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 static 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 static uint_t zio_taskq_batch_pct
= 80; /* 1 thread per cpu in pset */
171 static uint_t zio_taskq_batch_tpq
; /* threads per taskq */
172 static const boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
173 static const uint_t zio_taskq_basedc
= 80; /* base duty cycle */
175 static const boolean_t spa_create_process
= B_TRUE
; /* no process => no sysdc */
178 * Report any spa_load_verify errors found, but do not fail spa_load.
179 * This is used by zdb to analyze non-idle pools.
181 boolean_t spa_load_verify_dryrun
= B_FALSE
;
184 * Allow read spacemaps in case of readonly import (spa_mode == SPA_MODE_READ).
185 * This is used by zdb for spacemaps verification.
187 boolean_t spa_mode_readable_spacemaps
= B_FALSE
;
190 * This (illegal) pool name is used when temporarily importing a spa_t in order
191 * to get the vdev stats associated with the imported devices.
193 #define TRYIMPORT_NAME "$import"
196 * For debugging purposes: print out vdev tree during pool import.
198 static int spa_load_print_vdev_tree
= B_FALSE
;
201 * A non-zero value for zfs_max_missing_tvds means that we allow importing
202 * pools with missing top-level vdevs. This is strictly intended for advanced
203 * pool recovery cases since missing data is almost inevitable. Pools with
204 * missing devices can only be imported read-only for safety reasons, and their
205 * fail-mode will be automatically set to "continue".
207 * With 1 missing vdev we should be able to import the pool and mount all
208 * datasets. User data that was not modified after the missing device has been
209 * added should be recoverable. This means that snapshots created prior to the
210 * addition of that device should be completely intact.
212 * With 2 missing vdevs, some datasets may fail to mount since there are
213 * dataset statistics that are stored as regular metadata. Some data might be
214 * recoverable if those vdevs were added recently.
216 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
217 * may be missing entirely. Chances of data recovery are very low. Note that
218 * there are also risks of performing an inadvertent rewind as we might be
219 * missing all the vdevs with the latest uberblocks.
221 unsigned long zfs_max_missing_tvds
= 0;
224 * The parameters below are similar to zfs_max_missing_tvds but are only
225 * intended for a preliminary open of the pool with an untrusted config which
226 * might be incomplete or out-dated.
228 * We are more tolerant for pools opened from a cachefile since we could have
229 * an out-dated cachefile where a device removal was not registered.
230 * We could have set the limit arbitrarily high but in the case where devices
231 * are really missing we would want to return the proper error codes; we chose
232 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
233 * and we get a chance to retrieve the trusted config.
235 uint64_t zfs_max_missing_tvds_cachefile
= SPA_DVAS_PER_BP
- 1;
238 * In the case where config was assembled by scanning device paths (/dev/dsks
239 * by default) we are less tolerant since all the existing devices should have
240 * been detected and we want spa_load to return the right error codes.
242 uint64_t zfs_max_missing_tvds_scan
= 0;
245 * Debugging aid that pauses spa_sync() towards the end.
247 static const boolean_t zfs_pause_spa_sync
= B_FALSE
;
250 * Variables to indicate the livelist condense zthr func should wait at certain
251 * points for the livelist to be removed - used to test condense/destroy races
253 static int zfs_livelist_condense_zthr_pause
= 0;
254 static int zfs_livelist_condense_sync_pause
= 0;
257 * Variables to track whether or not condense cancellation has been
258 * triggered in testing.
260 static int zfs_livelist_condense_sync_cancel
= 0;
261 static int zfs_livelist_condense_zthr_cancel
= 0;
264 * Variable to track whether or not extra ALLOC blkptrs were added to a
265 * livelist entry while it was being condensed (caused by the way we track
266 * remapped blkptrs in dbuf_remap_impl)
268 static int zfs_livelist_condense_new_alloc
= 0;
271 * ==========================================================================
272 * SPA properties routines
273 * ==========================================================================
277 * Add a (source=src, propname=propval) list to an nvlist.
280 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
281 uint64_t intval
, zprop_source_t src
)
283 const char *propname
= zpool_prop_to_name(prop
);
286 propval
= fnvlist_alloc();
287 fnvlist_add_uint64(propval
, ZPROP_SOURCE
, src
);
290 fnvlist_add_string(propval
, ZPROP_VALUE
, strval
);
292 fnvlist_add_uint64(propval
, ZPROP_VALUE
, intval
);
294 fnvlist_add_nvlist(nvl
, propname
, propval
);
295 nvlist_free(propval
);
299 * Get property values from the spa configuration.
302 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
304 vdev_t
*rvd
= spa
->spa_root_vdev
;
305 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
306 uint64_t size
, alloc
, cap
, version
;
307 const zprop_source_t src
= ZPROP_SRC_NONE
;
308 spa_config_dirent_t
*dp
;
309 metaslab_class_t
*mc
= spa_normal_class(spa
);
311 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
314 alloc
= metaslab_class_get_alloc(mc
);
315 alloc
+= metaslab_class_get_alloc(spa_special_class(spa
));
316 alloc
+= metaslab_class_get_alloc(spa_dedup_class(spa
));
317 alloc
+= metaslab_class_get_alloc(spa_embedded_log_class(spa
));
319 size
= metaslab_class_get_space(mc
);
320 size
+= metaslab_class_get_space(spa_special_class(spa
));
321 size
+= metaslab_class_get_space(spa_dedup_class(spa
));
322 size
+= metaslab_class_get_space(spa_embedded_log_class(spa
));
324 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
325 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
326 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
327 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
329 spa_prop_add_list(*nvp
, ZPOOL_PROP_CHECKPOINT
, NULL
,
330 spa
->spa_checkpoint_info
.sci_dspace
, src
);
332 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
333 metaslab_class_fragmentation(mc
), src
);
334 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
335 metaslab_class_expandable_space(mc
), src
);
336 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
337 (spa_mode(spa
) == SPA_MODE_READ
), src
);
339 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
340 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
342 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
343 ddt_get_pool_dedup_ratio(spa
), src
);
345 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
346 rvd
->vdev_state
, src
);
348 version
= spa_version(spa
);
349 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
)) {
350 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
351 version
, ZPROP_SRC_DEFAULT
);
353 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
354 version
, ZPROP_SRC_LOCAL
);
356 spa_prop_add_list(*nvp
, ZPOOL_PROP_LOAD_GUID
,
357 NULL
, spa_load_guid(spa
), src
);
362 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
363 * when opening pools before this version freedir will be NULL.
365 if (pool
->dp_free_dir
!= NULL
) {
366 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
367 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
370 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
374 if (pool
->dp_leak_dir
!= NULL
) {
375 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
376 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
379 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
384 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
386 if (spa
->spa_comment
!= NULL
) {
387 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
391 if (spa
->spa_compatibility
!= NULL
) {
392 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMPATIBILITY
,
393 spa
->spa_compatibility
, 0, ZPROP_SRC_LOCAL
);
396 if (spa
->spa_root
!= NULL
)
397 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
400 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
401 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
402 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
404 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
405 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
408 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
)) {
409 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
410 DNODE_MAX_SIZE
, ZPROP_SRC_NONE
);
412 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
413 DNODE_MIN_SIZE
, ZPROP_SRC_NONE
);
416 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
417 if (dp
->scd_path
== NULL
) {
418 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
419 "none", 0, ZPROP_SRC_LOCAL
);
420 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
421 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
422 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
428 * Get zpool property values.
431 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
433 objset_t
*mos
= spa
->spa_meta_objset
;
439 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
443 dp
= spa_get_dsl(spa
);
444 dsl_pool_config_enter(dp
, FTAG
);
445 mutex_enter(&spa
->spa_props_lock
);
448 * Get properties from the spa config.
450 spa_prop_get_config(spa
, nvp
);
452 /* If no pool property object, no more prop to get. */
453 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0)
457 * Get properties from the MOS pool property object.
459 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
460 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
461 zap_cursor_advance(&zc
)) {
464 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
467 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPOOL_PROP_INVAL
)
470 switch (za
.za_integer_length
) {
472 /* integer property */
473 if (za
.za_first_integer
!=
474 zpool_prop_default_numeric(prop
))
475 src
= ZPROP_SRC_LOCAL
;
477 if (prop
== ZPOOL_PROP_BOOTFS
) {
478 dsl_dataset_t
*ds
= NULL
;
480 err
= dsl_dataset_hold_obj(dp
,
481 za
.za_first_integer
, FTAG
, &ds
);
485 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
487 dsl_dataset_name(ds
, strval
);
488 dsl_dataset_rele(ds
, FTAG
);
491 intval
= za
.za_first_integer
;
494 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
497 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
502 /* string property */
503 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
504 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
505 za
.za_name
, 1, za
.za_num_integers
, strval
);
507 kmem_free(strval
, za
.za_num_integers
);
510 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
511 kmem_free(strval
, za
.za_num_integers
);
518 zap_cursor_fini(&zc
);
520 mutex_exit(&spa
->spa_props_lock
);
521 dsl_pool_config_exit(dp
, FTAG
);
522 if (err
&& err
!= ENOENT
) {
532 * Validate the given pool properties nvlist and modify the list
533 * for the property values to be set.
536 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
539 int error
= 0, reset_bootfs
= 0;
541 boolean_t has_feature
= B_FALSE
;
544 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
546 char *strval
, *slash
, *check
, *fname
;
547 const char *propname
= nvpair_name(elem
);
548 zpool_prop_t prop
= zpool_name_to_prop(propname
);
551 case ZPOOL_PROP_INVAL
:
552 if (!zpool_prop_feature(propname
)) {
553 error
= SET_ERROR(EINVAL
);
558 * Sanitize the input.
560 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
561 error
= SET_ERROR(EINVAL
);
565 if (nvpair_value_uint64(elem
, &intval
) != 0) {
566 error
= SET_ERROR(EINVAL
);
571 error
= SET_ERROR(EINVAL
);
575 fname
= strchr(propname
, '@') + 1;
576 if (zfeature_lookup_name(fname
, NULL
) != 0) {
577 error
= SET_ERROR(EINVAL
);
581 has_feature
= B_TRUE
;
584 case ZPOOL_PROP_VERSION
:
585 error
= nvpair_value_uint64(elem
, &intval
);
587 (intval
< spa_version(spa
) ||
588 intval
> SPA_VERSION_BEFORE_FEATURES
||
590 error
= SET_ERROR(EINVAL
);
593 case ZPOOL_PROP_DELEGATION
:
594 case ZPOOL_PROP_AUTOREPLACE
:
595 case ZPOOL_PROP_LISTSNAPS
:
596 case ZPOOL_PROP_AUTOEXPAND
:
597 case ZPOOL_PROP_AUTOTRIM
:
598 error
= nvpair_value_uint64(elem
, &intval
);
599 if (!error
&& intval
> 1)
600 error
= SET_ERROR(EINVAL
);
603 case ZPOOL_PROP_MULTIHOST
:
604 error
= nvpair_value_uint64(elem
, &intval
);
605 if (!error
&& intval
> 1)
606 error
= SET_ERROR(EINVAL
);
609 uint32_t hostid
= zone_get_hostid(NULL
);
611 spa
->spa_hostid
= hostid
;
613 error
= SET_ERROR(ENOTSUP
);
618 case ZPOOL_PROP_BOOTFS
:
620 * If the pool version is less than SPA_VERSION_BOOTFS,
621 * or the pool is still being created (version == 0),
622 * the bootfs property cannot be set.
624 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
625 error
= SET_ERROR(ENOTSUP
);
630 * Make sure the vdev config is bootable
632 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
633 error
= SET_ERROR(ENOTSUP
);
639 error
= nvpair_value_string(elem
, &strval
);
644 if (strval
== NULL
|| strval
[0] == '\0') {
645 objnum
= zpool_prop_default_numeric(
650 error
= dmu_objset_hold(strval
, FTAG
, &os
);
655 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
656 error
= SET_ERROR(ENOTSUP
);
658 objnum
= dmu_objset_id(os
);
660 dmu_objset_rele(os
, FTAG
);
664 case ZPOOL_PROP_FAILUREMODE
:
665 error
= nvpair_value_uint64(elem
, &intval
);
666 if (!error
&& intval
> ZIO_FAILURE_MODE_PANIC
)
667 error
= SET_ERROR(EINVAL
);
670 * This is a special case which only occurs when
671 * the pool has completely failed. This allows
672 * the user to change the in-core failmode property
673 * without syncing it out to disk (I/Os might
674 * currently be blocked). We do this by returning
675 * EIO to the caller (spa_prop_set) to trick it
676 * into thinking we encountered a property validation
679 if (!error
&& spa_suspended(spa
)) {
680 spa
->spa_failmode
= intval
;
681 error
= SET_ERROR(EIO
);
685 case ZPOOL_PROP_CACHEFILE
:
686 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
689 if (strval
[0] == '\0')
692 if (strcmp(strval
, "none") == 0)
695 if (strval
[0] != '/') {
696 error
= SET_ERROR(EINVAL
);
700 slash
= strrchr(strval
, '/');
701 ASSERT(slash
!= NULL
);
703 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
704 strcmp(slash
, "/..") == 0)
705 error
= SET_ERROR(EINVAL
);
708 case ZPOOL_PROP_COMMENT
:
709 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
711 for (check
= strval
; *check
!= '\0'; check
++) {
712 if (!isprint(*check
)) {
713 error
= SET_ERROR(EINVAL
);
717 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
718 error
= SET_ERROR(E2BIG
);
729 (void) nvlist_remove_all(props
,
730 zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO
));
732 if (!error
&& reset_bootfs
) {
733 error
= nvlist_remove(props
,
734 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
737 error
= nvlist_add_uint64(props
,
738 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
746 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
749 spa_config_dirent_t
*dp
;
751 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
755 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
758 if (cachefile
[0] == '\0')
759 dp
->scd_path
= spa_strdup(spa_config_path
);
760 else if (strcmp(cachefile
, "none") == 0)
763 dp
->scd_path
= spa_strdup(cachefile
);
765 list_insert_head(&spa
->spa_config_list
, dp
);
767 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
771 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
774 nvpair_t
*elem
= NULL
;
775 boolean_t need_sync
= B_FALSE
;
777 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
780 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
781 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
783 if (prop
== ZPOOL_PROP_CACHEFILE
||
784 prop
== ZPOOL_PROP_ALTROOT
||
785 prop
== ZPOOL_PROP_READONLY
)
788 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPOOL_PROP_INVAL
) {
791 if (prop
== ZPOOL_PROP_VERSION
) {
792 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
794 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
795 ver
= SPA_VERSION_FEATURES
;
799 /* Save time if the version is already set. */
800 if (ver
== spa_version(spa
))
804 * In addition to the pool directory object, we might
805 * create the pool properties object, the features for
806 * read object, the features for write object, or the
807 * feature descriptions object.
809 error
= dsl_sync_task(spa
->spa_name
, NULL
,
810 spa_sync_version
, &ver
,
811 6, ZFS_SPACE_CHECK_RESERVED
);
822 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
823 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
830 * If the bootfs property value is dsobj, clear it.
833 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
835 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
836 VERIFY(zap_remove(spa
->spa_meta_objset
,
837 spa
->spa_pool_props_object
,
838 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
844 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
846 uint64_t *newguid __maybe_unused
= arg
;
847 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
848 vdev_t
*rvd
= spa
->spa_root_vdev
;
851 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
852 int error
= (spa_has_checkpoint(spa
)) ?
853 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
854 return (SET_ERROR(error
));
857 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
858 vdev_state
= rvd
->vdev_state
;
859 spa_config_exit(spa
, SCL_STATE
, FTAG
);
861 if (vdev_state
!= VDEV_STATE_HEALTHY
)
862 return (SET_ERROR(ENXIO
));
864 ASSERT3U(spa_guid(spa
), !=, *newguid
);
870 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
872 uint64_t *newguid
= arg
;
873 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
875 vdev_t
*rvd
= spa
->spa_root_vdev
;
877 oldguid
= spa_guid(spa
);
879 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
880 rvd
->vdev_guid
= *newguid
;
881 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
882 vdev_config_dirty(rvd
);
883 spa_config_exit(spa
, SCL_STATE
, FTAG
);
885 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
886 (u_longlong_t
)oldguid
, (u_longlong_t
)*newguid
);
890 * Change the GUID for the pool. This is done so that we can later
891 * re-import a pool built from a clone of our own vdevs. We will modify
892 * the root vdev's guid, our own pool guid, and then mark all of our
893 * vdevs dirty. Note that we must make sure that all our vdevs are
894 * online when we do this, or else any vdevs that weren't present
895 * would be orphaned from our pool. We are also going to issue a
896 * sysevent to update any watchers.
899 spa_change_guid(spa_t
*spa
)
904 mutex_enter(&spa
->spa_vdev_top_lock
);
905 mutex_enter(&spa_namespace_lock
);
906 guid
= spa_generate_guid(NULL
);
908 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
909 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
912 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
913 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
916 mutex_exit(&spa_namespace_lock
);
917 mutex_exit(&spa
->spa_vdev_top_lock
);
923 * ==========================================================================
924 * SPA state manipulation (open/create/destroy/import/export)
925 * ==========================================================================
929 spa_error_entry_compare(const void *a
, const void *b
)
931 const spa_error_entry_t
*sa
= (const spa_error_entry_t
*)a
;
932 const spa_error_entry_t
*sb
= (const spa_error_entry_t
*)b
;
935 ret
= memcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
936 sizeof (zbookmark_phys_t
));
938 return (TREE_ISIGN(ret
));
942 * Utility function which retrieves copies of the current logs and
943 * re-initializes them in the process.
946 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
948 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
950 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
951 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
953 avl_create(&spa
->spa_errlist_scrub
,
954 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
955 offsetof(spa_error_entry_t
, se_avl
));
956 avl_create(&spa
->spa_errlist_last
,
957 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
958 offsetof(spa_error_entry_t
, se_avl
));
962 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
964 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
965 enum zti_modes mode
= ztip
->zti_mode
;
966 uint_t value
= ztip
->zti_value
;
967 uint_t count
= ztip
->zti_count
;
968 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
969 uint_t cpus
, flags
= TASKQ_DYNAMIC
;
970 boolean_t batch
= B_FALSE
;
974 ASSERT3U(value
, >, 0);
979 flags
|= TASKQ_THREADS_CPU_PCT
;
980 value
= MIN(zio_taskq_batch_pct
, 100);
984 flags
|= TASKQ_THREADS_CPU_PCT
;
986 * We want more taskqs to reduce lock contention, but we want
987 * less for better request ordering and CPU utilization.
989 cpus
= MAX(1, boot_ncpus
* zio_taskq_batch_pct
/ 100);
990 if (zio_taskq_batch_tpq
> 0) {
991 count
= MAX(1, (cpus
+ zio_taskq_batch_tpq
/ 2) /
992 zio_taskq_batch_tpq
);
995 * Prefer 6 threads per taskq, but no more taskqs
996 * than threads in them on large systems. For 80%:
999 * cpus taskqs percent threads threads
1000 * ------- ------- ------- ------- -------
1011 count
= 1 + cpus
/ 6;
1012 while (count
* count
> cpus
)
1015 /* Limit each taskq within 100% to not trigger assertion. */
1016 count
= MAX(count
, (zio_taskq_batch_pct
+ 99) / 100);
1017 value
= (zio_taskq_batch_pct
+ count
/ 2) / count
;
1021 tqs
->stqs_count
= 0;
1022 tqs
->stqs_taskq
= NULL
;
1026 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
1028 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
1032 ASSERT3U(count
, >, 0);
1033 tqs
->stqs_count
= count
;
1034 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
1036 for (uint_t i
= 0; i
< count
; i
++) {
1041 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
1042 zio_type_name
[t
], zio_taskq_types
[q
], i
);
1044 (void) snprintf(name
, sizeof (name
), "%s_%s",
1045 zio_type_name
[t
], zio_taskq_types
[q
]);
1047 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
1049 flags
|= TASKQ_DC_BATCH
;
1051 (void) zio_taskq_basedc
;
1052 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
1053 spa
->spa_proc
, zio_taskq_basedc
, flags
);
1055 pri_t pri
= maxclsyspri
;
1057 * The write issue taskq can be extremely CPU
1058 * intensive. Run it at slightly less important
1059 * priority than the other taskqs.
1061 * Under Linux and FreeBSD this means incrementing
1062 * the priority value as opposed to platforms like
1063 * illumos where it should be decremented.
1065 * On FreeBSD, if priorities divided by four (RQ_PPQ)
1066 * are equal then a difference between them is
1069 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
) {
1070 #if defined(__linux__)
1072 #elif defined(__FreeBSD__)
1078 tq
= taskq_create_proc(name
, value
, pri
, 50,
1079 INT_MAX
, spa
->spa_proc
, flags
);
1082 tqs
->stqs_taskq
[i
] = tq
;
1087 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
1089 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1091 if (tqs
->stqs_taskq
== NULL
) {
1092 ASSERT3U(tqs
->stqs_count
, ==, 0);
1096 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
1097 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
1098 taskq_destroy(tqs
->stqs_taskq
[i
]);
1101 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
1102 tqs
->stqs_taskq
= NULL
;
1106 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1107 * Note that a type may have multiple discrete taskqs to avoid lock contention
1108 * on the taskq itself. In that case we choose which taskq at random by using
1109 * the low bits of gethrtime().
1112 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1113 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
1115 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1118 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1119 ASSERT3U(tqs
->stqs_count
, !=, 0);
1121 if (tqs
->stqs_count
== 1) {
1122 tq
= tqs
->stqs_taskq
[0];
1124 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1127 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
1131 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1134 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1135 task_func_t
*func
, void *arg
, uint_t flags
)
1137 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1141 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1142 ASSERT3U(tqs
->stqs_count
, !=, 0);
1144 if (tqs
->stqs_count
== 1) {
1145 tq
= tqs
->stqs_taskq
[0];
1147 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1150 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1152 taskq_wait_id(tq
, id
);
1156 spa_create_zio_taskqs(spa_t
*spa
)
1158 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1159 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1160 spa_taskqs_init(spa
, t
, q
);
1166 * Disabled until spa_thread() can be adapted for Linux.
1168 #undef HAVE_SPA_THREAD
1170 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1172 spa_thread(void *arg
)
1174 psetid_t zio_taskq_psrset_bind
= PS_NONE
;
1175 callb_cpr_t cprinfo
;
1178 user_t
*pu
= PTOU(curproc
);
1180 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1183 ASSERT(curproc
!= &p0
);
1184 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1185 "zpool-%s", spa
->spa_name
);
1186 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1188 /* bind this thread to the requested psrset */
1189 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1191 mutex_enter(&cpu_lock
);
1192 mutex_enter(&pidlock
);
1193 mutex_enter(&curproc
->p_lock
);
1195 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1196 0, NULL
, NULL
) == 0) {
1197 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1200 "Couldn't bind process for zfs pool \"%s\" to "
1201 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1204 mutex_exit(&curproc
->p_lock
);
1205 mutex_exit(&pidlock
);
1206 mutex_exit(&cpu_lock
);
1210 if (zio_taskq_sysdc
) {
1211 sysdc_thread_enter(curthread
, 100, 0);
1214 spa
->spa_proc
= curproc
;
1215 spa
->spa_did
= curthread
->t_did
;
1217 spa_create_zio_taskqs(spa
);
1219 mutex_enter(&spa
->spa_proc_lock
);
1220 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1222 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1223 cv_broadcast(&spa
->spa_proc_cv
);
1225 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1226 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1227 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1228 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1230 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1231 spa
->spa_proc_state
= SPA_PROC_GONE
;
1232 spa
->spa_proc
= &p0
;
1233 cv_broadcast(&spa
->spa_proc_cv
);
1234 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1236 mutex_enter(&curproc
->p_lock
);
1242 * Activate an uninitialized pool.
1245 spa_activate(spa_t
*spa
, spa_mode_t mode
)
1247 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1249 spa
->spa_state
= POOL_STATE_ACTIVE
;
1250 spa
->spa_mode
= mode
;
1251 spa
->spa_read_spacemaps
= spa_mode_readable_spacemaps
;
1253 spa
->spa_normal_class
= metaslab_class_create(spa
, &zfs_metaslab_ops
);
1254 spa
->spa_log_class
= metaslab_class_create(spa
, &zfs_metaslab_ops
);
1255 spa
->spa_embedded_log_class
=
1256 metaslab_class_create(spa
, &zfs_metaslab_ops
);
1257 spa
->spa_special_class
= metaslab_class_create(spa
, &zfs_metaslab_ops
);
1258 spa
->spa_dedup_class
= metaslab_class_create(spa
, &zfs_metaslab_ops
);
1260 /* Try to create a covering process */
1261 mutex_enter(&spa
->spa_proc_lock
);
1262 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1263 ASSERT(spa
->spa_proc
== &p0
);
1266 (void) spa_create_process
;
1267 #ifdef HAVE_SPA_THREAD
1268 /* Only create a process if we're going to be around a while. */
1269 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1270 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1272 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1273 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1274 cv_wait(&spa
->spa_proc_cv
,
1275 &spa
->spa_proc_lock
);
1277 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1278 ASSERT(spa
->spa_proc
!= &p0
);
1279 ASSERT(spa
->spa_did
!= 0);
1283 "Couldn't create process for zfs pool \"%s\"\n",
1288 #endif /* HAVE_SPA_THREAD */
1289 mutex_exit(&spa
->spa_proc_lock
);
1291 /* If we didn't create a process, we need to create our taskqs. */
1292 if (spa
->spa_proc
== &p0
) {
1293 spa_create_zio_taskqs(spa
);
1296 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1297 spa
->spa_txg_zio
[i
] = zio_root(spa
, NULL
, NULL
,
1301 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1302 offsetof(vdev_t
, vdev_config_dirty_node
));
1303 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1304 offsetof(objset_t
, os_evicting_node
));
1305 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1306 offsetof(vdev_t
, vdev_state_dirty_node
));
1308 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1309 offsetof(struct vdev
, vdev_txg_node
));
1311 avl_create(&spa
->spa_errlist_scrub
,
1312 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1313 offsetof(spa_error_entry_t
, se_avl
));
1314 avl_create(&spa
->spa_errlist_last
,
1315 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1316 offsetof(spa_error_entry_t
, se_avl
));
1318 spa_keystore_init(&spa
->spa_keystore
);
1321 * This taskq is used to perform zvol-minor-related tasks
1322 * asynchronously. This has several advantages, including easy
1323 * resolution of various deadlocks.
1325 * The taskq must be single threaded to ensure tasks are always
1326 * processed in the order in which they were dispatched.
1328 * A taskq per pool allows one to keep the pools independent.
1329 * This way if one pool is suspended, it will not impact another.
1331 * The preferred location to dispatch a zvol minor task is a sync
1332 * task. In this context, there is easy access to the spa_t and minimal
1333 * error handling is required because the sync task must succeed.
1335 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1339 * Taskq dedicated to prefetcher threads: this is used to prevent the
1340 * pool traverse code from monopolizing the global (and limited)
1341 * system_taskq by inappropriately scheduling long running tasks on it.
1343 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", 100,
1344 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1347 * The taskq to upgrade datasets in this pool. Currently used by
1348 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1350 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", 100,
1351 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1355 * Opposite of spa_activate().
1358 spa_deactivate(spa_t
*spa
)
1360 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1361 ASSERT(spa
->spa_dsl_pool
== NULL
);
1362 ASSERT(spa
->spa_root_vdev
== NULL
);
1363 ASSERT(spa
->spa_async_zio_root
== NULL
);
1364 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1366 spa_evicting_os_wait(spa
);
1368 if (spa
->spa_zvol_taskq
) {
1369 taskq_destroy(spa
->spa_zvol_taskq
);
1370 spa
->spa_zvol_taskq
= NULL
;
1373 if (spa
->spa_prefetch_taskq
) {
1374 taskq_destroy(spa
->spa_prefetch_taskq
);
1375 spa
->spa_prefetch_taskq
= NULL
;
1378 if (spa
->spa_upgrade_taskq
) {
1379 taskq_destroy(spa
->spa_upgrade_taskq
);
1380 spa
->spa_upgrade_taskq
= NULL
;
1383 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1385 list_destroy(&spa
->spa_config_dirty_list
);
1386 list_destroy(&spa
->spa_evicting_os_list
);
1387 list_destroy(&spa
->spa_state_dirty_list
);
1389 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1391 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1392 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1393 spa_taskqs_fini(spa
, t
, q
);
1397 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1398 ASSERT3P(spa
->spa_txg_zio
[i
], !=, NULL
);
1399 VERIFY0(zio_wait(spa
->spa_txg_zio
[i
]));
1400 spa
->spa_txg_zio
[i
] = NULL
;
1403 metaslab_class_destroy(spa
->spa_normal_class
);
1404 spa
->spa_normal_class
= NULL
;
1406 metaslab_class_destroy(spa
->spa_log_class
);
1407 spa
->spa_log_class
= NULL
;
1409 metaslab_class_destroy(spa
->spa_embedded_log_class
);
1410 spa
->spa_embedded_log_class
= NULL
;
1412 metaslab_class_destroy(spa
->spa_special_class
);
1413 spa
->spa_special_class
= NULL
;
1415 metaslab_class_destroy(spa
->spa_dedup_class
);
1416 spa
->spa_dedup_class
= NULL
;
1419 * If this was part of an import or the open otherwise failed, we may
1420 * still have errors left in the queues. Empty them just in case.
1422 spa_errlog_drain(spa
);
1423 avl_destroy(&spa
->spa_errlist_scrub
);
1424 avl_destroy(&spa
->spa_errlist_last
);
1426 spa_keystore_fini(&spa
->spa_keystore
);
1428 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1430 mutex_enter(&spa
->spa_proc_lock
);
1431 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1432 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1433 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1434 cv_broadcast(&spa
->spa_proc_cv
);
1435 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1436 ASSERT(spa
->spa_proc
!= &p0
);
1437 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1439 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1440 spa
->spa_proc_state
= SPA_PROC_NONE
;
1442 ASSERT(spa
->spa_proc
== &p0
);
1443 mutex_exit(&spa
->spa_proc_lock
);
1446 * We want to make sure spa_thread() has actually exited the ZFS
1447 * module, so that the module can't be unloaded out from underneath
1450 if (spa
->spa_did
!= 0) {
1451 thread_join(spa
->spa_did
);
1457 * Verify a pool configuration, and construct the vdev tree appropriately. This
1458 * will create all the necessary vdevs in the appropriate layout, with each vdev
1459 * in the CLOSED state. This will prep the pool before open/creation/import.
1460 * All vdev validation is done by the vdev_alloc() routine.
1463 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1464 uint_t id
, int atype
)
1470 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1473 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1476 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1479 if (error
== ENOENT
)
1485 return (SET_ERROR(EINVAL
));
1488 for (int c
= 0; c
< children
; c
++) {
1490 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1498 ASSERT(*vdp
!= NULL
);
1504 spa_should_flush_logs_on_unload(spa_t
*spa
)
1506 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
1509 if (!spa_writeable(spa
))
1512 if (!spa
->spa_sync_on
)
1515 if (spa_state(spa
) != POOL_STATE_EXPORTED
)
1518 if (zfs_keep_log_spacemaps_at_export
)
1525 * Opens a transaction that will set the flag that will instruct
1526 * spa_sync to attempt to flush all the metaslabs for that txg.
1529 spa_unload_log_sm_flush_all(spa_t
*spa
)
1531 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
1532 VERIFY0(dmu_tx_assign(tx
, TXG_WAIT
));
1534 ASSERT3U(spa
->spa_log_flushall_txg
, ==, 0);
1535 spa
->spa_log_flushall_txg
= dmu_tx_get_txg(tx
);
1538 txg_wait_synced(spa_get_dsl(spa
), spa
->spa_log_flushall_txg
);
1542 spa_unload_log_sm_metadata(spa_t
*spa
)
1544 void *cookie
= NULL
;
1546 while ((sls
= avl_destroy_nodes(&spa
->spa_sm_logs_by_txg
,
1547 &cookie
)) != NULL
) {
1548 VERIFY0(sls
->sls_mscount
);
1549 kmem_free(sls
, sizeof (spa_log_sm_t
));
1552 for (log_summary_entry_t
*e
= list_head(&spa
->spa_log_summary
);
1553 e
!= NULL
; e
= list_head(&spa
->spa_log_summary
)) {
1554 VERIFY0(e
->lse_mscount
);
1555 list_remove(&spa
->spa_log_summary
, e
);
1556 kmem_free(e
, sizeof (log_summary_entry_t
));
1559 spa
->spa_unflushed_stats
.sus_nblocks
= 0;
1560 spa
->spa_unflushed_stats
.sus_memused
= 0;
1561 spa
->spa_unflushed_stats
.sus_blocklimit
= 0;
1565 spa_destroy_aux_threads(spa_t
*spa
)
1567 if (spa
->spa_condense_zthr
!= NULL
) {
1568 zthr_destroy(spa
->spa_condense_zthr
);
1569 spa
->spa_condense_zthr
= NULL
;
1571 if (spa
->spa_checkpoint_discard_zthr
!= NULL
) {
1572 zthr_destroy(spa
->spa_checkpoint_discard_zthr
);
1573 spa
->spa_checkpoint_discard_zthr
= NULL
;
1575 if (spa
->spa_livelist_delete_zthr
!= NULL
) {
1576 zthr_destroy(spa
->spa_livelist_delete_zthr
);
1577 spa
->spa_livelist_delete_zthr
= NULL
;
1579 if (spa
->spa_livelist_condense_zthr
!= NULL
) {
1580 zthr_destroy(spa
->spa_livelist_condense_zthr
);
1581 spa
->spa_livelist_condense_zthr
= NULL
;
1586 * Opposite of spa_load().
1589 spa_unload(spa_t
*spa
)
1591 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1592 ASSERT(spa_state(spa
) != POOL_STATE_UNINITIALIZED
);
1594 spa_import_progress_remove(spa_guid(spa
));
1595 spa_load_note(spa
, "UNLOADING");
1597 spa_wake_waiters(spa
);
1600 * If the log space map feature is enabled and the pool is getting
1601 * exported (but not destroyed), we want to spend some time flushing
1602 * as many metaslabs as we can in an attempt to destroy log space
1603 * maps and save import time.
1605 if (spa_should_flush_logs_on_unload(spa
))
1606 spa_unload_log_sm_flush_all(spa
);
1611 spa_async_suspend(spa
);
1613 if (spa
->spa_root_vdev
) {
1614 vdev_t
*root_vdev
= spa
->spa_root_vdev
;
1615 vdev_initialize_stop_all(root_vdev
, VDEV_INITIALIZE_ACTIVE
);
1616 vdev_trim_stop_all(root_vdev
, VDEV_TRIM_ACTIVE
);
1617 vdev_autotrim_stop_all(spa
);
1618 vdev_rebuild_stop_all(spa
);
1624 if (spa
->spa_sync_on
) {
1625 txg_sync_stop(spa
->spa_dsl_pool
);
1626 spa
->spa_sync_on
= B_FALSE
;
1630 * This ensures that there is no async metaslab prefetching
1631 * while we attempt to unload the spa.
1633 if (spa
->spa_root_vdev
!= NULL
) {
1634 for (int c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++) {
1635 vdev_t
*vc
= spa
->spa_root_vdev
->vdev_child
[c
];
1636 if (vc
->vdev_mg
!= NULL
)
1637 taskq_wait(vc
->vdev_mg
->mg_taskq
);
1641 if (spa
->spa_mmp
.mmp_thread
)
1642 mmp_thread_stop(spa
);
1645 * Wait for any outstanding async I/O to complete.
1647 if (spa
->spa_async_zio_root
!= NULL
) {
1648 for (int i
= 0; i
< max_ncpus
; i
++)
1649 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1650 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1651 spa
->spa_async_zio_root
= NULL
;
1654 if (spa
->spa_vdev_removal
!= NULL
) {
1655 spa_vdev_removal_destroy(spa
->spa_vdev_removal
);
1656 spa
->spa_vdev_removal
= NULL
;
1659 spa_destroy_aux_threads(spa
);
1661 spa_condense_fini(spa
);
1663 bpobj_close(&spa
->spa_deferred_bpobj
);
1665 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1670 if (spa
->spa_root_vdev
)
1671 vdev_free(spa
->spa_root_vdev
);
1672 ASSERT(spa
->spa_root_vdev
== NULL
);
1675 * Close the dsl pool.
1677 if (spa
->spa_dsl_pool
) {
1678 dsl_pool_close(spa
->spa_dsl_pool
);
1679 spa
->spa_dsl_pool
= NULL
;
1680 spa
->spa_meta_objset
= NULL
;
1684 spa_unload_log_sm_metadata(spa
);
1687 * Drop and purge level 2 cache
1689 spa_l2cache_drop(spa
);
1691 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1692 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1693 if (spa
->spa_spares
.sav_vdevs
) {
1694 kmem_free(spa
->spa_spares
.sav_vdevs
,
1695 spa
->spa_spares
.sav_count
* sizeof (void *));
1696 spa
->spa_spares
.sav_vdevs
= NULL
;
1698 if (spa
->spa_spares
.sav_config
) {
1699 nvlist_free(spa
->spa_spares
.sav_config
);
1700 spa
->spa_spares
.sav_config
= NULL
;
1702 spa
->spa_spares
.sav_count
= 0;
1704 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1705 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1706 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1708 if (spa
->spa_l2cache
.sav_vdevs
) {
1709 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1710 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1711 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1713 if (spa
->spa_l2cache
.sav_config
) {
1714 nvlist_free(spa
->spa_l2cache
.sav_config
);
1715 spa
->spa_l2cache
.sav_config
= NULL
;
1717 spa
->spa_l2cache
.sav_count
= 0;
1719 spa
->spa_async_suspended
= 0;
1721 spa
->spa_indirect_vdevs_loaded
= B_FALSE
;
1723 if (spa
->spa_comment
!= NULL
) {
1724 spa_strfree(spa
->spa_comment
);
1725 spa
->spa_comment
= NULL
;
1727 if (spa
->spa_compatibility
!= NULL
) {
1728 spa_strfree(spa
->spa_compatibility
);
1729 spa
->spa_compatibility
= NULL
;
1732 spa_config_exit(spa
, SCL_ALL
, spa
);
1736 * Load (or re-load) the current list of vdevs describing the active spares for
1737 * this pool. When this is called, we have some form of basic information in
1738 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1739 * then re-generate a more complete list including status information.
1742 spa_load_spares(spa_t
*spa
)
1751 * zdb opens both the current state of the pool and the
1752 * checkpointed state (if present), with a different spa_t.
1754 * As spare vdevs are shared among open pools, we skip loading
1755 * them when we load the checkpointed state of the pool.
1757 if (!spa_writeable(spa
))
1761 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1764 * First, close and free any existing spare vdevs.
1766 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1767 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1769 /* Undo the call to spa_activate() below */
1770 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1771 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1772 spa_spare_remove(tvd
);
1777 if (spa
->spa_spares
.sav_vdevs
)
1778 kmem_free(spa
->spa_spares
.sav_vdevs
,
1779 spa
->spa_spares
.sav_count
* sizeof (void *));
1781 if (spa
->spa_spares
.sav_config
== NULL
)
1784 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1785 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
));
1787 spa
->spa_spares
.sav_count
= (int)nspares
;
1788 spa
->spa_spares
.sav_vdevs
= NULL
;
1794 * Construct the array of vdevs, opening them to get status in the
1795 * process. For each spare, there is potentially two different vdev_t
1796 * structures associated with it: one in the list of spares (used only
1797 * for basic validation purposes) and one in the active vdev
1798 * configuration (if it's spared in). During this phase we open and
1799 * validate each vdev on the spare list. If the vdev also exists in the
1800 * active configuration, then we also mark this vdev as an active spare.
1802 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1804 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1805 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1806 VDEV_ALLOC_SPARE
) == 0);
1809 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1811 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1812 B_FALSE
)) != NULL
) {
1813 if (!tvd
->vdev_isspare
)
1817 * We only mark the spare active if we were successfully
1818 * able to load the vdev. Otherwise, importing a pool
1819 * with a bad active spare would result in strange
1820 * behavior, because multiple pool would think the spare
1821 * is actively in use.
1823 * There is a vulnerability here to an equally bizarre
1824 * circumstance, where a dead active spare is later
1825 * brought back to life (onlined or otherwise). Given
1826 * the rarity of this scenario, and the extra complexity
1827 * it adds, we ignore the possibility.
1829 if (!vdev_is_dead(tvd
))
1830 spa_spare_activate(tvd
);
1834 vd
->vdev_aux
= &spa
->spa_spares
;
1836 if (vdev_open(vd
) != 0)
1839 if (vdev_validate_aux(vd
) == 0)
1844 * Recompute the stashed list of spares, with status information
1847 fnvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
);
1849 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1851 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1852 spares
[i
] = vdev_config_generate(spa
,
1853 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1854 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1855 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
1856 spa
->spa_spares
.sav_count
);
1857 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1858 nvlist_free(spares
[i
]);
1859 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1863 * Load (or re-load) the current list of vdevs describing the active l2cache for
1864 * this pool. When this is called, we have some form of basic information in
1865 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1866 * then re-generate a more complete list including status information.
1867 * Devices which are already active have their details maintained, and are
1871 spa_load_l2cache(spa_t
*spa
)
1873 nvlist_t
**l2cache
= NULL
;
1875 int i
, j
, oldnvdevs
;
1877 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1878 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1882 * zdb opens both the current state of the pool and the
1883 * checkpointed state (if present), with a different spa_t.
1885 * As L2 caches are part of the ARC which is shared among open
1886 * pools, we skip loading them when we load the checkpointed
1887 * state of the pool.
1889 if (!spa_writeable(spa
))
1893 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1895 oldvdevs
= sav
->sav_vdevs
;
1896 oldnvdevs
= sav
->sav_count
;
1897 sav
->sav_vdevs
= NULL
;
1900 if (sav
->sav_config
== NULL
) {
1906 VERIFY0(nvlist_lookup_nvlist_array(sav
->sav_config
,
1907 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
));
1908 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1911 * Process new nvlist of vdevs.
1913 for (i
= 0; i
< nl2cache
; i
++) {
1914 guid
= fnvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
);
1917 for (j
= 0; j
< oldnvdevs
; j
++) {
1919 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1921 * Retain previous vdev for add/remove ops.
1929 if (newvdevs
[i
] == NULL
) {
1933 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1934 VDEV_ALLOC_L2CACHE
) == 0);
1939 * Commit this vdev as an l2cache device,
1940 * even if it fails to open.
1942 spa_l2cache_add(vd
);
1947 spa_l2cache_activate(vd
);
1949 if (vdev_open(vd
) != 0)
1952 (void) vdev_validate_aux(vd
);
1954 if (!vdev_is_dead(vd
))
1955 l2arc_add_vdev(spa
, vd
);
1958 * Upon cache device addition to a pool or pool
1959 * creation with a cache device or if the header
1960 * of the device is invalid we issue an async
1961 * TRIM command for the whole device which will
1962 * execute if l2arc_trim_ahead > 0.
1964 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
1968 sav
->sav_vdevs
= newvdevs
;
1969 sav
->sav_count
= (int)nl2cache
;
1972 * Recompute the stashed list of l2cache devices, with status
1973 * information this time.
1975 fnvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
);
1977 if (sav
->sav_count
> 0)
1978 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *),
1980 for (i
= 0; i
< sav
->sav_count
; i
++)
1981 l2cache
[i
] = vdev_config_generate(spa
,
1982 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1983 fnvlist_add_nvlist_array(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1984 (const nvlist_t
* const *)l2cache
, sav
->sav_count
);
1988 * Purge vdevs that were dropped
1990 for (i
= 0; i
< oldnvdevs
; i
++) {
1995 ASSERT(vd
->vdev_isl2cache
);
1997 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1998 pool
!= 0ULL && l2arc_vdev_present(vd
))
1999 l2arc_remove_vdev(vd
);
2000 vdev_clear_stats(vd
);
2006 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
2008 for (i
= 0; i
< sav
->sav_count
; i
++)
2009 nvlist_free(l2cache
[i
]);
2011 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
2015 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
2018 char *packed
= NULL
;
2023 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
2027 nvsize
= *(uint64_t *)db
->db_data
;
2028 dmu_buf_rele(db
, FTAG
);
2030 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
2031 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
2034 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
2035 vmem_free(packed
, nvsize
);
2041 * Concrete top-level vdevs that are not missing and are not logs. At every
2042 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
2045 spa_healthy_core_tvds(spa_t
*spa
)
2047 vdev_t
*rvd
= spa
->spa_root_vdev
;
2050 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
2051 vdev_t
*vd
= rvd
->vdev_child
[i
];
2054 if (vdev_is_concrete(vd
) && !vdev_is_dead(vd
))
2062 * Checks to see if the given vdev could not be opened, in which case we post a
2063 * sysevent to notify the autoreplace code that the device has been removed.
2066 spa_check_removed(vdev_t
*vd
)
2068 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
2069 spa_check_removed(vd
->vdev_child
[c
]);
2071 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
2072 vdev_is_concrete(vd
)) {
2073 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
2074 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
2079 spa_check_for_missing_logs(spa_t
*spa
)
2081 vdev_t
*rvd
= spa
->spa_root_vdev
;
2084 * If we're doing a normal import, then build up any additional
2085 * diagnostic information about missing log devices.
2086 * We'll pass this up to the user for further processing.
2088 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
2089 nvlist_t
**child
, *nv
;
2092 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
2094 nv
= fnvlist_alloc();
2096 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2097 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2100 * We consider a device as missing only if it failed
2101 * to open (i.e. offline or faulted is not considered
2104 if (tvd
->vdev_islog
&&
2105 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2106 child
[idx
++] = vdev_config_generate(spa
, tvd
,
2107 B_FALSE
, VDEV_CONFIG_MISSING
);
2112 fnvlist_add_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
2113 (const nvlist_t
* const *)child
, idx
);
2114 fnvlist_add_nvlist(spa
->spa_load_info
,
2115 ZPOOL_CONFIG_MISSING_DEVICES
, nv
);
2117 for (uint64_t i
= 0; i
< idx
; i
++)
2118 nvlist_free(child
[i
]);
2121 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
2124 spa_load_failed(spa
, "some log devices are missing");
2125 vdev_dbgmsg_print_tree(rvd
, 2);
2126 return (SET_ERROR(ENXIO
));
2129 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2130 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2132 if (tvd
->vdev_islog
&&
2133 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2134 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2135 spa_load_note(spa
, "some log devices are "
2136 "missing, ZIL is dropped.");
2137 vdev_dbgmsg_print_tree(rvd
, 2);
2147 * Check for missing log devices
2150 spa_check_logs(spa_t
*spa
)
2152 boolean_t rv
= B_FALSE
;
2153 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2155 switch (spa
->spa_log_state
) {
2158 case SPA_LOG_MISSING
:
2159 /* need to recheck in case slog has been restored */
2160 case SPA_LOG_UNKNOWN
:
2161 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
2162 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
2164 spa_set_log_state(spa
, SPA_LOG_MISSING
);
2171 * Passivate any log vdevs (note, does not apply to embedded log metaslabs).
2174 spa_passivate_log(spa_t
*spa
)
2176 vdev_t
*rvd
= spa
->spa_root_vdev
;
2177 boolean_t slog_found
= B_FALSE
;
2179 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2181 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2182 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2184 if (tvd
->vdev_islog
) {
2185 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
2186 metaslab_group_passivate(tvd
->vdev_mg
);
2187 slog_found
= B_TRUE
;
2191 return (slog_found
);
2195 * Activate any log vdevs (note, does not apply to embedded log metaslabs).
2198 spa_activate_log(spa_t
*spa
)
2200 vdev_t
*rvd
= spa
->spa_root_vdev
;
2202 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2204 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2205 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2207 if (tvd
->vdev_islog
) {
2208 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
2209 metaslab_group_activate(tvd
->vdev_mg
);
2215 spa_reset_logs(spa_t
*spa
)
2219 error
= dmu_objset_find(spa_name(spa
), zil_reset
,
2220 NULL
, DS_FIND_CHILDREN
);
2223 * We successfully offlined the log device, sync out the
2224 * current txg so that the "stubby" block can be removed
2227 txg_wait_synced(spa
->spa_dsl_pool
, 0);
2233 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
2235 for (int i
= 0; i
< sav
->sav_count
; i
++)
2236 spa_check_removed(sav
->sav_vdevs
[i
]);
2240 spa_claim_notify(zio_t
*zio
)
2242 spa_t
*spa
= zio
->io_spa
;
2247 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
2248 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
2249 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
2250 mutex_exit(&spa
->spa_props_lock
);
2253 typedef struct spa_load_error
{
2254 boolean_t sle_verify_data
;
2255 uint64_t sle_meta_count
;
2256 uint64_t sle_data_count
;
2260 spa_load_verify_done(zio_t
*zio
)
2262 blkptr_t
*bp
= zio
->io_bp
;
2263 spa_load_error_t
*sle
= zio
->io_private
;
2264 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2265 int error
= zio
->io_error
;
2266 spa_t
*spa
= zio
->io_spa
;
2268 abd_free(zio
->io_abd
);
2270 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2271 type
!= DMU_OT_INTENT_LOG
)
2272 atomic_inc_64(&sle
->sle_meta_count
);
2274 atomic_inc_64(&sle
->sle_data_count
);
2277 mutex_enter(&spa
->spa_scrub_lock
);
2278 spa
->spa_load_verify_bytes
-= BP_GET_PSIZE(bp
);
2279 cv_broadcast(&spa
->spa_scrub_io_cv
);
2280 mutex_exit(&spa
->spa_scrub_lock
);
2284 * Maximum number of inflight bytes is the log2 fraction of the arc size.
2285 * By default, we set it to 1/16th of the arc.
2287 static int spa_load_verify_shift
= 4;
2288 static int spa_load_verify_metadata
= B_TRUE
;
2289 static int spa_load_verify_data
= B_TRUE
;
2292 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2293 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2296 spa_load_error_t
*sle
= rio
->io_private
;
2298 (void) zilog
, (void) dnp
;
2300 if (zb
->zb_level
== ZB_DNODE_LEVEL
|| BP_IS_HOLE(bp
) ||
2301 BP_IS_EMBEDDED(bp
) || BP_IS_REDACTED(bp
))
2304 * Note: normally this routine will not be called if
2305 * spa_load_verify_metadata is not set. However, it may be useful
2306 * to manually set the flag after the traversal has begun.
2308 if (!spa_load_verify_metadata
)
2310 if (!BP_IS_METADATA(bp
) &&
2311 (!spa_load_verify_data
|| !sle
->sle_verify_data
))
2314 uint64_t maxinflight_bytes
=
2315 arc_target_bytes() >> spa_load_verify_shift
;
2316 size_t size
= BP_GET_PSIZE(bp
);
2318 mutex_enter(&spa
->spa_scrub_lock
);
2319 while (spa
->spa_load_verify_bytes
>= maxinflight_bytes
)
2320 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2321 spa
->spa_load_verify_bytes
+= size
;
2322 mutex_exit(&spa
->spa_scrub_lock
);
2324 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2325 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2326 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2327 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2332 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2334 (void) dp
, (void) arg
;
2336 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2337 return (SET_ERROR(ENAMETOOLONG
));
2343 spa_load_verify(spa_t
*spa
)
2346 spa_load_error_t sle
= { 0 };
2347 zpool_load_policy_t policy
;
2348 boolean_t verify_ok
= B_FALSE
;
2351 zpool_get_load_policy(spa
->spa_config
, &policy
);
2353 if (policy
.zlp_rewind
& ZPOOL_NEVER_REWIND
||
2354 policy
.zlp_maxmeta
== UINT64_MAX
)
2357 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2358 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2359 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2361 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2366 * Verify data only if we are rewinding or error limit was set.
2367 * Otherwise nothing except dbgmsg care about it to waste time.
2369 sle
.sle_verify_data
= (policy
.zlp_rewind
& ZPOOL_REWIND_MASK
) ||
2370 (policy
.zlp_maxdata
< UINT64_MAX
);
2372 rio
= zio_root(spa
, NULL
, &sle
,
2373 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2375 if (spa_load_verify_metadata
) {
2376 if (spa
->spa_extreme_rewind
) {
2377 spa_load_note(spa
, "performing a complete scan of the "
2378 "pool since extreme rewind is on. This may take "
2379 "a very long time.\n (spa_load_verify_data=%u, "
2380 "spa_load_verify_metadata=%u)",
2381 spa_load_verify_data
, spa_load_verify_metadata
);
2384 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2385 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2386 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2389 (void) zio_wait(rio
);
2390 ASSERT0(spa
->spa_load_verify_bytes
);
2392 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2393 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2395 if (sle
.sle_meta_count
!= 0 || sle
.sle_data_count
!= 0) {
2396 spa_load_note(spa
, "spa_load_verify found %llu metadata errors "
2397 "and %llu data errors", (u_longlong_t
)sle
.sle_meta_count
,
2398 (u_longlong_t
)sle
.sle_data_count
);
2401 if (spa_load_verify_dryrun
||
2402 (!error
&& sle
.sle_meta_count
<= policy
.zlp_maxmeta
&&
2403 sle
.sle_data_count
<= policy
.zlp_maxdata
)) {
2407 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2408 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2410 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2411 fnvlist_add_uint64(spa
->spa_load_info
, ZPOOL_CONFIG_LOAD_TIME
,
2412 spa
->spa_load_txg_ts
);
2413 fnvlist_add_int64(spa
->spa_load_info
, ZPOOL_CONFIG_REWIND_TIME
,
2415 fnvlist_add_uint64(spa
->spa_load_info
,
2416 ZPOOL_CONFIG_LOAD_META_ERRORS
, sle
.sle_meta_count
);
2417 fnvlist_add_uint64(spa
->spa_load_info
,
2418 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
);
2420 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2423 if (spa_load_verify_dryrun
)
2427 if (error
!= ENXIO
&& error
!= EIO
)
2428 error
= SET_ERROR(EIO
);
2432 return (verify_ok
? 0 : EIO
);
2436 * Find a value in the pool props object.
2439 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2441 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2442 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2446 * Find a value in the pool directory object.
2449 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
, boolean_t log_enoent
)
2451 int error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2452 name
, sizeof (uint64_t), 1, val
);
2454 if (error
!= 0 && (error
!= ENOENT
|| log_enoent
)) {
2455 spa_load_failed(spa
, "couldn't get '%s' value in MOS directory "
2456 "[error=%d]", name
, error
);
2463 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2465 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2466 return (SET_ERROR(err
));
2470 spa_livelist_delete_check(spa_t
*spa
)
2472 return (spa
->spa_livelists_to_delete
!= 0);
2476 spa_livelist_delete_cb_check(void *arg
, zthr_t
*z
)
2480 return (spa_livelist_delete_check(spa
));
2484 delete_blkptr_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
2487 zio_free(spa
, tx
->tx_txg
, bp
);
2488 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
, DD_USED_HEAD
,
2489 -bp_get_dsize_sync(spa
, bp
),
2490 -BP_GET_PSIZE(bp
), -BP_GET_UCSIZE(bp
), tx
);
2495 dsl_get_next_livelist_obj(objset_t
*os
, uint64_t zap_obj
, uint64_t *llp
)
2500 zap_cursor_init(&zc
, os
, zap_obj
);
2501 err
= zap_cursor_retrieve(&zc
, &za
);
2502 zap_cursor_fini(&zc
);
2504 *llp
= za
.za_first_integer
;
2509 * Components of livelist deletion that must be performed in syncing
2510 * context: freeing block pointers and updating the pool-wide data
2511 * structures to indicate how much work is left to do
2513 typedef struct sublist_delete_arg
{
2518 } sublist_delete_arg_t
;
2521 sublist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2523 sublist_delete_arg_t
*sda
= arg
;
2524 spa_t
*spa
= sda
->spa
;
2525 dsl_deadlist_t
*ll
= sda
->ll
;
2526 uint64_t key
= sda
->key
;
2527 bplist_t
*to_free
= sda
->to_free
;
2529 bplist_iterate(to_free
, delete_blkptr_cb
, spa
, tx
);
2530 dsl_deadlist_remove_entry(ll
, key
, tx
);
2533 typedef struct livelist_delete_arg
{
2537 } livelist_delete_arg_t
;
2540 livelist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2542 livelist_delete_arg_t
*lda
= arg
;
2543 spa_t
*spa
= lda
->spa
;
2544 uint64_t ll_obj
= lda
->ll_obj
;
2545 uint64_t zap_obj
= lda
->zap_obj
;
2546 objset_t
*mos
= spa
->spa_meta_objset
;
2549 /* free the livelist and decrement the feature count */
2550 VERIFY0(zap_remove_int(mos
, zap_obj
, ll_obj
, tx
));
2551 dsl_deadlist_free(mos
, ll_obj
, tx
);
2552 spa_feature_decr(spa
, SPA_FEATURE_LIVELIST
, tx
);
2553 VERIFY0(zap_count(mos
, zap_obj
, &count
));
2555 /* no more livelists to delete */
2556 VERIFY0(zap_remove(mos
, DMU_POOL_DIRECTORY_OBJECT
,
2557 DMU_POOL_DELETED_CLONES
, tx
));
2558 VERIFY0(zap_destroy(mos
, zap_obj
, tx
));
2559 spa
->spa_livelists_to_delete
= 0;
2560 spa_notify_waiters(spa
);
2565 * Load in the value for the livelist to be removed and open it. Then,
2566 * load its first sublist and determine which block pointers should actually
2567 * be freed. Then, call a synctask which performs the actual frees and updates
2568 * the pool-wide livelist data.
2571 spa_livelist_delete_cb(void *arg
, zthr_t
*z
)
2574 uint64_t ll_obj
= 0, count
;
2575 objset_t
*mos
= spa
->spa_meta_objset
;
2576 uint64_t zap_obj
= spa
->spa_livelists_to_delete
;
2578 * Determine the next livelist to delete. This function should only
2579 * be called if there is at least one deleted clone.
2581 VERIFY0(dsl_get_next_livelist_obj(mos
, zap_obj
, &ll_obj
));
2582 VERIFY0(zap_count(mos
, ll_obj
, &count
));
2585 dsl_deadlist_entry_t
*dle
;
2587 ll
= kmem_zalloc(sizeof (dsl_deadlist_t
), KM_SLEEP
);
2588 dsl_deadlist_open(ll
, mos
, ll_obj
);
2589 dle
= dsl_deadlist_first(ll
);
2590 ASSERT3P(dle
, !=, NULL
);
2591 bplist_create(&to_free
);
2592 int err
= dsl_process_sub_livelist(&dle
->dle_bpobj
, &to_free
,
2595 sublist_delete_arg_t sync_arg
= {
2598 .key
= dle
->dle_mintxg
,
2601 zfs_dbgmsg("deleting sublist (id %llu) from"
2602 " livelist %llu, %lld remaining",
2603 (u_longlong_t
)dle
->dle_bpobj
.bpo_object
,
2604 (u_longlong_t
)ll_obj
, (longlong_t
)count
- 1);
2605 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
,
2606 sublist_delete_sync
, &sync_arg
, 0,
2607 ZFS_SPACE_CHECK_DESTROY
));
2609 VERIFY3U(err
, ==, EINTR
);
2611 bplist_clear(&to_free
);
2612 bplist_destroy(&to_free
);
2613 dsl_deadlist_close(ll
);
2614 kmem_free(ll
, sizeof (dsl_deadlist_t
));
2616 livelist_delete_arg_t sync_arg
= {
2621 zfs_dbgmsg("deletion of livelist %llu completed",
2622 (u_longlong_t
)ll_obj
);
2623 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
, livelist_delete_sync
,
2624 &sync_arg
, 0, ZFS_SPACE_CHECK_DESTROY
));
2629 spa_start_livelist_destroy_thread(spa_t
*spa
)
2631 ASSERT3P(spa
->spa_livelist_delete_zthr
, ==, NULL
);
2632 spa
->spa_livelist_delete_zthr
=
2633 zthr_create("z_livelist_destroy",
2634 spa_livelist_delete_cb_check
, spa_livelist_delete_cb
, spa
,
2638 typedef struct livelist_new_arg
{
2641 } livelist_new_arg_t
;
2644 livelist_track_new_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
2648 livelist_new_arg_t
*lna
= arg
;
2650 bplist_append(lna
->frees
, bp
);
2652 bplist_append(lna
->allocs
, bp
);
2653 zfs_livelist_condense_new_alloc
++;
2658 typedef struct livelist_condense_arg
{
2661 uint64_t first_size
;
2663 } livelist_condense_arg_t
;
2666 spa_livelist_condense_sync(void *arg
, dmu_tx_t
*tx
)
2668 livelist_condense_arg_t
*lca
= arg
;
2669 spa_t
*spa
= lca
->spa
;
2671 dsl_dataset_t
*ds
= spa
->spa_to_condense
.ds
;
2673 /* Have we been cancelled? */
2674 if (spa
->spa_to_condense
.cancelled
) {
2675 zfs_livelist_condense_sync_cancel
++;
2679 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
2680 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
2681 dsl_deadlist_t
*ll
= &ds
->ds_dir
->dd_livelist
;
2684 * It's possible that the livelist was changed while the zthr was
2685 * running. Therefore, we need to check for new blkptrs in the two
2686 * entries being condensed and continue to track them in the livelist.
2687 * Because of the way we handle remapped blkptrs (see dbuf_remap_impl),
2688 * it's possible that the newly added blkptrs are FREEs or ALLOCs so
2689 * we need to sort them into two different bplists.
2691 uint64_t first_obj
= first
->dle_bpobj
.bpo_object
;
2692 uint64_t next_obj
= next
->dle_bpobj
.bpo_object
;
2693 uint64_t cur_first_size
= first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
2694 uint64_t cur_next_size
= next
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
2696 bplist_create(&new_frees
);
2697 livelist_new_arg_t new_bps
= {
2698 .allocs
= &lca
->to_keep
,
2699 .frees
= &new_frees
,
2702 if (cur_first_size
> lca
->first_size
) {
2703 VERIFY0(livelist_bpobj_iterate_from_nofree(&first
->dle_bpobj
,
2704 livelist_track_new_cb
, &new_bps
, lca
->first_size
));
2706 if (cur_next_size
> lca
->next_size
) {
2707 VERIFY0(livelist_bpobj_iterate_from_nofree(&next
->dle_bpobj
,
2708 livelist_track_new_cb
, &new_bps
, lca
->next_size
));
2711 dsl_deadlist_clear_entry(first
, ll
, tx
);
2712 ASSERT(bpobj_is_empty(&first
->dle_bpobj
));
2713 dsl_deadlist_remove_entry(ll
, next
->dle_mintxg
, tx
);
2715 bplist_iterate(&lca
->to_keep
, dsl_deadlist_insert_alloc_cb
, ll
, tx
);
2716 bplist_iterate(&new_frees
, dsl_deadlist_insert_free_cb
, ll
, tx
);
2717 bplist_destroy(&new_frees
);
2719 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
2720 dsl_dataset_name(ds
, dsname
);
2721 zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu "
2722 "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu "
2723 "(%llu blkptrs)", (u_longlong_t
)tx
->tx_txg
, dsname
,
2724 (u_longlong_t
)ds
->ds_object
, (u_longlong_t
)first_obj
,
2725 (u_longlong_t
)cur_first_size
, (u_longlong_t
)next_obj
,
2726 (u_longlong_t
)cur_next_size
,
2727 (u_longlong_t
)first
->dle_bpobj
.bpo_object
,
2728 (u_longlong_t
)first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
);
2730 dmu_buf_rele(ds
->ds_dbuf
, spa
);
2731 spa
->spa_to_condense
.ds
= NULL
;
2732 bplist_clear(&lca
->to_keep
);
2733 bplist_destroy(&lca
->to_keep
);
2734 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
2735 spa
->spa_to_condense
.syncing
= B_FALSE
;
2739 spa_livelist_condense_cb(void *arg
, zthr_t
*t
)
2741 while (zfs_livelist_condense_zthr_pause
&&
2742 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
2746 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
2747 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
2748 uint64_t first_size
, next_size
;
2750 livelist_condense_arg_t
*lca
=
2751 kmem_alloc(sizeof (livelist_condense_arg_t
), KM_SLEEP
);
2752 bplist_create(&lca
->to_keep
);
2755 * Process the livelists (matching FREEs and ALLOCs) in open context
2756 * so we have minimal work in syncing context to condense.
2758 * We save bpobj sizes (first_size and next_size) to use later in
2759 * syncing context to determine if entries were added to these sublists
2760 * while in open context. This is possible because the clone is still
2761 * active and open for normal writes and we want to make sure the new,
2762 * unprocessed blockpointers are inserted into the livelist normally.
2764 * Note that dsl_process_sub_livelist() both stores the size number of
2765 * blockpointers and iterates over them while the bpobj's lock held, so
2766 * the sizes returned to us are consistent which what was actually
2769 int err
= dsl_process_sub_livelist(&first
->dle_bpobj
, &lca
->to_keep
, t
,
2772 err
= dsl_process_sub_livelist(&next
->dle_bpobj
, &lca
->to_keep
,
2776 while (zfs_livelist_condense_sync_pause
&&
2777 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
2780 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
2781 dmu_tx_mark_netfree(tx
);
2782 dmu_tx_hold_space(tx
, 1);
2783 err
= dmu_tx_assign(tx
, TXG_NOWAIT
| TXG_NOTHROTTLE
);
2786 * Prevent the condense zthr restarting before
2787 * the synctask completes.
2789 spa
->spa_to_condense
.syncing
= B_TRUE
;
2791 lca
->first_size
= first_size
;
2792 lca
->next_size
= next_size
;
2793 dsl_sync_task_nowait(spa_get_dsl(spa
),
2794 spa_livelist_condense_sync
, lca
, tx
);
2800 * Condensing can not continue: either it was externally stopped or
2801 * we were unable to assign to a tx because the pool has run out of
2802 * space. In the second case, we'll just end up trying to condense
2803 * again in a later txg.
2806 bplist_clear(&lca
->to_keep
);
2807 bplist_destroy(&lca
->to_keep
);
2808 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
2809 dmu_buf_rele(spa
->spa_to_condense
.ds
->ds_dbuf
, spa
);
2810 spa
->spa_to_condense
.ds
= NULL
;
2812 zfs_livelist_condense_zthr_cancel
++;
2816 * Check that there is something to condense but that a condense is not
2817 * already in progress and that condensing has not been cancelled.
2820 spa_livelist_condense_cb_check(void *arg
, zthr_t
*z
)
2824 if ((spa
->spa_to_condense
.ds
!= NULL
) &&
2825 (spa
->spa_to_condense
.syncing
== B_FALSE
) &&
2826 (spa
->spa_to_condense
.cancelled
== B_FALSE
)) {
2833 spa_start_livelist_condensing_thread(spa_t
*spa
)
2835 spa
->spa_to_condense
.ds
= NULL
;
2836 spa
->spa_to_condense
.first
= NULL
;
2837 spa
->spa_to_condense
.next
= NULL
;
2838 spa
->spa_to_condense
.syncing
= B_FALSE
;
2839 spa
->spa_to_condense
.cancelled
= B_FALSE
;
2841 ASSERT3P(spa
->spa_livelist_condense_zthr
, ==, NULL
);
2842 spa
->spa_livelist_condense_zthr
=
2843 zthr_create("z_livelist_condense",
2844 spa_livelist_condense_cb_check
,
2845 spa_livelist_condense_cb
, spa
, minclsyspri
);
2849 spa_spawn_aux_threads(spa_t
*spa
)
2851 ASSERT(spa_writeable(spa
));
2853 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2855 spa_start_indirect_condensing_thread(spa
);
2856 spa_start_livelist_destroy_thread(spa
);
2857 spa_start_livelist_condensing_thread(spa
);
2859 ASSERT3P(spa
->spa_checkpoint_discard_zthr
, ==, NULL
);
2860 spa
->spa_checkpoint_discard_zthr
=
2861 zthr_create("z_checkpoint_discard",
2862 spa_checkpoint_discard_thread_check
,
2863 spa_checkpoint_discard_thread
, spa
, minclsyspri
);
2867 * Fix up config after a partly-completed split. This is done with the
2868 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2869 * pool have that entry in their config, but only the splitting one contains
2870 * a list of all the guids of the vdevs that are being split off.
2872 * This function determines what to do with that list: either rejoin
2873 * all the disks to the pool, or complete the splitting process. To attempt
2874 * the rejoin, each disk that is offlined is marked online again, and
2875 * we do a reopen() call. If the vdev label for every disk that was
2876 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2877 * then we call vdev_split() on each disk, and complete the split.
2879 * Otherwise we leave the config alone, with all the vdevs in place in
2880 * the original pool.
2883 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2890 boolean_t attempt_reopen
;
2892 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2895 /* check that the config is complete */
2896 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2897 &glist
, &gcount
) != 0)
2900 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2902 /* attempt to online all the vdevs & validate */
2903 attempt_reopen
= B_TRUE
;
2904 for (i
= 0; i
< gcount
; i
++) {
2905 if (glist
[i
] == 0) /* vdev is hole */
2908 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2909 if (vd
[i
] == NULL
) {
2911 * Don't bother attempting to reopen the disks;
2912 * just do the split.
2914 attempt_reopen
= B_FALSE
;
2916 /* attempt to re-online it */
2917 vd
[i
]->vdev_offline
= B_FALSE
;
2921 if (attempt_reopen
) {
2922 vdev_reopen(spa
->spa_root_vdev
);
2924 /* check each device to see what state it's in */
2925 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2926 if (vd
[i
] != NULL
&&
2927 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2934 * If every disk has been moved to the new pool, or if we never
2935 * even attempted to look at them, then we split them off for
2938 if (!attempt_reopen
|| gcount
== extracted
) {
2939 for (i
= 0; i
< gcount
; i
++)
2942 vdev_reopen(spa
->spa_root_vdev
);
2945 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2949 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
)
2951 char *ereport
= FM_EREPORT_ZFS_POOL
;
2954 spa
->spa_load_state
= state
;
2955 (void) spa_import_progress_set_state(spa_guid(spa
),
2956 spa_load_state(spa
));
2958 gethrestime(&spa
->spa_loaded_ts
);
2959 error
= spa_load_impl(spa
, type
, &ereport
);
2962 * Don't count references from objsets that are already closed
2963 * and are making their way through the eviction process.
2965 spa_evicting_os_wait(spa
);
2966 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
2968 if (error
!= EEXIST
) {
2969 spa
->spa_loaded_ts
.tv_sec
= 0;
2970 spa
->spa_loaded_ts
.tv_nsec
= 0;
2972 if (error
!= EBADF
) {
2973 (void) zfs_ereport_post(ereport
, spa
,
2974 NULL
, NULL
, NULL
, 0);
2977 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2980 (void) spa_import_progress_set_state(spa_guid(spa
),
2981 spa_load_state(spa
));
2988 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2989 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2990 * spa's per-vdev ZAP list.
2993 vdev_count_verify_zaps(vdev_t
*vd
)
2995 spa_t
*spa
= vd
->vdev_spa
;
2998 if (vd
->vdev_top_zap
!= 0) {
3000 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3001 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
3003 if (vd
->vdev_leaf_zap
!= 0) {
3005 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3006 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
3009 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3010 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
3016 #define vdev_count_verify_zaps(vd) ((void) sizeof (vd), 0)
3020 * Determine whether the activity check is required.
3023 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
3027 uint64_t hostid
= 0;
3028 uint64_t tryconfig_txg
= 0;
3029 uint64_t tryconfig_timestamp
= 0;
3030 uint16_t tryconfig_mmp_seq
= 0;
3033 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3034 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
3035 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
3037 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
3038 &tryconfig_timestamp
);
3039 (void) nvlist_lookup_uint16(nvinfo
, ZPOOL_CONFIG_MMP_SEQ
,
3040 &tryconfig_mmp_seq
);
3043 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
3046 * Disable the MMP activity check - This is used by zdb which
3047 * is intended to be used on potentially active pools.
3049 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
3053 * Skip the activity check when the MMP feature is disabled.
3055 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
3059 * If the tryconfig_ values are nonzero, they are the results of an
3060 * earlier tryimport. If they all match the uberblock we just found,
3061 * then the pool has not changed and we return false so we do not test
3064 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
3065 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
&&
3066 tryconfig_mmp_seq
&& tryconfig_mmp_seq
==
3067 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0))
3071 * Allow the activity check to be skipped when importing the pool
3072 * on the same host which last imported it. Since the hostid from
3073 * configuration may be stale use the one read from the label.
3075 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
3076 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
3078 if (hostid
== spa_get_hostid(spa
))
3082 * Skip the activity test when the pool was cleanly exported.
3084 if (state
!= POOL_STATE_ACTIVE
)
3091 * Nanoseconds the activity check must watch for changes on-disk.
3094 spa_activity_check_duration(spa_t
*spa
, uberblock_t
*ub
)
3096 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
3097 uint64_t multihost_interval
= MSEC2NSEC(
3098 MMP_INTERVAL_OK(zfs_multihost_interval
));
3099 uint64_t import_delay
= MAX(NANOSEC
, import_intervals
*
3100 multihost_interval
);
3103 * Local tunables determine a minimum duration except for the case
3104 * where we know when the remote host will suspend the pool if MMP
3105 * writes do not land.
3107 * See Big Theory comment at the top of mmp.c for the reasoning behind
3108 * these cases and times.
3111 ASSERT(MMP_IMPORT_SAFETY_FACTOR
>= 100);
3113 if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3114 MMP_FAIL_INT(ub
) > 0) {
3116 /* MMP on remote host will suspend pool after failed writes */
3117 import_delay
= MMP_FAIL_INT(ub
) * MSEC2NSEC(MMP_INTERVAL(ub
)) *
3118 MMP_IMPORT_SAFETY_FACTOR
/ 100;
3120 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3121 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3122 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3123 (u_longlong_t
)MMP_FAIL_INT(ub
),
3124 (u_longlong_t
)MMP_INTERVAL(ub
),
3125 (u_longlong_t
)import_intervals
);
3127 } else if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3128 MMP_FAIL_INT(ub
) == 0) {
3130 /* MMP on remote host will never suspend pool */
3131 import_delay
= MAX(import_delay
, (MSEC2NSEC(MMP_INTERVAL(ub
)) +
3132 ub
->ub_mmp_delay
) * import_intervals
);
3134 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3135 "mmp_interval=%llu ub_mmp_delay=%llu "
3136 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3137 (u_longlong_t
)MMP_INTERVAL(ub
),
3138 (u_longlong_t
)ub
->ub_mmp_delay
,
3139 (u_longlong_t
)import_intervals
);
3141 } else if (MMP_VALID(ub
)) {
3143 * zfs-0.7 compatibility case
3146 import_delay
= MAX(import_delay
, (multihost_interval
+
3147 ub
->ub_mmp_delay
) * import_intervals
);
3149 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3150 "import_intervals=%llu leaves=%u",
3151 (u_longlong_t
)import_delay
,
3152 (u_longlong_t
)ub
->ub_mmp_delay
,
3153 (u_longlong_t
)import_intervals
,
3154 vdev_count_leaves(spa
));
3156 /* Using local tunings is the only reasonable option */
3157 zfs_dbgmsg("pool last imported on non-MMP aware "
3158 "host using import_delay=%llu multihost_interval=%llu "
3159 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3160 (u_longlong_t
)multihost_interval
,
3161 (u_longlong_t
)import_intervals
);
3164 return (import_delay
);
3168 * Perform the import activity check. If the user canceled the import or
3169 * we detected activity then fail.
3172 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
3174 uint64_t txg
= ub
->ub_txg
;
3175 uint64_t timestamp
= ub
->ub_timestamp
;
3176 uint64_t mmp_config
= ub
->ub_mmp_config
;
3177 uint16_t mmp_seq
= MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0;
3178 uint64_t import_delay
;
3179 hrtime_t import_expire
;
3180 nvlist_t
*mmp_label
= NULL
;
3181 vdev_t
*rvd
= spa
->spa_root_vdev
;
3186 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
3187 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3191 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3192 * during the earlier tryimport. If the txg recorded there is 0 then
3193 * the pool is known to be active on another host.
3195 * Otherwise, the pool might be in use on another host. Check for
3196 * changes in the uberblocks on disk if necessary.
3198 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3199 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
3200 ZPOOL_CONFIG_LOAD_INFO
);
3202 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
3203 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
3204 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3205 error
= SET_ERROR(EREMOTEIO
);
3210 import_delay
= spa_activity_check_duration(spa
, ub
);
3212 /* Add a small random factor in case of simultaneous imports (0-25%) */
3213 import_delay
+= import_delay
* random_in_range(250) / 1000;
3215 import_expire
= gethrtime() + import_delay
;
3217 while (gethrtime() < import_expire
) {
3218 (void) spa_import_progress_set_mmp_check(spa_guid(spa
),
3219 NSEC2SEC(import_expire
- gethrtime()));
3221 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3223 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
||
3224 mmp_seq
!= (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0)) {
3225 zfs_dbgmsg("multihost activity detected "
3226 "txg %llu ub_txg %llu "
3227 "timestamp %llu ub_timestamp %llu "
3228 "mmp_config %#llx ub_mmp_config %#llx",
3229 (u_longlong_t
)txg
, (u_longlong_t
)ub
->ub_txg
,
3230 (u_longlong_t
)timestamp
,
3231 (u_longlong_t
)ub
->ub_timestamp
,
3232 (u_longlong_t
)mmp_config
,
3233 (u_longlong_t
)ub
->ub_mmp_config
);
3235 error
= SET_ERROR(EREMOTEIO
);
3240 nvlist_free(mmp_label
);
3244 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
3246 error
= SET_ERROR(EINTR
);
3254 mutex_destroy(&mtx
);
3258 * If the pool is determined to be active store the status in the
3259 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3260 * available from configuration read from disk store them as well.
3261 * This allows 'zpool import' to generate a more useful message.
3263 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3264 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3265 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3267 if (error
== EREMOTEIO
) {
3268 char *hostname
= "<unknown>";
3269 uint64_t hostid
= 0;
3272 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
3273 hostname
= fnvlist_lookup_string(mmp_label
,
3274 ZPOOL_CONFIG_HOSTNAME
);
3275 fnvlist_add_string(spa
->spa_load_info
,
3276 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
3279 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
3280 hostid
= fnvlist_lookup_uint64(mmp_label
,
3281 ZPOOL_CONFIG_HOSTID
);
3282 fnvlist_add_uint64(spa
->spa_load_info
,
3283 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
3287 fnvlist_add_uint64(spa
->spa_load_info
,
3288 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
3289 fnvlist_add_uint64(spa
->spa_load_info
,
3290 ZPOOL_CONFIG_MMP_TXG
, 0);
3292 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
3296 nvlist_free(mmp_label
);
3302 spa_verify_host(spa_t
*spa
, nvlist_t
*mos_config
)
3306 uint64_t myhostid
= 0;
3308 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
3309 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
3310 hostname
= fnvlist_lookup_string(mos_config
,
3311 ZPOOL_CONFIG_HOSTNAME
);
3313 myhostid
= zone_get_hostid(NULL
);
3315 if (hostid
!= 0 && myhostid
!= 0 && hostid
!= myhostid
) {
3316 cmn_err(CE_WARN
, "pool '%s' could not be "
3317 "loaded as it was last accessed by "
3318 "another system (host: %s hostid: 0x%llx). "
3319 "See: https://openzfs.github.io/openzfs-docs/msg/"
3321 spa_name(spa
), hostname
, (u_longlong_t
)hostid
);
3322 spa_load_failed(spa
, "hostid verification failed: pool "
3323 "last accessed by host: %s (hostid: 0x%llx)",
3324 hostname
, (u_longlong_t
)hostid
);
3325 return (SET_ERROR(EBADF
));
3333 spa_ld_parse_config(spa_t
*spa
, spa_import_type_t type
)
3336 nvlist_t
*nvtree
, *nvl
, *config
= spa
->spa_config
;
3341 char *compatibility
;
3344 * Versioning wasn't explicitly added to the label until later, so if
3345 * it's not present treat it as the initial version.
3347 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
3348 &spa
->spa_ubsync
.ub_version
) != 0)
3349 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
3351 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
3352 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3353 ZPOOL_CONFIG_POOL_GUID
);
3354 return (SET_ERROR(EINVAL
));
3358 * If we are doing an import, ensure that the pool is not already
3359 * imported by checking if its pool guid already exists in the
3362 * The only case that we allow an already imported pool to be
3363 * imported again, is when the pool is checkpointed and we want to
3364 * look at its checkpointed state from userland tools like zdb.
3367 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3368 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3369 spa_guid_exists(pool_guid
, 0)) {
3371 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3372 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3373 spa_guid_exists(pool_guid
, 0) &&
3374 !spa_importing_readonly_checkpoint(spa
)) {
3376 spa_load_failed(spa
, "a pool with guid %llu is already open",
3377 (u_longlong_t
)pool_guid
);
3378 return (SET_ERROR(EEXIST
));
3381 spa
->spa_config_guid
= pool_guid
;
3383 nvlist_free(spa
->spa_load_info
);
3384 spa
->spa_load_info
= fnvlist_alloc();
3386 ASSERT(spa
->spa_comment
== NULL
);
3387 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
3388 spa
->spa_comment
= spa_strdup(comment
);
3390 ASSERT(spa
->spa_compatibility
== NULL
);
3391 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMPATIBILITY
,
3392 &compatibility
) == 0)
3393 spa
->spa_compatibility
= spa_strdup(compatibility
);
3395 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
3396 &spa
->spa_config_txg
);
3398 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) == 0)
3399 spa
->spa_config_splitting
= fnvlist_dup(nvl
);
3401 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
)) {
3402 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3403 ZPOOL_CONFIG_VDEV_TREE
);
3404 return (SET_ERROR(EINVAL
));
3408 * Create "The Godfather" zio to hold all async IOs
3410 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3412 for (int i
= 0; i
< max_ncpus
; i
++) {
3413 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3414 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3415 ZIO_FLAG_GODFATHER
);
3419 * Parse the configuration into a vdev tree. We explicitly set the
3420 * value that will be returned by spa_version() since parsing the
3421 * configuration requires knowing the version number.
3423 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3424 parse
= (type
== SPA_IMPORT_EXISTING
?
3425 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
3426 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
3427 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3430 spa_load_failed(spa
, "unable to parse config [error=%d]",
3435 ASSERT(spa
->spa_root_vdev
== rvd
);
3436 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
3437 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
3439 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3440 ASSERT(spa_guid(spa
) == pool_guid
);
3447 * Recursively open all vdevs in the vdev tree. This function is called twice:
3448 * first with the untrusted config, then with the trusted config.
3451 spa_ld_open_vdevs(spa_t
*spa
)
3456 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3457 * missing/unopenable for the root vdev to be still considered openable.
3459 if (spa
->spa_trust_config
) {
3460 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds
;
3461 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_CACHEFILE
) {
3462 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_cachefile
;
3463 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_SCAN
) {
3464 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_scan
;
3466 spa
->spa_missing_tvds_allowed
= 0;
3469 spa
->spa_missing_tvds_allowed
=
3470 MAX(zfs_max_missing_tvds
, spa
->spa_missing_tvds_allowed
);
3472 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3473 error
= vdev_open(spa
->spa_root_vdev
);
3474 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3476 if (spa
->spa_missing_tvds
!= 0) {
3477 spa_load_note(spa
, "vdev tree has %lld missing top-level "
3478 "vdevs.", (u_longlong_t
)spa
->spa_missing_tvds
);
3479 if (spa
->spa_trust_config
&& (spa
->spa_mode
& SPA_MODE_WRITE
)) {
3481 * Although theoretically we could allow users to open
3482 * incomplete pools in RW mode, we'd need to add a lot
3483 * of extra logic (e.g. adjust pool space to account
3484 * for missing vdevs).
3485 * This limitation also prevents users from accidentally
3486 * opening the pool in RW mode during data recovery and
3487 * damaging it further.
3489 spa_load_note(spa
, "pools with missing top-level "
3490 "vdevs can only be opened in read-only mode.");
3491 error
= SET_ERROR(ENXIO
);
3493 spa_load_note(spa
, "current settings allow for maximum "
3494 "%lld missing top-level vdevs at this stage.",
3495 (u_longlong_t
)spa
->spa_missing_tvds_allowed
);
3499 spa_load_failed(spa
, "unable to open vdev tree [error=%d]",
3502 if (spa
->spa_missing_tvds
!= 0 || error
!= 0)
3503 vdev_dbgmsg_print_tree(spa
->spa_root_vdev
, 2);
3509 * We need to validate the vdev labels against the configuration that
3510 * we have in hand. This function is called twice: first with an untrusted
3511 * config, then with a trusted config. The validation is more strict when the
3512 * config is trusted.
3515 spa_ld_validate_vdevs(spa_t
*spa
)
3518 vdev_t
*rvd
= spa
->spa_root_vdev
;
3520 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3521 error
= vdev_validate(rvd
);
3522 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3525 spa_load_failed(spa
, "vdev_validate failed [error=%d]", error
);
3529 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
3530 spa_load_failed(spa
, "cannot open vdev tree after invalidating "
3532 vdev_dbgmsg_print_tree(rvd
, 2);
3533 return (SET_ERROR(ENXIO
));
3540 spa_ld_select_uberblock_done(spa_t
*spa
, uberblock_t
*ub
)
3542 spa
->spa_state
= POOL_STATE_ACTIVE
;
3543 spa
->spa_ubsync
= spa
->spa_uberblock
;
3544 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
3545 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
3546 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
3547 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
3548 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
3549 spa
->spa_prev_software_version
= ub
->ub_software_version
;
3553 spa_ld_select_uberblock(spa_t
*spa
, spa_import_type_t type
)
3555 vdev_t
*rvd
= spa
->spa_root_vdev
;
3557 uberblock_t
*ub
= &spa
->spa_uberblock
;
3558 boolean_t activity_check
= B_FALSE
;
3561 * If we are opening the checkpointed state of the pool by
3562 * rewinding to it, at this point we will have written the
3563 * checkpointed uberblock to the vdev labels, so searching
3564 * the labels will find the right uberblock. However, if
3565 * we are opening the checkpointed state read-only, we have
3566 * not modified the labels. Therefore, we must ignore the
3567 * labels and continue using the spa_uberblock that was set
3568 * by spa_ld_checkpoint_rewind.
3570 * Note that it would be fine to ignore the labels when
3571 * rewinding (opening writeable) as well. However, if we
3572 * crash just after writing the labels, we will end up
3573 * searching the labels. Doing so in the common case means
3574 * that this code path gets exercised normally, rather than
3575 * just in the edge case.
3577 if (ub
->ub_checkpoint_txg
!= 0 &&
3578 spa_importing_readonly_checkpoint(spa
)) {
3579 spa_ld_select_uberblock_done(spa
, ub
);
3584 * Find the best uberblock.
3586 vdev_uberblock_load(rvd
, ub
, &label
);
3589 * If we weren't able to find a single valid uberblock, return failure.
3591 if (ub
->ub_txg
== 0) {
3593 spa_load_failed(spa
, "no valid uberblock found");
3594 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
3597 if (spa
->spa_load_max_txg
!= UINT64_MAX
) {
3598 (void) spa_import_progress_set_max_txg(spa_guid(spa
),
3599 (u_longlong_t
)spa
->spa_load_max_txg
);
3601 spa_load_note(spa
, "using uberblock with txg=%llu",
3602 (u_longlong_t
)ub
->ub_txg
);
3606 * For pools which have the multihost property on determine if the
3607 * pool is truly inactive and can be safely imported. Prevent
3608 * hosts which don't have a hostid set from importing the pool.
3610 activity_check
= spa_activity_check_required(spa
, ub
, label
,
3612 if (activity_check
) {
3613 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
3614 spa_get_hostid(spa
) == 0) {
3616 fnvlist_add_uint64(spa
->spa_load_info
,
3617 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3618 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3621 int error
= spa_activity_check(spa
, ub
, spa
->spa_config
);
3627 fnvlist_add_uint64(spa
->spa_load_info
,
3628 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
3629 fnvlist_add_uint64(spa
->spa_load_info
,
3630 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
3631 fnvlist_add_uint16(spa
->spa_load_info
,
3632 ZPOOL_CONFIG_MMP_SEQ
,
3633 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0));
3637 * If the pool has an unsupported version we can't open it.
3639 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
3641 spa_load_failed(spa
, "version %llu is not supported",
3642 (u_longlong_t
)ub
->ub_version
);
3643 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
3646 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3650 * If we weren't able to find what's necessary for reading the
3651 * MOS in the label, return failure.
3653 if (label
== NULL
) {
3654 spa_load_failed(spa
, "label config unavailable");
3655 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3659 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
3662 spa_load_failed(spa
, "invalid label: '%s' missing",
3663 ZPOOL_CONFIG_FEATURES_FOR_READ
);
3664 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3669 * Update our in-core representation with the definitive values
3672 nvlist_free(spa
->spa_label_features
);
3673 spa
->spa_label_features
= fnvlist_dup(features
);
3679 * Look through entries in the label nvlist's features_for_read. If
3680 * there is a feature listed there which we don't understand then we
3681 * cannot open a pool.
3683 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3684 nvlist_t
*unsup_feat
;
3686 unsup_feat
= fnvlist_alloc();
3688 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
3690 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
3691 if (!zfeature_is_supported(nvpair_name(nvp
))) {
3692 fnvlist_add_string(unsup_feat
,
3693 nvpair_name(nvp
), "");
3697 if (!nvlist_empty(unsup_feat
)) {
3698 fnvlist_add_nvlist(spa
->spa_load_info
,
3699 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
3700 nvlist_free(unsup_feat
);
3701 spa_load_failed(spa
, "some features are unsupported");
3702 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
3706 nvlist_free(unsup_feat
);
3709 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
3710 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3711 spa_try_repair(spa
, spa
->spa_config
);
3712 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3713 nvlist_free(spa
->spa_config_splitting
);
3714 spa
->spa_config_splitting
= NULL
;
3718 * Initialize internal SPA structures.
3720 spa_ld_select_uberblock_done(spa
, ub
);
3726 spa_ld_open_rootbp(spa_t
*spa
)
3729 vdev_t
*rvd
= spa
->spa_root_vdev
;
3731 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
3733 spa_load_failed(spa
, "unable to open rootbp in dsl_pool_init "
3734 "[error=%d]", error
);
3735 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3737 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
3743 spa_ld_trusted_config(spa_t
*spa
, spa_import_type_t type
,
3744 boolean_t reloading
)
3746 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
3747 nvlist_t
*nv
, *mos_config
, *policy
;
3748 int error
= 0, copy_error
;
3749 uint64_t healthy_tvds
, healthy_tvds_mos
;
3750 uint64_t mos_config_txg
;
3752 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
, B_TRUE
)
3754 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3757 * If we're assembling a pool from a split, the config provided is
3758 * already trusted so there is nothing to do.
3760 if (type
== SPA_IMPORT_ASSEMBLE
)
3763 healthy_tvds
= spa_healthy_core_tvds(spa
);
3765 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
)
3767 spa_load_failed(spa
, "unable to retrieve MOS config");
3768 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3772 * If we are doing an open, pool owner wasn't verified yet, thus do
3773 * the verification here.
3775 if (spa
->spa_load_state
== SPA_LOAD_OPEN
) {
3776 error
= spa_verify_host(spa
, mos_config
);
3778 nvlist_free(mos_config
);
3783 nv
= fnvlist_lookup_nvlist(mos_config
, ZPOOL_CONFIG_VDEV_TREE
);
3785 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3788 * Build a new vdev tree from the trusted config
3790 error
= spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
);
3792 nvlist_free(mos_config
);
3793 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3794 spa_load_failed(spa
, "spa_config_parse failed [error=%d]",
3796 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3800 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3801 * obtained by scanning /dev/dsk, then it will have the right vdev
3802 * paths. We update the trusted MOS config with this information.
3803 * We first try to copy the paths with vdev_copy_path_strict, which
3804 * succeeds only when both configs have exactly the same vdev tree.
3805 * If that fails, we fall back to a more flexible method that has a
3806 * best effort policy.
3808 copy_error
= vdev_copy_path_strict(rvd
, mrvd
);
3809 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3810 spa_load_note(spa
, "provided vdev tree:");
3811 vdev_dbgmsg_print_tree(rvd
, 2);
3812 spa_load_note(spa
, "MOS vdev tree:");
3813 vdev_dbgmsg_print_tree(mrvd
, 2);
3815 if (copy_error
!= 0) {
3816 spa_load_note(spa
, "vdev_copy_path_strict failed, falling "
3817 "back to vdev_copy_path_relaxed");
3818 vdev_copy_path_relaxed(rvd
, mrvd
);
3823 spa
->spa_root_vdev
= mrvd
;
3825 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3828 * We will use spa_config if we decide to reload the spa or if spa_load
3829 * fails and we rewind. We must thus regenerate the config using the
3830 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3831 * pass settings on how to load the pool and is not stored in the MOS.
3832 * We copy it over to our new, trusted config.
3834 mos_config_txg
= fnvlist_lookup_uint64(mos_config
,
3835 ZPOOL_CONFIG_POOL_TXG
);
3836 nvlist_free(mos_config
);
3837 mos_config
= spa_config_generate(spa
, NULL
, mos_config_txg
, B_FALSE
);
3838 if (nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_LOAD_POLICY
,
3840 fnvlist_add_nvlist(mos_config
, ZPOOL_LOAD_POLICY
, policy
);
3841 spa_config_set(spa
, mos_config
);
3842 spa
->spa_config_source
= SPA_CONFIG_SRC_MOS
;
3845 * Now that we got the config from the MOS, we should be more strict
3846 * in checking blkptrs and can make assumptions about the consistency
3847 * of the vdev tree. spa_trust_config must be set to true before opening
3848 * vdevs in order for them to be writeable.
3850 spa
->spa_trust_config
= B_TRUE
;
3853 * Open and validate the new vdev tree
3855 error
= spa_ld_open_vdevs(spa
);
3859 error
= spa_ld_validate_vdevs(spa
);
3863 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3864 spa_load_note(spa
, "final vdev tree:");
3865 vdev_dbgmsg_print_tree(rvd
, 2);
3868 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
&&
3869 !spa
->spa_extreme_rewind
&& zfs_max_missing_tvds
== 0) {
3871 * Sanity check to make sure that we are indeed loading the
3872 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3873 * in the config provided and they happened to be the only ones
3874 * to have the latest uberblock, we could involuntarily perform
3875 * an extreme rewind.
3877 healthy_tvds_mos
= spa_healthy_core_tvds(spa
);
3878 if (healthy_tvds_mos
- healthy_tvds
>=
3879 SPA_SYNC_MIN_VDEVS
) {
3880 spa_load_note(spa
, "config provided misses too many "
3881 "top-level vdevs compared to MOS (%lld vs %lld). ",
3882 (u_longlong_t
)healthy_tvds
,
3883 (u_longlong_t
)healthy_tvds_mos
);
3884 spa_load_note(spa
, "vdev tree:");
3885 vdev_dbgmsg_print_tree(rvd
, 2);
3887 spa_load_failed(spa
, "config was already "
3888 "provided from MOS. Aborting.");
3889 return (spa_vdev_err(rvd
,
3890 VDEV_AUX_CORRUPT_DATA
, EIO
));
3892 spa_load_note(spa
, "spa must be reloaded using MOS "
3894 return (SET_ERROR(EAGAIN
));
3898 error
= spa_check_for_missing_logs(spa
);
3900 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
3902 if (rvd
->vdev_guid_sum
!= spa
->spa_uberblock
.ub_guid_sum
) {
3903 spa_load_failed(spa
, "uberblock guid sum doesn't match MOS "
3904 "guid sum (%llu != %llu)",
3905 (u_longlong_t
)spa
->spa_uberblock
.ub_guid_sum
,
3906 (u_longlong_t
)rvd
->vdev_guid_sum
);
3907 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3915 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
3918 vdev_t
*rvd
= spa
->spa_root_vdev
;
3921 * Everything that we read before spa_remove_init() must be stored
3922 * on concreted vdevs. Therefore we do this as early as possible.
3924 error
= spa_remove_init(spa
);
3926 spa_load_failed(spa
, "spa_remove_init failed [error=%d]",
3928 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3932 * Retrieve information needed to condense indirect vdev mappings.
3934 error
= spa_condense_init(spa
);
3936 spa_load_failed(spa
, "spa_condense_init failed [error=%d]",
3938 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3945 spa_ld_check_features(spa_t
*spa
, boolean_t
*missing_feat_writep
)
3948 vdev_t
*rvd
= spa
->spa_root_vdev
;
3950 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
3951 boolean_t missing_feat_read
= B_FALSE
;
3952 nvlist_t
*unsup_feat
, *enabled_feat
;
3954 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
3955 &spa
->spa_feat_for_read_obj
, B_TRUE
) != 0) {
3956 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3959 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
3960 &spa
->spa_feat_for_write_obj
, B_TRUE
) != 0) {
3961 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3964 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
3965 &spa
->spa_feat_desc_obj
, B_TRUE
) != 0) {
3966 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3969 enabled_feat
= fnvlist_alloc();
3970 unsup_feat
= fnvlist_alloc();
3972 if (!spa_features_check(spa
, B_FALSE
,
3973 unsup_feat
, enabled_feat
))
3974 missing_feat_read
= B_TRUE
;
3976 if (spa_writeable(spa
) ||
3977 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) {
3978 if (!spa_features_check(spa
, B_TRUE
,
3979 unsup_feat
, enabled_feat
)) {
3980 *missing_feat_writep
= B_TRUE
;
3984 fnvlist_add_nvlist(spa
->spa_load_info
,
3985 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
3987 if (!nvlist_empty(unsup_feat
)) {
3988 fnvlist_add_nvlist(spa
->spa_load_info
,
3989 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
3992 fnvlist_free(enabled_feat
);
3993 fnvlist_free(unsup_feat
);
3995 if (!missing_feat_read
) {
3996 fnvlist_add_boolean(spa
->spa_load_info
,
3997 ZPOOL_CONFIG_CAN_RDONLY
);
4001 * If the state is SPA_LOAD_TRYIMPORT, our objective is
4002 * twofold: to determine whether the pool is available for
4003 * import in read-write mode and (if it is not) whether the
4004 * pool is available for import in read-only mode. If the pool
4005 * is available for import in read-write mode, it is displayed
4006 * as available in userland; if it is not available for import
4007 * in read-only mode, it is displayed as unavailable in
4008 * userland. If the pool is available for import in read-only
4009 * mode but not read-write mode, it is displayed as unavailable
4010 * in userland with a special note that the pool is actually
4011 * available for open in read-only mode.
4013 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
4014 * missing a feature for write, we must first determine whether
4015 * the pool can be opened read-only before returning to
4016 * userland in order to know whether to display the
4017 * abovementioned note.
4019 if (missing_feat_read
|| (*missing_feat_writep
&&
4020 spa_writeable(spa
))) {
4021 spa_load_failed(spa
, "pool uses unsupported features");
4022 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
4027 * Load refcounts for ZFS features from disk into an in-memory
4028 * cache during SPA initialization.
4030 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
4033 error
= feature_get_refcount_from_disk(spa
,
4034 &spa_feature_table
[i
], &refcount
);
4036 spa
->spa_feat_refcount_cache
[i
] = refcount
;
4037 } else if (error
== ENOTSUP
) {
4038 spa
->spa_feat_refcount_cache
[i
] =
4039 SPA_FEATURE_DISABLED
;
4041 spa_load_failed(spa
, "error getting refcount "
4042 "for feature %s [error=%d]",
4043 spa_feature_table
[i
].fi_guid
, error
);
4044 return (spa_vdev_err(rvd
,
4045 VDEV_AUX_CORRUPT_DATA
, EIO
));
4050 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
4051 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
4052 &spa
->spa_feat_enabled_txg_obj
, B_TRUE
) != 0)
4053 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4057 * Encryption was added before bookmark_v2, even though bookmark_v2
4058 * is now a dependency. If this pool has encryption enabled without
4059 * bookmark_v2, trigger an errata message.
4061 if (spa_feature_is_enabled(spa
, SPA_FEATURE_ENCRYPTION
) &&
4062 !spa_feature_is_enabled(spa
, SPA_FEATURE_BOOKMARK_V2
)) {
4063 spa
->spa_errata
= ZPOOL_ERRATA_ZOL_8308_ENCRYPTION
;
4070 spa_ld_load_special_directories(spa_t
*spa
)
4073 vdev_t
*rvd
= spa
->spa_root_vdev
;
4075 spa
->spa_is_initializing
= B_TRUE
;
4076 error
= dsl_pool_open(spa
->spa_dsl_pool
);
4077 spa
->spa_is_initializing
= B_FALSE
;
4079 spa_load_failed(spa
, "dsl_pool_open failed [error=%d]", error
);
4080 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4087 spa_ld_get_props(spa_t
*spa
)
4091 vdev_t
*rvd
= spa
->spa_root_vdev
;
4093 /* Grab the checksum salt from the MOS. */
4094 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4095 DMU_POOL_CHECKSUM_SALT
, 1,
4096 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
4097 spa
->spa_cksum_salt
.zcs_bytes
);
4098 if (error
== ENOENT
) {
4099 /* Generate a new salt for subsequent use */
4100 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4101 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4102 } else if (error
!= 0) {
4103 spa_load_failed(spa
, "unable to retrieve checksum salt from "
4104 "MOS [error=%d]", error
);
4105 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4108 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
, B_TRUE
) != 0)
4109 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4110 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
4112 spa_load_failed(spa
, "error opening deferred-frees bpobj "
4113 "[error=%d]", error
);
4114 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4118 * Load the bit that tells us to use the new accounting function
4119 * (raid-z deflation). If we have an older pool, this will not
4122 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
, B_FALSE
);
4123 if (error
!= 0 && error
!= ENOENT
)
4124 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4126 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
4127 &spa
->spa_creation_version
, B_FALSE
);
4128 if (error
!= 0 && error
!= ENOENT
)
4129 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4132 * Load the persistent error log. If we have an older pool, this will
4135 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
,
4137 if (error
!= 0 && error
!= ENOENT
)
4138 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4140 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
4141 &spa
->spa_errlog_scrub
, B_FALSE
);
4142 if (error
!= 0 && error
!= ENOENT
)
4143 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4146 * Load the livelist deletion field. If a livelist is queued for
4147 * deletion, indicate that in the spa
4149 error
= spa_dir_prop(spa
, DMU_POOL_DELETED_CLONES
,
4150 &spa
->spa_livelists_to_delete
, B_FALSE
);
4151 if (error
!= 0 && error
!= ENOENT
)
4152 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4155 * Load the history object. If we have an older pool, this
4156 * will not be present.
4158 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
, B_FALSE
);
4159 if (error
!= 0 && error
!= ENOENT
)
4160 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4163 * Load the per-vdev ZAP map. If we have an older pool, this will not
4164 * be present; in this case, defer its creation to a later time to
4165 * avoid dirtying the MOS this early / out of sync context. See
4166 * spa_sync_config_object.
4169 /* The sentinel is only available in the MOS config. */
4170 nvlist_t
*mos_config
;
4171 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0) {
4172 spa_load_failed(spa
, "unable to retrieve MOS config");
4173 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4176 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
4177 &spa
->spa_all_vdev_zaps
, B_FALSE
);
4179 if (error
== ENOENT
) {
4180 VERIFY(!nvlist_exists(mos_config
,
4181 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
4182 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
4183 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4184 } else if (error
!= 0) {
4185 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4186 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
4188 * An older version of ZFS overwrote the sentinel value, so
4189 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4190 * destruction to later; see spa_sync_config_object.
4192 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
4194 * We're assuming that no vdevs have had their ZAPs created
4195 * before this. Better be sure of it.
4197 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4199 nvlist_free(mos_config
);
4201 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4203 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
,
4205 if (error
&& error
!= ENOENT
)
4206 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4209 uint64_t autoreplace
= 0;
4211 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
4212 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
4213 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
4214 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
4215 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
4216 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
4217 spa_prop_find(spa
, ZPOOL_PROP_AUTOTRIM
, &spa
->spa_autotrim
);
4218 spa
->spa_autoreplace
= (autoreplace
!= 0);
4222 * If we are importing a pool with missing top-level vdevs,
4223 * we enforce that the pool doesn't panic or get suspended on
4224 * error since the likelihood of missing data is extremely high.
4226 if (spa
->spa_missing_tvds
> 0 &&
4227 spa
->spa_failmode
!= ZIO_FAILURE_MODE_CONTINUE
&&
4228 spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4229 spa_load_note(spa
, "forcing failmode to 'continue' "
4230 "as some top level vdevs are missing");
4231 spa
->spa_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
4238 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
4241 vdev_t
*rvd
= spa
->spa_root_vdev
;
4244 * If we're assembling the pool from the split-off vdevs of
4245 * an existing pool, we don't want to attach the spares & cache
4250 * Load any hot spares for this pool.
4252 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
,
4254 if (error
!= 0 && error
!= ENOENT
)
4255 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4256 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4257 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
4258 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
4259 &spa
->spa_spares
.sav_config
) != 0) {
4260 spa_load_failed(spa
, "error loading spares nvlist");
4261 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4264 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4265 spa_load_spares(spa
);
4266 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4267 } else if (error
== 0) {
4268 spa
->spa_spares
.sav_sync
= B_TRUE
;
4272 * Load any level 2 ARC devices for this pool.
4274 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
4275 &spa
->spa_l2cache
.sav_object
, B_FALSE
);
4276 if (error
!= 0 && error
!= ENOENT
)
4277 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4278 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4279 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
4280 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
4281 &spa
->spa_l2cache
.sav_config
) != 0) {
4282 spa_load_failed(spa
, "error loading l2cache nvlist");
4283 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4286 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4287 spa_load_l2cache(spa
);
4288 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4289 } else if (error
== 0) {
4290 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4297 spa_ld_load_vdev_metadata(spa_t
*spa
)
4300 vdev_t
*rvd
= spa
->spa_root_vdev
;
4303 * If the 'multihost' property is set, then never allow a pool to
4304 * be imported when the system hostid is zero. The exception to
4305 * this rule is zdb which is always allowed to access pools.
4307 if (spa_multihost(spa
) && spa_get_hostid(spa
) == 0 &&
4308 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
4309 fnvlist_add_uint64(spa
->spa_load_info
,
4310 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
4311 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
4315 * If the 'autoreplace' property is set, then post a resource notifying
4316 * the ZFS DE that it should not issue any faults for unopenable
4317 * devices. We also iterate over the vdevs, and post a sysevent for any
4318 * unopenable vdevs so that the normal autoreplace handler can take
4321 if (spa
->spa_autoreplace
&& spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4322 spa_check_removed(spa
->spa_root_vdev
);
4324 * For the import case, this is done in spa_import(), because
4325 * at this point we're using the spare definitions from
4326 * the MOS config, not necessarily from the userland config.
4328 if (spa
->spa_load_state
!= SPA_LOAD_IMPORT
) {
4329 spa_aux_check_removed(&spa
->spa_spares
);
4330 spa_aux_check_removed(&spa
->spa_l2cache
);
4335 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4337 error
= vdev_load(rvd
);
4339 spa_load_failed(spa
, "vdev_load failed [error=%d]", error
);
4340 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4343 error
= spa_ld_log_spacemaps(spa
);
4345 spa_load_failed(spa
, "spa_ld_log_sm_data failed [error=%d]",
4347 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4351 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4353 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4354 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
, B_FALSE
);
4355 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4361 spa_ld_load_dedup_tables(spa_t
*spa
)
4364 vdev_t
*rvd
= spa
->spa_root_vdev
;
4366 error
= ddt_load(spa
);
4368 spa_load_failed(spa
, "ddt_load failed [error=%d]", error
);
4369 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4376 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
4378 vdev_t
*rvd
= spa
->spa_root_vdev
;
4380 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
)) {
4381 boolean_t missing
= spa_check_logs(spa
);
4383 if (spa
->spa_missing_tvds
!= 0) {
4384 spa_load_note(spa
, "spa_check_logs failed "
4385 "so dropping the logs");
4387 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
4388 spa_load_failed(spa
, "spa_check_logs failed");
4389 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
,
4399 spa_ld_verify_pool_data(spa_t
*spa
)
4402 vdev_t
*rvd
= spa
->spa_root_vdev
;
4405 * We've successfully opened the pool, verify that we're ready
4406 * to start pushing transactions.
4408 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4409 error
= spa_load_verify(spa
);
4411 spa_load_failed(spa
, "spa_load_verify failed "
4412 "[error=%d]", error
);
4413 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
4422 spa_ld_claim_log_blocks(spa_t
*spa
)
4425 dsl_pool_t
*dp
= spa_get_dsl(spa
);
4428 * Claim log blocks that haven't been committed yet.
4429 * This must all happen in a single txg.
4430 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4431 * invoked from zil_claim_log_block()'s i/o done callback.
4432 * Price of rollback is that we abandon the log.
4434 spa
->spa_claiming
= B_TRUE
;
4436 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
4437 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
4438 zil_claim
, tx
, DS_FIND_CHILDREN
);
4441 spa
->spa_claiming
= B_FALSE
;
4443 spa_set_log_state(spa
, SPA_LOG_GOOD
);
4447 spa_ld_check_for_config_update(spa_t
*spa
, uint64_t config_cache_txg
,
4448 boolean_t update_config_cache
)
4450 vdev_t
*rvd
= spa
->spa_root_vdev
;
4451 int need_update
= B_FALSE
;
4454 * If the config cache is stale, or we have uninitialized
4455 * metaslabs (see spa_vdev_add()), then update the config.
4457 * If this is a verbatim import, trust the current
4458 * in-core spa_config and update the disk labels.
4460 if (update_config_cache
|| config_cache_txg
!= spa
->spa_config_txg
||
4461 spa
->spa_load_state
== SPA_LOAD_IMPORT
||
4462 spa
->spa_load_state
== SPA_LOAD_RECOVER
||
4463 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
4464 need_update
= B_TRUE
;
4466 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
4467 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
4468 need_update
= B_TRUE
;
4471 * Update the config cache asynchronously in case we're the
4472 * root pool, in which case the config cache isn't writable yet.
4475 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4479 spa_ld_prepare_for_reload(spa_t
*spa
)
4481 spa_mode_t mode
= spa
->spa_mode
;
4482 int async_suspended
= spa
->spa_async_suspended
;
4485 spa_deactivate(spa
);
4486 spa_activate(spa
, mode
);
4489 * We save the value of spa_async_suspended as it gets reset to 0 by
4490 * spa_unload(). We want to restore it back to the original value before
4491 * returning as we might be calling spa_async_resume() later.
4493 spa
->spa_async_suspended
= async_suspended
;
4497 spa_ld_read_checkpoint_txg(spa_t
*spa
)
4499 uberblock_t checkpoint
;
4502 ASSERT0(spa
->spa_checkpoint_txg
);
4503 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4505 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4506 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4507 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4509 if (error
== ENOENT
)
4515 ASSERT3U(checkpoint
.ub_txg
, !=, 0);
4516 ASSERT3U(checkpoint
.ub_checkpoint_txg
, !=, 0);
4517 ASSERT3U(checkpoint
.ub_timestamp
, !=, 0);
4518 spa
->spa_checkpoint_txg
= checkpoint
.ub_txg
;
4519 spa
->spa_checkpoint_info
.sci_timestamp
= checkpoint
.ub_timestamp
;
4525 spa_ld_mos_init(spa_t
*spa
, spa_import_type_t type
)
4529 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4530 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4533 * Never trust the config that is provided unless we are assembling
4534 * a pool following a split.
4535 * This means don't trust blkptrs and the vdev tree in general. This
4536 * also effectively puts the spa in read-only mode since
4537 * spa_writeable() checks for spa_trust_config to be true.
4538 * We will later load a trusted config from the MOS.
4540 if (type
!= SPA_IMPORT_ASSEMBLE
)
4541 spa
->spa_trust_config
= B_FALSE
;
4544 * Parse the config provided to create a vdev tree.
4546 error
= spa_ld_parse_config(spa
, type
);
4550 spa_import_progress_add(spa
);
4553 * Now that we have the vdev tree, try to open each vdev. This involves
4554 * opening the underlying physical device, retrieving its geometry and
4555 * probing the vdev with a dummy I/O. The state of each vdev will be set
4556 * based on the success of those operations. After this we'll be ready
4557 * to read from the vdevs.
4559 error
= spa_ld_open_vdevs(spa
);
4564 * Read the label of each vdev and make sure that the GUIDs stored
4565 * there match the GUIDs in the config provided.
4566 * If we're assembling a new pool that's been split off from an
4567 * existing pool, the labels haven't yet been updated so we skip
4568 * validation for now.
4570 if (type
!= SPA_IMPORT_ASSEMBLE
) {
4571 error
= spa_ld_validate_vdevs(spa
);
4577 * Read all vdev labels to find the best uberblock (i.e. latest,
4578 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
4579 * get the list of features required to read blkptrs in the MOS from
4580 * the vdev label with the best uberblock and verify that our version
4581 * of zfs supports them all.
4583 error
= spa_ld_select_uberblock(spa
, type
);
4588 * Pass that uberblock to the dsl_pool layer which will open the root
4589 * blkptr. This blkptr points to the latest version of the MOS and will
4590 * allow us to read its contents.
4592 error
= spa_ld_open_rootbp(spa
);
4600 spa_ld_checkpoint_rewind(spa_t
*spa
)
4602 uberblock_t checkpoint
;
4605 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4606 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4608 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4609 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4610 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4613 spa_load_failed(spa
, "unable to retrieve checkpointed "
4614 "uberblock from the MOS config [error=%d]", error
);
4616 if (error
== ENOENT
)
4617 error
= ZFS_ERR_NO_CHECKPOINT
;
4622 ASSERT3U(checkpoint
.ub_txg
, <, spa
->spa_uberblock
.ub_txg
);
4623 ASSERT3U(checkpoint
.ub_txg
, ==, checkpoint
.ub_checkpoint_txg
);
4626 * We need to update the txg and timestamp of the checkpointed
4627 * uberblock to be higher than the latest one. This ensures that
4628 * the checkpointed uberblock is selected if we were to close and
4629 * reopen the pool right after we've written it in the vdev labels.
4630 * (also see block comment in vdev_uberblock_compare)
4632 checkpoint
.ub_txg
= spa
->spa_uberblock
.ub_txg
+ 1;
4633 checkpoint
.ub_timestamp
= gethrestime_sec();
4636 * Set current uberblock to be the checkpointed uberblock.
4638 spa
->spa_uberblock
= checkpoint
;
4641 * If we are doing a normal rewind, then the pool is open for
4642 * writing and we sync the "updated" checkpointed uberblock to
4643 * disk. Once this is done, we've basically rewound the whole
4644 * pool and there is no way back.
4646 * There are cases when we don't want to attempt and sync the
4647 * checkpointed uberblock to disk because we are opening a
4648 * pool as read-only. Specifically, verifying the checkpointed
4649 * state with zdb, and importing the checkpointed state to get
4650 * a "preview" of its content.
4652 if (spa_writeable(spa
)) {
4653 vdev_t
*rvd
= spa
->spa_root_vdev
;
4655 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4656 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
4658 int children
= rvd
->vdev_children
;
4659 int c0
= random_in_range(children
);
4661 for (int c
= 0; c
< children
; c
++) {
4662 vdev_t
*vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
4664 /* Stop when revisiting the first vdev */
4665 if (c
> 0 && svd
[0] == vd
)
4668 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
4669 !vdev_is_concrete(vd
))
4672 svd
[svdcount
++] = vd
;
4673 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
4676 error
= vdev_config_sync(svd
, svdcount
, spa
->spa_first_txg
);
4678 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
4679 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4682 spa_load_failed(spa
, "failed to write checkpointed "
4683 "uberblock to the vdev labels [error=%d]", error
);
4692 spa_ld_mos_with_trusted_config(spa_t
*spa
, spa_import_type_t type
,
4693 boolean_t
*update_config_cache
)
4698 * Parse the config for pool, open and validate vdevs,
4699 * select an uberblock, and use that uberblock to open
4702 error
= spa_ld_mos_init(spa
, type
);
4707 * Retrieve the trusted config stored in the MOS and use it to create
4708 * a new, exact version of the vdev tree, then reopen all vdevs.
4710 error
= spa_ld_trusted_config(spa
, type
, B_FALSE
);
4711 if (error
== EAGAIN
) {
4712 if (update_config_cache
!= NULL
)
4713 *update_config_cache
= B_TRUE
;
4716 * Redo the loading process with the trusted config if it is
4717 * too different from the untrusted config.
4719 spa_ld_prepare_for_reload(spa
);
4720 spa_load_note(spa
, "RELOADING");
4721 error
= spa_ld_mos_init(spa
, type
);
4725 error
= spa_ld_trusted_config(spa
, type
, B_TRUE
);
4729 } else if (error
!= 0) {
4737 * Load an existing storage pool, using the config provided. This config
4738 * describes which vdevs are part of the pool and is later validated against
4739 * partial configs present in each vdev's label and an entire copy of the
4740 * config stored in the MOS.
4743 spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
4746 boolean_t missing_feat_write
= B_FALSE
;
4747 boolean_t checkpoint_rewind
=
4748 (spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4749 boolean_t update_config_cache
= B_FALSE
;
4751 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4752 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4754 spa_load_note(spa
, "LOADING");
4756 error
= spa_ld_mos_with_trusted_config(spa
, type
, &update_config_cache
);
4761 * If we are rewinding to the checkpoint then we need to repeat
4762 * everything we've done so far in this function but this time
4763 * selecting the checkpointed uberblock and using that to open
4766 if (checkpoint_rewind
) {
4768 * If we are rewinding to the checkpoint update config cache
4771 update_config_cache
= B_TRUE
;
4774 * Extract the checkpointed uberblock from the current MOS
4775 * and use this as the pool's uberblock from now on. If the
4776 * pool is imported as writeable we also write the checkpoint
4777 * uberblock to the labels, making the rewind permanent.
4779 error
= spa_ld_checkpoint_rewind(spa
);
4784 * Redo the loading process again with the
4785 * checkpointed uberblock.
4787 spa_ld_prepare_for_reload(spa
);
4788 spa_load_note(spa
, "LOADING checkpointed uberblock");
4789 error
= spa_ld_mos_with_trusted_config(spa
, type
, NULL
);
4795 * Retrieve the checkpoint txg if the pool has a checkpoint.
4797 error
= spa_ld_read_checkpoint_txg(spa
);
4802 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
4803 * from the pool and their contents were re-mapped to other vdevs. Note
4804 * that everything that we read before this step must have been
4805 * rewritten on concrete vdevs after the last device removal was
4806 * initiated. Otherwise we could be reading from indirect vdevs before
4807 * we have loaded their mappings.
4809 error
= spa_ld_open_indirect_vdev_metadata(spa
);
4814 * Retrieve the full list of active features from the MOS and check if
4815 * they are all supported.
4817 error
= spa_ld_check_features(spa
, &missing_feat_write
);
4822 * Load several special directories from the MOS needed by the dsl_pool
4825 error
= spa_ld_load_special_directories(spa
);
4830 * Retrieve pool properties from the MOS.
4832 error
= spa_ld_get_props(spa
);
4837 * Retrieve the list of auxiliary devices - cache devices and spares -
4840 error
= spa_ld_open_aux_vdevs(spa
, type
);
4845 * Load the metadata for all vdevs. Also check if unopenable devices
4846 * should be autoreplaced.
4848 error
= spa_ld_load_vdev_metadata(spa
);
4852 error
= spa_ld_load_dedup_tables(spa
);
4857 * Verify the logs now to make sure we don't have any unexpected errors
4858 * when we claim log blocks later.
4860 error
= spa_ld_verify_logs(spa
, type
, ereport
);
4864 if (missing_feat_write
) {
4865 ASSERT(spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
);
4868 * At this point, we know that we can open the pool in
4869 * read-only mode but not read-write mode. We now have enough
4870 * information and can return to userland.
4872 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
4877 * Traverse the last txgs to make sure the pool was left off in a safe
4878 * state. When performing an extreme rewind, we verify the whole pool,
4879 * which can take a very long time.
4881 error
= spa_ld_verify_pool_data(spa
);
4886 * Calculate the deflated space for the pool. This must be done before
4887 * we write anything to the pool because we'd need to update the space
4888 * accounting using the deflated sizes.
4890 spa_update_dspace(spa
);
4893 * We have now retrieved all the information we needed to open the
4894 * pool. If we are importing the pool in read-write mode, a few
4895 * additional steps must be performed to finish the import.
4897 if (spa_writeable(spa
) && (spa
->spa_load_state
== SPA_LOAD_RECOVER
||
4898 spa
->spa_load_max_txg
== UINT64_MAX
)) {
4899 uint64_t config_cache_txg
= spa
->spa_config_txg
;
4901 ASSERT(spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
);
4904 * In case of a checkpoint rewind, log the original txg
4905 * of the checkpointed uberblock.
4907 if (checkpoint_rewind
) {
4908 spa_history_log_internal(spa
, "checkpoint rewind",
4909 NULL
, "rewound state to txg=%llu",
4910 (u_longlong_t
)spa
->spa_uberblock
.ub_checkpoint_txg
);
4914 * Traverse the ZIL and claim all blocks.
4916 spa_ld_claim_log_blocks(spa
);
4919 * Kick-off the syncing thread.
4921 spa
->spa_sync_on
= B_TRUE
;
4922 txg_sync_start(spa
->spa_dsl_pool
);
4923 mmp_thread_start(spa
);
4926 * Wait for all claims to sync. We sync up to the highest
4927 * claimed log block birth time so that claimed log blocks
4928 * don't appear to be from the future. spa_claim_max_txg
4929 * will have been set for us by ZIL traversal operations
4932 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
4935 * Check if we need to request an update of the config. On the
4936 * next sync, we would update the config stored in vdev labels
4937 * and the cachefile (by default /etc/zfs/zpool.cache).
4939 spa_ld_check_for_config_update(spa
, config_cache_txg
,
4940 update_config_cache
);
4943 * Check if a rebuild was in progress and if so resume it.
4944 * Then check all DTLs to see if anything needs resilvering.
4945 * The resilver will be deferred if a rebuild was started.
4947 if (vdev_rebuild_active(spa
->spa_root_vdev
)) {
4948 vdev_rebuild_restart(spa
);
4949 } else if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4950 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
4951 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
4955 * Log the fact that we booted up (so that we can detect if
4956 * we rebooted in the middle of an operation).
4958 spa_history_log_version(spa
, "open", NULL
);
4960 spa_restart_removal(spa
);
4961 spa_spawn_aux_threads(spa
);
4964 * Delete any inconsistent datasets.
4967 * Since we may be issuing deletes for clones here,
4968 * we make sure to do so after we've spawned all the
4969 * auxiliary threads above (from which the livelist
4970 * deletion zthr is part of).
4972 (void) dmu_objset_find(spa_name(spa
),
4973 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
4976 * Clean up any stale temporary dataset userrefs.
4978 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
4980 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4981 vdev_initialize_restart(spa
->spa_root_vdev
);
4982 vdev_trim_restart(spa
->spa_root_vdev
);
4983 vdev_autotrim_restart(spa
);
4984 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4987 spa_import_progress_remove(spa_guid(spa
));
4988 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
4990 spa_load_note(spa
, "LOADED");
4996 spa_load_retry(spa_t
*spa
, spa_load_state_t state
)
4998 spa_mode_t mode
= spa
->spa_mode
;
5001 spa_deactivate(spa
);
5003 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
5005 spa_activate(spa
, mode
);
5006 spa_async_suspend(spa
);
5008 spa_load_note(spa
, "spa_load_retry: rewind, max txg: %llu",
5009 (u_longlong_t
)spa
->spa_load_max_txg
);
5011 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
));
5015 * If spa_load() fails this function will try loading prior txg's. If
5016 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
5017 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
5018 * function will not rewind the pool and will return the same error as
5022 spa_load_best(spa_t
*spa
, spa_load_state_t state
, uint64_t max_request
,
5025 nvlist_t
*loadinfo
= NULL
;
5026 nvlist_t
*config
= NULL
;
5027 int load_error
, rewind_error
;
5028 uint64_t safe_rewind_txg
;
5031 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
5032 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
5033 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5035 spa
->spa_load_max_txg
= max_request
;
5036 if (max_request
!= UINT64_MAX
)
5037 spa
->spa_extreme_rewind
= B_TRUE
;
5040 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
);
5041 if (load_error
== 0)
5043 if (load_error
== ZFS_ERR_NO_CHECKPOINT
) {
5045 * When attempting checkpoint-rewind on a pool with no
5046 * checkpoint, we should not attempt to load uberblocks
5047 * from previous txgs when spa_load fails.
5049 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
5050 spa_import_progress_remove(spa_guid(spa
));
5051 return (load_error
);
5054 if (spa
->spa_root_vdev
!= NULL
)
5055 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5057 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
5058 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
5060 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
5061 nvlist_free(config
);
5062 spa_import_progress_remove(spa_guid(spa
));
5063 return (load_error
);
5066 if (state
== SPA_LOAD_RECOVER
) {
5067 /* Price of rolling back is discarding txgs, including log */
5068 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5071 * If we aren't rolling back save the load info from our first
5072 * import attempt so that we can restore it after attempting
5075 loadinfo
= spa
->spa_load_info
;
5076 spa
->spa_load_info
= fnvlist_alloc();
5079 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
5080 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
5081 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
5082 TXG_INITIAL
: safe_rewind_txg
;
5085 * Continue as long as we're finding errors, we're still within
5086 * the acceptable rewind range, and we're still finding uberblocks
5088 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
5089 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
5090 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
5091 spa
->spa_extreme_rewind
= B_TRUE
;
5092 rewind_error
= spa_load_retry(spa
, state
);
5095 spa
->spa_extreme_rewind
= B_FALSE
;
5096 spa
->spa_load_max_txg
= UINT64_MAX
;
5098 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
5099 spa_config_set(spa
, config
);
5101 nvlist_free(config
);
5103 if (state
== SPA_LOAD_RECOVER
) {
5104 ASSERT3P(loadinfo
, ==, NULL
);
5105 spa_import_progress_remove(spa_guid(spa
));
5106 return (rewind_error
);
5108 /* Store the rewind info as part of the initial load info */
5109 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
5110 spa
->spa_load_info
);
5112 /* Restore the initial load info */
5113 fnvlist_free(spa
->spa_load_info
);
5114 spa
->spa_load_info
= loadinfo
;
5116 spa_import_progress_remove(spa_guid(spa
));
5117 return (load_error
);
5124 * The import case is identical to an open except that the configuration is sent
5125 * down from userland, instead of grabbed from the configuration cache. For the
5126 * case of an open, the pool configuration will exist in the
5127 * POOL_STATE_UNINITIALIZED state.
5129 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5130 * the same time open the pool, without having to keep around the spa_t in some
5134 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
5138 spa_load_state_t state
= SPA_LOAD_OPEN
;
5140 int locked
= B_FALSE
;
5141 int firstopen
= B_FALSE
;
5146 * As disgusting as this is, we need to support recursive calls to this
5147 * function because dsl_dir_open() is called during spa_load(), and ends
5148 * up calling spa_open() again. The real fix is to figure out how to
5149 * avoid dsl_dir_open() calling this in the first place.
5151 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
5152 mutex_enter(&spa_namespace_lock
);
5156 if ((spa
= spa_lookup(pool
)) == NULL
) {
5158 mutex_exit(&spa_namespace_lock
);
5159 return (SET_ERROR(ENOENT
));
5162 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
5163 zpool_load_policy_t policy
;
5167 zpool_get_load_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
5169 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
5170 state
= SPA_LOAD_RECOVER
;
5172 spa_activate(spa
, spa_mode_global
);
5174 if (state
!= SPA_LOAD_RECOVER
)
5175 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
5176 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
5178 zfs_dbgmsg("spa_open_common: opening %s", pool
);
5179 error
= spa_load_best(spa
, state
, policy
.zlp_txg
,
5182 if (error
== EBADF
) {
5184 * If vdev_validate() returns failure (indicated by
5185 * EBADF), it indicates that one of the vdevs indicates
5186 * that the pool has been exported or destroyed. If
5187 * this is the case, the config cache is out of sync and
5188 * we should remove the pool from the namespace.
5191 spa_deactivate(spa
);
5192 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
5195 mutex_exit(&spa_namespace_lock
);
5196 return (SET_ERROR(ENOENT
));
5201 * We can't open the pool, but we still have useful
5202 * information: the state of each vdev after the
5203 * attempted vdev_open(). Return this to the user.
5205 if (config
!= NULL
&& spa
->spa_config
) {
5206 *config
= fnvlist_dup(spa
->spa_config
);
5207 fnvlist_add_nvlist(*config
,
5208 ZPOOL_CONFIG_LOAD_INFO
,
5209 spa
->spa_load_info
);
5212 spa_deactivate(spa
);
5213 spa
->spa_last_open_failed
= error
;
5215 mutex_exit(&spa_namespace_lock
);
5221 spa_open_ref(spa
, tag
);
5224 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5227 * If we've recovered the pool, pass back any information we
5228 * gathered while doing the load.
5230 if (state
== SPA_LOAD_RECOVER
) {
5231 fnvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
5232 spa
->spa_load_info
);
5236 spa
->spa_last_open_failed
= 0;
5237 spa
->spa_last_ubsync_txg
= 0;
5238 spa
->spa_load_txg
= 0;
5239 mutex_exit(&spa_namespace_lock
);
5243 zvol_create_minors_recursive(spa_name(spa
));
5251 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
5254 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
5258 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
5260 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
5264 * Lookup the given spa_t, incrementing the inject count in the process,
5265 * preventing it from being exported or destroyed.
5268 spa_inject_addref(char *name
)
5272 mutex_enter(&spa_namespace_lock
);
5273 if ((spa
= spa_lookup(name
)) == NULL
) {
5274 mutex_exit(&spa_namespace_lock
);
5277 spa
->spa_inject_ref
++;
5278 mutex_exit(&spa_namespace_lock
);
5284 spa_inject_delref(spa_t
*spa
)
5286 mutex_enter(&spa_namespace_lock
);
5287 spa
->spa_inject_ref
--;
5288 mutex_exit(&spa_namespace_lock
);
5292 * Add spares device information to the nvlist.
5295 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
5305 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5307 if (spa
->spa_spares
.sav_count
== 0)
5310 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5311 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5312 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
));
5314 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5315 (const nvlist_t
* const *)spares
, nspares
);
5316 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5317 &spares
, &nspares
));
5320 * Go through and find any spares which have since been
5321 * repurposed as an active spare. If this is the case, update
5322 * their status appropriately.
5324 for (i
= 0; i
< nspares
; i
++) {
5325 guid
= fnvlist_lookup_uint64(spares
[i
],
5327 if (spa_spare_exists(guid
, &pool
, NULL
) &&
5329 VERIFY0(nvlist_lookup_uint64_array(spares
[i
],
5330 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
,
5332 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
5333 vs
->vs_aux
= VDEV_AUX_SPARED
;
5340 * Add l2cache device information to the nvlist, including vdev stats.
5343 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
5346 uint_t i
, j
, nl2cache
;
5353 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5355 if (spa
->spa_l2cache
.sav_count
== 0)
5358 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5359 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5360 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
));
5361 if (nl2cache
!= 0) {
5362 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5363 (const nvlist_t
* const *)l2cache
, nl2cache
);
5364 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5365 &l2cache
, &nl2cache
));
5368 * Update level 2 cache device stats.
5371 for (i
= 0; i
< nl2cache
; i
++) {
5372 guid
= fnvlist_lookup_uint64(l2cache
[i
],
5376 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
5378 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
5379 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
5385 VERIFY0(nvlist_lookup_uint64_array(l2cache
[i
],
5386 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
));
5387 vdev_get_stats(vd
, vs
);
5388 vdev_config_generate_stats(vd
, l2cache
[i
]);
5395 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
5400 if (spa
->spa_feat_for_read_obj
!= 0) {
5401 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5402 spa
->spa_feat_for_read_obj
);
5403 zap_cursor_retrieve(&zc
, &za
) == 0;
5404 zap_cursor_advance(&zc
)) {
5405 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5406 za
.za_num_integers
== 1);
5407 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5408 za
.za_first_integer
));
5410 zap_cursor_fini(&zc
);
5413 if (spa
->spa_feat_for_write_obj
!= 0) {
5414 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5415 spa
->spa_feat_for_write_obj
);
5416 zap_cursor_retrieve(&zc
, &za
) == 0;
5417 zap_cursor_advance(&zc
)) {
5418 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5419 za
.za_num_integers
== 1);
5420 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5421 za
.za_first_integer
));
5423 zap_cursor_fini(&zc
);
5428 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
5432 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5433 zfeature_info_t feature
= spa_feature_table
[i
];
5436 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
5439 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
5444 * Store a list of pool features and their reference counts in the
5447 * The first time this is called on a spa, allocate a new nvlist, fetch
5448 * the pool features and reference counts from disk, then save the list
5449 * in the spa. In subsequent calls on the same spa use the saved nvlist
5450 * and refresh its values from the cached reference counts. This
5451 * ensures we don't block here on I/O on a suspended pool so 'zpool
5452 * clear' can resume the pool.
5455 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
5459 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5461 mutex_enter(&spa
->spa_feat_stats_lock
);
5462 features
= spa
->spa_feat_stats
;
5464 if (features
!= NULL
) {
5465 spa_feature_stats_from_cache(spa
, features
);
5467 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
5468 spa
->spa_feat_stats
= features
;
5469 spa_feature_stats_from_disk(spa
, features
);
5472 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
5475 mutex_exit(&spa
->spa_feat_stats_lock
);
5479 spa_get_stats(const char *name
, nvlist_t
**config
,
5480 char *altroot
, size_t buflen
)
5486 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
5490 * This still leaves a window of inconsistency where the spares
5491 * or l2cache devices could change and the config would be
5492 * self-inconsistent.
5494 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5496 if (*config
!= NULL
) {
5497 uint64_t loadtimes
[2];
5499 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
5500 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
5501 fnvlist_add_uint64_array(*config
,
5502 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2);
5504 fnvlist_add_uint64(*config
,
5505 ZPOOL_CONFIG_ERRCOUNT
,
5506 spa_get_errlog_size(spa
));
5508 if (spa_suspended(spa
)) {
5509 fnvlist_add_uint64(*config
,
5510 ZPOOL_CONFIG_SUSPENDED
,
5512 fnvlist_add_uint64(*config
,
5513 ZPOOL_CONFIG_SUSPENDED_REASON
,
5514 spa
->spa_suspended
);
5517 spa_add_spares(spa
, *config
);
5518 spa_add_l2cache(spa
, *config
);
5519 spa_add_feature_stats(spa
, *config
);
5524 * We want to get the alternate root even for faulted pools, so we cheat
5525 * and call spa_lookup() directly.
5529 mutex_enter(&spa_namespace_lock
);
5530 spa
= spa_lookup(name
);
5532 spa_altroot(spa
, altroot
, buflen
);
5536 mutex_exit(&spa_namespace_lock
);
5538 spa_altroot(spa
, altroot
, buflen
);
5543 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5544 spa_close(spa
, FTAG
);
5551 * Validate that the auxiliary device array is well formed. We must have an
5552 * array of nvlists, each which describes a valid leaf vdev. If this is an
5553 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
5554 * specified, as long as they are well-formed.
5557 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
5558 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
5559 vdev_labeltype_t label
)
5566 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5569 * It's acceptable to have no devs specified.
5571 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
5575 return (SET_ERROR(EINVAL
));
5578 * Make sure the pool is formatted with a version that supports this
5581 if (spa_version(spa
) < version
)
5582 return (SET_ERROR(ENOTSUP
));
5585 * Set the pending device list so we correctly handle device in-use
5588 sav
->sav_pending
= dev
;
5589 sav
->sav_npending
= ndev
;
5591 for (i
= 0; i
< ndev
; i
++) {
5592 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
5596 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5598 error
= SET_ERROR(EINVAL
);
5604 if ((error
= vdev_open(vd
)) == 0 &&
5605 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
5606 fnvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
5613 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
5620 sav
->sav_pending
= NULL
;
5621 sav
->sav_npending
= 0;
5626 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
5630 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5632 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5633 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
5634 VDEV_LABEL_SPARE
)) != 0) {
5638 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5639 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
5640 VDEV_LABEL_L2CACHE
));
5644 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
5649 if (sav
->sav_config
!= NULL
) {
5655 * Generate new dev list by concatenating with the
5658 VERIFY0(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
5659 &olddevs
, &oldndevs
));
5661 newdevs
= kmem_alloc(sizeof (void *) *
5662 (ndevs
+ oldndevs
), KM_SLEEP
);
5663 for (i
= 0; i
< oldndevs
; i
++)
5664 newdevs
[i
] = fnvlist_dup(olddevs
[i
]);
5665 for (i
= 0; i
< ndevs
; i
++)
5666 newdevs
[i
+ oldndevs
] = fnvlist_dup(devs
[i
]);
5668 fnvlist_remove(sav
->sav_config
, config
);
5670 fnvlist_add_nvlist_array(sav
->sav_config
, config
,
5671 (const nvlist_t
* const *)newdevs
, ndevs
+ oldndevs
);
5672 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
5673 nvlist_free(newdevs
[i
]);
5674 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
5677 * Generate a new dev list.
5679 sav
->sav_config
= fnvlist_alloc();
5680 fnvlist_add_nvlist_array(sav
->sav_config
, config
,
5681 (const nvlist_t
* const *)devs
, ndevs
);
5686 * Stop and drop level 2 ARC devices
5689 spa_l2cache_drop(spa_t
*spa
)
5693 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
5695 for (i
= 0; i
< sav
->sav_count
; i
++) {
5698 vd
= sav
->sav_vdevs
[i
];
5701 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
5702 pool
!= 0ULL && l2arc_vdev_present(vd
))
5703 l2arc_remove_vdev(vd
);
5708 * Verify encryption parameters for spa creation. If we are encrypting, we must
5709 * have the encryption feature flag enabled.
5712 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
5713 boolean_t has_encryption
)
5715 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
5716 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
5718 return (SET_ERROR(ENOTSUP
));
5720 return (dmu_objset_create_crypt_check(NULL
, dcp
, NULL
));
5727 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
5728 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
5731 char *altroot
= NULL
;
5736 uint64_t txg
= TXG_INITIAL
;
5737 nvlist_t
**spares
, **l2cache
;
5738 uint_t nspares
, nl2cache
;
5739 uint64_t version
, obj
, ndraid
= 0;
5740 boolean_t has_features
;
5741 boolean_t has_encryption
;
5742 boolean_t has_allocclass
;
5748 if (props
== NULL
||
5749 nvlist_lookup_string(props
, "tname", &poolname
) != 0)
5750 poolname
= (char *)pool
;
5753 * If this pool already exists, return failure.
5755 mutex_enter(&spa_namespace_lock
);
5756 if (spa_lookup(poolname
) != NULL
) {
5757 mutex_exit(&spa_namespace_lock
);
5758 return (SET_ERROR(EEXIST
));
5762 * Allocate a new spa_t structure.
5764 nvl
= fnvlist_alloc();
5765 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
5766 (void) nvlist_lookup_string(props
,
5767 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5768 spa
= spa_add(poolname
, nvl
, altroot
);
5770 spa_activate(spa
, spa_mode_global
);
5772 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
5773 spa_deactivate(spa
);
5775 mutex_exit(&spa_namespace_lock
);
5780 * Temporary pool names should never be written to disk.
5782 if (poolname
!= pool
)
5783 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
5785 has_features
= B_FALSE
;
5786 has_encryption
= B_FALSE
;
5787 has_allocclass
= B_FALSE
;
5788 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
5789 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
5790 if (zpool_prop_feature(nvpair_name(elem
))) {
5791 has_features
= B_TRUE
;
5793 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
5794 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
5795 if (feat
== SPA_FEATURE_ENCRYPTION
)
5796 has_encryption
= B_TRUE
;
5797 if (feat
== SPA_FEATURE_ALLOCATION_CLASSES
)
5798 has_allocclass
= B_TRUE
;
5802 /* verify encryption params, if they were provided */
5804 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
5806 spa_deactivate(spa
);
5808 mutex_exit(&spa_namespace_lock
);
5812 if (!has_allocclass
&& zfs_special_devs(nvroot
, NULL
)) {
5813 spa_deactivate(spa
);
5815 mutex_exit(&spa_namespace_lock
);
5819 if (has_features
|| nvlist_lookup_uint64(props
,
5820 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
5821 version
= SPA_VERSION
;
5823 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
5825 spa
->spa_first_txg
= txg
;
5826 spa
->spa_uberblock
.ub_txg
= txg
- 1;
5827 spa
->spa_uberblock
.ub_version
= version
;
5828 spa
->spa_ubsync
= spa
->spa_uberblock
;
5829 spa
->spa_load_state
= SPA_LOAD_CREATE
;
5830 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
5831 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
5832 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
5833 spa
->spa_indirect_vdevs_loaded
= B_TRUE
;
5836 * Create "The Godfather" zio to hold all async IOs
5838 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
5840 for (int i
= 0; i
< max_ncpus
; i
++) {
5841 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
5842 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
5843 ZIO_FLAG_GODFATHER
);
5847 * Create the root vdev.
5849 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5851 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
5853 ASSERT(error
!= 0 || rvd
!= NULL
);
5854 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
5856 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
5857 error
= SET_ERROR(EINVAL
);
5860 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
5861 (error
= vdev_draid_spare_create(nvroot
, rvd
, &ndraid
, 0)) == 0 &&
5862 (error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) == 0) {
5864 * instantiate the metaslab groups (this will dirty the vdevs)
5865 * we can no longer error exit past this point
5867 for (int c
= 0; error
== 0 && c
< rvd
->vdev_children
; c
++) {
5868 vdev_t
*vd
= rvd
->vdev_child
[c
];
5870 vdev_metaslab_set_size(vd
);
5871 vdev_expand(vd
, txg
);
5875 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5879 spa_deactivate(spa
);
5881 mutex_exit(&spa_namespace_lock
);
5886 * Get the list of spares, if specified.
5888 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5889 &spares
, &nspares
) == 0) {
5890 spa
->spa_spares
.sav_config
= fnvlist_alloc();
5891 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
5892 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
5894 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5895 spa_load_spares(spa
);
5896 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5897 spa
->spa_spares
.sav_sync
= B_TRUE
;
5901 * Get the list of level 2 cache devices, if specified.
5903 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5904 &l2cache
, &nl2cache
) == 0) {
5905 VERIFY0(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
5906 NV_UNIQUE_NAME
, KM_SLEEP
));
5907 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
5908 ZPOOL_CONFIG_L2CACHE
, (const nvlist_t
* const *)l2cache
,
5910 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5911 spa_load_l2cache(spa
);
5912 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5913 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5916 spa
->spa_is_initializing
= B_TRUE
;
5917 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
5918 spa
->spa_is_initializing
= B_FALSE
;
5921 * Create DDTs (dedup tables).
5925 spa_update_dspace(spa
);
5927 tx
= dmu_tx_create_assigned(dp
, txg
);
5930 * Create the pool's history object.
5932 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
5933 spa_history_create_obj(spa
, tx
);
5935 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
5936 spa_history_log_version(spa
, "create", tx
);
5939 * Create the pool config object.
5941 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
5942 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
5943 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
5945 if (zap_add(spa
->spa_meta_objset
,
5946 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
5947 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
5948 cmn_err(CE_PANIC
, "failed to add pool config");
5951 if (zap_add(spa
->spa_meta_objset
,
5952 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
5953 sizeof (uint64_t), 1, &version
, tx
) != 0) {
5954 cmn_err(CE_PANIC
, "failed to add pool version");
5957 /* Newly created pools with the right version are always deflated. */
5958 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
5959 spa
->spa_deflate
= TRUE
;
5960 if (zap_add(spa
->spa_meta_objset
,
5961 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
5962 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
5963 cmn_err(CE_PANIC
, "failed to add deflate");
5968 * Create the deferred-free bpobj. Turn off compression
5969 * because sync-to-convergence takes longer if the blocksize
5972 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
5973 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
5974 ZIO_COMPRESS_OFF
, tx
);
5975 if (zap_add(spa
->spa_meta_objset
,
5976 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
5977 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
5978 cmn_err(CE_PANIC
, "failed to add bpobj");
5980 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
5981 spa
->spa_meta_objset
, obj
));
5984 * Generate some random noise for salted checksums to operate on.
5986 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
5987 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
5990 * Set pool properties.
5992 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
5993 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
5994 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
5995 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
5996 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
5997 spa
->spa_autotrim
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM
);
5999 if (props
!= NULL
) {
6000 spa_configfile_set(spa
, props
, B_FALSE
);
6001 spa_sync_props(props
, tx
);
6004 for (int i
= 0; i
< ndraid
; i
++)
6005 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
6009 spa
->spa_sync_on
= B_TRUE
;
6011 mmp_thread_start(spa
);
6012 txg_wait_synced(dp
, txg
);
6014 spa_spawn_aux_threads(spa
);
6016 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
6019 * Don't count references from objsets that are already closed
6020 * and are making their way through the eviction process.
6022 spa_evicting_os_wait(spa
);
6023 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
6024 spa
->spa_load_state
= SPA_LOAD_NONE
;
6026 mutex_exit(&spa_namespace_lock
);
6032 * Import a non-root pool into the system.
6035 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
6038 char *altroot
= NULL
;
6039 spa_load_state_t state
= SPA_LOAD_IMPORT
;
6040 zpool_load_policy_t policy
;
6041 spa_mode_t mode
= spa_mode_global
;
6042 uint64_t readonly
= B_FALSE
;
6045 nvlist_t
**spares
, **l2cache
;
6046 uint_t nspares
, nl2cache
;
6049 * If a pool with this name exists, return failure.
6051 mutex_enter(&spa_namespace_lock
);
6052 if (spa_lookup(pool
) != NULL
) {
6053 mutex_exit(&spa_namespace_lock
);
6054 return (SET_ERROR(EEXIST
));
6058 * Create and initialize the spa structure.
6060 (void) nvlist_lookup_string(props
,
6061 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
6062 (void) nvlist_lookup_uint64(props
,
6063 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
6065 mode
= SPA_MODE_READ
;
6066 spa
= spa_add(pool
, config
, altroot
);
6067 spa
->spa_import_flags
= flags
;
6070 * Verbatim import - Take a pool and insert it into the namespace
6071 * as if it had been loaded at boot.
6073 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
6075 spa_configfile_set(spa
, props
, B_FALSE
);
6077 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
6078 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6079 zfs_dbgmsg("spa_import: verbatim import of %s", pool
);
6080 mutex_exit(&spa_namespace_lock
);
6084 spa_activate(spa
, mode
);
6087 * Don't start async tasks until we know everything is healthy.
6089 spa_async_suspend(spa
);
6091 zpool_get_load_policy(config
, &policy
);
6092 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
6093 state
= SPA_LOAD_RECOVER
;
6095 spa
->spa_config_source
= SPA_CONFIG_SRC_TRYIMPORT
;
6097 if (state
!= SPA_LOAD_RECOVER
) {
6098 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
6099 zfs_dbgmsg("spa_import: importing %s", pool
);
6101 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
6102 "(RECOVERY MODE)", pool
, (longlong_t
)policy
.zlp_txg
);
6104 error
= spa_load_best(spa
, state
, policy
.zlp_txg
, policy
.zlp_rewind
);
6107 * Propagate anything learned while loading the pool and pass it
6108 * back to caller (i.e. rewind info, missing devices, etc).
6110 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
, spa
->spa_load_info
);
6112 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6114 * Toss any existing sparelist, as it doesn't have any validity
6115 * anymore, and conflicts with spa_has_spare().
6117 if (spa
->spa_spares
.sav_config
) {
6118 nvlist_free(spa
->spa_spares
.sav_config
);
6119 spa
->spa_spares
.sav_config
= NULL
;
6120 spa_load_spares(spa
);
6122 if (spa
->spa_l2cache
.sav_config
) {
6123 nvlist_free(spa
->spa_l2cache
.sav_config
);
6124 spa
->spa_l2cache
.sav_config
= NULL
;
6125 spa_load_l2cache(spa
);
6128 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
6129 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6132 spa_configfile_set(spa
, props
, B_FALSE
);
6134 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
6135 (error
= spa_prop_set(spa
, props
)))) {
6137 spa_deactivate(spa
);
6139 mutex_exit(&spa_namespace_lock
);
6143 spa_async_resume(spa
);
6146 * Override any spares and level 2 cache devices as specified by
6147 * the user, as these may have correct device names/devids, etc.
6149 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
6150 &spares
, &nspares
) == 0) {
6151 if (spa
->spa_spares
.sav_config
)
6152 fnvlist_remove(spa
->spa_spares
.sav_config
,
6153 ZPOOL_CONFIG_SPARES
);
6155 spa
->spa_spares
.sav_config
= fnvlist_alloc();
6156 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
6157 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
6159 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6160 spa_load_spares(spa
);
6161 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6162 spa
->spa_spares
.sav_sync
= B_TRUE
;
6164 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
6165 &l2cache
, &nl2cache
) == 0) {
6166 if (spa
->spa_l2cache
.sav_config
)
6167 fnvlist_remove(spa
->spa_l2cache
.sav_config
,
6168 ZPOOL_CONFIG_L2CACHE
);
6170 spa
->spa_l2cache
.sav_config
= fnvlist_alloc();
6171 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
6172 ZPOOL_CONFIG_L2CACHE
, (const nvlist_t
* const *)l2cache
,
6174 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6175 spa_load_l2cache(spa
);
6176 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6177 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6181 * Check for any removed devices.
6183 if (spa
->spa_autoreplace
) {
6184 spa_aux_check_removed(&spa
->spa_spares
);
6185 spa_aux_check_removed(&spa
->spa_l2cache
);
6188 if (spa_writeable(spa
)) {
6190 * Update the config cache to include the newly-imported pool.
6192 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6196 * It's possible that the pool was expanded while it was exported.
6197 * We kick off an async task to handle this for us.
6199 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
6201 spa_history_log_version(spa
, "import", NULL
);
6203 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6205 mutex_exit(&spa_namespace_lock
);
6207 zvol_create_minors_recursive(pool
);
6213 spa_tryimport(nvlist_t
*tryconfig
)
6215 nvlist_t
*config
= NULL
;
6216 char *poolname
, *cachefile
;
6220 zpool_load_policy_t policy
;
6222 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
6225 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
6229 * Create and initialize the spa structure.
6231 mutex_enter(&spa_namespace_lock
);
6232 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
6233 spa_activate(spa
, SPA_MODE_READ
);
6236 * Rewind pool if a max txg was provided.
6238 zpool_get_load_policy(spa
->spa_config
, &policy
);
6239 if (policy
.zlp_txg
!= UINT64_MAX
) {
6240 spa
->spa_load_max_txg
= policy
.zlp_txg
;
6241 spa
->spa_extreme_rewind
= B_TRUE
;
6242 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6243 poolname
, (longlong_t
)policy
.zlp_txg
);
6245 zfs_dbgmsg("spa_tryimport: importing %s", poolname
);
6248 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_CACHEFILE
, &cachefile
)
6250 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile
);
6251 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
6253 spa
->spa_config_source
= SPA_CONFIG_SRC_SCAN
;
6256 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
);
6259 * If 'tryconfig' was at least parsable, return the current config.
6261 if (spa
->spa_root_vdev
!= NULL
) {
6262 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
6263 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, poolname
);
6264 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, state
);
6265 fnvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
6266 spa
->spa_uberblock
.ub_timestamp
);
6267 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
6268 spa
->spa_load_info
);
6269 fnvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
6273 * If the bootfs property exists on this pool then we
6274 * copy it out so that external consumers can tell which
6275 * pools are bootable.
6277 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
6278 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6281 * We have to play games with the name since the
6282 * pool was opened as TRYIMPORT_NAME.
6284 if (dsl_dsobj_to_dsname(spa_name(spa
),
6285 spa
->spa_bootfs
, tmpname
) == 0) {
6289 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6291 cp
= strchr(tmpname
, '/');
6293 (void) strlcpy(dsname
, tmpname
,
6296 (void) snprintf(dsname
, MAXPATHLEN
,
6297 "%s/%s", poolname
, ++cp
);
6299 fnvlist_add_string(config
, ZPOOL_CONFIG_BOOTFS
,
6301 kmem_free(dsname
, MAXPATHLEN
);
6303 kmem_free(tmpname
, MAXPATHLEN
);
6307 * Add the list of hot spares and level 2 cache devices.
6309 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6310 spa_add_spares(spa
, config
);
6311 spa_add_l2cache(spa
, config
);
6312 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6316 spa_deactivate(spa
);
6318 mutex_exit(&spa_namespace_lock
);
6324 * Pool export/destroy
6326 * The act of destroying or exporting a pool is very simple. We make sure there
6327 * is no more pending I/O and any references to the pool are gone. Then, we
6328 * update the pool state and sync all the labels to disk, removing the
6329 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6330 * we don't sync the labels or remove the configuration cache.
6333 spa_export_common(const char *pool
, int new_state
, nvlist_t
**oldconfig
,
6334 boolean_t force
, boolean_t hardforce
)
6342 if (!(spa_mode_global
& SPA_MODE_WRITE
))
6343 return (SET_ERROR(EROFS
));
6345 mutex_enter(&spa_namespace_lock
);
6346 if ((spa
= spa_lookup(pool
)) == NULL
) {
6347 mutex_exit(&spa_namespace_lock
);
6348 return (SET_ERROR(ENOENT
));
6351 if (spa
->spa_is_exporting
) {
6352 /* the pool is being exported by another thread */
6353 mutex_exit(&spa_namespace_lock
);
6354 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS
));
6356 spa
->spa_is_exporting
= B_TRUE
;
6359 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6360 * reacquire the namespace lock, and see if we can export.
6362 spa_open_ref(spa
, FTAG
);
6363 mutex_exit(&spa_namespace_lock
);
6364 spa_async_suspend(spa
);
6365 if (spa
->spa_zvol_taskq
) {
6366 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
6367 taskq_wait(spa
->spa_zvol_taskq
);
6369 mutex_enter(&spa_namespace_lock
);
6370 spa_close(spa
, FTAG
);
6372 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
6375 * The pool will be in core if it's openable, in which case we can
6376 * modify its state. Objsets may be open only because they're dirty,
6377 * so we have to force it to sync before checking spa_refcnt.
6379 if (spa
->spa_sync_on
) {
6380 txg_wait_synced(spa
->spa_dsl_pool
, 0);
6381 spa_evicting_os_wait(spa
);
6385 * A pool cannot be exported or destroyed if there are active
6386 * references. If we are resetting a pool, allow references by
6387 * fault injection handlers.
6389 if (!spa_refcount_zero(spa
) || (spa
->spa_inject_ref
!= 0)) {
6390 error
= SET_ERROR(EBUSY
);
6394 if (spa
->spa_sync_on
) {
6396 * A pool cannot be exported if it has an active shared spare.
6397 * This is to prevent other pools stealing the active spare
6398 * from an exported pool. At user's own will, such pool can
6399 * be forcedly exported.
6401 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
6402 spa_has_active_shared_spare(spa
)) {
6403 error
= SET_ERROR(EXDEV
);
6408 * We're about to export or destroy this pool. Make sure
6409 * we stop all initialization and trim activity here before
6410 * we set the spa_final_txg. This will ensure that all
6411 * dirty data resulting from the initialization is
6412 * committed to disk before we unload the pool.
6414 if (spa
->spa_root_vdev
!= NULL
) {
6415 vdev_t
*rvd
= spa
->spa_root_vdev
;
6416 vdev_initialize_stop_all(rvd
, VDEV_INITIALIZE_ACTIVE
);
6417 vdev_trim_stop_all(rvd
, VDEV_TRIM_ACTIVE
);
6418 vdev_autotrim_stop_all(spa
);
6419 vdev_rebuild_stop_all(spa
);
6423 * We want this to be reflected on every label,
6424 * so mark them all dirty. spa_unload() will do the
6425 * final sync that pushes these changes out.
6427 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
6428 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6429 spa
->spa_state
= new_state
;
6430 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
6432 vdev_config_dirty(spa
->spa_root_vdev
);
6433 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6438 if (new_state
== POOL_STATE_DESTROYED
)
6439 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
6440 else if (new_state
== POOL_STATE_EXPORTED
)
6441 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
6443 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6445 spa_deactivate(spa
);
6448 if (oldconfig
&& spa
->spa_config
)
6449 *oldconfig
= fnvlist_dup(spa
->spa_config
);
6451 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
6453 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
6457 * If spa_remove() is not called for this spa_t and
6458 * there is any possibility that it can be reused,
6459 * we make sure to reset the exporting flag.
6461 spa
->spa_is_exporting
= B_FALSE
;
6464 mutex_exit(&spa_namespace_lock
);
6468 spa
->spa_is_exporting
= B_FALSE
;
6469 spa_async_resume(spa
);
6470 mutex_exit(&spa_namespace_lock
);
6475 * Destroy a storage pool.
6478 spa_destroy(const char *pool
)
6480 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
6485 * Export a storage pool.
6488 spa_export(const char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
6489 boolean_t hardforce
)
6491 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
6496 * Similar to spa_export(), this unloads the spa_t without actually removing it
6497 * from the namespace in any way.
6500 spa_reset(const char *pool
)
6502 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
6507 * ==========================================================================
6508 * Device manipulation
6509 * ==========================================================================
6513 * This is called as a synctask to increment the draid feature flag
6516 spa_draid_feature_incr(void *arg
, dmu_tx_t
*tx
)
6518 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6519 int draid
= (int)(uintptr_t)arg
;
6521 for (int c
= 0; c
< draid
; c
++)
6522 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
6526 * Add a device to a storage pool.
6529 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
6531 uint64_t txg
, ndraid
= 0;
6533 vdev_t
*rvd
= spa
->spa_root_vdev
;
6535 nvlist_t
**spares
, **l2cache
;
6536 uint_t nspares
, nl2cache
;
6538 ASSERT(spa_writeable(spa
));
6540 txg
= spa_vdev_enter(spa
);
6542 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
6543 VDEV_ALLOC_ADD
)) != 0)
6544 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6546 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
6548 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
6552 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
6556 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
6557 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6559 if (vd
->vdev_children
!= 0 &&
6560 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0) {
6561 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6565 * The virtual dRAID spares must be added after vdev tree is created
6566 * and the vdev guids are generated. The guid of their associated
6567 * dRAID is stored in the config and used when opening the spare.
6569 if ((error
= vdev_draid_spare_create(nvroot
, vd
, &ndraid
,
6570 rvd
->vdev_children
)) == 0) {
6571 if (ndraid
> 0 && nvlist_lookup_nvlist_array(nvroot
,
6572 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) != 0)
6575 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6579 * We must validate the spares and l2cache devices after checking the
6580 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
6582 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
6583 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6586 * If we are in the middle of a device removal, we can only add
6587 * devices which match the existing devices in the pool.
6588 * If we are in the middle of a removal, or have some indirect
6589 * vdevs, we can not add raidz or dRAID top levels.
6591 if (spa
->spa_vdev_removal
!= NULL
||
6592 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
6593 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6594 tvd
= vd
->vdev_child
[c
];
6595 if (spa
->spa_vdev_removal
!= NULL
&&
6596 tvd
->vdev_ashift
!= spa
->spa_max_ashift
) {
6597 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6599 /* Fail if top level vdev is raidz or a dRAID */
6600 if (vdev_get_nparity(tvd
) != 0)
6601 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6604 * Need the top level mirror to be
6605 * a mirror of leaf vdevs only
6607 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
6608 for (uint64_t cid
= 0;
6609 cid
< tvd
->vdev_children
; cid
++) {
6610 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
6611 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
6612 return (spa_vdev_exit(spa
, vd
,
6620 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6621 tvd
= vd
->vdev_child
[c
];
6622 vdev_remove_child(vd
, tvd
);
6623 tvd
->vdev_id
= rvd
->vdev_children
;
6624 vdev_add_child(rvd
, tvd
);
6625 vdev_config_dirty(tvd
);
6629 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
6630 ZPOOL_CONFIG_SPARES
);
6631 spa_load_spares(spa
);
6632 spa
->spa_spares
.sav_sync
= B_TRUE
;
6635 if (nl2cache
!= 0) {
6636 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
6637 ZPOOL_CONFIG_L2CACHE
);
6638 spa_load_l2cache(spa
);
6639 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6643 * We can't increment a feature while holding spa_vdev so we
6644 * have to do it in a synctask.
6649 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
6650 dsl_sync_task_nowait(spa
->spa_dsl_pool
, spa_draid_feature_incr
,
6651 (void *)(uintptr_t)ndraid
, tx
);
6656 * We have to be careful when adding new vdevs to an existing pool.
6657 * If other threads start allocating from these vdevs before we
6658 * sync the config cache, and we lose power, then upon reboot we may
6659 * fail to open the pool because there are DVAs that the config cache
6660 * can't translate. Therefore, we first add the vdevs without
6661 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
6662 * and then let spa_config_update() initialize the new metaslabs.
6664 * spa_load() checks for added-but-not-initialized vdevs, so that
6665 * if we lose power at any point in this sequence, the remaining
6666 * steps will be completed the next time we load the pool.
6668 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
6670 mutex_enter(&spa_namespace_lock
);
6671 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6672 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
6673 mutex_exit(&spa_namespace_lock
);
6679 * Attach a device to a mirror. The arguments are the path to any device
6680 * in the mirror, and the nvroot for the new device. If the path specifies
6681 * a device that is not mirrored, we automatically insert the mirror vdev.
6683 * If 'replacing' is specified, the new device is intended to replace the
6684 * existing device; in this case the two devices are made into their own
6685 * mirror using the 'replacing' vdev, which is functionally identical to
6686 * the mirror vdev (it actually reuses all the same ops) but has a few
6687 * extra rules: you can't attach to it after it's been created, and upon
6688 * completion of resilvering, the first disk (the one being replaced)
6689 * is automatically detached.
6691 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
6692 * should be performed instead of traditional healing reconstruction. From
6693 * an administrators perspective these are both resilver operations.
6696 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
,
6699 uint64_t txg
, dtl_max_txg
;
6700 vdev_t
*rvd
= spa
->spa_root_vdev
;
6701 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
6703 char *oldvdpath
, *newvdpath
;
6707 ASSERT(spa_writeable(spa
));
6709 txg
= spa_vdev_enter(spa
);
6711 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
6713 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
6714 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
6715 error
= (spa_has_checkpoint(spa
)) ?
6716 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
6717 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6721 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_DEVICE_REBUILD
))
6722 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6724 if (dsl_scan_resilvering(spa_get_dsl(spa
)))
6725 return (spa_vdev_exit(spa
, NULL
, txg
,
6726 ZFS_ERR_RESILVER_IN_PROGRESS
));
6728 if (vdev_rebuild_active(rvd
))
6729 return (spa_vdev_exit(spa
, NULL
, txg
,
6730 ZFS_ERR_REBUILD_IN_PROGRESS
));
6733 if (spa
->spa_vdev_removal
!= NULL
)
6734 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
6737 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
6739 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
6740 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6742 pvd
= oldvd
->vdev_parent
;
6744 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
6745 VDEV_ALLOC_ATTACH
)) != 0)
6746 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
6748 if (newrootvd
->vdev_children
!= 1)
6749 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
6751 newvd
= newrootvd
->vdev_child
[0];
6753 if (!newvd
->vdev_ops
->vdev_op_leaf
)
6754 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
6756 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
6757 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
6760 * Spares can't replace logs
6762 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
6763 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6766 * A dRAID spare can only replace a child of its parent dRAID vdev.
6768 if (newvd
->vdev_ops
== &vdev_draid_spare_ops
&&
6769 oldvd
->vdev_top
!= vdev_draid_spare_get_parent(newvd
)) {
6770 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6775 * For rebuilds, the top vdev must support reconstruction
6776 * using only space maps. This means the only allowable
6777 * vdevs types are the root vdev, a mirror, or dRAID.
6780 if (pvd
->vdev_top
!= NULL
)
6781 tvd
= pvd
->vdev_top
;
6783 if (tvd
->vdev_ops
!= &vdev_mirror_ops
&&
6784 tvd
->vdev_ops
!= &vdev_root_ops
&&
6785 tvd
->vdev_ops
!= &vdev_draid_ops
) {
6786 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6792 * For attach, the only allowable parent is a mirror or the root
6795 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
6796 pvd
->vdev_ops
!= &vdev_root_ops
)
6797 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6799 pvops
= &vdev_mirror_ops
;
6802 * Active hot spares can only be replaced by inactive hot
6805 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
6806 oldvd
->vdev_isspare
&&
6807 !spa_has_spare(spa
, newvd
->vdev_guid
))
6808 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6811 * If the source is a hot spare, and the parent isn't already a
6812 * spare, then we want to create a new hot spare. Otherwise, we
6813 * want to create a replacing vdev. The user is not allowed to
6814 * attach to a spared vdev child unless the 'isspare' state is
6815 * the same (spare replaces spare, non-spare replaces
6818 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
6819 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
6820 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6821 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
6822 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
6823 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6826 if (newvd
->vdev_isspare
)
6827 pvops
= &vdev_spare_ops
;
6829 pvops
= &vdev_replacing_ops
;
6833 * Make sure the new device is big enough.
6835 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
6836 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
6839 * The new device cannot have a higher alignment requirement
6840 * than the top-level vdev.
6842 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
6843 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
6846 * If this is an in-place replacement, update oldvd's path and devid
6847 * to make it distinguishable from newvd, and unopenable from now on.
6849 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
6850 spa_strfree(oldvd
->vdev_path
);
6851 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
6853 (void) snprintf(oldvd
->vdev_path
, strlen(newvd
->vdev_path
) + 5,
6854 "%s/%s", newvd
->vdev_path
, "old");
6855 if (oldvd
->vdev_devid
!= NULL
) {
6856 spa_strfree(oldvd
->vdev_devid
);
6857 oldvd
->vdev_devid
= NULL
;
6862 * If the parent is not a mirror, or if we're replacing, insert the new
6863 * mirror/replacing/spare vdev above oldvd.
6865 if (pvd
->vdev_ops
!= pvops
)
6866 pvd
= vdev_add_parent(oldvd
, pvops
);
6868 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
6869 ASSERT(pvd
->vdev_ops
== pvops
);
6870 ASSERT(oldvd
->vdev_parent
== pvd
);
6873 * Extract the new device from its root and add it to pvd.
6875 vdev_remove_child(newrootvd
, newvd
);
6876 newvd
->vdev_id
= pvd
->vdev_children
;
6877 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
6878 vdev_add_child(pvd
, newvd
);
6881 * Reevaluate the parent vdev state.
6883 vdev_propagate_state(pvd
);
6885 tvd
= newvd
->vdev_top
;
6886 ASSERT(pvd
->vdev_top
== tvd
);
6887 ASSERT(tvd
->vdev_parent
== rvd
);
6889 vdev_config_dirty(tvd
);
6892 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6893 * for any dmu_sync-ed blocks. It will propagate upward when
6894 * spa_vdev_exit() calls vdev_dtl_reassess().
6896 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
6898 vdev_dtl_dirty(newvd
, DTL_MISSING
,
6899 TXG_INITIAL
, dtl_max_txg
- TXG_INITIAL
);
6901 if (newvd
->vdev_isspare
) {
6902 spa_spare_activate(newvd
);
6903 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
6906 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
6907 newvdpath
= spa_strdup(newvd
->vdev_path
);
6908 newvd_isspare
= newvd
->vdev_isspare
;
6911 * Mark newvd's DTL dirty in this txg.
6913 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
6916 * Schedule the resilver or rebuild to restart in the future. We do
6917 * this to ensure that dmu_sync-ed blocks have been stitched into the
6918 * respective datasets.
6921 newvd
->vdev_rebuild_txg
= txg
;
6925 newvd
->vdev_resilver_txg
= txg
;
6927 if (dsl_scan_resilvering(spa_get_dsl(spa
)) &&
6928 spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
)) {
6929 vdev_defer_resilver(newvd
);
6931 dsl_scan_restart_resilver(spa
->spa_dsl_pool
,
6936 if (spa
->spa_bootfs
)
6937 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
6939 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
6944 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
6946 spa_history_log_internal(spa
, "vdev attach", NULL
,
6947 "%s vdev=%s %s vdev=%s",
6948 replacing
&& newvd_isspare
? "spare in" :
6949 replacing
? "replace" : "attach", newvdpath
,
6950 replacing
? "for" : "to", oldvdpath
);
6952 spa_strfree(oldvdpath
);
6953 spa_strfree(newvdpath
);
6959 * Detach a device from a mirror or replacing vdev.
6961 * If 'replace_done' is specified, only detach if the parent
6962 * is a replacing vdev.
6965 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
6969 vdev_t
*rvd __maybe_unused
= spa
->spa_root_vdev
;
6970 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
6971 boolean_t unspare
= B_FALSE
;
6972 uint64_t unspare_guid
= 0;
6975 ASSERT(spa_writeable(spa
));
6977 txg
= spa_vdev_detach_enter(spa
, guid
);
6979 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
6982 * Besides being called directly from the userland through the
6983 * ioctl interface, spa_vdev_detach() can be potentially called
6984 * at the end of spa_vdev_resilver_done().
6986 * In the regular case, when we have a checkpoint this shouldn't
6987 * happen as we never empty the DTLs of a vdev during the scrub
6988 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6989 * should never get here when we have a checkpoint.
6991 * That said, even in a case when we checkpoint the pool exactly
6992 * as spa_vdev_resilver_done() calls this function everything
6993 * should be fine as the resilver will return right away.
6995 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
6996 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
6997 error
= (spa_has_checkpoint(spa
)) ?
6998 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
6999 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7003 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
7005 if (!vd
->vdev_ops
->vdev_op_leaf
)
7006 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7008 pvd
= vd
->vdev_parent
;
7011 * If the parent/child relationship is not as expected, don't do it.
7012 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
7013 * vdev that's replacing B with C. The user's intent in replacing
7014 * is to go from M(A,B) to M(A,C). If the user decides to cancel
7015 * the replace by detaching C, the expected behavior is to end up
7016 * M(A,B). But suppose that right after deciding to detach C,
7017 * the replacement of B completes. We would have M(A,C), and then
7018 * ask to detach C, which would leave us with just A -- not what
7019 * the user wanted. To prevent this, we make sure that the
7020 * parent/child relationship hasn't changed -- in this example,
7021 * that C's parent is still the replacing vdev R.
7023 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
7024 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
7027 * Only 'replacing' or 'spare' vdevs can be replaced.
7029 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
7030 pvd
->vdev_ops
!= &vdev_spare_ops
)
7031 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7033 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
7034 spa_version(spa
) >= SPA_VERSION_SPARES
);
7037 * Only mirror, replacing, and spare vdevs support detach.
7039 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
7040 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
7041 pvd
->vdev_ops
!= &vdev_spare_ops
)
7042 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7045 * If this device has the only valid copy of some data,
7046 * we cannot safely detach it.
7048 if (vdev_dtl_required(vd
))
7049 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
7051 ASSERT(pvd
->vdev_children
>= 2);
7054 * If we are detaching the second disk from a replacing vdev, then
7055 * check to see if we changed the original vdev's path to have "/old"
7056 * at the end in spa_vdev_attach(). If so, undo that change now.
7058 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
7059 vd
->vdev_path
!= NULL
) {
7060 size_t len
= strlen(vd
->vdev_path
);
7062 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
7063 cvd
= pvd
->vdev_child
[c
];
7065 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
7068 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
7069 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
7070 spa_strfree(cvd
->vdev_path
);
7071 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
7078 * If we are detaching the original disk from a normal spare, then it
7079 * implies that the spare should become a real disk, and be removed
7080 * from the active spare list for the pool. dRAID spares on the
7081 * other hand are coupled to the pool and thus should never be removed
7082 * from the spares list.
7084 if (pvd
->vdev_ops
== &vdev_spare_ops
&& vd
->vdev_id
== 0) {
7085 vdev_t
*last_cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7087 if (last_cvd
->vdev_isspare
&&
7088 last_cvd
->vdev_ops
!= &vdev_draid_spare_ops
) {
7094 * Erase the disk labels so the disk can be used for other things.
7095 * This must be done after all other error cases are handled,
7096 * but before we disembowel vd (so we can still do I/O to it).
7097 * But if we can't do it, don't treat the error as fatal --
7098 * it may be that the unwritability of the disk is the reason
7099 * it's being detached!
7101 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
7104 * Remove vd from its parent and compact the parent's children.
7106 vdev_remove_child(pvd
, vd
);
7107 vdev_compact_children(pvd
);
7110 * Remember one of the remaining children so we can get tvd below.
7112 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7115 * If we need to remove the remaining child from the list of hot spares,
7116 * do it now, marking the vdev as no longer a spare in the process.
7117 * We must do this before vdev_remove_parent(), because that can
7118 * change the GUID if it creates a new toplevel GUID. For a similar
7119 * reason, we must remove the spare now, in the same txg as the detach;
7120 * otherwise someone could attach a new sibling, change the GUID, and
7121 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
7124 ASSERT(cvd
->vdev_isspare
);
7125 spa_spare_remove(cvd
);
7126 unspare_guid
= cvd
->vdev_guid
;
7127 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
7128 cvd
->vdev_unspare
= B_TRUE
;
7132 * If the parent mirror/replacing vdev only has one child,
7133 * the parent is no longer needed. Remove it from the tree.
7135 if (pvd
->vdev_children
== 1) {
7136 if (pvd
->vdev_ops
== &vdev_spare_ops
)
7137 cvd
->vdev_unspare
= B_FALSE
;
7138 vdev_remove_parent(cvd
);
7142 * We don't set tvd until now because the parent we just removed
7143 * may have been the previous top-level vdev.
7145 tvd
= cvd
->vdev_top
;
7146 ASSERT(tvd
->vdev_parent
== rvd
);
7149 * Reevaluate the parent vdev state.
7151 vdev_propagate_state(cvd
);
7154 * If the 'autoexpand' property is set on the pool then automatically
7155 * try to expand the size of the pool. For example if the device we
7156 * just detached was smaller than the others, it may be possible to
7157 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
7158 * first so that we can obtain the updated sizes of the leaf vdevs.
7160 if (spa
->spa_autoexpand
) {
7162 vdev_expand(tvd
, txg
);
7165 vdev_config_dirty(tvd
);
7168 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
7169 * vd->vdev_detached is set and free vd's DTL object in syncing context.
7170 * But first make sure we're not on any *other* txg's DTL list, to
7171 * prevent vd from being accessed after it's freed.
7173 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
7174 for (int t
= 0; t
< TXG_SIZE
; t
++)
7175 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
7176 vd
->vdev_detached
= B_TRUE
;
7177 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
7179 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
7180 spa_notify_waiters(spa
);
7182 /* hang on to the spa before we release the lock */
7183 spa_open_ref(spa
, FTAG
);
7185 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
7187 spa_history_log_internal(spa
, "detach", NULL
,
7189 spa_strfree(vdpath
);
7192 * If this was the removal of the original device in a hot spare vdev,
7193 * then we want to go through and remove the device from the hot spare
7194 * list of every other pool.
7197 spa_t
*altspa
= NULL
;
7199 mutex_enter(&spa_namespace_lock
);
7200 while ((altspa
= spa_next(altspa
)) != NULL
) {
7201 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
7205 spa_open_ref(altspa
, FTAG
);
7206 mutex_exit(&spa_namespace_lock
);
7207 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
7208 mutex_enter(&spa_namespace_lock
);
7209 spa_close(altspa
, FTAG
);
7211 mutex_exit(&spa_namespace_lock
);
7213 /* search the rest of the vdevs for spares to remove */
7214 spa_vdev_resilver_done(spa
);
7217 /* all done with the spa; OK to release */
7218 mutex_enter(&spa_namespace_lock
);
7219 spa_close(spa
, FTAG
);
7220 mutex_exit(&spa_namespace_lock
);
7226 spa_vdev_initialize_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7229 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7231 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7233 /* Look up vdev and ensure it's a leaf. */
7234 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7235 if (vd
== NULL
|| vd
->vdev_detached
) {
7236 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7237 return (SET_ERROR(ENODEV
));
7238 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7239 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7240 return (SET_ERROR(EINVAL
));
7241 } else if (!vdev_writeable(vd
)) {
7242 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7243 return (SET_ERROR(EROFS
));
7245 mutex_enter(&vd
->vdev_initialize_lock
);
7246 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7249 * When we activate an initialize action we check to see
7250 * if the vdev_initialize_thread is NULL. We do this instead
7251 * of using the vdev_initialize_state since there might be
7252 * a previous initialization process which has completed but
7253 * the thread is not exited.
7255 if (cmd_type
== POOL_INITIALIZE_START
&&
7256 (vd
->vdev_initialize_thread
!= NULL
||
7257 vd
->vdev_top
->vdev_removing
)) {
7258 mutex_exit(&vd
->vdev_initialize_lock
);
7259 return (SET_ERROR(EBUSY
));
7260 } else if (cmd_type
== POOL_INITIALIZE_CANCEL
&&
7261 (vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
&&
7262 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_SUSPENDED
)) {
7263 mutex_exit(&vd
->vdev_initialize_lock
);
7264 return (SET_ERROR(ESRCH
));
7265 } else if (cmd_type
== POOL_INITIALIZE_SUSPEND
&&
7266 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
) {
7267 mutex_exit(&vd
->vdev_initialize_lock
);
7268 return (SET_ERROR(ESRCH
));
7272 case POOL_INITIALIZE_START
:
7273 vdev_initialize(vd
);
7275 case POOL_INITIALIZE_CANCEL
:
7276 vdev_initialize_stop(vd
, VDEV_INITIALIZE_CANCELED
, vd_list
);
7278 case POOL_INITIALIZE_SUSPEND
:
7279 vdev_initialize_stop(vd
, VDEV_INITIALIZE_SUSPENDED
, vd_list
);
7282 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7284 mutex_exit(&vd
->vdev_initialize_lock
);
7290 spa_vdev_initialize(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
,
7291 nvlist_t
*vdev_errlist
)
7293 int total_errors
= 0;
7296 list_create(&vd_list
, sizeof (vdev_t
),
7297 offsetof(vdev_t
, vdev_initialize_node
));
7300 * We hold the namespace lock through the whole function
7301 * to prevent any changes to the pool while we're starting or
7302 * stopping initialization. The config and state locks are held so that
7303 * we can properly assess the vdev state before we commit to
7304 * the initializing operation.
7306 mutex_enter(&spa_namespace_lock
);
7308 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7309 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7310 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7312 int error
= spa_vdev_initialize_impl(spa
, vdev_guid
, cmd_type
,
7315 char guid_as_str
[MAXNAMELEN
];
7317 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7318 "%llu", (unsigned long long)vdev_guid
);
7319 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7324 /* Wait for all initialize threads to stop. */
7325 vdev_initialize_stop_wait(spa
, &vd_list
);
7327 /* Sync out the initializing state */
7328 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7329 mutex_exit(&spa_namespace_lock
);
7331 list_destroy(&vd_list
);
7333 return (total_errors
);
7337 spa_vdev_trim_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7338 uint64_t rate
, boolean_t partial
, boolean_t secure
, list_t
*vd_list
)
7340 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7342 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7344 /* Look up vdev and ensure it's a leaf. */
7345 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7346 if (vd
== NULL
|| vd
->vdev_detached
) {
7347 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7348 return (SET_ERROR(ENODEV
));
7349 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7350 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7351 return (SET_ERROR(EINVAL
));
7352 } else if (!vdev_writeable(vd
)) {
7353 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7354 return (SET_ERROR(EROFS
));
7355 } else if (!vd
->vdev_has_trim
) {
7356 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7357 return (SET_ERROR(EOPNOTSUPP
));
7358 } else if (secure
&& !vd
->vdev_has_securetrim
) {
7359 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7360 return (SET_ERROR(EOPNOTSUPP
));
7362 mutex_enter(&vd
->vdev_trim_lock
);
7363 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7366 * When we activate a TRIM action we check to see if the
7367 * vdev_trim_thread is NULL. We do this instead of using the
7368 * vdev_trim_state since there might be a previous TRIM process
7369 * which has completed but the thread is not exited.
7371 if (cmd_type
== POOL_TRIM_START
&&
7372 (vd
->vdev_trim_thread
!= NULL
|| vd
->vdev_top
->vdev_removing
)) {
7373 mutex_exit(&vd
->vdev_trim_lock
);
7374 return (SET_ERROR(EBUSY
));
7375 } else if (cmd_type
== POOL_TRIM_CANCEL
&&
7376 (vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
&&
7377 vd
->vdev_trim_state
!= VDEV_TRIM_SUSPENDED
)) {
7378 mutex_exit(&vd
->vdev_trim_lock
);
7379 return (SET_ERROR(ESRCH
));
7380 } else if (cmd_type
== POOL_TRIM_SUSPEND
&&
7381 vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
) {
7382 mutex_exit(&vd
->vdev_trim_lock
);
7383 return (SET_ERROR(ESRCH
));
7387 case POOL_TRIM_START
:
7388 vdev_trim(vd
, rate
, partial
, secure
);
7390 case POOL_TRIM_CANCEL
:
7391 vdev_trim_stop(vd
, VDEV_TRIM_CANCELED
, vd_list
);
7393 case POOL_TRIM_SUSPEND
:
7394 vdev_trim_stop(vd
, VDEV_TRIM_SUSPENDED
, vd_list
);
7397 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7399 mutex_exit(&vd
->vdev_trim_lock
);
7405 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
7406 * TRIM threads for each child vdev. These threads pass over all of the free
7407 * space in the vdev's metaslabs and issues TRIM commands for that space.
7410 spa_vdev_trim(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
, uint64_t rate
,
7411 boolean_t partial
, boolean_t secure
, nvlist_t
*vdev_errlist
)
7413 int total_errors
= 0;
7416 list_create(&vd_list
, sizeof (vdev_t
),
7417 offsetof(vdev_t
, vdev_trim_node
));
7420 * We hold the namespace lock through the whole function
7421 * to prevent any changes to the pool while we're starting or
7422 * stopping TRIM. The config and state locks are held so that
7423 * we can properly assess the vdev state before we commit to
7424 * the TRIM operation.
7426 mutex_enter(&spa_namespace_lock
);
7428 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7429 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7430 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7432 int error
= spa_vdev_trim_impl(spa
, vdev_guid
, cmd_type
,
7433 rate
, partial
, secure
, &vd_list
);
7435 char guid_as_str
[MAXNAMELEN
];
7437 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7438 "%llu", (unsigned long long)vdev_guid
);
7439 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7444 /* Wait for all TRIM threads to stop. */
7445 vdev_trim_stop_wait(spa
, &vd_list
);
7447 /* Sync out the TRIM state */
7448 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7449 mutex_exit(&spa_namespace_lock
);
7451 list_destroy(&vd_list
);
7453 return (total_errors
);
7457 * Split a set of devices from their mirrors, and create a new pool from them.
7460 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
7461 nvlist_t
*props
, boolean_t exp
)
7464 uint64_t txg
, *glist
;
7466 uint_t c
, children
, lastlog
;
7467 nvlist_t
**child
, *nvl
, *tmp
;
7469 char *altroot
= NULL
;
7470 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
7471 boolean_t activate_slog
;
7473 ASSERT(spa_writeable(spa
));
7475 txg
= spa_vdev_enter(spa
);
7477 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7478 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7479 error
= (spa_has_checkpoint(spa
)) ?
7480 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7481 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7484 /* clear the log and flush everything up to now */
7485 activate_slog
= spa_passivate_log(spa
);
7486 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
7487 error
= spa_reset_logs(spa
);
7488 txg
= spa_vdev_config_enter(spa
);
7491 spa_activate_log(spa
);
7494 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7496 /* check new spa name before going any further */
7497 if (spa_lookup(newname
) != NULL
)
7498 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
7501 * scan through all the children to ensure they're all mirrors
7503 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
7504 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
7506 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7508 /* first, check to ensure we've got the right child count */
7509 rvd
= spa
->spa_root_vdev
;
7511 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
7512 vdev_t
*vd
= rvd
->vdev_child
[c
];
7514 /* don't count the holes & logs as children */
7515 if (vd
->vdev_islog
|| (vd
->vdev_ops
!= &vdev_indirect_ops
&&
7516 !vdev_is_concrete(vd
))) {
7524 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
7525 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7527 /* next, ensure no spare or cache devices are part of the split */
7528 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
7529 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
7530 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7532 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
7533 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
7535 /* then, loop over each vdev and validate it */
7536 for (c
= 0; c
< children
; c
++) {
7537 uint64_t is_hole
= 0;
7539 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
7543 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
7544 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
7547 error
= SET_ERROR(EINVAL
);
7552 /* deal with indirect vdevs */
7553 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ops
==
7557 /* which disk is going to be split? */
7558 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
7560 error
= SET_ERROR(EINVAL
);
7564 /* look it up in the spa */
7565 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
7566 if (vml
[c
] == NULL
) {
7567 error
= SET_ERROR(ENODEV
);
7571 /* make sure there's nothing stopping the split */
7572 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
7573 vml
[c
]->vdev_islog
||
7574 !vdev_is_concrete(vml
[c
]) ||
7575 vml
[c
]->vdev_isspare
||
7576 vml
[c
]->vdev_isl2cache
||
7577 !vdev_writeable(vml
[c
]) ||
7578 vml
[c
]->vdev_children
!= 0 ||
7579 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
7580 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
7581 error
= SET_ERROR(EINVAL
);
7585 if (vdev_dtl_required(vml
[c
]) ||
7586 vdev_resilver_needed(vml
[c
], NULL
, NULL
)) {
7587 error
= SET_ERROR(EBUSY
);
7591 /* we need certain info from the top level */
7592 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
7593 vml
[c
]->vdev_top
->vdev_ms_array
);
7594 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
7595 vml
[c
]->vdev_top
->vdev_ms_shift
);
7596 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
7597 vml
[c
]->vdev_top
->vdev_asize
);
7598 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
7599 vml
[c
]->vdev_top
->vdev_ashift
);
7601 /* transfer per-vdev ZAPs */
7602 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
7603 VERIFY0(nvlist_add_uint64(child
[c
],
7604 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
7606 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
7607 VERIFY0(nvlist_add_uint64(child
[c
],
7608 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
7609 vml
[c
]->vdev_parent
->vdev_top_zap
));
7613 kmem_free(vml
, children
* sizeof (vdev_t
*));
7614 kmem_free(glist
, children
* sizeof (uint64_t));
7615 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7618 /* stop writers from using the disks */
7619 for (c
= 0; c
< children
; c
++) {
7621 vml
[c
]->vdev_offline
= B_TRUE
;
7623 vdev_reopen(spa
->spa_root_vdev
);
7626 * Temporarily record the splitting vdevs in the spa config. This
7627 * will disappear once the config is regenerated.
7629 nvl
= fnvlist_alloc();
7630 fnvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
, glist
, children
);
7631 kmem_free(glist
, children
* sizeof (uint64_t));
7633 mutex_enter(&spa
->spa_props_lock
);
7634 fnvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
, nvl
);
7635 mutex_exit(&spa
->spa_props_lock
);
7636 spa
->spa_config_splitting
= nvl
;
7637 vdev_config_dirty(spa
->spa_root_vdev
);
7639 /* configure and create the new pool */
7640 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
);
7641 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
7642 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
);
7643 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
, spa_version(spa
));
7644 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, spa
->spa_config_txg
);
7645 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
7646 spa_generate_guid(NULL
));
7647 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
7648 (void) nvlist_lookup_string(props
,
7649 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
7651 /* add the new pool to the namespace */
7652 newspa
= spa_add(newname
, config
, altroot
);
7653 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
7654 newspa
->spa_config_txg
= spa
->spa_config_txg
;
7655 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
7657 /* release the spa config lock, retaining the namespace lock */
7658 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
7660 if (zio_injection_enabled
)
7661 zio_handle_panic_injection(spa
, FTAG
, 1);
7663 spa_activate(newspa
, spa_mode_global
);
7664 spa_async_suspend(newspa
);
7667 * Temporarily stop the initializing and TRIM activity. We set the
7668 * state to ACTIVE so that we know to resume initializing or TRIM
7669 * once the split has completed.
7671 list_t vd_initialize_list
;
7672 list_create(&vd_initialize_list
, sizeof (vdev_t
),
7673 offsetof(vdev_t
, vdev_initialize_node
));
7675 list_t vd_trim_list
;
7676 list_create(&vd_trim_list
, sizeof (vdev_t
),
7677 offsetof(vdev_t
, vdev_trim_node
));
7679 for (c
= 0; c
< children
; c
++) {
7680 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
7681 mutex_enter(&vml
[c
]->vdev_initialize_lock
);
7682 vdev_initialize_stop(vml
[c
],
7683 VDEV_INITIALIZE_ACTIVE
, &vd_initialize_list
);
7684 mutex_exit(&vml
[c
]->vdev_initialize_lock
);
7686 mutex_enter(&vml
[c
]->vdev_trim_lock
);
7687 vdev_trim_stop(vml
[c
], VDEV_TRIM_ACTIVE
, &vd_trim_list
);
7688 mutex_exit(&vml
[c
]->vdev_trim_lock
);
7692 vdev_initialize_stop_wait(spa
, &vd_initialize_list
);
7693 vdev_trim_stop_wait(spa
, &vd_trim_list
);
7695 list_destroy(&vd_initialize_list
);
7696 list_destroy(&vd_trim_list
);
7698 newspa
->spa_config_source
= SPA_CONFIG_SRC_SPLIT
;
7699 newspa
->spa_is_splitting
= B_TRUE
;
7701 /* create the new pool from the disks of the original pool */
7702 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
);
7706 /* if that worked, generate a real config for the new pool */
7707 if (newspa
->spa_root_vdev
!= NULL
) {
7708 newspa
->spa_config_splitting
= fnvlist_alloc();
7709 fnvlist_add_uint64(newspa
->spa_config_splitting
,
7710 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
));
7711 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
7716 if (props
!= NULL
) {
7717 spa_configfile_set(newspa
, props
, B_FALSE
);
7718 error
= spa_prop_set(newspa
, props
);
7723 /* flush everything */
7724 txg
= spa_vdev_config_enter(newspa
);
7725 vdev_config_dirty(newspa
->spa_root_vdev
);
7726 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
7728 if (zio_injection_enabled
)
7729 zio_handle_panic_injection(spa
, FTAG
, 2);
7731 spa_async_resume(newspa
);
7733 /* finally, update the original pool's config */
7734 txg
= spa_vdev_config_enter(spa
);
7735 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
7736 error
= dmu_tx_assign(tx
, TXG_WAIT
);
7739 for (c
= 0; c
< children
; c
++) {
7740 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
7741 vdev_t
*tvd
= vml
[c
]->vdev_top
;
7744 * Need to be sure the detachable VDEV is not
7745 * on any *other* txg's DTL list to prevent it
7746 * from being accessed after it's freed.
7748 for (int t
= 0; t
< TXG_SIZE
; t
++) {
7749 (void) txg_list_remove_this(
7750 &tvd
->vdev_dtl_list
, vml
[c
], t
);
7755 spa_history_log_internal(spa
, "detach", tx
,
7756 "vdev=%s", vml
[c
]->vdev_path
);
7761 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
7762 vdev_config_dirty(spa
->spa_root_vdev
);
7763 spa
->spa_config_splitting
= NULL
;
7767 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
7769 if (zio_injection_enabled
)
7770 zio_handle_panic_injection(spa
, FTAG
, 3);
7772 /* split is complete; log a history record */
7773 spa_history_log_internal(newspa
, "split", NULL
,
7774 "from pool %s", spa_name(spa
));
7776 newspa
->spa_is_splitting
= B_FALSE
;
7777 kmem_free(vml
, children
* sizeof (vdev_t
*));
7779 /* if we're not going to mount the filesystems in userland, export */
7781 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
7788 spa_deactivate(newspa
);
7791 txg
= spa_vdev_config_enter(spa
);
7793 /* re-online all offlined disks */
7794 for (c
= 0; c
< children
; c
++) {
7796 vml
[c
]->vdev_offline
= B_FALSE
;
7799 /* restart initializing or trimming disks as necessary */
7800 spa_async_request(spa
, SPA_ASYNC_INITIALIZE_RESTART
);
7801 spa_async_request(spa
, SPA_ASYNC_TRIM_RESTART
);
7802 spa_async_request(spa
, SPA_ASYNC_AUTOTRIM_RESTART
);
7804 vdev_reopen(spa
->spa_root_vdev
);
7806 nvlist_free(spa
->spa_config_splitting
);
7807 spa
->spa_config_splitting
= NULL
;
7808 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
7810 kmem_free(vml
, children
* sizeof (vdev_t
*));
7815 * Find any device that's done replacing, or a vdev marked 'unspare' that's
7816 * currently spared, so we can detach it.
7819 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
7821 vdev_t
*newvd
, *oldvd
;
7823 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
7824 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
7830 * Check for a completed replacement. We always consider the first
7831 * vdev in the list to be the oldest vdev, and the last one to be
7832 * the newest (see spa_vdev_attach() for how that works). In
7833 * the case where the newest vdev is faulted, we will not automatically
7834 * remove it after a resilver completes. This is OK as it will require
7835 * user intervention to determine which disk the admin wishes to keep.
7837 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
7838 ASSERT(vd
->vdev_children
> 1);
7840 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
7841 oldvd
= vd
->vdev_child
[0];
7843 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
7844 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
7845 !vdev_dtl_required(oldvd
))
7850 * Check for a completed resilver with the 'unspare' flag set.
7851 * Also potentially update faulted state.
7853 if (vd
->vdev_ops
== &vdev_spare_ops
) {
7854 vdev_t
*first
= vd
->vdev_child
[0];
7855 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
7857 if (last
->vdev_unspare
) {
7860 } else if (first
->vdev_unspare
) {
7867 if (oldvd
!= NULL
&&
7868 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
7869 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
7870 !vdev_dtl_required(oldvd
))
7873 vdev_propagate_state(vd
);
7876 * If there are more than two spares attached to a disk,
7877 * and those spares are not required, then we want to
7878 * attempt to free them up now so that they can be used
7879 * by other pools. Once we're back down to a single
7880 * disk+spare, we stop removing them.
7882 if (vd
->vdev_children
> 2) {
7883 newvd
= vd
->vdev_child
[1];
7885 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
7886 vdev_dtl_empty(last
, DTL_MISSING
) &&
7887 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
7888 !vdev_dtl_required(newvd
))
7897 spa_vdev_resilver_done(spa_t
*spa
)
7899 vdev_t
*vd
, *pvd
, *ppvd
;
7900 uint64_t guid
, sguid
, pguid
, ppguid
;
7902 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7904 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
7905 pvd
= vd
->vdev_parent
;
7906 ppvd
= pvd
->vdev_parent
;
7907 guid
= vd
->vdev_guid
;
7908 pguid
= pvd
->vdev_guid
;
7909 ppguid
= ppvd
->vdev_guid
;
7912 * If we have just finished replacing a hot spared device, then
7913 * we need to detach the parent's first child (the original hot
7916 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
7917 ppvd
->vdev_children
== 2) {
7918 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
7919 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
7921 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
7923 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7924 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
7926 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
7928 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7931 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7934 * If a detach was not performed above replace waiters will not have
7935 * been notified. In which case we must do so now.
7937 spa_notify_waiters(spa
);
7941 * Update the stored path or FRU for this vdev.
7944 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
7948 boolean_t sync
= B_FALSE
;
7950 ASSERT(spa_writeable(spa
));
7952 spa_vdev_state_enter(spa
, SCL_ALL
);
7954 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
7955 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
7957 if (!vd
->vdev_ops
->vdev_op_leaf
)
7958 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
7961 if (strcmp(value
, vd
->vdev_path
) != 0) {
7962 spa_strfree(vd
->vdev_path
);
7963 vd
->vdev_path
= spa_strdup(value
);
7967 if (vd
->vdev_fru
== NULL
) {
7968 vd
->vdev_fru
= spa_strdup(value
);
7970 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
7971 spa_strfree(vd
->vdev_fru
);
7972 vd
->vdev_fru
= spa_strdup(value
);
7977 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
7981 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
7983 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
7987 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
7989 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
7993 * ==========================================================================
7995 * ==========================================================================
7998 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
8000 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8002 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
8003 return (SET_ERROR(EBUSY
));
8005 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
8009 spa_scan_stop(spa_t
*spa
)
8011 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8012 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
8013 return (SET_ERROR(EBUSY
));
8014 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
8018 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
8020 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8022 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
8023 return (SET_ERROR(ENOTSUP
));
8025 if (func
== POOL_SCAN_RESILVER
&&
8026 !spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
))
8027 return (SET_ERROR(ENOTSUP
));
8030 * If a resilver was requested, but there is no DTL on a
8031 * writeable leaf device, we have nothing to do.
8033 if (func
== POOL_SCAN_RESILVER
&&
8034 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
8035 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
8039 return (dsl_scan(spa
->spa_dsl_pool
, func
));
8043 * ==========================================================================
8044 * SPA async task processing
8045 * ==========================================================================
8049 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
8051 if (vd
->vdev_remove_wanted
) {
8052 vd
->vdev_remove_wanted
= B_FALSE
;
8053 vd
->vdev_delayed_close
= B_FALSE
;
8054 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
8057 * We want to clear the stats, but we don't want to do a full
8058 * vdev_clear() as that will cause us to throw away
8059 * degraded/faulted state as well as attempt to reopen the
8060 * device, all of which is a waste.
8062 vd
->vdev_stat
.vs_read_errors
= 0;
8063 vd
->vdev_stat
.vs_write_errors
= 0;
8064 vd
->vdev_stat
.vs_checksum_errors
= 0;
8066 vdev_state_dirty(vd
->vdev_top
);
8068 /* Tell userspace that the vdev is gone. */
8069 zfs_post_remove(spa
, vd
);
8072 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8073 spa_async_remove(spa
, vd
->vdev_child
[c
]);
8077 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
8079 if (vd
->vdev_probe_wanted
) {
8080 vd
->vdev_probe_wanted
= B_FALSE
;
8081 vdev_reopen(vd
); /* vdev_open() does the actual probe */
8084 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8085 spa_async_probe(spa
, vd
->vdev_child
[c
]);
8089 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
8091 if (!spa
->spa_autoexpand
)
8094 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
8095 vdev_t
*cvd
= vd
->vdev_child
[c
];
8096 spa_async_autoexpand(spa
, cvd
);
8099 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
8102 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
8106 spa_async_thread(void *arg
)
8108 spa_t
*spa
= (spa_t
*)arg
;
8109 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8112 ASSERT(spa
->spa_sync_on
);
8114 mutex_enter(&spa
->spa_async_lock
);
8115 tasks
= spa
->spa_async_tasks
;
8116 spa
->spa_async_tasks
= 0;
8117 mutex_exit(&spa
->spa_async_lock
);
8120 * See if the config needs to be updated.
8122 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
8123 uint64_t old_space
, new_space
;
8125 mutex_enter(&spa_namespace_lock
);
8126 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
8127 old_space
+= metaslab_class_get_space(spa_special_class(spa
));
8128 old_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8129 old_space
+= metaslab_class_get_space(
8130 spa_embedded_log_class(spa
));
8132 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
8134 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
8135 new_space
+= metaslab_class_get_space(spa_special_class(spa
));
8136 new_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8137 new_space
+= metaslab_class_get_space(
8138 spa_embedded_log_class(spa
));
8139 mutex_exit(&spa_namespace_lock
);
8142 * If the pool grew as a result of the config update,
8143 * then log an internal history event.
8145 if (new_space
!= old_space
) {
8146 spa_history_log_internal(spa
, "vdev online", NULL
,
8147 "pool '%s' size: %llu(+%llu)",
8148 spa_name(spa
), (u_longlong_t
)new_space
,
8149 (u_longlong_t
)(new_space
- old_space
));
8154 * See if any devices need to be marked REMOVED.
8156 if (tasks
& SPA_ASYNC_REMOVE
) {
8157 spa_vdev_state_enter(spa
, SCL_NONE
);
8158 spa_async_remove(spa
, spa
->spa_root_vdev
);
8159 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
8160 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
8161 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
8162 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
8163 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8166 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
8167 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8168 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
8169 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8173 * See if any devices need to be probed.
8175 if (tasks
& SPA_ASYNC_PROBE
) {
8176 spa_vdev_state_enter(spa
, SCL_NONE
);
8177 spa_async_probe(spa
, spa
->spa_root_vdev
);
8178 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8182 * If any devices are done replacing, detach them.
8184 if (tasks
& SPA_ASYNC_RESILVER_DONE
||
8185 tasks
& SPA_ASYNC_REBUILD_DONE
) {
8186 spa_vdev_resilver_done(spa
);
8190 * Kick off a resilver.
8192 if (tasks
& SPA_ASYNC_RESILVER
&&
8193 !vdev_rebuild_active(spa
->spa_root_vdev
) &&
8194 (!dsl_scan_resilvering(dp
) ||
8195 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_RESILVER_DEFER
)))
8196 dsl_scan_restart_resilver(dp
, 0);
8198 if (tasks
& SPA_ASYNC_INITIALIZE_RESTART
) {
8199 mutex_enter(&spa_namespace_lock
);
8200 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8201 vdev_initialize_restart(spa
->spa_root_vdev
);
8202 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8203 mutex_exit(&spa_namespace_lock
);
8206 if (tasks
& SPA_ASYNC_TRIM_RESTART
) {
8207 mutex_enter(&spa_namespace_lock
);
8208 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8209 vdev_trim_restart(spa
->spa_root_vdev
);
8210 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8211 mutex_exit(&spa_namespace_lock
);
8214 if (tasks
& SPA_ASYNC_AUTOTRIM_RESTART
) {
8215 mutex_enter(&spa_namespace_lock
);
8216 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8217 vdev_autotrim_restart(spa
);
8218 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8219 mutex_exit(&spa_namespace_lock
);
8223 * Kick off L2 cache whole device TRIM.
8225 if (tasks
& SPA_ASYNC_L2CACHE_TRIM
) {
8226 mutex_enter(&spa_namespace_lock
);
8227 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8228 vdev_trim_l2arc(spa
);
8229 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8230 mutex_exit(&spa_namespace_lock
);
8234 * Kick off L2 cache rebuilding.
8236 if (tasks
& SPA_ASYNC_L2CACHE_REBUILD
) {
8237 mutex_enter(&spa_namespace_lock
);
8238 spa_config_enter(spa
, SCL_L2ARC
, FTAG
, RW_READER
);
8239 l2arc_spa_rebuild_start(spa
);
8240 spa_config_exit(spa
, SCL_L2ARC
, FTAG
);
8241 mutex_exit(&spa_namespace_lock
);
8245 * Let the world know that we're done.
8247 mutex_enter(&spa
->spa_async_lock
);
8248 spa
->spa_async_thread
= NULL
;
8249 cv_broadcast(&spa
->spa_async_cv
);
8250 mutex_exit(&spa
->spa_async_lock
);
8255 spa_async_suspend(spa_t
*spa
)
8257 mutex_enter(&spa
->spa_async_lock
);
8258 spa
->spa_async_suspended
++;
8259 while (spa
->spa_async_thread
!= NULL
)
8260 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
8261 mutex_exit(&spa
->spa_async_lock
);
8263 spa_vdev_remove_suspend(spa
);
8265 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8266 if (condense_thread
!= NULL
)
8267 zthr_cancel(condense_thread
);
8269 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8270 if (discard_thread
!= NULL
)
8271 zthr_cancel(discard_thread
);
8273 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8274 if (ll_delete_thread
!= NULL
)
8275 zthr_cancel(ll_delete_thread
);
8277 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8278 if (ll_condense_thread
!= NULL
)
8279 zthr_cancel(ll_condense_thread
);
8283 spa_async_resume(spa_t
*spa
)
8285 mutex_enter(&spa
->spa_async_lock
);
8286 ASSERT(spa
->spa_async_suspended
!= 0);
8287 spa
->spa_async_suspended
--;
8288 mutex_exit(&spa
->spa_async_lock
);
8289 spa_restart_removal(spa
);
8291 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8292 if (condense_thread
!= NULL
)
8293 zthr_resume(condense_thread
);
8295 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8296 if (discard_thread
!= NULL
)
8297 zthr_resume(discard_thread
);
8299 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8300 if (ll_delete_thread
!= NULL
)
8301 zthr_resume(ll_delete_thread
);
8303 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8304 if (ll_condense_thread
!= NULL
)
8305 zthr_resume(ll_condense_thread
);
8309 spa_async_tasks_pending(spa_t
*spa
)
8311 uint_t non_config_tasks
;
8313 boolean_t config_task_suspended
;
8315 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
8316 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
8317 if (spa
->spa_ccw_fail_time
== 0) {
8318 config_task_suspended
= B_FALSE
;
8320 config_task_suspended
=
8321 (gethrtime() - spa
->spa_ccw_fail_time
) <
8322 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
8325 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
8329 spa_async_dispatch(spa_t
*spa
)
8331 mutex_enter(&spa
->spa_async_lock
);
8332 if (spa_async_tasks_pending(spa
) &&
8333 !spa
->spa_async_suspended
&&
8334 spa
->spa_async_thread
== NULL
)
8335 spa
->spa_async_thread
= thread_create(NULL
, 0,
8336 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
8337 mutex_exit(&spa
->spa_async_lock
);
8341 spa_async_request(spa_t
*spa
, int task
)
8343 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
8344 mutex_enter(&spa
->spa_async_lock
);
8345 spa
->spa_async_tasks
|= task
;
8346 mutex_exit(&spa
->spa_async_lock
);
8350 spa_async_tasks(spa_t
*spa
)
8352 return (spa
->spa_async_tasks
);
8356 * ==========================================================================
8357 * SPA syncing routines
8358 * ==========================================================================
8363 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8367 bpobj_enqueue(bpo
, bp
, bp_freed
, tx
);
8372 bpobj_enqueue_alloc_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8374 return (bpobj_enqueue_cb(arg
, bp
, B_FALSE
, tx
));
8378 bpobj_enqueue_free_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8380 return (bpobj_enqueue_cb(arg
, bp
, B_TRUE
, tx
));
8384 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8388 zio_nowait(zio_free_sync(pio
, pio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
8394 bpobj_spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8398 return (spa_free_sync_cb(arg
, bp
, tx
));
8402 * Note: this simple function is not inlined to make it easier to dtrace the
8403 * amount of time spent syncing frees.
8406 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
8408 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8409 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
8410 VERIFY(zio_wait(zio
) == 0);
8414 * Note: this simple function is not inlined to make it easier to dtrace the
8415 * amount of time spent syncing deferred frees.
8418 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
8420 if (spa_sync_pass(spa
) != 1)
8425 * If the log space map feature is active, we stop deferring
8426 * frees to the next TXG and therefore running this function
8427 * would be considered a no-op as spa_deferred_bpobj should
8428 * not have any entries.
8430 * That said we run this function anyway (instead of returning
8431 * immediately) for the edge-case scenario where we just
8432 * activated the log space map feature in this TXG but we have
8433 * deferred frees from the previous TXG.
8435 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8436 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
8437 bpobj_spa_free_sync_cb
, zio
, tx
), ==, 0);
8438 VERIFY0(zio_wait(zio
));
8442 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
8444 char *packed
= NULL
;
8449 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
8452 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
8453 * information. This avoids the dmu_buf_will_dirty() path and
8454 * saves us a pre-read to get data we don't actually care about.
8456 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
8457 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
8459 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
8461 bzero(packed
+ nvsize
, bufsize
- nvsize
);
8463 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
8465 vmem_free(packed
, bufsize
);
8467 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
8468 dmu_buf_will_dirty(db
, tx
);
8469 *(uint64_t *)db
->db_data
= nvsize
;
8470 dmu_buf_rele(db
, FTAG
);
8474 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
8475 const char *config
, const char *entry
)
8485 * Update the MOS nvlist describing the list of available devices.
8486 * spa_validate_aux() will have already made sure this nvlist is
8487 * valid and the vdevs are labeled appropriately.
8489 if (sav
->sav_object
== 0) {
8490 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
8491 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
8492 sizeof (uint64_t), tx
);
8493 VERIFY(zap_update(spa
->spa_meta_objset
,
8494 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
8495 &sav
->sav_object
, tx
) == 0);
8498 nvroot
= fnvlist_alloc();
8499 if (sav
->sav_count
== 0) {
8500 fnvlist_add_nvlist_array(nvroot
, config
,
8501 (const nvlist_t
* const *)NULL
, 0);
8503 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
8504 for (i
= 0; i
< sav
->sav_count
; i
++)
8505 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
8506 B_FALSE
, VDEV_CONFIG_L2CACHE
);
8507 fnvlist_add_nvlist_array(nvroot
, config
,
8508 (const nvlist_t
* const *)list
, sav
->sav_count
);
8509 for (i
= 0; i
< sav
->sav_count
; i
++)
8510 nvlist_free(list
[i
]);
8511 kmem_free(list
, sav
->sav_count
* sizeof (void *));
8514 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
8515 nvlist_free(nvroot
);
8517 sav
->sav_sync
= B_FALSE
;
8521 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
8522 * The all-vdev ZAP must be empty.
8525 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
8527 spa_t
*spa
= vd
->vdev_spa
;
8529 if (vd
->vdev_top_zap
!= 0) {
8530 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8531 vd
->vdev_top_zap
, tx
));
8533 if (vd
->vdev_leaf_zap
!= 0) {
8534 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8535 vd
->vdev_leaf_zap
, tx
));
8537 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
8538 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
8543 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
8548 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
8549 * its config may not be dirty but we still need to build per-vdev ZAPs.
8550 * Similarly, if the pool is being assembled (e.g. after a split), we
8551 * need to rebuild the AVZ although the config may not be dirty.
8553 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
8554 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
8557 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8559 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
8560 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
8561 spa
->spa_all_vdev_zaps
!= 0);
8563 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
8564 /* Make and build the new AVZ */
8565 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
8566 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
8567 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
8569 /* Diff old AVZ with new one */
8573 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
8574 spa
->spa_all_vdev_zaps
);
8575 zap_cursor_retrieve(&zc
, &za
) == 0;
8576 zap_cursor_advance(&zc
)) {
8577 uint64_t vdzap
= za
.za_first_integer
;
8578 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
8581 * ZAP is listed in old AVZ but not in new one;
8584 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
8589 zap_cursor_fini(&zc
);
8591 /* Destroy the old AVZ */
8592 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
8593 spa
->spa_all_vdev_zaps
, tx
));
8595 /* Replace the old AVZ in the dir obj with the new one */
8596 VERIFY0(zap_update(spa
->spa_meta_objset
,
8597 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
8598 sizeof (new_avz
), 1, &new_avz
, tx
));
8600 spa
->spa_all_vdev_zaps
= new_avz
;
8601 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
8605 /* Walk through the AVZ and destroy all listed ZAPs */
8606 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
8607 spa
->spa_all_vdev_zaps
);
8608 zap_cursor_retrieve(&zc
, &za
) == 0;
8609 zap_cursor_advance(&zc
)) {
8610 uint64_t zap
= za
.za_first_integer
;
8611 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
8614 zap_cursor_fini(&zc
);
8616 /* Destroy and unlink the AVZ itself */
8617 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
8618 spa
->spa_all_vdev_zaps
, tx
));
8619 VERIFY0(zap_remove(spa
->spa_meta_objset
,
8620 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
8621 spa
->spa_all_vdev_zaps
= 0;
8624 if (spa
->spa_all_vdev_zaps
== 0) {
8625 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
8626 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
8627 DMU_POOL_VDEV_ZAP_MAP
, tx
);
8629 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
8631 /* Create ZAPs for vdevs that don't have them. */
8632 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
8634 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
8635 dmu_tx_get_txg(tx
), B_FALSE
);
8638 * If we're upgrading the spa version then make sure that
8639 * the config object gets updated with the correct version.
8641 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
8642 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
8643 spa
->spa_uberblock
.ub_version
);
8645 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8647 nvlist_free(spa
->spa_config_syncing
);
8648 spa
->spa_config_syncing
= config
;
8650 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
8654 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
8656 uint64_t *versionp
= arg
;
8657 uint64_t version
= *versionp
;
8658 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
8661 * Setting the version is special cased when first creating the pool.
8663 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
8665 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
8666 ASSERT(version
>= spa_version(spa
));
8668 spa
->spa_uberblock
.ub_version
= version
;
8669 vdev_config_dirty(spa
->spa_root_vdev
);
8670 spa_history_log_internal(spa
, "set", tx
, "version=%lld",
8671 (longlong_t
)version
);
8675 * Set zpool properties.
8678 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
8680 nvlist_t
*nvp
= arg
;
8681 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
8682 objset_t
*mos
= spa
->spa_meta_objset
;
8683 nvpair_t
*elem
= NULL
;
8685 mutex_enter(&spa
->spa_props_lock
);
8687 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
8689 char *strval
, *fname
;
8691 const char *propname
;
8692 zprop_type_t proptype
;
8695 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
8696 case ZPOOL_PROP_INVAL
:
8698 * We checked this earlier in spa_prop_validate().
8700 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
8702 fname
= strchr(nvpair_name(elem
), '@') + 1;
8703 VERIFY0(zfeature_lookup_name(fname
, &fid
));
8705 spa_feature_enable(spa
, fid
, tx
);
8706 spa_history_log_internal(spa
, "set", tx
,
8707 "%s=enabled", nvpair_name(elem
));
8710 case ZPOOL_PROP_VERSION
:
8711 intval
= fnvpair_value_uint64(elem
);
8713 * The version is synced separately before other
8714 * properties and should be correct by now.
8716 ASSERT3U(spa_version(spa
), >=, intval
);
8719 case ZPOOL_PROP_ALTROOT
:
8721 * 'altroot' is a non-persistent property. It should
8722 * have been set temporarily at creation or import time.
8724 ASSERT(spa
->spa_root
!= NULL
);
8727 case ZPOOL_PROP_READONLY
:
8728 case ZPOOL_PROP_CACHEFILE
:
8730 * 'readonly' and 'cachefile' are also non-persistent
8734 case ZPOOL_PROP_COMMENT
:
8735 strval
= fnvpair_value_string(elem
);
8736 if (spa
->spa_comment
!= NULL
)
8737 spa_strfree(spa
->spa_comment
);
8738 spa
->spa_comment
= spa_strdup(strval
);
8740 * We need to dirty the configuration on all the vdevs
8741 * so that their labels get updated. We also need to
8742 * update the cache file to keep it in sync with the
8743 * MOS version. It's unnecessary to do this for pool
8744 * creation since the vdev's configuration has already
8747 if (tx
->tx_txg
!= TXG_INITIAL
) {
8748 vdev_config_dirty(spa
->spa_root_vdev
);
8749 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
8751 spa_history_log_internal(spa
, "set", tx
,
8752 "%s=%s", nvpair_name(elem
), strval
);
8754 case ZPOOL_PROP_COMPATIBILITY
:
8755 strval
= fnvpair_value_string(elem
);
8756 if (spa
->spa_compatibility
!= NULL
)
8757 spa_strfree(spa
->spa_compatibility
);
8758 spa
->spa_compatibility
= spa_strdup(strval
);
8760 * Dirty the configuration on vdevs as above.
8762 if (tx
->tx_txg
!= TXG_INITIAL
) {
8763 vdev_config_dirty(spa
->spa_root_vdev
);
8764 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
8767 spa_history_log_internal(spa
, "set", tx
,
8768 "%s=%s", nvpair_name(elem
), strval
);
8773 * Set pool property values in the poolprops mos object.
8775 if (spa
->spa_pool_props_object
== 0) {
8776 spa
->spa_pool_props_object
=
8777 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
8778 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
8782 /* normalize the property name */
8783 propname
= zpool_prop_to_name(prop
);
8784 proptype
= zpool_prop_get_type(prop
);
8786 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
8787 ASSERT(proptype
== PROP_TYPE_STRING
);
8788 strval
= fnvpair_value_string(elem
);
8789 VERIFY0(zap_update(mos
,
8790 spa
->spa_pool_props_object
, propname
,
8791 1, strlen(strval
) + 1, strval
, tx
));
8792 spa_history_log_internal(spa
, "set", tx
,
8793 "%s=%s", nvpair_name(elem
), strval
);
8794 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
8795 intval
= fnvpair_value_uint64(elem
);
8797 if (proptype
== PROP_TYPE_INDEX
) {
8799 VERIFY0(zpool_prop_index_to_string(
8800 prop
, intval
, &unused
));
8802 VERIFY0(zap_update(mos
,
8803 spa
->spa_pool_props_object
, propname
,
8804 8, 1, &intval
, tx
));
8805 spa_history_log_internal(spa
, "set", tx
,
8806 "%s=%lld", nvpair_name(elem
),
8807 (longlong_t
)intval
);
8809 ASSERT(0); /* not allowed */
8813 case ZPOOL_PROP_DELEGATION
:
8814 spa
->spa_delegation
= intval
;
8816 case ZPOOL_PROP_BOOTFS
:
8817 spa
->spa_bootfs
= intval
;
8819 case ZPOOL_PROP_FAILUREMODE
:
8820 spa
->spa_failmode
= intval
;
8822 case ZPOOL_PROP_AUTOTRIM
:
8823 spa
->spa_autotrim
= intval
;
8824 spa_async_request(spa
,
8825 SPA_ASYNC_AUTOTRIM_RESTART
);
8827 case ZPOOL_PROP_AUTOEXPAND
:
8828 spa
->spa_autoexpand
= intval
;
8829 if (tx
->tx_txg
!= TXG_INITIAL
)
8830 spa_async_request(spa
,
8831 SPA_ASYNC_AUTOEXPAND
);
8833 case ZPOOL_PROP_MULTIHOST
:
8834 spa
->spa_multihost
= intval
;
8843 mutex_exit(&spa
->spa_props_lock
);
8847 * Perform one-time upgrade on-disk changes. spa_version() does not
8848 * reflect the new version this txg, so there must be no changes this
8849 * txg to anything that the upgrade code depends on after it executes.
8850 * Therefore this must be called after dsl_pool_sync() does the sync
8854 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
8856 if (spa_sync_pass(spa
) != 1)
8859 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8860 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
8862 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
8863 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
8864 dsl_pool_create_origin(dp
, tx
);
8866 /* Keeping the origin open increases spa_minref */
8867 spa
->spa_minref
+= 3;
8870 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
8871 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
8872 dsl_pool_upgrade_clones(dp
, tx
);
8875 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
8876 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
8877 dsl_pool_upgrade_dir_clones(dp
, tx
);
8879 /* Keeping the freedir open increases spa_minref */
8880 spa
->spa_minref
+= 3;
8883 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
8884 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
8885 spa_feature_create_zap_objects(spa
, tx
);
8889 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
8890 * when possibility to use lz4 compression for metadata was added
8891 * Old pools that have this feature enabled must be upgraded to have
8892 * this feature active
8894 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
8895 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
8896 SPA_FEATURE_LZ4_COMPRESS
);
8897 boolean_t lz4_ac
= spa_feature_is_active(spa
,
8898 SPA_FEATURE_LZ4_COMPRESS
);
8900 if (lz4_en
&& !lz4_ac
)
8901 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
8905 * If we haven't written the salt, do so now. Note that the
8906 * feature may not be activated yet, but that's fine since
8907 * the presence of this ZAP entry is backwards compatible.
8909 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
8910 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
8911 VERIFY0(zap_add(spa
->spa_meta_objset
,
8912 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
8913 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
8914 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
8917 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
8921 vdev_indirect_state_sync_verify(vdev_t
*vd
)
8923 vdev_indirect_mapping_t
*vim __maybe_unused
= vd
->vdev_indirect_mapping
;
8924 vdev_indirect_births_t
*vib __maybe_unused
= vd
->vdev_indirect_births
;
8926 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
8927 ASSERT(vim
!= NULL
);
8928 ASSERT(vib
!= NULL
);
8931 uint64_t obsolete_sm_object
= 0;
8932 ASSERT0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
8933 if (obsolete_sm_object
!= 0) {
8934 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
8935 ASSERT(vd
->vdev_removing
||
8936 vd
->vdev_ops
== &vdev_indirect_ops
);
8937 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
8938 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
8939 ASSERT3U(obsolete_sm_object
, ==,
8940 space_map_object(vd
->vdev_obsolete_sm
));
8941 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
8942 space_map_allocated(vd
->vdev_obsolete_sm
));
8944 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
8947 * Since frees / remaps to an indirect vdev can only
8948 * happen in syncing context, the obsolete segments
8949 * tree must be empty when we start syncing.
8951 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
8955 * Set the top-level vdev's max queue depth. Evaluate each top-level's
8956 * async write queue depth in case it changed. The max queue depth will
8957 * not change in the middle of syncing out this txg.
8960 spa_sync_adjust_vdev_max_queue_depth(spa_t
*spa
)
8962 ASSERT(spa_writeable(spa
));
8964 vdev_t
*rvd
= spa
->spa_root_vdev
;
8965 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
8966 zfs_vdev_queue_depth_pct
/ 100;
8967 metaslab_class_t
*normal
= spa_normal_class(spa
);
8968 metaslab_class_t
*special
= spa_special_class(spa
);
8969 metaslab_class_t
*dedup
= spa_dedup_class(spa
);
8971 uint64_t slots_per_allocator
= 0;
8972 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
8973 vdev_t
*tvd
= rvd
->vdev_child
[c
];
8975 metaslab_group_t
*mg
= tvd
->vdev_mg
;
8976 if (mg
== NULL
|| !metaslab_group_initialized(mg
))
8979 metaslab_class_t
*mc
= mg
->mg_class
;
8980 if (mc
!= normal
&& mc
!= special
&& mc
!= dedup
)
8984 * It is safe to do a lock-free check here because only async
8985 * allocations look at mg_max_alloc_queue_depth, and async
8986 * allocations all happen from spa_sync().
8988 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
8989 ASSERT0(zfs_refcount_count(
8990 &(mg
->mg_allocator
[i
].mga_alloc_queue_depth
)));
8992 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
8994 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
8995 mg
->mg_allocator
[i
].mga_cur_max_alloc_queue_depth
=
8996 zfs_vdev_def_queue_depth
;
8998 slots_per_allocator
+= zfs_vdev_def_queue_depth
;
9001 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9002 ASSERT0(zfs_refcount_count(&normal
->mc_allocator
[i
].
9004 ASSERT0(zfs_refcount_count(&special
->mc_allocator
[i
].
9006 ASSERT0(zfs_refcount_count(&dedup
->mc_allocator
[i
].
9008 normal
->mc_allocator
[i
].mca_alloc_max_slots
=
9009 slots_per_allocator
;
9010 special
->mc_allocator
[i
].mca_alloc_max_slots
=
9011 slots_per_allocator
;
9012 dedup
->mc_allocator
[i
].mca_alloc_max_slots
=
9013 slots_per_allocator
;
9015 normal
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9016 special
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9017 dedup
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9021 spa_sync_condense_indirect(spa_t
*spa
, dmu_tx_t
*tx
)
9023 ASSERT(spa_writeable(spa
));
9025 vdev_t
*rvd
= spa
->spa_root_vdev
;
9026 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
9027 vdev_t
*vd
= rvd
->vdev_child
[c
];
9028 vdev_indirect_state_sync_verify(vd
);
9030 if (vdev_indirect_should_condense(vd
)) {
9031 spa_condense_indirect_start_sync(vd
, tx
);
9038 spa_sync_iterate_to_convergence(spa_t
*spa
, dmu_tx_t
*tx
)
9040 objset_t
*mos
= spa
->spa_meta_objset
;
9041 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9042 uint64_t txg
= tx
->tx_txg
;
9043 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
9046 int pass
= ++spa
->spa_sync_pass
;
9048 spa_sync_config_object(spa
, tx
);
9049 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
9050 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
9051 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
9052 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
9053 spa_errlog_sync(spa
, txg
);
9054 dsl_pool_sync(dp
, txg
);
9056 if (pass
< zfs_sync_pass_deferred_free
||
9057 spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
)) {
9059 * If the log space map feature is active we don't
9060 * care about deferred frees and the deferred bpobj
9061 * as the log space map should effectively have the
9062 * same results (i.e. appending only to one object).
9064 spa_sync_frees(spa
, free_bpl
, tx
);
9067 * We can not defer frees in pass 1, because
9068 * we sync the deferred frees later in pass 1.
9070 ASSERT3U(pass
, >, 1);
9071 bplist_iterate(free_bpl
, bpobj_enqueue_alloc_cb
,
9072 &spa
->spa_deferred_bpobj
, tx
);
9076 dsl_scan_sync(dp
, tx
);
9078 spa_sync_upgrades(spa
, tx
);
9080 spa_flush_metaslabs(spa
, tx
);
9083 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
9088 * Note: We need to check if the MOS is dirty because we could
9089 * have marked the MOS dirty without updating the uberblock
9090 * (e.g. if we have sync tasks but no dirty user data). We need
9091 * to check the uberblock's rootbp because it is updated if we
9092 * have synced out dirty data (though in this case the MOS will
9093 * most likely also be dirty due to second order effects, we
9094 * don't want to rely on that here).
9097 spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
9098 !dmu_objset_is_dirty(mos
, txg
)) {
9100 * Nothing changed on the first pass, therefore this
9101 * TXG is a no-op. Avoid syncing deferred frees, so
9102 * that we can keep this TXG as a no-op.
9104 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9105 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9106 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
9107 ASSERT(txg_list_empty(&dp
->dp_early_sync_tasks
, txg
));
9111 spa_sync_deferred_frees(spa
, tx
);
9112 } while (dmu_objset_is_dirty(mos
, txg
));
9116 * Rewrite the vdev configuration (which includes the uberblock) to
9117 * commit the transaction group.
9119 * If there are no dirty vdevs, we sync the uberblock to a few random
9120 * top-level vdevs that are known to be visible in the config cache
9121 * (see spa_vdev_add() for a complete description). If there *are* dirty
9122 * vdevs, sync the uberblock to all vdevs.
9125 spa_sync_rewrite_vdev_config(spa_t
*spa
, dmu_tx_t
*tx
)
9127 vdev_t
*rvd
= spa
->spa_root_vdev
;
9128 uint64_t txg
= tx
->tx_txg
;
9134 * We hold SCL_STATE to prevent vdev open/close/etc.
9135 * while we're attempting to write the vdev labels.
9137 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9139 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
9140 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
9142 int children
= rvd
->vdev_children
;
9143 int c0
= random_in_range(children
);
9145 for (int c
= 0; c
< children
; c
++) {
9147 rvd
->vdev_child
[(c0
+ c
) % children
];
9149 /* Stop when revisiting the first vdev */
9150 if (c
> 0 && svd
[0] == vd
)
9153 if (vd
->vdev_ms_array
== 0 ||
9155 !vdev_is_concrete(vd
))
9158 svd
[svdcount
++] = vd
;
9159 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
9162 error
= vdev_config_sync(svd
, svdcount
, txg
);
9164 error
= vdev_config_sync(rvd
->vdev_child
,
9165 rvd
->vdev_children
, txg
);
9169 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
9171 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9175 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
9176 zio_resume_wait(spa
);
9181 * Sync the specified transaction group. New blocks may be dirtied as
9182 * part of the process, so we iterate until it converges.
9185 spa_sync(spa_t
*spa
, uint64_t txg
)
9189 VERIFY(spa_writeable(spa
));
9192 * Wait for i/os issued in open context that need to complete
9193 * before this txg syncs.
9195 (void) zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]);
9196 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
,
9200 * Lock out configuration changes.
9202 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
9204 spa
->spa_syncing_txg
= txg
;
9205 spa
->spa_sync_pass
= 0;
9207 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9208 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
9209 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
9210 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
9214 * If there are any pending vdev state changes, convert them
9215 * into config changes that go out with this transaction group.
9217 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9218 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
9220 * We need the write lock here because, for aux vdevs,
9221 * calling vdev_config_dirty() modifies sav_config.
9222 * This is ugly and will become unnecessary when we
9223 * eliminate the aux vdev wart by integrating all vdevs
9224 * into the root vdev tree.
9226 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9227 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
9228 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
9229 vdev_state_clean(vd
);
9230 vdev_config_dirty(vd
);
9232 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9233 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9235 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9237 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9238 dmu_tx_t
*tx
= dmu_tx_create_assigned(dp
, txg
);
9240 spa
->spa_sync_starttime
= gethrtime();
9241 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9242 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
9243 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
9244 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
9247 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9248 * set spa_deflate if we have no raid-z vdevs.
9250 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
9251 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
9252 vdev_t
*rvd
= spa
->spa_root_vdev
;
9255 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
9256 vd
= rvd
->vdev_child
[i
];
9257 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
9260 if (i
== rvd
->vdev_children
) {
9261 spa
->spa_deflate
= TRUE
;
9262 VERIFY0(zap_add(spa
->spa_meta_objset
,
9263 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
9264 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
9268 spa_sync_adjust_vdev_max_queue_depth(spa
);
9270 spa_sync_condense_indirect(spa
, tx
);
9272 spa_sync_iterate_to_convergence(spa
, tx
);
9275 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
9277 * Make sure that the number of ZAPs for all the vdevs matches
9278 * the number of ZAPs in the per-vdev ZAP list. This only gets
9279 * called if the config is dirty; otherwise there may be
9280 * outstanding AVZ operations that weren't completed in
9281 * spa_sync_config_object.
9283 uint64_t all_vdev_zap_entry_count
;
9284 ASSERT0(zap_count(spa
->spa_meta_objset
,
9285 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
9286 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
9287 all_vdev_zap_entry_count
);
9291 if (spa
->spa_vdev_removal
!= NULL
) {
9292 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
9295 spa_sync_rewrite_vdev_config(spa
, tx
);
9298 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9299 spa
->spa_deadman_tqid
= 0;
9302 * Clear the dirty config list.
9304 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
9305 vdev_config_clean(vd
);
9308 * Now that the new config has synced transactionally,
9309 * let it become visible to the config cache.
9311 if (spa
->spa_config_syncing
!= NULL
) {
9312 spa_config_set(spa
, spa
->spa_config_syncing
);
9313 spa
->spa_config_txg
= txg
;
9314 spa
->spa_config_syncing
= NULL
;
9317 dsl_pool_sync_done(dp
, txg
);
9319 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9320 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
9321 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
9322 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
9326 * Update usable space statistics.
9328 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
9330 vdev_sync_done(vd
, txg
);
9332 metaslab_class_evict_old(spa
->spa_normal_class
, txg
);
9333 metaslab_class_evict_old(spa
->spa_log_class
, txg
);
9335 spa_sync_close_syncing_log_sm(spa
);
9337 spa_update_dspace(spa
);
9340 * It had better be the case that we didn't dirty anything
9341 * since vdev_config_sync().
9343 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9344 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9345 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
9347 while (zfs_pause_spa_sync
)
9350 spa
->spa_sync_pass
= 0;
9353 * Update the last synced uberblock here. We want to do this at
9354 * the end of spa_sync() so that consumers of spa_last_synced_txg()
9355 * will be guaranteed that all the processing associated with
9356 * that txg has been completed.
9358 spa
->spa_ubsync
= spa
->spa_uberblock
;
9359 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
9361 spa_handle_ignored_writes(spa
);
9364 * If any async tasks have been requested, kick them off.
9366 spa_async_dispatch(spa
);
9370 * Sync all pools. We don't want to hold the namespace lock across these
9371 * operations, so we take a reference on the spa_t and drop the lock during the
9375 spa_sync_allpools(void)
9378 mutex_enter(&spa_namespace_lock
);
9379 while ((spa
= spa_next(spa
)) != NULL
) {
9380 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
9381 !spa_writeable(spa
) || spa_suspended(spa
))
9383 spa_open_ref(spa
, FTAG
);
9384 mutex_exit(&spa_namespace_lock
);
9385 txg_wait_synced(spa_get_dsl(spa
), 0);
9386 mutex_enter(&spa_namespace_lock
);
9387 spa_close(spa
, FTAG
);
9389 mutex_exit(&spa_namespace_lock
);
9393 * ==========================================================================
9394 * Miscellaneous routines
9395 * ==========================================================================
9399 * Remove all pools in the system.
9407 * Remove all cached state. All pools should be closed now,
9408 * so every spa in the AVL tree should be unreferenced.
9410 mutex_enter(&spa_namespace_lock
);
9411 while ((spa
= spa_next(NULL
)) != NULL
) {
9413 * Stop async tasks. The async thread may need to detach
9414 * a device that's been replaced, which requires grabbing
9415 * spa_namespace_lock, so we must drop it here.
9417 spa_open_ref(spa
, FTAG
);
9418 mutex_exit(&spa_namespace_lock
);
9419 spa_async_suspend(spa
);
9420 mutex_enter(&spa_namespace_lock
);
9421 spa_close(spa
, FTAG
);
9423 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
9425 spa_deactivate(spa
);
9429 mutex_exit(&spa_namespace_lock
);
9433 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
9438 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
9442 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
9443 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
9444 if (vd
->vdev_guid
== guid
)
9448 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
9449 vd
= spa
->spa_spares
.sav_vdevs
[i
];
9450 if (vd
->vdev_guid
== guid
)
9459 spa_upgrade(spa_t
*spa
, uint64_t version
)
9461 ASSERT(spa_writeable(spa
));
9463 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
9466 * This should only be called for a non-faulted pool, and since a
9467 * future version would result in an unopenable pool, this shouldn't be
9470 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
9471 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
9473 spa
->spa_uberblock
.ub_version
= version
;
9474 vdev_config_dirty(spa
->spa_root_vdev
);
9476 spa_config_exit(spa
, SCL_ALL
, FTAG
);
9478 txg_wait_synced(spa_get_dsl(spa
), 0);
9482 spa_has_aux_vdev(spa_t
*spa
, uint64_t guid
, spa_aux_vdev_t
*sav
)
9488 for (i
= 0; i
< sav
->sav_count
; i
++)
9489 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
9492 for (i
= 0; i
< sav
->sav_npending
; i
++) {
9493 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
9494 &vdev_guid
) == 0 && vdev_guid
== guid
)
9502 spa_has_l2cache(spa_t
*spa
, uint64_t guid
)
9504 return (spa_has_aux_vdev(spa
, guid
, &spa
->spa_l2cache
));
9508 spa_has_spare(spa_t
*spa
, uint64_t guid
)
9510 return (spa_has_aux_vdev(spa
, guid
, &spa
->spa_spares
));
9514 * Check if a pool has an active shared spare device.
9515 * Note: reference count of an active spare is 2, as a spare and as a replace
9518 spa_has_active_shared_spare(spa_t
*spa
)
9522 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
9524 for (i
= 0; i
< sav
->sav_count
; i
++) {
9525 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
9526 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
9535 spa_total_metaslabs(spa_t
*spa
)
9537 vdev_t
*rvd
= spa
->spa_root_vdev
;
9540 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
9541 vdev_t
*vd
= rvd
->vdev_child
[c
];
9542 if (!vdev_is_concrete(vd
))
9544 m
+= vd
->vdev_ms_count
;
9550 * Notify any waiting threads that some activity has switched from being in-
9551 * progress to not-in-progress so that the thread can wake up and determine
9552 * whether it is finished waiting.
9555 spa_notify_waiters(spa_t
*spa
)
9558 * Acquiring spa_activities_lock here prevents the cv_broadcast from
9559 * happening between the waiting thread's check and cv_wait.
9561 mutex_enter(&spa
->spa_activities_lock
);
9562 cv_broadcast(&spa
->spa_activities_cv
);
9563 mutex_exit(&spa
->spa_activities_lock
);
9567 * Notify any waiting threads that the pool is exporting, and then block until
9568 * they are finished using the spa_t.
9571 spa_wake_waiters(spa_t
*spa
)
9573 mutex_enter(&spa
->spa_activities_lock
);
9574 spa
->spa_waiters_cancel
= B_TRUE
;
9575 cv_broadcast(&spa
->spa_activities_cv
);
9576 while (spa
->spa_waiters
!= 0)
9577 cv_wait(&spa
->spa_waiters_cv
, &spa
->spa_activities_lock
);
9578 spa
->spa_waiters_cancel
= B_FALSE
;
9579 mutex_exit(&spa
->spa_activities_lock
);
9582 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
9584 spa_vdev_activity_in_progress_impl(vdev_t
*vd
, zpool_wait_activity_t activity
)
9586 spa_t
*spa
= vd
->vdev_spa
;
9588 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
));
9589 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
9590 ASSERT(activity
== ZPOOL_WAIT_INITIALIZE
||
9591 activity
== ZPOOL_WAIT_TRIM
);
9593 kmutex_t
*lock
= activity
== ZPOOL_WAIT_INITIALIZE
?
9594 &vd
->vdev_initialize_lock
: &vd
->vdev_trim_lock
;
9596 mutex_exit(&spa
->spa_activities_lock
);
9598 mutex_enter(&spa
->spa_activities_lock
);
9600 boolean_t in_progress
= (activity
== ZPOOL_WAIT_INITIALIZE
) ?
9601 (vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) :
9602 (vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
);
9608 for (int i
= 0; i
< vd
->vdev_children
; i
++) {
9609 if (spa_vdev_activity_in_progress_impl(vd
->vdev_child
[i
],
9618 * If use_guid is true, this checks whether the vdev specified by guid is
9619 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
9620 * is being initialized/trimmed. The caller must hold the config lock and
9621 * spa_activities_lock.
9624 spa_vdev_activity_in_progress(spa_t
*spa
, boolean_t use_guid
, uint64_t guid
,
9625 zpool_wait_activity_t activity
, boolean_t
*in_progress
)
9627 mutex_exit(&spa
->spa_activities_lock
);
9628 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9629 mutex_enter(&spa
->spa_activities_lock
);
9633 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
9634 if (vd
== NULL
|| !vd
->vdev_ops
->vdev_op_leaf
) {
9635 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9639 vd
= spa
->spa_root_vdev
;
9642 *in_progress
= spa_vdev_activity_in_progress_impl(vd
, activity
);
9644 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9649 * Locking for waiting threads
9650 * ---------------------------
9652 * Waiting threads need a way to check whether a given activity is in progress,
9653 * and then, if it is, wait for it to complete. Each activity will have some
9654 * in-memory representation of the relevant on-disk state which can be used to
9655 * determine whether or not the activity is in progress. The in-memory state and
9656 * the locking used to protect it will be different for each activity, and may
9657 * not be suitable for use with a cvar (e.g., some state is protected by the
9658 * config lock). To allow waiting threads to wait without any races, another
9659 * lock, spa_activities_lock, is used.
9661 * When the state is checked, both the activity-specific lock (if there is one)
9662 * and spa_activities_lock are held. In some cases, the activity-specific lock
9663 * is acquired explicitly (e.g. the config lock). In others, the locking is
9664 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
9665 * thread releases the activity-specific lock and, if the activity is in
9666 * progress, then cv_waits using spa_activities_lock.
9668 * The waiting thread is woken when another thread, one completing some
9669 * activity, updates the state of the activity and then calls
9670 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
9671 * needs to hold its activity-specific lock when updating the state, and this
9672 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
9674 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
9675 * and because it is held when the waiting thread checks the state of the
9676 * activity, it can never be the case that the completing thread both updates
9677 * the activity state and cv_broadcasts in between the waiting thread's check
9678 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
9680 * In order to prevent deadlock, when the waiting thread does its check, in some
9681 * cases it will temporarily drop spa_activities_lock in order to acquire the
9682 * activity-specific lock. The order in which spa_activities_lock and the
9683 * activity specific lock are acquired in the waiting thread is determined by
9684 * the order in which they are acquired in the completing thread; if the
9685 * completing thread calls spa_notify_waiters with the activity-specific lock
9686 * held, then the waiting thread must also acquire the activity-specific lock
9691 spa_activity_in_progress(spa_t
*spa
, zpool_wait_activity_t activity
,
9692 boolean_t use_tag
, uint64_t tag
, boolean_t
*in_progress
)
9696 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
9699 case ZPOOL_WAIT_CKPT_DISCARD
:
9701 (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
) &&
9702 zap_contains(spa_meta_objset(spa
),
9703 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ZPOOL_CHECKPOINT
) ==
9706 case ZPOOL_WAIT_FREE
:
9707 *in_progress
= ((spa_version(spa
) >= SPA_VERSION_DEADLISTS
&&
9708 !bpobj_is_empty(&spa
->spa_dsl_pool
->dp_free_bpobj
)) ||
9709 spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
) ||
9710 spa_livelist_delete_check(spa
));
9712 case ZPOOL_WAIT_INITIALIZE
:
9713 case ZPOOL_WAIT_TRIM
:
9714 error
= spa_vdev_activity_in_progress(spa
, use_tag
, tag
,
9715 activity
, in_progress
);
9717 case ZPOOL_WAIT_REPLACE
:
9718 mutex_exit(&spa
->spa_activities_lock
);
9719 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9720 mutex_enter(&spa
->spa_activities_lock
);
9722 *in_progress
= vdev_replace_in_progress(spa
->spa_root_vdev
);
9723 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9725 case ZPOOL_WAIT_REMOVE
:
9726 *in_progress
= (spa
->spa_removing_phys
.sr_state
==
9729 case ZPOOL_WAIT_RESILVER
:
9730 if ((*in_progress
= vdev_rebuild_active(spa
->spa_root_vdev
)))
9733 case ZPOOL_WAIT_SCRUB
:
9735 boolean_t scanning
, paused
, is_scrub
;
9736 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
9738 is_scrub
= (scn
->scn_phys
.scn_func
== POOL_SCAN_SCRUB
);
9739 scanning
= (scn
->scn_phys
.scn_state
== DSS_SCANNING
);
9740 paused
= dsl_scan_is_paused_scrub(scn
);
9741 *in_progress
= (scanning
&& !paused
&&
9742 is_scrub
== (activity
== ZPOOL_WAIT_SCRUB
));
9746 panic("unrecognized value for activity %d", activity
);
9753 spa_wait_common(const char *pool
, zpool_wait_activity_t activity
,
9754 boolean_t use_tag
, uint64_t tag
, boolean_t
*waited
)
9757 * The tag is used to distinguish between instances of an activity.
9758 * 'initialize' and 'trim' are the only activities that we use this for.
9759 * The other activities can only have a single instance in progress in a
9760 * pool at one time, making the tag unnecessary.
9762 * There can be multiple devices being replaced at once, but since they
9763 * all finish once resilvering finishes, we don't bother keeping track
9764 * of them individually, we just wait for them all to finish.
9766 if (use_tag
&& activity
!= ZPOOL_WAIT_INITIALIZE
&&
9767 activity
!= ZPOOL_WAIT_TRIM
)
9770 if (activity
< 0 || activity
>= ZPOOL_WAIT_NUM_ACTIVITIES
)
9774 int error
= spa_open(pool
, &spa
, FTAG
);
9779 * Increment the spa's waiter count so that we can call spa_close and
9780 * still ensure that the spa_t doesn't get freed before this thread is
9781 * finished with it when the pool is exported. We want to call spa_close
9782 * before we start waiting because otherwise the additional ref would
9783 * prevent the pool from being exported or destroyed throughout the
9784 * potentially long wait.
9786 mutex_enter(&spa
->spa_activities_lock
);
9788 spa_close(spa
, FTAG
);
9792 boolean_t in_progress
;
9793 error
= spa_activity_in_progress(spa
, activity
, use_tag
, tag
,
9796 if (error
|| !in_progress
|| spa
->spa_waiters_cancel
)
9801 if (cv_wait_sig(&spa
->spa_activities_cv
,
9802 &spa
->spa_activities_lock
) == 0) {
9809 cv_signal(&spa
->spa_waiters_cv
);
9810 mutex_exit(&spa
->spa_activities_lock
);
9816 * Wait for a particular instance of the specified activity to complete, where
9817 * the instance is identified by 'tag'
9820 spa_wait_tag(const char *pool
, zpool_wait_activity_t activity
, uint64_t tag
,
9823 return (spa_wait_common(pool
, activity
, B_TRUE
, tag
, waited
));
9827 * Wait for all instances of the specified activity complete
9830 spa_wait(const char *pool
, zpool_wait_activity_t activity
, boolean_t
*waited
)
9833 return (spa_wait_common(pool
, activity
, B_FALSE
, 0, waited
));
9837 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
9839 sysevent_t
*ev
= NULL
;
9843 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
9845 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
9846 ev
->resource
= resource
;
9849 (void) spa
, (void) vd
, (void) hist_nvl
, (void) name
;
9855 spa_event_post(sysevent_t
*ev
)
9859 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
9860 kmem_free(ev
, sizeof (*ev
));
9868 * Post a zevent corresponding to the given sysevent. The 'name' must be one
9869 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
9870 * filled in from the spa and (optionally) the vdev. This doesn't do anything
9871 * in the userland libzpool, as we don't want consumers to misinterpret ztest
9872 * or zdb as real changes.
9875 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
9877 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
9880 /* state manipulation functions */
9881 EXPORT_SYMBOL(spa_open
);
9882 EXPORT_SYMBOL(spa_open_rewind
);
9883 EXPORT_SYMBOL(spa_get_stats
);
9884 EXPORT_SYMBOL(spa_create
);
9885 EXPORT_SYMBOL(spa_import
);
9886 EXPORT_SYMBOL(spa_tryimport
);
9887 EXPORT_SYMBOL(spa_destroy
);
9888 EXPORT_SYMBOL(spa_export
);
9889 EXPORT_SYMBOL(spa_reset
);
9890 EXPORT_SYMBOL(spa_async_request
);
9891 EXPORT_SYMBOL(spa_async_suspend
);
9892 EXPORT_SYMBOL(spa_async_resume
);
9893 EXPORT_SYMBOL(spa_inject_addref
);
9894 EXPORT_SYMBOL(spa_inject_delref
);
9895 EXPORT_SYMBOL(spa_scan_stat_init
);
9896 EXPORT_SYMBOL(spa_scan_get_stats
);
9898 /* device manipulation */
9899 EXPORT_SYMBOL(spa_vdev_add
);
9900 EXPORT_SYMBOL(spa_vdev_attach
);
9901 EXPORT_SYMBOL(spa_vdev_detach
);
9902 EXPORT_SYMBOL(spa_vdev_setpath
);
9903 EXPORT_SYMBOL(spa_vdev_setfru
);
9904 EXPORT_SYMBOL(spa_vdev_split_mirror
);
9906 /* spare statech is global across all pools) */
9907 EXPORT_SYMBOL(spa_spare_add
);
9908 EXPORT_SYMBOL(spa_spare_remove
);
9909 EXPORT_SYMBOL(spa_spare_exists
);
9910 EXPORT_SYMBOL(spa_spare_activate
);
9912 /* L2ARC statech is global across all pools) */
9913 EXPORT_SYMBOL(spa_l2cache_add
);
9914 EXPORT_SYMBOL(spa_l2cache_remove
);
9915 EXPORT_SYMBOL(spa_l2cache_exists
);
9916 EXPORT_SYMBOL(spa_l2cache_activate
);
9917 EXPORT_SYMBOL(spa_l2cache_drop
);
9920 EXPORT_SYMBOL(spa_scan
);
9921 EXPORT_SYMBOL(spa_scan_stop
);
9924 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
9925 EXPORT_SYMBOL(spa_sync_allpools
);
9928 EXPORT_SYMBOL(spa_prop_set
);
9929 EXPORT_SYMBOL(spa_prop_get
);
9930 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
9932 /* asynchronous event notification */
9933 EXPORT_SYMBOL(spa_event_notify
);
9936 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_shift
, INT
, ZMOD_RW
,
9937 "log2 fraction of arc that can be used by inflight I/Os when "
9938 "verifying pool during import");
9941 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_metadata
, INT
, ZMOD_RW
,
9942 "Set to traverse metadata on pool import");
9944 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_data
, INT
, ZMOD_RW
,
9945 "Set to traverse data on pool import");
9947 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_print_vdev_tree
, INT
, ZMOD_RW
,
9948 "Print vdev tree to zfs_dbgmsg during pool import");
9950 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_pct
, UINT
, ZMOD_RD
,
9951 "Percentage of CPUs to run an IO worker thread");
9953 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_tpq
, UINT
, ZMOD_RD
,
9954 "Number of threads per IO worker taskqueue");
9957 ZFS_MODULE_PARAM(zfs
, zfs_
, max_missing_tvds
, ULONG
, ZMOD_RW
,
9958 "Allow importing pool with up to this number of missing top-level "
9959 "vdevs (in read-only mode)");
9962 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_pause
, INT
,
9963 ZMOD_RW
, "Set the livelist condense zthr to pause");
9965 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_pause
, INT
,
9966 ZMOD_RW
, "Set the livelist condense synctask to pause");
9969 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_cancel
,
9971 "Whether livelist condensing was canceled in the synctask");
9973 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_cancel
,
9975 "Whether livelist condensing was canceled in the zthr function");
9977 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, new_alloc
, INT
,
9979 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
9980 "was being condensed");