4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2018, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2016 Toomas Soome <tsoome@me.com>
30 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
31 * Copyright 2018 Joyent, Inc.
32 * Copyright (c) 2017, 2019, Datto Inc. All rights reserved.
33 * Copyright 2017 Joyent, Inc.
34 * Copyright (c) 2017, Intel Corporation.
35 * Copyright (c) 2021, Colm Buckley <colm@tuatha.org>
39 * SPA: Storage Pool Allocator
41 * This file contains all the routines used when modifying on-disk SPA state.
42 * This includes opening, importing, destroying, exporting a pool, and syncing a
46 #include <sys/zfs_context.h>
47 #include <sys/fm/fs/zfs.h>
48 #include <sys/spa_impl.h>
50 #include <sys/zio_checksum.h>
52 #include <sys/dmu_tx.h>
56 #include <sys/vdev_impl.h>
57 #include <sys/vdev_removal.h>
58 #include <sys/vdev_indirect_mapping.h>
59 #include <sys/vdev_indirect_births.h>
60 #include <sys/vdev_initialize.h>
61 #include <sys/vdev_rebuild.h>
62 #include <sys/vdev_trim.h>
63 #include <sys/vdev_disk.h>
64 #include <sys/vdev_draid.h>
65 #include <sys/metaslab.h>
66 #include <sys/metaslab_impl.h>
68 #include <sys/uberblock_impl.h>
71 #include <sys/bpobj.h>
72 #include <sys/dmu_traverse.h>
73 #include <sys/dmu_objset.h>
74 #include <sys/unique.h>
75 #include <sys/dsl_pool.h>
76 #include <sys/dsl_dataset.h>
77 #include <sys/dsl_dir.h>
78 #include <sys/dsl_prop.h>
79 #include <sys/dsl_synctask.h>
80 #include <sys/fs/zfs.h>
82 #include <sys/callb.h>
83 #include <sys/systeminfo.h>
84 #include <sys/spa_boot.h>
85 #include <sys/zfs_ioctl.h>
86 #include <sys/dsl_scan.h>
87 #include <sys/zfeature.h>
88 #include <sys/dsl_destroy.h>
92 #include <sys/fm/protocol.h>
93 #include <sys/fm/util.h>
94 #include <sys/callb.h>
96 #include <sys/vmsystm.h>
100 #include "zfs_comutil.h"
103 * The interval, in seconds, at which failed configuration cache file writes
106 int zfs_ccw_retry_interval
= 300;
108 typedef enum zti_modes
{
109 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
110 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
111 ZTI_MODE_SCALE
, /* Taskqs scale with CPUs. */
112 ZTI_MODE_NULL
, /* don't create a taskq */
116 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
117 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
118 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
119 #define ZTI_SCALE { ZTI_MODE_SCALE, 0, 1 }
120 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
122 #define ZTI_N(n) ZTI_P(n, 1)
123 #define ZTI_ONE ZTI_N(1)
125 typedef struct zio_taskq_info
{
126 zti_modes_t zti_mode
;
131 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
132 "iss", "iss_h", "int", "int_h"
136 * This table defines the taskq settings for each ZFS I/O type. When
137 * initializing a pool, we use this table to create an appropriately sized
138 * taskq. Some operations are low volume and therefore have a small, static
139 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
140 * macros. Other operations process a large amount of data; the ZTI_BATCH
141 * macro causes us to create a taskq oriented for throughput. Some operations
142 * are so high frequency and short-lived that the taskq itself can become a
143 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
144 * additional degree of parallelism specified by the number of threads per-
145 * taskq and the number of taskqs; when dispatching an event in this case, the
146 * particular taskq is chosen at random. ZTI_SCALE is similar to ZTI_BATCH,
147 * but with number of taskqs also scaling with number of CPUs.
149 * The different taskq priorities are to handle the different contexts (issue
150 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
151 * need to be handled with minimum delay.
153 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
154 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
155 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
156 { ZTI_N(8), ZTI_NULL
, ZTI_SCALE
, ZTI_NULL
}, /* READ */
157 { ZTI_BATCH
, ZTI_N(5), ZTI_SCALE
, ZTI_N(5) }, /* WRITE */
158 { ZTI_SCALE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
159 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
160 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
161 { ZTI_N(4), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* TRIM */
164 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
165 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
166 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
167 static int spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
);
168 static void spa_vdev_resilver_done(spa_t
*spa
);
170 uint_t zio_taskq_batch_pct
= 80; /* 1 thread per cpu in pset */
171 uint_t zio_taskq_batch_tpq
; /* threads per taskq */
172 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
173 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
175 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
178 * Report any spa_load_verify errors found, but do not fail spa_load.
179 * This is used by zdb to analyze non-idle pools.
181 boolean_t spa_load_verify_dryrun
= B_FALSE
;
184 * This (illegal) pool name is used when temporarily importing a spa_t in order
185 * to get the vdev stats associated with the imported devices.
187 #define TRYIMPORT_NAME "$import"
190 * For debugging purposes: print out vdev tree during pool import.
192 int spa_load_print_vdev_tree
= B_FALSE
;
195 * A non-zero value for zfs_max_missing_tvds means that we allow importing
196 * pools with missing top-level vdevs. This is strictly intended for advanced
197 * pool recovery cases since missing data is almost inevitable. Pools with
198 * missing devices can only be imported read-only for safety reasons, and their
199 * fail-mode will be automatically set to "continue".
201 * With 1 missing vdev we should be able to import the pool and mount all
202 * datasets. User data that was not modified after the missing device has been
203 * added should be recoverable. This means that snapshots created prior to the
204 * addition of that device should be completely intact.
206 * With 2 missing vdevs, some datasets may fail to mount since there are
207 * dataset statistics that are stored as regular metadata. Some data might be
208 * recoverable if those vdevs were added recently.
210 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
211 * may be missing entirely. Chances of data recovery are very low. Note that
212 * there are also risks of performing an inadvertent rewind as we might be
213 * missing all the vdevs with the latest uberblocks.
215 unsigned long zfs_max_missing_tvds
= 0;
218 * The parameters below are similar to zfs_max_missing_tvds but are only
219 * intended for a preliminary open of the pool with an untrusted config which
220 * might be incomplete or out-dated.
222 * We are more tolerant for pools opened from a cachefile since we could have
223 * an out-dated cachefile where a device removal was not registered.
224 * We could have set the limit arbitrarily high but in the case where devices
225 * are really missing we would want to return the proper error codes; we chose
226 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
227 * and we get a chance to retrieve the trusted config.
229 uint64_t zfs_max_missing_tvds_cachefile
= SPA_DVAS_PER_BP
- 1;
232 * In the case where config was assembled by scanning device paths (/dev/dsks
233 * by default) we are less tolerant since all the existing devices should have
234 * been detected and we want spa_load to return the right error codes.
236 uint64_t zfs_max_missing_tvds_scan
= 0;
239 * Debugging aid that pauses spa_sync() towards the end.
241 boolean_t zfs_pause_spa_sync
= B_FALSE
;
244 * Variables to indicate the livelist condense zthr func should wait at certain
245 * points for the livelist to be removed - used to test condense/destroy races
247 int zfs_livelist_condense_zthr_pause
= 0;
248 int zfs_livelist_condense_sync_pause
= 0;
251 * Variables to track whether or not condense cancellation has been
252 * triggered in testing.
254 int zfs_livelist_condense_sync_cancel
= 0;
255 int zfs_livelist_condense_zthr_cancel
= 0;
258 * Variable to track whether or not extra ALLOC blkptrs were added to a
259 * livelist entry while it was being condensed (caused by the way we track
260 * remapped blkptrs in dbuf_remap_impl)
262 int zfs_livelist_condense_new_alloc
= 0;
265 * ==========================================================================
266 * SPA properties routines
267 * ==========================================================================
271 * Add a (source=src, propname=propval) list to an nvlist.
274 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
275 uint64_t intval
, zprop_source_t src
)
277 const char *propname
= zpool_prop_to_name(prop
);
280 propval
= fnvlist_alloc();
281 fnvlist_add_uint64(propval
, ZPROP_SOURCE
, src
);
284 fnvlist_add_string(propval
, ZPROP_VALUE
, strval
);
286 fnvlist_add_uint64(propval
, ZPROP_VALUE
, intval
);
288 fnvlist_add_nvlist(nvl
, propname
, propval
);
289 nvlist_free(propval
);
293 * Get property values from the spa configuration.
296 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
298 vdev_t
*rvd
= spa
->spa_root_vdev
;
299 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
300 uint64_t size
, alloc
, cap
, version
;
301 const zprop_source_t src
= ZPROP_SRC_NONE
;
302 spa_config_dirent_t
*dp
;
303 metaslab_class_t
*mc
= spa_normal_class(spa
);
305 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
308 alloc
= metaslab_class_get_alloc(mc
);
309 alloc
+= metaslab_class_get_alloc(spa_special_class(spa
));
310 alloc
+= metaslab_class_get_alloc(spa_dedup_class(spa
));
311 alloc
+= metaslab_class_get_alloc(spa_embedded_log_class(spa
));
313 size
= metaslab_class_get_space(mc
);
314 size
+= metaslab_class_get_space(spa_special_class(spa
));
315 size
+= metaslab_class_get_space(spa_dedup_class(spa
));
316 size
+= metaslab_class_get_space(spa_embedded_log_class(spa
));
318 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
319 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
320 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
321 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
323 spa_prop_add_list(*nvp
, ZPOOL_PROP_CHECKPOINT
, NULL
,
324 spa
->spa_checkpoint_info
.sci_dspace
, src
);
326 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
327 metaslab_class_fragmentation(mc
), src
);
328 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
329 metaslab_class_expandable_space(mc
), src
);
330 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
331 (spa_mode(spa
) == SPA_MODE_READ
), src
);
333 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
334 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
336 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
337 ddt_get_pool_dedup_ratio(spa
), src
);
339 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
340 rvd
->vdev_state
, src
);
342 version
= spa_version(spa
);
343 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
)) {
344 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
345 version
, ZPROP_SRC_DEFAULT
);
347 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
348 version
, ZPROP_SRC_LOCAL
);
350 spa_prop_add_list(*nvp
, ZPOOL_PROP_LOAD_GUID
,
351 NULL
, spa_load_guid(spa
), src
);
356 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
357 * when opening pools before this version freedir will be NULL.
359 if (pool
->dp_free_dir
!= NULL
) {
360 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
361 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
364 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
368 if (pool
->dp_leak_dir
!= NULL
) {
369 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
370 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
373 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
378 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
380 if (spa
->spa_comment
!= NULL
) {
381 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
385 if (spa
->spa_compatibility
!= NULL
) {
386 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMPATIBILITY
,
387 spa
->spa_compatibility
, 0, ZPROP_SRC_LOCAL
);
390 if (spa
->spa_root
!= NULL
)
391 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
394 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
395 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
396 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
398 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
399 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
402 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
)) {
403 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
404 DNODE_MAX_SIZE
, ZPROP_SRC_NONE
);
406 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
407 DNODE_MIN_SIZE
, ZPROP_SRC_NONE
);
410 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
411 if (dp
->scd_path
== NULL
) {
412 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
413 "none", 0, ZPROP_SRC_LOCAL
);
414 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
415 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
416 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
422 * Get zpool property values.
425 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
427 objset_t
*mos
= spa
->spa_meta_objset
;
433 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
437 dp
= spa_get_dsl(spa
);
438 dsl_pool_config_enter(dp
, FTAG
);
439 mutex_enter(&spa
->spa_props_lock
);
442 * Get properties from the spa config.
444 spa_prop_get_config(spa
, nvp
);
446 /* If no pool property object, no more prop to get. */
447 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0)
451 * Get properties from the MOS pool property object.
453 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
454 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
455 zap_cursor_advance(&zc
)) {
458 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
461 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPOOL_PROP_INVAL
)
464 switch (za
.za_integer_length
) {
466 /* integer property */
467 if (za
.za_first_integer
!=
468 zpool_prop_default_numeric(prop
))
469 src
= ZPROP_SRC_LOCAL
;
471 if (prop
== ZPOOL_PROP_BOOTFS
) {
472 dsl_dataset_t
*ds
= NULL
;
474 err
= dsl_dataset_hold_obj(dp
,
475 za
.za_first_integer
, FTAG
, &ds
);
479 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
481 dsl_dataset_name(ds
, strval
);
482 dsl_dataset_rele(ds
, FTAG
);
485 intval
= za
.za_first_integer
;
488 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
491 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
496 /* string property */
497 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
498 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
499 za
.za_name
, 1, za
.za_num_integers
, strval
);
501 kmem_free(strval
, za
.za_num_integers
);
504 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
505 kmem_free(strval
, za
.za_num_integers
);
512 zap_cursor_fini(&zc
);
514 mutex_exit(&spa
->spa_props_lock
);
515 dsl_pool_config_exit(dp
, FTAG
);
516 if (err
&& err
!= ENOENT
) {
526 * Validate the given pool properties nvlist and modify the list
527 * for the property values to be set.
530 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
533 int error
= 0, reset_bootfs
= 0;
535 boolean_t has_feature
= B_FALSE
;
538 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
540 char *strval
, *slash
, *check
, *fname
;
541 const char *propname
= nvpair_name(elem
);
542 zpool_prop_t prop
= zpool_name_to_prop(propname
);
545 case ZPOOL_PROP_INVAL
:
546 if (!zpool_prop_feature(propname
)) {
547 error
= SET_ERROR(EINVAL
);
552 * Sanitize the input.
554 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
555 error
= SET_ERROR(EINVAL
);
559 if (nvpair_value_uint64(elem
, &intval
) != 0) {
560 error
= SET_ERROR(EINVAL
);
565 error
= SET_ERROR(EINVAL
);
569 fname
= strchr(propname
, '@') + 1;
570 if (zfeature_lookup_name(fname
, NULL
) != 0) {
571 error
= SET_ERROR(EINVAL
);
575 has_feature
= B_TRUE
;
578 case ZPOOL_PROP_VERSION
:
579 error
= nvpair_value_uint64(elem
, &intval
);
581 (intval
< spa_version(spa
) ||
582 intval
> SPA_VERSION_BEFORE_FEATURES
||
584 error
= SET_ERROR(EINVAL
);
587 case ZPOOL_PROP_DELEGATION
:
588 case ZPOOL_PROP_AUTOREPLACE
:
589 case ZPOOL_PROP_LISTSNAPS
:
590 case ZPOOL_PROP_AUTOEXPAND
:
591 case ZPOOL_PROP_AUTOTRIM
:
592 error
= nvpair_value_uint64(elem
, &intval
);
593 if (!error
&& intval
> 1)
594 error
= SET_ERROR(EINVAL
);
597 case ZPOOL_PROP_MULTIHOST
:
598 error
= nvpair_value_uint64(elem
, &intval
);
599 if (!error
&& intval
> 1)
600 error
= SET_ERROR(EINVAL
);
603 uint32_t hostid
= zone_get_hostid(NULL
);
605 spa
->spa_hostid
= hostid
;
607 error
= SET_ERROR(ENOTSUP
);
612 case ZPOOL_PROP_BOOTFS
:
614 * If the pool version is less than SPA_VERSION_BOOTFS,
615 * or the pool is still being created (version == 0),
616 * the bootfs property cannot be set.
618 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
619 error
= SET_ERROR(ENOTSUP
);
624 * Make sure the vdev config is bootable
626 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
627 error
= SET_ERROR(ENOTSUP
);
633 error
= nvpair_value_string(elem
, &strval
);
638 if (strval
== NULL
|| strval
[0] == '\0') {
639 objnum
= zpool_prop_default_numeric(
644 error
= dmu_objset_hold(strval
, FTAG
, &os
);
649 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
650 error
= SET_ERROR(ENOTSUP
);
652 objnum
= dmu_objset_id(os
);
654 dmu_objset_rele(os
, FTAG
);
658 case ZPOOL_PROP_FAILUREMODE
:
659 error
= nvpair_value_uint64(elem
, &intval
);
660 if (!error
&& intval
> ZIO_FAILURE_MODE_PANIC
)
661 error
= SET_ERROR(EINVAL
);
664 * This is a special case which only occurs when
665 * the pool has completely failed. This allows
666 * the user to change the in-core failmode property
667 * without syncing it out to disk (I/Os might
668 * currently be blocked). We do this by returning
669 * EIO to the caller (spa_prop_set) to trick it
670 * into thinking we encountered a property validation
673 if (!error
&& spa_suspended(spa
)) {
674 spa
->spa_failmode
= intval
;
675 error
= SET_ERROR(EIO
);
679 case ZPOOL_PROP_CACHEFILE
:
680 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
683 if (strval
[0] == '\0')
686 if (strcmp(strval
, "none") == 0)
689 if (strval
[0] != '/') {
690 error
= SET_ERROR(EINVAL
);
694 slash
= strrchr(strval
, '/');
695 ASSERT(slash
!= NULL
);
697 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
698 strcmp(slash
, "/..") == 0)
699 error
= SET_ERROR(EINVAL
);
702 case ZPOOL_PROP_COMMENT
:
703 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
705 for (check
= strval
; *check
!= '\0'; check
++) {
706 if (!isprint(*check
)) {
707 error
= SET_ERROR(EINVAL
);
711 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
712 error
= SET_ERROR(E2BIG
);
723 (void) nvlist_remove_all(props
,
724 zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO
));
726 if (!error
&& reset_bootfs
) {
727 error
= nvlist_remove(props
,
728 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
731 error
= nvlist_add_uint64(props
,
732 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
740 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
743 spa_config_dirent_t
*dp
;
745 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
749 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
752 if (cachefile
[0] == '\0')
753 dp
->scd_path
= spa_strdup(spa_config_path
);
754 else if (strcmp(cachefile
, "none") == 0)
757 dp
->scd_path
= spa_strdup(cachefile
);
759 list_insert_head(&spa
->spa_config_list
, dp
);
761 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
765 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
768 nvpair_t
*elem
= NULL
;
769 boolean_t need_sync
= B_FALSE
;
771 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
774 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
775 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
777 if (prop
== ZPOOL_PROP_CACHEFILE
||
778 prop
== ZPOOL_PROP_ALTROOT
||
779 prop
== ZPOOL_PROP_READONLY
)
782 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPOOL_PROP_INVAL
) {
785 if (prop
== ZPOOL_PROP_VERSION
) {
786 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
788 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
789 ver
= SPA_VERSION_FEATURES
;
793 /* Save time if the version is already set. */
794 if (ver
== spa_version(spa
))
798 * In addition to the pool directory object, we might
799 * create the pool properties object, the features for
800 * read object, the features for write object, or the
801 * feature descriptions object.
803 error
= dsl_sync_task(spa
->spa_name
, NULL
,
804 spa_sync_version
, &ver
,
805 6, ZFS_SPACE_CHECK_RESERVED
);
816 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
817 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
824 * If the bootfs property value is dsobj, clear it.
827 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
829 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
830 VERIFY(zap_remove(spa
->spa_meta_objset
,
831 spa
->spa_pool_props_object
,
832 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
839 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
841 uint64_t *newguid __maybe_unused
= arg
;
842 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
843 vdev_t
*rvd
= spa
->spa_root_vdev
;
846 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
847 int error
= (spa_has_checkpoint(spa
)) ?
848 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
849 return (SET_ERROR(error
));
852 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
853 vdev_state
= rvd
->vdev_state
;
854 spa_config_exit(spa
, SCL_STATE
, FTAG
);
856 if (vdev_state
!= VDEV_STATE_HEALTHY
)
857 return (SET_ERROR(ENXIO
));
859 ASSERT3U(spa_guid(spa
), !=, *newguid
);
865 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
867 uint64_t *newguid
= arg
;
868 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
870 vdev_t
*rvd
= spa
->spa_root_vdev
;
872 oldguid
= spa_guid(spa
);
874 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
875 rvd
->vdev_guid
= *newguid
;
876 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
877 vdev_config_dirty(rvd
);
878 spa_config_exit(spa
, SCL_STATE
, FTAG
);
880 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
881 (u_longlong_t
)oldguid
, (u_longlong_t
)*newguid
);
885 * Change the GUID for the pool. This is done so that we can later
886 * re-import a pool built from a clone of our own vdevs. We will modify
887 * the root vdev's guid, our own pool guid, and then mark all of our
888 * vdevs dirty. Note that we must make sure that all our vdevs are
889 * online when we do this, or else any vdevs that weren't present
890 * would be orphaned from our pool. We are also going to issue a
891 * sysevent to update any watchers.
894 spa_change_guid(spa_t
*spa
)
899 mutex_enter(&spa
->spa_vdev_top_lock
);
900 mutex_enter(&spa_namespace_lock
);
901 guid
= spa_generate_guid(NULL
);
903 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
904 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
907 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
908 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
911 mutex_exit(&spa_namespace_lock
);
912 mutex_exit(&spa
->spa_vdev_top_lock
);
918 * ==========================================================================
919 * SPA state manipulation (open/create/destroy/import/export)
920 * ==========================================================================
924 spa_error_entry_compare(const void *a
, const void *b
)
926 const spa_error_entry_t
*sa
= (const spa_error_entry_t
*)a
;
927 const spa_error_entry_t
*sb
= (const spa_error_entry_t
*)b
;
930 ret
= memcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
931 sizeof (zbookmark_phys_t
));
933 return (TREE_ISIGN(ret
));
937 * Utility function which retrieves copies of the current logs and
938 * re-initializes them in the process.
941 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
943 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
945 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
946 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
948 avl_create(&spa
->spa_errlist_scrub
,
949 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
950 offsetof(spa_error_entry_t
, se_avl
));
951 avl_create(&spa
->spa_errlist_last
,
952 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
953 offsetof(spa_error_entry_t
, se_avl
));
957 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
959 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
960 enum zti_modes mode
= ztip
->zti_mode
;
961 uint_t value
= ztip
->zti_value
;
962 uint_t count
= ztip
->zti_count
;
963 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
964 uint_t cpus
, flags
= TASKQ_DYNAMIC
;
965 boolean_t batch
= B_FALSE
;
969 ASSERT3U(value
, >, 0);
974 flags
|= TASKQ_THREADS_CPU_PCT
;
975 value
= MIN(zio_taskq_batch_pct
, 100);
979 flags
|= TASKQ_THREADS_CPU_PCT
;
981 * We want more taskqs to reduce lock contention, but we want
982 * less for better request ordering and CPU utilization.
984 cpus
= MAX(1, boot_ncpus
* zio_taskq_batch_pct
/ 100);
985 if (zio_taskq_batch_tpq
> 0) {
986 count
= MAX(1, (cpus
+ zio_taskq_batch_tpq
/ 2) /
987 zio_taskq_batch_tpq
);
990 * Prefer 6 threads per taskq, but no more taskqs
991 * than threads in them on large systems. For 80%:
994 * cpus taskqs percent threads threads
995 * ------- ------- ------- ------- -------
1006 count
= 1 + cpus
/ 6;
1007 while (count
* count
> cpus
)
1010 /* Limit each taskq within 100% to not trigger assertion. */
1011 count
= MAX(count
, (zio_taskq_batch_pct
+ 99) / 100);
1012 value
= (zio_taskq_batch_pct
+ count
/ 2) / count
;
1016 tqs
->stqs_count
= 0;
1017 tqs
->stqs_taskq
= NULL
;
1021 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
1023 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
1027 ASSERT3U(count
, >, 0);
1028 tqs
->stqs_count
= count
;
1029 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
1031 for (uint_t i
= 0; i
< count
; i
++) {
1036 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
1037 zio_type_name
[t
], zio_taskq_types
[q
], i
);
1039 (void) snprintf(name
, sizeof (name
), "%s_%s",
1040 zio_type_name
[t
], zio_taskq_types
[q
]);
1042 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
1044 flags
|= TASKQ_DC_BATCH
;
1046 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
1047 spa
->spa_proc
, zio_taskq_basedc
, flags
);
1049 pri_t pri
= maxclsyspri
;
1051 * The write issue taskq can be extremely CPU
1052 * intensive. Run it at slightly less important
1053 * priority than the other taskqs.
1055 * Under Linux and FreeBSD this means incrementing
1056 * the priority value as opposed to platforms like
1057 * illumos where it should be decremented.
1059 * On FreeBSD, if priorities divided by four (RQ_PPQ)
1060 * are equal then a difference between them is
1063 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
) {
1064 #if defined(__linux__)
1066 #elif defined(__FreeBSD__)
1072 tq
= taskq_create_proc(name
, value
, pri
, 50,
1073 INT_MAX
, spa
->spa_proc
, flags
);
1076 tqs
->stqs_taskq
[i
] = tq
;
1081 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
1083 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1085 if (tqs
->stqs_taskq
== NULL
) {
1086 ASSERT3U(tqs
->stqs_count
, ==, 0);
1090 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
1091 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
1092 taskq_destroy(tqs
->stqs_taskq
[i
]);
1095 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
1096 tqs
->stqs_taskq
= NULL
;
1100 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1101 * Note that a type may have multiple discrete taskqs to avoid lock contention
1102 * on the taskq itself. In that case we choose which taskq at random by using
1103 * the low bits of gethrtime().
1106 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1107 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
1109 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1112 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1113 ASSERT3U(tqs
->stqs_count
, !=, 0);
1115 if (tqs
->stqs_count
== 1) {
1116 tq
= tqs
->stqs_taskq
[0];
1118 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1121 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
1125 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1128 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1129 task_func_t
*func
, void *arg
, uint_t flags
)
1131 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1135 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1136 ASSERT3U(tqs
->stqs_count
, !=, 0);
1138 if (tqs
->stqs_count
== 1) {
1139 tq
= tqs
->stqs_taskq
[0];
1141 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1144 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1146 taskq_wait_id(tq
, id
);
1150 spa_create_zio_taskqs(spa_t
*spa
)
1152 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1153 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1154 spa_taskqs_init(spa
, t
, q
);
1160 * Disabled until spa_thread() can be adapted for Linux.
1162 #undef HAVE_SPA_THREAD
1164 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1166 spa_thread(void *arg
)
1168 psetid_t zio_taskq_psrset_bind
= PS_NONE
;
1169 callb_cpr_t cprinfo
;
1172 user_t
*pu
= PTOU(curproc
);
1174 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1177 ASSERT(curproc
!= &p0
);
1178 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1179 "zpool-%s", spa
->spa_name
);
1180 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1182 /* bind this thread to the requested psrset */
1183 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1185 mutex_enter(&cpu_lock
);
1186 mutex_enter(&pidlock
);
1187 mutex_enter(&curproc
->p_lock
);
1189 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1190 0, NULL
, NULL
) == 0) {
1191 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1194 "Couldn't bind process for zfs pool \"%s\" to "
1195 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1198 mutex_exit(&curproc
->p_lock
);
1199 mutex_exit(&pidlock
);
1200 mutex_exit(&cpu_lock
);
1204 if (zio_taskq_sysdc
) {
1205 sysdc_thread_enter(curthread
, 100, 0);
1208 spa
->spa_proc
= curproc
;
1209 spa
->spa_did
= curthread
->t_did
;
1211 spa_create_zio_taskqs(spa
);
1213 mutex_enter(&spa
->spa_proc_lock
);
1214 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1216 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1217 cv_broadcast(&spa
->spa_proc_cv
);
1219 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1220 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1221 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1222 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1224 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1225 spa
->spa_proc_state
= SPA_PROC_GONE
;
1226 spa
->spa_proc
= &p0
;
1227 cv_broadcast(&spa
->spa_proc_cv
);
1228 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1230 mutex_enter(&curproc
->p_lock
);
1236 * Activate an uninitialized pool.
1239 spa_activate(spa_t
*spa
, spa_mode_t mode
)
1241 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1243 spa
->spa_state
= POOL_STATE_ACTIVE
;
1244 spa
->spa_mode
= mode
;
1246 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1247 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1248 spa
->spa_embedded_log_class
=
1249 metaslab_class_create(spa
, zfs_metaslab_ops
);
1250 spa
->spa_special_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1251 spa
->spa_dedup_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1253 /* Try to create a covering process */
1254 mutex_enter(&spa
->spa_proc_lock
);
1255 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1256 ASSERT(spa
->spa_proc
== &p0
);
1259 #ifdef HAVE_SPA_THREAD
1260 /* Only create a process if we're going to be around a while. */
1261 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1262 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1264 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1265 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1266 cv_wait(&spa
->spa_proc_cv
,
1267 &spa
->spa_proc_lock
);
1269 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1270 ASSERT(spa
->spa_proc
!= &p0
);
1271 ASSERT(spa
->spa_did
!= 0);
1275 "Couldn't create process for zfs pool \"%s\"\n",
1280 #endif /* HAVE_SPA_THREAD */
1281 mutex_exit(&spa
->spa_proc_lock
);
1283 /* If we didn't create a process, we need to create our taskqs. */
1284 if (spa
->spa_proc
== &p0
) {
1285 spa_create_zio_taskqs(spa
);
1288 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1289 spa
->spa_txg_zio
[i
] = zio_root(spa
, NULL
, NULL
,
1293 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1294 offsetof(vdev_t
, vdev_config_dirty_node
));
1295 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1296 offsetof(objset_t
, os_evicting_node
));
1297 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1298 offsetof(vdev_t
, vdev_state_dirty_node
));
1300 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1301 offsetof(struct vdev
, vdev_txg_node
));
1303 avl_create(&spa
->spa_errlist_scrub
,
1304 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1305 offsetof(spa_error_entry_t
, se_avl
));
1306 avl_create(&spa
->spa_errlist_last
,
1307 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1308 offsetof(spa_error_entry_t
, se_avl
));
1310 spa_keystore_init(&spa
->spa_keystore
);
1313 * This taskq is used to perform zvol-minor-related tasks
1314 * asynchronously. This has several advantages, including easy
1315 * resolution of various deadlocks.
1317 * The taskq must be single threaded to ensure tasks are always
1318 * processed in the order in which they were dispatched.
1320 * A taskq per pool allows one to keep the pools independent.
1321 * This way if one pool is suspended, it will not impact another.
1323 * The preferred location to dispatch a zvol minor task is a sync
1324 * task. In this context, there is easy access to the spa_t and minimal
1325 * error handling is required because the sync task must succeed.
1327 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1331 * Taskq dedicated to prefetcher threads: this is used to prevent the
1332 * pool traverse code from monopolizing the global (and limited)
1333 * system_taskq by inappropriately scheduling long running tasks on it.
1335 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", 100,
1336 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1339 * The taskq to upgrade datasets in this pool. Currently used by
1340 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1342 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", 100,
1343 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1347 * Opposite of spa_activate().
1350 spa_deactivate(spa_t
*spa
)
1352 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1353 ASSERT(spa
->spa_dsl_pool
== NULL
);
1354 ASSERT(spa
->spa_root_vdev
== NULL
);
1355 ASSERT(spa
->spa_async_zio_root
== NULL
);
1356 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1358 spa_evicting_os_wait(spa
);
1360 if (spa
->spa_zvol_taskq
) {
1361 taskq_destroy(spa
->spa_zvol_taskq
);
1362 spa
->spa_zvol_taskq
= NULL
;
1365 if (spa
->spa_prefetch_taskq
) {
1366 taskq_destroy(spa
->spa_prefetch_taskq
);
1367 spa
->spa_prefetch_taskq
= NULL
;
1370 if (spa
->spa_upgrade_taskq
) {
1371 taskq_destroy(spa
->spa_upgrade_taskq
);
1372 spa
->spa_upgrade_taskq
= NULL
;
1375 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1377 list_destroy(&spa
->spa_config_dirty_list
);
1378 list_destroy(&spa
->spa_evicting_os_list
);
1379 list_destroy(&spa
->spa_state_dirty_list
);
1381 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1383 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1384 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1385 spa_taskqs_fini(spa
, t
, q
);
1389 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1390 ASSERT3P(spa
->spa_txg_zio
[i
], !=, NULL
);
1391 VERIFY0(zio_wait(spa
->spa_txg_zio
[i
]));
1392 spa
->spa_txg_zio
[i
] = NULL
;
1395 metaslab_class_destroy(spa
->spa_normal_class
);
1396 spa
->spa_normal_class
= NULL
;
1398 metaslab_class_destroy(spa
->spa_log_class
);
1399 spa
->spa_log_class
= NULL
;
1401 metaslab_class_destroy(spa
->spa_embedded_log_class
);
1402 spa
->spa_embedded_log_class
= NULL
;
1404 metaslab_class_destroy(spa
->spa_special_class
);
1405 spa
->spa_special_class
= NULL
;
1407 metaslab_class_destroy(spa
->spa_dedup_class
);
1408 spa
->spa_dedup_class
= NULL
;
1411 * If this was part of an import or the open otherwise failed, we may
1412 * still have errors left in the queues. Empty them just in case.
1414 spa_errlog_drain(spa
);
1415 avl_destroy(&spa
->spa_errlist_scrub
);
1416 avl_destroy(&spa
->spa_errlist_last
);
1418 spa_keystore_fini(&spa
->spa_keystore
);
1420 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1422 mutex_enter(&spa
->spa_proc_lock
);
1423 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1424 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1425 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1426 cv_broadcast(&spa
->spa_proc_cv
);
1427 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1428 ASSERT(spa
->spa_proc
!= &p0
);
1429 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1431 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1432 spa
->spa_proc_state
= SPA_PROC_NONE
;
1434 ASSERT(spa
->spa_proc
== &p0
);
1435 mutex_exit(&spa
->spa_proc_lock
);
1438 * We want to make sure spa_thread() has actually exited the ZFS
1439 * module, so that the module can't be unloaded out from underneath
1442 if (spa
->spa_did
!= 0) {
1443 thread_join(spa
->spa_did
);
1449 * Verify a pool configuration, and construct the vdev tree appropriately. This
1450 * will create all the necessary vdevs in the appropriate layout, with each vdev
1451 * in the CLOSED state. This will prep the pool before open/creation/import.
1452 * All vdev validation is done by the vdev_alloc() routine.
1455 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1456 uint_t id
, int atype
)
1462 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1465 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1468 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1471 if (error
== ENOENT
)
1477 return (SET_ERROR(EINVAL
));
1480 for (int c
= 0; c
< children
; c
++) {
1482 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1490 ASSERT(*vdp
!= NULL
);
1496 spa_should_flush_logs_on_unload(spa_t
*spa
)
1498 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
1501 if (!spa_writeable(spa
))
1504 if (!spa
->spa_sync_on
)
1507 if (spa_state(spa
) != POOL_STATE_EXPORTED
)
1510 if (zfs_keep_log_spacemaps_at_export
)
1517 * Opens a transaction that will set the flag that will instruct
1518 * spa_sync to attempt to flush all the metaslabs for that txg.
1521 spa_unload_log_sm_flush_all(spa_t
*spa
)
1523 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
1524 VERIFY0(dmu_tx_assign(tx
, TXG_WAIT
));
1526 ASSERT3U(spa
->spa_log_flushall_txg
, ==, 0);
1527 spa
->spa_log_flushall_txg
= dmu_tx_get_txg(tx
);
1530 txg_wait_synced(spa_get_dsl(spa
), spa
->spa_log_flushall_txg
);
1534 spa_unload_log_sm_metadata(spa_t
*spa
)
1536 void *cookie
= NULL
;
1538 while ((sls
= avl_destroy_nodes(&spa
->spa_sm_logs_by_txg
,
1539 &cookie
)) != NULL
) {
1540 VERIFY0(sls
->sls_mscount
);
1541 kmem_free(sls
, sizeof (spa_log_sm_t
));
1544 for (log_summary_entry_t
*e
= list_head(&spa
->spa_log_summary
);
1545 e
!= NULL
; e
= list_head(&spa
->spa_log_summary
)) {
1546 VERIFY0(e
->lse_mscount
);
1547 list_remove(&spa
->spa_log_summary
, e
);
1548 kmem_free(e
, sizeof (log_summary_entry_t
));
1551 spa
->spa_unflushed_stats
.sus_nblocks
= 0;
1552 spa
->spa_unflushed_stats
.sus_memused
= 0;
1553 spa
->spa_unflushed_stats
.sus_blocklimit
= 0;
1557 spa_destroy_aux_threads(spa_t
*spa
)
1559 if (spa
->spa_condense_zthr
!= NULL
) {
1560 zthr_destroy(spa
->spa_condense_zthr
);
1561 spa
->spa_condense_zthr
= NULL
;
1563 if (spa
->spa_checkpoint_discard_zthr
!= NULL
) {
1564 zthr_destroy(spa
->spa_checkpoint_discard_zthr
);
1565 spa
->spa_checkpoint_discard_zthr
= NULL
;
1567 if (spa
->spa_livelist_delete_zthr
!= NULL
) {
1568 zthr_destroy(spa
->spa_livelist_delete_zthr
);
1569 spa
->spa_livelist_delete_zthr
= NULL
;
1571 if (spa
->spa_livelist_condense_zthr
!= NULL
) {
1572 zthr_destroy(spa
->spa_livelist_condense_zthr
);
1573 spa
->spa_livelist_condense_zthr
= NULL
;
1578 * Opposite of spa_load().
1581 spa_unload(spa_t
*spa
)
1583 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1584 ASSERT(spa_state(spa
) != POOL_STATE_UNINITIALIZED
);
1586 spa_import_progress_remove(spa_guid(spa
));
1587 spa_load_note(spa
, "UNLOADING");
1589 spa_wake_waiters(spa
);
1592 * If the log space map feature is enabled and the pool is getting
1593 * exported (but not destroyed), we want to spend some time flushing
1594 * as many metaslabs as we can in an attempt to destroy log space
1595 * maps and save import time.
1597 if (spa_should_flush_logs_on_unload(spa
))
1598 spa_unload_log_sm_flush_all(spa
);
1603 spa_async_suspend(spa
);
1605 if (spa
->spa_root_vdev
) {
1606 vdev_t
*root_vdev
= spa
->spa_root_vdev
;
1607 vdev_initialize_stop_all(root_vdev
, VDEV_INITIALIZE_ACTIVE
);
1608 vdev_trim_stop_all(root_vdev
, VDEV_TRIM_ACTIVE
);
1609 vdev_autotrim_stop_all(spa
);
1610 vdev_rebuild_stop_all(spa
);
1616 if (spa
->spa_sync_on
) {
1617 txg_sync_stop(spa
->spa_dsl_pool
);
1618 spa
->spa_sync_on
= B_FALSE
;
1622 * This ensures that there is no async metaslab prefetching
1623 * while we attempt to unload the spa.
1625 if (spa
->spa_root_vdev
!= NULL
) {
1626 for (int c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++) {
1627 vdev_t
*vc
= spa
->spa_root_vdev
->vdev_child
[c
];
1628 if (vc
->vdev_mg
!= NULL
)
1629 taskq_wait(vc
->vdev_mg
->mg_taskq
);
1633 if (spa
->spa_mmp
.mmp_thread
)
1634 mmp_thread_stop(spa
);
1637 * Wait for any outstanding async I/O to complete.
1639 if (spa
->spa_async_zio_root
!= NULL
) {
1640 for (int i
= 0; i
< max_ncpus
; i
++)
1641 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1642 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1643 spa
->spa_async_zio_root
= NULL
;
1646 if (spa
->spa_vdev_removal
!= NULL
) {
1647 spa_vdev_removal_destroy(spa
->spa_vdev_removal
);
1648 spa
->spa_vdev_removal
= NULL
;
1651 spa_destroy_aux_threads(spa
);
1653 spa_condense_fini(spa
);
1655 bpobj_close(&spa
->spa_deferred_bpobj
);
1657 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1662 if (spa
->spa_root_vdev
)
1663 vdev_free(spa
->spa_root_vdev
);
1664 ASSERT(spa
->spa_root_vdev
== NULL
);
1667 * Close the dsl pool.
1669 if (spa
->spa_dsl_pool
) {
1670 dsl_pool_close(spa
->spa_dsl_pool
);
1671 spa
->spa_dsl_pool
= NULL
;
1672 spa
->spa_meta_objset
= NULL
;
1676 spa_unload_log_sm_metadata(spa
);
1679 * Drop and purge level 2 cache
1681 spa_l2cache_drop(spa
);
1683 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1684 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1685 if (spa
->spa_spares
.sav_vdevs
) {
1686 kmem_free(spa
->spa_spares
.sav_vdevs
,
1687 spa
->spa_spares
.sav_count
* sizeof (void *));
1688 spa
->spa_spares
.sav_vdevs
= NULL
;
1690 if (spa
->spa_spares
.sav_config
) {
1691 nvlist_free(spa
->spa_spares
.sav_config
);
1692 spa
->spa_spares
.sav_config
= NULL
;
1694 spa
->spa_spares
.sav_count
= 0;
1696 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1697 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1698 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1700 if (spa
->spa_l2cache
.sav_vdevs
) {
1701 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1702 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1703 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1705 if (spa
->spa_l2cache
.sav_config
) {
1706 nvlist_free(spa
->spa_l2cache
.sav_config
);
1707 spa
->spa_l2cache
.sav_config
= NULL
;
1709 spa
->spa_l2cache
.sav_count
= 0;
1711 spa
->spa_async_suspended
= 0;
1713 spa
->spa_indirect_vdevs_loaded
= B_FALSE
;
1715 if (spa
->spa_comment
!= NULL
) {
1716 spa_strfree(spa
->spa_comment
);
1717 spa
->spa_comment
= NULL
;
1719 if (spa
->spa_compatibility
!= NULL
) {
1720 spa_strfree(spa
->spa_compatibility
);
1721 spa
->spa_compatibility
= NULL
;
1724 spa_config_exit(spa
, SCL_ALL
, spa
);
1728 * Load (or re-load) the current list of vdevs describing the active spares for
1729 * this pool. When this is called, we have some form of basic information in
1730 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1731 * then re-generate a more complete list including status information.
1734 spa_load_spares(spa_t
*spa
)
1743 * zdb opens both the current state of the pool and the
1744 * checkpointed state (if present), with a different spa_t.
1746 * As spare vdevs are shared among open pools, we skip loading
1747 * them when we load the checkpointed state of the pool.
1749 if (!spa_writeable(spa
))
1753 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1756 * First, close and free any existing spare vdevs.
1758 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1759 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1761 /* Undo the call to spa_activate() below */
1762 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1763 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1764 spa_spare_remove(tvd
);
1769 if (spa
->spa_spares
.sav_vdevs
)
1770 kmem_free(spa
->spa_spares
.sav_vdevs
,
1771 spa
->spa_spares
.sav_count
* sizeof (void *));
1773 if (spa
->spa_spares
.sav_config
== NULL
)
1776 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1777 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
));
1779 spa
->spa_spares
.sav_count
= (int)nspares
;
1780 spa
->spa_spares
.sav_vdevs
= NULL
;
1786 * Construct the array of vdevs, opening them to get status in the
1787 * process. For each spare, there is potentially two different vdev_t
1788 * structures associated with it: one in the list of spares (used only
1789 * for basic validation purposes) and one in the active vdev
1790 * configuration (if it's spared in). During this phase we open and
1791 * validate each vdev on the spare list. If the vdev also exists in the
1792 * active configuration, then we also mark this vdev as an active spare.
1794 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1796 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1797 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1798 VDEV_ALLOC_SPARE
) == 0);
1801 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1803 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1804 B_FALSE
)) != NULL
) {
1805 if (!tvd
->vdev_isspare
)
1809 * We only mark the spare active if we were successfully
1810 * able to load the vdev. Otherwise, importing a pool
1811 * with a bad active spare would result in strange
1812 * behavior, because multiple pool would think the spare
1813 * is actively in use.
1815 * There is a vulnerability here to an equally bizarre
1816 * circumstance, where a dead active spare is later
1817 * brought back to life (onlined or otherwise). Given
1818 * the rarity of this scenario, and the extra complexity
1819 * it adds, we ignore the possibility.
1821 if (!vdev_is_dead(tvd
))
1822 spa_spare_activate(tvd
);
1826 vd
->vdev_aux
= &spa
->spa_spares
;
1828 if (vdev_open(vd
) != 0)
1831 if (vdev_validate_aux(vd
) == 0)
1836 * Recompute the stashed list of spares, with status information
1839 fnvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
);
1841 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1843 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1844 spares
[i
] = vdev_config_generate(spa
,
1845 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1846 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1847 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
);
1848 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1849 nvlist_free(spares
[i
]);
1850 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1854 * Load (or re-load) the current list of vdevs describing the active l2cache for
1855 * this pool. When this is called, we have some form of basic information in
1856 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1857 * then re-generate a more complete list including status information.
1858 * Devices which are already active have their details maintained, and are
1862 spa_load_l2cache(spa_t
*spa
)
1864 nvlist_t
**l2cache
= NULL
;
1866 int i
, j
, oldnvdevs
;
1868 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1869 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1873 * zdb opens both the current state of the pool and the
1874 * checkpointed state (if present), with a different spa_t.
1876 * As L2 caches are part of the ARC which is shared among open
1877 * pools, we skip loading them when we load the checkpointed
1878 * state of the pool.
1880 if (!spa_writeable(spa
))
1884 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1886 oldvdevs
= sav
->sav_vdevs
;
1887 oldnvdevs
= sav
->sav_count
;
1888 sav
->sav_vdevs
= NULL
;
1891 if (sav
->sav_config
== NULL
) {
1897 VERIFY0(nvlist_lookup_nvlist_array(sav
->sav_config
,
1898 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
));
1899 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1902 * Process new nvlist of vdevs.
1904 for (i
= 0; i
< nl2cache
; i
++) {
1905 guid
= fnvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
);
1908 for (j
= 0; j
< oldnvdevs
; j
++) {
1910 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1912 * Retain previous vdev for add/remove ops.
1920 if (newvdevs
[i
] == NULL
) {
1924 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1925 VDEV_ALLOC_L2CACHE
) == 0);
1930 * Commit this vdev as an l2cache device,
1931 * even if it fails to open.
1933 spa_l2cache_add(vd
);
1938 spa_l2cache_activate(vd
);
1940 if (vdev_open(vd
) != 0)
1943 (void) vdev_validate_aux(vd
);
1945 if (!vdev_is_dead(vd
))
1946 l2arc_add_vdev(spa
, vd
);
1949 * Upon cache device addition to a pool or pool
1950 * creation with a cache device or if the header
1951 * of the device is invalid we issue an async
1952 * TRIM command for the whole device which will
1953 * execute if l2arc_trim_ahead > 0.
1955 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
1959 sav
->sav_vdevs
= newvdevs
;
1960 sav
->sav_count
= (int)nl2cache
;
1963 * Recompute the stashed list of l2cache devices, with status
1964 * information this time.
1966 fnvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
);
1968 if (sav
->sav_count
> 0)
1969 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *),
1971 for (i
= 0; i
< sav
->sav_count
; i
++)
1972 l2cache
[i
] = vdev_config_generate(spa
,
1973 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1974 fnvlist_add_nvlist_array(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
, l2cache
,
1979 * Purge vdevs that were dropped
1981 for (i
= 0; i
< oldnvdevs
; i
++) {
1986 ASSERT(vd
->vdev_isl2cache
);
1988 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1989 pool
!= 0ULL && l2arc_vdev_present(vd
))
1990 l2arc_remove_vdev(vd
);
1991 vdev_clear_stats(vd
);
1997 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1999 for (i
= 0; i
< sav
->sav_count
; i
++)
2000 nvlist_free(l2cache
[i
]);
2002 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
2006 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
2009 char *packed
= NULL
;
2014 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
2018 nvsize
= *(uint64_t *)db
->db_data
;
2019 dmu_buf_rele(db
, FTAG
);
2021 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
2022 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
2025 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
2026 vmem_free(packed
, nvsize
);
2032 * Concrete top-level vdevs that are not missing and are not logs. At every
2033 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
2036 spa_healthy_core_tvds(spa_t
*spa
)
2038 vdev_t
*rvd
= spa
->spa_root_vdev
;
2041 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
2042 vdev_t
*vd
= rvd
->vdev_child
[i
];
2045 if (vdev_is_concrete(vd
) && !vdev_is_dead(vd
))
2053 * Checks to see if the given vdev could not be opened, in which case we post a
2054 * sysevent to notify the autoreplace code that the device has been removed.
2057 spa_check_removed(vdev_t
*vd
)
2059 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
2060 spa_check_removed(vd
->vdev_child
[c
]);
2062 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
2063 vdev_is_concrete(vd
)) {
2064 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
2065 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
2070 spa_check_for_missing_logs(spa_t
*spa
)
2072 vdev_t
*rvd
= spa
->spa_root_vdev
;
2075 * If we're doing a normal import, then build up any additional
2076 * diagnostic information about missing log devices.
2077 * We'll pass this up to the user for further processing.
2079 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
2080 nvlist_t
**child
, *nv
;
2083 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
2085 nv
= fnvlist_alloc();
2087 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2088 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2091 * We consider a device as missing only if it failed
2092 * to open (i.e. offline or faulted is not considered
2095 if (tvd
->vdev_islog
&&
2096 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2097 child
[idx
++] = vdev_config_generate(spa
, tvd
,
2098 B_FALSE
, VDEV_CONFIG_MISSING
);
2103 fnvlist_add_nvlist_array(nv
,
2104 ZPOOL_CONFIG_CHILDREN
, child
, idx
);
2105 fnvlist_add_nvlist(spa
->spa_load_info
,
2106 ZPOOL_CONFIG_MISSING_DEVICES
, nv
);
2108 for (uint64_t i
= 0; i
< idx
; i
++)
2109 nvlist_free(child
[i
]);
2112 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
2115 spa_load_failed(spa
, "some log devices are missing");
2116 vdev_dbgmsg_print_tree(rvd
, 2);
2117 return (SET_ERROR(ENXIO
));
2120 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2121 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2123 if (tvd
->vdev_islog
&&
2124 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2125 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2126 spa_load_note(spa
, "some log devices are "
2127 "missing, ZIL is dropped.");
2128 vdev_dbgmsg_print_tree(rvd
, 2);
2138 * Check for missing log devices
2141 spa_check_logs(spa_t
*spa
)
2143 boolean_t rv
= B_FALSE
;
2144 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2146 switch (spa
->spa_log_state
) {
2149 case SPA_LOG_MISSING
:
2150 /* need to recheck in case slog has been restored */
2151 case SPA_LOG_UNKNOWN
:
2152 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
2153 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
2155 spa_set_log_state(spa
, SPA_LOG_MISSING
);
2162 * Passivate any log vdevs (note, does not apply to embedded log metaslabs).
2165 spa_passivate_log(spa_t
*spa
)
2167 vdev_t
*rvd
= spa
->spa_root_vdev
;
2168 boolean_t slog_found
= B_FALSE
;
2170 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2172 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2173 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2175 if (tvd
->vdev_islog
) {
2176 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
2177 metaslab_group_passivate(tvd
->vdev_mg
);
2178 slog_found
= B_TRUE
;
2182 return (slog_found
);
2186 * Activate any log vdevs (note, does not apply to embedded log metaslabs).
2189 spa_activate_log(spa_t
*spa
)
2191 vdev_t
*rvd
= spa
->spa_root_vdev
;
2193 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2195 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2196 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2198 if (tvd
->vdev_islog
) {
2199 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
2200 metaslab_group_activate(tvd
->vdev_mg
);
2206 spa_reset_logs(spa_t
*spa
)
2210 error
= dmu_objset_find(spa_name(spa
), zil_reset
,
2211 NULL
, DS_FIND_CHILDREN
);
2214 * We successfully offlined the log device, sync out the
2215 * current txg so that the "stubby" block can be removed
2218 txg_wait_synced(spa
->spa_dsl_pool
, 0);
2224 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
2226 for (int i
= 0; i
< sav
->sav_count
; i
++)
2227 spa_check_removed(sav
->sav_vdevs
[i
]);
2231 spa_claim_notify(zio_t
*zio
)
2233 spa_t
*spa
= zio
->io_spa
;
2238 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
2239 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
2240 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
2241 mutex_exit(&spa
->spa_props_lock
);
2244 typedef struct spa_load_error
{
2245 uint64_t sle_meta_count
;
2246 uint64_t sle_data_count
;
2250 spa_load_verify_done(zio_t
*zio
)
2252 blkptr_t
*bp
= zio
->io_bp
;
2253 spa_load_error_t
*sle
= zio
->io_private
;
2254 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2255 int error
= zio
->io_error
;
2256 spa_t
*spa
= zio
->io_spa
;
2258 abd_free(zio
->io_abd
);
2260 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2261 type
!= DMU_OT_INTENT_LOG
)
2262 atomic_inc_64(&sle
->sle_meta_count
);
2264 atomic_inc_64(&sle
->sle_data_count
);
2267 mutex_enter(&spa
->spa_scrub_lock
);
2268 spa
->spa_load_verify_bytes
-= BP_GET_PSIZE(bp
);
2269 cv_broadcast(&spa
->spa_scrub_io_cv
);
2270 mutex_exit(&spa
->spa_scrub_lock
);
2274 * Maximum number of inflight bytes is the log2 fraction of the arc size.
2275 * By default, we set it to 1/16th of the arc.
2277 int spa_load_verify_shift
= 4;
2278 int spa_load_verify_metadata
= B_TRUE
;
2279 int spa_load_verify_data
= B_TRUE
;
2283 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2284 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2286 if (zb
->zb_level
== ZB_DNODE_LEVEL
|| BP_IS_HOLE(bp
) ||
2287 BP_IS_EMBEDDED(bp
) || BP_IS_REDACTED(bp
))
2290 * Note: normally this routine will not be called if
2291 * spa_load_verify_metadata is not set. However, it may be useful
2292 * to manually set the flag after the traversal has begun.
2294 if (!spa_load_verify_metadata
)
2296 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2299 uint64_t maxinflight_bytes
=
2300 arc_target_bytes() >> spa_load_verify_shift
;
2302 size_t size
= BP_GET_PSIZE(bp
);
2304 mutex_enter(&spa
->spa_scrub_lock
);
2305 while (spa
->spa_load_verify_bytes
>= maxinflight_bytes
)
2306 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2307 spa
->spa_load_verify_bytes
+= size
;
2308 mutex_exit(&spa
->spa_scrub_lock
);
2310 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2311 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2312 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2313 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2319 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2321 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2322 return (SET_ERROR(ENAMETOOLONG
));
2328 spa_load_verify(spa_t
*spa
)
2331 spa_load_error_t sle
= { 0 };
2332 zpool_load_policy_t policy
;
2333 boolean_t verify_ok
= B_FALSE
;
2336 zpool_get_load_policy(spa
->spa_config
, &policy
);
2338 if (policy
.zlp_rewind
& ZPOOL_NEVER_REWIND
)
2341 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2342 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2343 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2345 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2349 rio
= zio_root(spa
, NULL
, &sle
,
2350 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2352 if (spa_load_verify_metadata
) {
2353 if (spa
->spa_extreme_rewind
) {
2354 spa_load_note(spa
, "performing a complete scan of the "
2355 "pool since extreme rewind is on. This may take "
2356 "a very long time.\n (spa_load_verify_data=%u, "
2357 "spa_load_verify_metadata=%u)",
2358 spa_load_verify_data
, spa_load_verify_metadata
);
2361 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2362 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2363 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2366 (void) zio_wait(rio
);
2367 ASSERT0(spa
->spa_load_verify_bytes
);
2369 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2370 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2372 if (sle
.sle_meta_count
!= 0 || sle
.sle_data_count
!= 0) {
2373 spa_load_note(spa
, "spa_load_verify found %llu metadata errors "
2374 "and %llu data errors", (u_longlong_t
)sle
.sle_meta_count
,
2375 (u_longlong_t
)sle
.sle_data_count
);
2378 if (spa_load_verify_dryrun
||
2379 (!error
&& sle
.sle_meta_count
<= policy
.zlp_maxmeta
&&
2380 sle
.sle_data_count
<= policy
.zlp_maxdata
)) {
2384 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2385 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2387 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2388 fnvlist_add_uint64(spa
->spa_load_info
, ZPOOL_CONFIG_LOAD_TIME
,
2389 spa
->spa_load_txg_ts
);
2390 fnvlist_add_int64(spa
->spa_load_info
, ZPOOL_CONFIG_REWIND_TIME
,
2392 fnvlist_add_uint64(spa
->spa_load_info
,
2393 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
);
2395 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2398 if (spa_load_verify_dryrun
)
2402 if (error
!= ENXIO
&& error
!= EIO
)
2403 error
= SET_ERROR(EIO
);
2407 return (verify_ok
? 0 : EIO
);
2411 * Find a value in the pool props object.
2414 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2416 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2417 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2421 * Find a value in the pool directory object.
2424 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
, boolean_t log_enoent
)
2426 int error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2427 name
, sizeof (uint64_t), 1, val
);
2429 if (error
!= 0 && (error
!= ENOENT
|| log_enoent
)) {
2430 spa_load_failed(spa
, "couldn't get '%s' value in MOS directory "
2431 "[error=%d]", name
, error
);
2438 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2440 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2441 return (SET_ERROR(err
));
2445 spa_livelist_delete_check(spa_t
*spa
)
2447 return (spa
->spa_livelists_to_delete
!= 0);
2452 spa_livelist_delete_cb_check(void *arg
, zthr_t
*z
)
2455 return (spa_livelist_delete_check(spa
));
2459 delete_blkptr_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
2462 zio_free(spa
, tx
->tx_txg
, bp
);
2463 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
, DD_USED_HEAD
,
2464 -bp_get_dsize_sync(spa
, bp
),
2465 -BP_GET_PSIZE(bp
), -BP_GET_UCSIZE(bp
), tx
);
2470 dsl_get_next_livelist_obj(objset_t
*os
, uint64_t zap_obj
, uint64_t *llp
)
2475 zap_cursor_init(&zc
, os
, zap_obj
);
2476 err
= zap_cursor_retrieve(&zc
, &za
);
2477 zap_cursor_fini(&zc
);
2479 *llp
= za
.za_first_integer
;
2484 * Components of livelist deletion that must be performed in syncing
2485 * context: freeing block pointers and updating the pool-wide data
2486 * structures to indicate how much work is left to do
2488 typedef struct sublist_delete_arg
{
2493 } sublist_delete_arg_t
;
2496 sublist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2498 sublist_delete_arg_t
*sda
= arg
;
2499 spa_t
*spa
= sda
->spa
;
2500 dsl_deadlist_t
*ll
= sda
->ll
;
2501 uint64_t key
= sda
->key
;
2502 bplist_t
*to_free
= sda
->to_free
;
2504 bplist_iterate(to_free
, delete_blkptr_cb
, spa
, tx
);
2505 dsl_deadlist_remove_entry(ll
, key
, tx
);
2508 typedef struct livelist_delete_arg
{
2512 } livelist_delete_arg_t
;
2515 livelist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2517 livelist_delete_arg_t
*lda
= arg
;
2518 spa_t
*spa
= lda
->spa
;
2519 uint64_t ll_obj
= lda
->ll_obj
;
2520 uint64_t zap_obj
= lda
->zap_obj
;
2521 objset_t
*mos
= spa
->spa_meta_objset
;
2524 /* free the livelist and decrement the feature count */
2525 VERIFY0(zap_remove_int(mos
, zap_obj
, ll_obj
, tx
));
2526 dsl_deadlist_free(mos
, ll_obj
, tx
);
2527 spa_feature_decr(spa
, SPA_FEATURE_LIVELIST
, tx
);
2528 VERIFY0(zap_count(mos
, zap_obj
, &count
));
2530 /* no more livelists to delete */
2531 VERIFY0(zap_remove(mos
, DMU_POOL_DIRECTORY_OBJECT
,
2532 DMU_POOL_DELETED_CLONES
, tx
));
2533 VERIFY0(zap_destroy(mos
, zap_obj
, tx
));
2534 spa
->spa_livelists_to_delete
= 0;
2535 spa_notify_waiters(spa
);
2540 * Load in the value for the livelist to be removed and open it. Then,
2541 * load its first sublist and determine which block pointers should actually
2542 * be freed. Then, call a synctask which performs the actual frees and updates
2543 * the pool-wide livelist data.
2547 spa_livelist_delete_cb(void *arg
, zthr_t
*z
)
2550 uint64_t ll_obj
= 0, count
;
2551 objset_t
*mos
= spa
->spa_meta_objset
;
2552 uint64_t zap_obj
= spa
->spa_livelists_to_delete
;
2554 * Determine the next livelist to delete. This function should only
2555 * be called if there is at least one deleted clone.
2557 VERIFY0(dsl_get_next_livelist_obj(mos
, zap_obj
, &ll_obj
));
2558 VERIFY0(zap_count(mos
, ll_obj
, &count
));
2561 dsl_deadlist_entry_t
*dle
;
2563 ll
= kmem_zalloc(sizeof (dsl_deadlist_t
), KM_SLEEP
);
2564 dsl_deadlist_open(ll
, mos
, ll_obj
);
2565 dle
= dsl_deadlist_first(ll
);
2566 ASSERT3P(dle
, !=, NULL
);
2567 bplist_create(&to_free
);
2568 int err
= dsl_process_sub_livelist(&dle
->dle_bpobj
, &to_free
,
2571 sublist_delete_arg_t sync_arg
= {
2574 .key
= dle
->dle_mintxg
,
2577 zfs_dbgmsg("deleting sublist (id %llu) from"
2578 " livelist %llu, %lld remaining",
2579 (u_longlong_t
)dle
->dle_bpobj
.bpo_object
,
2580 (u_longlong_t
)ll_obj
, (longlong_t
)count
- 1);
2581 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
,
2582 sublist_delete_sync
, &sync_arg
, 0,
2583 ZFS_SPACE_CHECK_DESTROY
));
2585 VERIFY3U(err
, ==, EINTR
);
2587 bplist_clear(&to_free
);
2588 bplist_destroy(&to_free
);
2589 dsl_deadlist_close(ll
);
2590 kmem_free(ll
, sizeof (dsl_deadlist_t
));
2592 livelist_delete_arg_t sync_arg
= {
2597 zfs_dbgmsg("deletion of livelist %llu completed",
2598 (u_longlong_t
)ll_obj
);
2599 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
, livelist_delete_sync
,
2600 &sync_arg
, 0, ZFS_SPACE_CHECK_DESTROY
));
2605 spa_start_livelist_destroy_thread(spa_t
*spa
)
2607 ASSERT3P(spa
->spa_livelist_delete_zthr
, ==, NULL
);
2608 spa
->spa_livelist_delete_zthr
=
2609 zthr_create("z_livelist_destroy",
2610 spa_livelist_delete_cb_check
, spa_livelist_delete_cb
, spa
,
2614 typedef struct livelist_new_arg
{
2617 } livelist_new_arg_t
;
2620 livelist_track_new_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
2624 livelist_new_arg_t
*lna
= arg
;
2626 bplist_append(lna
->frees
, bp
);
2628 bplist_append(lna
->allocs
, bp
);
2629 zfs_livelist_condense_new_alloc
++;
2634 typedef struct livelist_condense_arg
{
2637 uint64_t first_size
;
2639 } livelist_condense_arg_t
;
2642 spa_livelist_condense_sync(void *arg
, dmu_tx_t
*tx
)
2644 livelist_condense_arg_t
*lca
= arg
;
2645 spa_t
*spa
= lca
->spa
;
2647 dsl_dataset_t
*ds
= spa
->spa_to_condense
.ds
;
2649 /* Have we been cancelled? */
2650 if (spa
->spa_to_condense
.cancelled
) {
2651 zfs_livelist_condense_sync_cancel
++;
2655 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
2656 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
2657 dsl_deadlist_t
*ll
= &ds
->ds_dir
->dd_livelist
;
2660 * It's possible that the livelist was changed while the zthr was
2661 * running. Therefore, we need to check for new blkptrs in the two
2662 * entries being condensed and continue to track them in the livelist.
2663 * Because of the way we handle remapped blkptrs (see dbuf_remap_impl),
2664 * it's possible that the newly added blkptrs are FREEs or ALLOCs so
2665 * we need to sort them into two different bplists.
2667 uint64_t first_obj
= first
->dle_bpobj
.bpo_object
;
2668 uint64_t next_obj
= next
->dle_bpobj
.bpo_object
;
2669 uint64_t cur_first_size
= first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
2670 uint64_t cur_next_size
= next
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
2672 bplist_create(&new_frees
);
2673 livelist_new_arg_t new_bps
= {
2674 .allocs
= &lca
->to_keep
,
2675 .frees
= &new_frees
,
2678 if (cur_first_size
> lca
->first_size
) {
2679 VERIFY0(livelist_bpobj_iterate_from_nofree(&first
->dle_bpobj
,
2680 livelist_track_new_cb
, &new_bps
, lca
->first_size
));
2682 if (cur_next_size
> lca
->next_size
) {
2683 VERIFY0(livelist_bpobj_iterate_from_nofree(&next
->dle_bpobj
,
2684 livelist_track_new_cb
, &new_bps
, lca
->next_size
));
2687 dsl_deadlist_clear_entry(first
, ll
, tx
);
2688 ASSERT(bpobj_is_empty(&first
->dle_bpobj
));
2689 dsl_deadlist_remove_entry(ll
, next
->dle_mintxg
, tx
);
2691 bplist_iterate(&lca
->to_keep
, dsl_deadlist_insert_alloc_cb
, ll
, tx
);
2692 bplist_iterate(&new_frees
, dsl_deadlist_insert_free_cb
, ll
, tx
);
2693 bplist_destroy(&new_frees
);
2695 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
2696 dsl_dataset_name(ds
, dsname
);
2697 zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu "
2698 "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu "
2699 "(%llu blkptrs)", (u_longlong_t
)tx
->tx_txg
, dsname
,
2700 (u_longlong_t
)ds
->ds_object
, (u_longlong_t
)first_obj
,
2701 (u_longlong_t
)cur_first_size
, (u_longlong_t
)next_obj
,
2702 (u_longlong_t
)cur_next_size
,
2703 (u_longlong_t
)first
->dle_bpobj
.bpo_object
,
2704 (u_longlong_t
)first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
);
2706 dmu_buf_rele(ds
->ds_dbuf
, spa
);
2707 spa
->spa_to_condense
.ds
= NULL
;
2708 bplist_clear(&lca
->to_keep
);
2709 bplist_destroy(&lca
->to_keep
);
2710 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
2711 spa
->spa_to_condense
.syncing
= B_FALSE
;
2715 spa_livelist_condense_cb(void *arg
, zthr_t
*t
)
2717 while (zfs_livelist_condense_zthr_pause
&&
2718 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
2722 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
2723 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
2724 uint64_t first_size
, next_size
;
2726 livelist_condense_arg_t
*lca
=
2727 kmem_alloc(sizeof (livelist_condense_arg_t
), KM_SLEEP
);
2728 bplist_create(&lca
->to_keep
);
2731 * Process the livelists (matching FREEs and ALLOCs) in open context
2732 * so we have minimal work in syncing context to condense.
2734 * We save bpobj sizes (first_size and next_size) to use later in
2735 * syncing context to determine if entries were added to these sublists
2736 * while in open context. This is possible because the clone is still
2737 * active and open for normal writes and we want to make sure the new,
2738 * unprocessed blockpointers are inserted into the livelist normally.
2740 * Note that dsl_process_sub_livelist() both stores the size number of
2741 * blockpointers and iterates over them while the bpobj's lock held, so
2742 * the sizes returned to us are consistent which what was actually
2745 int err
= dsl_process_sub_livelist(&first
->dle_bpobj
, &lca
->to_keep
, t
,
2748 err
= dsl_process_sub_livelist(&next
->dle_bpobj
, &lca
->to_keep
,
2752 while (zfs_livelist_condense_sync_pause
&&
2753 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
2756 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
2757 dmu_tx_mark_netfree(tx
);
2758 dmu_tx_hold_space(tx
, 1);
2759 err
= dmu_tx_assign(tx
, TXG_NOWAIT
| TXG_NOTHROTTLE
);
2762 * Prevent the condense zthr restarting before
2763 * the synctask completes.
2765 spa
->spa_to_condense
.syncing
= B_TRUE
;
2767 lca
->first_size
= first_size
;
2768 lca
->next_size
= next_size
;
2769 dsl_sync_task_nowait(spa_get_dsl(spa
),
2770 spa_livelist_condense_sync
, lca
, tx
);
2776 * Condensing can not continue: either it was externally stopped or
2777 * we were unable to assign to a tx because the pool has run out of
2778 * space. In the second case, we'll just end up trying to condense
2779 * again in a later txg.
2782 bplist_clear(&lca
->to_keep
);
2783 bplist_destroy(&lca
->to_keep
);
2784 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
2785 dmu_buf_rele(spa
->spa_to_condense
.ds
->ds_dbuf
, spa
);
2786 spa
->spa_to_condense
.ds
= NULL
;
2788 zfs_livelist_condense_zthr_cancel
++;
2793 * Check that there is something to condense but that a condense is not
2794 * already in progress and that condensing has not been cancelled.
2797 spa_livelist_condense_cb_check(void *arg
, zthr_t
*z
)
2800 if ((spa
->spa_to_condense
.ds
!= NULL
) &&
2801 (spa
->spa_to_condense
.syncing
== B_FALSE
) &&
2802 (spa
->spa_to_condense
.cancelled
== B_FALSE
)) {
2809 spa_start_livelist_condensing_thread(spa_t
*spa
)
2811 spa
->spa_to_condense
.ds
= NULL
;
2812 spa
->spa_to_condense
.first
= NULL
;
2813 spa
->spa_to_condense
.next
= NULL
;
2814 spa
->spa_to_condense
.syncing
= B_FALSE
;
2815 spa
->spa_to_condense
.cancelled
= B_FALSE
;
2817 ASSERT3P(spa
->spa_livelist_condense_zthr
, ==, NULL
);
2818 spa
->spa_livelist_condense_zthr
=
2819 zthr_create("z_livelist_condense",
2820 spa_livelist_condense_cb_check
,
2821 spa_livelist_condense_cb
, spa
, minclsyspri
);
2825 spa_spawn_aux_threads(spa_t
*spa
)
2827 ASSERT(spa_writeable(spa
));
2829 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2831 spa_start_indirect_condensing_thread(spa
);
2832 spa_start_livelist_destroy_thread(spa
);
2833 spa_start_livelist_condensing_thread(spa
);
2835 ASSERT3P(spa
->spa_checkpoint_discard_zthr
, ==, NULL
);
2836 spa
->spa_checkpoint_discard_zthr
=
2837 zthr_create("z_checkpoint_discard",
2838 spa_checkpoint_discard_thread_check
,
2839 spa_checkpoint_discard_thread
, spa
, minclsyspri
);
2843 * Fix up config after a partly-completed split. This is done with the
2844 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2845 * pool have that entry in their config, but only the splitting one contains
2846 * a list of all the guids of the vdevs that are being split off.
2848 * This function determines what to do with that list: either rejoin
2849 * all the disks to the pool, or complete the splitting process. To attempt
2850 * the rejoin, each disk that is offlined is marked online again, and
2851 * we do a reopen() call. If the vdev label for every disk that was
2852 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2853 * then we call vdev_split() on each disk, and complete the split.
2855 * Otherwise we leave the config alone, with all the vdevs in place in
2856 * the original pool.
2859 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2866 boolean_t attempt_reopen
;
2868 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2871 /* check that the config is complete */
2872 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2873 &glist
, &gcount
) != 0)
2876 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2878 /* attempt to online all the vdevs & validate */
2879 attempt_reopen
= B_TRUE
;
2880 for (i
= 0; i
< gcount
; i
++) {
2881 if (glist
[i
] == 0) /* vdev is hole */
2884 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2885 if (vd
[i
] == NULL
) {
2887 * Don't bother attempting to reopen the disks;
2888 * just do the split.
2890 attempt_reopen
= B_FALSE
;
2892 /* attempt to re-online it */
2893 vd
[i
]->vdev_offline
= B_FALSE
;
2897 if (attempt_reopen
) {
2898 vdev_reopen(spa
->spa_root_vdev
);
2900 /* check each device to see what state it's in */
2901 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2902 if (vd
[i
] != NULL
&&
2903 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2910 * If every disk has been moved to the new pool, or if we never
2911 * even attempted to look at them, then we split them off for
2914 if (!attempt_reopen
|| gcount
== extracted
) {
2915 for (i
= 0; i
< gcount
; i
++)
2918 vdev_reopen(spa
->spa_root_vdev
);
2921 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2925 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
)
2927 char *ereport
= FM_EREPORT_ZFS_POOL
;
2930 spa
->spa_load_state
= state
;
2931 (void) spa_import_progress_set_state(spa_guid(spa
),
2932 spa_load_state(spa
));
2934 gethrestime(&spa
->spa_loaded_ts
);
2935 error
= spa_load_impl(spa
, type
, &ereport
);
2938 * Don't count references from objsets that are already closed
2939 * and are making their way through the eviction process.
2941 spa_evicting_os_wait(spa
);
2942 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
2944 if (error
!= EEXIST
) {
2945 spa
->spa_loaded_ts
.tv_sec
= 0;
2946 spa
->spa_loaded_ts
.tv_nsec
= 0;
2948 if (error
!= EBADF
) {
2949 (void) zfs_ereport_post(ereport
, spa
,
2950 NULL
, NULL
, NULL
, 0);
2953 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2956 (void) spa_import_progress_set_state(spa_guid(spa
),
2957 spa_load_state(spa
));
2964 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2965 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2966 * spa's per-vdev ZAP list.
2969 vdev_count_verify_zaps(vdev_t
*vd
)
2971 spa_t
*spa
= vd
->vdev_spa
;
2974 if (vd
->vdev_top_zap
!= 0) {
2976 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2977 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2979 if (vd
->vdev_leaf_zap
!= 0) {
2981 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2982 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2985 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2986 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2994 * Determine whether the activity check is required.
2997 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
3001 uint64_t hostid
= 0;
3002 uint64_t tryconfig_txg
= 0;
3003 uint64_t tryconfig_timestamp
= 0;
3004 uint16_t tryconfig_mmp_seq
= 0;
3007 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3008 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
3009 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
3011 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
3012 &tryconfig_timestamp
);
3013 (void) nvlist_lookup_uint16(nvinfo
, ZPOOL_CONFIG_MMP_SEQ
,
3014 &tryconfig_mmp_seq
);
3017 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
3020 * Disable the MMP activity check - This is used by zdb which
3021 * is intended to be used on potentially active pools.
3023 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
3027 * Skip the activity check when the MMP feature is disabled.
3029 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
3033 * If the tryconfig_ values are nonzero, they are the results of an
3034 * earlier tryimport. If they all match the uberblock we just found,
3035 * then the pool has not changed and we return false so we do not test
3038 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
3039 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
&&
3040 tryconfig_mmp_seq
&& tryconfig_mmp_seq
==
3041 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0))
3045 * Allow the activity check to be skipped when importing the pool
3046 * on the same host which last imported it. Since the hostid from
3047 * configuration may be stale use the one read from the label.
3049 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
3050 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
3052 if (hostid
== spa_get_hostid(spa
))
3056 * Skip the activity test when the pool was cleanly exported.
3058 if (state
!= POOL_STATE_ACTIVE
)
3065 * Nanoseconds the activity check must watch for changes on-disk.
3068 spa_activity_check_duration(spa_t
*spa
, uberblock_t
*ub
)
3070 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
3071 uint64_t multihost_interval
= MSEC2NSEC(
3072 MMP_INTERVAL_OK(zfs_multihost_interval
));
3073 uint64_t import_delay
= MAX(NANOSEC
, import_intervals
*
3074 multihost_interval
);
3077 * Local tunables determine a minimum duration except for the case
3078 * where we know when the remote host will suspend the pool if MMP
3079 * writes do not land.
3081 * See Big Theory comment at the top of mmp.c for the reasoning behind
3082 * these cases and times.
3085 ASSERT(MMP_IMPORT_SAFETY_FACTOR
>= 100);
3087 if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3088 MMP_FAIL_INT(ub
) > 0) {
3090 /* MMP on remote host will suspend pool after failed writes */
3091 import_delay
= MMP_FAIL_INT(ub
) * MSEC2NSEC(MMP_INTERVAL(ub
)) *
3092 MMP_IMPORT_SAFETY_FACTOR
/ 100;
3094 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3095 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3096 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3097 (u_longlong_t
)MMP_FAIL_INT(ub
),
3098 (u_longlong_t
)MMP_INTERVAL(ub
),
3099 (u_longlong_t
)import_intervals
);
3101 } else if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3102 MMP_FAIL_INT(ub
) == 0) {
3104 /* MMP on remote host will never suspend pool */
3105 import_delay
= MAX(import_delay
, (MSEC2NSEC(MMP_INTERVAL(ub
)) +
3106 ub
->ub_mmp_delay
) * import_intervals
);
3108 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3109 "mmp_interval=%llu ub_mmp_delay=%llu "
3110 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3111 (u_longlong_t
)MMP_INTERVAL(ub
),
3112 (u_longlong_t
)ub
->ub_mmp_delay
,
3113 (u_longlong_t
)import_intervals
);
3115 } else if (MMP_VALID(ub
)) {
3117 * zfs-0.7 compatibility case
3120 import_delay
= MAX(import_delay
, (multihost_interval
+
3121 ub
->ub_mmp_delay
) * import_intervals
);
3123 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3124 "import_intervals=%llu leaves=%u",
3125 (u_longlong_t
)import_delay
,
3126 (u_longlong_t
)ub
->ub_mmp_delay
,
3127 (u_longlong_t
)import_intervals
,
3128 vdev_count_leaves(spa
));
3130 /* Using local tunings is the only reasonable option */
3131 zfs_dbgmsg("pool last imported on non-MMP aware "
3132 "host using import_delay=%llu multihost_interval=%llu "
3133 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3134 (u_longlong_t
)multihost_interval
,
3135 (u_longlong_t
)import_intervals
);
3138 return (import_delay
);
3142 * Perform the import activity check. If the user canceled the import or
3143 * we detected activity then fail.
3146 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
3148 uint64_t txg
= ub
->ub_txg
;
3149 uint64_t timestamp
= ub
->ub_timestamp
;
3150 uint64_t mmp_config
= ub
->ub_mmp_config
;
3151 uint16_t mmp_seq
= MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0;
3152 uint64_t import_delay
;
3153 hrtime_t import_expire
;
3154 nvlist_t
*mmp_label
= NULL
;
3155 vdev_t
*rvd
= spa
->spa_root_vdev
;
3160 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
3161 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3165 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3166 * during the earlier tryimport. If the txg recorded there is 0 then
3167 * the pool is known to be active on another host.
3169 * Otherwise, the pool might be in use on another host. Check for
3170 * changes in the uberblocks on disk if necessary.
3172 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3173 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
3174 ZPOOL_CONFIG_LOAD_INFO
);
3176 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
3177 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
3178 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3179 error
= SET_ERROR(EREMOTEIO
);
3184 import_delay
= spa_activity_check_duration(spa
, ub
);
3186 /* Add a small random factor in case of simultaneous imports (0-25%) */
3187 import_delay
+= import_delay
* random_in_range(250) / 1000;
3189 import_expire
= gethrtime() + import_delay
;
3191 while (gethrtime() < import_expire
) {
3192 (void) spa_import_progress_set_mmp_check(spa_guid(spa
),
3193 NSEC2SEC(import_expire
- gethrtime()));
3195 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3197 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
||
3198 mmp_seq
!= (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0)) {
3199 zfs_dbgmsg("multihost activity detected "
3200 "txg %llu ub_txg %llu "
3201 "timestamp %llu ub_timestamp %llu "
3202 "mmp_config %#llx ub_mmp_config %#llx",
3203 (u_longlong_t
)txg
, (u_longlong_t
)ub
->ub_txg
,
3204 (u_longlong_t
)timestamp
,
3205 (u_longlong_t
)ub
->ub_timestamp
,
3206 (u_longlong_t
)mmp_config
,
3207 (u_longlong_t
)ub
->ub_mmp_config
);
3209 error
= SET_ERROR(EREMOTEIO
);
3214 nvlist_free(mmp_label
);
3218 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
3220 error
= SET_ERROR(EINTR
);
3228 mutex_destroy(&mtx
);
3232 * If the pool is determined to be active store the status in the
3233 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3234 * available from configuration read from disk store them as well.
3235 * This allows 'zpool import' to generate a more useful message.
3237 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3238 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3239 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3241 if (error
== EREMOTEIO
) {
3242 char *hostname
= "<unknown>";
3243 uint64_t hostid
= 0;
3246 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
3247 hostname
= fnvlist_lookup_string(mmp_label
,
3248 ZPOOL_CONFIG_HOSTNAME
);
3249 fnvlist_add_string(spa
->spa_load_info
,
3250 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
3253 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
3254 hostid
= fnvlist_lookup_uint64(mmp_label
,
3255 ZPOOL_CONFIG_HOSTID
);
3256 fnvlist_add_uint64(spa
->spa_load_info
,
3257 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
3261 fnvlist_add_uint64(spa
->spa_load_info
,
3262 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
3263 fnvlist_add_uint64(spa
->spa_load_info
,
3264 ZPOOL_CONFIG_MMP_TXG
, 0);
3266 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
3270 nvlist_free(mmp_label
);
3276 spa_verify_host(spa_t
*spa
, nvlist_t
*mos_config
)
3280 uint64_t myhostid
= 0;
3282 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
3283 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
3284 hostname
= fnvlist_lookup_string(mos_config
,
3285 ZPOOL_CONFIG_HOSTNAME
);
3287 myhostid
= zone_get_hostid(NULL
);
3289 if (hostid
!= 0 && myhostid
!= 0 && hostid
!= myhostid
) {
3290 cmn_err(CE_WARN
, "pool '%s' could not be "
3291 "loaded as it was last accessed by "
3292 "another system (host: %s hostid: 0x%llx). "
3293 "See: https://openzfs.github.io/openzfs-docs/msg/"
3295 spa_name(spa
), hostname
, (u_longlong_t
)hostid
);
3296 spa_load_failed(spa
, "hostid verification failed: pool "
3297 "last accessed by host: %s (hostid: 0x%llx)",
3298 hostname
, (u_longlong_t
)hostid
);
3299 return (SET_ERROR(EBADF
));
3307 spa_ld_parse_config(spa_t
*spa
, spa_import_type_t type
)
3310 nvlist_t
*nvtree
, *nvl
, *config
= spa
->spa_config
;
3315 char *compatibility
;
3318 * Versioning wasn't explicitly added to the label until later, so if
3319 * it's not present treat it as the initial version.
3321 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
3322 &spa
->spa_ubsync
.ub_version
) != 0)
3323 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
3325 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
3326 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3327 ZPOOL_CONFIG_POOL_GUID
);
3328 return (SET_ERROR(EINVAL
));
3332 * If we are doing an import, ensure that the pool is not already
3333 * imported by checking if its pool guid already exists in the
3336 * The only case that we allow an already imported pool to be
3337 * imported again, is when the pool is checkpointed and we want to
3338 * look at its checkpointed state from userland tools like zdb.
3341 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3342 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3343 spa_guid_exists(pool_guid
, 0)) {
3345 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3346 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3347 spa_guid_exists(pool_guid
, 0) &&
3348 !spa_importing_readonly_checkpoint(spa
)) {
3350 spa_load_failed(spa
, "a pool with guid %llu is already open",
3351 (u_longlong_t
)pool_guid
);
3352 return (SET_ERROR(EEXIST
));
3355 spa
->spa_config_guid
= pool_guid
;
3357 nvlist_free(spa
->spa_load_info
);
3358 spa
->spa_load_info
= fnvlist_alloc();
3360 ASSERT(spa
->spa_comment
== NULL
);
3361 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
3362 spa
->spa_comment
= spa_strdup(comment
);
3364 ASSERT(spa
->spa_compatibility
== NULL
);
3365 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMPATIBILITY
,
3366 &compatibility
) == 0)
3367 spa
->spa_compatibility
= spa_strdup(compatibility
);
3369 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
3370 &spa
->spa_config_txg
);
3372 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) == 0)
3373 spa
->spa_config_splitting
= fnvlist_dup(nvl
);
3375 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
)) {
3376 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3377 ZPOOL_CONFIG_VDEV_TREE
);
3378 return (SET_ERROR(EINVAL
));
3382 * Create "The Godfather" zio to hold all async IOs
3384 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3386 for (int i
= 0; i
< max_ncpus
; i
++) {
3387 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3388 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3389 ZIO_FLAG_GODFATHER
);
3393 * Parse the configuration into a vdev tree. We explicitly set the
3394 * value that will be returned by spa_version() since parsing the
3395 * configuration requires knowing the version number.
3397 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3398 parse
= (type
== SPA_IMPORT_EXISTING
?
3399 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
3400 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
3401 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3404 spa_load_failed(spa
, "unable to parse config [error=%d]",
3409 ASSERT(spa
->spa_root_vdev
== rvd
);
3410 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
3411 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
3413 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3414 ASSERT(spa_guid(spa
) == pool_guid
);
3421 * Recursively open all vdevs in the vdev tree. This function is called twice:
3422 * first with the untrusted config, then with the trusted config.
3425 spa_ld_open_vdevs(spa_t
*spa
)
3430 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3431 * missing/unopenable for the root vdev to be still considered openable.
3433 if (spa
->spa_trust_config
) {
3434 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds
;
3435 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_CACHEFILE
) {
3436 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_cachefile
;
3437 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_SCAN
) {
3438 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_scan
;
3440 spa
->spa_missing_tvds_allowed
= 0;
3443 spa
->spa_missing_tvds_allowed
=
3444 MAX(zfs_max_missing_tvds
, spa
->spa_missing_tvds_allowed
);
3446 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3447 error
= vdev_open(spa
->spa_root_vdev
);
3448 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3450 if (spa
->spa_missing_tvds
!= 0) {
3451 spa_load_note(spa
, "vdev tree has %lld missing top-level "
3452 "vdevs.", (u_longlong_t
)spa
->spa_missing_tvds
);
3453 if (spa
->spa_trust_config
&& (spa
->spa_mode
& SPA_MODE_WRITE
)) {
3455 * Although theoretically we could allow users to open
3456 * incomplete pools in RW mode, we'd need to add a lot
3457 * of extra logic (e.g. adjust pool space to account
3458 * for missing vdevs).
3459 * This limitation also prevents users from accidentally
3460 * opening the pool in RW mode during data recovery and
3461 * damaging it further.
3463 spa_load_note(spa
, "pools with missing top-level "
3464 "vdevs can only be opened in read-only mode.");
3465 error
= SET_ERROR(ENXIO
);
3467 spa_load_note(spa
, "current settings allow for maximum "
3468 "%lld missing top-level vdevs at this stage.",
3469 (u_longlong_t
)spa
->spa_missing_tvds_allowed
);
3473 spa_load_failed(spa
, "unable to open vdev tree [error=%d]",
3476 if (spa
->spa_missing_tvds
!= 0 || error
!= 0)
3477 vdev_dbgmsg_print_tree(spa
->spa_root_vdev
, 2);
3483 * We need to validate the vdev labels against the configuration that
3484 * we have in hand. This function is called twice: first with an untrusted
3485 * config, then with a trusted config. The validation is more strict when the
3486 * config is trusted.
3489 spa_ld_validate_vdevs(spa_t
*spa
)
3492 vdev_t
*rvd
= spa
->spa_root_vdev
;
3494 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3495 error
= vdev_validate(rvd
);
3496 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3499 spa_load_failed(spa
, "vdev_validate failed [error=%d]", error
);
3503 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
3504 spa_load_failed(spa
, "cannot open vdev tree after invalidating "
3506 vdev_dbgmsg_print_tree(rvd
, 2);
3507 return (SET_ERROR(ENXIO
));
3514 spa_ld_select_uberblock_done(spa_t
*spa
, uberblock_t
*ub
)
3516 spa
->spa_state
= POOL_STATE_ACTIVE
;
3517 spa
->spa_ubsync
= spa
->spa_uberblock
;
3518 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
3519 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
3520 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
3521 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
3522 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
3523 spa
->spa_prev_software_version
= ub
->ub_software_version
;
3527 spa_ld_select_uberblock(spa_t
*spa
, spa_import_type_t type
)
3529 vdev_t
*rvd
= spa
->spa_root_vdev
;
3531 uberblock_t
*ub
= &spa
->spa_uberblock
;
3532 boolean_t activity_check
= B_FALSE
;
3535 * If we are opening the checkpointed state of the pool by
3536 * rewinding to it, at this point we will have written the
3537 * checkpointed uberblock to the vdev labels, so searching
3538 * the labels will find the right uberblock. However, if
3539 * we are opening the checkpointed state read-only, we have
3540 * not modified the labels. Therefore, we must ignore the
3541 * labels and continue using the spa_uberblock that was set
3542 * by spa_ld_checkpoint_rewind.
3544 * Note that it would be fine to ignore the labels when
3545 * rewinding (opening writeable) as well. However, if we
3546 * crash just after writing the labels, we will end up
3547 * searching the labels. Doing so in the common case means
3548 * that this code path gets exercised normally, rather than
3549 * just in the edge case.
3551 if (ub
->ub_checkpoint_txg
!= 0 &&
3552 spa_importing_readonly_checkpoint(spa
)) {
3553 spa_ld_select_uberblock_done(spa
, ub
);
3558 * Find the best uberblock.
3560 vdev_uberblock_load(rvd
, ub
, &label
);
3563 * If we weren't able to find a single valid uberblock, return failure.
3565 if (ub
->ub_txg
== 0) {
3567 spa_load_failed(spa
, "no valid uberblock found");
3568 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
3571 if (spa
->spa_load_max_txg
!= UINT64_MAX
) {
3572 (void) spa_import_progress_set_max_txg(spa_guid(spa
),
3573 (u_longlong_t
)spa
->spa_load_max_txg
);
3575 spa_load_note(spa
, "using uberblock with txg=%llu",
3576 (u_longlong_t
)ub
->ub_txg
);
3580 * For pools which have the multihost property on determine if the
3581 * pool is truly inactive and can be safely imported. Prevent
3582 * hosts which don't have a hostid set from importing the pool.
3584 activity_check
= spa_activity_check_required(spa
, ub
, label
,
3586 if (activity_check
) {
3587 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
3588 spa_get_hostid(spa
) == 0) {
3590 fnvlist_add_uint64(spa
->spa_load_info
,
3591 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3592 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3595 int error
= spa_activity_check(spa
, ub
, spa
->spa_config
);
3601 fnvlist_add_uint64(spa
->spa_load_info
,
3602 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
3603 fnvlist_add_uint64(spa
->spa_load_info
,
3604 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
3605 fnvlist_add_uint16(spa
->spa_load_info
,
3606 ZPOOL_CONFIG_MMP_SEQ
,
3607 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0));
3611 * If the pool has an unsupported version we can't open it.
3613 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
3615 spa_load_failed(spa
, "version %llu is not supported",
3616 (u_longlong_t
)ub
->ub_version
);
3617 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
3620 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3624 * If we weren't able to find what's necessary for reading the
3625 * MOS in the label, return failure.
3627 if (label
== NULL
) {
3628 spa_load_failed(spa
, "label config unavailable");
3629 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3633 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
3636 spa_load_failed(spa
, "invalid label: '%s' missing",
3637 ZPOOL_CONFIG_FEATURES_FOR_READ
);
3638 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3643 * Update our in-core representation with the definitive values
3646 nvlist_free(spa
->spa_label_features
);
3647 spa
->spa_label_features
= fnvlist_dup(features
);
3653 * Look through entries in the label nvlist's features_for_read. If
3654 * there is a feature listed there which we don't understand then we
3655 * cannot open a pool.
3657 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3658 nvlist_t
*unsup_feat
;
3660 unsup_feat
= fnvlist_alloc();
3662 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
3664 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
3665 if (!zfeature_is_supported(nvpair_name(nvp
))) {
3666 fnvlist_add_string(unsup_feat
,
3667 nvpair_name(nvp
), "");
3671 if (!nvlist_empty(unsup_feat
)) {
3672 fnvlist_add_nvlist(spa
->spa_load_info
,
3673 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
3674 nvlist_free(unsup_feat
);
3675 spa_load_failed(spa
, "some features are unsupported");
3676 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
3680 nvlist_free(unsup_feat
);
3683 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
3684 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3685 spa_try_repair(spa
, spa
->spa_config
);
3686 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3687 nvlist_free(spa
->spa_config_splitting
);
3688 spa
->spa_config_splitting
= NULL
;
3692 * Initialize internal SPA structures.
3694 spa_ld_select_uberblock_done(spa
, ub
);
3700 spa_ld_open_rootbp(spa_t
*spa
)
3703 vdev_t
*rvd
= spa
->spa_root_vdev
;
3705 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
3707 spa_load_failed(spa
, "unable to open rootbp in dsl_pool_init "
3708 "[error=%d]", error
);
3709 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3711 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
3717 spa_ld_trusted_config(spa_t
*spa
, spa_import_type_t type
,
3718 boolean_t reloading
)
3720 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
3721 nvlist_t
*nv
, *mos_config
, *policy
;
3722 int error
= 0, copy_error
;
3723 uint64_t healthy_tvds
, healthy_tvds_mos
;
3724 uint64_t mos_config_txg
;
3726 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
, B_TRUE
)
3728 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3731 * If we're assembling a pool from a split, the config provided is
3732 * already trusted so there is nothing to do.
3734 if (type
== SPA_IMPORT_ASSEMBLE
)
3737 healthy_tvds
= spa_healthy_core_tvds(spa
);
3739 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
)
3741 spa_load_failed(spa
, "unable to retrieve MOS config");
3742 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3746 * If we are doing an open, pool owner wasn't verified yet, thus do
3747 * the verification here.
3749 if (spa
->spa_load_state
== SPA_LOAD_OPEN
) {
3750 error
= spa_verify_host(spa
, mos_config
);
3752 nvlist_free(mos_config
);
3757 nv
= fnvlist_lookup_nvlist(mos_config
, ZPOOL_CONFIG_VDEV_TREE
);
3759 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3762 * Build a new vdev tree from the trusted config
3764 error
= spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
);
3766 nvlist_free(mos_config
);
3767 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3768 spa_load_failed(spa
, "spa_config_parse failed [error=%d]",
3770 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3774 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3775 * obtained by scanning /dev/dsk, then it will have the right vdev
3776 * paths. We update the trusted MOS config with this information.
3777 * We first try to copy the paths with vdev_copy_path_strict, which
3778 * succeeds only when both configs have exactly the same vdev tree.
3779 * If that fails, we fall back to a more flexible method that has a
3780 * best effort policy.
3782 copy_error
= vdev_copy_path_strict(rvd
, mrvd
);
3783 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3784 spa_load_note(spa
, "provided vdev tree:");
3785 vdev_dbgmsg_print_tree(rvd
, 2);
3786 spa_load_note(spa
, "MOS vdev tree:");
3787 vdev_dbgmsg_print_tree(mrvd
, 2);
3789 if (copy_error
!= 0) {
3790 spa_load_note(spa
, "vdev_copy_path_strict failed, falling "
3791 "back to vdev_copy_path_relaxed");
3792 vdev_copy_path_relaxed(rvd
, mrvd
);
3797 spa
->spa_root_vdev
= mrvd
;
3799 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3802 * We will use spa_config if we decide to reload the spa or if spa_load
3803 * fails and we rewind. We must thus regenerate the config using the
3804 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3805 * pass settings on how to load the pool and is not stored in the MOS.
3806 * We copy it over to our new, trusted config.
3808 mos_config_txg
= fnvlist_lookup_uint64(mos_config
,
3809 ZPOOL_CONFIG_POOL_TXG
);
3810 nvlist_free(mos_config
);
3811 mos_config
= spa_config_generate(spa
, NULL
, mos_config_txg
, B_FALSE
);
3812 if (nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_LOAD_POLICY
,
3814 fnvlist_add_nvlist(mos_config
, ZPOOL_LOAD_POLICY
, policy
);
3815 spa_config_set(spa
, mos_config
);
3816 spa
->spa_config_source
= SPA_CONFIG_SRC_MOS
;
3819 * Now that we got the config from the MOS, we should be more strict
3820 * in checking blkptrs and can make assumptions about the consistency
3821 * of the vdev tree. spa_trust_config must be set to true before opening
3822 * vdevs in order for them to be writeable.
3824 spa
->spa_trust_config
= B_TRUE
;
3827 * Open and validate the new vdev tree
3829 error
= spa_ld_open_vdevs(spa
);
3833 error
= spa_ld_validate_vdevs(spa
);
3837 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3838 spa_load_note(spa
, "final vdev tree:");
3839 vdev_dbgmsg_print_tree(rvd
, 2);
3842 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
&&
3843 !spa
->spa_extreme_rewind
&& zfs_max_missing_tvds
== 0) {
3845 * Sanity check to make sure that we are indeed loading the
3846 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3847 * in the config provided and they happened to be the only ones
3848 * to have the latest uberblock, we could involuntarily perform
3849 * an extreme rewind.
3851 healthy_tvds_mos
= spa_healthy_core_tvds(spa
);
3852 if (healthy_tvds_mos
- healthy_tvds
>=
3853 SPA_SYNC_MIN_VDEVS
) {
3854 spa_load_note(spa
, "config provided misses too many "
3855 "top-level vdevs compared to MOS (%lld vs %lld). ",
3856 (u_longlong_t
)healthy_tvds
,
3857 (u_longlong_t
)healthy_tvds_mos
);
3858 spa_load_note(spa
, "vdev tree:");
3859 vdev_dbgmsg_print_tree(rvd
, 2);
3861 spa_load_failed(spa
, "config was already "
3862 "provided from MOS. Aborting.");
3863 return (spa_vdev_err(rvd
,
3864 VDEV_AUX_CORRUPT_DATA
, EIO
));
3866 spa_load_note(spa
, "spa must be reloaded using MOS "
3868 return (SET_ERROR(EAGAIN
));
3872 error
= spa_check_for_missing_logs(spa
);
3874 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
3876 if (rvd
->vdev_guid_sum
!= spa
->spa_uberblock
.ub_guid_sum
) {
3877 spa_load_failed(spa
, "uberblock guid sum doesn't match MOS "
3878 "guid sum (%llu != %llu)",
3879 (u_longlong_t
)spa
->spa_uberblock
.ub_guid_sum
,
3880 (u_longlong_t
)rvd
->vdev_guid_sum
);
3881 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3889 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
3892 vdev_t
*rvd
= spa
->spa_root_vdev
;
3895 * Everything that we read before spa_remove_init() must be stored
3896 * on concreted vdevs. Therefore we do this as early as possible.
3898 error
= spa_remove_init(spa
);
3900 spa_load_failed(spa
, "spa_remove_init failed [error=%d]",
3902 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3906 * Retrieve information needed to condense indirect vdev mappings.
3908 error
= spa_condense_init(spa
);
3910 spa_load_failed(spa
, "spa_condense_init failed [error=%d]",
3912 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3919 spa_ld_check_features(spa_t
*spa
, boolean_t
*missing_feat_writep
)
3922 vdev_t
*rvd
= spa
->spa_root_vdev
;
3924 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
3925 boolean_t missing_feat_read
= B_FALSE
;
3926 nvlist_t
*unsup_feat
, *enabled_feat
;
3928 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
3929 &spa
->spa_feat_for_read_obj
, B_TRUE
) != 0) {
3930 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3933 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
3934 &spa
->spa_feat_for_write_obj
, B_TRUE
) != 0) {
3935 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3938 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
3939 &spa
->spa_feat_desc_obj
, B_TRUE
) != 0) {
3940 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3943 enabled_feat
= fnvlist_alloc();
3944 unsup_feat
= fnvlist_alloc();
3946 if (!spa_features_check(spa
, B_FALSE
,
3947 unsup_feat
, enabled_feat
))
3948 missing_feat_read
= B_TRUE
;
3950 if (spa_writeable(spa
) ||
3951 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) {
3952 if (!spa_features_check(spa
, B_TRUE
,
3953 unsup_feat
, enabled_feat
)) {
3954 *missing_feat_writep
= B_TRUE
;
3958 fnvlist_add_nvlist(spa
->spa_load_info
,
3959 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
3961 if (!nvlist_empty(unsup_feat
)) {
3962 fnvlist_add_nvlist(spa
->spa_load_info
,
3963 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
3966 fnvlist_free(enabled_feat
);
3967 fnvlist_free(unsup_feat
);
3969 if (!missing_feat_read
) {
3970 fnvlist_add_boolean(spa
->spa_load_info
,
3971 ZPOOL_CONFIG_CAN_RDONLY
);
3975 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3976 * twofold: to determine whether the pool is available for
3977 * import in read-write mode and (if it is not) whether the
3978 * pool is available for import in read-only mode. If the pool
3979 * is available for import in read-write mode, it is displayed
3980 * as available in userland; if it is not available for import
3981 * in read-only mode, it is displayed as unavailable in
3982 * userland. If the pool is available for import in read-only
3983 * mode but not read-write mode, it is displayed as unavailable
3984 * in userland with a special note that the pool is actually
3985 * available for open in read-only mode.
3987 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3988 * missing a feature for write, we must first determine whether
3989 * the pool can be opened read-only before returning to
3990 * userland in order to know whether to display the
3991 * abovementioned note.
3993 if (missing_feat_read
|| (*missing_feat_writep
&&
3994 spa_writeable(spa
))) {
3995 spa_load_failed(spa
, "pool uses unsupported features");
3996 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
4001 * Load refcounts for ZFS features from disk into an in-memory
4002 * cache during SPA initialization.
4004 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
4007 error
= feature_get_refcount_from_disk(spa
,
4008 &spa_feature_table
[i
], &refcount
);
4010 spa
->spa_feat_refcount_cache
[i
] = refcount
;
4011 } else if (error
== ENOTSUP
) {
4012 spa
->spa_feat_refcount_cache
[i
] =
4013 SPA_FEATURE_DISABLED
;
4015 spa_load_failed(spa
, "error getting refcount "
4016 "for feature %s [error=%d]",
4017 spa_feature_table
[i
].fi_guid
, error
);
4018 return (spa_vdev_err(rvd
,
4019 VDEV_AUX_CORRUPT_DATA
, EIO
));
4024 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
4025 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
4026 &spa
->spa_feat_enabled_txg_obj
, B_TRUE
) != 0)
4027 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4031 * Encryption was added before bookmark_v2, even though bookmark_v2
4032 * is now a dependency. If this pool has encryption enabled without
4033 * bookmark_v2, trigger an errata message.
4035 if (spa_feature_is_enabled(spa
, SPA_FEATURE_ENCRYPTION
) &&
4036 !spa_feature_is_enabled(spa
, SPA_FEATURE_BOOKMARK_V2
)) {
4037 spa
->spa_errata
= ZPOOL_ERRATA_ZOL_8308_ENCRYPTION
;
4044 spa_ld_load_special_directories(spa_t
*spa
)
4047 vdev_t
*rvd
= spa
->spa_root_vdev
;
4049 spa
->spa_is_initializing
= B_TRUE
;
4050 error
= dsl_pool_open(spa
->spa_dsl_pool
);
4051 spa
->spa_is_initializing
= B_FALSE
;
4053 spa_load_failed(spa
, "dsl_pool_open failed [error=%d]", error
);
4054 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4061 spa_ld_get_props(spa_t
*spa
)
4065 vdev_t
*rvd
= spa
->spa_root_vdev
;
4067 /* Grab the checksum salt from the MOS. */
4068 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4069 DMU_POOL_CHECKSUM_SALT
, 1,
4070 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
4071 spa
->spa_cksum_salt
.zcs_bytes
);
4072 if (error
== ENOENT
) {
4073 /* Generate a new salt for subsequent use */
4074 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4075 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4076 } else if (error
!= 0) {
4077 spa_load_failed(spa
, "unable to retrieve checksum salt from "
4078 "MOS [error=%d]", error
);
4079 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4082 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
, B_TRUE
) != 0)
4083 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4084 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
4086 spa_load_failed(spa
, "error opening deferred-frees bpobj "
4087 "[error=%d]", error
);
4088 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4092 * Load the bit that tells us to use the new accounting function
4093 * (raid-z deflation). If we have an older pool, this will not
4096 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
, B_FALSE
);
4097 if (error
!= 0 && error
!= ENOENT
)
4098 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4100 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
4101 &spa
->spa_creation_version
, B_FALSE
);
4102 if (error
!= 0 && error
!= ENOENT
)
4103 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4106 * Load the persistent error log. If we have an older pool, this will
4109 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
,
4111 if (error
!= 0 && error
!= ENOENT
)
4112 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4114 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
4115 &spa
->spa_errlog_scrub
, B_FALSE
);
4116 if (error
!= 0 && error
!= ENOENT
)
4117 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4120 * Load the livelist deletion field. If a livelist is queued for
4121 * deletion, indicate that in the spa
4123 error
= spa_dir_prop(spa
, DMU_POOL_DELETED_CLONES
,
4124 &spa
->spa_livelists_to_delete
, B_FALSE
);
4125 if (error
!= 0 && error
!= ENOENT
)
4126 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4129 * Load the history object. If we have an older pool, this
4130 * will not be present.
4132 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
, B_FALSE
);
4133 if (error
!= 0 && error
!= ENOENT
)
4134 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4137 * Load the per-vdev ZAP map. If we have an older pool, this will not
4138 * be present; in this case, defer its creation to a later time to
4139 * avoid dirtying the MOS this early / out of sync context. See
4140 * spa_sync_config_object.
4143 /* The sentinel is only available in the MOS config. */
4144 nvlist_t
*mos_config
;
4145 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0) {
4146 spa_load_failed(spa
, "unable to retrieve MOS config");
4147 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4150 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
4151 &spa
->spa_all_vdev_zaps
, B_FALSE
);
4153 if (error
== ENOENT
) {
4154 VERIFY(!nvlist_exists(mos_config
,
4155 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
4156 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
4157 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4158 } else if (error
!= 0) {
4159 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4160 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
4162 * An older version of ZFS overwrote the sentinel value, so
4163 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4164 * destruction to later; see spa_sync_config_object.
4166 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
4168 * We're assuming that no vdevs have had their ZAPs created
4169 * before this. Better be sure of it.
4171 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4173 nvlist_free(mos_config
);
4175 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4177 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
,
4179 if (error
&& error
!= ENOENT
)
4180 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4183 uint64_t autoreplace
= 0;
4185 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
4186 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
4187 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
4188 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
4189 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
4190 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
4191 spa_prop_find(spa
, ZPOOL_PROP_AUTOTRIM
, &spa
->spa_autotrim
);
4192 spa
->spa_autoreplace
= (autoreplace
!= 0);
4196 * If we are importing a pool with missing top-level vdevs,
4197 * we enforce that the pool doesn't panic or get suspended on
4198 * error since the likelihood of missing data is extremely high.
4200 if (spa
->spa_missing_tvds
> 0 &&
4201 spa
->spa_failmode
!= ZIO_FAILURE_MODE_CONTINUE
&&
4202 spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4203 spa_load_note(spa
, "forcing failmode to 'continue' "
4204 "as some top level vdevs are missing");
4205 spa
->spa_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
4212 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
4215 vdev_t
*rvd
= spa
->spa_root_vdev
;
4218 * If we're assembling the pool from the split-off vdevs of
4219 * an existing pool, we don't want to attach the spares & cache
4224 * Load any hot spares for this pool.
4226 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
,
4228 if (error
!= 0 && error
!= ENOENT
)
4229 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4230 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4231 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
4232 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
4233 &spa
->spa_spares
.sav_config
) != 0) {
4234 spa_load_failed(spa
, "error loading spares nvlist");
4235 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4238 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4239 spa_load_spares(spa
);
4240 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4241 } else if (error
== 0) {
4242 spa
->spa_spares
.sav_sync
= B_TRUE
;
4246 * Load any level 2 ARC devices for this pool.
4248 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
4249 &spa
->spa_l2cache
.sav_object
, B_FALSE
);
4250 if (error
!= 0 && error
!= ENOENT
)
4251 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4252 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4253 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
4254 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
4255 &spa
->spa_l2cache
.sav_config
) != 0) {
4256 spa_load_failed(spa
, "error loading l2cache nvlist");
4257 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4260 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4261 spa_load_l2cache(spa
);
4262 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4263 } else if (error
== 0) {
4264 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4271 spa_ld_load_vdev_metadata(spa_t
*spa
)
4274 vdev_t
*rvd
= spa
->spa_root_vdev
;
4277 * If the 'multihost' property is set, then never allow a pool to
4278 * be imported when the system hostid is zero. The exception to
4279 * this rule is zdb which is always allowed to access pools.
4281 if (spa_multihost(spa
) && spa_get_hostid(spa
) == 0 &&
4282 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
4283 fnvlist_add_uint64(spa
->spa_load_info
,
4284 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
4285 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
4289 * If the 'autoreplace' property is set, then post a resource notifying
4290 * the ZFS DE that it should not issue any faults for unopenable
4291 * devices. We also iterate over the vdevs, and post a sysevent for any
4292 * unopenable vdevs so that the normal autoreplace handler can take
4295 if (spa
->spa_autoreplace
&& spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4296 spa_check_removed(spa
->spa_root_vdev
);
4298 * For the import case, this is done in spa_import(), because
4299 * at this point we're using the spare definitions from
4300 * the MOS config, not necessarily from the userland config.
4302 if (spa
->spa_load_state
!= SPA_LOAD_IMPORT
) {
4303 spa_aux_check_removed(&spa
->spa_spares
);
4304 spa_aux_check_removed(&spa
->spa_l2cache
);
4309 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4311 error
= vdev_load(rvd
);
4313 spa_load_failed(spa
, "vdev_load failed [error=%d]", error
);
4314 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4317 error
= spa_ld_log_spacemaps(spa
);
4319 spa_load_failed(spa
, "spa_ld_log_sm_data failed [error=%d]",
4321 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4325 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4327 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4328 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
, B_FALSE
);
4329 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4335 spa_ld_load_dedup_tables(spa_t
*spa
)
4338 vdev_t
*rvd
= spa
->spa_root_vdev
;
4340 error
= ddt_load(spa
);
4342 spa_load_failed(spa
, "ddt_load failed [error=%d]", error
);
4343 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4350 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
4352 vdev_t
*rvd
= spa
->spa_root_vdev
;
4354 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
)) {
4355 boolean_t missing
= spa_check_logs(spa
);
4357 if (spa
->spa_missing_tvds
!= 0) {
4358 spa_load_note(spa
, "spa_check_logs failed "
4359 "so dropping the logs");
4361 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
4362 spa_load_failed(spa
, "spa_check_logs failed");
4363 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
,
4373 spa_ld_verify_pool_data(spa_t
*spa
)
4376 vdev_t
*rvd
= spa
->spa_root_vdev
;
4379 * We've successfully opened the pool, verify that we're ready
4380 * to start pushing transactions.
4382 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4383 error
= spa_load_verify(spa
);
4385 spa_load_failed(spa
, "spa_load_verify failed "
4386 "[error=%d]", error
);
4387 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
4396 spa_ld_claim_log_blocks(spa_t
*spa
)
4399 dsl_pool_t
*dp
= spa_get_dsl(spa
);
4402 * Claim log blocks that haven't been committed yet.
4403 * This must all happen in a single txg.
4404 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4405 * invoked from zil_claim_log_block()'s i/o done callback.
4406 * Price of rollback is that we abandon the log.
4408 spa
->spa_claiming
= B_TRUE
;
4410 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
4411 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
4412 zil_claim
, tx
, DS_FIND_CHILDREN
);
4415 spa
->spa_claiming
= B_FALSE
;
4417 spa_set_log_state(spa
, SPA_LOG_GOOD
);
4421 spa_ld_check_for_config_update(spa_t
*spa
, uint64_t config_cache_txg
,
4422 boolean_t update_config_cache
)
4424 vdev_t
*rvd
= spa
->spa_root_vdev
;
4425 int need_update
= B_FALSE
;
4428 * If the config cache is stale, or we have uninitialized
4429 * metaslabs (see spa_vdev_add()), then update the config.
4431 * If this is a verbatim import, trust the current
4432 * in-core spa_config and update the disk labels.
4434 if (update_config_cache
|| config_cache_txg
!= spa
->spa_config_txg
||
4435 spa
->spa_load_state
== SPA_LOAD_IMPORT
||
4436 spa
->spa_load_state
== SPA_LOAD_RECOVER
||
4437 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
4438 need_update
= B_TRUE
;
4440 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
4441 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
4442 need_update
= B_TRUE
;
4445 * Update the config cache asynchronously in case we're the
4446 * root pool, in which case the config cache isn't writable yet.
4449 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4453 spa_ld_prepare_for_reload(spa_t
*spa
)
4455 spa_mode_t mode
= spa
->spa_mode
;
4456 int async_suspended
= spa
->spa_async_suspended
;
4459 spa_deactivate(spa
);
4460 spa_activate(spa
, mode
);
4463 * We save the value of spa_async_suspended as it gets reset to 0 by
4464 * spa_unload(). We want to restore it back to the original value before
4465 * returning as we might be calling spa_async_resume() later.
4467 spa
->spa_async_suspended
= async_suspended
;
4471 spa_ld_read_checkpoint_txg(spa_t
*spa
)
4473 uberblock_t checkpoint
;
4476 ASSERT0(spa
->spa_checkpoint_txg
);
4477 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4479 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4480 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4481 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4483 if (error
== ENOENT
)
4489 ASSERT3U(checkpoint
.ub_txg
, !=, 0);
4490 ASSERT3U(checkpoint
.ub_checkpoint_txg
, !=, 0);
4491 ASSERT3U(checkpoint
.ub_timestamp
, !=, 0);
4492 spa
->spa_checkpoint_txg
= checkpoint
.ub_txg
;
4493 spa
->spa_checkpoint_info
.sci_timestamp
= checkpoint
.ub_timestamp
;
4499 spa_ld_mos_init(spa_t
*spa
, spa_import_type_t type
)
4503 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4504 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4507 * Never trust the config that is provided unless we are assembling
4508 * a pool following a split.
4509 * This means don't trust blkptrs and the vdev tree in general. This
4510 * also effectively puts the spa in read-only mode since
4511 * spa_writeable() checks for spa_trust_config to be true.
4512 * We will later load a trusted config from the MOS.
4514 if (type
!= SPA_IMPORT_ASSEMBLE
)
4515 spa
->spa_trust_config
= B_FALSE
;
4518 * Parse the config provided to create a vdev tree.
4520 error
= spa_ld_parse_config(spa
, type
);
4524 spa_import_progress_add(spa
);
4527 * Now that we have the vdev tree, try to open each vdev. This involves
4528 * opening the underlying physical device, retrieving its geometry and
4529 * probing the vdev with a dummy I/O. The state of each vdev will be set
4530 * based on the success of those operations. After this we'll be ready
4531 * to read from the vdevs.
4533 error
= spa_ld_open_vdevs(spa
);
4538 * Read the label of each vdev and make sure that the GUIDs stored
4539 * there match the GUIDs in the config provided.
4540 * If we're assembling a new pool that's been split off from an
4541 * existing pool, the labels haven't yet been updated so we skip
4542 * validation for now.
4544 if (type
!= SPA_IMPORT_ASSEMBLE
) {
4545 error
= spa_ld_validate_vdevs(spa
);
4551 * Read all vdev labels to find the best uberblock (i.e. latest,
4552 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
4553 * get the list of features required to read blkptrs in the MOS from
4554 * the vdev label with the best uberblock and verify that our version
4555 * of zfs supports them all.
4557 error
= spa_ld_select_uberblock(spa
, type
);
4562 * Pass that uberblock to the dsl_pool layer which will open the root
4563 * blkptr. This blkptr points to the latest version of the MOS and will
4564 * allow us to read its contents.
4566 error
= spa_ld_open_rootbp(spa
);
4574 spa_ld_checkpoint_rewind(spa_t
*spa
)
4576 uberblock_t checkpoint
;
4579 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4580 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4582 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4583 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4584 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4587 spa_load_failed(spa
, "unable to retrieve checkpointed "
4588 "uberblock from the MOS config [error=%d]", error
);
4590 if (error
== ENOENT
)
4591 error
= ZFS_ERR_NO_CHECKPOINT
;
4596 ASSERT3U(checkpoint
.ub_txg
, <, spa
->spa_uberblock
.ub_txg
);
4597 ASSERT3U(checkpoint
.ub_txg
, ==, checkpoint
.ub_checkpoint_txg
);
4600 * We need to update the txg and timestamp of the checkpointed
4601 * uberblock to be higher than the latest one. This ensures that
4602 * the checkpointed uberblock is selected if we were to close and
4603 * reopen the pool right after we've written it in the vdev labels.
4604 * (also see block comment in vdev_uberblock_compare)
4606 checkpoint
.ub_txg
= spa
->spa_uberblock
.ub_txg
+ 1;
4607 checkpoint
.ub_timestamp
= gethrestime_sec();
4610 * Set current uberblock to be the checkpointed uberblock.
4612 spa
->spa_uberblock
= checkpoint
;
4615 * If we are doing a normal rewind, then the pool is open for
4616 * writing and we sync the "updated" checkpointed uberblock to
4617 * disk. Once this is done, we've basically rewound the whole
4618 * pool and there is no way back.
4620 * There are cases when we don't want to attempt and sync the
4621 * checkpointed uberblock to disk because we are opening a
4622 * pool as read-only. Specifically, verifying the checkpointed
4623 * state with zdb, and importing the checkpointed state to get
4624 * a "preview" of its content.
4626 if (spa_writeable(spa
)) {
4627 vdev_t
*rvd
= spa
->spa_root_vdev
;
4629 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4630 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
4632 int children
= rvd
->vdev_children
;
4633 int c0
= random_in_range(children
);
4635 for (int c
= 0; c
< children
; c
++) {
4636 vdev_t
*vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
4638 /* Stop when revisiting the first vdev */
4639 if (c
> 0 && svd
[0] == vd
)
4642 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
4643 !vdev_is_concrete(vd
))
4646 svd
[svdcount
++] = vd
;
4647 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
4650 error
= vdev_config_sync(svd
, svdcount
, spa
->spa_first_txg
);
4652 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
4653 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4656 spa_load_failed(spa
, "failed to write checkpointed "
4657 "uberblock to the vdev labels [error=%d]", error
);
4666 spa_ld_mos_with_trusted_config(spa_t
*spa
, spa_import_type_t type
,
4667 boolean_t
*update_config_cache
)
4672 * Parse the config for pool, open and validate vdevs,
4673 * select an uberblock, and use that uberblock to open
4676 error
= spa_ld_mos_init(spa
, type
);
4681 * Retrieve the trusted config stored in the MOS and use it to create
4682 * a new, exact version of the vdev tree, then reopen all vdevs.
4684 error
= spa_ld_trusted_config(spa
, type
, B_FALSE
);
4685 if (error
== EAGAIN
) {
4686 if (update_config_cache
!= NULL
)
4687 *update_config_cache
= B_TRUE
;
4690 * Redo the loading process with the trusted config if it is
4691 * too different from the untrusted config.
4693 spa_ld_prepare_for_reload(spa
);
4694 spa_load_note(spa
, "RELOADING");
4695 error
= spa_ld_mos_init(spa
, type
);
4699 error
= spa_ld_trusted_config(spa
, type
, B_TRUE
);
4703 } else if (error
!= 0) {
4711 * Load an existing storage pool, using the config provided. This config
4712 * describes which vdevs are part of the pool and is later validated against
4713 * partial configs present in each vdev's label and an entire copy of the
4714 * config stored in the MOS.
4717 spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
4720 boolean_t missing_feat_write
= B_FALSE
;
4721 boolean_t checkpoint_rewind
=
4722 (spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4723 boolean_t update_config_cache
= B_FALSE
;
4725 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4726 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4728 spa_load_note(spa
, "LOADING");
4730 error
= spa_ld_mos_with_trusted_config(spa
, type
, &update_config_cache
);
4735 * If we are rewinding to the checkpoint then we need to repeat
4736 * everything we've done so far in this function but this time
4737 * selecting the checkpointed uberblock and using that to open
4740 if (checkpoint_rewind
) {
4742 * If we are rewinding to the checkpoint update config cache
4745 update_config_cache
= B_TRUE
;
4748 * Extract the checkpointed uberblock from the current MOS
4749 * and use this as the pool's uberblock from now on. If the
4750 * pool is imported as writeable we also write the checkpoint
4751 * uberblock to the labels, making the rewind permanent.
4753 error
= spa_ld_checkpoint_rewind(spa
);
4758 * Redo the loading process again with the
4759 * checkpointed uberblock.
4761 spa_ld_prepare_for_reload(spa
);
4762 spa_load_note(spa
, "LOADING checkpointed uberblock");
4763 error
= spa_ld_mos_with_trusted_config(spa
, type
, NULL
);
4769 * Retrieve the checkpoint txg if the pool has a checkpoint.
4771 error
= spa_ld_read_checkpoint_txg(spa
);
4776 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
4777 * from the pool and their contents were re-mapped to other vdevs. Note
4778 * that everything that we read before this step must have been
4779 * rewritten on concrete vdevs after the last device removal was
4780 * initiated. Otherwise we could be reading from indirect vdevs before
4781 * we have loaded their mappings.
4783 error
= spa_ld_open_indirect_vdev_metadata(spa
);
4788 * Retrieve the full list of active features from the MOS and check if
4789 * they are all supported.
4791 error
= spa_ld_check_features(spa
, &missing_feat_write
);
4796 * Load several special directories from the MOS needed by the dsl_pool
4799 error
= spa_ld_load_special_directories(spa
);
4804 * Retrieve pool properties from the MOS.
4806 error
= spa_ld_get_props(spa
);
4811 * Retrieve the list of auxiliary devices - cache devices and spares -
4814 error
= spa_ld_open_aux_vdevs(spa
, type
);
4819 * Load the metadata for all vdevs. Also check if unopenable devices
4820 * should be autoreplaced.
4822 error
= spa_ld_load_vdev_metadata(spa
);
4826 error
= spa_ld_load_dedup_tables(spa
);
4831 * Verify the logs now to make sure we don't have any unexpected errors
4832 * when we claim log blocks later.
4834 error
= spa_ld_verify_logs(spa
, type
, ereport
);
4838 if (missing_feat_write
) {
4839 ASSERT(spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
);
4842 * At this point, we know that we can open the pool in
4843 * read-only mode but not read-write mode. We now have enough
4844 * information and can return to userland.
4846 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
4851 * Traverse the last txgs to make sure the pool was left off in a safe
4852 * state. When performing an extreme rewind, we verify the whole pool,
4853 * which can take a very long time.
4855 error
= spa_ld_verify_pool_data(spa
);
4860 * Calculate the deflated space for the pool. This must be done before
4861 * we write anything to the pool because we'd need to update the space
4862 * accounting using the deflated sizes.
4864 spa_update_dspace(spa
);
4867 * We have now retrieved all the information we needed to open the
4868 * pool. If we are importing the pool in read-write mode, a few
4869 * additional steps must be performed to finish the import.
4871 if (spa_writeable(spa
) && (spa
->spa_load_state
== SPA_LOAD_RECOVER
||
4872 spa
->spa_load_max_txg
== UINT64_MAX
)) {
4873 uint64_t config_cache_txg
= spa
->spa_config_txg
;
4875 ASSERT(spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
);
4878 * In case of a checkpoint rewind, log the original txg
4879 * of the checkpointed uberblock.
4881 if (checkpoint_rewind
) {
4882 spa_history_log_internal(spa
, "checkpoint rewind",
4883 NULL
, "rewound state to txg=%llu",
4884 (u_longlong_t
)spa
->spa_uberblock
.ub_checkpoint_txg
);
4888 * Traverse the ZIL and claim all blocks.
4890 spa_ld_claim_log_blocks(spa
);
4893 * Kick-off the syncing thread.
4895 spa
->spa_sync_on
= B_TRUE
;
4896 txg_sync_start(spa
->spa_dsl_pool
);
4897 mmp_thread_start(spa
);
4900 * Wait for all claims to sync. We sync up to the highest
4901 * claimed log block birth time so that claimed log blocks
4902 * don't appear to be from the future. spa_claim_max_txg
4903 * will have been set for us by ZIL traversal operations
4906 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
4909 * Check if we need to request an update of the config. On the
4910 * next sync, we would update the config stored in vdev labels
4911 * and the cachefile (by default /etc/zfs/zpool.cache).
4913 spa_ld_check_for_config_update(spa
, config_cache_txg
,
4914 update_config_cache
);
4917 * Check if a rebuild was in progress and if so resume it.
4918 * Then check all DTLs to see if anything needs resilvering.
4919 * The resilver will be deferred if a rebuild was started.
4921 if (vdev_rebuild_active(spa
->spa_root_vdev
)) {
4922 vdev_rebuild_restart(spa
);
4923 } else if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4924 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
4925 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
4929 * Log the fact that we booted up (so that we can detect if
4930 * we rebooted in the middle of an operation).
4932 spa_history_log_version(spa
, "open", NULL
);
4934 spa_restart_removal(spa
);
4935 spa_spawn_aux_threads(spa
);
4938 * Delete any inconsistent datasets.
4941 * Since we may be issuing deletes for clones here,
4942 * we make sure to do so after we've spawned all the
4943 * auxiliary threads above (from which the livelist
4944 * deletion zthr is part of).
4946 (void) dmu_objset_find(spa_name(spa
),
4947 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
4950 * Clean up any stale temporary dataset userrefs.
4952 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
4954 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4955 vdev_initialize_restart(spa
->spa_root_vdev
);
4956 vdev_trim_restart(spa
->spa_root_vdev
);
4957 vdev_autotrim_restart(spa
);
4958 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4961 spa_import_progress_remove(spa_guid(spa
));
4962 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
4964 spa_load_note(spa
, "LOADED");
4970 spa_load_retry(spa_t
*spa
, spa_load_state_t state
)
4972 spa_mode_t mode
= spa
->spa_mode
;
4975 spa_deactivate(spa
);
4977 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
4979 spa_activate(spa
, mode
);
4980 spa_async_suspend(spa
);
4982 spa_load_note(spa
, "spa_load_retry: rewind, max txg: %llu",
4983 (u_longlong_t
)spa
->spa_load_max_txg
);
4985 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
));
4989 * If spa_load() fails this function will try loading prior txg's. If
4990 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
4991 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
4992 * function will not rewind the pool and will return the same error as
4996 spa_load_best(spa_t
*spa
, spa_load_state_t state
, uint64_t max_request
,
4999 nvlist_t
*loadinfo
= NULL
;
5000 nvlist_t
*config
= NULL
;
5001 int load_error
, rewind_error
;
5002 uint64_t safe_rewind_txg
;
5005 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
5006 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
5007 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5009 spa
->spa_load_max_txg
= max_request
;
5010 if (max_request
!= UINT64_MAX
)
5011 spa
->spa_extreme_rewind
= B_TRUE
;
5014 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
);
5015 if (load_error
== 0)
5017 if (load_error
== ZFS_ERR_NO_CHECKPOINT
) {
5019 * When attempting checkpoint-rewind on a pool with no
5020 * checkpoint, we should not attempt to load uberblocks
5021 * from previous txgs when spa_load fails.
5023 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
5024 spa_import_progress_remove(spa_guid(spa
));
5025 return (load_error
);
5028 if (spa
->spa_root_vdev
!= NULL
)
5029 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5031 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
5032 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
5034 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
5035 nvlist_free(config
);
5036 spa_import_progress_remove(spa_guid(spa
));
5037 return (load_error
);
5040 if (state
== SPA_LOAD_RECOVER
) {
5041 /* Price of rolling back is discarding txgs, including log */
5042 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5045 * If we aren't rolling back save the load info from our first
5046 * import attempt so that we can restore it after attempting
5049 loadinfo
= spa
->spa_load_info
;
5050 spa
->spa_load_info
= fnvlist_alloc();
5053 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
5054 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
5055 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
5056 TXG_INITIAL
: safe_rewind_txg
;
5059 * Continue as long as we're finding errors, we're still within
5060 * the acceptable rewind range, and we're still finding uberblocks
5062 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
5063 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
5064 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
5065 spa
->spa_extreme_rewind
= B_TRUE
;
5066 rewind_error
= spa_load_retry(spa
, state
);
5069 spa
->spa_extreme_rewind
= B_FALSE
;
5070 spa
->spa_load_max_txg
= UINT64_MAX
;
5072 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
5073 spa_config_set(spa
, config
);
5075 nvlist_free(config
);
5077 if (state
== SPA_LOAD_RECOVER
) {
5078 ASSERT3P(loadinfo
, ==, NULL
);
5079 spa_import_progress_remove(spa_guid(spa
));
5080 return (rewind_error
);
5082 /* Store the rewind info as part of the initial load info */
5083 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
5084 spa
->spa_load_info
);
5086 /* Restore the initial load info */
5087 fnvlist_free(spa
->spa_load_info
);
5088 spa
->spa_load_info
= loadinfo
;
5090 spa_import_progress_remove(spa_guid(spa
));
5091 return (load_error
);
5098 * The import case is identical to an open except that the configuration is sent
5099 * down from userland, instead of grabbed from the configuration cache. For the
5100 * case of an open, the pool configuration will exist in the
5101 * POOL_STATE_UNINITIALIZED state.
5103 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5104 * the same time open the pool, without having to keep around the spa_t in some
5108 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
5112 spa_load_state_t state
= SPA_LOAD_OPEN
;
5114 int locked
= B_FALSE
;
5115 int firstopen
= B_FALSE
;
5120 * As disgusting as this is, we need to support recursive calls to this
5121 * function because dsl_dir_open() is called during spa_load(), and ends
5122 * up calling spa_open() again. The real fix is to figure out how to
5123 * avoid dsl_dir_open() calling this in the first place.
5125 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
5126 mutex_enter(&spa_namespace_lock
);
5130 if ((spa
= spa_lookup(pool
)) == NULL
) {
5132 mutex_exit(&spa_namespace_lock
);
5133 return (SET_ERROR(ENOENT
));
5136 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
5137 zpool_load_policy_t policy
;
5141 zpool_get_load_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
5143 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
5144 state
= SPA_LOAD_RECOVER
;
5146 spa_activate(spa
, spa_mode_global
);
5148 if (state
!= SPA_LOAD_RECOVER
)
5149 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
5150 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
5152 zfs_dbgmsg("spa_open_common: opening %s", pool
);
5153 error
= spa_load_best(spa
, state
, policy
.zlp_txg
,
5156 if (error
== EBADF
) {
5158 * If vdev_validate() returns failure (indicated by
5159 * EBADF), it indicates that one of the vdevs indicates
5160 * that the pool has been exported or destroyed. If
5161 * this is the case, the config cache is out of sync and
5162 * we should remove the pool from the namespace.
5165 spa_deactivate(spa
);
5166 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
5169 mutex_exit(&spa_namespace_lock
);
5170 return (SET_ERROR(ENOENT
));
5175 * We can't open the pool, but we still have useful
5176 * information: the state of each vdev after the
5177 * attempted vdev_open(). Return this to the user.
5179 if (config
!= NULL
&& spa
->spa_config
) {
5180 *config
= fnvlist_dup(spa
->spa_config
);
5181 fnvlist_add_nvlist(*config
,
5182 ZPOOL_CONFIG_LOAD_INFO
,
5183 spa
->spa_load_info
);
5186 spa_deactivate(spa
);
5187 spa
->spa_last_open_failed
= error
;
5189 mutex_exit(&spa_namespace_lock
);
5195 spa_open_ref(spa
, tag
);
5198 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5201 * If we've recovered the pool, pass back any information we
5202 * gathered while doing the load.
5204 if (state
== SPA_LOAD_RECOVER
) {
5205 fnvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
5206 spa
->spa_load_info
);
5210 spa
->spa_last_open_failed
= 0;
5211 spa
->spa_last_ubsync_txg
= 0;
5212 spa
->spa_load_txg
= 0;
5213 mutex_exit(&spa_namespace_lock
);
5217 zvol_create_minors_recursive(spa_name(spa
));
5225 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
5228 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
5232 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
5234 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
5238 * Lookup the given spa_t, incrementing the inject count in the process,
5239 * preventing it from being exported or destroyed.
5242 spa_inject_addref(char *name
)
5246 mutex_enter(&spa_namespace_lock
);
5247 if ((spa
= spa_lookup(name
)) == NULL
) {
5248 mutex_exit(&spa_namespace_lock
);
5251 spa
->spa_inject_ref
++;
5252 mutex_exit(&spa_namespace_lock
);
5258 spa_inject_delref(spa_t
*spa
)
5260 mutex_enter(&spa_namespace_lock
);
5261 spa
->spa_inject_ref
--;
5262 mutex_exit(&spa_namespace_lock
);
5266 * Add spares device information to the nvlist.
5269 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
5279 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5281 if (spa
->spa_spares
.sav_count
== 0)
5284 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5285 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5286 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
));
5288 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, spares
,
5290 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5291 &spares
, &nspares
));
5294 * Go through and find any spares which have since been
5295 * repurposed as an active spare. If this is the case, update
5296 * their status appropriately.
5298 for (i
= 0; i
< nspares
; i
++) {
5299 guid
= fnvlist_lookup_uint64(spares
[i
],
5301 if (spa_spare_exists(guid
, &pool
, NULL
) &&
5303 VERIFY0(nvlist_lookup_uint64_array(spares
[i
],
5304 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
,
5306 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
5307 vs
->vs_aux
= VDEV_AUX_SPARED
;
5314 * Add l2cache device information to the nvlist, including vdev stats.
5317 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
5320 uint_t i
, j
, nl2cache
;
5327 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5329 if (spa
->spa_l2cache
.sav_count
== 0)
5332 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5333 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5334 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
));
5335 if (nl2cache
!= 0) {
5336 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, l2cache
,
5338 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5339 &l2cache
, &nl2cache
));
5342 * Update level 2 cache device stats.
5345 for (i
= 0; i
< nl2cache
; i
++) {
5346 guid
= fnvlist_lookup_uint64(l2cache
[i
],
5350 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
5352 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
5353 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
5359 VERIFY0(nvlist_lookup_uint64_array(l2cache
[i
],
5360 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
));
5361 vdev_get_stats(vd
, vs
);
5362 vdev_config_generate_stats(vd
, l2cache
[i
]);
5369 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
5374 if (spa
->spa_feat_for_read_obj
!= 0) {
5375 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5376 spa
->spa_feat_for_read_obj
);
5377 zap_cursor_retrieve(&zc
, &za
) == 0;
5378 zap_cursor_advance(&zc
)) {
5379 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5380 za
.za_num_integers
== 1);
5381 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5382 za
.za_first_integer
));
5384 zap_cursor_fini(&zc
);
5387 if (spa
->spa_feat_for_write_obj
!= 0) {
5388 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5389 spa
->spa_feat_for_write_obj
);
5390 zap_cursor_retrieve(&zc
, &za
) == 0;
5391 zap_cursor_advance(&zc
)) {
5392 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5393 za
.za_num_integers
== 1);
5394 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5395 za
.za_first_integer
));
5397 zap_cursor_fini(&zc
);
5402 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
5406 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5407 zfeature_info_t feature
= spa_feature_table
[i
];
5410 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
5413 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
5418 * Store a list of pool features and their reference counts in the
5421 * The first time this is called on a spa, allocate a new nvlist, fetch
5422 * the pool features and reference counts from disk, then save the list
5423 * in the spa. In subsequent calls on the same spa use the saved nvlist
5424 * and refresh its values from the cached reference counts. This
5425 * ensures we don't block here on I/O on a suspended pool so 'zpool
5426 * clear' can resume the pool.
5429 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
5433 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5435 mutex_enter(&spa
->spa_feat_stats_lock
);
5436 features
= spa
->spa_feat_stats
;
5438 if (features
!= NULL
) {
5439 spa_feature_stats_from_cache(spa
, features
);
5441 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
5442 spa
->spa_feat_stats
= features
;
5443 spa_feature_stats_from_disk(spa
, features
);
5446 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
5449 mutex_exit(&spa
->spa_feat_stats_lock
);
5453 spa_get_stats(const char *name
, nvlist_t
**config
,
5454 char *altroot
, size_t buflen
)
5460 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
5464 * This still leaves a window of inconsistency where the spares
5465 * or l2cache devices could change and the config would be
5466 * self-inconsistent.
5468 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5470 if (*config
!= NULL
) {
5471 uint64_t loadtimes
[2];
5473 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
5474 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
5475 fnvlist_add_uint64_array(*config
,
5476 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2);
5478 fnvlist_add_uint64(*config
,
5479 ZPOOL_CONFIG_ERRCOUNT
,
5480 spa_get_errlog_size(spa
));
5482 if (spa_suspended(spa
)) {
5483 fnvlist_add_uint64(*config
,
5484 ZPOOL_CONFIG_SUSPENDED
,
5486 fnvlist_add_uint64(*config
,
5487 ZPOOL_CONFIG_SUSPENDED_REASON
,
5488 spa
->spa_suspended
);
5491 spa_add_spares(spa
, *config
);
5492 spa_add_l2cache(spa
, *config
);
5493 spa_add_feature_stats(spa
, *config
);
5498 * We want to get the alternate root even for faulted pools, so we cheat
5499 * and call spa_lookup() directly.
5503 mutex_enter(&spa_namespace_lock
);
5504 spa
= spa_lookup(name
);
5506 spa_altroot(spa
, altroot
, buflen
);
5510 mutex_exit(&spa_namespace_lock
);
5512 spa_altroot(spa
, altroot
, buflen
);
5517 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5518 spa_close(spa
, FTAG
);
5525 * Validate that the auxiliary device array is well formed. We must have an
5526 * array of nvlists, each which describes a valid leaf vdev. If this is an
5527 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
5528 * specified, as long as they are well-formed.
5531 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
5532 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
5533 vdev_labeltype_t label
)
5540 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5543 * It's acceptable to have no devs specified.
5545 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
5549 return (SET_ERROR(EINVAL
));
5552 * Make sure the pool is formatted with a version that supports this
5555 if (spa_version(spa
) < version
)
5556 return (SET_ERROR(ENOTSUP
));
5559 * Set the pending device list so we correctly handle device in-use
5562 sav
->sav_pending
= dev
;
5563 sav
->sav_npending
= ndev
;
5565 for (i
= 0; i
< ndev
; i
++) {
5566 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
5570 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5572 error
= SET_ERROR(EINVAL
);
5578 if ((error
= vdev_open(vd
)) == 0 &&
5579 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
5580 fnvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
5587 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
5594 sav
->sav_pending
= NULL
;
5595 sav
->sav_npending
= 0;
5600 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
5604 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5606 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5607 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
5608 VDEV_LABEL_SPARE
)) != 0) {
5612 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5613 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
5614 VDEV_LABEL_L2CACHE
));
5618 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
5623 if (sav
->sav_config
!= NULL
) {
5629 * Generate new dev list by concatenating with the
5632 VERIFY0(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
5633 &olddevs
, &oldndevs
));
5635 newdevs
= kmem_alloc(sizeof (void *) *
5636 (ndevs
+ oldndevs
), KM_SLEEP
);
5637 for (i
= 0; i
< oldndevs
; i
++)
5638 newdevs
[i
] = fnvlist_dup(olddevs
[i
]);
5639 for (i
= 0; i
< ndevs
; i
++)
5640 newdevs
[i
+ oldndevs
] = fnvlist_dup(devs
[i
]);
5642 fnvlist_remove(sav
->sav_config
, config
);
5644 fnvlist_add_nvlist_array(sav
->sav_config
, config
, newdevs
,
5646 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
5647 nvlist_free(newdevs
[i
]);
5648 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
5651 * Generate a new dev list.
5653 sav
->sav_config
= fnvlist_alloc();
5654 fnvlist_add_nvlist_array(sav
->sav_config
, config
, devs
, ndevs
);
5659 * Stop and drop level 2 ARC devices
5662 spa_l2cache_drop(spa_t
*spa
)
5666 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
5668 for (i
= 0; i
< sav
->sav_count
; i
++) {
5671 vd
= sav
->sav_vdevs
[i
];
5674 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
5675 pool
!= 0ULL && l2arc_vdev_present(vd
))
5676 l2arc_remove_vdev(vd
);
5681 * Verify encryption parameters for spa creation. If we are encrypting, we must
5682 * have the encryption feature flag enabled.
5685 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
5686 boolean_t has_encryption
)
5688 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
5689 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
5691 return (SET_ERROR(ENOTSUP
));
5693 return (dmu_objset_create_crypt_check(NULL
, dcp
, NULL
));
5700 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
5701 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
5704 char *altroot
= NULL
;
5709 uint64_t txg
= TXG_INITIAL
;
5710 nvlist_t
**spares
, **l2cache
;
5711 uint_t nspares
, nl2cache
;
5712 uint64_t version
, obj
, ndraid
= 0;
5713 boolean_t has_features
;
5714 boolean_t has_encryption
;
5715 boolean_t has_allocclass
;
5721 if (props
== NULL
||
5722 nvlist_lookup_string(props
, "tname", &poolname
) != 0)
5723 poolname
= (char *)pool
;
5726 * If this pool already exists, return failure.
5728 mutex_enter(&spa_namespace_lock
);
5729 if (spa_lookup(poolname
) != NULL
) {
5730 mutex_exit(&spa_namespace_lock
);
5731 return (SET_ERROR(EEXIST
));
5735 * Allocate a new spa_t structure.
5737 nvl
= fnvlist_alloc();
5738 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
5739 (void) nvlist_lookup_string(props
,
5740 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5741 spa
= spa_add(poolname
, nvl
, altroot
);
5743 spa_activate(spa
, spa_mode_global
);
5745 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
5746 spa_deactivate(spa
);
5748 mutex_exit(&spa_namespace_lock
);
5753 * Temporary pool names should never be written to disk.
5755 if (poolname
!= pool
)
5756 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
5758 has_features
= B_FALSE
;
5759 has_encryption
= B_FALSE
;
5760 has_allocclass
= B_FALSE
;
5761 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
5762 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
5763 if (zpool_prop_feature(nvpair_name(elem
))) {
5764 has_features
= B_TRUE
;
5766 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
5767 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
5768 if (feat
== SPA_FEATURE_ENCRYPTION
)
5769 has_encryption
= B_TRUE
;
5770 if (feat
== SPA_FEATURE_ALLOCATION_CLASSES
)
5771 has_allocclass
= B_TRUE
;
5775 /* verify encryption params, if they were provided */
5777 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
5779 spa_deactivate(spa
);
5781 mutex_exit(&spa_namespace_lock
);
5785 if (!has_allocclass
&& zfs_special_devs(nvroot
, NULL
)) {
5786 spa_deactivate(spa
);
5788 mutex_exit(&spa_namespace_lock
);
5792 if (has_features
|| nvlist_lookup_uint64(props
,
5793 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
5794 version
= SPA_VERSION
;
5796 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
5798 spa
->spa_first_txg
= txg
;
5799 spa
->spa_uberblock
.ub_txg
= txg
- 1;
5800 spa
->spa_uberblock
.ub_version
= version
;
5801 spa
->spa_ubsync
= spa
->spa_uberblock
;
5802 spa
->spa_load_state
= SPA_LOAD_CREATE
;
5803 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
5804 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
5805 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
5806 spa
->spa_indirect_vdevs_loaded
= B_TRUE
;
5809 * Create "The Godfather" zio to hold all async IOs
5811 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
5813 for (int i
= 0; i
< max_ncpus
; i
++) {
5814 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
5815 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
5816 ZIO_FLAG_GODFATHER
);
5820 * Create the root vdev.
5822 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5824 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
5826 ASSERT(error
!= 0 || rvd
!= NULL
);
5827 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
5829 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
5830 error
= SET_ERROR(EINVAL
);
5833 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
5834 (error
= vdev_draid_spare_create(nvroot
, rvd
, &ndraid
, 0)) == 0 &&
5835 (error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) == 0) {
5837 * instantiate the metaslab groups (this will dirty the vdevs)
5838 * we can no longer error exit past this point
5840 for (int c
= 0; error
== 0 && c
< rvd
->vdev_children
; c
++) {
5841 vdev_t
*vd
= rvd
->vdev_child
[c
];
5843 vdev_metaslab_set_size(vd
);
5844 vdev_expand(vd
, txg
);
5848 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5852 spa_deactivate(spa
);
5854 mutex_exit(&spa_namespace_lock
);
5859 * Get the list of spares, if specified.
5861 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5862 &spares
, &nspares
) == 0) {
5863 spa
->spa_spares
.sav_config
= fnvlist_alloc();
5864 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
5865 ZPOOL_CONFIG_SPARES
, spares
, nspares
);
5866 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5867 spa_load_spares(spa
);
5868 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5869 spa
->spa_spares
.sav_sync
= B_TRUE
;
5873 * Get the list of level 2 cache devices, if specified.
5875 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5876 &l2cache
, &nl2cache
) == 0) {
5877 spa
->spa_l2cache
.sav_config
= fnvlist_alloc();
5878 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
5879 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
);
5880 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5881 spa_load_l2cache(spa
);
5882 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5883 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5886 spa
->spa_is_initializing
= B_TRUE
;
5887 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
5888 spa
->spa_is_initializing
= B_FALSE
;
5891 * Create DDTs (dedup tables).
5895 spa_update_dspace(spa
);
5897 tx
= dmu_tx_create_assigned(dp
, txg
);
5900 * Create the pool's history object.
5902 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
5903 spa_history_create_obj(spa
, tx
);
5905 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
5906 spa_history_log_version(spa
, "create", tx
);
5909 * Create the pool config object.
5911 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
5912 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
5913 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
5915 if (zap_add(spa
->spa_meta_objset
,
5916 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
5917 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
5918 cmn_err(CE_PANIC
, "failed to add pool config");
5921 if (zap_add(spa
->spa_meta_objset
,
5922 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
5923 sizeof (uint64_t), 1, &version
, tx
) != 0) {
5924 cmn_err(CE_PANIC
, "failed to add pool version");
5927 /* Newly created pools with the right version are always deflated. */
5928 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
5929 spa
->spa_deflate
= TRUE
;
5930 if (zap_add(spa
->spa_meta_objset
,
5931 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
5932 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
5933 cmn_err(CE_PANIC
, "failed to add deflate");
5938 * Create the deferred-free bpobj. Turn off compression
5939 * because sync-to-convergence takes longer if the blocksize
5942 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
5943 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
5944 ZIO_COMPRESS_OFF
, tx
);
5945 if (zap_add(spa
->spa_meta_objset
,
5946 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
5947 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
5948 cmn_err(CE_PANIC
, "failed to add bpobj");
5950 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
5951 spa
->spa_meta_objset
, obj
));
5954 * Generate some random noise for salted checksums to operate on.
5956 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
5957 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
5960 * Set pool properties.
5962 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
5963 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
5964 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
5965 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
5966 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
5967 spa
->spa_autotrim
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM
);
5969 if (props
!= NULL
) {
5970 spa_configfile_set(spa
, props
, B_FALSE
);
5971 spa_sync_props(props
, tx
);
5974 for (int i
= 0; i
< ndraid
; i
++)
5975 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
5979 spa
->spa_sync_on
= B_TRUE
;
5981 mmp_thread_start(spa
);
5982 txg_wait_synced(dp
, txg
);
5984 spa_spawn_aux_threads(spa
);
5986 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
5989 * Don't count references from objsets that are already closed
5990 * and are making their way through the eviction process.
5992 spa_evicting_os_wait(spa
);
5993 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
5994 spa
->spa_load_state
= SPA_LOAD_NONE
;
5996 mutex_exit(&spa_namespace_lock
);
6002 * Import a non-root pool into the system.
6005 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
6008 char *altroot
= NULL
;
6009 spa_load_state_t state
= SPA_LOAD_IMPORT
;
6010 zpool_load_policy_t policy
;
6011 spa_mode_t mode
= spa_mode_global
;
6012 uint64_t readonly
= B_FALSE
;
6015 nvlist_t
**spares
, **l2cache
;
6016 uint_t nspares
, nl2cache
;
6019 * If a pool with this name exists, return failure.
6021 mutex_enter(&spa_namespace_lock
);
6022 if (spa_lookup(pool
) != NULL
) {
6023 mutex_exit(&spa_namespace_lock
);
6024 return (SET_ERROR(EEXIST
));
6028 * Create and initialize the spa structure.
6030 (void) nvlist_lookup_string(props
,
6031 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
6032 (void) nvlist_lookup_uint64(props
,
6033 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
6035 mode
= SPA_MODE_READ
;
6036 spa
= spa_add(pool
, config
, altroot
);
6037 spa
->spa_import_flags
= flags
;
6040 * Verbatim import - Take a pool and insert it into the namespace
6041 * as if it had been loaded at boot.
6043 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
6045 spa_configfile_set(spa
, props
, B_FALSE
);
6047 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
6048 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6049 zfs_dbgmsg("spa_import: verbatim import of %s", pool
);
6050 mutex_exit(&spa_namespace_lock
);
6054 spa_activate(spa
, mode
);
6057 * Don't start async tasks until we know everything is healthy.
6059 spa_async_suspend(spa
);
6061 zpool_get_load_policy(config
, &policy
);
6062 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
6063 state
= SPA_LOAD_RECOVER
;
6065 spa
->spa_config_source
= SPA_CONFIG_SRC_TRYIMPORT
;
6067 if (state
!= SPA_LOAD_RECOVER
) {
6068 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
6069 zfs_dbgmsg("spa_import: importing %s", pool
);
6071 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
6072 "(RECOVERY MODE)", pool
, (longlong_t
)policy
.zlp_txg
);
6074 error
= spa_load_best(spa
, state
, policy
.zlp_txg
, policy
.zlp_rewind
);
6077 * Propagate anything learned while loading the pool and pass it
6078 * back to caller (i.e. rewind info, missing devices, etc).
6080 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
, spa
->spa_load_info
);
6082 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6084 * Toss any existing sparelist, as it doesn't have any validity
6085 * anymore, and conflicts with spa_has_spare().
6087 if (spa
->spa_spares
.sav_config
) {
6088 nvlist_free(spa
->spa_spares
.sav_config
);
6089 spa
->spa_spares
.sav_config
= NULL
;
6090 spa_load_spares(spa
);
6092 if (spa
->spa_l2cache
.sav_config
) {
6093 nvlist_free(spa
->spa_l2cache
.sav_config
);
6094 spa
->spa_l2cache
.sav_config
= NULL
;
6095 spa_load_l2cache(spa
);
6098 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
6099 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6102 spa_configfile_set(spa
, props
, B_FALSE
);
6104 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
6105 (error
= spa_prop_set(spa
, props
)))) {
6107 spa_deactivate(spa
);
6109 mutex_exit(&spa_namespace_lock
);
6113 spa_async_resume(spa
);
6116 * Override any spares and level 2 cache devices as specified by
6117 * the user, as these may have correct device names/devids, etc.
6119 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
6120 &spares
, &nspares
) == 0) {
6121 if (spa
->spa_spares
.sav_config
)
6122 fnvlist_remove(spa
->spa_spares
.sav_config
,
6123 ZPOOL_CONFIG_SPARES
);
6125 spa
->spa_spares
.sav_config
= fnvlist_alloc();
6126 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
6127 ZPOOL_CONFIG_SPARES
, spares
, nspares
);
6128 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6129 spa_load_spares(spa
);
6130 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6131 spa
->spa_spares
.sav_sync
= B_TRUE
;
6133 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
6134 &l2cache
, &nl2cache
) == 0) {
6135 if (spa
->spa_l2cache
.sav_config
)
6136 fnvlist_remove(spa
->spa_l2cache
.sav_config
,
6137 ZPOOL_CONFIG_L2CACHE
);
6139 spa
->spa_l2cache
.sav_config
= fnvlist_alloc();
6140 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
6141 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
);
6142 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6143 spa_load_l2cache(spa
);
6144 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6145 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6149 * Check for any removed devices.
6151 if (spa
->spa_autoreplace
) {
6152 spa_aux_check_removed(&spa
->spa_spares
);
6153 spa_aux_check_removed(&spa
->spa_l2cache
);
6156 if (spa_writeable(spa
)) {
6158 * Update the config cache to include the newly-imported pool.
6160 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6164 * It's possible that the pool was expanded while it was exported.
6165 * We kick off an async task to handle this for us.
6167 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
6169 spa_history_log_version(spa
, "import", NULL
);
6171 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6173 mutex_exit(&spa_namespace_lock
);
6175 zvol_create_minors_recursive(pool
);
6181 spa_tryimport(nvlist_t
*tryconfig
)
6183 nvlist_t
*config
= NULL
;
6184 char *poolname
, *cachefile
;
6188 zpool_load_policy_t policy
;
6190 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
6193 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
6197 * Create and initialize the spa structure.
6199 mutex_enter(&spa_namespace_lock
);
6200 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
6201 spa_activate(spa
, SPA_MODE_READ
);
6204 * Rewind pool if a max txg was provided.
6206 zpool_get_load_policy(spa
->spa_config
, &policy
);
6207 if (policy
.zlp_txg
!= UINT64_MAX
) {
6208 spa
->spa_load_max_txg
= policy
.zlp_txg
;
6209 spa
->spa_extreme_rewind
= B_TRUE
;
6210 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6211 poolname
, (longlong_t
)policy
.zlp_txg
);
6213 zfs_dbgmsg("spa_tryimport: importing %s", poolname
);
6216 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_CACHEFILE
, &cachefile
)
6218 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile
);
6219 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
6221 spa
->spa_config_source
= SPA_CONFIG_SRC_SCAN
;
6224 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
);
6227 * If 'tryconfig' was at least parsable, return the current config.
6229 if (spa
->spa_root_vdev
!= NULL
) {
6230 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
6231 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, poolname
);
6232 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, state
);
6233 fnvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
6234 spa
->spa_uberblock
.ub_timestamp
);
6235 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
6236 spa
->spa_load_info
);
6237 fnvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
6241 * If the bootfs property exists on this pool then we
6242 * copy it out so that external consumers can tell which
6243 * pools are bootable.
6245 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
6246 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6249 * We have to play games with the name since the
6250 * pool was opened as TRYIMPORT_NAME.
6252 if (dsl_dsobj_to_dsname(spa_name(spa
),
6253 spa
->spa_bootfs
, tmpname
) == 0) {
6257 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6259 cp
= strchr(tmpname
, '/');
6261 (void) strlcpy(dsname
, tmpname
,
6264 (void) snprintf(dsname
, MAXPATHLEN
,
6265 "%s/%s", poolname
, ++cp
);
6267 fnvlist_add_string(config
, ZPOOL_CONFIG_BOOTFS
,
6269 kmem_free(dsname
, MAXPATHLEN
);
6271 kmem_free(tmpname
, MAXPATHLEN
);
6275 * Add the list of hot spares and level 2 cache devices.
6277 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6278 spa_add_spares(spa
, config
);
6279 spa_add_l2cache(spa
, config
);
6280 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6284 spa_deactivate(spa
);
6286 mutex_exit(&spa_namespace_lock
);
6292 * Pool export/destroy
6294 * The act of destroying or exporting a pool is very simple. We make sure there
6295 * is no more pending I/O and any references to the pool are gone. Then, we
6296 * update the pool state and sync all the labels to disk, removing the
6297 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6298 * we don't sync the labels or remove the configuration cache.
6301 spa_export_common(const char *pool
, int new_state
, nvlist_t
**oldconfig
,
6302 boolean_t force
, boolean_t hardforce
)
6310 if (!(spa_mode_global
& SPA_MODE_WRITE
))
6311 return (SET_ERROR(EROFS
));
6313 mutex_enter(&spa_namespace_lock
);
6314 if ((spa
= spa_lookup(pool
)) == NULL
) {
6315 mutex_exit(&spa_namespace_lock
);
6316 return (SET_ERROR(ENOENT
));
6319 if (spa
->spa_is_exporting
) {
6320 /* the pool is being exported by another thread */
6321 mutex_exit(&spa_namespace_lock
);
6322 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS
));
6324 spa
->spa_is_exporting
= B_TRUE
;
6327 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6328 * reacquire the namespace lock, and see if we can export.
6330 spa_open_ref(spa
, FTAG
);
6331 mutex_exit(&spa_namespace_lock
);
6332 spa_async_suspend(spa
);
6333 if (spa
->spa_zvol_taskq
) {
6334 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
6335 taskq_wait(spa
->spa_zvol_taskq
);
6337 mutex_enter(&spa_namespace_lock
);
6338 spa_close(spa
, FTAG
);
6340 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
6343 * The pool will be in core if it's openable, in which case we can
6344 * modify its state. Objsets may be open only because they're dirty,
6345 * so we have to force it to sync before checking spa_refcnt.
6347 if (spa
->spa_sync_on
) {
6348 txg_wait_synced(spa
->spa_dsl_pool
, 0);
6349 spa_evicting_os_wait(spa
);
6353 * A pool cannot be exported or destroyed if there are active
6354 * references. If we are resetting a pool, allow references by
6355 * fault injection handlers.
6357 if (!spa_refcount_zero(spa
) || (spa
->spa_inject_ref
!= 0)) {
6358 error
= SET_ERROR(EBUSY
);
6362 if (spa
->spa_sync_on
) {
6364 * A pool cannot be exported if it has an active shared spare.
6365 * This is to prevent other pools stealing the active spare
6366 * from an exported pool. At user's own will, such pool can
6367 * be forcedly exported.
6369 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
6370 spa_has_active_shared_spare(spa
)) {
6371 error
= SET_ERROR(EXDEV
);
6376 * We're about to export or destroy this pool. Make sure
6377 * we stop all initialization and trim activity here before
6378 * we set the spa_final_txg. This will ensure that all
6379 * dirty data resulting from the initialization is
6380 * committed to disk before we unload the pool.
6382 if (spa
->spa_root_vdev
!= NULL
) {
6383 vdev_t
*rvd
= spa
->spa_root_vdev
;
6384 vdev_initialize_stop_all(rvd
, VDEV_INITIALIZE_ACTIVE
);
6385 vdev_trim_stop_all(rvd
, VDEV_TRIM_ACTIVE
);
6386 vdev_autotrim_stop_all(spa
);
6387 vdev_rebuild_stop_all(spa
);
6391 * We want this to be reflected on every label,
6392 * so mark them all dirty. spa_unload() will do the
6393 * final sync that pushes these changes out.
6395 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
6396 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6397 spa
->spa_state
= new_state
;
6398 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
6400 vdev_config_dirty(spa
->spa_root_vdev
);
6401 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6406 if (new_state
== POOL_STATE_DESTROYED
)
6407 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
6408 else if (new_state
== POOL_STATE_EXPORTED
)
6409 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
6411 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6413 spa_deactivate(spa
);
6416 if (oldconfig
&& spa
->spa_config
)
6417 *oldconfig
= fnvlist_dup(spa
->spa_config
);
6419 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
6421 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
6425 * If spa_remove() is not called for this spa_t and
6426 * there is any possibility that it can be reused,
6427 * we make sure to reset the exporting flag.
6429 spa
->spa_is_exporting
= B_FALSE
;
6432 mutex_exit(&spa_namespace_lock
);
6436 spa
->spa_is_exporting
= B_FALSE
;
6437 spa_async_resume(spa
);
6438 mutex_exit(&spa_namespace_lock
);
6443 * Destroy a storage pool.
6446 spa_destroy(const char *pool
)
6448 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
6453 * Export a storage pool.
6456 spa_export(const char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
6457 boolean_t hardforce
)
6459 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
6464 * Similar to spa_export(), this unloads the spa_t without actually removing it
6465 * from the namespace in any way.
6468 spa_reset(const char *pool
)
6470 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
6475 * ==========================================================================
6476 * Device manipulation
6477 * ==========================================================================
6481 * This is called as a synctask to increment the draid feature flag
6484 spa_draid_feature_incr(void *arg
, dmu_tx_t
*tx
)
6486 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6487 int draid
= (int)(uintptr_t)arg
;
6489 for (int c
= 0; c
< draid
; c
++)
6490 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
6494 * Add a device to a storage pool.
6497 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
6499 uint64_t txg
, ndraid
= 0;
6501 vdev_t
*rvd
= spa
->spa_root_vdev
;
6503 nvlist_t
**spares
, **l2cache
;
6504 uint_t nspares
, nl2cache
;
6506 ASSERT(spa_writeable(spa
));
6508 txg
= spa_vdev_enter(spa
);
6510 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
6511 VDEV_ALLOC_ADD
)) != 0)
6512 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6514 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
6516 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
6520 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
6524 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
6525 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6527 if (vd
->vdev_children
!= 0 &&
6528 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0) {
6529 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6533 * The virtual dRAID spares must be added after vdev tree is created
6534 * and the vdev guids are generated. The guid of their associated
6535 * dRAID is stored in the config and used when opening the spare.
6537 if ((error
= vdev_draid_spare_create(nvroot
, vd
, &ndraid
,
6538 rvd
->vdev_children
)) == 0) {
6539 if (ndraid
> 0 && nvlist_lookup_nvlist_array(nvroot
,
6540 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) != 0)
6543 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6547 * We must validate the spares and l2cache devices after checking the
6548 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
6550 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
6551 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6554 * If we are in the middle of a device removal, we can only add
6555 * devices which match the existing devices in the pool.
6556 * If we are in the middle of a removal, or have some indirect
6557 * vdevs, we can not add raidz or dRAID top levels.
6559 if (spa
->spa_vdev_removal
!= NULL
||
6560 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
6561 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6562 tvd
= vd
->vdev_child
[c
];
6563 if (spa
->spa_vdev_removal
!= NULL
&&
6564 tvd
->vdev_ashift
!= spa
->spa_max_ashift
) {
6565 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6567 /* Fail if top level vdev is raidz or a dRAID */
6568 if (vdev_get_nparity(tvd
) != 0)
6569 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6572 * Need the top level mirror to be
6573 * a mirror of leaf vdevs only
6575 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
6576 for (uint64_t cid
= 0;
6577 cid
< tvd
->vdev_children
; cid
++) {
6578 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
6579 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
6580 return (spa_vdev_exit(spa
, vd
,
6588 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6589 tvd
= vd
->vdev_child
[c
];
6590 vdev_remove_child(vd
, tvd
);
6591 tvd
->vdev_id
= rvd
->vdev_children
;
6592 vdev_add_child(rvd
, tvd
);
6593 vdev_config_dirty(tvd
);
6597 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
6598 ZPOOL_CONFIG_SPARES
);
6599 spa_load_spares(spa
);
6600 spa
->spa_spares
.sav_sync
= B_TRUE
;
6603 if (nl2cache
!= 0) {
6604 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
6605 ZPOOL_CONFIG_L2CACHE
);
6606 spa_load_l2cache(spa
);
6607 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6611 * We can't increment a feature while holding spa_vdev so we
6612 * have to do it in a synctask.
6617 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
6618 dsl_sync_task_nowait(spa
->spa_dsl_pool
, spa_draid_feature_incr
,
6619 (void *)(uintptr_t)ndraid
, tx
);
6624 * We have to be careful when adding new vdevs to an existing pool.
6625 * If other threads start allocating from these vdevs before we
6626 * sync the config cache, and we lose power, then upon reboot we may
6627 * fail to open the pool because there are DVAs that the config cache
6628 * can't translate. Therefore, we first add the vdevs without
6629 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
6630 * and then let spa_config_update() initialize the new metaslabs.
6632 * spa_load() checks for added-but-not-initialized vdevs, so that
6633 * if we lose power at any point in this sequence, the remaining
6634 * steps will be completed the next time we load the pool.
6636 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
6638 mutex_enter(&spa_namespace_lock
);
6639 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6640 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
6641 mutex_exit(&spa_namespace_lock
);
6647 * Attach a device to a mirror. The arguments are the path to any device
6648 * in the mirror, and the nvroot for the new device. If the path specifies
6649 * a device that is not mirrored, we automatically insert the mirror vdev.
6651 * If 'replacing' is specified, the new device is intended to replace the
6652 * existing device; in this case the two devices are made into their own
6653 * mirror using the 'replacing' vdev, which is functionally identical to
6654 * the mirror vdev (it actually reuses all the same ops) but has a few
6655 * extra rules: you can't attach to it after it's been created, and upon
6656 * completion of resilvering, the first disk (the one being replaced)
6657 * is automatically detached.
6659 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
6660 * should be performed instead of traditional healing reconstruction. From
6661 * an administrators perspective these are both resilver operations.
6664 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
,
6667 uint64_t txg
, dtl_max_txg
;
6668 vdev_t
*rvd
= spa
->spa_root_vdev
;
6669 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
6671 char *oldvdpath
, *newvdpath
;
6675 ASSERT(spa_writeable(spa
));
6677 txg
= spa_vdev_enter(spa
);
6679 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
6681 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
6682 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
6683 error
= (spa_has_checkpoint(spa
)) ?
6684 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
6685 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6689 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_DEVICE_REBUILD
))
6690 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6692 if (dsl_scan_resilvering(spa_get_dsl(spa
)))
6693 return (spa_vdev_exit(spa
, NULL
, txg
,
6694 ZFS_ERR_RESILVER_IN_PROGRESS
));
6696 if (vdev_rebuild_active(rvd
))
6697 return (spa_vdev_exit(spa
, NULL
, txg
,
6698 ZFS_ERR_REBUILD_IN_PROGRESS
));
6701 if (spa
->spa_vdev_removal
!= NULL
)
6702 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
6705 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
6707 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
6708 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6710 pvd
= oldvd
->vdev_parent
;
6712 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
6713 VDEV_ALLOC_ATTACH
)) != 0)
6714 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
6716 if (newrootvd
->vdev_children
!= 1)
6717 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
6719 newvd
= newrootvd
->vdev_child
[0];
6721 if (!newvd
->vdev_ops
->vdev_op_leaf
)
6722 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
6724 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
6725 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
6728 * Spares can't replace logs
6730 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
6731 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6734 * A dRAID spare can only replace a child of its parent dRAID vdev.
6736 if (newvd
->vdev_ops
== &vdev_draid_spare_ops
&&
6737 oldvd
->vdev_top
!= vdev_draid_spare_get_parent(newvd
)) {
6738 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6743 * For rebuilds, the top vdev must support reconstruction
6744 * using only space maps. This means the only allowable
6745 * vdevs types are the root vdev, a mirror, or dRAID.
6748 if (pvd
->vdev_top
!= NULL
)
6749 tvd
= pvd
->vdev_top
;
6751 if (tvd
->vdev_ops
!= &vdev_mirror_ops
&&
6752 tvd
->vdev_ops
!= &vdev_root_ops
&&
6753 tvd
->vdev_ops
!= &vdev_draid_ops
) {
6754 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6760 * For attach, the only allowable parent is a mirror or the root
6763 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
6764 pvd
->vdev_ops
!= &vdev_root_ops
)
6765 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6767 pvops
= &vdev_mirror_ops
;
6770 * Active hot spares can only be replaced by inactive hot
6773 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
6774 oldvd
->vdev_isspare
&&
6775 !spa_has_spare(spa
, newvd
->vdev_guid
))
6776 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6779 * If the source is a hot spare, and the parent isn't already a
6780 * spare, then we want to create a new hot spare. Otherwise, we
6781 * want to create a replacing vdev. The user is not allowed to
6782 * attach to a spared vdev child unless the 'isspare' state is
6783 * the same (spare replaces spare, non-spare replaces
6786 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
6787 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
6788 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6789 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
6790 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
6791 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6794 if (newvd
->vdev_isspare
)
6795 pvops
= &vdev_spare_ops
;
6797 pvops
= &vdev_replacing_ops
;
6801 * Make sure the new device is big enough.
6803 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
6804 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
6807 * The new device cannot have a higher alignment requirement
6808 * than the top-level vdev.
6810 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
6811 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6814 * If this is an in-place replacement, update oldvd's path and devid
6815 * to make it distinguishable from newvd, and unopenable from now on.
6817 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
6818 spa_strfree(oldvd
->vdev_path
);
6819 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
6821 (void) snprintf(oldvd
->vdev_path
, strlen(newvd
->vdev_path
) + 5,
6822 "%s/%s", newvd
->vdev_path
, "old");
6823 if (oldvd
->vdev_devid
!= NULL
) {
6824 spa_strfree(oldvd
->vdev_devid
);
6825 oldvd
->vdev_devid
= NULL
;
6830 * If the parent is not a mirror, or if we're replacing, insert the new
6831 * mirror/replacing/spare vdev above oldvd.
6833 if (pvd
->vdev_ops
!= pvops
)
6834 pvd
= vdev_add_parent(oldvd
, pvops
);
6836 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
6837 ASSERT(pvd
->vdev_ops
== pvops
);
6838 ASSERT(oldvd
->vdev_parent
== pvd
);
6841 * Extract the new device from its root and add it to pvd.
6843 vdev_remove_child(newrootvd
, newvd
);
6844 newvd
->vdev_id
= pvd
->vdev_children
;
6845 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
6846 vdev_add_child(pvd
, newvd
);
6849 * Reevaluate the parent vdev state.
6851 vdev_propagate_state(pvd
);
6853 tvd
= newvd
->vdev_top
;
6854 ASSERT(pvd
->vdev_top
== tvd
);
6855 ASSERT(tvd
->vdev_parent
== rvd
);
6857 vdev_config_dirty(tvd
);
6860 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6861 * for any dmu_sync-ed blocks. It will propagate upward when
6862 * spa_vdev_exit() calls vdev_dtl_reassess().
6864 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
6866 vdev_dtl_dirty(newvd
, DTL_MISSING
,
6867 TXG_INITIAL
, dtl_max_txg
- TXG_INITIAL
);
6869 if (newvd
->vdev_isspare
) {
6870 spa_spare_activate(newvd
);
6871 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
6874 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
6875 newvdpath
= spa_strdup(newvd
->vdev_path
);
6876 newvd_isspare
= newvd
->vdev_isspare
;
6879 * Mark newvd's DTL dirty in this txg.
6881 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
6884 * Schedule the resilver or rebuild to restart in the future. We do
6885 * this to ensure that dmu_sync-ed blocks have been stitched into the
6886 * respective datasets.
6889 newvd
->vdev_rebuild_txg
= txg
;
6893 newvd
->vdev_resilver_txg
= txg
;
6895 if (dsl_scan_resilvering(spa_get_dsl(spa
)) &&
6896 spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
)) {
6897 vdev_defer_resilver(newvd
);
6899 dsl_scan_restart_resilver(spa
->spa_dsl_pool
,
6904 if (spa
->spa_bootfs
)
6905 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
6907 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
6912 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
6914 spa_history_log_internal(spa
, "vdev attach", NULL
,
6915 "%s vdev=%s %s vdev=%s",
6916 replacing
&& newvd_isspare
? "spare in" :
6917 replacing
? "replace" : "attach", newvdpath
,
6918 replacing
? "for" : "to", oldvdpath
);
6920 spa_strfree(oldvdpath
);
6921 spa_strfree(newvdpath
);
6927 * Detach a device from a mirror or replacing vdev.
6929 * If 'replace_done' is specified, only detach if the parent
6930 * is a replacing vdev.
6933 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
6937 vdev_t
*rvd __maybe_unused
= spa
->spa_root_vdev
;
6938 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
6939 boolean_t unspare
= B_FALSE
;
6940 uint64_t unspare_guid
= 0;
6943 ASSERT(spa_writeable(spa
));
6945 txg
= spa_vdev_detach_enter(spa
, guid
);
6947 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
6950 * Besides being called directly from the userland through the
6951 * ioctl interface, spa_vdev_detach() can be potentially called
6952 * at the end of spa_vdev_resilver_done().
6954 * In the regular case, when we have a checkpoint this shouldn't
6955 * happen as we never empty the DTLs of a vdev during the scrub
6956 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6957 * should never get here when we have a checkpoint.
6959 * That said, even in a case when we checkpoint the pool exactly
6960 * as spa_vdev_resilver_done() calls this function everything
6961 * should be fine as the resilver will return right away.
6963 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
6964 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
6965 error
= (spa_has_checkpoint(spa
)) ?
6966 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
6967 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6971 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
6973 if (!vd
->vdev_ops
->vdev_op_leaf
)
6974 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6976 pvd
= vd
->vdev_parent
;
6979 * If the parent/child relationship is not as expected, don't do it.
6980 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6981 * vdev that's replacing B with C. The user's intent in replacing
6982 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6983 * the replace by detaching C, the expected behavior is to end up
6984 * M(A,B). But suppose that right after deciding to detach C,
6985 * the replacement of B completes. We would have M(A,C), and then
6986 * ask to detach C, which would leave us with just A -- not what
6987 * the user wanted. To prevent this, we make sure that the
6988 * parent/child relationship hasn't changed -- in this example,
6989 * that C's parent is still the replacing vdev R.
6991 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
6992 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
6995 * Only 'replacing' or 'spare' vdevs can be replaced.
6997 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
6998 pvd
->vdev_ops
!= &vdev_spare_ops
)
6999 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7001 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
7002 spa_version(spa
) >= SPA_VERSION_SPARES
);
7005 * Only mirror, replacing, and spare vdevs support detach.
7007 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
7008 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
7009 pvd
->vdev_ops
!= &vdev_spare_ops
)
7010 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7013 * If this device has the only valid copy of some data,
7014 * we cannot safely detach it.
7016 if (vdev_dtl_required(vd
))
7017 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
7019 ASSERT(pvd
->vdev_children
>= 2);
7022 * If we are detaching the second disk from a replacing vdev, then
7023 * check to see if we changed the original vdev's path to have "/old"
7024 * at the end in spa_vdev_attach(). If so, undo that change now.
7026 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
7027 vd
->vdev_path
!= NULL
) {
7028 size_t len
= strlen(vd
->vdev_path
);
7030 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
7031 cvd
= pvd
->vdev_child
[c
];
7033 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
7036 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
7037 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
7038 spa_strfree(cvd
->vdev_path
);
7039 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
7046 * If we are detaching the original disk from a normal spare, then it
7047 * implies that the spare should become a real disk, and be removed
7048 * from the active spare list for the pool. dRAID spares on the
7049 * other hand are coupled to the pool and thus should never be removed
7050 * from the spares list.
7052 if (pvd
->vdev_ops
== &vdev_spare_ops
&& vd
->vdev_id
== 0) {
7053 vdev_t
*last_cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7055 if (last_cvd
->vdev_isspare
&&
7056 last_cvd
->vdev_ops
!= &vdev_draid_spare_ops
) {
7062 * Erase the disk labels so the disk can be used for other things.
7063 * This must be done after all other error cases are handled,
7064 * but before we disembowel vd (so we can still do I/O to it).
7065 * But if we can't do it, don't treat the error as fatal --
7066 * it may be that the unwritability of the disk is the reason
7067 * it's being detached!
7069 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
7072 * Remove vd from its parent and compact the parent's children.
7074 vdev_remove_child(pvd
, vd
);
7075 vdev_compact_children(pvd
);
7078 * Remember one of the remaining children so we can get tvd below.
7080 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7083 * If we need to remove the remaining child from the list of hot spares,
7084 * do it now, marking the vdev as no longer a spare in the process.
7085 * We must do this before vdev_remove_parent(), because that can
7086 * change the GUID if it creates a new toplevel GUID. For a similar
7087 * reason, we must remove the spare now, in the same txg as the detach;
7088 * otherwise someone could attach a new sibling, change the GUID, and
7089 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
7092 ASSERT(cvd
->vdev_isspare
);
7093 spa_spare_remove(cvd
);
7094 unspare_guid
= cvd
->vdev_guid
;
7095 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
7096 cvd
->vdev_unspare
= B_TRUE
;
7100 * If the parent mirror/replacing vdev only has one child,
7101 * the parent is no longer needed. Remove it from the tree.
7103 if (pvd
->vdev_children
== 1) {
7104 if (pvd
->vdev_ops
== &vdev_spare_ops
)
7105 cvd
->vdev_unspare
= B_FALSE
;
7106 vdev_remove_parent(cvd
);
7110 * We don't set tvd until now because the parent we just removed
7111 * may have been the previous top-level vdev.
7113 tvd
= cvd
->vdev_top
;
7114 ASSERT(tvd
->vdev_parent
== rvd
);
7117 * Reevaluate the parent vdev state.
7119 vdev_propagate_state(cvd
);
7122 * If the 'autoexpand' property is set on the pool then automatically
7123 * try to expand the size of the pool. For example if the device we
7124 * just detached was smaller than the others, it may be possible to
7125 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
7126 * first so that we can obtain the updated sizes of the leaf vdevs.
7128 if (spa
->spa_autoexpand
) {
7130 vdev_expand(tvd
, txg
);
7133 vdev_config_dirty(tvd
);
7136 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
7137 * vd->vdev_detached is set and free vd's DTL object in syncing context.
7138 * But first make sure we're not on any *other* txg's DTL list, to
7139 * prevent vd from being accessed after it's freed.
7141 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
7142 for (int t
= 0; t
< TXG_SIZE
; t
++)
7143 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
7144 vd
->vdev_detached
= B_TRUE
;
7145 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
7147 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
7148 spa_notify_waiters(spa
);
7150 /* hang on to the spa before we release the lock */
7151 spa_open_ref(spa
, FTAG
);
7153 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
7155 spa_history_log_internal(spa
, "detach", NULL
,
7157 spa_strfree(vdpath
);
7160 * If this was the removal of the original device in a hot spare vdev,
7161 * then we want to go through and remove the device from the hot spare
7162 * list of every other pool.
7165 spa_t
*altspa
= NULL
;
7167 mutex_enter(&spa_namespace_lock
);
7168 while ((altspa
= spa_next(altspa
)) != NULL
) {
7169 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
7173 spa_open_ref(altspa
, FTAG
);
7174 mutex_exit(&spa_namespace_lock
);
7175 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
7176 mutex_enter(&spa_namespace_lock
);
7177 spa_close(altspa
, FTAG
);
7179 mutex_exit(&spa_namespace_lock
);
7181 /* search the rest of the vdevs for spares to remove */
7182 spa_vdev_resilver_done(spa
);
7185 /* all done with the spa; OK to release */
7186 mutex_enter(&spa_namespace_lock
);
7187 spa_close(spa
, FTAG
);
7188 mutex_exit(&spa_namespace_lock
);
7194 spa_vdev_initialize_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7197 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7199 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7201 /* Look up vdev and ensure it's a leaf. */
7202 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7203 if (vd
== NULL
|| vd
->vdev_detached
) {
7204 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7205 return (SET_ERROR(ENODEV
));
7206 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7207 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7208 return (SET_ERROR(EINVAL
));
7209 } else if (!vdev_writeable(vd
)) {
7210 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7211 return (SET_ERROR(EROFS
));
7213 mutex_enter(&vd
->vdev_initialize_lock
);
7214 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7217 * When we activate an initialize action we check to see
7218 * if the vdev_initialize_thread is NULL. We do this instead
7219 * of using the vdev_initialize_state since there might be
7220 * a previous initialization process which has completed but
7221 * the thread is not exited.
7223 if (cmd_type
== POOL_INITIALIZE_START
&&
7224 (vd
->vdev_initialize_thread
!= NULL
||
7225 vd
->vdev_top
->vdev_removing
)) {
7226 mutex_exit(&vd
->vdev_initialize_lock
);
7227 return (SET_ERROR(EBUSY
));
7228 } else if (cmd_type
== POOL_INITIALIZE_CANCEL
&&
7229 (vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
&&
7230 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_SUSPENDED
)) {
7231 mutex_exit(&vd
->vdev_initialize_lock
);
7232 return (SET_ERROR(ESRCH
));
7233 } else if (cmd_type
== POOL_INITIALIZE_SUSPEND
&&
7234 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
) {
7235 mutex_exit(&vd
->vdev_initialize_lock
);
7236 return (SET_ERROR(ESRCH
));
7240 case POOL_INITIALIZE_START
:
7241 vdev_initialize(vd
);
7243 case POOL_INITIALIZE_CANCEL
:
7244 vdev_initialize_stop(vd
, VDEV_INITIALIZE_CANCELED
, vd_list
);
7246 case POOL_INITIALIZE_SUSPEND
:
7247 vdev_initialize_stop(vd
, VDEV_INITIALIZE_SUSPENDED
, vd_list
);
7250 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7252 mutex_exit(&vd
->vdev_initialize_lock
);
7258 spa_vdev_initialize(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
,
7259 nvlist_t
*vdev_errlist
)
7261 int total_errors
= 0;
7264 list_create(&vd_list
, sizeof (vdev_t
),
7265 offsetof(vdev_t
, vdev_initialize_node
));
7268 * We hold the namespace lock through the whole function
7269 * to prevent any changes to the pool while we're starting or
7270 * stopping initialization. The config and state locks are held so that
7271 * we can properly assess the vdev state before we commit to
7272 * the initializing operation.
7274 mutex_enter(&spa_namespace_lock
);
7276 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7277 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7278 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7280 int error
= spa_vdev_initialize_impl(spa
, vdev_guid
, cmd_type
,
7283 char guid_as_str
[MAXNAMELEN
];
7285 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7286 "%llu", (unsigned long long)vdev_guid
);
7287 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7292 /* Wait for all initialize threads to stop. */
7293 vdev_initialize_stop_wait(spa
, &vd_list
);
7295 /* Sync out the initializing state */
7296 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7297 mutex_exit(&spa_namespace_lock
);
7299 list_destroy(&vd_list
);
7301 return (total_errors
);
7305 spa_vdev_trim_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7306 uint64_t rate
, boolean_t partial
, boolean_t secure
, list_t
*vd_list
)
7308 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7310 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7312 /* Look up vdev and ensure it's a leaf. */
7313 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7314 if (vd
== NULL
|| vd
->vdev_detached
) {
7315 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7316 return (SET_ERROR(ENODEV
));
7317 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7318 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7319 return (SET_ERROR(EINVAL
));
7320 } else if (!vdev_writeable(vd
)) {
7321 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7322 return (SET_ERROR(EROFS
));
7323 } else if (!vd
->vdev_has_trim
) {
7324 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7325 return (SET_ERROR(EOPNOTSUPP
));
7326 } else if (secure
&& !vd
->vdev_has_securetrim
) {
7327 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7328 return (SET_ERROR(EOPNOTSUPP
));
7330 mutex_enter(&vd
->vdev_trim_lock
);
7331 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7334 * When we activate a TRIM action we check to see if the
7335 * vdev_trim_thread is NULL. We do this instead of using the
7336 * vdev_trim_state since there might be a previous TRIM process
7337 * which has completed but the thread is not exited.
7339 if (cmd_type
== POOL_TRIM_START
&&
7340 (vd
->vdev_trim_thread
!= NULL
|| vd
->vdev_top
->vdev_removing
)) {
7341 mutex_exit(&vd
->vdev_trim_lock
);
7342 return (SET_ERROR(EBUSY
));
7343 } else if (cmd_type
== POOL_TRIM_CANCEL
&&
7344 (vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
&&
7345 vd
->vdev_trim_state
!= VDEV_TRIM_SUSPENDED
)) {
7346 mutex_exit(&vd
->vdev_trim_lock
);
7347 return (SET_ERROR(ESRCH
));
7348 } else if (cmd_type
== POOL_TRIM_SUSPEND
&&
7349 vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
) {
7350 mutex_exit(&vd
->vdev_trim_lock
);
7351 return (SET_ERROR(ESRCH
));
7355 case POOL_TRIM_START
:
7356 vdev_trim(vd
, rate
, partial
, secure
);
7358 case POOL_TRIM_CANCEL
:
7359 vdev_trim_stop(vd
, VDEV_TRIM_CANCELED
, vd_list
);
7361 case POOL_TRIM_SUSPEND
:
7362 vdev_trim_stop(vd
, VDEV_TRIM_SUSPENDED
, vd_list
);
7365 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7367 mutex_exit(&vd
->vdev_trim_lock
);
7373 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
7374 * TRIM threads for each child vdev. These threads pass over all of the free
7375 * space in the vdev's metaslabs and issues TRIM commands for that space.
7378 spa_vdev_trim(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
, uint64_t rate
,
7379 boolean_t partial
, boolean_t secure
, nvlist_t
*vdev_errlist
)
7381 int total_errors
= 0;
7384 list_create(&vd_list
, sizeof (vdev_t
),
7385 offsetof(vdev_t
, vdev_trim_node
));
7388 * We hold the namespace lock through the whole function
7389 * to prevent any changes to the pool while we're starting or
7390 * stopping TRIM. The config and state locks are held so that
7391 * we can properly assess the vdev state before we commit to
7392 * the TRIM operation.
7394 mutex_enter(&spa_namespace_lock
);
7396 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7397 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7398 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7400 int error
= spa_vdev_trim_impl(spa
, vdev_guid
, cmd_type
,
7401 rate
, partial
, secure
, &vd_list
);
7403 char guid_as_str
[MAXNAMELEN
];
7405 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7406 "%llu", (unsigned long long)vdev_guid
);
7407 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7412 /* Wait for all TRIM threads to stop. */
7413 vdev_trim_stop_wait(spa
, &vd_list
);
7415 /* Sync out the TRIM state */
7416 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7417 mutex_exit(&spa_namespace_lock
);
7419 list_destroy(&vd_list
);
7421 return (total_errors
);
7425 * Split a set of devices from their mirrors, and create a new pool from them.
7428 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
7429 nvlist_t
*props
, boolean_t exp
)
7432 uint64_t txg
, *glist
;
7434 uint_t c
, children
, lastlog
;
7435 nvlist_t
**child
, *nvl
, *tmp
;
7437 char *altroot
= NULL
;
7438 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
7439 boolean_t activate_slog
;
7441 ASSERT(spa_writeable(spa
));
7443 txg
= spa_vdev_enter(spa
);
7445 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7446 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7447 error
= (spa_has_checkpoint(spa
)) ?
7448 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7449 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7452 /* clear the log and flush everything up to now */
7453 activate_slog
= spa_passivate_log(spa
);
7454 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
7455 error
= spa_reset_logs(spa
);
7456 txg
= spa_vdev_config_enter(spa
);
7459 spa_activate_log(spa
);
7462 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7464 /* check new spa name before going any further */
7465 if (spa_lookup(newname
) != NULL
)
7466 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
7469 * scan through all the children to ensure they're all mirrors
7471 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
7472 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
7474 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7476 /* first, check to ensure we've got the right child count */
7477 rvd
= spa
->spa_root_vdev
;
7479 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
7480 vdev_t
*vd
= rvd
->vdev_child
[c
];
7482 /* don't count the holes & logs as children */
7483 if (vd
->vdev_islog
|| (vd
->vdev_ops
!= &vdev_indirect_ops
&&
7484 !vdev_is_concrete(vd
))) {
7492 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
7493 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7495 /* next, ensure no spare or cache devices are part of the split */
7496 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
7497 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
7498 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7500 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
7501 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
7503 /* then, loop over each vdev and validate it */
7504 for (c
= 0; c
< children
; c
++) {
7505 uint64_t is_hole
= 0;
7507 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
7511 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
7512 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
7515 error
= SET_ERROR(EINVAL
);
7520 /* deal with indirect vdevs */
7521 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ops
==
7525 /* which disk is going to be split? */
7526 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
7528 error
= SET_ERROR(EINVAL
);
7532 /* look it up in the spa */
7533 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
7534 if (vml
[c
] == NULL
) {
7535 error
= SET_ERROR(ENODEV
);
7539 /* make sure there's nothing stopping the split */
7540 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
7541 vml
[c
]->vdev_islog
||
7542 !vdev_is_concrete(vml
[c
]) ||
7543 vml
[c
]->vdev_isspare
||
7544 vml
[c
]->vdev_isl2cache
||
7545 !vdev_writeable(vml
[c
]) ||
7546 vml
[c
]->vdev_children
!= 0 ||
7547 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
7548 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
7549 error
= SET_ERROR(EINVAL
);
7553 if (vdev_dtl_required(vml
[c
]) ||
7554 vdev_resilver_needed(vml
[c
], NULL
, NULL
)) {
7555 error
= SET_ERROR(EBUSY
);
7559 /* we need certain info from the top level */
7560 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
7561 vml
[c
]->vdev_top
->vdev_ms_array
);
7562 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
7563 vml
[c
]->vdev_top
->vdev_ms_shift
);
7564 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
7565 vml
[c
]->vdev_top
->vdev_asize
);
7566 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
7567 vml
[c
]->vdev_top
->vdev_ashift
);
7569 /* transfer per-vdev ZAPs */
7570 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
7571 VERIFY0(nvlist_add_uint64(child
[c
],
7572 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
7574 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
7575 VERIFY0(nvlist_add_uint64(child
[c
],
7576 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
7577 vml
[c
]->vdev_parent
->vdev_top_zap
));
7581 kmem_free(vml
, children
* sizeof (vdev_t
*));
7582 kmem_free(glist
, children
* sizeof (uint64_t));
7583 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7586 /* stop writers from using the disks */
7587 for (c
= 0; c
< children
; c
++) {
7589 vml
[c
]->vdev_offline
= B_TRUE
;
7591 vdev_reopen(spa
->spa_root_vdev
);
7594 * Temporarily record the splitting vdevs in the spa config. This
7595 * will disappear once the config is regenerated.
7597 nvl
= fnvlist_alloc();
7598 fnvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
, glist
, children
);
7599 kmem_free(glist
, children
* sizeof (uint64_t));
7601 mutex_enter(&spa
->spa_props_lock
);
7602 fnvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
, nvl
);
7603 mutex_exit(&spa
->spa_props_lock
);
7604 spa
->spa_config_splitting
= nvl
;
7605 vdev_config_dirty(spa
->spa_root_vdev
);
7607 /* configure and create the new pool */
7608 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
);
7609 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
7610 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
);
7611 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
, spa_version(spa
));
7612 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, spa
->spa_config_txg
);
7613 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
7614 spa_generate_guid(NULL
));
7615 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
7616 (void) nvlist_lookup_string(props
,
7617 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
7619 /* add the new pool to the namespace */
7620 newspa
= spa_add(newname
, config
, altroot
);
7621 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
7622 newspa
->spa_config_txg
= spa
->spa_config_txg
;
7623 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
7625 /* release the spa config lock, retaining the namespace lock */
7626 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
7628 if (zio_injection_enabled
)
7629 zio_handle_panic_injection(spa
, FTAG
, 1);
7631 spa_activate(newspa
, spa_mode_global
);
7632 spa_async_suspend(newspa
);
7635 * Temporarily stop the initializing and TRIM activity. We set the
7636 * state to ACTIVE so that we know to resume initializing or TRIM
7637 * once the split has completed.
7639 list_t vd_initialize_list
;
7640 list_create(&vd_initialize_list
, sizeof (vdev_t
),
7641 offsetof(vdev_t
, vdev_initialize_node
));
7643 list_t vd_trim_list
;
7644 list_create(&vd_trim_list
, sizeof (vdev_t
),
7645 offsetof(vdev_t
, vdev_trim_node
));
7647 for (c
= 0; c
< children
; c
++) {
7648 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
7649 mutex_enter(&vml
[c
]->vdev_initialize_lock
);
7650 vdev_initialize_stop(vml
[c
],
7651 VDEV_INITIALIZE_ACTIVE
, &vd_initialize_list
);
7652 mutex_exit(&vml
[c
]->vdev_initialize_lock
);
7654 mutex_enter(&vml
[c
]->vdev_trim_lock
);
7655 vdev_trim_stop(vml
[c
], VDEV_TRIM_ACTIVE
, &vd_trim_list
);
7656 mutex_exit(&vml
[c
]->vdev_trim_lock
);
7660 vdev_initialize_stop_wait(spa
, &vd_initialize_list
);
7661 vdev_trim_stop_wait(spa
, &vd_trim_list
);
7663 list_destroy(&vd_initialize_list
);
7664 list_destroy(&vd_trim_list
);
7666 newspa
->spa_config_source
= SPA_CONFIG_SRC_SPLIT
;
7667 newspa
->spa_is_splitting
= B_TRUE
;
7669 /* create the new pool from the disks of the original pool */
7670 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
);
7674 /* if that worked, generate a real config for the new pool */
7675 if (newspa
->spa_root_vdev
!= NULL
) {
7676 newspa
->spa_config_splitting
= fnvlist_alloc();
7677 fnvlist_add_uint64(newspa
->spa_config_splitting
,
7678 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
));
7679 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
7684 if (props
!= NULL
) {
7685 spa_configfile_set(newspa
, props
, B_FALSE
);
7686 error
= spa_prop_set(newspa
, props
);
7691 /* flush everything */
7692 txg
= spa_vdev_config_enter(newspa
);
7693 vdev_config_dirty(newspa
->spa_root_vdev
);
7694 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
7696 if (zio_injection_enabled
)
7697 zio_handle_panic_injection(spa
, FTAG
, 2);
7699 spa_async_resume(newspa
);
7701 /* finally, update the original pool's config */
7702 txg
= spa_vdev_config_enter(spa
);
7703 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
7704 error
= dmu_tx_assign(tx
, TXG_WAIT
);
7707 for (c
= 0; c
< children
; c
++) {
7708 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
7709 vdev_t
*tvd
= vml
[c
]->vdev_top
;
7712 * Need to be sure the detachable VDEV is not
7713 * on any *other* txg's DTL list to prevent it
7714 * from being accessed after it's freed.
7716 for (int t
= 0; t
< TXG_SIZE
; t
++) {
7717 (void) txg_list_remove_this(
7718 &tvd
->vdev_dtl_list
, vml
[c
], t
);
7723 spa_history_log_internal(spa
, "detach", tx
,
7724 "vdev=%s", vml
[c
]->vdev_path
);
7729 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
7730 vdev_config_dirty(spa
->spa_root_vdev
);
7731 spa
->spa_config_splitting
= NULL
;
7735 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
7737 if (zio_injection_enabled
)
7738 zio_handle_panic_injection(spa
, FTAG
, 3);
7740 /* split is complete; log a history record */
7741 spa_history_log_internal(newspa
, "split", NULL
,
7742 "from pool %s", spa_name(spa
));
7744 newspa
->spa_is_splitting
= B_FALSE
;
7745 kmem_free(vml
, children
* sizeof (vdev_t
*));
7747 /* if we're not going to mount the filesystems in userland, export */
7749 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
7756 spa_deactivate(newspa
);
7759 txg
= spa_vdev_config_enter(spa
);
7761 /* re-online all offlined disks */
7762 for (c
= 0; c
< children
; c
++) {
7764 vml
[c
]->vdev_offline
= B_FALSE
;
7767 /* restart initializing or trimming disks as necessary */
7768 spa_async_request(spa
, SPA_ASYNC_INITIALIZE_RESTART
);
7769 spa_async_request(spa
, SPA_ASYNC_TRIM_RESTART
);
7770 spa_async_request(spa
, SPA_ASYNC_AUTOTRIM_RESTART
);
7772 vdev_reopen(spa
->spa_root_vdev
);
7774 nvlist_free(spa
->spa_config_splitting
);
7775 spa
->spa_config_splitting
= NULL
;
7776 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
7778 kmem_free(vml
, children
* sizeof (vdev_t
*));
7783 * Find any device that's done replacing, or a vdev marked 'unspare' that's
7784 * currently spared, so we can detach it.
7787 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
7789 vdev_t
*newvd
, *oldvd
;
7791 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
7792 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
7798 * Check for a completed replacement. We always consider the first
7799 * vdev in the list to be the oldest vdev, and the last one to be
7800 * the newest (see spa_vdev_attach() for how that works). In
7801 * the case where the newest vdev is faulted, we will not automatically
7802 * remove it after a resilver completes. This is OK as it will require
7803 * user intervention to determine which disk the admin wishes to keep.
7805 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
7806 ASSERT(vd
->vdev_children
> 1);
7808 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
7809 oldvd
= vd
->vdev_child
[0];
7811 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
7812 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
7813 !vdev_dtl_required(oldvd
))
7818 * Check for a completed resilver with the 'unspare' flag set.
7819 * Also potentially update faulted state.
7821 if (vd
->vdev_ops
== &vdev_spare_ops
) {
7822 vdev_t
*first
= vd
->vdev_child
[0];
7823 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
7825 if (last
->vdev_unspare
) {
7828 } else if (first
->vdev_unspare
) {
7835 if (oldvd
!= NULL
&&
7836 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
7837 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
7838 !vdev_dtl_required(oldvd
))
7841 vdev_propagate_state(vd
);
7844 * If there are more than two spares attached to a disk,
7845 * and those spares are not required, then we want to
7846 * attempt to free them up now so that they can be used
7847 * by other pools. Once we're back down to a single
7848 * disk+spare, we stop removing them.
7850 if (vd
->vdev_children
> 2) {
7851 newvd
= vd
->vdev_child
[1];
7853 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
7854 vdev_dtl_empty(last
, DTL_MISSING
) &&
7855 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
7856 !vdev_dtl_required(newvd
))
7865 spa_vdev_resilver_done(spa_t
*spa
)
7867 vdev_t
*vd
, *pvd
, *ppvd
;
7868 uint64_t guid
, sguid
, pguid
, ppguid
;
7870 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7872 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
7873 pvd
= vd
->vdev_parent
;
7874 ppvd
= pvd
->vdev_parent
;
7875 guid
= vd
->vdev_guid
;
7876 pguid
= pvd
->vdev_guid
;
7877 ppguid
= ppvd
->vdev_guid
;
7880 * If we have just finished replacing a hot spared device, then
7881 * we need to detach the parent's first child (the original hot
7884 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
7885 ppvd
->vdev_children
== 2) {
7886 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
7887 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
7889 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
7891 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7892 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
7894 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
7896 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7899 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7902 * If a detach was not performed above replace waiters will not have
7903 * been notified. In which case we must do so now.
7905 spa_notify_waiters(spa
);
7909 * Update the stored path or FRU for this vdev.
7912 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
7916 boolean_t sync
= B_FALSE
;
7918 ASSERT(spa_writeable(spa
));
7920 spa_vdev_state_enter(spa
, SCL_ALL
);
7922 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
7923 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
7925 if (!vd
->vdev_ops
->vdev_op_leaf
)
7926 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
7929 if (strcmp(value
, vd
->vdev_path
) != 0) {
7930 spa_strfree(vd
->vdev_path
);
7931 vd
->vdev_path
= spa_strdup(value
);
7935 if (vd
->vdev_fru
== NULL
) {
7936 vd
->vdev_fru
= spa_strdup(value
);
7938 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
7939 spa_strfree(vd
->vdev_fru
);
7940 vd
->vdev_fru
= spa_strdup(value
);
7945 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
7949 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
7951 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
7955 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
7957 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
7961 * ==========================================================================
7963 * ==========================================================================
7966 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
7968 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
7970 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
7971 return (SET_ERROR(EBUSY
));
7973 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
7977 spa_scan_stop(spa_t
*spa
)
7979 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
7980 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
7981 return (SET_ERROR(EBUSY
));
7982 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
7986 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
7988 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
7990 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
7991 return (SET_ERROR(ENOTSUP
));
7993 if (func
== POOL_SCAN_RESILVER
&&
7994 !spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
))
7995 return (SET_ERROR(ENOTSUP
));
7998 * If a resilver was requested, but there is no DTL on a
7999 * writeable leaf device, we have nothing to do.
8001 if (func
== POOL_SCAN_RESILVER
&&
8002 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
8003 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
8007 return (dsl_scan(spa
->spa_dsl_pool
, func
));
8011 * ==========================================================================
8012 * SPA async task processing
8013 * ==========================================================================
8017 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
8019 if (vd
->vdev_remove_wanted
) {
8020 vd
->vdev_remove_wanted
= B_FALSE
;
8021 vd
->vdev_delayed_close
= B_FALSE
;
8022 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
8025 * We want to clear the stats, but we don't want to do a full
8026 * vdev_clear() as that will cause us to throw away
8027 * degraded/faulted state as well as attempt to reopen the
8028 * device, all of which is a waste.
8030 vd
->vdev_stat
.vs_read_errors
= 0;
8031 vd
->vdev_stat
.vs_write_errors
= 0;
8032 vd
->vdev_stat
.vs_checksum_errors
= 0;
8034 vdev_state_dirty(vd
->vdev_top
);
8036 /* Tell userspace that the vdev is gone. */
8037 zfs_post_remove(spa
, vd
);
8040 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8041 spa_async_remove(spa
, vd
->vdev_child
[c
]);
8045 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
8047 if (vd
->vdev_probe_wanted
) {
8048 vd
->vdev_probe_wanted
= B_FALSE
;
8049 vdev_reopen(vd
); /* vdev_open() does the actual probe */
8052 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8053 spa_async_probe(spa
, vd
->vdev_child
[c
]);
8057 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
8059 if (!spa
->spa_autoexpand
)
8062 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
8063 vdev_t
*cvd
= vd
->vdev_child
[c
];
8064 spa_async_autoexpand(spa
, cvd
);
8067 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
8070 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
8074 spa_async_thread(void *arg
)
8076 spa_t
*spa
= (spa_t
*)arg
;
8077 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8080 ASSERT(spa
->spa_sync_on
);
8082 mutex_enter(&spa
->spa_async_lock
);
8083 tasks
= spa
->spa_async_tasks
;
8084 spa
->spa_async_tasks
= 0;
8085 mutex_exit(&spa
->spa_async_lock
);
8088 * See if the config needs to be updated.
8090 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
8091 uint64_t old_space
, new_space
;
8093 mutex_enter(&spa_namespace_lock
);
8094 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
8095 old_space
+= metaslab_class_get_space(spa_special_class(spa
));
8096 old_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8097 old_space
+= metaslab_class_get_space(
8098 spa_embedded_log_class(spa
));
8100 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
8102 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
8103 new_space
+= metaslab_class_get_space(spa_special_class(spa
));
8104 new_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8105 new_space
+= metaslab_class_get_space(
8106 spa_embedded_log_class(spa
));
8107 mutex_exit(&spa_namespace_lock
);
8110 * If the pool grew as a result of the config update,
8111 * then log an internal history event.
8113 if (new_space
!= old_space
) {
8114 spa_history_log_internal(spa
, "vdev online", NULL
,
8115 "pool '%s' size: %llu(+%llu)",
8116 spa_name(spa
), (u_longlong_t
)new_space
,
8117 (u_longlong_t
)(new_space
- old_space
));
8122 * See if any devices need to be marked REMOVED.
8124 if (tasks
& SPA_ASYNC_REMOVE
) {
8125 spa_vdev_state_enter(spa
, SCL_NONE
);
8126 spa_async_remove(spa
, spa
->spa_root_vdev
);
8127 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
8128 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
8129 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
8130 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
8131 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8134 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
8135 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8136 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
8137 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8141 * See if any devices need to be probed.
8143 if (tasks
& SPA_ASYNC_PROBE
) {
8144 spa_vdev_state_enter(spa
, SCL_NONE
);
8145 spa_async_probe(spa
, spa
->spa_root_vdev
);
8146 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8150 * If any devices are done replacing, detach them.
8152 if (tasks
& SPA_ASYNC_RESILVER_DONE
||
8153 tasks
& SPA_ASYNC_REBUILD_DONE
) {
8154 spa_vdev_resilver_done(spa
);
8158 * Kick off a resilver.
8160 if (tasks
& SPA_ASYNC_RESILVER
&&
8161 !vdev_rebuild_active(spa
->spa_root_vdev
) &&
8162 (!dsl_scan_resilvering(dp
) ||
8163 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_RESILVER_DEFER
)))
8164 dsl_scan_restart_resilver(dp
, 0);
8166 if (tasks
& SPA_ASYNC_INITIALIZE_RESTART
) {
8167 mutex_enter(&spa_namespace_lock
);
8168 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8169 vdev_initialize_restart(spa
->spa_root_vdev
);
8170 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8171 mutex_exit(&spa_namespace_lock
);
8174 if (tasks
& SPA_ASYNC_TRIM_RESTART
) {
8175 mutex_enter(&spa_namespace_lock
);
8176 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8177 vdev_trim_restart(spa
->spa_root_vdev
);
8178 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8179 mutex_exit(&spa_namespace_lock
);
8182 if (tasks
& SPA_ASYNC_AUTOTRIM_RESTART
) {
8183 mutex_enter(&spa_namespace_lock
);
8184 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8185 vdev_autotrim_restart(spa
);
8186 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8187 mutex_exit(&spa_namespace_lock
);
8191 * Kick off L2 cache whole device TRIM.
8193 if (tasks
& SPA_ASYNC_L2CACHE_TRIM
) {
8194 mutex_enter(&spa_namespace_lock
);
8195 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8196 vdev_trim_l2arc(spa
);
8197 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8198 mutex_exit(&spa_namespace_lock
);
8202 * Kick off L2 cache rebuilding.
8204 if (tasks
& SPA_ASYNC_L2CACHE_REBUILD
) {
8205 mutex_enter(&spa_namespace_lock
);
8206 spa_config_enter(spa
, SCL_L2ARC
, FTAG
, RW_READER
);
8207 l2arc_spa_rebuild_start(spa
);
8208 spa_config_exit(spa
, SCL_L2ARC
, FTAG
);
8209 mutex_exit(&spa_namespace_lock
);
8213 * Let the world know that we're done.
8215 mutex_enter(&spa
->spa_async_lock
);
8216 spa
->spa_async_thread
= NULL
;
8217 cv_broadcast(&spa
->spa_async_cv
);
8218 mutex_exit(&spa
->spa_async_lock
);
8223 spa_async_suspend(spa_t
*spa
)
8225 mutex_enter(&spa
->spa_async_lock
);
8226 spa
->spa_async_suspended
++;
8227 while (spa
->spa_async_thread
!= NULL
)
8228 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
8229 mutex_exit(&spa
->spa_async_lock
);
8231 spa_vdev_remove_suspend(spa
);
8233 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8234 if (condense_thread
!= NULL
)
8235 zthr_cancel(condense_thread
);
8237 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8238 if (discard_thread
!= NULL
)
8239 zthr_cancel(discard_thread
);
8241 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8242 if (ll_delete_thread
!= NULL
)
8243 zthr_cancel(ll_delete_thread
);
8245 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8246 if (ll_condense_thread
!= NULL
)
8247 zthr_cancel(ll_condense_thread
);
8251 spa_async_resume(spa_t
*spa
)
8253 mutex_enter(&spa
->spa_async_lock
);
8254 ASSERT(spa
->spa_async_suspended
!= 0);
8255 spa
->spa_async_suspended
--;
8256 mutex_exit(&spa
->spa_async_lock
);
8257 spa_restart_removal(spa
);
8259 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8260 if (condense_thread
!= NULL
)
8261 zthr_resume(condense_thread
);
8263 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8264 if (discard_thread
!= NULL
)
8265 zthr_resume(discard_thread
);
8267 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8268 if (ll_delete_thread
!= NULL
)
8269 zthr_resume(ll_delete_thread
);
8271 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8272 if (ll_condense_thread
!= NULL
)
8273 zthr_resume(ll_condense_thread
);
8277 spa_async_tasks_pending(spa_t
*spa
)
8279 uint_t non_config_tasks
;
8281 boolean_t config_task_suspended
;
8283 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
8284 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
8285 if (spa
->spa_ccw_fail_time
== 0) {
8286 config_task_suspended
= B_FALSE
;
8288 config_task_suspended
=
8289 (gethrtime() - spa
->spa_ccw_fail_time
) <
8290 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
8293 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
8297 spa_async_dispatch(spa_t
*spa
)
8299 mutex_enter(&spa
->spa_async_lock
);
8300 if (spa_async_tasks_pending(spa
) &&
8301 !spa
->spa_async_suspended
&&
8302 spa
->spa_async_thread
== NULL
)
8303 spa
->spa_async_thread
= thread_create(NULL
, 0,
8304 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
8305 mutex_exit(&spa
->spa_async_lock
);
8309 spa_async_request(spa_t
*spa
, int task
)
8311 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
8312 mutex_enter(&spa
->spa_async_lock
);
8313 spa
->spa_async_tasks
|= task
;
8314 mutex_exit(&spa
->spa_async_lock
);
8318 spa_async_tasks(spa_t
*spa
)
8320 return (spa
->spa_async_tasks
);
8324 * ==========================================================================
8325 * SPA syncing routines
8326 * ==========================================================================
8331 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8335 bpobj_enqueue(bpo
, bp
, bp_freed
, tx
);
8340 bpobj_enqueue_alloc_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8342 return (bpobj_enqueue_cb(arg
, bp
, B_FALSE
, tx
));
8346 bpobj_enqueue_free_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8348 return (bpobj_enqueue_cb(arg
, bp
, B_TRUE
, tx
));
8352 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8356 zio_nowait(zio_free_sync(pio
, pio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
8362 bpobj_spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8366 return (spa_free_sync_cb(arg
, bp
, tx
));
8370 * Note: this simple function is not inlined to make it easier to dtrace the
8371 * amount of time spent syncing frees.
8374 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
8376 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8377 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
8378 VERIFY(zio_wait(zio
) == 0);
8382 * Note: this simple function is not inlined to make it easier to dtrace the
8383 * amount of time spent syncing deferred frees.
8386 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
8388 if (spa_sync_pass(spa
) != 1)
8393 * If the log space map feature is active, we stop deferring
8394 * frees to the next TXG and therefore running this function
8395 * would be considered a no-op as spa_deferred_bpobj should
8396 * not have any entries.
8398 * That said we run this function anyway (instead of returning
8399 * immediately) for the edge-case scenario where we just
8400 * activated the log space map feature in this TXG but we have
8401 * deferred frees from the previous TXG.
8403 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8404 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
8405 bpobj_spa_free_sync_cb
, zio
, tx
), ==, 0);
8406 VERIFY0(zio_wait(zio
));
8410 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
8412 char *packed
= NULL
;
8417 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
8420 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
8421 * information. This avoids the dmu_buf_will_dirty() path and
8422 * saves us a pre-read to get data we don't actually care about.
8424 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
8425 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
8427 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
8429 bzero(packed
+ nvsize
, bufsize
- nvsize
);
8431 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
8433 vmem_free(packed
, bufsize
);
8435 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
8436 dmu_buf_will_dirty(db
, tx
);
8437 *(uint64_t *)db
->db_data
= nvsize
;
8438 dmu_buf_rele(db
, FTAG
);
8442 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
8443 const char *config
, const char *entry
)
8453 * Update the MOS nvlist describing the list of available devices.
8454 * spa_validate_aux() will have already made sure this nvlist is
8455 * valid and the vdevs are labeled appropriately.
8457 if (sav
->sav_object
== 0) {
8458 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
8459 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
8460 sizeof (uint64_t), tx
);
8461 VERIFY(zap_update(spa
->spa_meta_objset
,
8462 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
8463 &sav
->sav_object
, tx
) == 0);
8466 nvroot
= fnvlist_alloc();
8467 if (sav
->sav_count
== 0) {
8468 fnvlist_add_nvlist_array(nvroot
, config
, NULL
, 0);
8470 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
8471 for (i
= 0; i
< sav
->sav_count
; i
++)
8472 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
8473 B_FALSE
, VDEV_CONFIG_L2CACHE
);
8474 fnvlist_add_nvlist_array(nvroot
, config
, list
, sav
->sav_count
);
8475 for (i
= 0; i
< sav
->sav_count
; i
++)
8476 nvlist_free(list
[i
]);
8477 kmem_free(list
, sav
->sav_count
* sizeof (void *));
8480 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
8481 nvlist_free(nvroot
);
8483 sav
->sav_sync
= B_FALSE
;
8487 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
8488 * The all-vdev ZAP must be empty.
8491 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
8493 spa_t
*spa
= vd
->vdev_spa
;
8495 if (vd
->vdev_top_zap
!= 0) {
8496 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8497 vd
->vdev_top_zap
, tx
));
8499 if (vd
->vdev_leaf_zap
!= 0) {
8500 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8501 vd
->vdev_leaf_zap
, tx
));
8503 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
8504 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
8509 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
8514 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
8515 * its config may not be dirty but we still need to build per-vdev ZAPs.
8516 * Similarly, if the pool is being assembled (e.g. after a split), we
8517 * need to rebuild the AVZ although the config may not be dirty.
8519 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
8520 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
8523 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8525 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
8526 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
8527 spa
->spa_all_vdev_zaps
!= 0);
8529 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
8530 /* Make and build the new AVZ */
8531 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
8532 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
8533 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
8535 /* Diff old AVZ with new one */
8539 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
8540 spa
->spa_all_vdev_zaps
);
8541 zap_cursor_retrieve(&zc
, &za
) == 0;
8542 zap_cursor_advance(&zc
)) {
8543 uint64_t vdzap
= za
.za_first_integer
;
8544 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
8547 * ZAP is listed in old AVZ but not in new one;
8550 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
8555 zap_cursor_fini(&zc
);
8557 /* Destroy the old AVZ */
8558 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
8559 spa
->spa_all_vdev_zaps
, tx
));
8561 /* Replace the old AVZ in the dir obj with the new one */
8562 VERIFY0(zap_update(spa
->spa_meta_objset
,
8563 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
8564 sizeof (new_avz
), 1, &new_avz
, tx
));
8566 spa
->spa_all_vdev_zaps
= new_avz
;
8567 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
8571 /* Walk through the AVZ and destroy all listed ZAPs */
8572 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
8573 spa
->spa_all_vdev_zaps
);
8574 zap_cursor_retrieve(&zc
, &za
) == 0;
8575 zap_cursor_advance(&zc
)) {
8576 uint64_t zap
= za
.za_first_integer
;
8577 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
8580 zap_cursor_fini(&zc
);
8582 /* Destroy and unlink the AVZ itself */
8583 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
8584 spa
->spa_all_vdev_zaps
, tx
));
8585 VERIFY0(zap_remove(spa
->spa_meta_objset
,
8586 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
8587 spa
->spa_all_vdev_zaps
= 0;
8590 if (spa
->spa_all_vdev_zaps
== 0) {
8591 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
8592 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
8593 DMU_POOL_VDEV_ZAP_MAP
, tx
);
8595 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
8597 /* Create ZAPs for vdevs that don't have them. */
8598 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
8600 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
8601 dmu_tx_get_txg(tx
), B_FALSE
);
8604 * If we're upgrading the spa version then make sure that
8605 * the config object gets updated with the correct version.
8607 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
8608 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
8609 spa
->spa_uberblock
.ub_version
);
8611 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8613 nvlist_free(spa
->spa_config_syncing
);
8614 spa
->spa_config_syncing
= config
;
8616 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
8620 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
8622 uint64_t *versionp
= arg
;
8623 uint64_t version
= *versionp
;
8624 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
8627 * Setting the version is special cased when first creating the pool.
8629 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
8631 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
8632 ASSERT(version
>= spa_version(spa
));
8634 spa
->spa_uberblock
.ub_version
= version
;
8635 vdev_config_dirty(spa
->spa_root_vdev
);
8636 spa_history_log_internal(spa
, "set", tx
, "version=%lld",
8637 (longlong_t
)version
);
8641 * Set zpool properties.
8644 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
8646 nvlist_t
*nvp
= arg
;
8647 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
8648 objset_t
*mos
= spa
->spa_meta_objset
;
8649 nvpair_t
*elem
= NULL
;
8651 mutex_enter(&spa
->spa_props_lock
);
8653 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
8655 char *strval
, *fname
;
8657 const char *propname
;
8658 zprop_type_t proptype
;
8661 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
8662 case ZPOOL_PROP_INVAL
:
8664 * We checked this earlier in spa_prop_validate().
8666 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
8668 fname
= strchr(nvpair_name(elem
), '@') + 1;
8669 VERIFY0(zfeature_lookup_name(fname
, &fid
));
8671 spa_feature_enable(spa
, fid
, tx
);
8672 spa_history_log_internal(spa
, "set", tx
,
8673 "%s=enabled", nvpair_name(elem
));
8676 case ZPOOL_PROP_VERSION
:
8677 intval
= fnvpair_value_uint64(elem
);
8679 * The version is synced separately before other
8680 * properties and should be correct by now.
8682 ASSERT3U(spa_version(spa
), >=, intval
);
8685 case ZPOOL_PROP_ALTROOT
:
8687 * 'altroot' is a non-persistent property. It should
8688 * have been set temporarily at creation or import time.
8690 ASSERT(spa
->spa_root
!= NULL
);
8693 case ZPOOL_PROP_READONLY
:
8694 case ZPOOL_PROP_CACHEFILE
:
8696 * 'readonly' and 'cachefile' are also non-persistent
8700 case ZPOOL_PROP_COMMENT
:
8701 strval
= fnvpair_value_string(elem
);
8702 if (spa
->spa_comment
!= NULL
)
8703 spa_strfree(spa
->spa_comment
);
8704 spa
->spa_comment
= spa_strdup(strval
);
8706 * We need to dirty the configuration on all the vdevs
8707 * so that their labels get updated. We also need to
8708 * update the cache file to keep it in sync with the
8709 * MOS version. It's unnecessary to do this for pool
8710 * creation since the vdev's configuration has already
8713 if (tx
->tx_txg
!= TXG_INITIAL
) {
8714 vdev_config_dirty(spa
->spa_root_vdev
);
8715 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
8717 spa_history_log_internal(spa
, "set", tx
,
8718 "%s=%s", nvpair_name(elem
), strval
);
8720 case ZPOOL_PROP_COMPATIBILITY
:
8721 strval
= fnvpair_value_string(elem
);
8722 if (spa
->spa_compatibility
!= NULL
)
8723 spa_strfree(spa
->spa_compatibility
);
8724 spa
->spa_compatibility
= spa_strdup(strval
);
8726 * Dirty the configuration on vdevs as above.
8728 if (tx
->tx_txg
!= TXG_INITIAL
) {
8729 vdev_config_dirty(spa
->spa_root_vdev
);
8730 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
8733 spa_history_log_internal(spa
, "set", tx
,
8734 "%s=%s", nvpair_name(elem
), strval
);
8739 * Set pool property values in the poolprops mos object.
8741 if (spa
->spa_pool_props_object
== 0) {
8742 spa
->spa_pool_props_object
=
8743 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
8744 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
8748 /* normalize the property name */
8749 propname
= zpool_prop_to_name(prop
);
8750 proptype
= zpool_prop_get_type(prop
);
8752 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
8753 ASSERT(proptype
== PROP_TYPE_STRING
);
8754 strval
= fnvpair_value_string(elem
);
8755 VERIFY0(zap_update(mos
,
8756 spa
->spa_pool_props_object
, propname
,
8757 1, strlen(strval
) + 1, strval
, tx
));
8758 spa_history_log_internal(spa
, "set", tx
,
8759 "%s=%s", nvpair_name(elem
), strval
);
8760 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
8761 intval
= fnvpair_value_uint64(elem
);
8763 if (proptype
== PROP_TYPE_INDEX
) {
8765 VERIFY0(zpool_prop_index_to_string(
8766 prop
, intval
, &unused
));
8768 VERIFY0(zap_update(mos
,
8769 spa
->spa_pool_props_object
, propname
,
8770 8, 1, &intval
, tx
));
8771 spa_history_log_internal(spa
, "set", tx
,
8772 "%s=%lld", nvpair_name(elem
),
8773 (longlong_t
)intval
);
8775 ASSERT(0); /* not allowed */
8779 case ZPOOL_PROP_DELEGATION
:
8780 spa
->spa_delegation
= intval
;
8782 case ZPOOL_PROP_BOOTFS
:
8783 spa
->spa_bootfs
= intval
;
8785 case ZPOOL_PROP_FAILUREMODE
:
8786 spa
->spa_failmode
= intval
;
8788 case ZPOOL_PROP_AUTOTRIM
:
8789 spa
->spa_autotrim
= intval
;
8790 spa_async_request(spa
,
8791 SPA_ASYNC_AUTOTRIM_RESTART
);
8793 case ZPOOL_PROP_AUTOEXPAND
:
8794 spa
->spa_autoexpand
= intval
;
8795 if (tx
->tx_txg
!= TXG_INITIAL
)
8796 spa_async_request(spa
,
8797 SPA_ASYNC_AUTOEXPAND
);
8799 case ZPOOL_PROP_MULTIHOST
:
8800 spa
->spa_multihost
= intval
;
8809 mutex_exit(&spa
->spa_props_lock
);
8813 * Perform one-time upgrade on-disk changes. spa_version() does not
8814 * reflect the new version this txg, so there must be no changes this
8815 * txg to anything that the upgrade code depends on after it executes.
8816 * Therefore this must be called after dsl_pool_sync() does the sync
8820 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
8822 if (spa_sync_pass(spa
) != 1)
8825 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8826 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
8828 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
8829 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
8830 dsl_pool_create_origin(dp
, tx
);
8832 /* Keeping the origin open increases spa_minref */
8833 spa
->spa_minref
+= 3;
8836 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
8837 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
8838 dsl_pool_upgrade_clones(dp
, tx
);
8841 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
8842 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
8843 dsl_pool_upgrade_dir_clones(dp
, tx
);
8845 /* Keeping the freedir open increases spa_minref */
8846 spa
->spa_minref
+= 3;
8849 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
8850 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
8851 spa_feature_create_zap_objects(spa
, tx
);
8855 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
8856 * when possibility to use lz4 compression for metadata was added
8857 * Old pools that have this feature enabled must be upgraded to have
8858 * this feature active
8860 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
8861 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
8862 SPA_FEATURE_LZ4_COMPRESS
);
8863 boolean_t lz4_ac
= spa_feature_is_active(spa
,
8864 SPA_FEATURE_LZ4_COMPRESS
);
8866 if (lz4_en
&& !lz4_ac
)
8867 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
8871 * If we haven't written the salt, do so now. Note that the
8872 * feature may not be activated yet, but that's fine since
8873 * the presence of this ZAP entry is backwards compatible.
8875 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
8876 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
8877 VERIFY0(zap_add(spa
->spa_meta_objset
,
8878 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
8879 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
8880 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
8883 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
8887 vdev_indirect_state_sync_verify(vdev_t
*vd
)
8889 vdev_indirect_mapping_t
*vim __maybe_unused
= vd
->vdev_indirect_mapping
;
8890 vdev_indirect_births_t
*vib __maybe_unused
= vd
->vdev_indirect_births
;
8892 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
8893 ASSERT(vim
!= NULL
);
8894 ASSERT(vib
!= NULL
);
8897 uint64_t obsolete_sm_object
= 0;
8898 ASSERT0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
8899 if (obsolete_sm_object
!= 0) {
8900 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
8901 ASSERT(vd
->vdev_removing
||
8902 vd
->vdev_ops
== &vdev_indirect_ops
);
8903 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
8904 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
8905 ASSERT3U(obsolete_sm_object
, ==,
8906 space_map_object(vd
->vdev_obsolete_sm
));
8907 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
8908 space_map_allocated(vd
->vdev_obsolete_sm
));
8910 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
8913 * Since frees / remaps to an indirect vdev can only
8914 * happen in syncing context, the obsolete segments
8915 * tree must be empty when we start syncing.
8917 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
8921 * Set the top-level vdev's max queue depth. Evaluate each top-level's
8922 * async write queue depth in case it changed. The max queue depth will
8923 * not change in the middle of syncing out this txg.
8926 spa_sync_adjust_vdev_max_queue_depth(spa_t
*spa
)
8928 ASSERT(spa_writeable(spa
));
8930 vdev_t
*rvd
= spa
->spa_root_vdev
;
8931 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
8932 zfs_vdev_queue_depth_pct
/ 100;
8933 metaslab_class_t
*normal
= spa_normal_class(spa
);
8934 metaslab_class_t
*special
= spa_special_class(spa
);
8935 metaslab_class_t
*dedup
= spa_dedup_class(spa
);
8937 uint64_t slots_per_allocator
= 0;
8938 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
8939 vdev_t
*tvd
= rvd
->vdev_child
[c
];
8941 metaslab_group_t
*mg
= tvd
->vdev_mg
;
8942 if (mg
== NULL
|| !metaslab_group_initialized(mg
))
8945 metaslab_class_t
*mc
= mg
->mg_class
;
8946 if (mc
!= normal
&& mc
!= special
&& mc
!= dedup
)
8950 * It is safe to do a lock-free check here because only async
8951 * allocations look at mg_max_alloc_queue_depth, and async
8952 * allocations all happen from spa_sync().
8954 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
8955 ASSERT0(zfs_refcount_count(
8956 &(mg
->mg_allocator
[i
].mga_alloc_queue_depth
)));
8958 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
8960 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
8961 mg
->mg_allocator
[i
].mga_cur_max_alloc_queue_depth
=
8962 zfs_vdev_def_queue_depth
;
8964 slots_per_allocator
+= zfs_vdev_def_queue_depth
;
8967 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
8968 ASSERT0(zfs_refcount_count(&normal
->mc_allocator
[i
].
8970 ASSERT0(zfs_refcount_count(&special
->mc_allocator
[i
].
8972 ASSERT0(zfs_refcount_count(&dedup
->mc_allocator
[i
].
8974 normal
->mc_allocator
[i
].mca_alloc_max_slots
=
8975 slots_per_allocator
;
8976 special
->mc_allocator
[i
].mca_alloc_max_slots
=
8977 slots_per_allocator
;
8978 dedup
->mc_allocator
[i
].mca_alloc_max_slots
=
8979 slots_per_allocator
;
8981 normal
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
8982 special
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
8983 dedup
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
8987 spa_sync_condense_indirect(spa_t
*spa
, dmu_tx_t
*tx
)
8989 ASSERT(spa_writeable(spa
));
8991 vdev_t
*rvd
= spa
->spa_root_vdev
;
8992 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
8993 vdev_t
*vd
= rvd
->vdev_child
[c
];
8994 vdev_indirect_state_sync_verify(vd
);
8996 if (vdev_indirect_should_condense(vd
)) {
8997 spa_condense_indirect_start_sync(vd
, tx
);
9004 spa_sync_iterate_to_convergence(spa_t
*spa
, dmu_tx_t
*tx
)
9006 objset_t
*mos
= spa
->spa_meta_objset
;
9007 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9008 uint64_t txg
= tx
->tx_txg
;
9009 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
9012 int pass
= ++spa
->spa_sync_pass
;
9014 spa_sync_config_object(spa
, tx
);
9015 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
9016 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
9017 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
9018 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
9019 spa_errlog_sync(spa
, txg
);
9020 dsl_pool_sync(dp
, txg
);
9022 if (pass
< zfs_sync_pass_deferred_free
||
9023 spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
)) {
9025 * If the log space map feature is active we don't
9026 * care about deferred frees and the deferred bpobj
9027 * as the log space map should effectively have the
9028 * same results (i.e. appending only to one object).
9030 spa_sync_frees(spa
, free_bpl
, tx
);
9033 * We can not defer frees in pass 1, because
9034 * we sync the deferred frees later in pass 1.
9036 ASSERT3U(pass
, >, 1);
9037 bplist_iterate(free_bpl
, bpobj_enqueue_alloc_cb
,
9038 &spa
->spa_deferred_bpobj
, tx
);
9042 dsl_scan_sync(dp
, tx
);
9044 spa_sync_upgrades(spa
, tx
);
9046 spa_flush_metaslabs(spa
, tx
);
9049 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
9054 * Note: We need to check if the MOS is dirty because we could
9055 * have marked the MOS dirty without updating the uberblock
9056 * (e.g. if we have sync tasks but no dirty user data). We need
9057 * to check the uberblock's rootbp because it is updated if we
9058 * have synced out dirty data (though in this case the MOS will
9059 * most likely also be dirty due to second order effects, we
9060 * don't want to rely on that here).
9063 spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
9064 !dmu_objset_is_dirty(mos
, txg
)) {
9066 * Nothing changed on the first pass, therefore this
9067 * TXG is a no-op. Avoid syncing deferred frees, so
9068 * that we can keep this TXG as a no-op.
9070 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9071 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9072 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
9073 ASSERT(txg_list_empty(&dp
->dp_early_sync_tasks
, txg
));
9077 spa_sync_deferred_frees(spa
, tx
);
9078 } while (dmu_objset_is_dirty(mos
, txg
));
9082 * Rewrite the vdev configuration (which includes the uberblock) to
9083 * commit the transaction group.
9085 * If there are no dirty vdevs, we sync the uberblock to a few random
9086 * top-level vdevs that are known to be visible in the config cache
9087 * (see spa_vdev_add() for a complete description). If there *are* dirty
9088 * vdevs, sync the uberblock to all vdevs.
9091 spa_sync_rewrite_vdev_config(spa_t
*spa
, dmu_tx_t
*tx
)
9093 vdev_t
*rvd
= spa
->spa_root_vdev
;
9094 uint64_t txg
= tx
->tx_txg
;
9100 * We hold SCL_STATE to prevent vdev open/close/etc.
9101 * while we're attempting to write the vdev labels.
9103 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9105 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
9106 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
9108 int children
= rvd
->vdev_children
;
9109 int c0
= random_in_range(children
);
9111 for (int c
= 0; c
< children
; c
++) {
9113 rvd
->vdev_child
[(c0
+ c
) % children
];
9115 /* Stop when revisiting the first vdev */
9116 if (c
> 0 && svd
[0] == vd
)
9119 if (vd
->vdev_ms_array
== 0 ||
9121 !vdev_is_concrete(vd
))
9124 svd
[svdcount
++] = vd
;
9125 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
9128 error
= vdev_config_sync(svd
, svdcount
, txg
);
9130 error
= vdev_config_sync(rvd
->vdev_child
,
9131 rvd
->vdev_children
, txg
);
9135 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
9137 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9141 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
9142 zio_resume_wait(spa
);
9147 * Sync the specified transaction group. New blocks may be dirtied as
9148 * part of the process, so we iterate until it converges.
9151 spa_sync(spa_t
*spa
, uint64_t txg
)
9155 VERIFY(spa_writeable(spa
));
9158 * Wait for i/os issued in open context that need to complete
9159 * before this txg syncs.
9161 (void) zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]);
9162 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
,
9166 * Lock out configuration changes.
9168 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
9170 spa
->spa_syncing_txg
= txg
;
9171 spa
->spa_sync_pass
= 0;
9173 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9174 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
9175 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
9176 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
9180 * If there are any pending vdev state changes, convert them
9181 * into config changes that go out with this transaction group.
9183 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9184 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
9186 * We need the write lock here because, for aux vdevs,
9187 * calling vdev_config_dirty() modifies sav_config.
9188 * This is ugly and will become unnecessary when we
9189 * eliminate the aux vdev wart by integrating all vdevs
9190 * into the root vdev tree.
9192 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9193 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
9194 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
9195 vdev_state_clean(vd
);
9196 vdev_config_dirty(vd
);
9198 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9199 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9201 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9203 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9204 dmu_tx_t
*tx
= dmu_tx_create_assigned(dp
, txg
);
9206 spa
->spa_sync_starttime
= gethrtime();
9207 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9208 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
9209 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
9210 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
9213 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9214 * set spa_deflate if we have no raid-z vdevs.
9216 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
9217 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
9218 vdev_t
*rvd
= spa
->spa_root_vdev
;
9221 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
9222 vd
= rvd
->vdev_child
[i
];
9223 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
9226 if (i
== rvd
->vdev_children
) {
9227 spa
->spa_deflate
= TRUE
;
9228 VERIFY0(zap_add(spa
->spa_meta_objset
,
9229 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
9230 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
9234 spa_sync_adjust_vdev_max_queue_depth(spa
);
9236 spa_sync_condense_indirect(spa
, tx
);
9238 spa_sync_iterate_to_convergence(spa
, tx
);
9241 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
9243 * Make sure that the number of ZAPs for all the vdevs matches
9244 * the number of ZAPs in the per-vdev ZAP list. This only gets
9245 * called if the config is dirty; otherwise there may be
9246 * outstanding AVZ operations that weren't completed in
9247 * spa_sync_config_object.
9249 uint64_t all_vdev_zap_entry_count
;
9250 ASSERT0(zap_count(spa
->spa_meta_objset
,
9251 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
9252 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
9253 all_vdev_zap_entry_count
);
9257 if (spa
->spa_vdev_removal
!= NULL
) {
9258 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
9261 spa_sync_rewrite_vdev_config(spa
, tx
);
9264 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9265 spa
->spa_deadman_tqid
= 0;
9268 * Clear the dirty config list.
9270 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
9271 vdev_config_clean(vd
);
9274 * Now that the new config has synced transactionally,
9275 * let it become visible to the config cache.
9277 if (spa
->spa_config_syncing
!= NULL
) {
9278 spa_config_set(spa
, spa
->spa_config_syncing
);
9279 spa
->spa_config_txg
= txg
;
9280 spa
->spa_config_syncing
= NULL
;
9283 dsl_pool_sync_done(dp
, txg
);
9285 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9286 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
9287 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
9288 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
9292 * Update usable space statistics.
9294 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
9296 vdev_sync_done(vd
, txg
);
9298 metaslab_class_evict_old(spa
->spa_normal_class
, txg
);
9299 metaslab_class_evict_old(spa
->spa_log_class
, txg
);
9301 spa_sync_close_syncing_log_sm(spa
);
9303 spa_update_dspace(spa
);
9306 * It had better be the case that we didn't dirty anything
9307 * since vdev_config_sync().
9309 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9310 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9311 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
9313 while (zfs_pause_spa_sync
)
9316 spa
->spa_sync_pass
= 0;
9319 * Update the last synced uberblock here. We want to do this at
9320 * the end of spa_sync() so that consumers of spa_last_synced_txg()
9321 * will be guaranteed that all the processing associated with
9322 * that txg has been completed.
9324 spa
->spa_ubsync
= spa
->spa_uberblock
;
9325 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
9327 spa_handle_ignored_writes(spa
);
9330 * If any async tasks have been requested, kick them off.
9332 spa_async_dispatch(spa
);
9336 * Sync all pools. We don't want to hold the namespace lock across these
9337 * operations, so we take a reference on the spa_t and drop the lock during the
9341 spa_sync_allpools(void)
9344 mutex_enter(&spa_namespace_lock
);
9345 while ((spa
= spa_next(spa
)) != NULL
) {
9346 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
9347 !spa_writeable(spa
) || spa_suspended(spa
))
9349 spa_open_ref(spa
, FTAG
);
9350 mutex_exit(&spa_namespace_lock
);
9351 txg_wait_synced(spa_get_dsl(spa
), 0);
9352 mutex_enter(&spa_namespace_lock
);
9353 spa_close(spa
, FTAG
);
9355 mutex_exit(&spa_namespace_lock
);
9359 * ==========================================================================
9360 * Miscellaneous routines
9361 * ==========================================================================
9365 * Remove all pools in the system.
9373 * Remove all cached state. All pools should be closed now,
9374 * so every spa in the AVL tree should be unreferenced.
9376 mutex_enter(&spa_namespace_lock
);
9377 while ((spa
= spa_next(NULL
)) != NULL
) {
9379 * Stop async tasks. The async thread may need to detach
9380 * a device that's been replaced, which requires grabbing
9381 * spa_namespace_lock, so we must drop it here.
9383 spa_open_ref(spa
, FTAG
);
9384 mutex_exit(&spa_namespace_lock
);
9385 spa_async_suspend(spa
);
9386 mutex_enter(&spa_namespace_lock
);
9387 spa_close(spa
, FTAG
);
9389 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
9391 spa_deactivate(spa
);
9395 mutex_exit(&spa_namespace_lock
);
9399 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
9404 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
9408 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
9409 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
9410 if (vd
->vdev_guid
== guid
)
9414 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
9415 vd
= spa
->spa_spares
.sav_vdevs
[i
];
9416 if (vd
->vdev_guid
== guid
)
9425 spa_upgrade(spa_t
*spa
, uint64_t version
)
9427 ASSERT(spa_writeable(spa
));
9429 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
9432 * This should only be called for a non-faulted pool, and since a
9433 * future version would result in an unopenable pool, this shouldn't be
9436 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
9437 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
9439 spa
->spa_uberblock
.ub_version
= version
;
9440 vdev_config_dirty(spa
->spa_root_vdev
);
9442 spa_config_exit(spa
, SCL_ALL
, FTAG
);
9444 txg_wait_synced(spa_get_dsl(spa
), 0);
9448 spa_has_spare(spa_t
*spa
, uint64_t guid
)
9452 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
9454 for (i
= 0; i
< sav
->sav_count
; i
++)
9455 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
9458 for (i
= 0; i
< sav
->sav_npending
; i
++) {
9459 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
9460 &spareguid
) == 0 && spareguid
== guid
)
9468 * Check if a pool has an active shared spare device.
9469 * Note: reference count of an active spare is 2, as a spare and as a replace
9472 spa_has_active_shared_spare(spa_t
*spa
)
9476 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
9478 for (i
= 0; i
< sav
->sav_count
; i
++) {
9479 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
9480 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
9489 spa_total_metaslabs(spa_t
*spa
)
9491 vdev_t
*rvd
= spa
->spa_root_vdev
;
9494 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
9495 vdev_t
*vd
= rvd
->vdev_child
[c
];
9496 if (!vdev_is_concrete(vd
))
9498 m
+= vd
->vdev_ms_count
;
9504 * Notify any waiting threads that some activity has switched from being in-
9505 * progress to not-in-progress so that the thread can wake up and determine
9506 * whether it is finished waiting.
9509 spa_notify_waiters(spa_t
*spa
)
9512 * Acquiring spa_activities_lock here prevents the cv_broadcast from
9513 * happening between the waiting thread's check and cv_wait.
9515 mutex_enter(&spa
->spa_activities_lock
);
9516 cv_broadcast(&spa
->spa_activities_cv
);
9517 mutex_exit(&spa
->spa_activities_lock
);
9521 * Notify any waiting threads that the pool is exporting, and then block until
9522 * they are finished using the spa_t.
9525 spa_wake_waiters(spa_t
*spa
)
9527 mutex_enter(&spa
->spa_activities_lock
);
9528 spa
->spa_waiters_cancel
= B_TRUE
;
9529 cv_broadcast(&spa
->spa_activities_cv
);
9530 while (spa
->spa_waiters
!= 0)
9531 cv_wait(&spa
->spa_waiters_cv
, &spa
->spa_activities_lock
);
9532 spa
->spa_waiters_cancel
= B_FALSE
;
9533 mutex_exit(&spa
->spa_activities_lock
);
9536 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
9538 spa_vdev_activity_in_progress_impl(vdev_t
*vd
, zpool_wait_activity_t activity
)
9540 spa_t
*spa
= vd
->vdev_spa
;
9542 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
));
9543 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
9544 ASSERT(activity
== ZPOOL_WAIT_INITIALIZE
||
9545 activity
== ZPOOL_WAIT_TRIM
);
9547 kmutex_t
*lock
= activity
== ZPOOL_WAIT_INITIALIZE
?
9548 &vd
->vdev_initialize_lock
: &vd
->vdev_trim_lock
;
9550 mutex_exit(&spa
->spa_activities_lock
);
9552 mutex_enter(&spa
->spa_activities_lock
);
9554 boolean_t in_progress
= (activity
== ZPOOL_WAIT_INITIALIZE
) ?
9555 (vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) :
9556 (vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
);
9562 for (int i
= 0; i
< vd
->vdev_children
; i
++) {
9563 if (spa_vdev_activity_in_progress_impl(vd
->vdev_child
[i
],
9572 * If use_guid is true, this checks whether the vdev specified by guid is
9573 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
9574 * is being initialized/trimmed. The caller must hold the config lock and
9575 * spa_activities_lock.
9578 spa_vdev_activity_in_progress(spa_t
*spa
, boolean_t use_guid
, uint64_t guid
,
9579 zpool_wait_activity_t activity
, boolean_t
*in_progress
)
9581 mutex_exit(&spa
->spa_activities_lock
);
9582 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9583 mutex_enter(&spa
->spa_activities_lock
);
9587 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
9588 if (vd
== NULL
|| !vd
->vdev_ops
->vdev_op_leaf
) {
9589 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9593 vd
= spa
->spa_root_vdev
;
9596 *in_progress
= spa_vdev_activity_in_progress_impl(vd
, activity
);
9598 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9603 * Locking for waiting threads
9604 * ---------------------------
9606 * Waiting threads need a way to check whether a given activity is in progress,
9607 * and then, if it is, wait for it to complete. Each activity will have some
9608 * in-memory representation of the relevant on-disk state which can be used to
9609 * determine whether or not the activity is in progress. The in-memory state and
9610 * the locking used to protect it will be different for each activity, and may
9611 * not be suitable for use with a cvar (e.g., some state is protected by the
9612 * config lock). To allow waiting threads to wait without any races, another
9613 * lock, spa_activities_lock, is used.
9615 * When the state is checked, both the activity-specific lock (if there is one)
9616 * and spa_activities_lock are held. In some cases, the activity-specific lock
9617 * is acquired explicitly (e.g. the config lock). In others, the locking is
9618 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
9619 * thread releases the activity-specific lock and, if the activity is in
9620 * progress, then cv_waits using spa_activities_lock.
9622 * The waiting thread is woken when another thread, one completing some
9623 * activity, updates the state of the activity and then calls
9624 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
9625 * needs to hold its activity-specific lock when updating the state, and this
9626 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
9628 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
9629 * and because it is held when the waiting thread checks the state of the
9630 * activity, it can never be the case that the completing thread both updates
9631 * the activity state and cv_broadcasts in between the waiting thread's check
9632 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
9634 * In order to prevent deadlock, when the waiting thread does its check, in some
9635 * cases it will temporarily drop spa_activities_lock in order to acquire the
9636 * activity-specific lock. The order in which spa_activities_lock and the
9637 * activity specific lock are acquired in the waiting thread is determined by
9638 * the order in which they are acquired in the completing thread; if the
9639 * completing thread calls spa_notify_waiters with the activity-specific lock
9640 * held, then the waiting thread must also acquire the activity-specific lock
9645 spa_activity_in_progress(spa_t
*spa
, zpool_wait_activity_t activity
,
9646 boolean_t use_tag
, uint64_t tag
, boolean_t
*in_progress
)
9650 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
9653 case ZPOOL_WAIT_CKPT_DISCARD
:
9655 (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
) &&
9656 zap_contains(spa_meta_objset(spa
),
9657 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ZPOOL_CHECKPOINT
) ==
9660 case ZPOOL_WAIT_FREE
:
9661 *in_progress
= ((spa_version(spa
) >= SPA_VERSION_DEADLISTS
&&
9662 !bpobj_is_empty(&spa
->spa_dsl_pool
->dp_free_bpobj
)) ||
9663 spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
) ||
9664 spa_livelist_delete_check(spa
));
9666 case ZPOOL_WAIT_INITIALIZE
:
9667 case ZPOOL_WAIT_TRIM
:
9668 error
= spa_vdev_activity_in_progress(spa
, use_tag
, tag
,
9669 activity
, in_progress
);
9671 case ZPOOL_WAIT_REPLACE
:
9672 mutex_exit(&spa
->spa_activities_lock
);
9673 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9674 mutex_enter(&spa
->spa_activities_lock
);
9676 *in_progress
= vdev_replace_in_progress(spa
->spa_root_vdev
);
9677 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9679 case ZPOOL_WAIT_REMOVE
:
9680 *in_progress
= (spa
->spa_removing_phys
.sr_state
==
9683 case ZPOOL_WAIT_RESILVER
:
9684 if ((*in_progress
= vdev_rebuild_active(spa
->spa_root_vdev
)))
9687 case ZPOOL_WAIT_SCRUB
:
9689 boolean_t scanning
, paused
, is_scrub
;
9690 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
9692 is_scrub
= (scn
->scn_phys
.scn_func
== POOL_SCAN_SCRUB
);
9693 scanning
= (scn
->scn_phys
.scn_state
== DSS_SCANNING
);
9694 paused
= dsl_scan_is_paused_scrub(scn
);
9695 *in_progress
= (scanning
&& !paused
&&
9696 is_scrub
== (activity
== ZPOOL_WAIT_SCRUB
));
9700 panic("unrecognized value for activity %d", activity
);
9707 spa_wait_common(const char *pool
, zpool_wait_activity_t activity
,
9708 boolean_t use_tag
, uint64_t tag
, boolean_t
*waited
)
9711 * The tag is used to distinguish between instances of an activity.
9712 * 'initialize' and 'trim' are the only activities that we use this for.
9713 * The other activities can only have a single instance in progress in a
9714 * pool at one time, making the tag unnecessary.
9716 * There can be multiple devices being replaced at once, but since they
9717 * all finish once resilvering finishes, we don't bother keeping track
9718 * of them individually, we just wait for them all to finish.
9720 if (use_tag
&& activity
!= ZPOOL_WAIT_INITIALIZE
&&
9721 activity
!= ZPOOL_WAIT_TRIM
)
9724 if (activity
< 0 || activity
>= ZPOOL_WAIT_NUM_ACTIVITIES
)
9728 int error
= spa_open(pool
, &spa
, FTAG
);
9733 * Increment the spa's waiter count so that we can call spa_close and
9734 * still ensure that the spa_t doesn't get freed before this thread is
9735 * finished with it when the pool is exported. We want to call spa_close
9736 * before we start waiting because otherwise the additional ref would
9737 * prevent the pool from being exported or destroyed throughout the
9738 * potentially long wait.
9740 mutex_enter(&spa
->spa_activities_lock
);
9742 spa_close(spa
, FTAG
);
9746 boolean_t in_progress
;
9747 error
= spa_activity_in_progress(spa
, activity
, use_tag
, tag
,
9750 if (error
|| !in_progress
|| spa
->spa_waiters_cancel
)
9755 if (cv_wait_sig(&spa
->spa_activities_cv
,
9756 &spa
->spa_activities_lock
) == 0) {
9763 cv_signal(&spa
->spa_waiters_cv
);
9764 mutex_exit(&spa
->spa_activities_lock
);
9770 * Wait for a particular instance of the specified activity to complete, where
9771 * the instance is identified by 'tag'
9774 spa_wait_tag(const char *pool
, zpool_wait_activity_t activity
, uint64_t tag
,
9777 return (spa_wait_common(pool
, activity
, B_TRUE
, tag
, waited
));
9781 * Wait for all instances of the specified activity complete
9784 spa_wait(const char *pool
, zpool_wait_activity_t activity
, boolean_t
*waited
)
9787 return (spa_wait_common(pool
, activity
, B_FALSE
, 0, waited
));
9791 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
9793 sysevent_t
*ev
= NULL
;
9797 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
9799 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
9800 ev
->resource
= resource
;
9807 spa_event_post(sysevent_t
*ev
)
9811 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
9812 kmem_free(ev
, sizeof (*ev
));
9818 * Post a zevent corresponding to the given sysevent. The 'name' must be one
9819 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
9820 * filled in from the spa and (optionally) the vdev. This doesn't do anything
9821 * in the userland libzpool, as we don't want consumers to misinterpret ztest
9822 * or zdb as real changes.
9825 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
9827 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
9830 /* state manipulation functions */
9831 EXPORT_SYMBOL(spa_open
);
9832 EXPORT_SYMBOL(spa_open_rewind
);
9833 EXPORT_SYMBOL(spa_get_stats
);
9834 EXPORT_SYMBOL(spa_create
);
9835 EXPORT_SYMBOL(spa_import
);
9836 EXPORT_SYMBOL(spa_tryimport
);
9837 EXPORT_SYMBOL(spa_destroy
);
9838 EXPORT_SYMBOL(spa_export
);
9839 EXPORT_SYMBOL(spa_reset
);
9840 EXPORT_SYMBOL(spa_async_request
);
9841 EXPORT_SYMBOL(spa_async_suspend
);
9842 EXPORT_SYMBOL(spa_async_resume
);
9843 EXPORT_SYMBOL(spa_inject_addref
);
9844 EXPORT_SYMBOL(spa_inject_delref
);
9845 EXPORT_SYMBOL(spa_scan_stat_init
);
9846 EXPORT_SYMBOL(spa_scan_get_stats
);
9848 /* device manipulation */
9849 EXPORT_SYMBOL(spa_vdev_add
);
9850 EXPORT_SYMBOL(spa_vdev_attach
);
9851 EXPORT_SYMBOL(spa_vdev_detach
);
9852 EXPORT_SYMBOL(spa_vdev_setpath
);
9853 EXPORT_SYMBOL(spa_vdev_setfru
);
9854 EXPORT_SYMBOL(spa_vdev_split_mirror
);
9856 /* spare statech is global across all pools) */
9857 EXPORT_SYMBOL(spa_spare_add
);
9858 EXPORT_SYMBOL(spa_spare_remove
);
9859 EXPORT_SYMBOL(spa_spare_exists
);
9860 EXPORT_SYMBOL(spa_spare_activate
);
9862 /* L2ARC statech is global across all pools) */
9863 EXPORT_SYMBOL(spa_l2cache_add
);
9864 EXPORT_SYMBOL(spa_l2cache_remove
);
9865 EXPORT_SYMBOL(spa_l2cache_exists
);
9866 EXPORT_SYMBOL(spa_l2cache_activate
);
9867 EXPORT_SYMBOL(spa_l2cache_drop
);
9870 EXPORT_SYMBOL(spa_scan
);
9871 EXPORT_SYMBOL(spa_scan_stop
);
9874 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
9875 EXPORT_SYMBOL(spa_sync_allpools
);
9878 EXPORT_SYMBOL(spa_prop_set
);
9879 EXPORT_SYMBOL(spa_prop_get
);
9880 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
9882 /* asynchronous event notification */
9883 EXPORT_SYMBOL(spa_event_notify
);
9886 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_shift
, INT
, ZMOD_RW
,
9887 "log2 fraction of arc that can be used by inflight I/Os when "
9888 "verifying pool during import");
9890 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_metadata
, INT
, ZMOD_RW
,
9891 "Set to traverse metadata on pool import");
9893 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_data
, INT
, ZMOD_RW
,
9894 "Set to traverse data on pool import");
9896 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_print_vdev_tree
, INT
, ZMOD_RW
,
9897 "Print vdev tree to zfs_dbgmsg during pool import");
9899 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_pct
, UINT
, ZMOD_RD
,
9900 "Percentage of CPUs to run an IO worker thread");
9902 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_tpq
, UINT
, ZMOD_RD
,
9903 "Number of threads per IO worker taskqueue");
9905 ZFS_MODULE_PARAM(zfs
, zfs_
, max_missing_tvds
, ULONG
, ZMOD_RW
,
9906 "Allow importing pool with up to this number of missing top-level "
9907 "vdevs (in read-only mode)");
9909 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_pause
, INT
, ZMOD_RW
,
9910 "Set the livelist condense zthr to pause");
9912 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_pause
, INT
, ZMOD_RW
,
9913 "Set the livelist condense synctask to pause");
9915 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_cancel
, INT
, ZMOD_RW
,
9916 "Whether livelist condensing was canceled in the synctask");
9918 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_cancel
, INT
, ZMOD_RW
,
9919 "Whether livelist condensing was canceled in the zthr function");
9921 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, new_alloc
, INT
, ZMOD_RW
,
9922 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
9923 "was being condensed");