannot be exported or destroyed if there are active
6267 * references. If we are resetting a pool, allow references by
6268 * fault injection handlers.
6270 if (!spa_refcount_zero(spa
) ||
6271 (spa
->spa_inject_ref
!= 0 &&
6272 new_state
!= POOL_STATE_UNINITIALIZED
)) {
6273 spa_async_resume(spa
);
6274 spa
->spa_is_exporting
= B_FALSE
;
6275 mutex_exit(&spa_namespace_lock
);
6276 return (SET_ERROR(EBUSY
));
6279 if (spa
->spa_sync_on
) {
6281 * A pool cannot be exported if it has an active shared spare.
6282 * This is to prevent other pools stealing the active spare
6283 * from an exported pool. At user's own will, such pool can
6284 * be forcedly exported.
6286 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
6287 spa_has_active_shared_spare(spa
)) {
6288 spa_async_resume(spa
);
6289 spa
->spa_is_exporting
= B_FALSE
;
6290 mutex_exit(&spa_namespace_lock
);
6291 return (SET_ERROR(EXDEV
));
6295 * We're about to export or destroy this pool. Make sure
6296 * we stop all initialization and trim activity here before
6297 * we set the spa_final_txg. This will ensure that all
6298 * dirty data resulting from the initialization is
6299 * committed to disk before we unload the pool.
6301 if (spa
->spa_root_vdev
!= NULL
) {
6302 vdev_t
*rvd
= spa
->spa_root_vdev
;
6303 vdev_initialize_stop_all(rvd
, VDEV_INITIALIZE_ACTIVE
);
6304 vdev_trim_stop_all(rvd
, VDEV_TRIM_ACTIVE
);
6305 vdev_autotrim_stop_all(spa
);
6306 vdev_rebuild_stop_all(spa
);
6310 * We want this to be reflected on every label,
6311 * so mark them all dirty. spa_unload() will do the
6312 * final sync that pushes these changes out.
6314 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
6315 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6316 spa
->spa_state
= new_state
;
6317 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
6319 vdev_config_dirty(spa
->spa_root_vdev
);
6320 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6325 if (new_state
== POOL_STATE_DESTROYED
)
6326 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
6327 else if (new_state
== POOL_STATE_EXPORTED
)
6328 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
6330 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6332 spa_deactivate(spa
);
6335 if (oldconfig
&& spa
->spa_config
)
6336 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
6338 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
6340 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
6344 * If spa_remove() is not called for this spa_t and
6345 * there is any possibility that it can be reused,
6346 * we make sure to reset the exporting flag.
6348 spa
->spa_is_exporting
= B_FALSE
;
6351 mutex_exit(&spa_namespace_lock
);
6356 * Destroy a storage pool.
6359 spa_destroy(char *pool
)
6361 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
6366 * Export a storage pool.
6369 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
6370 boolean_t hardforce
)
6372 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
6377 * Similar to spa_export(), this unloads the spa_t without actually removing it
6378 * from the namespace in any way.
6381 spa_reset(char *pool
)
6383 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
6388 * ==========================================================================
6389 * Device manipulation
6390 * ==========================================================================
6394 * Add a device to a storage pool.
6397 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
6401 vdev_t
*rvd
= spa
->spa_root_vdev
;
6403 nvlist_t
**spares
, **l2cache
;
6404 uint_t nspares
, nl2cache
;
6406 ASSERT(spa_writeable(spa
));
6408 txg
= spa_vdev_enter(spa
);
6410 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
6411 VDEV_ALLOC_ADD
)) != 0)
6412 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6414 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
6416 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
6420 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
6424 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
6425 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6427 if (vd
->vdev_children
!= 0 &&
6428 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
6429 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6432 * We must validate the spares and l2cache devices after checking the
6433 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
6435 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
6436 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6439 * If we are in the middle of a device removal, we can only add
6440 * devices which match the existing devices in the pool.
6441 * If we are in the middle of a removal, or have some indirect
6442 * vdevs, we can not add raidz toplevels.
6444 if (spa
->spa_vdev_removal
!= NULL
||
6445 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
6446 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6447 tvd
= vd
->vdev_child
[c
];
6448 if (spa
->spa_vdev_removal
!= NULL
&&
6449 tvd
->vdev_ashift
!= spa
->spa_max_ashift
) {
6450 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6452 /* Fail if top level vdev is raidz */
6453 if (tvd
->vdev_ops
== &vdev_raidz_ops
) {
6454 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6457 * Need the top level mirror to be
6458 * a mirror of leaf vdevs only
6460 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
6461 for (uint64_t cid
= 0;
6462 cid
< tvd
->vdev_children
; cid
++) {
6463 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
6464 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
6465 return (spa_vdev_exit(spa
, vd
,
6473 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6474 tvd
= vd
->vdev_child
[c
];
6475 vdev_remove_child(vd
, tvd
);
6476 tvd
->vdev_id
= rvd
->vdev_children
;
6477 vdev_add_child(rvd
, tvd
);
6478 vdev_config_dirty(tvd
);
6482 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
6483 ZPOOL_CONFIG_SPARES
);
6484 spa_load_spares(spa
);
6485 spa
->spa_spares
.sav_sync
= B_TRUE
;
6488 if (nl2cache
!= 0) {
6489 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
6490 ZPOOL_CONFIG_L2CACHE
);
6491 spa_load_l2cache(spa
);
6492 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6496 * We have to be careful when adding new vdevs to an existing pool.
6497 * If other threads start allocating from these vdevs before we
6498 * sync the config cache, and we lose power, then upon reboot we may
6499 * fail to open the pool because there are DVAs that the config cache
6500 * can't translate. Therefore, we first add the vdevs without
6501 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
6502 * and then let spa_config_update() initialize the new metaslabs.
6504 * spa_load() checks for added-but-not-initialized vdevs, so that
6505 * if we lose power at any point in this sequence, the remaining
6506 * steps will be completed the next time we load the pool.
6508 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
6510 mutex_enter(&spa_namespace_lock
);
6511 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6512 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
6513 mutex_exit(&spa_namespace_lock
);
6519 * Attach a device to a mirror. The arguments are the path to any device
6520 * in the mirror, and the nvroot for the new device. If the path specifies
6521 * a device that is not mirrored, we automatically insert the mirror vdev.
6523 * If 'replacing' is specified, the new device is intended to replace the
6524 * existing device; in this case the two devices are made into their own
6525 * mirror using the 'replacing' vdev, which is functionally identical to
6526 * the mirror vdev (it actually reuses all the same ops) but has a few
6527 * extra rules: you can't attach to it after it's been created, and upon
6528 * completion of resilvering, the first disk (the one being replaced)
6529 * is automatically detached.
6531 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
6532 * should be performed instead of traditional healing reconstruction. From
6533 * an administrators perspective these are both resilver operations.
6536 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
,
6539 uint64_t txg
, dtl_max_txg
;
6540 vdev_t
*rvd
= spa
->spa_root_vdev
;
6541 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
6543 char *oldvdpath
, *newvdpath
;
6547 ASSERT(spa_writeable(spa
));
6549 txg
= spa_vdev_enter(spa
);
6551 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
6553 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
6554 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
6555 error
= (spa_has_checkpoint(spa
)) ?
6556 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
6557 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6561 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_DEVICE_REBUILD
))
6562 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6564 if (dsl_scan_resilvering(spa_get_dsl(spa
)))
6565 return (spa_vdev_exit(spa
, NULL
, txg
,
6566 ZFS_ERR_RESILVER_IN_PROGRESS
));
6568 if (vdev_rebuild_active(rvd
))
6569 return (spa_vdev_exit(spa
, NULL
, txg
,
6570 ZFS_ERR_REBUILD_IN_PROGRESS
));
6573 if (spa
->spa_vdev_removal
!= NULL
)
6574 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
6577 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
6579 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
6580 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6582 pvd
= oldvd
->vdev_parent
;
6584 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
6585 VDEV_ALLOC_ATTACH
)) != 0)
6586 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
6588 if (newrootvd
->vdev_children
!= 1)
6589 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
6591 newvd
= newrootvd
->vdev_child
[0];
6593 if (!newvd
->vdev_ops
->vdev_op_leaf
)
6594 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
6596 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
6597 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
6600 * Spares can't replace logs
6602 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
6603 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6607 * For rebuilds, the parent vdev must support reconstruction
6608 * using only space maps. This means the only allowable
6609 * parents are the root vdev or a mirror vdev.
6611 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
6612 pvd
->vdev_ops
!= &vdev_root_ops
) {
6613 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6619 * For attach, the only allowable parent is a mirror or the root
6622 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
6623 pvd
->vdev_ops
!= &vdev_root_ops
)
6624 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6626 pvops
= &vdev_mirror_ops
;
6629 * Active hot spares can only be replaced by inactive hot
6632 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
6633 oldvd
->vdev_isspare
&&
6634 !spa_has_spare(spa
, newvd
->vdev_guid
))
6635 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6638 * If the source is a hot spare, and the parent isn't already a
6639 * spare, then we want to create a new hot spare. Otherwise, we
6640 * want to create a replacing vdev. The user is not allowed to
6641 * attach to a spared vdev child unless the 'isspare' state is
6642 * the same (spare replaces spare, non-spare replaces
6645 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
6646 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
6647 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6648 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
6649 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
6650 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6653 if (newvd
->vdev_isspare
)
6654 pvops
= &vdev_spare_ops
;
6656 pvops
= &vdev_replacing_ops
;
6660 * Make sure the new device is big enough.
6662 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
6663 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
6666 * The new device cannot have a higher alignment requirement
6667 * than the top-level vdev.
6669 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
6670 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6673 * If this is an in-place replacement, update oldvd's path and devid
6674 * to make it distinguishable from newvd, and unopenable from now on.
6676 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
6677 spa_strfree(oldvd
->vdev_path
);
6678 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
6680 (void) snprintf(oldvd
->vdev_path
, strlen(newvd
->vdev_path
) + 5,
6681 "%s/%s", newvd
->vdev_path
, "old");
6682 if (oldvd
->vdev_devid
!= NULL
) {
6683 spa_strfree(oldvd
->vdev_devid
);
6684 oldvd
->vdev_devid
= NULL
;
6689 * If the parent is not a mirror, or if we're replacing, insert the new
6690 * mirror/replacing/spare vdev above oldvd.
6692 if (pvd
->vdev_ops
!= pvops
)
6693 pvd
= vdev_add_parent(oldvd
, pvops
);
6695 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
6696 ASSERT(pvd
->vdev_ops
== pvops
);
6697 ASSERT(oldvd
->vdev_parent
== pvd
);
6700 * Extract the new device from its root and add it to pvd.
6702 vdev_remove_child(newrootvd
, newvd
);
6703 newvd
->vdev_id
= pvd
->vdev_children
;
6704 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
6705 vdev_add_child(pvd
, newvd
);
6708 * Reevaluate the parent vdev state.
6710 vdev_propagate_state(pvd
);
6712 tvd
= newvd
->vdev_top
;
6713 ASSERT(pvd
->vdev_top
== tvd
);
6714 ASSERT(tvd
->vdev_parent
== rvd
);
6716 vdev_config_dirty(tvd
);
6719 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6720 * for any dmu_sync-ed blocks. It will propagate upward when
6721 * spa_vdev_exit() calls vdev_dtl_reassess().
6723 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
6725 vdev_dtl_dirty(newvd
, DTL_MISSING
,
6726 TXG_INITIAL
, dtl_max_txg
- TXG_INITIAL
);
6728 if (newvd
->vdev_isspare
) {
6729 spa_spare_activate(newvd
);
6730 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
6733 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
6734 newvdpath
= spa_strdup(newvd
->vdev_path
);
6735 newvd_isspare
= newvd
->vdev_isspare
;
6738 * Mark newvd's DTL dirty in this txg.
6740 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
6743 * Schedule the resilver or rebuild to restart in the future. We do
6744 * this to ensure that dmu_sync-ed blocks have been stitched into the
6745 * respective datasets.
6748 newvd
->vdev_rebuild_txg
= txg
;
6752 newvd
->vdev_resilver_txg
= txg
;
6754 if (dsl_scan_resilvering(spa_get_dsl(spa
)) &&
6755 spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
)) {
6756 vdev_defer_resilver(newvd
);
6758 dsl_scan_restart_resilver(spa
->spa_dsl_pool
,
6763 if (spa
->spa_bootfs
)
6764 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
6766 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
6771 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
6773 spa_history_log_internal(spa
, "vdev attach", NULL
,
6774 "%s vdev=%s %s vdev=%s",
6775 replacing
&& newvd_isspare
? "spare in" :
6776 replacing
? "replace" : "attach", newvdpath
,
6777 replacing
? "for" : "to", oldvdpath
);
6779 spa_strfree(oldvdpath
);
6780 spa_strfree(newvdpath
);
6786 * Detach a device from a mirror or replacing vdev.
6788 * If 'replace_done' is specified, only detach if the parent
6789 * is a replacing vdev.
6792 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
6796 vdev_t
*rvd __maybe_unused
= spa
->spa_root_vdev
;
6797 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
6798 boolean_t unspare
= B_FALSE
;
6799 uint64_t unspare_guid
= 0;
6802 ASSERT(spa_writeable(spa
));
6804 txg
= spa_vdev_detach_enter(spa
, guid
);
6806 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
6809 * Besides being called directly from the userland through the
6810 * ioctl interface, spa_vdev_detach() can be potentially called
6811 * at the end of spa_vdev_resilver_done().
6813 * In the regular case, when we have a checkpoint this shouldn't
6814 * happen as we never empty the DTLs of a vdev during the scrub
6815 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6816 * should never get here when we have a checkpoint.
6818 * That said, even in a case when we checkpoint the pool exactly
6819 * as spa_vdev_resilver_done() calls this function everything
6820 * should be fine as the resilver will return right away.
6822 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
6823 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
6824 error
= (spa_has_checkpoint(spa
)) ?
6825 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
6826 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6830 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
6832 if (!vd
->vdev_ops
->vdev_op_leaf
)
6833 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6835 pvd
= vd
->vdev_parent
;
6838 * If the parent/child relationship is not as expected, don't do it.
6839 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6840 * vdev that's replacing B with C. The user's intent in replacing
6841 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6842 * the replace by detaching C, the expected behavior is to end up
6843 * M(A,B). But suppose that right after deciding to detach C,
6844 * the replacement of B completes. We would have M(A,C), and then
6845 * ask to detach C, which would leave us with just A -- not what
6846 * the user wanted. To prevent this, we make sure that the
6847 * parent/child relationship hasn't changed -- in this example,
6848 * that C's parent is still the replacing vdev R.
6850 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
6851 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
6854 * Only 'replacing' or 'spare' vdevs can be replaced.
6856 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
6857 pvd
->vdev_ops
!= &vdev_spare_ops
)
6858 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6860 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
6861 spa_version(spa
) >= SPA_VERSION_SPARES
);
6864 * Only mirror, replacing, and spare vdevs support detach.
6866 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
6867 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
6868 pvd
->vdev_ops
!= &vdev_spare_ops
)
6869 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6872 * If this device has the only valid copy of some data,
6873 * we cannot safely detach it.
6875 if (vdev_dtl_required(vd
))
6876 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
6878 ASSERT(pvd
->vdev_children
>= 2);
6881 * If we are detaching the second disk from a replacing vdev, then
6882 * check to see if we changed the original vdev's path to have "/old"
6883 * at the end in spa_vdev_attach(). If so, undo that change now.
6885 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
6886 vd
->vdev_path
!= NULL
) {
6887 size_t len
= strlen(vd
->vdev_path
);
6889 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
6890 cvd
= pvd
->vdev_child
[c
];
6892 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
6895 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
6896 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
6897 spa_strfree(cvd
->vdev_path
);
6898 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
6905 * If we are detaching the original disk from a spare, then it implies
6906 * that the spare should become a real disk, and be removed from the
6907 * active spare list for the pool.
6909 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
6911 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
6915 * Erase the disk labels so the disk can be used for other things.
6916 * This must be done after all other error cases are handled,
6917 * but before we disembowel vd (so we can still do I/O to it).
6918 * But if we can't do it, don't treat the error as fatal --
6919 * it may be that the unwritability of the disk is the reason
6920 * it's being detached!
6922 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
6925 * Remove vd from its parent and compact the parent's children.
6927 vdev_remove_child(pvd
, vd
);
6928 vdev_compact_children(pvd
);
6931 * Remember one of the remaining children so we can get tvd below.
6933 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
6936 * If we need to remove the remaining child from the list of hot spares,
6937 * do it now, marking the vdev as no longer a spare in the process.
6938 * We must do this before vdev_remove_parent(), because that can
6939 * change the GUID if it creates a new toplevel GUID. For a similar
6940 * reason, we must remove the spare now, in the same txg as the detach;
6941 * otherwise someone could attach a new sibling, change the GUID, and
6942 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6945 ASSERT(cvd
->vdev_isspare
);
6946 spa_spare_remove(cvd
);
6947 unspare_guid
= cvd
->vdev_guid
;
6948 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
6949 cvd
->vdev_unspare
= B_TRUE
;
6953 * If the parent mirror/replacing vdev only has one child,
6954 * the parent is no longer needed. Remove it from the tree.
6956 if (pvd
->vdev_children
== 1) {
6957 if (pvd
->vdev_ops
== &vdev_spare_ops
)
6958 cvd
->vdev_unspare
= B_FALSE
;
6959 vdev_remove_parent(cvd
);
6963 * We don't set tvd until now because the parent we just removed
6964 * may have been the previous top-level vdev.
6966 tvd
= cvd
->vdev_top
;
6967 ASSERT(tvd
->vdev_parent
== rvd
);
6970 * Reevaluate the parent vdev state.
6972 vdev_propagate_state(cvd
);
6975 * If the 'autoexpand' property is set on the pool then automatically
6976 * try to expand the size of the pool. For example if the device we
6977 * just detached was smaller than the others, it may be possible to
6978 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6979 * first so that we can obtain the updated sizes of the leaf vdevs.
6981 if (spa
->spa_autoexpand
) {
6983 vdev_expand(tvd
, txg
);
6986 vdev_config_dirty(tvd
);
6989 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
6990 * vd->vdev_detached is set and free vd's DTL object in syncing context.
6991 * But first make sure we're not on any *other* txg's DTL list, to
6992 * prevent vd from being accessed after it's freed.
6994 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
6995 for (int t
= 0; t
< TXG_SIZE
; t
++)
6996 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
6997 vd
->vdev_detached
= B_TRUE
;
6998 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
7000 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
7001 spa_notify_waiters(spa
);
7003 /* hang on to the spa before we release the lock */
7004 spa_open_ref(spa
, FTAG
);
7006 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
7008 spa_history_log_internal(spa
, "detach", NULL
,
7010 spa_strfree(vdpath
);
7013 * If this was the removal of the original device in a hot spare vdev,
7014 * then we want to go through and remove the device from the hot spare
7015 * list of every other pool.
7018 spa_t
*altspa
= NULL
;
7020 mutex_enter(&spa_namespace_lock
);
7021 while ((altspa
= spa_next(altspa
)) != NULL
) {
7022 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
7026 spa_open_ref(altspa
, FTAG
);
7027 mutex_exit(&spa_namespace_lock
);
7028 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
7029 mutex_enter(&spa_namespace_lock
);
7030 spa_close(altspa
, FTAG
);
7032 mutex_exit(&spa_namespace_lock
);
7034 /* search the rest of the vdevs for spares to remove */
7035 spa_vdev_resilver_done(spa
);
7038 /* all done with the spa; OK to release */
7039 mutex_enter(&spa_namespace_lock
);
7040 spa_close(spa
, FTAG
);
7041 mutex_exit(&spa_namespace_lock
);
7047 spa_vdev_initialize_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7050 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7052 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7054 /* Look up vdev and ensure it's a leaf. */
7055 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7056 if (vd
== NULL
|| vd
->vdev_detached
) {
7057 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7058 return (SET_ERROR(ENODEV
));
7059 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7060 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7061 return (SET_ERROR(EINVAL
));
7062 } else if (!vdev_writeable(vd
)) {
7063 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7064 return (SET_ERROR(EROFS
));
7066 mutex_enter(&vd
->vdev_initialize_lock
);
7067 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7070 * When we activate an initialize action we check to see
7071 * if the vdev_initialize_thread is NULL. We do this instead
7072 * of using the vdev_initialize_state since there might be
7073 * a previous initialization process which has completed but
7074 * the thread is not exited.
7076 if (cmd_type
== POOL_INITIALIZE_START
&&
7077 (vd
->vdev_initialize_thread
!= NULL
||
7078 vd
->vdev_top
->vdev_removing
)) {
7079 mutex_exit(&vd
->vdev_initialize_lock
);
7080 return (SET_ERROR(EBUSY
));
7081 } else if (cmd_type
== POOL_INITIALIZE_CANCEL
&&
7082 (vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
&&
7083 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_SUSPENDED
)) {
7084 mutex_exit(&vd
->vdev_initialize_lock
);
7085 return (SET_ERROR(ESRCH
));
7086 } else if (cmd_type
== POOL_INITIALIZE_SUSPEND
&&
7087 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
) {
7088 mutex_exit(&vd
->vdev_initialize_lock
);
7089 return (SET_ERROR(ESRCH
));
7093 case POOL_INITIALIZE_START
:
7094 vdev_initialize(vd
);
7096 case POOL_INITIALIZE_CANCEL
:
7097 vdev_initialize_stop(vd
, VDEV_INITIALIZE_CANCELED
, vd_list
);
7099 case POOL_INITIALIZE_SUSPEND
:
7100 vdev_initialize_stop(vd
, VDEV_INITIALIZE_SUSPENDED
, vd_list
);
7103 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7105 mutex_exit(&vd
->vdev_initialize_lock
);
7111 spa_vdev_initialize(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
,
7112 nvlist_t
*vdev_errlist
)
7114 int total_errors
= 0;
7117 list_create(&vd_list
, sizeof (vdev_t
),
7118 offsetof(vdev_t
, vdev_initialize_node
));
7121 * We hold the namespace lock through the whole function
7122 * to prevent any changes to the pool while we're starting or
7123 * stopping initialization. The config and state locks are held so that
7124 * we can properly assess the vdev state before we commit to
7125 * the initializing operation.
7127 mutex_enter(&spa_namespace_lock
);
7129 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7130 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7131 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7133 int error
= spa_vdev_initialize_impl(spa
, vdev_guid
, cmd_type
,
7136 char guid_as_str
[MAXNAMELEN
];
7138 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7139 "%llu", (unsigned long long)vdev_guid
);
7140 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7145 /* Wait for all initialize threads to stop. */
7146 vdev_initialize_stop_wait(spa
, &vd_list
);
7148 /* Sync out the initializing state */
7149 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7150 mutex_exit(&spa_namespace_lock
);
7152 list_destroy(&vd_list
);
7154 return (total_errors
);
7158 spa_vdev_trim_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7159 uint64_t rate
, boolean_t partial
, boolean_t secure
, list_t
*vd_list
)
7161 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7163 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7165 /* Look up vdev and ensure it's a leaf. */
7166 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7167 if (vd
== NULL
|| vd
->vdev_detached
) {
7168 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7169 return (SET_ERROR(ENODEV
));
7170 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7171 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7172 return (SET_ERROR(EINVAL
));
7173 } else if (!vdev_writeable(vd
)) {
7174 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7175 return (SET_ERROR(EROFS
));
7176 } else if (!vd
->vdev_has_trim
) {
7177 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7178 return (SET_ERROR(EOPNOTSUPP
));
7179 } else if (secure
&& !vd
->vdev_has_securetrim
) {
7180 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7181 return (SET_ERROR(EOPNOTSUPP
));
7183 mutex_enter(&vd
->vdev_trim_lock
);
7184 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7187 * When we activate a TRIM action we check to see if the
7188 * vdev_trim_thread is NULL. We do this instead of using the
7189 * vdev_trim_state since there might be a previous TRIM process
7190 * which has completed but the thread is not exited.
7192 if (cmd_type
== POOL_TRIM_START
&&
7193 (vd
->vdev_trim_thread
!= NULL
|| vd
->vdev_top
->vdev_removing
)) {
7194 mutex_exit(&vd
->vdev_trim_lock
);
7195 return (SET_ERROR(EBUSY
));
7196 } else if (cmd_type
== POOL_TRIM_CANCEL
&&
7197 (vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
&&
7198 vd
->vdev_trim_state
!= VDEV_TRIM_SUSPENDED
)) {
7199 mutex_exit(&vd
->vdev_trim_lock
);
7200 return (SET_ERROR(ESRCH
));
7201 } else if (cmd_type
== POOL_TRIM_SUSPEND
&&
7202 vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
) {
7203 mutex_exit(&vd
->vdev_trim_lock
);
7204 return (SET_ERROR(ESRCH
));
7208 case POOL_TRIM_START
:
7209 vdev_trim(vd
, rate
, partial
, secure
);
7211 case POOL_TRIM_CANCEL
:
7212 vdev_trim_stop(vd
, VDEV_TRIM_CANCELED
, vd_list
);
7214 case POOL_TRIM_SUSPEND
:
7215 vdev_trim_stop(vd
, VDEV_TRIM_SUSPENDED
, vd_list
);
7218 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7220 mutex_exit(&vd
->vdev_trim_lock
);
7226 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
7227 * TRIM threads for each child vdev. These threads pass over all of the free
7228 * space in the vdev's metaslabs and issues TRIM commands for that space.
7231 spa_vdev_trim(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
, uint64_t rate
,
7232 boolean_t partial
, boolean_t secure
, nvlist_t
*vdev_errlist
)
7234 int total_errors
= 0;
7237 list_create(&vd_list
, sizeof (vdev_t
),
7238 offsetof(vdev_t
, vdev_trim_node
));
7241 * We hold the namespace lock through the whole function
7242 * to prevent any changes to the pool while we're starting or
7243 * stopping TRIM. The config and state locks are held so that
7244 * we can properly assess the vdev state before we commit to
7245 * the TRIM operation.
7247 mutex_enter(&spa_namespace_lock
);
7249 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7250 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7251 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7253 int error
= spa_vdev_trim_impl(spa
, vdev_guid
, cmd_type
,
7254 rate
, partial
, secure
, &vd_list
);
7256 char guid_as_str
[MAXNAMELEN
];
7258 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7259 "%llu", (unsigned long long)vdev_guid
);
7260 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7265 /* Wait for all TRIM threads to stop. */
7266 vdev_trim_stop_wait(spa
, &vd_list
);
7268 /* Sync out the TRIM state */
7269 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7270 mutex_exit(&spa_namespace_lock
);
7272 list_destroy(&vd_list
);
7274 return (total_errors
);
7278 * Split a set of devices from their mirrors, and create a new pool from them.
7281 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
7282 nvlist_t
*props
, boolean_t exp
)
7285 uint64_t txg
, *glist
;
7287 uint_t c
, children
, lastlog
;
7288 nvlist_t
**child
, *nvl
, *tmp
;
7290 char *altroot
= NULL
;
7291 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
7292 boolean_t activate_slog
;
7294 ASSERT(spa_writeable(spa
));
7296 txg
= spa_vdev_enter(spa
);
7298 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7299 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7300 error
= (spa_has_checkpoint(spa
)) ?
7301 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7302 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7305 /* clear the log and flush everything up to now */
7306 activate_slog
= spa_passivate_log(spa
);
7307 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
7308 error
= spa_reset_logs(spa
);
7309 txg
= spa_vdev_config_enter(spa
);
7312 spa_activate_log(spa
);
7315 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7317 /* check new spa name before going any further */
7318 if (spa_lookup(newname
) != NULL
)
7319 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
7322 * scan through all the children to ensure they're all mirrors
7324 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
7325 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
7327 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7329 /* first, check to ensure we've got the right child count */
7330 rvd
= spa
->spa_root_vdev
;
7332 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
7333 vdev_t
*vd
= rvd
->vdev_child
[c
];
7335 /* don't count the holes & logs as children */
7336 if (vd
->vdev_islog
|| (vd
->vdev_ops
!= &vdev_indirect_ops
&&
7337 !vdev_is_concrete(vd
))) {
7345 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
7346 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7348 /* next, ensure no spare or cache devices are part of the split */
7349 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
7350 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
7351 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7353 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
7354 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
7356 /* then, loop over each vdev and validate it */
7357 for (c
= 0; c
< children
; c
++) {
7358 uint64_t is_hole
= 0;
7360 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
7364 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
7365 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
7368 error
= SET_ERROR(EINVAL
);
7373 /* deal with indirect vdevs */
7374 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ops
==
7378 /* which disk is going to be split? */
7379 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
7381 error
= SET_ERROR(EINVAL
);
7385 /* look it up in the spa */
7386 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
7387 if (vml
[c
] == NULL
) {
7388 error
= SET_ERROR(ENODEV
);
7392 /* make sure there's nothing stopping the split */
7393 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
7394 vml
[c
]->vdev_islog
||
7395 !vdev_is_concrete(vml
[c
]) ||
7396 vml
[c
]->vdev_isspare
||
7397 vml
[c
]->vdev_isl2cache
||
7398 !vdev_writeable(vml
[c
]) ||
7399 vml
[c
]->vdev_children
!= 0 ||
7400 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
7401 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
7402 error
= SET_ERROR(EINVAL
);
7406 if (vdev_dtl_required(vml
[c
]) ||
7407 vdev_resilver_needed(vml
[c
], NULL
, NULL
)) {
7408 error
= SET_ERROR(EBUSY
);
7412 /* we need certain info from the top level */
7413 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
7414 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
7415 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
7416 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
7417 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
7418 vml
[c
]->vdev_top
->vdev_asize
) == 0);
7419 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
7420 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
7422 /* transfer per-vdev ZAPs */
7423 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
7424 VERIFY0(nvlist_add_uint64(child
[c
],
7425 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
7427 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
7428 VERIFY0(nvlist_add_uint64(child
[c
],
7429 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
7430 vml
[c
]->vdev_parent
->vdev_top_zap
));
7434 kmem_free(vml
, children
* sizeof (vdev_t
*));
7435 kmem_free(glist
, children
* sizeof (uint64_t));
7436 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7439 /* stop writers from using the disks */
7440 for (c
= 0; c
< children
; c
++) {
7442 vml
[c
]->vdev_offline
= B_TRUE
;
7444 vdev_reopen(spa
->spa_root_vdev
);
7447 * Temporarily record the splitting vdevs in the spa config. This
7448 * will disappear once the config is regenerated.
7450 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
7451 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
7452 glist
, children
) == 0);
7453 kmem_free(glist
, children
* sizeof (uint64_t));
7455 mutex_enter(&spa
->spa_props_lock
);
7456 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
7458 mutex_exit(&spa
->spa_props_lock
);
7459 spa
->spa_config_splitting
= nvl
;
7460 vdev_config_dirty(spa
->spa_root_vdev
);
7462 /* configure and create the new pool */
7463 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
7464 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
7465 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
7466 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
7467 spa_version(spa
)) == 0);
7468 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
7469 spa
->spa_config_txg
) == 0);
7470 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
7471 spa_generate_guid(NULL
)) == 0);
7472 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
7473 (void) nvlist_lookup_string(props
,
7474 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
7476 /* add the new pool to the namespace */
7477 newspa
= spa_add(newname
, config
, altroot
);
7478 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
7479 newspa
->spa_config_txg
= spa
->spa_config_txg
;
7480 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
7482 /* release the spa config lock, retaining the namespace lock */
7483 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
7485 if (zio_injection_enabled
)
7486 zio_handle_panic_injection(spa
, FTAG
, 1);
7488 spa_activate(newspa
, spa_mode_global
);
7489 spa_async_suspend(newspa
);
7492 * Temporarily stop the initializing and TRIM activity. We set the
7493 * state to ACTIVE so that we know to resume initializing or TRIM
7494 * once the split has completed.
7496 list_t vd_initialize_list
;
7497 list_create(&vd_initialize_list
, sizeof (vdev_t
),
7498 offsetof(vdev_t
, vdev_initialize_node
));
7500 list_t vd_trim_list
;
7501 list_create(&vd_trim_list
, sizeof (vdev_t
),
7502 offsetof(vdev_t
, vdev_trim_node
));
7504 for (c
= 0; c
< children
; c
++) {
7505 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
7506 mutex_enter(&vml
[c
]->vdev_initialize_lock
);
7507 vdev_initialize_stop(vml
[c
],
7508 VDEV_INITIALIZE_ACTIVE
, &vd_initialize_list
);
7509 mutex_exit(&vml
[c
]->vdev_initialize_lock
);
7511 mutex_enter(&vml
[c
]->vdev_trim_lock
);
7512 vdev_trim_stop(vml
[c
], VDEV_TRIM_ACTIVE
, &vd_trim_list
);
7513 mutex_exit(&vml
[c
]->vdev_trim_lock
);
7517 vdev_initialize_stop_wait(spa
, &vd_initialize_list
);
7518 vdev_trim_stop_wait(spa
, &vd_trim_list
);
7520 list_destroy(&vd_initialize_list
);
7521 list_destroy(&vd_trim_list
);
7523 newspa
->spa_config_source
= SPA_CONFIG_SRC_SPLIT
;
7524 newspa
->spa_is_splitting
= B_TRUE
;
7526 /* create the new pool from the disks of the original pool */
7527 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
);
7531 /* if that worked, generate a real config for the new pool */
7532 if (newspa
->spa_root_vdev
!= NULL
) {
7533 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
7534 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
7535 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
7536 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
7537 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
7542 if (props
!= NULL
) {
7543 spa_configfile_set(newspa
, props
, B_FALSE
);
7544 error
= spa_prop_set(newspa
, props
);
7549 /* flush everything */
7550 txg
= spa_vdev_config_enter(newspa
);
7551 vdev_config_dirty(newspa
->spa_root_vdev
);
7552 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
7554 if (zio_injection_enabled
)
7555 zio_handle_panic_injection(spa
, FTAG
, 2);
7557 spa_async_resume(newspa
);
7559 /* finally, update the original pool's config */
7560 txg
= spa_vdev_config_enter(spa
);
7561 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
7562 error
= dmu_tx_assign(tx
, TXG_WAIT
);
7565 for (c
= 0; c
< children
; c
++) {
7566 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
7567 vdev_t
*tvd
= vml
[c
]->vdev_top
;
7570 * Need to be sure the detachable VDEV is not
7571 * on any *other* txg's DTL list to prevent it
7572 * from being accessed after it's freed.
7574 for (int t
= 0; t
< TXG_SIZE
; t
++) {
7575 (void) txg_list_remove_this(
7576 &tvd
->vdev_dtl_list
, vml
[c
], t
);
7581 spa_history_log_internal(spa
, "detach", tx
,
7582 "vdev=%s", vml
[c
]->vdev_path
);
7587 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
7588 vdev_config_dirty(spa
->spa_root_vdev
);
7589 spa
->spa_config_splitting
= NULL
;
7593 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
7595 if (zio_injection_enabled
)
7596 zio_handle_panic_injection(spa
, FTAG
, 3);
7598 /* split is complete; log a history record */
7599 spa_history_log_internal(newspa
, "split", NULL
,
7600 "from pool %s", spa_name(spa
));
7602 newspa
->spa_is_splitting
= B_FALSE
;
7603 kmem_free(vml
, children
* sizeof (vdev_t
*));
7605 /* if we're not going to mount the filesystems in userland, export */
7607 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
7614 spa_deactivate(newspa
);
7617 txg
= spa_vdev_config_enter(spa
);
7619 /* re-online all offlined disks */
7620 for (c
= 0; c
< children
; c
++) {
7622 vml
[c
]->vdev_offline
= B_FALSE
;
7625 /* restart initializing or trimming disks as necessary */
7626 spa_async_request(spa
, SPA_ASYNC_INITIALIZE_RESTART
);
7627 spa_async_request(spa
, SPA_ASYNC_TRIM_RESTART
);
7628 spa_async_request(spa
, SPA_ASYNC_AUTOTRIM_RESTART
);
7630 vdev_reopen(spa
->spa_root_vdev
);
7632 nvlist_free(spa
->spa_config_splitting
);
7633 spa
->spa_config_splitting
= NULL
;
7634 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
7636 kmem_free(vml
, children
* sizeof (vdev_t
*));
7641 * Find any device that's done replacing, or a vdev marked 'unspare' that's
7642 * currently spared, so we can detach it.
7645 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
7647 vdev_t
*newvd
, *oldvd
;
7649 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
7650 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
7656 * Check for a completed replacement. We always consider the first
7657 * vdev in the list to be the oldest vdev, and the last one to be
7658 * the newest (see spa_vdev_attach() for how that works). In
7659 * the case where the newest vdev is faulted, we will not automatically
7660 * remove it after a resilver completes. This is OK as it will require
7661 * user intervention to determine which disk the admin wishes to keep.
7663 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
7664 ASSERT(vd
->vdev_children
> 1);
7666 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
7667 oldvd
= vd
->vdev_child
[0];
7669 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
7670 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
7671 !vdev_dtl_required(oldvd
))
7676 * Check for a completed resilver with the 'unspare' flag set.
7677 * Also potentially update faulted state.
7679 if (vd
->vdev_ops
== &vdev_spare_ops
) {
7680 vdev_t
*first
= vd
->vdev_child
[0];
7681 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
7683 if (last
->vdev_unspare
) {
7686 } else if (first
->vdev_unspare
) {
7693 if (oldvd
!= NULL
&&
7694 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
7695 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
7696 !vdev_dtl_required(oldvd
))
7699 vdev_propagate_state(vd
);
7702 * If there are more than two spares attached to a disk,
7703 * and those spares are not required, then we want to
7704 * attempt to free them up now so that they can be used
7705 * by other pools. Once we're back down to a single
7706 * disk+spare, we stop removing them.
7708 if (vd
->vdev_children
> 2) {
7709 newvd
= vd
->vdev_child
[1];
7711 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
7712 vdev_dtl_empty(last
, DTL_MISSING
) &&
7713 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
7714 !vdev_dtl_required(newvd
))
7723 spa_vdev_resilver_done(spa_t
*spa
)
7725 vdev_t
*vd
, *pvd
, *ppvd
;
7726 uint64_t guid
, sguid
, pguid
, ppguid
;
7728 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7730 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
7731 pvd
= vd
->vdev_parent
;
7732 ppvd
= pvd
->vdev_parent
;
7733 guid
= vd
->vdev_guid
;
7734 pguid
= pvd
->vdev_guid
;
7735 ppguid
= ppvd
->vdev_guid
;
7738 * If we have just finished replacing a hot spared device, then
7739 * we need to detach the parent's first child (the original hot
7742 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
7743 ppvd
->vdev_children
== 2) {
7744 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
7745 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
7747 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
7749 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7750 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
7752 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
7754 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7757 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7760 * If a detach was not performed above replace waiters will not have
7761 * been notified. In which case we must do so now.
7763 spa_notify_waiters(spa
);
7767 * Update the stored path or FRU for this vdev.
7770 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
7774 boolean_t sync
= B_FALSE
;
7776 ASSERT(spa_writeable(spa
));
7778 spa_vdev_state_enter(spa
, SCL_ALL
);
7780 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
7781 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
7783 if (!vd
->vdev_ops
->vdev_op_leaf
)
7784 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
7787 if (strcmp(value
, vd
->vdev_path
) != 0) {
7788 spa_strfree(vd
->vdev_path
);
7789 vd
->vdev_path
= spa_strdup(value
);
7793 if (vd
->vdev_fru
== NULL
) {
7794 vd
->vdev_fru
= spa_strdup(value
);
7796 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
7797 spa_strfree(vd
->vdev_fru
);
7798 vd
->vdev_fru
= spa_strdup(value
);
7803 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
7807 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
7809 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
7813 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
7815 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
7819 * ==========================================================================
7821 * ==========================================================================
7824 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
7826 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
7828 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
7829 return (SET_ERROR(EBUSY
));
7831 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
7835 spa_scan_stop(spa_t
*spa
)
7837 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
7838 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
7839 return (SET_ERROR(EBUSY
));
7840 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
7844 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
7846 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
7848 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
7849 return (SET_ERROR(ENOTSUP
));
7851 if (func
== POOL_SCAN_RESILVER
&&
7852 !spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
))
7853 return (SET_ERROR(ENOTSUP
));
7856 * If a resilver was requested, but there is no DTL on a
7857 * writeable leaf device, we have nothing to do.
7859 if (func
== POOL_SCAN_RESILVER
&&
7860 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
7861 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
7865 return (dsl_scan(spa
->spa_dsl_pool
, func
));
7869 * ==========================================================================
7870 * SPA async task processing
7871 * ==========================================================================
7875 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
7877 if (vd
->vdev_remove_wanted
) {
7878 vd
->vdev_remove_wanted
= B_FALSE
;
7879 vd
->vdev_delayed_close
= B_FALSE
;
7880 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
7883 * We want to clear the stats, but we don't want to do a full
7884 * vdev_clear() as that will cause us to throw away
7885 * degraded/faulted state as well as attempt to reopen the
7886 * device, all of which is a waste.
7888 vd
->vdev_stat
.vs_read_errors
= 0;
7889 vd
->vdev_stat
.vs_write_errors
= 0;
7890 vd
->vdev_stat
.vs_checksum_errors
= 0;
7892 vdev_state_dirty(vd
->vdev_top
);
7895 for (int c
= 0; c
< vd
->vdev_children
; c
++)
7896 spa_async_remove(spa
, vd
->vdev_child
[c
]);
7900 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
7902 if (vd
->vdev_probe_wanted
) {
7903 vd
->vdev_probe_wanted
= B_FALSE
;
7904 vdev_reopen(vd
); /* vdev_open() does the actual probe */
7907 for (int c
= 0; c
< vd
->vdev_children
; c
++)
7908 spa_async_probe(spa
, vd
->vdev_child
[c
]);
7912 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
7914 if (!spa
->spa_autoexpand
)
7917 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
7918 vdev_t
*cvd
= vd
->vdev_child
[c
];
7919 spa_async_autoexpand(spa
, cvd
);
7922 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
7925 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
7929 spa_async_thread(void *arg
)
7931 spa_t
*spa
= (spa_t
*)arg
;
7932 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7935 ASSERT(spa
->spa_sync_on
);
7937 mutex_enter(&spa
->spa_async_lock
);
7938 tasks
= spa
->spa_async_tasks
;
7939 spa
->spa_async_tasks
= 0;
7940 mutex_exit(&spa
->spa_async_lock
);
7943 * See if the config needs to be updated.
7945 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
7946 uint64_t old_space
, new_space
;
7948 mutex_enter(&spa_namespace_lock
);
7949 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
7950 old_space
+= metaslab_class_get_space(spa_special_class(spa
));
7951 old_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
7953 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
7955 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
7956 new_space
+= metaslab_class_get_space(spa_special_class(spa
));
7957 new_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
7958 mutex_exit(&spa_namespace_lock
);
7961 * If the pool grew as a result of the config update,
7962 * then log an internal history event.
7964 if (new_space
!= old_space
) {
7965 spa_history_log_internal(spa
, "vdev online", NULL
,
7966 "pool '%s' size: %llu(+%llu)",
7967 spa_name(spa
), (u_longlong_t
)new_space
,
7968 (u_longlong_t
)(new_space
- old_space
));
7973 * See if any devices need to be marked REMOVED.
7975 if (tasks
& SPA_ASYNC_REMOVE
) {
7976 spa_vdev_state_enter(spa
, SCL_NONE
);
7977 spa_async_remove(spa
, spa
->spa_root_vdev
);
7978 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
7979 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
7980 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
7981 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
7982 (void) spa_vdev_state_exit(spa
, NULL
, 0);
7985 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
7986 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
7987 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
7988 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7992 * See if any devices need to be probed.
7994 if (tasks
& SPA_ASYNC_PROBE
) {
7995 spa_vdev_state_enter(spa
, SCL_NONE
);
7996 spa_async_probe(spa
, spa
->spa_root_vdev
);
7997 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8001 * If any devices are done replacing, detach them.
8003 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
8004 spa_vdev_resilver_done(spa
);
8007 * If any devices are done replacing, detach them. Then if no
8008 * top-level vdevs are rebuilding attempt to kick off a scrub.
8010 if (tasks
& SPA_ASYNC_REBUILD_DONE
) {
8011 spa_vdev_resilver_done(spa
);
8013 if (!vdev_rebuild_active(spa
->spa_root_vdev
))
8014 (void) dsl_scan(spa
->spa_dsl_pool
, POOL_SCAN_SCRUB
);
8018 * Kick off a resilver.
8020 if (tasks
& SPA_ASYNC_RESILVER
&&
8021 !vdev_rebuild_active(spa
->spa_root_vdev
) &&
8022 (!dsl_scan_resilvering(dp
) ||
8023 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_RESILVER_DEFER
)))
8024 dsl_scan_restart_resilver(dp
, 0);
8026 if (tasks
& SPA_ASYNC_INITIALIZE_RESTART
) {
8027 mutex_enter(&spa_namespace_lock
);
8028 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8029 vdev_initialize_restart(spa
->spa_root_vdev
);
8030 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8031 mutex_exit(&spa_namespace_lock
);
8034 if (tasks
& SPA_ASYNC_TRIM_RESTART
) {
8035 mutex_enter(&spa_namespace_lock
);
8036 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8037 vdev_trim_restart(spa
->spa_root_vdev
);
8038 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8039 mutex_exit(&spa_namespace_lock
);
8042 if (tasks
& SPA_ASYNC_AUTOTRIM_RESTART
) {
8043 mutex_enter(&spa_namespace_lock
);
8044 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8045 vdev_autotrim_restart(spa
);
8046 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8047 mutex_exit(&spa_namespace_lock
);
8051 * Kick off L2 cache whole device TRIM.
8053 if (tasks
& SPA_ASYNC_L2CACHE_TRIM
) {
8054 mutex_enter(&spa_namespace_lock
);
8055 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8056 vdev_trim_l2arc(spa
);
8057 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8058 mutex_exit(&spa_namespace_lock
);
8062 * Kick off L2 cache rebuilding.
8064 if (tasks
& SPA_ASYNC_L2CACHE_REBUILD
) {
8065 mutex_enter(&spa_namespace_lock
);
8066 spa_config_enter(spa
, SCL_L2ARC
, FTAG
, RW_READER
);
8067 l2arc_spa_rebuild_start(spa
);
8068 spa_config_exit(spa
, SCL_L2ARC
, FTAG
);
8069 mutex_exit(&spa_namespace_lock
);
8073 * Let the world know that we're done.
8075 mutex_enter(&spa
->spa_async_lock
);
8076 spa
->spa_async_thread
= NULL
;
8077 cv_broadcast(&spa
->spa_async_cv
);
8078 mutex_exit(&spa
->spa_async_lock
);
8083 spa_async_suspend(spa_t
*spa
)
8085 mutex_enter(&spa
->spa_async_lock
);
8086 spa
->spa_async_suspended
++;
8087 while (spa
->spa_async_thread
!= NULL
)
8088 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
8089 mutex_exit(&spa
->spa_async_lock
);
8091 spa_vdev_remove_suspend(spa
);
8093 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8094 if (condense_thread
!= NULL
)
8095 zthr_cancel(condense_thread
);
8097 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8098 if (discard_thread
!= NULL
)
8099 zthr_cancel(discard_thread
);
8101 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8102 if (ll_delete_thread
!= NULL
)
8103 zthr_cancel(ll_delete_thread
);
8105 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8106 if (ll_condense_thread
!= NULL
)
8107 zthr_cancel(ll_condense_thread
);
8111 spa_async_resume(spa_t
*spa
)
8113 mutex_enter(&spa
->spa_async_lock
);
8114 ASSERT(spa
->spa_async_suspended
!= 0);
8115 spa
->spa_async_suspended
--;
8116 mutex_exit(&spa
->spa_async_lock
);
8117 spa_restart_removal(spa
);
8119 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8120 if (condense_thread
!= NULL
)
8121 zthr_resume(condense_thread
);
8123 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8124 if (discard_thread
!= NULL
)
8125 zthr_resume(discard_thread
);
8127 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8128 if (ll_delete_thread
!= NULL
)
8129 zthr_resume(ll_delete_thread
);
8131 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8132 if (ll_condense_thread
!= NULL
)
8133 zthr_resume(ll_condense_thread
);
8137 spa_async_tasks_pending(spa_t
*spa
)
8139 uint_t non_config_tasks
;
8141 boolean_t config_task_suspended
;
8143 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
8144 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
8145 if (spa
->spa_ccw_fail_time
== 0) {
8146 config_task_suspended
= B_FALSE
;
8148 config_task_suspended
=
8149 (gethrtime() - spa
->spa_ccw_fail_time
) <
8150 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
8153 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
8157 spa_async_dispatch(spa_t
*spa
)
8159 mutex_enter(&spa
->spa_async_lock
);
8160 if (spa_async_tasks_pending(spa
) &&
8161 !spa
->spa_async_suspended
&&
8162 spa
->spa_async_thread
== NULL
)
8163 spa
->spa_async_thread
= thread_create(NULL
, 0,
8164 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
8165 mutex_exit(&spa
->spa_async_lock
);
8169 spa_async_request(spa_t
*spa
, int task
)
8171 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
8172 mutex_enter(&spa
->spa_async_lock
);
8173 spa
->spa_async_tasks
|= task
;
8174 mutex_exit(&spa
->spa_async_lock
);
8178 spa_async_tasks(spa_t
*spa
)
8180 return (spa
->spa_async_tasks
);
8184 * ==========================================================================
8185 * SPA syncing routines
8186 * ==========================================================================
8191 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8195 bpobj_enqueue(bpo
, bp
, bp_freed
, tx
);
8200 bpobj_enqueue_alloc_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8202 return (bpobj_enqueue_cb(arg
, bp
, B_FALSE
, tx
));
8206 bpobj_enqueue_free_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8208 return (bpobj_enqueue_cb(arg
, bp
, B_TRUE
, tx
));
8212 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8216 zio_nowait(zio_free_sync(pio
, pio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
8222 bpobj_spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8226 return (spa_free_sync_cb(arg
, bp
, tx
));
8230 * Note: this simple function is not inlined to make it easier to dtrace the
8231 * amount of time spent syncing frees.
8234 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
8236 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8237 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
8238 VERIFY(zio_wait(zio
) == 0);
8242 * Note: this simple function is not inlined to make it easier to dtrace the
8243 * amount of time spent syncing deferred frees.
8246 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
8248 if (spa_sync_pass(spa
) != 1)
8253 * If the log space map feature is active, we stop deferring
8254 * frees to the next TXG and therefore running this function
8255 * would be considered a no-op as spa_deferred_bpobj should
8256 * not have any entries.
8258 * That said we run this function anyway (instead of returning
8259 * immediately) for the edge-case scenario where we just
8260 * activated the log space map feature in this TXG but we have
8261 * deferred frees from the previous TXG.
8263 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8264 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
8265 bpobj_spa_free_sync_cb
, zio
, tx
), ==, 0);
8266 VERIFY0(zio_wait(zio
));
8270 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
8272 char *packed
= NULL
;
8277 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
8280 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
8281 * information. This avoids the dmu_buf_will_dirty() path and
8282 * saves us a pre-read to get data we don't actually care about.
8284 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
8285 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
8287 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
8289 bzero(packed
+ nvsize
, bufsize
- nvsize
);
8291 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
8293 vmem_free(packed
, bufsize
);
8295 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
8296 dmu_buf_will_dirty(db
, tx
);
8297 *(uint64_t *)db
->db_data
= nvsize
;
8298 dmu_buf_rele(db
, FTAG
);
8302 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
8303 const char *config
, const char *entry
)
8313 * Update the MOS nvlist describing the list of available devices.
8314 * spa_validate_aux() will have already made sure this nvlist is
8315 * valid and the vdevs are labeled appropriately.
8317 if (sav
->sav_object
== 0) {
8318 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
8319 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
8320 sizeof (uint64_t), tx
);
8321 VERIFY(zap_update(spa
->spa_meta_objset
,
8322 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
8323 &sav
->sav_object
, tx
) == 0);
8326 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
8327 if (sav
->sav_count
== 0) {
8328 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
8330 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
8331 for (i
= 0; i
< sav
->sav_count
; i
++)
8332 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
8333 B_FALSE
, VDEV_CONFIG_L2CACHE
);
8334 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
8335 sav
->sav_count
) == 0);
8336 for (i
= 0; i
< sav
->sav_count
; i
++)
8337 nvlist_free(list
[i
]);
8338 kmem_free(list
, sav
->sav_count
* sizeof (void *));
8341 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
8342 nvlist_free(nvroot
);
8344 sav
->sav_sync
= B_FALSE
;
8348 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
8349 * The all-vdev ZAP must be empty.
8352 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
8354 spa_t
*spa
= vd
->vdev_spa
;
8356 if (vd
->vdev_top_zap
!= 0) {
8357 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8358 vd
->vdev_top_zap
, tx
));
8360 if (vd
->vdev_leaf_zap
!= 0) {
8361 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8362 vd
->vdev_leaf_zap
, tx
));
8364 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
8365 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
8370 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
8375 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
8376 * its config may not be dirty but we still need to build per-vdev ZAPs.
8377 * Similarly, if the pool is being assembled (e.g. after a split), we
8378 * need to rebuild the AVZ although the config may not be dirty.
8380 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
8381 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
8384 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8386 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
8387 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
8388 spa
->spa_all_vdev_zaps
!= 0);
8390 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
8391 /* Make and build the new AVZ */
8392 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
8393 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
8394 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
8396 /* Diff old AVZ with new one */
8400 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
8401 spa
->spa_all_vdev_zaps
);
8402 zap_cursor_retrieve(&zc
, &za
) == 0;
8403 zap_cursor_advance(&zc
)) {
8404 uint64_t vdzap
= za
.za_first_integer
;
8405 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
8408 * ZAP is listed in old AVZ but not in new one;
8411 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
8416 zap_cursor_fini(&zc
);
8418 /* Destroy the old AVZ */
8419 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
8420 spa
->spa_all_vdev_zaps
, tx
));
8422 /* Replace the old AVZ in the dir obj with the new one */
8423 VERIFY0(zap_update(spa
->spa_meta_objset
,
8424 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
8425 sizeof (new_avz
), 1, &new_avz
, tx
));
8427 spa
->spa_all_vdev_zaps
= new_avz
;
8428 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
8432 /* Walk through the AVZ and destroy all listed ZAPs */
8433 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
8434 spa
->spa_all_vdev_zaps
);
8435 zap_cursor_retrieve(&zc
, &za
) == 0;
8436 zap_cursor_advance(&zc
)) {
8437 uint64_t zap
= za
.za_first_integer
;
8438 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
8441 zap_cursor_fini(&zc
);
8443 /* Destroy and unlink the AVZ itself */
8444 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
8445 spa
->spa_all_vdev_zaps
, tx
));
8446 VERIFY0(zap_remove(spa
->spa_meta_objset
,
8447 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
8448 spa
->spa_all_vdev_zaps
= 0;
8451 if (spa
->spa_all_vdev_zaps
== 0) {
8452 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
8453 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
8454 DMU_POOL_VDEV_ZAP_MAP
, tx
);
8456 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
8458 /* Create ZAPs for vdevs that don't have them. */
8459 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
8461 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
8462 dmu_tx_get_txg(tx
), B_FALSE
);
8465 * If we're upgrading the spa version then make sure that
8466 * the config object gets updated with the correct version.
8468 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
8469 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
8470 spa
->spa_uberblock
.ub_version
);
8472 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8474 nvlist_free(spa
->spa_config_syncing
);
8475 spa
->spa_config_syncing
= config
;
8477 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
8481 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
8483 uint64_t *versionp
= arg
;
8484 uint64_t version
= *versionp
;
8485 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
8488 * Setting the version is special cased when first creating the pool.
8490 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
8492 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
8493 ASSERT(version
>= spa_version(spa
));
8495 spa
->spa_uberblock
.ub_version
= version
;
8496 vdev_config_dirty(spa
->spa_root_vdev
);
8497 spa_history_log_internal(spa
, "set", tx
, "version=%lld",
8498 (longlong_t
)version
);
8502 * Set zpool properties.
8505 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
8507 nvlist_t
*nvp
= arg
;
8508 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
8509 objset_t
*mos
= spa
->spa_meta_objset
;
8510 nvpair_t
*elem
= NULL
;
8512 mutex_enter(&spa
->spa_props_lock
);
8514 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
8516 char *strval
, *fname
;
8518 const char *propname
;
8519 zprop_type_t proptype
;
8522 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
8523 case ZPOOL_PROP_INVAL
:
8525 * We checked this earlier in spa_prop_validate().
8527 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
8529 fname
= strchr(nvpair_name(elem
), '@') + 1;
8530 VERIFY0(zfeature_lookup_name(fname
, &fid
));
8532 spa_feature_enable(spa
, fid
, tx
);
8533 spa_history_log_internal(spa
, "set", tx
,
8534 "%s=enabled", nvpair_name(elem
));
8537 case ZPOOL_PROP_VERSION
:
8538 intval
= fnvpair_value_uint64(elem
);
8540 * The version is synced separately before other
8541 * properties and should be correct by now.
8543 ASSERT3U(spa_version(spa
), >=, intval
);
8546 case ZPOOL_PROP_ALTROOT
:
8548 * 'altroot' is a non-persistent property. It should
8549 * have been set temporarily at creation or import time.
8551 ASSERT(spa
->spa_root
!= NULL
);
8554 case ZPOOL_PROP_READONLY
:
8555 case ZPOOL_PROP_CACHEFILE
:
8557 * 'readonly' and 'cachefile' are also non-persistent
8561 case ZPOOL_PROP_COMMENT
:
8562 strval
= fnvpair_value_string(elem
);
8563 if (spa
->spa_comment
!= NULL
)
8564 spa_strfree(spa
->spa_comment
);
8565 spa
->spa_comment
= spa_strdup(strval
);
8567 * We need to dirty the configuration on all the vdevs
8568 * so that their labels get updated. It's unnecessary
8569 * to do this for pool creation since the vdev's
8570 * configuration has already been dirtied.
8572 if (tx
->tx_txg
!= TXG_INITIAL
)
8573 vdev_config_dirty(spa
->spa_root_vdev
);
8574 spa_history_log_internal(spa
, "set", tx
,
8575 "%s=%s", nvpair_name(elem
), strval
);
8579 * Set pool property values in the poolprops mos object.
8581 if (spa
->spa_pool_props_object
== 0) {
8582 spa
->spa_pool_props_object
=
8583 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
8584 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
8588 /* normalize the property name */
8589 propname
= zpool_prop_to_name(prop
);
8590 proptype
= zpool_prop_get_type(prop
);
8592 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
8593 ASSERT(proptype
== PROP_TYPE_STRING
);
8594 strval
= fnvpair_value_string(elem
);
8595 VERIFY0(zap_update(mos
,
8596 spa
->spa_pool_props_object
, propname
,
8597 1, strlen(strval
) + 1, strval
, tx
));
8598 spa_history_log_internal(spa
, "set", tx
,
8599 "%s=%s", nvpair_name(elem
), strval
);
8600 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
8601 intval
= fnvpair_value_uint64(elem
);
8603 if (proptype
== PROP_TYPE_INDEX
) {
8605 VERIFY0(zpool_prop_index_to_string(
8606 prop
, intval
, &unused
));
8608 VERIFY0(zap_update(mos
,
8609 spa
->spa_pool_props_object
, propname
,
8610 8, 1, &intval
, tx
));
8611 spa_history_log_internal(spa
, "set", tx
,
8612 "%s=%lld", nvpair_name(elem
),
8613 (longlong_t
)intval
);
8615 ASSERT(0); /* not allowed */
8619 case ZPOOL_PROP_DELEGATION
:
8620 spa
->spa_delegation
= intval
;
8622 case ZPOOL_PROP_BOOTFS
:
8623 spa
->spa_bootfs
= intval
;
8625 case ZPOOL_PROP_FAILUREMODE
:
8626 spa
->spa_failmode
= intval
;
8628 case ZPOOL_PROP_AUTOTRIM
:
8629 spa
->spa_autotrim
= intval
;
8630 spa_async_request(spa
,
8631 SPA_ASYNC_AUTOTRIM_RESTART
);
8633 case ZPOOL_PROP_AUTOEXPAND
:
8634 spa
->spa_autoexpand
= intval
;
8635 if (tx
->tx_txg
!= TXG_INITIAL
)
8636 spa_async_request(spa
,
8637 SPA_ASYNC_AUTOEXPAND
);
8639 case ZPOOL_PROP_MULTIHOST
:
8640 spa
->spa_multihost
= intval
;
8649 mutex_exit(&spa
->spa_props_lock
);
8653 * Perform one-time upgrade on-disk changes. spa_version() does not
8654 * reflect the new version this txg, so there must be no changes this
8655 * txg to anything that the upgrade code depends on after it executes.
8656 * Therefore this must be called after dsl_pool_sync() does the sync
8660 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
8662 if (spa_sync_pass(spa
) != 1)
8665 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8666 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
8668 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
8669 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
8670 dsl_pool_create_origin(dp
, tx
);
8672 /* Keeping the origin open increases spa_minref */
8673 spa
->spa_minref
+= 3;
8676 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
8677 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
8678 dsl_pool_upgrade_clones(dp
, tx
);
8681 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
8682 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
8683 dsl_pool_upgrade_dir_clones(dp
, tx
);
8685 /* Keeping the freedir open increases spa_minref */
8686 spa
->spa_minref
+= 3;
8689 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
8690 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
8691 spa_feature_create_zap_objects(spa
, tx
);
8695 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
8696 * when possibility to use lz4 compression for metadata was added
8697 * Old pools that have this feature enabled must be upgraded to have
8698 * this feature active
8700 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
8701 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
8702 SPA_FEATURE_LZ4_COMPRESS
);
8703 boolean_t lz4_ac
= spa_feature_is_active(spa
,
8704 SPA_FEATURE_LZ4_COMPRESS
);
8706 if (lz4_en
&& !lz4_ac
)
8707 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
8711 * If we haven't written the salt, do so now. Note that the
8712 * feature may not be activated yet, but that's fine since
8713 * the presence of this ZAP entry is backwards compatible.
8715 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
8716 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
8717 VERIFY0(zap_add(spa
->spa_meta_objset
,
8718 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
8719 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
8720 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
8723 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
8727 vdev_indirect_state_sync_verify(vdev_t
*vd
)
8729 vdev_indirect_mapping_t
*vim __maybe_unused
= vd
->vdev_indirect_mapping
;
8730 vdev_indirect_births_t
*vib __maybe_unused
= vd
->vdev_indirect_births
;
8732 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
8733 ASSERT(vim
!= NULL
);
8734 ASSERT(vib
!= NULL
);
8737 uint64_t obsolete_sm_object
= 0;
8738 ASSERT0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
8739 if (obsolete_sm_object
!= 0) {
8740 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
8741 ASSERT(vd
->vdev_removing
||
8742 vd
->vdev_ops
== &vdev_indirect_ops
);
8743 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
8744 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
8745 ASSERT3U(obsolete_sm_object
, ==,
8746 space_map_object(vd
->vdev_obsolete_sm
));
8747 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
8748 space_map_allocated(vd
->vdev_obsolete_sm
));
8750 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
8753 * Since frees / remaps to an indirect vdev can only
8754 * happen in syncing context, the obsolete segments
8755 * tree must be empty when we start syncing.
8757 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
8761 * Set the top-level vdev's max queue depth. Evaluate each top-level's
8762 * async write queue depth in case it changed. The max queue depth will
8763 * not change in the middle of syncing out this txg.
8766 spa_sync_adjust_vdev_max_queue_depth(spa_t
*spa
)
8768 ASSERT(spa_writeable(spa
));
8770 vdev_t
*rvd
= spa
->spa_root_vdev
;
8771 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
8772 zfs_vdev_queue_depth_pct
/ 100;
8773 metaslab_class_t
*normal
= spa_normal_class(spa
);
8774 metaslab_class_t
*special
= spa_special_class(spa
);
8775 metaslab_class_t
*dedup
= spa_dedup_class(spa
);
8777 uint64_t slots_per_allocator
= 0;
8778 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
8779 vdev_t
*tvd
= rvd
->vdev_child
[c
];
8781 metaslab_group_t
*mg
= tvd
->vdev_mg
;
8782 if (mg
== NULL
|| !metaslab_group_initialized(mg
))
8785 metaslab_class_t
*mc
= mg
->mg_class
;
8786 if (mc
!= normal
&& mc
!= special
&& mc
!= dedup
)
8790 * It is safe to do a lock-free check here because only async
8791 * allocations look at mg_max_alloc_queue_depth, and async
8792 * allocations all happen from spa_sync().
8794 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
8795 ASSERT0(zfs_refcount_count(
8796 &(mg
->mg_allocator
[i
].mga_alloc_queue_depth
)));
8798 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
8800 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
8801 mg
->mg_allocator
[i
].mga_cur_max_alloc_queue_depth
=
8802 zfs_vdev_def_queue_depth
;
8804 slots_per_allocator
+= zfs_vdev_def_queue_depth
;
8807 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
8808 ASSERT0(zfs_refcount_count(&normal
->mc_alloc_slots
[i
]));
8809 ASSERT0(zfs_refcount_count(&special
->mc_alloc_slots
[i
]));
8810 ASSERT0(zfs_refcount_count(&dedup
->mc_alloc_slots
[i
]));
8811 normal
->mc_alloc_max_slots
[i
] = slots_per_allocator
;
8812 special
->mc_alloc_max_slots
[i
] = slots_per_allocator
;
8813 dedup
->mc_alloc_max_slots
[i
] = slots_per_allocator
;
8815 normal
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
8816 special
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
8817 dedup
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
8821 spa_sync_condense_indirect(spa_t
*spa
, dmu_tx_t
*tx
)
8823 ASSERT(spa_writeable(spa
));
8825 vdev_t
*rvd
= spa
->spa_root_vdev
;
8826 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
8827 vdev_t
*vd
= rvd
->vdev_child
[c
];
8828 vdev_indirect_state_sync_verify(vd
);
8830 if (vdev_indirect_should_condense(vd
)) {
8831 spa_condense_indirect_start_sync(vd
, tx
);
8838 spa_sync_iterate_to_convergence(spa_t
*spa
, dmu_tx_t
*tx
)
8840 objset_t
*mos
= spa
->spa_meta_objset
;
8841 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8842 uint64_t txg
= tx
->tx_txg
;
8843 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
8846 int pass
= ++spa
->spa_sync_pass
;
8848 spa_sync_config_object(spa
, tx
);
8849 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
8850 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
8851 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
8852 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
8853 spa_errlog_sync(spa
, txg
);
8854 dsl_pool_sync(dp
, txg
);
8856 if (pass
< zfs_sync_pass_deferred_free
||
8857 spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
)) {
8859 * If the log space map feature is active we don't
8860 * care about deferred frees and the deferred bpobj
8861 * as the log space map should effectively have the
8862 * same results (i.e. appending only to one object).
8864 spa_sync_frees(spa
, free_bpl
, tx
);
8867 * We can not defer frees in pass 1, because
8868 * we sync the deferred frees later in pass 1.
8870 ASSERT3U(pass
, >, 1);
8871 bplist_iterate(free_bpl
, bpobj_enqueue_alloc_cb
,
8872 &spa
->spa_deferred_bpobj
, tx
);
8876 dsl_scan_sync(dp
, tx
);
8878 spa_sync_upgrades(spa
, tx
);
8880 spa_flush_metaslabs(spa
, tx
);
8883 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
8888 * Note: We need to check if the MOS is dirty because we could
8889 * have marked the MOS dirty without updating the uberblock
8890 * (e.g. if we have sync tasks but no dirty user data). We need
8891 * to check the uberblock's rootbp because it is updated if we
8892 * have synced out dirty data (though in this case the MOS will
8893 * most likely also be dirty due to second order effects, we
8894 * don't want to rely on that here).
8897 spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
8898 !dmu_objset_is_dirty(mos
, txg
)) {
8900 * Nothing changed on the first pass, therefore this
8901 * TXG is a no-op. Avoid syncing deferred frees, so
8902 * that we can keep this TXG as a no-op.
8904 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
8905 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
8906 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
8907 ASSERT(txg_list_empty(&dp
->dp_early_sync_tasks
, txg
));
8911 spa_sync_deferred_frees(spa
, tx
);
8912 } while (dmu_objset_is_dirty(mos
, txg
));
8916 * Rewrite the vdev configuration (which includes the uberblock) to
8917 * commit the transaction group.
8919 * If there are no dirty vdevs, we sync the uberblock to a few random
8920 * top-level vdevs that are known to be visible in the config cache
8921 * (see spa_vdev_add() for a complete description). If there *are* dirty
8922 * vdevs, sync the uberblock to all vdevs.
8925 spa_sync_rewrite_vdev_config(spa_t
*spa
, dmu_tx_t
*tx
)
8927 vdev_t
*rvd
= spa
->spa_root_vdev
;
8928 uint64_t txg
= tx
->tx_txg
;
8934 * We hold SCL_STATE to prevent vdev open/close/etc.
8935 * while we're attempting to write the vdev labels.
8937 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8939 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
8940 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
8942 int children
= rvd
->vdev_children
;
8943 int c0
= spa_get_random(children
);
8945 for (int c
= 0; c
< children
; c
++) {
8947 rvd
->vdev_child
[(c0
+ c
) % children
];
8949 /* Stop when revisiting the first vdev */
8950 if (c
> 0 && svd
[0] == vd
)
8953 if (vd
->vdev_ms_array
== 0 ||
8955 !vdev_is_concrete(vd
))
8958 svd
[svdcount
++] = vd
;
8959 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
8962 error
= vdev_config_sync(svd
, svdcount
, txg
);
8964 error
= vdev_config_sync(rvd
->vdev_child
,
8965 rvd
->vdev_children
, txg
);
8969 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
8971 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8975 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
8976 zio_resume_wait(spa
);
8981 * Sync the specified transaction group. New blocks may be dirtied as
8982 * part of the process, so we iterate until it converges.
8985 spa_sync(spa_t
*spa
, uint64_t txg
)
8989 VERIFY(spa_writeable(spa
));
8992 * Wait for i/os issued in open context that need to complete
8993 * before this txg syncs.
8995 (void) zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]);
8996 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
,
9000 * Lock out configuration changes.
9002 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
9004 spa
->spa_syncing_txg
= txg
;
9005 spa
->spa_sync_pass
= 0;
9007 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9008 mutex_enter(&spa
->spa_alloc_locks
[i
]);
9009 VERIFY0(avl_numnodes(&spa
->spa_alloc_trees
[i
]));
9010 mutex_exit(&spa
->spa_alloc_locks
[i
]);
9014 * If there are any pending vdev state changes, convert them
9015 * into config changes that go out with this transaction group.
9017 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9018 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
9020 * We need the write lock here because, for aux vdevs,
9021 * calling vdev_config_dirty() modifies sav_config.
9022 * This is ugly and will become unnecessary when we
9023 * eliminate the aux vdev wart by integrating all vdevs
9024 * into the root vdev tree.
9026 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9027 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
9028 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
9029 vdev_state_clean(vd
);
9030 vdev_config_dirty(vd
);
9032 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9033 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9035 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9037 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9038 dmu_tx_t
*tx
= dmu_tx_create_assigned(dp
, txg
);
9040 spa
->spa_sync_starttime
= gethrtime();
9041 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9042 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
9043 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
9044 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
9047 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9048 * set spa_deflate if we have no raid-z vdevs.
9050 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
9051 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
9052 vdev_t
*rvd
= spa
->spa_root_vdev
;
9055 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
9056 vd
= rvd
->vdev_child
[i
];
9057 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
9060 if (i
== rvd
->vdev_children
) {
9061 spa
->spa_deflate
= TRUE
;
9062 VERIFY0(zap_add(spa
->spa_meta_objset
,
9063 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
9064 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
9068 spa_sync_adjust_vdev_max_queue_depth(spa
);
9070 spa_sync_condense_indirect(spa
, tx
);
9072 spa_sync_iterate_to_convergence(spa
, tx
);
9075 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
9077 * Make sure that the number of ZAPs for all the vdevs matches
9078 * the number of ZAPs in the per-vdev ZAP list. This only gets
9079 * called if the config is dirty; otherwise there may be
9080 * outstanding AVZ operations that weren't completed in
9081 * spa_sync_config_object.
9083 uint64_t all_vdev_zap_entry_count
;
9084 ASSERT0(zap_count(spa
->spa_meta_objset
,
9085 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
9086 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
9087 all_vdev_zap_entry_count
);
9091 if (spa
->spa_vdev_removal
!= NULL
) {
9092 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
9095 spa_sync_rewrite_vdev_config(spa
, tx
);
9098 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9099 spa
->spa_deadman_tqid
= 0;
9102 * Clear the dirty config list.
9104 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
9105 vdev_config_clean(vd
);
9108 * Now that the new config has synced transactionally,
9109 * let it become visible to the config cache.
9111 if (spa
->spa_config_syncing
!= NULL
) {
9112 spa_config_set(spa
, spa
->spa_config_syncing
);
9113 spa
->spa_config_txg
= txg
;
9114 spa
->spa_config_syncing
= NULL
;
9117 dsl_pool_sync_done(dp
, txg
);
9119 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9120 mutex_enter(&spa
->spa_alloc_locks
[i
]);
9121 VERIFY0(avl_numnodes(&spa
->spa_alloc_trees
[i
]));
9122 mutex_exit(&spa
->spa_alloc_locks
[i
]);
9126 * Update usable space statistics.
9128 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
9130 vdev_sync_done(vd
, txg
);
9132 metaslab_class_evict_old(spa
->spa_normal_class
, txg
);
9133 metaslab_class_evict_old(spa
->spa_log_class
, txg
);
9135 spa_sync_close_syncing_log_sm(spa
);
9137 spa_update_dspace(spa
);
9140 * It had better be the case that we didn't dirty anything
9141 * since vdev_config_sync().
9143 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9144 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9145 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
9147 while (zfs_pause_spa_sync
)
9150 spa
->spa_sync_pass
= 0;
9153 * Update the last synced uberblock here. We want to do this at
9154 * the end of spa_sync() so that consumers of spa_last_synced_txg()
9155 * will be guaranteed that all the processing associated with
9156 * that txg has been completed.
9158 spa
->spa_ubsync
= spa
->spa_uberblock
;
9159 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
9161 spa_handle_ignored_writes(spa
);
9164 * If any async tasks have been requested, kick them off.
9166 spa_async_dispatch(spa
);
9170 * Sync all pools. We don't want to hold the namespace lock across these
9171 * operations, so we take a reference on the spa_t and drop the lock during the
9175 spa_sync_allpools(void)
9178 mutex_enter(&spa_namespace_lock
);
9179 while ((spa
= spa_next(spa
)) != NULL
) {
9180 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
9181 !spa_writeable(spa
) || spa_suspended(spa
))
9183 spa_open_ref(spa
, FTAG
);
9184 mutex_exit(&spa_namespace_lock
);
9185 txg_wait_synced(spa_get_dsl(spa
), 0);
9186 mutex_enter(&spa_namespace_lock
);
9187 spa_close(spa
, FTAG
);
9189 mutex_exit(&spa_namespace_lock
);
9193 * ==========================================================================
9194 * Miscellaneous routines
9195 * ==========================================================================
9199 * Remove all pools in the system.
9207 * Remove all cached state. All pools should be closed now,
9208 * so every spa in the AVL tree should be unreferenced.
9210 mutex_enter(&spa_namespace_lock
);
9211 while ((spa
= spa_next(NULL
)) != NULL
) {
9213 * Stop async tasks. The async thread may need to detach
9214 * a device that's been replaced, which requires grabbing
9215 * spa_namespace_lock, so we must drop it here.
9217 spa_open_ref(spa
, FTAG
);
9218 mutex_exit(&spa_namespace_lock
);
9219 spa_async_suspend(spa
);
9220 mutex_enter(&spa_namespace_lock
);
9221 spa_close(spa
, FTAG
);
9223 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
9225 spa_deactivate(spa
);
9229 mutex_exit(&spa_namespace_lock
);
9233 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
9238 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
9242 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
9243 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
9244 if (vd
->vdev_guid
== guid
)
9248 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
9249 vd
= spa
->spa_spares
.sav_vdevs
[i
];
9250 if (vd
->vdev_guid
== guid
)
9259 spa_upgrade(spa_t
*spa
, uint64_t version
)
9261 ASSERT(spa_writeable(spa
));
9263 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
9266 * This should only be called for a non-faulted pool, and since a
9267 * future version would result in an unopenable pool, this shouldn't be
9270 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
9271 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
9273 spa
->spa_uberblock
.ub_version
= version
;
9274 vdev_config_dirty(spa
->spa_root_vdev
);
9276 spa_config_exit(spa
, SCL_ALL
, FTAG
);
9278 txg_wait_synced(spa_get_dsl(spa
), 0);
9282 spa_has_spare(spa_t
*spa
, uint64_t guid
)
9286 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
9288 for (i
= 0; i
< sav
->sav_count
; i
++)
9289 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
9292 for (i
= 0; i
< sav
->sav_npending
; i
++) {
9293 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
9294 &spareguid
) == 0 && spareguid
== guid
)
9302 * Check if a pool has an active shared spare device.
9303 * Note: reference count of an active spare is 2, as a spare and as a replace
9306 spa_has_active_shared_spare(spa_t
*spa
)
9310 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
9312 for (i
= 0; i
< sav
->sav_count
; i
++) {
9313 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
9314 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
9323 spa_total_metaslabs(spa_t
*spa
)
9325 vdev_t
*rvd
= spa
->spa_root_vdev
;
9328 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
9329 vdev_t
*vd
= rvd
->vdev_child
[c
];
9330 if (!vdev_is_concrete(vd
))
9332 m
+= vd
->vdev_ms_count
;
9338 * Notify any waiting threads that some activity has switched from being in-
9339 * progress to not-in-progress so that the thread can wake up and determine
9340 * whether it is finished waiting.
9343 spa_notify_waiters(spa_t
*spa
)
9346 * Acquiring spa_activities_lock here prevents the cv_broadcast from
9347 * happening between the waiting thread's check and cv_wait.
9349 mutex_enter(&spa
->spa_activities_lock
);
9350 cv_broadcast(&spa
->spa_activities_cv
);
9351 mutex_exit(&spa
->spa_activities_lock
);
9355 * Notify any waiting threads that the pool is exporting, and then block until
9356 * they are finished using the spa_t.
9359 spa_wake_waiters(spa_t
*spa
)
9361 mutex_enter(&spa
->spa_activities_lock
);
9362 spa
->spa_waiters_cancel
= B_TRUE
;
9363 cv_broadcast(&spa
->spa_activities_cv
);
9364 while (spa
->spa_waiters
!= 0)
9365 cv_wait(&spa
->spa_waiters_cv
, &spa
->spa_activities_lock
);
9366 spa
->spa_waiters_cancel
= B_FALSE
;
9367 mutex_exit(&spa
->spa_activities_lock
);
9370 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
9372 spa_vdev_activity_in_progress_impl(vdev_t
*vd
, zpool_wait_activity_t activity
)
9374 spa_t
*spa
= vd
->vdev_spa
;
9376 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
));
9377 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
9378 ASSERT(activity
== ZPOOL_WAIT_INITIALIZE
||
9379 activity
== ZPOOL_WAIT_TRIM
);
9381 kmutex_t
*lock
= activity
== ZPOOL_WAIT_INITIALIZE
?
9382 &vd
->vdev_initialize_lock
: &vd
->vdev_trim_lock
;
9384 mutex_exit(&spa
->spa_activities_lock
);
9386 mutex_enter(&spa
->spa_activities_lock
);
9388 boolean_t in_progress
= (activity
== ZPOOL_WAIT_INITIALIZE
) ?
9389 (vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) :
9390 (vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
);
9396 for (int i
= 0; i
< vd
->vdev_children
; i
++) {
9397 if (spa_vdev_activity_in_progress_impl(vd
->vdev_child
[i
],
9406 * If use_guid is true, this checks whether the vdev specified by guid is
9407 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
9408 * is being initialized/trimmed. The caller must hold the config lock and
9409 * spa_activities_lock.
9412 spa_vdev_activity_in_progress(spa_t
*spa
, boolean_t use_guid
, uint64_t guid
,
9413 zpool_wait_activity_t activity
, boolean_t
*in_progress
)
9415 mutex_exit(&spa
->spa_activities_lock
);
9416 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9417 mutex_enter(&spa
->spa_activities_lock
);
9421 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
9422 if (vd
== NULL
|| !vd
->vdev_ops
->vdev_op_leaf
) {
9423 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9427 vd
= spa
->spa_root_vdev
;
9430 *in_progress
= spa_vdev_activity_in_progress_impl(vd
, activity
);
9432 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9437 * Locking for waiting threads
9438 * ---------------------------
9440 * Waiting threads need a way to check whether a given activity is in progress,
9441 * and then, if it is, wait for it to complete. Each activity will have some
9442 * in-memory representation of the relevant on-disk state which can be used to
9443 * determine whether or not the activity is in progress. The in-memory state and
9444 * the locking used to protect it will be different for each activity, and may
9445 * not be suitable for use with a cvar (e.g., some state is protected by the
9446 * config lock). To allow waiting threads to wait without any races, another
9447 * lock, spa_activities_lock, is used.
9449 * When the state is checked, both the activity-specific lock (if there is one)
9450 * and spa_activities_lock are held. In some cases, the activity-specific lock
9451 * is acquired explicitly (e.g. the config lock). In others, the locking is
9452 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
9453 * thread releases the activity-specific lock and, if the activity is in
9454 * progress, then cv_waits using spa_activities_lock.
9456 * The waiting thread is woken when another thread, one completing some
9457 * activity, updates the state of the activity and then calls
9458 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
9459 * needs to hold its activity-specific lock when updating the state, and this
9460 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
9462 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
9463 * and because it is held when the waiting thread checks the state of the
9464 * activity, it can never be the case that the completing thread both updates
9465 * the activity state and cv_broadcasts in between the waiting thread's check
9466 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
9468 * In order to prevent deadlock, when the waiting thread does its check, in some
9469 * cases it will temporarily drop spa_activities_lock in order to acquire the
9470 * activity-specific lock. The order in which spa_activities_lock and the
9471 * activity specific lock are acquired in the waiting thread is determined by
9472 * the order in which they are acquired in the completing thread; if the
9473 * completing thread calls spa_notify_waiters with the activity-specific lock
9474 * held, then the waiting thread must also acquire the activity-specific lock
9479 spa_activity_in_progress(spa_t
*spa
, zpool_wait_activity_t activity
,
9480 boolean_t use_tag
, uint64_t tag
, boolean_t
*in_progress
)
9484 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
9487 case ZPOOL_WAIT_CKPT_DISCARD
:
9489 (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
) &&
9490 zap_contains(spa_meta_objset(spa
),
9491 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ZPOOL_CHECKPOINT
) ==
9494 case ZPOOL_WAIT_FREE
:
9495 *in_progress
= ((spa_version(spa
) >= SPA_VERSION_DEADLISTS
&&
9496 !bpobj_is_empty(&spa
->spa_dsl_pool
->dp_free_bpobj
)) ||
9497 spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
) ||
9498 spa_livelist_delete_check(spa
));
9500 case ZPOOL_WAIT_INITIALIZE
:
9501 case ZPOOL_WAIT_TRIM
:
9502 error
= spa_vdev_activity_in_progress(spa
, use_tag
, tag
,
9503 activity
, in_progress
);
9505 case ZPOOL_WAIT_REPLACE
:
9506 mutex_exit(&spa
->spa_activities_lock
);
9507 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9508 mutex_enter(&spa
->spa_activities_lock
);
9510 *in_progress
= vdev_replace_in_progress(spa
->spa_root_vdev
);
9511 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9513 case ZPOOL_WAIT_REMOVE
:
9514 *in_progress
= (spa
->spa_removing_phys
.sr_state
==
9517 case ZPOOL_WAIT_RESILVER
:
9518 if ((*in_progress
= vdev_rebuild_active(spa
->spa_root_vdev
)))
9521 case ZPOOL_WAIT_SCRUB
:
9523 boolean_t scanning
, paused
, is_scrub
;
9524 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
9526 is_scrub
= (scn
->scn_phys
.scn_func
== POOL_SCAN_SCRUB
);
9527 scanning
= (scn
->scn_phys
.scn_state
== DSS_SCANNING
);
9528 paused
= dsl_scan_is_paused_scrub(scn
);
9529 *in_progress
= (scanning
&& !paused
&&
9530 is_scrub
== (activity
== ZPOOL_WAIT_SCRUB
));
9534 panic("unrecognized value for activity %d", activity
);
9541 spa_wait_common(const char *pool
, zpool_wait_activity_t activity
,
9542 boolean_t use_tag
, uint64_t tag
, boolean_t
*waited
)
9545 * The tag is used to distinguish between instances of an activity.
9546 * 'initialize' and 'trim' are the only activities that we use this for.
9547 * The other activities can only have a single instance in progress in a
9548 * pool at one time, making the tag unnecessary.
9550 * There can be multiple devices being replaced at once, but since they
9551 * all finish once resilvering finishes, we don't bother keeping track
9552 * of them individually, we just wait for them all to finish.
9554 if (use_tag
&& activity
!= ZPOOL_WAIT_INITIALIZE
&&
9555 activity
!= ZPOOL_WAIT_TRIM
)
9558 if (activity
< 0 || activity
>= ZPOOL_WAIT_NUM_ACTIVITIES
)
9562 int error
= spa_open(pool
, &spa
, FTAG
);
9567 * Increment the spa's waiter count so that we can call spa_close and
9568 * still ensure that the spa_t doesn't get freed before this thread is
9569 * finished with it when the pool is exported. We want to call spa_close
9570 * before we start waiting because otherwise the additional ref would
9571 * prevent the pool from being exported or destroyed throughout the
9572 * potentially long wait.
9574 mutex_enter(&spa
->spa_activities_lock
);
9576 spa_close(spa
, FTAG
);
9580 boolean_t in_progress
;
9581 error
= spa_activity_in_progress(spa
, activity
, use_tag
, tag
,
9584 if (error
|| !in_progress
|| spa
->spa_waiters_cancel
)
9589 if (cv_wait_sig(&spa
->spa_activities_cv
,
9590 &spa
->spa_activities_lock
) == 0) {
9597 cv_signal(&spa
->spa_waiters_cv
);
9598 mutex_exit(&spa
->spa_activities_lock
);
9604 * Wait for a particular instance of the specified activity to complete, where
9605 * the instance is identified by 'tag'
9608 spa_wait_tag(const char *pool
, zpool_wait_activity_t activity
, uint64_t tag
,
9611 return (spa_wait_common(pool
, activity
, B_TRUE
, tag
, waited
));
9615 * Wait for all instances of the specified activity complete
9618 spa_wait(const char *pool
, zpool_wait_activity_t activity
, boolean_t
*waited
)
9621 return (spa_wait_common(pool
, activity
, B_FALSE
, 0, waited
));
9625 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
9627 sysevent_t
*ev
= NULL
;
9631 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
9633 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
9634 ev
->resource
= resource
;
9641 spa_event_post(sysevent_t
*ev
)
9645 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
9646 kmem_free(ev
, sizeof (*ev
));
9652 * Post a zevent corresponding to the given sysevent. The 'name' must be one
9653 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
9654 * filled in from the spa and (optionally) the vdev. This doesn't do anything
9655 * in the userland libzpool, as we don't want consumers to misinterpret ztest
9656 * or zdb as real changes.
9659 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
9661 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
9664 /* state manipulation functions */
9665 EXPORT_SYMBOL(spa_open
);
9666 EXPORT_SYMBOL(spa_open_rewind
);
9667 EXPORT_SYMBOL(spa_get_stats
);
9668 EXPORT_SYMBOL(spa_create
);
9669 EXPORT_SYMBOL(spa_import
);
9670 EXPORT_SYMBOL(spa_tryimport
);
9671 EXPORT_SYMBOL(spa_destroy
);
9672 EXPORT_SYMBOL(spa_export
);
9673 EXPORT_SYMBOL(spa_reset
);
9674 EXPORT_SYMBOL(spa_async_request
);
9675 EXPORT_SYMBOL(spa_async_suspend
);
9676 EXPORT_SYMBOL(spa_async_resume
);
9677 EXPORT_SYMBOL(spa_inject_addref
);
9678 EXPORT_SYMBOL(spa_inject_delref
);
9679 EXPORT_SYMBOL(spa_scan_stat_init
);
9680 EXPORT_SYMBOL(spa_scan_get_stats
);
9682 /* device manipulation */
9683 EXPORT_SYMBOL(spa_vdev_add
);
9684 EXPORT_SYMBOL(spa_vdev_attach
);
9685 EXPORT_SYMBOL(spa_vdev_detach
);
9686 EXPORT_SYMBOL(spa_vdev_setpath
);
9687 EXPORT_SYMBOL(spa_vdev_setfru
);
9688 EXPORT_SYMBOL(spa_vdev_split_mirror
);
9690 /* spare statech is global across all pools) */
9691 EXPORT_SYMBOL(spa_spare_add
);
9692 EXPORT_SYMBOL(spa_spare_remove
);
9693 EXPORT_SYMBOL(spa_spare_exists
);
9694 EXPORT_SYMBOL(spa_spare_activate
);
9696 /* L2ARC statech is global across all pools) */
9697 EXPORT_SYMBOL(spa_l2cache_add
);
9698 EXPORT_SYMBOL(spa_l2cache_remove
);
9699 EXPORT_SYMBOL(spa_l2cache_exists
);
9700 EXPORT_SYMBOL(spa_l2cache_activate
);
9701 EXPORT_SYMBOL(spa_l2cache_drop
);
9704 EXPORT_SYMBOL(spa_scan
);
9705 EXPORT_SYMBOL(spa_scan_stop
);
9708 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
9709 EXPORT_SYMBOL(spa_sync_allpools
);
9712 EXPORT_SYMBOL(spa_prop_set
);
9713 EXPORT_SYMBOL(spa_prop_get
);
9714 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
9716 /* asynchronous event notification */
9717 EXPORT_SYMBOL(spa_event_notify
);
9720 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_shift
, INT
, ZMOD_RW
,
9721 "log2(fraction of arc that can be used by inflight I/Os when "
9722 "verifying pool during import");
9724 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_metadata
, INT
, ZMOD_RW
,
9725 "Set to traverse metadata on pool import");
9727 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_data
, INT
, ZMOD_RW
,
9728 "Set to traverse data on pool import");
9730 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_print_vdev_tree
, INT
, ZMOD_RW
,
9731 "Print vdev tree to zfs_dbgmsg during pool import");
9733 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_pct
, UINT
, ZMOD_RD
,
9734 "Percentage of CPUs to run an IO worker thread");
9736 ZFS_MODULE_PARAM(zfs
, zfs_
, max_missing_tvds
, ULONG
, ZMOD_RW
,
9737 "Allow importing pool with up to this number of missing top-level "
9738 "vdevs (in read-only mode)");
9740 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_pause
, INT
, ZMOD_RW
,
9741 "Set the livelist condense zthr to pause");
9743 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_pause
, INT
, ZMOD_RW
,
9744 "Set the livelist condense synctask to pause");
9746 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_cancel
, INT
, ZMOD_RW
,
9747 "Whether livelist condensing was canceled in the synctask");
9749 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_cancel
, INT
, ZMOD_RW
,
9750 "Whether livelist condensing was canceled in the zthr function");
9752 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, new_alloc
, INT
, ZMOD_RW
,
9753 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
9754 "was being condensed");