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 https://opensource.org/licenses/CDDL-1.0.
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>
36 * Copyright (c) 2023 Hewlett Packard Enterprise Development LP.
40 * SPA: Storage Pool Allocator
42 * This file contains all the routines used when modifying on-disk SPA state.
43 * This includes opening, importing, destroying, exporting a pool, and syncing a
47 #include <sys/zfs_context.h>
48 #include <sys/fm/fs/zfs.h>
49 #include <sys/spa_impl.h>
51 #include <sys/zio_checksum.h>
53 #include <sys/dmu_tx.h>
58 #include <sys/vdev_impl.h>
59 #include <sys/vdev_removal.h>
60 #include <sys/vdev_indirect_mapping.h>
61 #include <sys/vdev_indirect_births.h>
62 #include <sys/vdev_initialize.h>
63 #include <sys/vdev_rebuild.h>
64 #include <sys/vdev_trim.h>
65 #include <sys/vdev_disk.h>
66 #include <sys/vdev_raidz.h>
67 #include <sys/vdev_draid.h>
68 #include <sys/metaslab.h>
69 #include <sys/metaslab_impl.h>
71 #include <sys/uberblock_impl.h>
74 #include <sys/bpobj.h>
75 #include <sys/dmu_traverse.h>
76 #include <sys/dmu_objset.h>
77 #include <sys/unique.h>
78 #include <sys/dsl_pool.h>
79 #include <sys/dsl_dataset.h>
80 #include <sys/dsl_dir.h>
81 #include <sys/dsl_prop.h>
82 #include <sys/dsl_synctask.h>
83 #include <sys/fs/zfs.h>
85 #include <sys/callb.h>
86 #include <sys/systeminfo.h>
87 #include <sys/zfs_ioctl.h>
88 #include <sys/dsl_scan.h>
89 #include <sys/zfeature.h>
90 #include <sys/dsl_destroy.h>
94 #include <sys/fm/protocol.h>
95 #include <sys/fm/util.h>
96 #include <sys/callb.h>
98 #include <sys/vmsystm.h>
101 #include "zfs_prop.h"
102 #include "zfs_comutil.h"
103 #include <cityhash.h>
106 * spa_thread() existed on Illumos as a parent thread for the various worker
107 * threads that actually run the pool, as a way to both reference the entire
108 * pool work as a single object, and to share properties like scheduling
109 * options. It has not yet been adapted to Linux or FreeBSD. This define is
110 * used to mark related parts of the code to make things easier for the reader,
111 * and to compile this code out. It can be removed when someone implements it,
112 * moves it to some Illumos-specific place, or removes it entirely.
114 #undef HAVE_SPA_THREAD
117 * The "System Duty Cycle" scheduling class is an Illumos feature to help
118 * prevent CPU-intensive kernel threads from affecting latency on interactive
119 * threads. It doesn't exist on Linux or FreeBSD, so the supporting code is
120 * gated behind a define. On Illumos SDC depends on spa_thread(), but
121 * spa_thread() also has other uses, so this is a separate define.
126 * The interval, in seconds, at which failed configuration cache file writes
129 int zfs_ccw_retry_interval
= 300;
131 typedef enum zti_modes
{
132 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
133 ZTI_MODE_SCALE
, /* Taskqs scale with CPUs. */
134 ZTI_MODE_SYNC
, /* sync thread assigned */
135 ZTI_MODE_NULL
, /* don't create a taskq */
139 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
140 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
141 #define ZTI_SCALE { ZTI_MODE_SCALE, 0, 1 }
142 #define ZTI_SYNC { ZTI_MODE_SYNC, 0, 1 }
143 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
145 #define ZTI_N(n) ZTI_P(n, 1)
146 #define ZTI_ONE ZTI_N(1)
148 typedef struct zio_taskq_info
{
149 zti_modes_t zti_mode
;
154 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
155 "iss", "iss_h", "int", "int_h"
159 * This table defines the taskq settings for each ZFS I/O type. When
160 * initializing a pool, we use this table to create an appropriately sized
161 * taskq. Some operations are low volume and therefore have a small, static
162 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
163 * macros. Other operations process a large amount of data; the ZTI_SCALE
164 * macro causes us to create a taskq oriented for throughput. Some operations
165 * are so high frequency and short-lived that the taskq itself can become a
166 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
167 * additional degree of parallelism specified by the number of threads per-
168 * taskq and the number of taskqs; when dispatching an event in this case, the
169 * particular taskq is chosen at random. ZTI_SCALE uses a number of taskqs
170 * that scales with the number of CPUs.
172 * The different taskq priorities are to handle the different contexts (issue
173 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
174 * need to be handled with minimum delay.
176 static const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
177 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
178 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
179 { ZTI_N(8), ZTI_NULL
, ZTI_SCALE
, ZTI_NULL
}, /* READ */
180 { ZTI_SYNC
, ZTI_N(5), ZTI_SCALE
, ZTI_N(5) }, /* WRITE */
181 { ZTI_SCALE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
182 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
183 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
184 { ZTI_N(4), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* TRIM */
187 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
188 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
189 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
190 static int spa_load_impl(spa_t
*spa
, spa_import_type_t type
,
191 const char **ereport
);
192 static void spa_vdev_resilver_done(spa_t
*spa
);
195 * Percentage of all CPUs that can be used by the metaslab preload taskq.
197 static uint_t metaslab_preload_pct
= 50;
199 static uint_t zio_taskq_batch_pct
= 80; /* 1 thread per cpu in pset */
200 static uint_t zio_taskq_batch_tpq
; /* threads per taskq */
203 static const boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
204 static const uint_t zio_taskq_basedc
= 80; /* base duty cycle */
207 #ifdef HAVE_SPA_THREAD
208 static const boolean_t spa_create_process
= B_TRUE
; /* no process => no sysdc */
211 static uint_t zio_taskq_wr_iss_ncpus
= 0;
214 * Report any spa_load_verify errors found, but do not fail spa_load.
215 * This is used by zdb to analyze non-idle pools.
217 boolean_t spa_load_verify_dryrun
= B_FALSE
;
220 * Allow read spacemaps in case of readonly import (spa_mode == SPA_MODE_READ).
221 * This is used by zdb for spacemaps verification.
223 boolean_t spa_mode_readable_spacemaps
= B_FALSE
;
226 * This (illegal) pool name is used when temporarily importing a spa_t in order
227 * to get the vdev stats associated with the imported devices.
229 #define TRYIMPORT_NAME "$import"
232 * For debugging purposes: print out vdev tree during pool import.
234 static int spa_load_print_vdev_tree
= B_FALSE
;
237 * A non-zero value for zfs_max_missing_tvds means that we allow importing
238 * pools with missing top-level vdevs. This is strictly intended for advanced
239 * pool recovery cases since missing data is almost inevitable. Pools with
240 * missing devices can only be imported read-only for safety reasons, and their
241 * fail-mode will be automatically set to "continue".
243 * With 1 missing vdev we should be able to import the pool and mount all
244 * datasets. User data that was not modified after the missing device has been
245 * added should be recoverable. This means that snapshots created prior to the
246 * addition of that device should be completely intact.
248 * With 2 missing vdevs, some datasets may fail to mount since there are
249 * dataset statistics that are stored as regular metadata. Some data might be
250 * recoverable if those vdevs were added recently.
252 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
253 * may be missing entirely. Chances of data recovery are very low. Note that
254 * there are also risks of performing an inadvertent rewind as we might be
255 * missing all the vdevs with the latest uberblocks.
257 uint64_t zfs_max_missing_tvds
= 0;
260 * The parameters below are similar to zfs_max_missing_tvds but are only
261 * intended for a preliminary open of the pool with an untrusted config which
262 * might be incomplete or out-dated.
264 * We are more tolerant for pools opened from a cachefile since we could have
265 * an out-dated cachefile where a device removal was not registered.
266 * We could have set the limit arbitrarily high but in the case where devices
267 * are really missing we would want to return the proper error codes; we chose
268 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
269 * and we get a chance to retrieve the trusted config.
271 uint64_t zfs_max_missing_tvds_cachefile
= SPA_DVAS_PER_BP
- 1;
274 * In the case where config was assembled by scanning device paths (/dev/dsks
275 * by default) we are less tolerant since all the existing devices should have
276 * been detected and we want spa_load to return the right error codes.
278 uint64_t zfs_max_missing_tvds_scan
= 0;
281 * Debugging aid that pauses spa_sync() towards the end.
283 static const boolean_t zfs_pause_spa_sync
= B_FALSE
;
286 * Variables to indicate the livelist condense zthr func should wait at certain
287 * points for the livelist to be removed - used to test condense/destroy races
289 static int zfs_livelist_condense_zthr_pause
= 0;
290 static int zfs_livelist_condense_sync_pause
= 0;
293 * Variables to track whether or not condense cancellation has been
294 * triggered in testing.
296 static int zfs_livelist_condense_sync_cancel
= 0;
297 static int zfs_livelist_condense_zthr_cancel
= 0;
300 * Variable to track whether or not extra ALLOC blkptrs were added to a
301 * livelist entry while it was being condensed (caused by the way we track
302 * remapped blkptrs in dbuf_remap_impl)
304 static int zfs_livelist_condense_new_alloc
= 0;
307 * ==========================================================================
308 * SPA properties routines
309 * ==========================================================================
313 * Add a (source=src, propname=propval) list to an nvlist.
316 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, const char *strval
,
317 uint64_t intval
, zprop_source_t src
)
319 const char *propname
= zpool_prop_to_name(prop
);
322 propval
= fnvlist_alloc();
323 fnvlist_add_uint64(propval
, ZPROP_SOURCE
, src
);
326 fnvlist_add_string(propval
, ZPROP_VALUE
, strval
);
328 fnvlist_add_uint64(propval
, ZPROP_VALUE
, intval
);
330 fnvlist_add_nvlist(nvl
, propname
, propval
);
331 nvlist_free(propval
);
335 * Add a user property (source=src, propname=propval) to an nvlist.
338 spa_prop_add_user(nvlist_t
*nvl
, const char *propname
, char *strval
,
343 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
344 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
345 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
346 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
347 nvlist_free(propval
);
351 * Get property values from the spa configuration.
354 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
356 vdev_t
*rvd
= spa
->spa_root_vdev
;
357 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
358 uint64_t size
, alloc
, cap
, version
;
359 const zprop_source_t src
= ZPROP_SRC_NONE
;
360 spa_config_dirent_t
*dp
;
361 metaslab_class_t
*mc
= spa_normal_class(spa
);
363 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
366 alloc
= metaslab_class_get_alloc(mc
);
367 alloc
+= metaslab_class_get_alloc(spa_special_class(spa
));
368 alloc
+= metaslab_class_get_alloc(spa_dedup_class(spa
));
369 alloc
+= metaslab_class_get_alloc(spa_embedded_log_class(spa
));
371 size
= metaslab_class_get_space(mc
);
372 size
+= metaslab_class_get_space(spa_special_class(spa
));
373 size
+= metaslab_class_get_space(spa_dedup_class(spa
));
374 size
+= metaslab_class_get_space(spa_embedded_log_class(spa
));
376 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
377 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
378 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
379 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
381 spa_prop_add_list(*nvp
, ZPOOL_PROP_CHECKPOINT
, NULL
,
382 spa
->spa_checkpoint_info
.sci_dspace
, src
);
384 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
385 metaslab_class_fragmentation(mc
), src
);
386 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
387 metaslab_class_expandable_space(mc
), src
);
388 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
389 (spa_mode(spa
) == SPA_MODE_READ
), src
);
391 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
392 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
394 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
395 ddt_get_pool_dedup_ratio(spa
), src
);
396 spa_prop_add_list(*nvp
, ZPOOL_PROP_BCLONEUSED
, NULL
,
397 brt_get_used(spa
), src
);
398 spa_prop_add_list(*nvp
, ZPOOL_PROP_BCLONESAVED
, NULL
,
399 brt_get_saved(spa
), src
);
400 spa_prop_add_list(*nvp
, ZPOOL_PROP_BCLONERATIO
, NULL
,
401 brt_get_ratio(spa
), src
);
403 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
404 rvd
->vdev_state
, src
);
406 version
= spa_version(spa
);
407 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
)) {
408 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
409 version
, ZPROP_SRC_DEFAULT
);
411 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
412 version
, ZPROP_SRC_LOCAL
);
414 spa_prop_add_list(*nvp
, ZPOOL_PROP_LOAD_GUID
,
415 NULL
, spa_load_guid(spa
), src
);
420 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
421 * when opening pools before this version freedir will be NULL.
423 if (pool
->dp_free_dir
!= NULL
) {
424 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
425 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
428 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
432 if (pool
->dp_leak_dir
!= NULL
) {
433 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
434 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
437 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
442 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
444 if (spa
->spa_comment
!= NULL
) {
445 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
449 if (spa
->spa_compatibility
!= NULL
) {
450 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMPATIBILITY
,
451 spa
->spa_compatibility
, 0, ZPROP_SRC_LOCAL
);
454 if (spa
->spa_root
!= NULL
)
455 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
458 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
459 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
460 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
462 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
463 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
466 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
)) {
467 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
468 DNODE_MAX_SIZE
, ZPROP_SRC_NONE
);
470 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
471 DNODE_MIN_SIZE
, ZPROP_SRC_NONE
);
474 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
475 if (dp
->scd_path
== NULL
) {
476 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
477 "none", 0, ZPROP_SRC_LOCAL
);
478 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
479 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
480 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
486 * Get zpool property values.
489 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
491 objset_t
*mos
= spa
->spa_meta_objset
;
497 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
501 dp
= spa_get_dsl(spa
);
502 dsl_pool_config_enter(dp
, FTAG
);
503 mutex_enter(&spa
->spa_props_lock
);
506 * Get properties from the spa config.
508 spa_prop_get_config(spa
, nvp
);
510 /* If no pool property object, no more prop to get. */
511 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0)
515 * Get properties from the MOS pool property object.
517 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
518 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
519 zap_cursor_advance(&zc
)) {
522 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
525 if ((prop
= zpool_name_to_prop(za
.za_name
)) ==
526 ZPOOL_PROP_INVAL
&& !zfs_prop_user(za
.za_name
))
529 switch (za
.za_integer_length
) {
531 /* integer property */
532 if (za
.za_first_integer
!=
533 zpool_prop_default_numeric(prop
))
534 src
= ZPROP_SRC_LOCAL
;
536 if (prop
== ZPOOL_PROP_BOOTFS
) {
537 dsl_dataset_t
*ds
= NULL
;
539 err
= dsl_dataset_hold_obj(dp
,
540 za
.za_first_integer
, FTAG
, &ds
);
544 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
546 dsl_dataset_name(ds
, strval
);
547 dsl_dataset_rele(ds
, FTAG
);
550 intval
= za
.za_first_integer
;
553 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
556 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
561 /* string property */
562 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
563 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
564 za
.za_name
, 1, za
.za_num_integers
, strval
);
566 kmem_free(strval
, za
.za_num_integers
);
569 if (prop
!= ZPOOL_PROP_INVAL
) {
570 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
572 src
= ZPROP_SRC_LOCAL
;
573 spa_prop_add_user(*nvp
, za
.za_name
, strval
,
576 kmem_free(strval
, za
.za_num_integers
);
583 zap_cursor_fini(&zc
);
585 mutex_exit(&spa
->spa_props_lock
);
586 dsl_pool_config_exit(dp
, FTAG
);
587 if (err
&& err
!= ENOENT
) {
597 * Validate the given pool properties nvlist and modify the list
598 * for the property values to be set.
601 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
604 int error
= 0, reset_bootfs
= 0;
606 boolean_t has_feature
= B_FALSE
;
609 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
611 const char *strval
, *slash
, *check
, *fname
;
612 const char *propname
= nvpair_name(elem
);
613 zpool_prop_t prop
= zpool_name_to_prop(propname
);
616 case ZPOOL_PROP_INVAL
:
618 * Sanitize the input.
620 if (zfs_prop_user(propname
)) {
621 if (strlen(propname
) >= ZAP_MAXNAMELEN
) {
622 error
= SET_ERROR(ENAMETOOLONG
);
626 if (strlen(fnvpair_value_string(elem
)) >=
628 error
= SET_ERROR(E2BIG
);
631 } else if (zpool_prop_feature(propname
)) {
632 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
633 error
= SET_ERROR(EINVAL
);
637 if (nvpair_value_uint64(elem
, &intval
) != 0) {
638 error
= SET_ERROR(EINVAL
);
643 error
= SET_ERROR(EINVAL
);
647 fname
= strchr(propname
, '@') + 1;
648 if (zfeature_lookup_name(fname
, NULL
) != 0) {
649 error
= SET_ERROR(EINVAL
);
653 has_feature
= B_TRUE
;
655 error
= SET_ERROR(EINVAL
);
660 case ZPOOL_PROP_VERSION
:
661 error
= nvpair_value_uint64(elem
, &intval
);
663 (intval
< spa_version(spa
) ||
664 intval
> SPA_VERSION_BEFORE_FEATURES
||
666 error
= SET_ERROR(EINVAL
);
669 case ZPOOL_PROP_DELEGATION
:
670 case ZPOOL_PROP_AUTOREPLACE
:
671 case ZPOOL_PROP_LISTSNAPS
:
672 case ZPOOL_PROP_AUTOEXPAND
:
673 case ZPOOL_PROP_AUTOTRIM
:
674 error
= nvpair_value_uint64(elem
, &intval
);
675 if (!error
&& intval
> 1)
676 error
= SET_ERROR(EINVAL
);
679 case ZPOOL_PROP_MULTIHOST
:
680 error
= nvpair_value_uint64(elem
, &intval
);
681 if (!error
&& intval
> 1)
682 error
= SET_ERROR(EINVAL
);
685 uint32_t hostid
= zone_get_hostid(NULL
);
687 spa
->spa_hostid
= hostid
;
689 error
= SET_ERROR(ENOTSUP
);
694 case ZPOOL_PROP_BOOTFS
:
696 * If the pool version is less than SPA_VERSION_BOOTFS,
697 * or the pool is still being created (version == 0),
698 * the bootfs property cannot be set.
700 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
701 error
= SET_ERROR(ENOTSUP
);
706 * Make sure the vdev config is bootable
708 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
709 error
= SET_ERROR(ENOTSUP
);
715 error
= nvpair_value_string(elem
, &strval
);
720 if (strval
== NULL
|| strval
[0] == '\0') {
721 objnum
= zpool_prop_default_numeric(
726 error
= dmu_objset_hold(strval
, FTAG
, &os
);
731 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
732 error
= SET_ERROR(ENOTSUP
);
734 objnum
= dmu_objset_id(os
);
736 dmu_objset_rele(os
, FTAG
);
740 case ZPOOL_PROP_FAILUREMODE
:
741 error
= nvpair_value_uint64(elem
, &intval
);
742 if (!error
&& intval
> ZIO_FAILURE_MODE_PANIC
)
743 error
= SET_ERROR(EINVAL
);
746 * This is a special case which only occurs when
747 * the pool has completely failed. This allows
748 * the user to change the in-core failmode property
749 * without syncing it out to disk (I/Os might
750 * currently be blocked). We do this by returning
751 * EIO to the caller (spa_prop_set) to trick it
752 * into thinking we encountered a property validation
755 if (!error
&& spa_suspended(spa
)) {
756 spa
->spa_failmode
= intval
;
757 error
= SET_ERROR(EIO
);
761 case ZPOOL_PROP_CACHEFILE
:
762 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
765 if (strval
[0] == '\0')
768 if (strcmp(strval
, "none") == 0)
771 if (strval
[0] != '/') {
772 error
= SET_ERROR(EINVAL
);
776 slash
= strrchr(strval
, '/');
777 ASSERT(slash
!= NULL
);
779 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
780 strcmp(slash
, "/..") == 0)
781 error
= SET_ERROR(EINVAL
);
784 case ZPOOL_PROP_COMMENT
:
785 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
787 for (check
= strval
; *check
!= '\0'; check
++) {
788 if (!isprint(*check
)) {
789 error
= SET_ERROR(EINVAL
);
793 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
794 error
= SET_ERROR(E2BIG
);
805 (void) nvlist_remove_all(props
,
806 zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO
));
808 if (!error
&& reset_bootfs
) {
809 error
= nvlist_remove(props
,
810 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
813 error
= nvlist_add_uint64(props
,
814 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
822 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
824 const char *cachefile
;
825 spa_config_dirent_t
*dp
;
827 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
831 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
834 if (cachefile
[0] == '\0')
835 dp
->scd_path
= spa_strdup(spa_config_path
);
836 else if (strcmp(cachefile
, "none") == 0)
839 dp
->scd_path
= spa_strdup(cachefile
);
841 list_insert_head(&spa
->spa_config_list
, dp
);
843 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
847 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
850 nvpair_t
*elem
= NULL
;
851 boolean_t need_sync
= B_FALSE
;
853 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
856 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
857 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
859 if (prop
== ZPOOL_PROP_CACHEFILE
||
860 prop
== ZPOOL_PROP_ALTROOT
||
861 prop
== ZPOOL_PROP_READONLY
)
864 if (prop
== ZPOOL_PROP_INVAL
&&
865 zfs_prop_user(nvpair_name(elem
))) {
870 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPOOL_PROP_INVAL
) {
873 if (prop
== ZPOOL_PROP_VERSION
) {
874 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
876 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
877 ver
= SPA_VERSION_FEATURES
;
881 /* Save time if the version is already set. */
882 if (ver
== spa_version(spa
))
886 * In addition to the pool directory object, we might
887 * create the pool properties object, the features for
888 * read object, the features for write object, or the
889 * feature descriptions object.
891 error
= dsl_sync_task(spa
->spa_name
, NULL
,
892 spa_sync_version
, &ver
,
893 6, ZFS_SPACE_CHECK_RESERVED
);
904 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
905 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
912 * If the bootfs property value is dsobj, clear it.
915 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
917 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
918 VERIFY(zap_remove(spa
->spa_meta_objset
,
919 spa
->spa_pool_props_object
,
920 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
926 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
928 uint64_t *newguid __maybe_unused
= arg
;
929 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
930 vdev_t
*rvd
= spa
->spa_root_vdev
;
933 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
934 int error
= (spa_has_checkpoint(spa
)) ?
935 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
936 return (SET_ERROR(error
));
939 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
940 vdev_state
= rvd
->vdev_state
;
941 spa_config_exit(spa
, SCL_STATE
, FTAG
);
943 if (vdev_state
!= VDEV_STATE_HEALTHY
)
944 return (SET_ERROR(ENXIO
));
946 ASSERT3U(spa_guid(spa
), !=, *newguid
);
952 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
954 uint64_t *newguid
= arg
;
955 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
957 vdev_t
*rvd
= spa
->spa_root_vdev
;
959 oldguid
= spa_guid(spa
);
961 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
962 rvd
->vdev_guid
= *newguid
;
963 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
964 vdev_config_dirty(rvd
);
965 spa_config_exit(spa
, SCL_STATE
, FTAG
);
967 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
968 (u_longlong_t
)oldguid
, (u_longlong_t
)*newguid
);
972 * Change the GUID for the pool. This is done so that we can later
973 * re-import a pool built from a clone of our own vdevs. We will modify
974 * the root vdev's guid, our own pool guid, and then mark all of our
975 * vdevs dirty. Note that we must make sure that all our vdevs are
976 * online when we do this, or else any vdevs that weren't present
977 * would be orphaned from our pool. We are also going to issue a
978 * sysevent to update any watchers.
981 spa_change_guid(spa_t
*spa
)
986 mutex_enter(&spa
->spa_vdev_top_lock
);
987 mutex_enter(&spa_namespace_lock
);
988 guid
= spa_generate_guid(NULL
);
990 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
991 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
995 * Clear the kobj flag from all the vdevs to allow
996 * vdev_cache_process_kobj_evt() to post events to all the
997 * vdevs since GUID is updated.
999 vdev_clear_kobj_evt(spa
->spa_root_vdev
);
1000 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
1001 vdev_clear_kobj_evt(spa
->spa_l2cache
.sav_vdevs
[i
]);
1003 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
, B_TRUE
);
1004 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
1007 mutex_exit(&spa_namespace_lock
);
1008 mutex_exit(&spa
->spa_vdev_top_lock
);
1014 * ==========================================================================
1015 * SPA state manipulation (open/create/destroy/import/export)
1016 * ==========================================================================
1020 spa_error_entry_compare(const void *a
, const void *b
)
1022 const spa_error_entry_t
*sa
= (const spa_error_entry_t
*)a
;
1023 const spa_error_entry_t
*sb
= (const spa_error_entry_t
*)b
;
1026 ret
= memcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
1027 sizeof (zbookmark_phys_t
));
1029 return (TREE_ISIGN(ret
));
1033 * Utility function which retrieves copies of the current logs and
1034 * re-initializes them in the process.
1037 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
1039 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
1041 memcpy(last
, &spa
->spa_errlist_last
, sizeof (avl_tree_t
));
1042 memcpy(scrub
, &spa
->spa_errlist_scrub
, sizeof (avl_tree_t
));
1044 avl_create(&spa
->spa_errlist_scrub
,
1045 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1046 offsetof(spa_error_entry_t
, se_avl
));
1047 avl_create(&spa
->spa_errlist_last
,
1048 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1049 offsetof(spa_error_entry_t
, se_avl
));
1053 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
1055 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
1056 enum zti_modes mode
= ztip
->zti_mode
;
1057 uint_t value
= ztip
->zti_value
;
1058 uint_t count
= ztip
->zti_count
;
1059 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1060 uint_t cpus
, flags
= TASKQ_DYNAMIC
;
1063 case ZTI_MODE_FIXED
:
1064 ASSERT3U(value
, >, 0);
1070 * Create one wr_iss taskq for every 'zio_taskq_wr_iss_ncpus',
1071 * not to exceed the number of spa allocators.
1073 if (zio_taskq_wr_iss_ncpus
== 0) {
1074 count
= MAX(boot_ncpus
/ spa
->spa_alloc_count
, 1);
1077 boot_ncpus
/ MAX(1, zio_taskq_wr_iss_ncpus
));
1079 count
= MAX(count
, (zio_taskq_batch_pct
+ 99) / 100);
1080 count
= MIN(count
, spa
->spa_alloc_count
);
1083 * zio_taskq_batch_pct is unbounded and may exceed 100%, but no
1084 * single taskq may have more threads than 100% of online cpus.
1086 value
= (zio_taskq_batch_pct
+ count
/ 2) / count
;
1087 value
= MIN(value
, 100);
1088 flags
|= TASKQ_THREADS_CPU_PCT
;
1091 case ZTI_MODE_SCALE
:
1092 flags
|= TASKQ_THREADS_CPU_PCT
;
1094 * We want more taskqs to reduce lock contention, but we want
1095 * less for better request ordering and CPU utilization.
1097 cpus
= MAX(1, boot_ncpus
* zio_taskq_batch_pct
/ 100);
1098 if (zio_taskq_batch_tpq
> 0) {
1099 count
= MAX(1, (cpus
+ zio_taskq_batch_tpq
/ 2) /
1100 zio_taskq_batch_tpq
);
1103 * Prefer 6 threads per taskq, but no more taskqs
1104 * than threads in them on large systems. For 80%:
1107 * cpus taskqs percent threads threads
1108 * ------- ------- ------- ------- -------
1119 count
= 1 + cpus
/ 6;
1120 while (count
* count
> cpus
)
1123 /* Limit each taskq within 100% to not trigger assertion. */
1124 count
= MAX(count
, (zio_taskq_batch_pct
+ 99) / 100);
1125 value
= (zio_taskq_batch_pct
+ count
/ 2) / count
;
1129 tqs
->stqs_count
= 0;
1130 tqs
->stqs_taskq
= NULL
;
1134 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
1135 "spa_taskqs_init()",
1136 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
1140 ASSERT3U(count
, >, 0);
1141 tqs
->stqs_count
= count
;
1142 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
1144 for (uint_t i
= 0; i
< count
; i
++) {
1149 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
1150 zio_type_name
[t
], zio_taskq_types
[q
], i
);
1152 (void) snprintf(name
, sizeof (name
), "%s_%s",
1153 zio_type_name
[t
], zio_taskq_types
[q
]);
1156 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
1157 (void) zio_taskq_basedc
;
1158 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
1159 spa
->spa_proc
, zio_taskq_basedc
, flags
);
1162 pri_t pri
= maxclsyspri
;
1164 * The write issue taskq can be extremely CPU
1165 * intensive. Run it at slightly less important
1166 * priority than the other taskqs.
1168 * Under Linux and FreeBSD this means incrementing
1169 * the priority value as opposed to platforms like
1170 * illumos where it should be decremented.
1172 * On FreeBSD, if priorities divided by four (RQ_PPQ)
1173 * are equal then a difference between them is
1176 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
) {
1177 #if defined(__linux__)
1179 #elif defined(__FreeBSD__)
1185 tq
= taskq_create_proc(name
, value
, pri
, 50,
1186 INT_MAX
, spa
->spa_proc
, flags
);
1191 tqs
->stqs_taskq
[i
] = tq
;
1196 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
1198 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1200 if (tqs
->stqs_taskq
== NULL
) {
1201 ASSERT3U(tqs
->stqs_count
, ==, 0);
1205 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
1206 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
1207 taskq_destroy(tqs
->stqs_taskq
[i
]);
1210 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
1211 tqs
->stqs_taskq
= NULL
;
1215 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1216 * Note that a type may have multiple discrete taskqs to avoid lock contention
1217 * on the taskq itself.
1220 spa_taskq_dispatch_select(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1223 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1226 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1227 ASSERT3U(tqs
->stqs_count
, !=, 0);
1229 if ((t
== ZIO_TYPE_WRITE
) && (q
== ZIO_TASKQ_ISSUE
) &&
1230 (zio
!= NULL
) && (zio
->io_wr_iss_tq
!= NULL
)) {
1231 /* dispatch to assigned write issue taskq */
1232 tq
= zio
->io_wr_iss_tq
;
1236 if (tqs
->stqs_count
== 1) {
1237 tq
= tqs
->stqs_taskq
[0];
1239 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1245 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1246 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
,
1249 taskq_t
*tq
= spa_taskq_dispatch_select(spa
, t
, q
, zio
);
1250 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
1254 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1257 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1258 task_func_t
*func
, void *arg
, uint_t flags
)
1260 taskq_t
*tq
= spa_taskq_dispatch_select(spa
, t
, q
, NULL
);
1261 taskqid_t id
= taskq_dispatch(tq
, func
, arg
, flags
);
1263 taskq_wait_id(tq
, id
);
1267 spa_create_zio_taskqs(spa_t
*spa
)
1269 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1270 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1271 spa_taskqs_init(spa
, t
, q
);
1276 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1278 spa_thread(void *arg
)
1280 psetid_t zio_taskq_psrset_bind
= PS_NONE
;
1281 callb_cpr_t cprinfo
;
1284 user_t
*pu
= PTOU(curproc
);
1286 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1289 ASSERT(curproc
!= &p0
);
1290 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1291 "zpool-%s", spa
->spa_name
);
1292 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1294 /* bind this thread to the requested psrset */
1295 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1297 mutex_enter(&cpu_lock
);
1298 mutex_enter(&pidlock
);
1299 mutex_enter(&curproc
->p_lock
);
1301 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1302 0, NULL
, NULL
) == 0) {
1303 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1306 "Couldn't bind process for zfs pool \"%s\" to "
1307 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1310 mutex_exit(&curproc
->p_lock
);
1311 mutex_exit(&pidlock
);
1312 mutex_exit(&cpu_lock
);
1317 if (zio_taskq_sysdc
) {
1318 sysdc_thread_enter(curthread
, 100, 0);
1322 spa
->spa_proc
= curproc
;
1323 spa
->spa_did
= curthread
->t_did
;
1325 spa_create_zio_taskqs(spa
);
1327 mutex_enter(&spa
->spa_proc_lock
);
1328 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1330 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1331 cv_broadcast(&spa
->spa_proc_cv
);
1333 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1334 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1335 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1336 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1338 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1339 spa
->spa_proc_state
= SPA_PROC_GONE
;
1340 spa
->spa_proc
= &p0
;
1341 cv_broadcast(&spa
->spa_proc_cv
);
1342 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1344 mutex_enter(&curproc
->p_lock
);
1349 extern metaslab_ops_t
*metaslab_allocator(spa_t
*spa
);
1352 * Activate an uninitialized pool.
1355 spa_activate(spa_t
*spa
, spa_mode_t mode
)
1357 metaslab_ops_t
*msp
= metaslab_allocator(spa
);
1358 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1360 spa
->spa_state
= POOL_STATE_ACTIVE
;
1361 spa
->spa_mode
= mode
;
1362 spa
->spa_read_spacemaps
= spa_mode_readable_spacemaps
;
1364 spa
->spa_normal_class
= metaslab_class_create(spa
, msp
);
1365 spa
->spa_log_class
= metaslab_class_create(spa
, msp
);
1366 spa
->spa_embedded_log_class
= metaslab_class_create(spa
, msp
);
1367 spa
->spa_special_class
= metaslab_class_create(spa
, msp
);
1368 spa
->spa_dedup_class
= metaslab_class_create(spa
, msp
);
1370 /* Try to create a covering process */
1371 mutex_enter(&spa
->spa_proc_lock
);
1372 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1373 ASSERT(spa
->spa_proc
== &p0
);
1376 #ifdef HAVE_SPA_THREAD
1377 /* Only create a process if we're going to be around a while. */
1378 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1379 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1381 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1382 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1383 cv_wait(&spa
->spa_proc_cv
,
1384 &spa
->spa_proc_lock
);
1386 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1387 ASSERT(spa
->spa_proc
!= &p0
);
1388 ASSERT(spa
->spa_did
!= 0);
1392 "Couldn't create process for zfs pool \"%s\"\n",
1397 #endif /* HAVE_SPA_THREAD */
1398 mutex_exit(&spa
->spa_proc_lock
);
1400 /* If we didn't create a process, we need to create our taskqs. */
1401 if (spa
->spa_proc
== &p0
) {
1402 spa_create_zio_taskqs(spa
);
1405 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1406 spa
->spa_txg_zio
[i
] = zio_root(spa
, NULL
, NULL
,
1410 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1411 offsetof(vdev_t
, vdev_config_dirty_node
));
1412 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1413 offsetof(objset_t
, os_evicting_node
));
1414 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1415 offsetof(vdev_t
, vdev_state_dirty_node
));
1417 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1418 offsetof(struct vdev
, vdev_txg_node
));
1420 avl_create(&spa
->spa_errlist_scrub
,
1421 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1422 offsetof(spa_error_entry_t
, se_avl
));
1423 avl_create(&spa
->spa_errlist_last
,
1424 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1425 offsetof(spa_error_entry_t
, se_avl
));
1426 avl_create(&spa
->spa_errlist_healed
,
1427 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1428 offsetof(spa_error_entry_t
, se_avl
));
1430 spa_activate_os(spa
);
1432 spa_keystore_init(&spa
->spa_keystore
);
1435 * This taskq is used to perform zvol-minor-related tasks
1436 * asynchronously. This has several advantages, including easy
1437 * resolution of various deadlocks.
1439 * The taskq must be single threaded to ensure tasks are always
1440 * processed in the order in which they were dispatched.
1442 * A taskq per pool allows one to keep the pools independent.
1443 * This way if one pool is suspended, it will not impact another.
1445 * The preferred location to dispatch a zvol minor task is a sync
1446 * task. In this context, there is easy access to the spa_t and minimal
1447 * error handling is required because the sync task must succeed.
1449 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1453 * The taskq to preload metaslabs.
1455 spa
->spa_metaslab_taskq
= taskq_create("z_metaslab",
1456 metaslab_preload_pct
, maxclsyspri
, 1, INT_MAX
,
1457 TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1460 * Taskq dedicated to prefetcher threads: this is used to prevent the
1461 * pool traverse code from monopolizing the global (and limited)
1462 * system_taskq by inappropriately scheduling long running tasks on it.
1464 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", 100,
1465 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1468 * The taskq to upgrade datasets in this pool. Currently used by
1469 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1471 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", 100,
1472 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1476 * Opposite of spa_activate().
1479 spa_deactivate(spa_t
*spa
)
1481 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1482 ASSERT(spa
->spa_dsl_pool
== NULL
);
1483 ASSERT(spa
->spa_root_vdev
== NULL
);
1484 ASSERT(spa
->spa_async_zio_root
== NULL
);
1485 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1487 spa_evicting_os_wait(spa
);
1489 if (spa
->spa_zvol_taskq
) {
1490 taskq_destroy(spa
->spa_zvol_taskq
);
1491 spa
->spa_zvol_taskq
= NULL
;
1494 if (spa
->spa_metaslab_taskq
) {
1495 taskq_destroy(spa
->spa_metaslab_taskq
);
1496 spa
->spa_metaslab_taskq
= NULL
;
1499 if (spa
->spa_prefetch_taskq
) {
1500 taskq_destroy(spa
->spa_prefetch_taskq
);
1501 spa
->spa_prefetch_taskq
= NULL
;
1504 if (spa
->spa_upgrade_taskq
) {
1505 taskq_destroy(spa
->spa_upgrade_taskq
);
1506 spa
->spa_upgrade_taskq
= NULL
;
1509 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1511 list_destroy(&spa
->spa_config_dirty_list
);
1512 list_destroy(&spa
->spa_evicting_os_list
);
1513 list_destroy(&spa
->spa_state_dirty_list
);
1515 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1517 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1518 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1519 spa_taskqs_fini(spa
, t
, q
);
1523 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1524 ASSERT3P(spa
->spa_txg_zio
[i
], !=, NULL
);
1525 VERIFY0(zio_wait(spa
->spa_txg_zio
[i
]));
1526 spa
->spa_txg_zio
[i
] = NULL
;
1529 metaslab_class_destroy(spa
->spa_normal_class
);
1530 spa
->spa_normal_class
= NULL
;
1532 metaslab_class_destroy(spa
->spa_log_class
);
1533 spa
->spa_log_class
= NULL
;
1535 metaslab_class_destroy(spa
->spa_embedded_log_class
);
1536 spa
->spa_embedded_log_class
= NULL
;
1538 metaslab_class_destroy(spa
->spa_special_class
);
1539 spa
->spa_special_class
= NULL
;
1541 metaslab_class_destroy(spa
->spa_dedup_class
);
1542 spa
->spa_dedup_class
= NULL
;
1545 * If this was part of an import or the open otherwise failed, we may
1546 * still have errors left in the queues. Empty them just in case.
1548 spa_errlog_drain(spa
);
1549 avl_destroy(&spa
->spa_errlist_scrub
);
1550 avl_destroy(&spa
->spa_errlist_last
);
1551 avl_destroy(&spa
->spa_errlist_healed
);
1553 spa_keystore_fini(&spa
->spa_keystore
);
1555 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1557 mutex_enter(&spa
->spa_proc_lock
);
1558 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1559 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1560 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1561 cv_broadcast(&spa
->spa_proc_cv
);
1562 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1563 ASSERT(spa
->spa_proc
!= &p0
);
1564 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1566 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1567 spa
->spa_proc_state
= SPA_PROC_NONE
;
1569 ASSERT(spa
->spa_proc
== &p0
);
1570 mutex_exit(&spa
->spa_proc_lock
);
1573 * We want to make sure spa_thread() has actually exited the ZFS
1574 * module, so that the module can't be unloaded out from underneath
1577 if (spa
->spa_did
!= 0) {
1578 thread_join(spa
->spa_did
);
1582 spa_deactivate_os(spa
);
1587 * Verify a pool configuration, and construct the vdev tree appropriately. This
1588 * will create all the necessary vdevs in the appropriate layout, with each vdev
1589 * in the CLOSED state. This will prep the pool before open/creation/import.
1590 * All vdev validation is done by the vdev_alloc() routine.
1593 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1594 uint_t id
, int atype
)
1600 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1603 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1606 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1609 if (error
== ENOENT
)
1615 return (SET_ERROR(EINVAL
));
1618 for (int c
= 0; c
< children
; c
++) {
1620 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1628 ASSERT(*vdp
!= NULL
);
1634 spa_should_flush_logs_on_unload(spa_t
*spa
)
1636 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
1639 if (!spa_writeable(spa
))
1642 if (!spa
->spa_sync_on
)
1645 if (spa_state(spa
) != POOL_STATE_EXPORTED
)
1648 if (zfs_keep_log_spacemaps_at_export
)
1655 * Opens a transaction that will set the flag that will instruct
1656 * spa_sync to attempt to flush all the metaslabs for that txg.
1659 spa_unload_log_sm_flush_all(spa_t
*spa
)
1661 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
1662 VERIFY0(dmu_tx_assign(tx
, TXG_WAIT
));
1664 ASSERT3U(spa
->spa_log_flushall_txg
, ==, 0);
1665 spa
->spa_log_flushall_txg
= dmu_tx_get_txg(tx
);
1668 txg_wait_synced(spa_get_dsl(spa
), spa
->spa_log_flushall_txg
);
1672 spa_unload_log_sm_metadata(spa_t
*spa
)
1674 void *cookie
= NULL
;
1676 log_summary_entry_t
*e
;
1678 while ((sls
= avl_destroy_nodes(&spa
->spa_sm_logs_by_txg
,
1679 &cookie
)) != NULL
) {
1680 VERIFY0(sls
->sls_mscount
);
1681 kmem_free(sls
, sizeof (spa_log_sm_t
));
1684 while ((e
= list_remove_head(&spa
->spa_log_summary
)) != NULL
) {
1685 VERIFY0(e
->lse_mscount
);
1686 kmem_free(e
, sizeof (log_summary_entry_t
));
1689 spa
->spa_unflushed_stats
.sus_nblocks
= 0;
1690 spa
->spa_unflushed_stats
.sus_memused
= 0;
1691 spa
->spa_unflushed_stats
.sus_blocklimit
= 0;
1695 spa_destroy_aux_threads(spa_t
*spa
)
1697 if (spa
->spa_condense_zthr
!= NULL
) {
1698 zthr_destroy(spa
->spa_condense_zthr
);
1699 spa
->spa_condense_zthr
= NULL
;
1701 if (spa
->spa_checkpoint_discard_zthr
!= NULL
) {
1702 zthr_destroy(spa
->spa_checkpoint_discard_zthr
);
1703 spa
->spa_checkpoint_discard_zthr
= NULL
;
1705 if (spa
->spa_livelist_delete_zthr
!= NULL
) {
1706 zthr_destroy(spa
->spa_livelist_delete_zthr
);
1707 spa
->spa_livelist_delete_zthr
= NULL
;
1709 if (spa
->spa_livelist_condense_zthr
!= NULL
) {
1710 zthr_destroy(spa
->spa_livelist_condense_zthr
);
1711 spa
->spa_livelist_condense_zthr
= NULL
;
1713 if (spa
->spa_raidz_expand_zthr
!= NULL
) {
1714 zthr_destroy(spa
->spa_raidz_expand_zthr
);
1715 spa
->spa_raidz_expand_zthr
= NULL
;
1720 * Opposite of spa_load().
1723 spa_unload(spa_t
*spa
)
1725 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1726 ASSERT(spa_state(spa
) != POOL_STATE_UNINITIALIZED
);
1728 spa_import_progress_remove(spa_guid(spa
));
1729 spa_load_note(spa
, "UNLOADING");
1731 spa_wake_waiters(spa
);
1734 * If we have set the spa_final_txg, we have already performed the
1735 * tasks below in spa_export_common(). We should not redo it here since
1736 * we delay the final TXGs beyond what spa_final_txg is set at.
1738 if (spa
->spa_final_txg
== UINT64_MAX
) {
1740 * If the log space map feature is enabled and the pool is
1741 * getting exported (but not destroyed), we want to spend some
1742 * time flushing as many metaslabs as we can in an attempt to
1743 * destroy log space maps and save import time.
1745 if (spa_should_flush_logs_on_unload(spa
))
1746 spa_unload_log_sm_flush_all(spa
);
1751 spa_async_suspend(spa
);
1753 if (spa
->spa_root_vdev
) {
1754 vdev_t
*root_vdev
= spa
->spa_root_vdev
;
1755 vdev_initialize_stop_all(root_vdev
,
1756 VDEV_INITIALIZE_ACTIVE
);
1757 vdev_trim_stop_all(root_vdev
, VDEV_TRIM_ACTIVE
);
1758 vdev_autotrim_stop_all(spa
);
1759 vdev_rebuild_stop_all(spa
);
1766 if (spa
->spa_sync_on
) {
1767 txg_sync_stop(spa
->spa_dsl_pool
);
1768 spa
->spa_sync_on
= B_FALSE
;
1772 * This ensures that there is no async metaslab prefetching
1773 * while we attempt to unload the spa.
1775 taskq_wait(spa
->spa_metaslab_taskq
);
1777 if (spa
->spa_mmp
.mmp_thread
)
1778 mmp_thread_stop(spa
);
1781 * Wait for any outstanding async I/O to complete.
1783 if (spa
->spa_async_zio_root
!= NULL
) {
1784 for (int i
= 0; i
< max_ncpus
; i
++)
1785 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1786 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1787 spa
->spa_async_zio_root
= NULL
;
1790 if (spa
->spa_vdev_removal
!= NULL
) {
1791 spa_vdev_removal_destroy(spa
->spa_vdev_removal
);
1792 spa
->spa_vdev_removal
= NULL
;
1795 spa_destroy_aux_threads(spa
);
1797 spa_condense_fini(spa
);
1799 bpobj_close(&spa
->spa_deferred_bpobj
);
1801 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1806 if (spa
->spa_root_vdev
)
1807 vdev_free(spa
->spa_root_vdev
);
1808 ASSERT(spa
->spa_root_vdev
== NULL
);
1811 * Close the dsl pool.
1813 if (spa
->spa_dsl_pool
) {
1814 dsl_pool_close(spa
->spa_dsl_pool
);
1815 spa
->spa_dsl_pool
= NULL
;
1816 spa
->spa_meta_objset
= NULL
;
1821 spa_unload_log_sm_metadata(spa
);
1824 * Drop and purge level 2 cache
1826 spa_l2cache_drop(spa
);
1828 if (spa
->spa_spares
.sav_vdevs
) {
1829 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1830 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1831 kmem_free(spa
->spa_spares
.sav_vdevs
,
1832 spa
->spa_spares
.sav_count
* sizeof (void *));
1833 spa
->spa_spares
.sav_vdevs
= NULL
;
1835 if (spa
->spa_spares
.sav_config
) {
1836 nvlist_free(spa
->spa_spares
.sav_config
);
1837 spa
->spa_spares
.sav_config
= NULL
;
1839 spa
->spa_spares
.sav_count
= 0;
1841 if (spa
->spa_l2cache
.sav_vdevs
) {
1842 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1843 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1844 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1846 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1847 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1848 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1850 if (spa
->spa_l2cache
.sav_config
) {
1851 nvlist_free(spa
->spa_l2cache
.sav_config
);
1852 spa
->spa_l2cache
.sav_config
= NULL
;
1854 spa
->spa_l2cache
.sav_count
= 0;
1856 spa
->spa_async_suspended
= 0;
1858 spa
->spa_indirect_vdevs_loaded
= B_FALSE
;
1860 if (spa
->spa_comment
!= NULL
) {
1861 spa_strfree(spa
->spa_comment
);
1862 spa
->spa_comment
= NULL
;
1864 if (spa
->spa_compatibility
!= NULL
) {
1865 spa_strfree(spa
->spa_compatibility
);
1866 spa
->spa_compatibility
= NULL
;
1869 spa
->spa_raidz_expand
= NULL
;
1871 spa_config_exit(spa
, SCL_ALL
, spa
);
1875 * Load (or re-load) the current list of vdevs describing the active spares for
1876 * this pool. When this is called, we have some form of basic information in
1877 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1878 * then re-generate a more complete list including status information.
1881 spa_load_spares(spa_t
*spa
)
1890 * zdb opens both the current state of the pool and the
1891 * checkpointed state (if present), with a different spa_t.
1893 * As spare vdevs are shared among open pools, we skip loading
1894 * them when we load the checkpointed state of the pool.
1896 if (!spa_writeable(spa
))
1900 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1903 * First, close and free any existing spare vdevs.
1905 if (spa
->spa_spares
.sav_vdevs
) {
1906 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1907 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1909 /* Undo the call to spa_activate() below */
1910 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1911 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1912 spa_spare_remove(tvd
);
1917 kmem_free(spa
->spa_spares
.sav_vdevs
,
1918 spa
->spa_spares
.sav_count
* sizeof (void *));
1921 if (spa
->spa_spares
.sav_config
== NULL
)
1924 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1925 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
));
1927 spa
->spa_spares
.sav_count
= (int)nspares
;
1928 spa
->spa_spares
.sav_vdevs
= NULL
;
1934 * Construct the array of vdevs, opening them to get status in the
1935 * process. For each spare, there is potentially two different vdev_t
1936 * structures associated with it: one in the list of spares (used only
1937 * for basic validation purposes) and one in the active vdev
1938 * configuration (if it's spared in). During this phase we open and
1939 * validate each vdev on the spare list. If the vdev also exists in the
1940 * active configuration, then we also mark this vdev as an active spare.
1942 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1944 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1945 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1946 VDEV_ALLOC_SPARE
) == 0);
1949 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1951 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1952 B_FALSE
)) != NULL
) {
1953 if (!tvd
->vdev_isspare
)
1957 * We only mark the spare active if we were successfully
1958 * able to load the vdev. Otherwise, importing a pool
1959 * with a bad active spare would result in strange
1960 * behavior, because multiple pool would think the spare
1961 * is actively in use.
1963 * There is a vulnerability here to an equally bizarre
1964 * circumstance, where a dead active spare is later
1965 * brought back to life (onlined or otherwise). Given
1966 * the rarity of this scenario, and the extra complexity
1967 * it adds, we ignore the possibility.
1969 if (!vdev_is_dead(tvd
))
1970 spa_spare_activate(tvd
);
1974 vd
->vdev_aux
= &spa
->spa_spares
;
1976 if (vdev_open(vd
) != 0)
1979 if (vdev_validate_aux(vd
) == 0)
1984 * Recompute the stashed list of spares, with status information
1987 fnvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
);
1989 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1991 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1992 spares
[i
] = vdev_config_generate(spa
,
1993 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1994 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1995 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
1996 spa
->spa_spares
.sav_count
);
1997 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1998 nvlist_free(spares
[i
]);
1999 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
2003 * Load (or re-load) the current list of vdevs describing the active l2cache for
2004 * this pool. When this is called, we have some form of basic information in
2005 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
2006 * then re-generate a more complete list including status information.
2007 * Devices which are already active have their details maintained, and are
2011 spa_load_l2cache(spa_t
*spa
)
2013 nvlist_t
**l2cache
= NULL
;
2015 int i
, j
, oldnvdevs
;
2017 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
2018 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
2022 * zdb opens both the current state of the pool and the
2023 * checkpointed state (if present), with a different spa_t.
2025 * As L2 caches are part of the ARC which is shared among open
2026 * pools, we skip loading them when we load the checkpointed
2027 * state of the pool.
2029 if (!spa_writeable(spa
))
2033 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
2035 oldvdevs
= sav
->sav_vdevs
;
2036 oldnvdevs
= sav
->sav_count
;
2037 sav
->sav_vdevs
= NULL
;
2040 if (sav
->sav_config
== NULL
) {
2046 VERIFY0(nvlist_lookup_nvlist_array(sav
->sav_config
,
2047 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
));
2048 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
2051 * Process new nvlist of vdevs.
2053 for (i
= 0; i
< nl2cache
; i
++) {
2054 guid
= fnvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
);
2057 for (j
= 0; j
< oldnvdevs
; j
++) {
2059 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
2061 * Retain previous vdev for add/remove ops.
2069 if (newvdevs
[i
] == NULL
) {
2073 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
2074 VDEV_ALLOC_L2CACHE
) == 0);
2079 * Commit this vdev as an l2cache device,
2080 * even if it fails to open.
2082 spa_l2cache_add(vd
);
2087 spa_l2cache_activate(vd
);
2089 if (vdev_open(vd
) != 0)
2092 (void) vdev_validate_aux(vd
);
2094 if (!vdev_is_dead(vd
))
2095 l2arc_add_vdev(spa
, vd
);
2098 * Upon cache device addition to a pool or pool
2099 * creation with a cache device or if the header
2100 * of the device is invalid we issue an async
2101 * TRIM command for the whole device which will
2102 * execute if l2arc_trim_ahead > 0.
2104 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
2108 sav
->sav_vdevs
= newvdevs
;
2109 sav
->sav_count
= (int)nl2cache
;
2112 * Recompute the stashed list of l2cache devices, with status
2113 * information this time.
2115 fnvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
);
2117 if (sav
->sav_count
> 0)
2118 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *),
2120 for (i
= 0; i
< sav
->sav_count
; i
++)
2121 l2cache
[i
] = vdev_config_generate(spa
,
2122 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
2123 fnvlist_add_nvlist_array(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
2124 (const nvlist_t
* const *)l2cache
, sav
->sav_count
);
2128 * Purge vdevs that were dropped
2131 for (i
= 0; i
< oldnvdevs
; i
++) {
2136 ASSERT(vd
->vdev_isl2cache
);
2138 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
2139 pool
!= 0ULL && l2arc_vdev_present(vd
))
2140 l2arc_remove_vdev(vd
);
2141 vdev_clear_stats(vd
);
2146 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
2149 for (i
= 0; i
< sav
->sav_count
; i
++)
2150 nvlist_free(l2cache
[i
]);
2152 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
2156 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
2159 char *packed
= NULL
;
2164 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
2168 nvsize
= *(uint64_t *)db
->db_data
;
2169 dmu_buf_rele(db
, FTAG
);
2171 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
2172 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
2175 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
2176 vmem_free(packed
, nvsize
);
2182 * Concrete top-level vdevs that are not missing and are not logs. At every
2183 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
2186 spa_healthy_core_tvds(spa_t
*spa
)
2188 vdev_t
*rvd
= spa
->spa_root_vdev
;
2191 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
2192 vdev_t
*vd
= rvd
->vdev_child
[i
];
2195 if (vdev_is_concrete(vd
) && !vdev_is_dead(vd
))
2203 * Checks to see if the given vdev could not be opened, in which case we post a
2204 * sysevent to notify the autoreplace code that the device has been removed.
2207 spa_check_removed(vdev_t
*vd
)
2209 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
2210 spa_check_removed(vd
->vdev_child
[c
]);
2212 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
2213 vdev_is_concrete(vd
)) {
2214 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
2215 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
2220 spa_check_for_missing_logs(spa_t
*spa
)
2222 vdev_t
*rvd
= spa
->spa_root_vdev
;
2225 * If we're doing a normal import, then build up any additional
2226 * diagnostic information about missing log devices.
2227 * We'll pass this up to the user for further processing.
2229 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
2230 nvlist_t
**child
, *nv
;
2233 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
2235 nv
= fnvlist_alloc();
2237 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2238 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2241 * We consider a device as missing only if it failed
2242 * to open (i.e. offline or faulted is not considered
2245 if (tvd
->vdev_islog
&&
2246 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2247 child
[idx
++] = vdev_config_generate(spa
, tvd
,
2248 B_FALSE
, VDEV_CONFIG_MISSING
);
2253 fnvlist_add_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
2254 (const nvlist_t
* const *)child
, idx
);
2255 fnvlist_add_nvlist(spa
->spa_load_info
,
2256 ZPOOL_CONFIG_MISSING_DEVICES
, nv
);
2258 for (uint64_t i
= 0; i
< idx
; i
++)
2259 nvlist_free(child
[i
]);
2262 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
2265 spa_load_failed(spa
, "some log devices are missing");
2266 vdev_dbgmsg_print_tree(rvd
, 2);
2267 return (SET_ERROR(ENXIO
));
2270 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2271 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2273 if (tvd
->vdev_islog
&&
2274 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2275 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2276 spa_load_note(spa
, "some log devices are "
2277 "missing, ZIL is dropped.");
2278 vdev_dbgmsg_print_tree(rvd
, 2);
2288 * Check for missing log devices
2291 spa_check_logs(spa_t
*spa
)
2293 boolean_t rv
= B_FALSE
;
2294 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2296 switch (spa
->spa_log_state
) {
2299 case SPA_LOG_MISSING
:
2300 /* need to recheck in case slog has been restored */
2301 case SPA_LOG_UNKNOWN
:
2302 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
2303 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
2305 spa_set_log_state(spa
, SPA_LOG_MISSING
);
2312 * Passivate any log vdevs (note, does not apply to embedded log metaslabs).
2315 spa_passivate_log(spa_t
*spa
)
2317 vdev_t
*rvd
= spa
->spa_root_vdev
;
2318 boolean_t slog_found
= B_FALSE
;
2320 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2322 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2323 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2325 if (tvd
->vdev_islog
) {
2326 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
2327 metaslab_group_passivate(tvd
->vdev_mg
);
2328 slog_found
= B_TRUE
;
2332 return (slog_found
);
2336 * Activate any log vdevs (note, does not apply to embedded log metaslabs).
2339 spa_activate_log(spa_t
*spa
)
2341 vdev_t
*rvd
= spa
->spa_root_vdev
;
2343 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2345 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2346 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2348 if (tvd
->vdev_islog
) {
2349 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
2350 metaslab_group_activate(tvd
->vdev_mg
);
2356 spa_reset_logs(spa_t
*spa
)
2360 error
= dmu_objset_find(spa_name(spa
), zil_reset
,
2361 NULL
, DS_FIND_CHILDREN
);
2364 * We successfully offlined the log device, sync out the
2365 * current txg so that the "stubby" block can be removed
2368 txg_wait_synced(spa
->spa_dsl_pool
, 0);
2374 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
2376 for (int i
= 0; i
< sav
->sav_count
; i
++)
2377 spa_check_removed(sav
->sav_vdevs
[i
]);
2381 spa_claim_notify(zio_t
*zio
)
2383 spa_t
*spa
= zio
->io_spa
;
2388 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
2389 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
2390 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
2391 mutex_exit(&spa
->spa_props_lock
);
2394 typedef struct spa_load_error
{
2395 boolean_t sle_verify_data
;
2396 uint64_t sle_meta_count
;
2397 uint64_t sle_data_count
;
2401 spa_load_verify_done(zio_t
*zio
)
2403 blkptr_t
*bp
= zio
->io_bp
;
2404 spa_load_error_t
*sle
= zio
->io_private
;
2405 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2406 int error
= zio
->io_error
;
2407 spa_t
*spa
= zio
->io_spa
;
2409 abd_free(zio
->io_abd
);
2411 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2412 type
!= DMU_OT_INTENT_LOG
)
2413 atomic_inc_64(&sle
->sle_meta_count
);
2415 atomic_inc_64(&sle
->sle_data_count
);
2418 mutex_enter(&spa
->spa_scrub_lock
);
2419 spa
->spa_load_verify_bytes
-= BP_GET_PSIZE(bp
);
2420 cv_broadcast(&spa
->spa_scrub_io_cv
);
2421 mutex_exit(&spa
->spa_scrub_lock
);
2425 * Maximum number of inflight bytes is the log2 fraction of the arc size.
2426 * By default, we set it to 1/16th of the arc.
2428 static uint_t spa_load_verify_shift
= 4;
2429 static int spa_load_verify_metadata
= B_TRUE
;
2430 static int spa_load_verify_data
= B_TRUE
;
2433 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2434 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2437 spa_load_error_t
*sle
= rio
->io_private
;
2439 (void) zilog
, (void) dnp
;
2442 * Note: normally this routine will not be called if
2443 * spa_load_verify_metadata is not set. However, it may be useful
2444 * to manually set the flag after the traversal has begun.
2446 if (!spa_load_verify_metadata
)
2450 * Sanity check the block pointer in order to detect obvious damage
2451 * before using the contents in subsequent checks or in zio_read().
2452 * When damaged consider it to be a metadata error since we cannot
2453 * trust the BP_GET_TYPE and BP_GET_LEVEL values.
2455 if (!zfs_blkptr_verify(spa
, bp
, BLK_CONFIG_NEEDED
, BLK_VERIFY_LOG
)) {
2456 atomic_inc_64(&sle
->sle_meta_count
);
2460 if (zb
->zb_level
== ZB_DNODE_LEVEL
|| BP_IS_HOLE(bp
) ||
2461 BP_IS_EMBEDDED(bp
) || BP_IS_REDACTED(bp
))
2464 if (!BP_IS_METADATA(bp
) &&
2465 (!spa_load_verify_data
|| !sle
->sle_verify_data
))
2468 uint64_t maxinflight_bytes
=
2469 arc_target_bytes() >> spa_load_verify_shift
;
2470 size_t size
= BP_GET_PSIZE(bp
);
2472 mutex_enter(&spa
->spa_scrub_lock
);
2473 while (spa
->spa_load_verify_bytes
>= maxinflight_bytes
)
2474 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2475 spa
->spa_load_verify_bytes
+= size
;
2476 mutex_exit(&spa
->spa_scrub_lock
);
2478 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2479 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2480 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2481 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2486 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2488 (void) dp
, (void) arg
;
2490 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2491 return (SET_ERROR(ENAMETOOLONG
));
2497 spa_load_verify(spa_t
*spa
)
2500 spa_load_error_t sle
= { 0 };
2501 zpool_load_policy_t policy
;
2502 boolean_t verify_ok
= B_FALSE
;
2505 zpool_get_load_policy(spa
->spa_config
, &policy
);
2507 if (policy
.zlp_rewind
& ZPOOL_NEVER_REWIND
||
2508 policy
.zlp_maxmeta
== UINT64_MAX
)
2511 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2512 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2513 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2515 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2520 * Verify data only if we are rewinding or error limit was set.
2521 * Otherwise nothing except dbgmsg care about it to waste time.
2523 sle
.sle_verify_data
= (policy
.zlp_rewind
& ZPOOL_REWIND_MASK
) ||
2524 (policy
.zlp_maxdata
< UINT64_MAX
);
2526 rio
= zio_root(spa
, NULL
, &sle
,
2527 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2529 if (spa_load_verify_metadata
) {
2530 if (spa
->spa_extreme_rewind
) {
2531 spa_load_note(spa
, "performing a complete scan of the "
2532 "pool since extreme rewind is on. This may take "
2533 "a very long time.\n (spa_load_verify_data=%u, "
2534 "spa_load_verify_metadata=%u)",
2535 spa_load_verify_data
, spa_load_verify_metadata
);
2538 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2539 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2540 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2543 (void) zio_wait(rio
);
2544 ASSERT0(spa
->spa_load_verify_bytes
);
2546 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2547 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2549 if (sle
.sle_meta_count
!= 0 || sle
.sle_data_count
!= 0) {
2550 spa_load_note(spa
, "spa_load_verify found %llu metadata errors "
2551 "and %llu data errors", (u_longlong_t
)sle
.sle_meta_count
,
2552 (u_longlong_t
)sle
.sle_data_count
);
2555 if (spa_load_verify_dryrun
||
2556 (!error
&& sle
.sle_meta_count
<= policy
.zlp_maxmeta
&&
2557 sle
.sle_data_count
<= policy
.zlp_maxdata
)) {
2561 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2562 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2564 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2565 fnvlist_add_uint64(spa
->spa_load_info
, ZPOOL_CONFIG_LOAD_TIME
,
2566 spa
->spa_load_txg_ts
);
2567 fnvlist_add_int64(spa
->spa_load_info
, ZPOOL_CONFIG_REWIND_TIME
,
2569 fnvlist_add_uint64(spa
->spa_load_info
,
2570 ZPOOL_CONFIG_LOAD_META_ERRORS
, sle
.sle_meta_count
);
2571 fnvlist_add_uint64(spa
->spa_load_info
,
2572 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
);
2574 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2577 if (spa_load_verify_dryrun
)
2581 if (error
!= ENXIO
&& error
!= EIO
)
2582 error
= SET_ERROR(EIO
);
2586 return (verify_ok
? 0 : EIO
);
2590 * Find a value in the pool props object.
2593 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2595 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2596 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2600 * Find a value in the pool directory object.
2603 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
, boolean_t log_enoent
)
2605 int error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2606 name
, sizeof (uint64_t), 1, val
);
2608 if (error
!= 0 && (error
!= ENOENT
|| log_enoent
)) {
2609 spa_load_failed(spa
, "couldn't get '%s' value in MOS directory "
2610 "[error=%d]", name
, error
);
2617 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2619 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2620 return (SET_ERROR(err
));
2624 spa_livelist_delete_check(spa_t
*spa
)
2626 return (spa
->spa_livelists_to_delete
!= 0);
2630 spa_livelist_delete_cb_check(void *arg
, zthr_t
*z
)
2634 return (spa_livelist_delete_check(spa
));
2638 delete_blkptr_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
2641 zio_free(spa
, tx
->tx_txg
, bp
);
2642 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
, DD_USED_HEAD
,
2643 -bp_get_dsize_sync(spa
, bp
),
2644 -BP_GET_PSIZE(bp
), -BP_GET_UCSIZE(bp
), tx
);
2649 dsl_get_next_livelist_obj(objset_t
*os
, uint64_t zap_obj
, uint64_t *llp
)
2654 zap_cursor_init(&zc
, os
, zap_obj
);
2655 err
= zap_cursor_retrieve(&zc
, &za
);
2656 zap_cursor_fini(&zc
);
2658 *llp
= za
.za_first_integer
;
2663 * Components of livelist deletion that must be performed in syncing
2664 * context: freeing block pointers and updating the pool-wide data
2665 * structures to indicate how much work is left to do
2667 typedef struct sublist_delete_arg
{
2672 } sublist_delete_arg_t
;
2675 sublist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2677 sublist_delete_arg_t
*sda
= arg
;
2678 spa_t
*spa
= sda
->spa
;
2679 dsl_deadlist_t
*ll
= sda
->ll
;
2680 uint64_t key
= sda
->key
;
2681 bplist_t
*to_free
= sda
->to_free
;
2683 bplist_iterate(to_free
, delete_blkptr_cb
, spa
, tx
);
2684 dsl_deadlist_remove_entry(ll
, key
, tx
);
2687 typedef struct livelist_delete_arg
{
2691 } livelist_delete_arg_t
;
2694 livelist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2696 livelist_delete_arg_t
*lda
= arg
;
2697 spa_t
*spa
= lda
->spa
;
2698 uint64_t ll_obj
= lda
->ll_obj
;
2699 uint64_t zap_obj
= lda
->zap_obj
;
2700 objset_t
*mos
= spa
->spa_meta_objset
;
2703 /* free the livelist and decrement the feature count */
2704 VERIFY0(zap_remove_int(mos
, zap_obj
, ll_obj
, tx
));
2705 dsl_deadlist_free(mos
, ll_obj
, tx
);
2706 spa_feature_decr(spa
, SPA_FEATURE_LIVELIST
, tx
);
2707 VERIFY0(zap_count(mos
, zap_obj
, &count
));
2709 /* no more livelists to delete */
2710 VERIFY0(zap_remove(mos
, DMU_POOL_DIRECTORY_OBJECT
,
2711 DMU_POOL_DELETED_CLONES
, tx
));
2712 VERIFY0(zap_destroy(mos
, zap_obj
, tx
));
2713 spa
->spa_livelists_to_delete
= 0;
2714 spa_notify_waiters(spa
);
2719 * Load in the value for the livelist to be removed and open it. Then,
2720 * load its first sublist and determine which block pointers should actually
2721 * be freed. Then, call a synctask which performs the actual frees and updates
2722 * the pool-wide livelist data.
2725 spa_livelist_delete_cb(void *arg
, zthr_t
*z
)
2728 uint64_t ll_obj
= 0, count
;
2729 objset_t
*mos
= spa
->spa_meta_objset
;
2730 uint64_t zap_obj
= spa
->spa_livelists_to_delete
;
2732 * Determine the next livelist to delete. This function should only
2733 * be called if there is at least one deleted clone.
2735 VERIFY0(dsl_get_next_livelist_obj(mos
, zap_obj
, &ll_obj
));
2736 VERIFY0(zap_count(mos
, ll_obj
, &count
));
2739 dsl_deadlist_entry_t
*dle
;
2741 ll
= kmem_zalloc(sizeof (dsl_deadlist_t
), KM_SLEEP
);
2742 dsl_deadlist_open(ll
, mos
, ll_obj
);
2743 dle
= dsl_deadlist_first(ll
);
2744 ASSERT3P(dle
, !=, NULL
);
2745 bplist_create(&to_free
);
2746 int err
= dsl_process_sub_livelist(&dle
->dle_bpobj
, &to_free
,
2749 sublist_delete_arg_t sync_arg
= {
2752 .key
= dle
->dle_mintxg
,
2755 zfs_dbgmsg("deleting sublist (id %llu) from"
2756 " livelist %llu, %lld remaining",
2757 (u_longlong_t
)dle
->dle_bpobj
.bpo_object
,
2758 (u_longlong_t
)ll_obj
, (longlong_t
)count
- 1);
2759 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
,
2760 sublist_delete_sync
, &sync_arg
, 0,
2761 ZFS_SPACE_CHECK_DESTROY
));
2763 VERIFY3U(err
, ==, EINTR
);
2765 bplist_clear(&to_free
);
2766 bplist_destroy(&to_free
);
2767 dsl_deadlist_close(ll
);
2768 kmem_free(ll
, sizeof (dsl_deadlist_t
));
2770 livelist_delete_arg_t sync_arg
= {
2775 zfs_dbgmsg("deletion of livelist %llu completed",
2776 (u_longlong_t
)ll_obj
);
2777 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
, livelist_delete_sync
,
2778 &sync_arg
, 0, ZFS_SPACE_CHECK_DESTROY
));
2783 spa_start_livelist_destroy_thread(spa_t
*spa
)
2785 ASSERT3P(spa
->spa_livelist_delete_zthr
, ==, NULL
);
2786 spa
->spa_livelist_delete_zthr
=
2787 zthr_create("z_livelist_destroy",
2788 spa_livelist_delete_cb_check
, spa_livelist_delete_cb
, spa
,
2792 typedef struct livelist_new_arg
{
2795 } livelist_new_arg_t
;
2798 livelist_track_new_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
2802 livelist_new_arg_t
*lna
= arg
;
2804 bplist_append(lna
->frees
, bp
);
2806 bplist_append(lna
->allocs
, bp
);
2807 zfs_livelist_condense_new_alloc
++;
2812 typedef struct livelist_condense_arg
{
2815 uint64_t first_size
;
2817 } livelist_condense_arg_t
;
2820 spa_livelist_condense_sync(void *arg
, dmu_tx_t
*tx
)
2822 livelist_condense_arg_t
*lca
= arg
;
2823 spa_t
*spa
= lca
->spa
;
2825 dsl_dataset_t
*ds
= spa
->spa_to_condense
.ds
;
2827 /* Have we been cancelled? */
2828 if (spa
->spa_to_condense
.cancelled
) {
2829 zfs_livelist_condense_sync_cancel
++;
2833 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
2834 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
2835 dsl_deadlist_t
*ll
= &ds
->ds_dir
->dd_livelist
;
2838 * It's possible that the livelist was changed while the zthr was
2839 * running. Therefore, we need to check for new blkptrs in the two
2840 * entries being condensed and continue to track them in the livelist.
2841 * Because of the way we handle remapped blkptrs (see dbuf_remap_impl),
2842 * it's possible that the newly added blkptrs are FREEs or ALLOCs so
2843 * we need to sort them into two different bplists.
2845 uint64_t first_obj
= first
->dle_bpobj
.bpo_object
;
2846 uint64_t next_obj
= next
->dle_bpobj
.bpo_object
;
2847 uint64_t cur_first_size
= first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
2848 uint64_t cur_next_size
= next
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
2850 bplist_create(&new_frees
);
2851 livelist_new_arg_t new_bps
= {
2852 .allocs
= &lca
->to_keep
,
2853 .frees
= &new_frees
,
2856 if (cur_first_size
> lca
->first_size
) {
2857 VERIFY0(livelist_bpobj_iterate_from_nofree(&first
->dle_bpobj
,
2858 livelist_track_new_cb
, &new_bps
, lca
->first_size
));
2860 if (cur_next_size
> lca
->next_size
) {
2861 VERIFY0(livelist_bpobj_iterate_from_nofree(&next
->dle_bpobj
,
2862 livelist_track_new_cb
, &new_bps
, lca
->next_size
));
2865 dsl_deadlist_clear_entry(first
, ll
, tx
);
2866 ASSERT(bpobj_is_empty(&first
->dle_bpobj
));
2867 dsl_deadlist_remove_entry(ll
, next
->dle_mintxg
, tx
);
2869 bplist_iterate(&lca
->to_keep
, dsl_deadlist_insert_alloc_cb
, ll
, tx
);
2870 bplist_iterate(&new_frees
, dsl_deadlist_insert_free_cb
, ll
, tx
);
2871 bplist_destroy(&new_frees
);
2873 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
2874 dsl_dataset_name(ds
, dsname
);
2875 zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu "
2876 "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu "
2877 "(%llu blkptrs)", (u_longlong_t
)tx
->tx_txg
, dsname
,
2878 (u_longlong_t
)ds
->ds_object
, (u_longlong_t
)first_obj
,
2879 (u_longlong_t
)cur_first_size
, (u_longlong_t
)next_obj
,
2880 (u_longlong_t
)cur_next_size
,
2881 (u_longlong_t
)first
->dle_bpobj
.bpo_object
,
2882 (u_longlong_t
)first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
);
2884 dmu_buf_rele(ds
->ds_dbuf
, spa
);
2885 spa
->spa_to_condense
.ds
= NULL
;
2886 bplist_clear(&lca
->to_keep
);
2887 bplist_destroy(&lca
->to_keep
);
2888 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
2889 spa
->spa_to_condense
.syncing
= B_FALSE
;
2893 spa_livelist_condense_cb(void *arg
, zthr_t
*t
)
2895 while (zfs_livelist_condense_zthr_pause
&&
2896 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
2900 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
2901 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
2902 uint64_t first_size
, next_size
;
2904 livelist_condense_arg_t
*lca
=
2905 kmem_alloc(sizeof (livelist_condense_arg_t
), KM_SLEEP
);
2906 bplist_create(&lca
->to_keep
);
2909 * Process the livelists (matching FREEs and ALLOCs) in open context
2910 * so we have minimal work in syncing context to condense.
2912 * We save bpobj sizes (first_size and next_size) to use later in
2913 * syncing context to determine if entries were added to these sublists
2914 * while in open context. This is possible because the clone is still
2915 * active and open for normal writes and we want to make sure the new,
2916 * unprocessed blockpointers are inserted into the livelist normally.
2918 * Note that dsl_process_sub_livelist() both stores the size number of
2919 * blockpointers and iterates over them while the bpobj's lock held, so
2920 * the sizes returned to us are consistent which what was actually
2923 int err
= dsl_process_sub_livelist(&first
->dle_bpobj
, &lca
->to_keep
, t
,
2926 err
= dsl_process_sub_livelist(&next
->dle_bpobj
, &lca
->to_keep
,
2930 while (zfs_livelist_condense_sync_pause
&&
2931 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
2934 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
2935 dmu_tx_mark_netfree(tx
);
2936 dmu_tx_hold_space(tx
, 1);
2937 err
= dmu_tx_assign(tx
, TXG_NOWAIT
| TXG_NOTHROTTLE
);
2940 * Prevent the condense zthr restarting before
2941 * the synctask completes.
2943 spa
->spa_to_condense
.syncing
= B_TRUE
;
2945 lca
->first_size
= first_size
;
2946 lca
->next_size
= next_size
;
2947 dsl_sync_task_nowait(spa_get_dsl(spa
),
2948 spa_livelist_condense_sync
, lca
, tx
);
2954 * Condensing can not continue: either it was externally stopped or
2955 * we were unable to assign to a tx because the pool has run out of
2956 * space. In the second case, we'll just end up trying to condense
2957 * again in a later txg.
2960 bplist_clear(&lca
->to_keep
);
2961 bplist_destroy(&lca
->to_keep
);
2962 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
2963 dmu_buf_rele(spa
->spa_to_condense
.ds
->ds_dbuf
, spa
);
2964 spa
->spa_to_condense
.ds
= NULL
;
2966 zfs_livelist_condense_zthr_cancel
++;
2970 * Check that there is something to condense but that a condense is not
2971 * already in progress and that condensing has not been cancelled.
2974 spa_livelist_condense_cb_check(void *arg
, zthr_t
*z
)
2978 if ((spa
->spa_to_condense
.ds
!= NULL
) &&
2979 (spa
->spa_to_condense
.syncing
== B_FALSE
) &&
2980 (spa
->spa_to_condense
.cancelled
== B_FALSE
)) {
2987 spa_start_livelist_condensing_thread(spa_t
*spa
)
2989 spa
->spa_to_condense
.ds
= NULL
;
2990 spa
->spa_to_condense
.first
= NULL
;
2991 spa
->spa_to_condense
.next
= NULL
;
2992 spa
->spa_to_condense
.syncing
= B_FALSE
;
2993 spa
->spa_to_condense
.cancelled
= B_FALSE
;
2995 ASSERT3P(spa
->spa_livelist_condense_zthr
, ==, NULL
);
2996 spa
->spa_livelist_condense_zthr
=
2997 zthr_create("z_livelist_condense",
2998 spa_livelist_condense_cb_check
,
2999 spa_livelist_condense_cb
, spa
, minclsyspri
);
3003 spa_spawn_aux_threads(spa_t
*spa
)
3005 ASSERT(spa_writeable(spa
));
3007 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
3009 spa_start_raidz_expansion_thread(spa
);
3010 spa_start_indirect_condensing_thread(spa
);
3011 spa_start_livelist_destroy_thread(spa
);
3012 spa_start_livelist_condensing_thread(spa
);
3014 ASSERT3P(spa
->spa_checkpoint_discard_zthr
, ==, NULL
);
3015 spa
->spa_checkpoint_discard_zthr
=
3016 zthr_create("z_checkpoint_discard",
3017 spa_checkpoint_discard_thread_check
,
3018 spa_checkpoint_discard_thread
, spa
, minclsyspri
);
3022 * Fix up config after a partly-completed split. This is done with the
3023 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
3024 * pool have that entry in their config, but only the splitting one contains
3025 * a list of all the guids of the vdevs that are being split off.
3027 * This function determines what to do with that list: either rejoin
3028 * all the disks to the pool, or complete the splitting process. To attempt
3029 * the rejoin, each disk that is offlined is marked online again, and
3030 * we do a reopen() call. If the vdev label for every disk that was
3031 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
3032 * then we call vdev_split() on each disk, and complete the split.
3034 * Otherwise we leave the config alone, with all the vdevs in place in
3035 * the original pool.
3038 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
3045 boolean_t attempt_reopen
;
3047 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
3050 /* check that the config is complete */
3051 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
3052 &glist
, &gcount
) != 0)
3055 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
3057 /* attempt to online all the vdevs & validate */
3058 attempt_reopen
= B_TRUE
;
3059 for (i
= 0; i
< gcount
; i
++) {
3060 if (glist
[i
] == 0) /* vdev is hole */
3063 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
3064 if (vd
[i
] == NULL
) {
3066 * Don't bother attempting to reopen the disks;
3067 * just do the split.
3069 attempt_reopen
= B_FALSE
;
3071 /* attempt to re-online it */
3072 vd
[i
]->vdev_offline
= B_FALSE
;
3076 if (attempt_reopen
) {
3077 vdev_reopen(spa
->spa_root_vdev
);
3079 /* check each device to see what state it's in */
3080 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
3081 if (vd
[i
] != NULL
&&
3082 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
3089 * If every disk has been moved to the new pool, or if we never
3090 * even attempted to look at them, then we split them off for
3093 if (!attempt_reopen
|| gcount
== extracted
) {
3094 for (i
= 0; i
< gcount
; i
++)
3097 vdev_reopen(spa
->spa_root_vdev
);
3100 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
3104 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
)
3106 const char *ereport
= FM_EREPORT_ZFS_POOL
;
3109 spa
->spa_load_state
= state
;
3110 (void) spa_import_progress_set_state(spa_guid(spa
),
3111 spa_load_state(spa
));
3112 spa_import_progress_set_notes(spa
, "spa_load()");
3114 gethrestime(&spa
->spa_loaded_ts
);
3115 error
= spa_load_impl(spa
, type
, &ereport
);
3118 * Don't count references from objsets that are already closed
3119 * and are making their way through the eviction process.
3121 spa_evicting_os_wait(spa
);
3122 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
3124 if (error
!= EEXIST
) {
3125 spa
->spa_loaded_ts
.tv_sec
= 0;
3126 spa
->spa_loaded_ts
.tv_nsec
= 0;
3128 if (error
!= EBADF
) {
3129 (void) zfs_ereport_post(ereport
, spa
,
3130 NULL
, NULL
, NULL
, 0);
3133 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
3136 (void) spa_import_progress_set_state(spa_guid(spa
),
3137 spa_load_state(spa
));
3144 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
3145 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
3146 * spa's per-vdev ZAP list.
3149 vdev_count_verify_zaps(vdev_t
*vd
)
3151 spa_t
*spa
= vd
->vdev_spa
;
3154 if (spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_AVZ_V2
) &&
3155 vd
->vdev_root_zap
!= 0) {
3157 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3158 spa
->spa_all_vdev_zaps
, vd
->vdev_root_zap
));
3160 if (vd
->vdev_top_zap
!= 0) {
3162 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3163 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
3165 if (vd
->vdev_leaf_zap
!= 0) {
3167 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3168 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
3171 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3172 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
3178 #define vdev_count_verify_zaps(vd) ((void) sizeof (vd), 0)
3182 * Determine whether the activity check is required.
3185 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
3189 uint64_t hostid
= 0;
3190 uint64_t tryconfig_txg
= 0;
3191 uint64_t tryconfig_timestamp
= 0;
3192 uint16_t tryconfig_mmp_seq
= 0;
3195 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3196 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
3197 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
3199 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
3200 &tryconfig_timestamp
);
3201 (void) nvlist_lookup_uint16(nvinfo
, ZPOOL_CONFIG_MMP_SEQ
,
3202 &tryconfig_mmp_seq
);
3205 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
3208 * Disable the MMP activity check - This is used by zdb which
3209 * is intended to be used on potentially active pools.
3211 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
3215 * Skip the activity check when the MMP feature is disabled.
3217 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
3221 * If the tryconfig_ values are nonzero, they are the results of an
3222 * earlier tryimport. If they all match the uberblock we just found,
3223 * then the pool has not changed and we return false so we do not test
3226 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
3227 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
&&
3228 tryconfig_mmp_seq
&& tryconfig_mmp_seq
==
3229 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0))
3233 * Allow the activity check to be skipped when importing the pool
3234 * on the same host which last imported it. Since the hostid from
3235 * configuration may be stale use the one read from the label.
3237 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
3238 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
3240 if (hostid
== spa_get_hostid(spa
))
3244 * Skip the activity test when the pool was cleanly exported.
3246 if (state
!= POOL_STATE_ACTIVE
)
3253 * Nanoseconds the activity check must watch for changes on-disk.
3256 spa_activity_check_duration(spa_t
*spa
, uberblock_t
*ub
)
3258 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
3259 uint64_t multihost_interval
= MSEC2NSEC(
3260 MMP_INTERVAL_OK(zfs_multihost_interval
));
3261 uint64_t import_delay
= MAX(NANOSEC
, import_intervals
*
3262 multihost_interval
);
3265 * Local tunables determine a minimum duration except for the case
3266 * where we know when the remote host will suspend the pool if MMP
3267 * writes do not land.
3269 * See Big Theory comment at the top of mmp.c for the reasoning behind
3270 * these cases and times.
3273 ASSERT(MMP_IMPORT_SAFETY_FACTOR
>= 100);
3275 if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3276 MMP_FAIL_INT(ub
) > 0) {
3278 /* MMP on remote host will suspend pool after failed writes */
3279 import_delay
= MMP_FAIL_INT(ub
) * MSEC2NSEC(MMP_INTERVAL(ub
)) *
3280 MMP_IMPORT_SAFETY_FACTOR
/ 100;
3282 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3283 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3284 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3285 (u_longlong_t
)MMP_FAIL_INT(ub
),
3286 (u_longlong_t
)MMP_INTERVAL(ub
),
3287 (u_longlong_t
)import_intervals
);
3289 } else if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3290 MMP_FAIL_INT(ub
) == 0) {
3292 /* MMP on remote host will never suspend pool */
3293 import_delay
= MAX(import_delay
, (MSEC2NSEC(MMP_INTERVAL(ub
)) +
3294 ub
->ub_mmp_delay
) * import_intervals
);
3296 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3297 "mmp_interval=%llu ub_mmp_delay=%llu "
3298 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3299 (u_longlong_t
)MMP_INTERVAL(ub
),
3300 (u_longlong_t
)ub
->ub_mmp_delay
,
3301 (u_longlong_t
)import_intervals
);
3303 } else if (MMP_VALID(ub
)) {
3305 * zfs-0.7 compatibility case
3308 import_delay
= MAX(import_delay
, (multihost_interval
+
3309 ub
->ub_mmp_delay
) * import_intervals
);
3311 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3312 "import_intervals=%llu leaves=%u",
3313 (u_longlong_t
)import_delay
,
3314 (u_longlong_t
)ub
->ub_mmp_delay
,
3315 (u_longlong_t
)import_intervals
,
3316 vdev_count_leaves(spa
));
3318 /* Using local tunings is the only reasonable option */
3319 zfs_dbgmsg("pool last imported on non-MMP aware "
3320 "host using import_delay=%llu multihost_interval=%llu "
3321 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3322 (u_longlong_t
)multihost_interval
,
3323 (u_longlong_t
)import_intervals
);
3326 return (import_delay
);
3330 * Perform the import activity check. If the user canceled the import or
3331 * we detected activity then fail.
3334 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
3336 uint64_t txg
= ub
->ub_txg
;
3337 uint64_t timestamp
= ub
->ub_timestamp
;
3338 uint64_t mmp_config
= ub
->ub_mmp_config
;
3339 uint16_t mmp_seq
= MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0;
3340 uint64_t import_delay
;
3341 hrtime_t import_expire
, now
;
3342 nvlist_t
*mmp_label
= NULL
;
3343 vdev_t
*rvd
= spa
->spa_root_vdev
;
3348 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
3349 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3353 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3354 * during the earlier tryimport. If the txg recorded there is 0 then
3355 * the pool is known to be active on another host.
3357 * Otherwise, the pool might be in use on another host. Check for
3358 * changes in the uberblocks on disk if necessary.
3360 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3361 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
3362 ZPOOL_CONFIG_LOAD_INFO
);
3364 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
3365 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
3366 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3367 error
= SET_ERROR(EREMOTEIO
);
3372 import_delay
= spa_activity_check_duration(spa
, ub
);
3374 /* Add a small random factor in case of simultaneous imports (0-25%) */
3375 import_delay
+= import_delay
* random_in_range(250) / 1000;
3377 import_expire
= gethrtime() + import_delay
;
3379 spa_import_progress_set_notes(spa
, "Checking MMP activity, waiting "
3380 "%llu ms", (u_longlong_t
)NSEC2MSEC(import_delay
));
3382 int interations
= 0;
3383 while ((now
= gethrtime()) < import_expire
) {
3384 if (interations
++ % 30 == 0) {
3385 spa_import_progress_set_notes(spa
, "Checking MMP "
3386 "activity, %llu ms remaining",
3387 (u_longlong_t
)NSEC2MSEC(import_expire
- now
));
3390 (void) spa_import_progress_set_mmp_check(spa_guid(spa
),
3391 NSEC2SEC(import_expire
- gethrtime()));
3393 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3395 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
||
3396 mmp_seq
!= (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0)) {
3397 zfs_dbgmsg("multihost activity detected "
3398 "txg %llu ub_txg %llu "
3399 "timestamp %llu ub_timestamp %llu "
3400 "mmp_config %#llx ub_mmp_config %#llx",
3401 (u_longlong_t
)txg
, (u_longlong_t
)ub
->ub_txg
,
3402 (u_longlong_t
)timestamp
,
3403 (u_longlong_t
)ub
->ub_timestamp
,
3404 (u_longlong_t
)mmp_config
,
3405 (u_longlong_t
)ub
->ub_mmp_config
);
3407 error
= SET_ERROR(EREMOTEIO
);
3412 nvlist_free(mmp_label
);
3416 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
3418 error
= SET_ERROR(EINTR
);
3426 mutex_destroy(&mtx
);
3430 * If the pool is determined to be active store the status in the
3431 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3432 * available from configuration read from disk store them as well.
3433 * This allows 'zpool import' to generate a more useful message.
3435 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3436 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3437 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3439 if (error
== EREMOTEIO
) {
3440 const char *hostname
= "<unknown>";
3441 uint64_t hostid
= 0;
3444 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
3445 hostname
= fnvlist_lookup_string(mmp_label
,
3446 ZPOOL_CONFIG_HOSTNAME
);
3447 fnvlist_add_string(spa
->spa_load_info
,
3448 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
3451 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
3452 hostid
= fnvlist_lookup_uint64(mmp_label
,
3453 ZPOOL_CONFIG_HOSTID
);
3454 fnvlist_add_uint64(spa
->spa_load_info
,
3455 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
3459 fnvlist_add_uint64(spa
->spa_load_info
,
3460 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
3461 fnvlist_add_uint64(spa
->spa_load_info
,
3462 ZPOOL_CONFIG_MMP_TXG
, 0);
3464 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
3468 nvlist_free(mmp_label
);
3474 spa_verify_host(spa_t
*spa
, nvlist_t
*mos_config
)
3477 const char *hostname
;
3478 uint64_t myhostid
= 0;
3480 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
3481 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
3482 hostname
= fnvlist_lookup_string(mos_config
,
3483 ZPOOL_CONFIG_HOSTNAME
);
3485 myhostid
= zone_get_hostid(NULL
);
3487 if (hostid
!= 0 && myhostid
!= 0 && hostid
!= myhostid
) {
3488 cmn_err(CE_WARN
, "pool '%s' could not be "
3489 "loaded as it was last accessed by "
3490 "another system (host: %s hostid: 0x%llx). "
3491 "See: https://openzfs.github.io/openzfs-docs/msg/"
3493 spa_name(spa
), hostname
, (u_longlong_t
)hostid
);
3494 spa_load_failed(spa
, "hostid verification failed: pool "
3495 "last accessed by host: %s (hostid: 0x%llx)",
3496 hostname
, (u_longlong_t
)hostid
);
3497 return (SET_ERROR(EBADF
));
3505 spa_ld_parse_config(spa_t
*spa
, spa_import_type_t type
)
3508 nvlist_t
*nvtree
, *nvl
, *config
= spa
->spa_config
;
3512 const char *comment
;
3513 const char *compatibility
;
3516 * Versioning wasn't explicitly added to the label until later, so if
3517 * it's not present treat it as the initial version.
3519 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
3520 &spa
->spa_ubsync
.ub_version
) != 0)
3521 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
3523 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
3524 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3525 ZPOOL_CONFIG_POOL_GUID
);
3526 return (SET_ERROR(EINVAL
));
3530 * If we are doing an import, ensure that the pool is not already
3531 * imported by checking if its pool guid already exists in the
3534 * The only case that we allow an already imported pool to be
3535 * imported again, is when the pool is checkpointed and we want to
3536 * look at its checkpointed state from userland tools like zdb.
3539 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3540 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3541 spa_guid_exists(pool_guid
, 0)) {
3543 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3544 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3545 spa_guid_exists(pool_guid
, 0) &&
3546 !spa_importing_readonly_checkpoint(spa
)) {
3548 spa_load_failed(spa
, "a pool with guid %llu is already open",
3549 (u_longlong_t
)pool_guid
);
3550 return (SET_ERROR(EEXIST
));
3553 spa
->spa_config_guid
= pool_guid
;
3555 nvlist_free(spa
->spa_load_info
);
3556 spa
->spa_load_info
= fnvlist_alloc();
3558 ASSERT(spa
->spa_comment
== NULL
);
3559 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
3560 spa
->spa_comment
= spa_strdup(comment
);
3562 ASSERT(spa
->spa_compatibility
== NULL
);
3563 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMPATIBILITY
,
3564 &compatibility
) == 0)
3565 spa
->spa_compatibility
= spa_strdup(compatibility
);
3567 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
3568 &spa
->spa_config_txg
);
3570 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) == 0)
3571 spa
->spa_config_splitting
= fnvlist_dup(nvl
);
3573 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
)) {
3574 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3575 ZPOOL_CONFIG_VDEV_TREE
);
3576 return (SET_ERROR(EINVAL
));
3580 * Create "The Godfather" zio to hold all async IOs
3582 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3584 for (int i
= 0; i
< max_ncpus
; i
++) {
3585 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3586 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3587 ZIO_FLAG_GODFATHER
);
3591 * Parse the configuration into a vdev tree. We explicitly set the
3592 * value that will be returned by spa_version() since parsing the
3593 * configuration requires knowing the version number.
3595 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3596 parse
= (type
== SPA_IMPORT_EXISTING
?
3597 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
3598 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
3599 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3602 spa_load_failed(spa
, "unable to parse config [error=%d]",
3607 ASSERT(spa
->spa_root_vdev
== rvd
);
3608 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
3609 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
3611 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3612 ASSERT(spa_guid(spa
) == pool_guid
);
3619 * Recursively open all vdevs in the vdev tree. This function is called twice:
3620 * first with the untrusted config, then with the trusted config.
3623 spa_ld_open_vdevs(spa_t
*spa
)
3628 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3629 * missing/unopenable for the root vdev to be still considered openable.
3631 if (spa
->spa_trust_config
) {
3632 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds
;
3633 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_CACHEFILE
) {
3634 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_cachefile
;
3635 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_SCAN
) {
3636 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_scan
;
3638 spa
->spa_missing_tvds_allowed
= 0;
3641 spa
->spa_missing_tvds_allowed
=
3642 MAX(zfs_max_missing_tvds
, spa
->spa_missing_tvds_allowed
);
3644 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3645 error
= vdev_open(spa
->spa_root_vdev
);
3646 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3648 if (spa
->spa_missing_tvds
!= 0) {
3649 spa_load_note(spa
, "vdev tree has %lld missing top-level "
3650 "vdevs.", (u_longlong_t
)spa
->spa_missing_tvds
);
3651 if (spa
->spa_trust_config
&& (spa
->spa_mode
& SPA_MODE_WRITE
)) {
3653 * Although theoretically we could allow users to open
3654 * incomplete pools in RW mode, we'd need to add a lot
3655 * of extra logic (e.g. adjust pool space to account
3656 * for missing vdevs).
3657 * This limitation also prevents users from accidentally
3658 * opening the pool in RW mode during data recovery and
3659 * damaging it further.
3661 spa_load_note(spa
, "pools with missing top-level "
3662 "vdevs can only be opened in read-only mode.");
3663 error
= SET_ERROR(ENXIO
);
3665 spa_load_note(spa
, "current settings allow for maximum "
3666 "%lld missing top-level vdevs at this stage.",
3667 (u_longlong_t
)spa
->spa_missing_tvds_allowed
);
3671 spa_load_failed(spa
, "unable to open vdev tree [error=%d]",
3674 if (spa
->spa_missing_tvds
!= 0 || error
!= 0)
3675 vdev_dbgmsg_print_tree(spa
->spa_root_vdev
, 2);
3681 * We need to validate the vdev labels against the configuration that
3682 * we have in hand. This function is called twice: first with an untrusted
3683 * config, then with a trusted config. The validation is more strict when the
3684 * config is trusted.
3687 spa_ld_validate_vdevs(spa_t
*spa
)
3690 vdev_t
*rvd
= spa
->spa_root_vdev
;
3692 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3693 error
= vdev_validate(rvd
);
3694 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3697 spa_load_failed(spa
, "vdev_validate failed [error=%d]", error
);
3701 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
3702 spa_load_failed(spa
, "cannot open vdev tree after invalidating "
3704 vdev_dbgmsg_print_tree(rvd
, 2);
3705 return (SET_ERROR(ENXIO
));
3712 spa_ld_select_uberblock_done(spa_t
*spa
, uberblock_t
*ub
)
3714 spa
->spa_state
= POOL_STATE_ACTIVE
;
3715 spa
->spa_ubsync
= spa
->spa_uberblock
;
3716 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
3717 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
3718 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
3719 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
3720 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
3721 spa
->spa_prev_software_version
= ub
->ub_software_version
;
3725 spa_ld_select_uberblock(spa_t
*spa
, spa_import_type_t type
)
3727 vdev_t
*rvd
= spa
->spa_root_vdev
;
3729 uberblock_t
*ub
= &spa
->spa_uberblock
;
3730 boolean_t activity_check
= B_FALSE
;
3733 * If we are opening the checkpointed state of the pool by
3734 * rewinding to it, at this point we will have written the
3735 * checkpointed uberblock to the vdev labels, so searching
3736 * the labels will find the right uberblock. However, if
3737 * we are opening the checkpointed state read-only, we have
3738 * not modified the labels. Therefore, we must ignore the
3739 * labels and continue using the spa_uberblock that was set
3740 * by spa_ld_checkpoint_rewind.
3742 * Note that it would be fine to ignore the labels when
3743 * rewinding (opening writeable) as well. However, if we
3744 * crash just after writing the labels, we will end up
3745 * searching the labels. Doing so in the common case means
3746 * that this code path gets exercised normally, rather than
3747 * just in the edge case.
3749 if (ub
->ub_checkpoint_txg
!= 0 &&
3750 spa_importing_readonly_checkpoint(spa
)) {
3751 spa_ld_select_uberblock_done(spa
, ub
);
3756 * Find the best uberblock.
3758 vdev_uberblock_load(rvd
, ub
, &label
);
3761 * If we weren't able to find a single valid uberblock, return failure.
3763 if (ub
->ub_txg
== 0) {
3765 spa_load_failed(spa
, "no valid uberblock found");
3766 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
3769 if (spa
->spa_load_max_txg
!= UINT64_MAX
) {
3770 (void) spa_import_progress_set_max_txg(spa_guid(spa
),
3771 (u_longlong_t
)spa
->spa_load_max_txg
);
3773 spa_load_note(spa
, "using uberblock with txg=%llu",
3774 (u_longlong_t
)ub
->ub_txg
);
3775 if (ub
->ub_raidz_reflow_info
!= 0) {
3776 spa_load_note(spa
, "uberblock raidz_reflow_info: "
3777 "state=%u offset=%llu",
3778 (int)RRSS_GET_STATE(ub
),
3779 (u_longlong_t
)RRSS_GET_OFFSET(ub
));
3784 * For pools which have the multihost property on determine if the
3785 * pool is truly inactive and can be safely imported. Prevent
3786 * hosts which don't have a hostid set from importing the pool.
3788 activity_check
= spa_activity_check_required(spa
, ub
, label
,
3790 if (activity_check
) {
3791 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
3792 spa_get_hostid(spa
) == 0) {
3794 fnvlist_add_uint64(spa
->spa_load_info
,
3795 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3796 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3799 int error
= spa_activity_check(spa
, ub
, spa
->spa_config
);
3805 fnvlist_add_uint64(spa
->spa_load_info
,
3806 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
3807 fnvlist_add_uint64(spa
->spa_load_info
,
3808 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
3809 fnvlist_add_uint16(spa
->spa_load_info
,
3810 ZPOOL_CONFIG_MMP_SEQ
,
3811 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0));
3815 * If the pool has an unsupported version we can't open it.
3817 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
3819 spa_load_failed(spa
, "version %llu is not supported",
3820 (u_longlong_t
)ub
->ub_version
);
3821 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
3824 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3828 * If we weren't able to find what's necessary for reading the
3829 * MOS in the label, return failure.
3831 if (label
== NULL
) {
3832 spa_load_failed(spa
, "label config unavailable");
3833 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3837 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
3840 spa_load_failed(spa
, "invalid label: '%s' missing",
3841 ZPOOL_CONFIG_FEATURES_FOR_READ
);
3842 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3847 * Update our in-core representation with the definitive values
3850 nvlist_free(spa
->spa_label_features
);
3851 spa
->spa_label_features
= fnvlist_dup(features
);
3857 * Look through entries in the label nvlist's features_for_read. If
3858 * there is a feature listed there which we don't understand then we
3859 * cannot open a pool.
3861 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3862 nvlist_t
*unsup_feat
;
3864 unsup_feat
= fnvlist_alloc();
3866 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
3868 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
3869 if (!zfeature_is_supported(nvpair_name(nvp
))) {
3870 fnvlist_add_string(unsup_feat
,
3871 nvpair_name(nvp
), "");
3875 if (!nvlist_empty(unsup_feat
)) {
3876 fnvlist_add_nvlist(spa
->spa_load_info
,
3877 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
3878 nvlist_free(unsup_feat
);
3879 spa_load_failed(spa
, "some features are unsupported");
3880 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
3884 nvlist_free(unsup_feat
);
3887 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
3888 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3889 spa_try_repair(spa
, spa
->spa_config
);
3890 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3891 nvlist_free(spa
->spa_config_splitting
);
3892 spa
->spa_config_splitting
= NULL
;
3896 * Initialize internal SPA structures.
3898 spa_ld_select_uberblock_done(spa
, ub
);
3904 spa_ld_open_rootbp(spa_t
*spa
)
3907 vdev_t
*rvd
= spa
->spa_root_vdev
;
3909 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
3911 spa_load_failed(spa
, "unable to open rootbp in dsl_pool_init "
3912 "[error=%d]", error
);
3913 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3915 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
3921 spa_ld_trusted_config(spa_t
*spa
, spa_import_type_t type
,
3922 boolean_t reloading
)
3924 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
3925 nvlist_t
*nv
, *mos_config
, *policy
;
3926 int error
= 0, copy_error
;
3927 uint64_t healthy_tvds
, healthy_tvds_mos
;
3928 uint64_t mos_config_txg
;
3930 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
, B_TRUE
)
3932 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3935 * If we're assembling a pool from a split, the config provided is
3936 * already trusted so there is nothing to do.
3938 if (type
== SPA_IMPORT_ASSEMBLE
)
3941 healthy_tvds
= spa_healthy_core_tvds(spa
);
3943 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
)
3945 spa_load_failed(spa
, "unable to retrieve MOS config");
3946 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3950 * If we are doing an open, pool owner wasn't verified yet, thus do
3951 * the verification here.
3953 if (spa
->spa_load_state
== SPA_LOAD_OPEN
) {
3954 error
= spa_verify_host(spa
, mos_config
);
3956 nvlist_free(mos_config
);
3961 nv
= fnvlist_lookup_nvlist(mos_config
, ZPOOL_CONFIG_VDEV_TREE
);
3963 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3966 * Build a new vdev tree from the trusted config
3968 error
= spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
);
3970 nvlist_free(mos_config
);
3971 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3972 spa_load_failed(spa
, "spa_config_parse failed [error=%d]",
3974 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3978 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3979 * obtained by scanning /dev/dsk, then it will have the right vdev
3980 * paths. We update the trusted MOS config with this information.
3981 * We first try to copy the paths with vdev_copy_path_strict, which
3982 * succeeds only when both configs have exactly the same vdev tree.
3983 * If that fails, we fall back to a more flexible method that has a
3984 * best effort policy.
3986 copy_error
= vdev_copy_path_strict(rvd
, mrvd
);
3987 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3988 spa_load_note(spa
, "provided vdev tree:");
3989 vdev_dbgmsg_print_tree(rvd
, 2);
3990 spa_load_note(spa
, "MOS vdev tree:");
3991 vdev_dbgmsg_print_tree(mrvd
, 2);
3993 if (copy_error
!= 0) {
3994 spa_load_note(spa
, "vdev_copy_path_strict failed, falling "
3995 "back to vdev_copy_path_relaxed");
3996 vdev_copy_path_relaxed(rvd
, mrvd
);
4001 spa
->spa_root_vdev
= mrvd
;
4003 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4006 * If 'zpool import' used a cached config, then the on-disk hostid and
4007 * hostname may be different to the cached config in ways that should
4008 * prevent import. Userspace can't discover this without a scan, but
4009 * we know, so we add these values to LOAD_INFO so the caller can know
4012 * Note that we have to do this before the config is regenerated,
4013 * because the new config will have the hostid and hostname for this
4014 * host, in readiness for import.
4016 if (nvlist_exists(mos_config
, ZPOOL_CONFIG_HOSTID
))
4017 fnvlist_add_uint64(spa
->spa_load_info
, ZPOOL_CONFIG_HOSTID
,
4018 fnvlist_lookup_uint64(mos_config
, ZPOOL_CONFIG_HOSTID
));
4019 if (nvlist_exists(mos_config
, ZPOOL_CONFIG_HOSTNAME
))
4020 fnvlist_add_string(spa
->spa_load_info
, ZPOOL_CONFIG_HOSTNAME
,
4021 fnvlist_lookup_string(mos_config
, ZPOOL_CONFIG_HOSTNAME
));
4024 * We will use spa_config if we decide to reload the spa or if spa_load
4025 * fails and we rewind. We must thus regenerate the config using the
4026 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
4027 * pass settings on how to load the pool and is not stored in the MOS.
4028 * We copy it over to our new, trusted config.
4030 mos_config_txg
= fnvlist_lookup_uint64(mos_config
,
4031 ZPOOL_CONFIG_POOL_TXG
);
4032 nvlist_free(mos_config
);
4033 mos_config
= spa_config_generate(spa
, NULL
, mos_config_txg
, B_FALSE
);
4034 if (nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_LOAD_POLICY
,
4036 fnvlist_add_nvlist(mos_config
, ZPOOL_LOAD_POLICY
, policy
);
4037 spa_config_set(spa
, mos_config
);
4038 spa
->spa_config_source
= SPA_CONFIG_SRC_MOS
;
4041 * Now that we got the config from the MOS, we should be more strict
4042 * in checking blkptrs and can make assumptions about the consistency
4043 * of the vdev tree. spa_trust_config must be set to true before opening
4044 * vdevs in order for them to be writeable.
4046 spa
->spa_trust_config
= B_TRUE
;
4049 * Open and validate the new vdev tree
4051 error
= spa_ld_open_vdevs(spa
);
4055 error
= spa_ld_validate_vdevs(spa
);
4059 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
4060 spa_load_note(spa
, "final vdev tree:");
4061 vdev_dbgmsg_print_tree(rvd
, 2);
4064 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
&&
4065 !spa
->spa_extreme_rewind
&& zfs_max_missing_tvds
== 0) {
4067 * Sanity check to make sure that we are indeed loading the
4068 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
4069 * in the config provided and they happened to be the only ones
4070 * to have the latest uberblock, we could involuntarily perform
4071 * an extreme rewind.
4073 healthy_tvds_mos
= spa_healthy_core_tvds(spa
);
4074 if (healthy_tvds_mos
- healthy_tvds
>=
4075 SPA_SYNC_MIN_VDEVS
) {
4076 spa_load_note(spa
, "config provided misses too many "
4077 "top-level vdevs compared to MOS (%lld vs %lld). ",
4078 (u_longlong_t
)healthy_tvds
,
4079 (u_longlong_t
)healthy_tvds_mos
);
4080 spa_load_note(spa
, "vdev tree:");
4081 vdev_dbgmsg_print_tree(rvd
, 2);
4083 spa_load_failed(spa
, "config was already "
4084 "provided from MOS. Aborting.");
4085 return (spa_vdev_err(rvd
,
4086 VDEV_AUX_CORRUPT_DATA
, EIO
));
4088 spa_load_note(spa
, "spa must be reloaded using MOS "
4090 return (SET_ERROR(EAGAIN
));
4094 error
= spa_check_for_missing_logs(spa
);
4096 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
4098 if (rvd
->vdev_guid_sum
!= spa
->spa_uberblock
.ub_guid_sum
) {
4099 spa_load_failed(spa
, "uberblock guid sum doesn't match MOS "
4100 "guid sum (%llu != %llu)",
4101 (u_longlong_t
)spa
->spa_uberblock
.ub_guid_sum
,
4102 (u_longlong_t
)rvd
->vdev_guid_sum
);
4103 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
4111 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
4114 vdev_t
*rvd
= spa
->spa_root_vdev
;
4117 * Everything that we read before spa_remove_init() must be stored
4118 * on concreted vdevs. Therefore we do this as early as possible.
4120 error
= spa_remove_init(spa
);
4122 spa_load_failed(spa
, "spa_remove_init failed [error=%d]",
4124 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4128 * Retrieve information needed to condense indirect vdev mappings.
4130 error
= spa_condense_init(spa
);
4132 spa_load_failed(spa
, "spa_condense_init failed [error=%d]",
4134 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4141 spa_ld_check_features(spa_t
*spa
, boolean_t
*missing_feat_writep
)
4144 vdev_t
*rvd
= spa
->spa_root_vdev
;
4146 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
4147 boolean_t missing_feat_read
= B_FALSE
;
4148 nvlist_t
*unsup_feat
, *enabled_feat
;
4150 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
4151 &spa
->spa_feat_for_read_obj
, B_TRUE
) != 0) {
4152 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4155 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
4156 &spa
->spa_feat_for_write_obj
, B_TRUE
) != 0) {
4157 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4160 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
4161 &spa
->spa_feat_desc_obj
, B_TRUE
) != 0) {
4162 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4165 enabled_feat
= fnvlist_alloc();
4166 unsup_feat
= fnvlist_alloc();
4168 if (!spa_features_check(spa
, B_FALSE
,
4169 unsup_feat
, enabled_feat
))
4170 missing_feat_read
= B_TRUE
;
4172 if (spa_writeable(spa
) ||
4173 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) {
4174 if (!spa_features_check(spa
, B_TRUE
,
4175 unsup_feat
, enabled_feat
)) {
4176 *missing_feat_writep
= B_TRUE
;
4180 fnvlist_add_nvlist(spa
->spa_load_info
,
4181 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
4183 if (!nvlist_empty(unsup_feat
)) {
4184 fnvlist_add_nvlist(spa
->spa_load_info
,
4185 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
4188 fnvlist_free(enabled_feat
);
4189 fnvlist_free(unsup_feat
);
4191 if (!missing_feat_read
) {
4192 fnvlist_add_boolean(spa
->spa_load_info
,
4193 ZPOOL_CONFIG_CAN_RDONLY
);
4197 * If the state is SPA_LOAD_TRYIMPORT, our objective is
4198 * twofold: to determine whether the pool is available for
4199 * import in read-write mode and (if it is not) whether the
4200 * pool is available for import in read-only mode. If the pool
4201 * is available for import in read-write mode, it is displayed
4202 * as available in userland; if it is not available for import
4203 * in read-only mode, it is displayed as unavailable in
4204 * userland. If the pool is available for import in read-only
4205 * mode but not read-write mode, it is displayed as unavailable
4206 * in userland with a special note that the pool is actually
4207 * available for open in read-only mode.
4209 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
4210 * missing a feature for write, we must first determine whether
4211 * the pool can be opened read-only before returning to
4212 * userland in order to know whether to display the
4213 * abovementioned note.
4215 if (missing_feat_read
|| (*missing_feat_writep
&&
4216 spa_writeable(spa
))) {
4217 spa_load_failed(spa
, "pool uses unsupported features");
4218 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
4223 * Load refcounts for ZFS features from disk into an in-memory
4224 * cache during SPA initialization.
4226 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
4229 error
= feature_get_refcount_from_disk(spa
,
4230 &spa_feature_table
[i
], &refcount
);
4232 spa
->spa_feat_refcount_cache
[i
] = refcount
;
4233 } else if (error
== ENOTSUP
) {
4234 spa
->spa_feat_refcount_cache
[i
] =
4235 SPA_FEATURE_DISABLED
;
4237 spa_load_failed(spa
, "error getting refcount "
4238 "for feature %s [error=%d]",
4239 spa_feature_table
[i
].fi_guid
, error
);
4240 return (spa_vdev_err(rvd
,
4241 VDEV_AUX_CORRUPT_DATA
, EIO
));
4246 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
4247 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
4248 &spa
->spa_feat_enabled_txg_obj
, B_TRUE
) != 0)
4249 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4253 * Encryption was added before bookmark_v2, even though bookmark_v2
4254 * is now a dependency. If this pool has encryption enabled without
4255 * bookmark_v2, trigger an errata message.
4257 if (spa_feature_is_enabled(spa
, SPA_FEATURE_ENCRYPTION
) &&
4258 !spa_feature_is_enabled(spa
, SPA_FEATURE_BOOKMARK_V2
)) {
4259 spa
->spa_errata
= ZPOOL_ERRATA_ZOL_8308_ENCRYPTION
;
4266 spa_ld_load_special_directories(spa_t
*spa
)
4269 vdev_t
*rvd
= spa
->spa_root_vdev
;
4271 spa
->spa_is_initializing
= B_TRUE
;
4272 error
= dsl_pool_open(spa
->spa_dsl_pool
);
4273 spa
->spa_is_initializing
= B_FALSE
;
4275 spa_load_failed(spa
, "dsl_pool_open failed [error=%d]", error
);
4276 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4283 spa_ld_get_props(spa_t
*spa
)
4287 vdev_t
*rvd
= spa
->spa_root_vdev
;
4289 /* Grab the checksum salt from the MOS. */
4290 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4291 DMU_POOL_CHECKSUM_SALT
, 1,
4292 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
4293 spa
->spa_cksum_salt
.zcs_bytes
);
4294 if (error
== ENOENT
) {
4295 /* Generate a new salt for subsequent use */
4296 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4297 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4298 } else if (error
!= 0) {
4299 spa_load_failed(spa
, "unable to retrieve checksum salt from "
4300 "MOS [error=%d]", error
);
4301 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4304 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
, B_TRUE
) != 0)
4305 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4306 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
4308 spa_load_failed(spa
, "error opening deferred-frees bpobj "
4309 "[error=%d]", error
);
4310 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4314 * Load the bit that tells us to use the new accounting function
4315 * (raid-z deflation). If we have an older pool, this will not
4318 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
, B_FALSE
);
4319 if (error
!= 0 && error
!= ENOENT
)
4320 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4322 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
4323 &spa
->spa_creation_version
, B_FALSE
);
4324 if (error
!= 0 && error
!= ENOENT
)
4325 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4328 * Load the persistent error log. If we have an older pool, this will
4331 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
,
4333 if (error
!= 0 && error
!= ENOENT
)
4334 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4336 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
4337 &spa
->spa_errlog_scrub
, B_FALSE
);
4338 if (error
!= 0 && error
!= ENOENT
)
4339 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4342 * Load the livelist deletion field. If a livelist is queued for
4343 * deletion, indicate that in the spa
4345 error
= spa_dir_prop(spa
, DMU_POOL_DELETED_CLONES
,
4346 &spa
->spa_livelists_to_delete
, B_FALSE
);
4347 if (error
!= 0 && error
!= ENOENT
)
4348 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4351 * Load the history object. If we have an older pool, this
4352 * will not be present.
4354 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
, B_FALSE
);
4355 if (error
!= 0 && error
!= ENOENT
)
4356 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4359 * Load the per-vdev ZAP map. If we have an older pool, this will not
4360 * be present; in this case, defer its creation to a later time to
4361 * avoid dirtying the MOS this early / out of sync context. See
4362 * spa_sync_config_object.
4365 /* The sentinel is only available in the MOS config. */
4366 nvlist_t
*mos_config
;
4367 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0) {
4368 spa_load_failed(spa
, "unable to retrieve MOS config");
4369 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4372 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
4373 &spa
->spa_all_vdev_zaps
, B_FALSE
);
4375 if (error
== ENOENT
) {
4376 VERIFY(!nvlist_exists(mos_config
,
4377 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
4378 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
4379 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4380 } else if (error
!= 0) {
4381 nvlist_free(mos_config
);
4382 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4383 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
4385 * An older version of ZFS overwrote the sentinel value, so
4386 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4387 * destruction to later; see spa_sync_config_object.
4389 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
4391 * We're assuming that no vdevs have had their ZAPs created
4392 * before this. Better be sure of it.
4394 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4396 nvlist_free(mos_config
);
4398 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4400 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
,
4402 if (error
&& error
!= ENOENT
)
4403 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4406 uint64_t autoreplace
= 0;
4408 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
4409 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
4410 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
4411 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
4412 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
4413 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
4414 spa_prop_find(spa
, ZPOOL_PROP_AUTOTRIM
, &spa
->spa_autotrim
);
4415 spa
->spa_autoreplace
= (autoreplace
!= 0);
4419 * If we are importing a pool with missing top-level vdevs,
4420 * we enforce that the pool doesn't panic or get suspended on
4421 * error since the likelihood of missing data is extremely high.
4423 if (spa
->spa_missing_tvds
> 0 &&
4424 spa
->spa_failmode
!= ZIO_FAILURE_MODE_CONTINUE
&&
4425 spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4426 spa_load_note(spa
, "forcing failmode to 'continue' "
4427 "as some top level vdevs are missing");
4428 spa
->spa_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
4435 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
4438 vdev_t
*rvd
= spa
->spa_root_vdev
;
4441 * If we're assembling the pool from the split-off vdevs of
4442 * an existing pool, we don't want to attach the spares & cache
4447 * Load any hot spares for this pool.
4449 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
,
4451 if (error
!= 0 && error
!= ENOENT
)
4452 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4453 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4454 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
4455 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
4456 &spa
->spa_spares
.sav_config
) != 0) {
4457 spa_load_failed(spa
, "error loading spares nvlist");
4458 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4461 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4462 spa_load_spares(spa
);
4463 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4464 } else if (error
== 0) {
4465 spa
->spa_spares
.sav_sync
= B_TRUE
;
4469 * Load any level 2 ARC devices for this pool.
4471 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
4472 &spa
->spa_l2cache
.sav_object
, B_FALSE
);
4473 if (error
!= 0 && error
!= ENOENT
)
4474 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4475 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4476 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
4477 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
4478 &spa
->spa_l2cache
.sav_config
) != 0) {
4479 spa_load_failed(spa
, "error loading l2cache nvlist");
4480 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4483 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4484 spa_load_l2cache(spa
);
4485 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4486 } else if (error
== 0) {
4487 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4494 spa_ld_load_vdev_metadata(spa_t
*spa
)
4497 vdev_t
*rvd
= spa
->spa_root_vdev
;
4500 * If the 'multihost' property is set, then never allow a pool to
4501 * be imported when the system hostid is zero. The exception to
4502 * this rule is zdb which is always allowed to access pools.
4504 if (spa_multihost(spa
) && spa_get_hostid(spa
) == 0 &&
4505 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
4506 fnvlist_add_uint64(spa
->spa_load_info
,
4507 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
4508 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
4512 * If the 'autoreplace' property is set, then post a resource notifying
4513 * the ZFS DE that it should not issue any faults for unopenable
4514 * devices. We also iterate over the vdevs, and post a sysevent for any
4515 * unopenable vdevs so that the normal autoreplace handler can take
4518 if (spa
->spa_autoreplace
&& spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4519 spa_check_removed(spa
->spa_root_vdev
);
4521 * For the import case, this is done in spa_import(), because
4522 * at this point we're using the spare definitions from
4523 * the MOS config, not necessarily from the userland config.
4525 if (spa
->spa_load_state
!= SPA_LOAD_IMPORT
) {
4526 spa_aux_check_removed(&spa
->spa_spares
);
4527 spa_aux_check_removed(&spa
->spa_l2cache
);
4532 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4534 error
= vdev_load(rvd
);
4536 spa_load_failed(spa
, "vdev_load failed [error=%d]", error
);
4537 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4540 error
= spa_ld_log_spacemaps(spa
);
4542 spa_load_failed(spa
, "spa_ld_log_spacemaps failed [error=%d]",
4544 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4548 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4550 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4551 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
, B_FALSE
);
4552 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4558 spa_ld_load_dedup_tables(spa_t
*spa
)
4561 vdev_t
*rvd
= spa
->spa_root_vdev
;
4563 error
= ddt_load(spa
);
4565 spa_load_failed(spa
, "ddt_load failed [error=%d]", error
);
4566 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4573 spa_ld_load_brt(spa_t
*spa
)
4576 vdev_t
*rvd
= spa
->spa_root_vdev
;
4578 error
= brt_load(spa
);
4580 spa_load_failed(spa
, "brt_load failed [error=%d]", error
);
4581 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4588 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, const char **ereport
)
4590 vdev_t
*rvd
= spa
->spa_root_vdev
;
4592 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
)) {
4593 boolean_t missing
= spa_check_logs(spa
);
4595 if (spa
->spa_missing_tvds
!= 0) {
4596 spa_load_note(spa
, "spa_check_logs failed "
4597 "so dropping the logs");
4599 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
4600 spa_load_failed(spa
, "spa_check_logs failed");
4601 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
,
4611 spa_ld_verify_pool_data(spa_t
*spa
)
4614 vdev_t
*rvd
= spa
->spa_root_vdev
;
4617 * We've successfully opened the pool, verify that we're ready
4618 * to start pushing transactions.
4620 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4621 error
= spa_load_verify(spa
);
4623 spa_load_failed(spa
, "spa_load_verify failed "
4624 "[error=%d]", error
);
4625 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
4634 spa_ld_claim_log_blocks(spa_t
*spa
)
4637 dsl_pool_t
*dp
= spa_get_dsl(spa
);
4640 * Claim log blocks that haven't been committed yet.
4641 * This must all happen in a single txg.
4642 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4643 * invoked from zil_claim_log_block()'s i/o done callback.
4644 * Price of rollback is that we abandon the log.
4646 spa
->spa_claiming
= B_TRUE
;
4648 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
4649 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
4650 zil_claim
, tx
, DS_FIND_CHILDREN
);
4653 spa
->spa_claiming
= B_FALSE
;
4655 spa_set_log_state(spa
, SPA_LOG_GOOD
);
4659 spa_ld_check_for_config_update(spa_t
*spa
, uint64_t config_cache_txg
,
4660 boolean_t update_config_cache
)
4662 vdev_t
*rvd
= spa
->spa_root_vdev
;
4663 int need_update
= B_FALSE
;
4666 * If the config cache is stale, or we have uninitialized
4667 * metaslabs (see spa_vdev_add()), then update the config.
4669 * If this is a verbatim import, trust the current
4670 * in-core spa_config and update the disk labels.
4672 if (update_config_cache
|| config_cache_txg
!= spa
->spa_config_txg
||
4673 spa
->spa_load_state
== SPA_LOAD_IMPORT
||
4674 spa
->spa_load_state
== SPA_LOAD_RECOVER
||
4675 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
4676 need_update
= B_TRUE
;
4678 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
4679 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
4680 need_update
= B_TRUE
;
4683 * Update the config cache asynchronously in case we're the
4684 * root pool, in which case the config cache isn't writable yet.
4687 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4691 spa_ld_prepare_for_reload(spa_t
*spa
)
4693 spa_mode_t mode
= spa
->spa_mode
;
4694 int async_suspended
= spa
->spa_async_suspended
;
4697 spa_deactivate(spa
);
4698 spa_activate(spa
, mode
);
4701 * We save the value of spa_async_suspended as it gets reset to 0 by
4702 * spa_unload(). We want to restore it back to the original value before
4703 * returning as we might be calling spa_async_resume() later.
4705 spa
->spa_async_suspended
= async_suspended
;
4709 spa_ld_read_checkpoint_txg(spa_t
*spa
)
4711 uberblock_t checkpoint
;
4714 ASSERT0(spa
->spa_checkpoint_txg
);
4715 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4717 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4718 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4719 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4721 if (error
== ENOENT
)
4727 ASSERT3U(checkpoint
.ub_txg
, !=, 0);
4728 ASSERT3U(checkpoint
.ub_checkpoint_txg
, !=, 0);
4729 ASSERT3U(checkpoint
.ub_timestamp
, !=, 0);
4730 spa
->spa_checkpoint_txg
= checkpoint
.ub_txg
;
4731 spa
->spa_checkpoint_info
.sci_timestamp
= checkpoint
.ub_timestamp
;
4737 spa_ld_mos_init(spa_t
*spa
, spa_import_type_t type
)
4741 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4742 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4745 * Never trust the config that is provided unless we are assembling
4746 * a pool following a split.
4747 * This means don't trust blkptrs and the vdev tree in general. This
4748 * also effectively puts the spa in read-only mode since
4749 * spa_writeable() checks for spa_trust_config to be true.
4750 * We will later load a trusted config from the MOS.
4752 if (type
!= SPA_IMPORT_ASSEMBLE
)
4753 spa
->spa_trust_config
= B_FALSE
;
4756 * Parse the config provided to create a vdev tree.
4758 error
= spa_ld_parse_config(spa
, type
);
4762 spa_import_progress_add(spa
);
4765 * Now that we have the vdev tree, try to open each vdev. This involves
4766 * opening the underlying physical device, retrieving its geometry and
4767 * probing the vdev with a dummy I/O. The state of each vdev will be set
4768 * based on the success of those operations. After this we'll be ready
4769 * to read from the vdevs.
4771 error
= spa_ld_open_vdevs(spa
);
4776 * Read the label of each vdev and make sure that the GUIDs stored
4777 * there match the GUIDs in the config provided.
4778 * If we're assembling a new pool that's been split off from an
4779 * existing pool, the labels haven't yet been updated so we skip
4780 * validation for now.
4782 if (type
!= SPA_IMPORT_ASSEMBLE
) {
4783 error
= spa_ld_validate_vdevs(spa
);
4789 * Read all vdev labels to find the best uberblock (i.e. latest,
4790 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
4791 * get the list of features required to read blkptrs in the MOS from
4792 * the vdev label with the best uberblock and verify that our version
4793 * of zfs supports them all.
4795 error
= spa_ld_select_uberblock(spa
, type
);
4800 * Pass that uberblock to the dsl_pool layer which will open the root
4801 * blkptr. This blkptr points to the latest version of the MOS and will
4802 * allow us to read its contents.
4804 error
= spa_ld_open_rootbp(spa
);
4812 spa_ld_checkpoint_rewind(spa_t
*spa
)
4814 uberblock_t checkpoint
;
4817 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4818 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4820 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4821 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4822 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4825 spa_load_failed(spa
, "unable to retrieve checkpointed "
4826 "uberblock from the MOS config [error=%d]", error
);
4828 if (error
== ENOENT
)
4829 error
= ZFS_ERR_NO_CHECKPOINT
;
4834 ASSERT3U(checkpoint
.ub_txg
, <, spa
->spa_uberblock
.ub_txg
);
4835 ASSERT3U(checkpoint
.ub_txg
, ==, checkpoint
.ub_checkpoint_txg
);
4838 * We need to update the txg and timestamp of the checkpointed
4839 * uberblock to be higher than the latest one. This ensures that
4840 * the checkpointed uberblock is selected if we were to close and
4841 * reopen the pool right after we've written it in the vdev labels.
4842 * (also see block comment in vdev_uberblock_compare)
4844 checkpoint
.ub_txg
= spa
->spa_uberblock
.ub_txg
+ 1;
4845 checkpoint
.ub_timestamp
= gethrestime_sec();
4848 * Set current uberblock to be the checkpointed uberblock.
4850 spa
->spa_uberblock
= checkpoint
;
4853 * If we are doing a normal rewind, then the pool is open for
4854 * writing and we sync the "updated" checkpointed uberblock to
4855 * disk. Once this is done, we've basically rewound the whole
4856 * pool and there is no way back.
4858 * There are cases when we don't want to attempt and sync the
4859 * checkpointed uberblock to disk because we are opening a
4860 * pool as read-only. Specifically, verifying the checkpointed
4861 * state with zdb, and importing the checkpointed state to get
4862 * a "preview" of its content.
4864 if (spa_writeable(spa
)) {
4865 vdev_t
*rvd
= spa
->spa_root_vdev
;
4867 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4868 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
4870 int children
= rvd
->vdev_children
;
4871 int c0
= random_in_range(children
);
4873 for (int c
= 0; c
< children
; c
++) {
4874 vdev_t
*vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
4876 /* Stop when revisiting the first vdev */
4877 if (c
> 0 && svd
[0] == vd
)
4880 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
4881 !vdev_is_concrete(vd
))
4884 svd
[svdcount
++] = vd
;
4885 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
4888 error
= vdev_config_sync(svd
, svdcount
, spa
->spa_first_txg
);
4890 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
4891 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4894 spa_load_failed(spa
, "failed to write checkpointed "
4895 "uberblock to the vdev labels [error=%d]", error
);
4904 spa_ld_mos_with_trusted_config(spa_t
*spa
, spa_import_type_t type
,
4905 boolean_t
*update_config_cache
)
4910 * Parse the config for pool, open and validate vdevs,
4911 * select an uberblock, and use that uberblock to open
4914 error
= spa_ld_mos_init(spa
, type
);
4919 * Retrieve the trusted config stored in the MOS and use it to create
4920 * a new, exact version of the vdev tree, then reopen all vdevs.
4922 error
= spa_ld_trusted_config(spa
, type
, B_FALSE
);
4923 if (error
== EAGAIN
) {
4924 if (update_config_cache
!= NULL
)
4925 *update_config_cache
= B_TRUE
;
4928 * Redo the loading process with the trusted config if it is
4929 * too different from the untrusted config.
4931 spa_ld_prepare_for_reload(spa
);
4932 spa_load_note(spa
, "RELOADING");
4933 error
= spa_ld_mos_init(spa
, type
);
4937 error
= spa_ld_trusted_config(spa
, type
, B_TRUE
);
4941 } else if (error
!= 0) {
4949 * Load an existing storage pool, using the config provided. This config
4950 * describes which vdevs are part of the pool and is later validated against
4951 * partial configs present in each vdev's label and an entire copy of the
4952 * config stored in the MOS.
4955 spa_load_impl(spa_t
*spa
, spa_import_type_t type
, const char **ereport
)
4958 boolean_t missing_feat_write
= B_FALSE
;
4959 boolean_t checkpoint_rewind
=
4960 (spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4961 boolean_t update_config_cache
= B_FALSE
;
4963 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4964 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4966 spa_load_note(spa
, "LOADING");
4968 error
= spa_ld_mos_with_trusted_config(spa
, type
, &update_config_cache
);
4973 * If we are rewinding to the checkpoint then we need to repeat
4974 * everything we've done so far in this function but this time
4975 * selecting the checkpointed uberblock and using that to open
4978 if (checkpoint_rewind
) {
4980 * If we are rewinding to the checkpoint update config cache
4983 update_config_cache
= B_TRUE
;
4986 * Extract the checkpointed uberblock from the current MOS
4987 * and use this as the pool's uberblock from now on. If the
4988 * pool is imported as writeable we also write the checkpoint
4989 * uberblock to the labels, making the rewind permanent.
4991 error
= spa_ld_checkpoint_rewind(spa
);
4996 * Redo the loading process again with the
4997 * checkpointed uberblock.
4999 spa_ld_prepare_for_reload(spa
);
5000 spa_load_note(spa
, "LOADING checkpointed uberblock");
5001 error
= spa_ld_mos_with_trusted_config(spa
, type
, NULL
);
5007 * Retrieve the checkpoint txg if the pool has a checkpoint.
5009 spa_import_progress_set_notes(spa
, "Loading checkpoint txg");
5010 error
= spa_ld_read_checkpoint_txg(spa
);
5015 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
5016 * from the pool and their contents were re-mapped to other vdevs. Note
5017 * that everything that we read before this step must have been
5018 * rewritten on concrete vdevs after the last device removal was
5019 * initiated. Otherwise we could be reading from indirect vdevs before
5020 * we have loaded their mappings.
5022 spa_import_progress_set_notes(spa
, "Loading indirect vdev metadata");
5023 error
= spa_ld_open_indirect_vdev_metadata(spa
);
5028 * Retrieve the full list of active features from the MOS and check if
5029 * they are all supported.
5031 spa_import_progress_set_notes(spa
, "Checking feature flags");
5032 error
= spa_ld_check_features(spa
, &missing_feat_write
);
5037 * Load several special directories from the MOS needed by the dsl_pool
5040 spa_import_progress_set_notes(spa
, "Loading special MOS directories");
5041 error
= spa_ld_load_special_directories(spa
);
5046 * Retrieve pool properties from the MOS.
5048 spa_import_progress_set_notes(spa
, "Loading properties");
5049 error
= spa_ld_get_props(spa
);
5054 * Retrieve the list of auxiliary devices - cache devices and spares -
5057 spa_import_progress_set_notes(spa
, "Loading AUX vdevs");
5058 error
= spa_ld_open_aux_vdevs(spa
, type
);
5063 * Load the metadata for all vdevs. Also check if unopenable devices
5064 * should be autoreplaced.
5066 spa_import_progress_set_notes(spa
, "Loading vdev metadata");
5067 error
= spa_ld_load_vdev_metadata(spa
);
5071 spa_import_progress_set_notes(spa
, "Loading dedup tables");
5072 error
= spa_ld_load_dedup_tables(spa
);
5076 spa_import_progress_set_notes(spa
, "Loading BRT");
5077 error
= spa_ld_load_brt(spa
);
5082 * Verify the logs now to make sure we don't have any unexpected errors
5083 * when we claim log blocks later.
5085 spa_import_progress_set_notes(spa
, "Verifying Log Devices");
5086 error
= spa_ld_verify_logs(spa
, type
, ereport
);
5090 if (missing_feat_write
) {
5091 ASSERT(spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
);
5094 * At this point, we know that we can open the pool in
5095 * read-only mode but not read-write mode. We now have enough
5096 * information and can return to userland.
5098 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
5103 * Traverse the last txgs to make sure the pool was left off in a safe
5104 * state. When performing an extreme rewind, we verify the whole pool,
5105 * which can take a very long time.
5107 spa_import_progress_set_notes(spa
, "Verifying pool data");
5108 error
= spa_ld_verify_pool_data(spa
);
5113 * Calculate the deflated space for the pool. This must be done before
5114 * we write anything to the pool because we'd need to update the space
5115 * accounting using the deflated sizes.
5117 spa_import_progress_set_notes(spa
, "Calculating deflated space");
5118 spa_update_dspace(spa
);
5121 * We have now retrieved all the information we needed to open the
5122 * pool. If we are importing the pool in read-write mode, a few
5123 * additional steps must be performed to finish the import.
5125 spa_import_progress_set_notes(spa
, "Starting import");
5126 if (spa_writeable(spa
) && (spa
->spa_load_state
== SPA_LOAD_RECOVER
||
5127 spa
->spa_load_max_txg
== UINT64_MAX
)) {
5128 uint64_t config_cache_txg
= spa
->spa_config_txg
;
5130 ASSERT(spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
);
5133 * Before we do any zio_write's, complete the raidz expansion
5134 * scratch space copying, if necessary.
5136 if (RRSS_GET_STATE(&spa
->spa_uberblock
) == RRSS_SCRATCH_VALID
)
5137 vdev_raidz_reflow_copy_scratch(spa
);
5140 * In case of a checkpoint rewind, log the original txg
5141 * of the checkpointed uberblock.
5143 if (checkpoint_rewind
) {
5144 spa_history_log_internal(spa
, "checkpoint rewind",
5145 NULL
, "rewound state to txg=%llu",
5146 (u_longlong_t
)spa
->spa_uberblock
.ub_checkpoint_txg
);
5149 spa_import_progress_set_notes(spa
, "Claiming ZIL blocks");
5151 * Traverse the ZIL and claim all blocks.
5153 spa_ld_claim_log_blocks(spa
);
5156 * Kick-off the syncing thread.
5158 spa
->spa_sync_on
= B_TRUE
;
5159 txg_sync_start(spa
->spa_dsl_pool
);
5160 mmp_thread_start(spa
);
5163 * Wait for all claims to sync. We sync up to the highest
5164 * claimed log block birth time so that claimed log blocks
5165 * don't appear to be from the future. spa_claim_max_txg
5166 * will have been set for us by ZIL traversal operations
5169 spa_import_progress_set_notes(spa
, "Syncing ZIL claims");
5170 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
5173 * Check if we need to request an update of the config. On the
5174 * next sync, we would update the config stored in vdev labels
5175 * and the cachefile (by default /etc/zfs/zpool.cache).
5177 spa_import_progress_set_notes(spa
, "Updating configs");
5178 spa_ld_check_for_config_update(spa
, config_cache_txg
,
5179 update_config_cache
);
5182 * Check if a rebuild was in progress and if so resume it.
5183 * Then check all DTLs to see if anything needs resilvering.
5184 * The resilver will be deferred if a rebuild was started.
5186 spa_import_progress_set_notes(spa
, "Starting resilvers");
5187 if (vdev_rebuild_active(spa
->spa_root_vdev
)) {
5188 vdev_rebuild_restart(spa
);
5189 } else if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
5190 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5191 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
5195 * Log the fact that we booted up (so that we can detect if
5196 * we rebooted in the middle of an operation).
5198 spa_history_log_version(spa
, "open", NULL
);
5200 spa_import_progress_set_notes(spa
,
5201 "Restarting device removals");
5202 spa_restart_removal(spa
);
5203 spa_spawn_aux_threads(spa
);
5206 * Delete any inconsistent datasets.
5209 * Since we may be issuing deletes for clones here,
5210 * we make sure to do so after we've spawned all the
5211 * auxiliary threads above (from which the livelist
5212 * deletion zthr is part of).
5214 spa_import_progress_set_notes(spa
,
5215 "Cleaning up inconsistent objsets");
5216 (void) dmu_objset_find(spa_name(spa
),
5217 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
5220 * Clean up any stale temporary dataset userrefs.
5222 spa_import_progress_set_notes(spa
,
5223 "Cleaning up temporary userrefs");
5224 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
5226 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5227 spa_import_progress_set_notes(spa
, "Restarting initialize");
5228 vdev_initialize_restart(spa
->spa_root_vdev
);
5229 spa_import_progress_set_notes(spa
, "Restarting TRIM");
5230 vdev_trim_restart(spa
->spa_root_vdev
);
5231 vdev_autotrim_restart(spa
);
5232 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5233 spa_import_progress_set_notes(spa
, "Finished importing");
5236 spa_import_progress_remove(spa_guid(spa
));
5237 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
5239 spa_load_note(spa
, "LOADED");
5245 spa_load_retry(spa_t
*spa
, spa_load_state_t state
)
5247 spa_mode_t mode
= spa
->spa_mode
;
5250 spa_deactivate(spa
);
5252 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
5254 spa_activate(spa
, mode
);
5255 spa_async_suspend(spa
);
5257 spa_load_note(spa
, "spa_load_retry: rewind, max txg: %llu",
5258 (u_longlong_t
)spa
->spa_load_max_txg
);
5260 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
));
5264 * If spa_load() fails this function will try loading prior txg's. If
5265 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
5266 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
5267 * function will not rewind the pool and will return the same error as
5271 spa_load_best(spa_t
*spa
, spa_load_state_t state
, uint64_t max_request
,
5274 nvlist_t
*loadinfo
= NULL
;
5275 nvlist_t
*config
= NULL
;
5276 int load_error
, rewind_error
;
5277 uint64_t safe_rewind_txg
;
5280 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
5281 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
5282 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5284 spa
->spa_load_max_txg
= max_request
;
5285 if (max_request
!= UINT64_MAX
)
5286 spa
->spa_extreme_rewind
= B_TRUE
;
5289 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
);
5290 if (load_error
== 0)
5292 if (load_error
== ZFS_ERR_NO_CHECKPOINT
) {
5294 * When attempting checkpoint-rewind on a pool with no
5295 * checkpoint, we should not attempt to load uberblocks
5296 * from previous txgs when spa_load fails.
5298 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
5299 spa_import_progress_remove(spa_guid(spa
));
5300 return (load_error
);
5303 if (spa
->spa_root_vdev
!= NULL
)
5304 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5306 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
5307 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
5309 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
5310 nvlist_free(config
);
5311 spa_import_progress_remove(spa_guid(spa
));
5312 return (load_error
);
5315 if (state
== SPA_LOAD_RECOVER
) {
5316 /* Price of rolling back is discarding txgs, including log */
5317 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5320 * If we aren't rolling back save the load info from our first
5321 * import attempt so that we can restore it after attempting
5324 loadinfo
= spa
->spa_load_info
;
5325 spa
->spa_load_info
= fnvlist_alloc();
5328 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
5329 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
5330 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
5331 TXG_INITIAL
: safe_rewind_txg
;
5334 * Continue as long as we're finding errors, we're still within
5335 * the acceptable rewind range, and we're still finding uberblocks
5337 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
5338 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
5339 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
5340 spa
->spa_extreme_rewind
= B_TRUE
;
5341 rewind_error
= spa_load_retry(spa
, state
);
5344 spa
->spa_extreme_rewind
= B_FALSE
;
5345 spa
->spa_load_max_txg
= UINT64_MAX
;
5347 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
5348 spa_config_set(spa
, config
);
5350 nvlist_free(config
);
5352 if (state
== SPA_LOAD_RECOVER
) {
5353 ASSERT3P(loadinfo
, ==, NULL
);
5354 spa_import_progress_remove(spa_guid(spa
));
5355 return (rewind_error
);
5357 /* Store the rewind info as part of the initial load info */
5358 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
5359 spa
->spa_load_info
);
5361 /* Restore the initial load info */
5362 fnvlist_free(spa
->spa_load_info
);
5363 spa
->spa_load_info
= loadinfo
;
5365 spa_import_progress_remove(spa_guid(spa
));
5366 return (load_error
);
5373 * The import case is identical to an open except that the configuration is sent
5374 * down from userland, instead of grabbed from the configuration cache. For the
5375 * case of an open, the pool configuration will exist in the
5376 * POOL_STATE_UNINITIALIZED state.
5378 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5379 * the same time open the pool, without having to keep around the spa_t in some
5383 spa_open_common(const char *pool
, spa_t
**spapp
, const void *tag
,
5384 nvlist_t
*nvpolicy
, nvlist_t
**config
)
5387 spa_load_state_t state
= SPA_LOAD_OPEN
;
5389 int locked
= B_FALSE
;
5390 int firstopen
= B_FALSE
;
5395 * As disgusting as this is, we need to support recursive calls to this
5396 * function because dsl_dir_open() is called during spa_load(), and ends
5397 * up calling spa_open() again. The real fix is to figure out how to
5398 * avoid dsl_dir_open() calling this in the first place.
5400 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
5401 mutex_enter(&spa_namespace_lock
);
5405 if ((spa
= spa_lookup(pool
)) == NULL
) {
5407 mutex_exit(&spa_namespace_lock
);
5408 return (SET_ERROR(ENOENT
));
5411 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
5412 zpool_load_policy_t policy
;
5416 zpool_get_load_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
5418 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
5419 state
= SPA_LOAD_RECOVER
;
5421 spa_activate(spa
, spa_mode_global
);
5423 if (state
!= SPA_LOAD_RECOVER
)
5424 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
5425 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
5427 zfs_dbgmsg("spa_open_common: opening %s", pool
);
5428 error
= spa_load_best(spa
, state
, policy
.zlp_txg
,
5431 if (error
== EBADF
) {
5433 * If vdev_validate() returns failure (indicated by
5434 * EBADF), it indicates that one of the vdevs indicates
5435 * that the pool has been exported or destroyed. If
5436 * this is the case, the config cache is out of sync and
5437 * we should remove the pool from the namespace.
5440 spa_deactivate(spa
);
5441 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
, B_FALSE
);
5444 mutex_exit(&spa_namespace_lock
);
5445 return (SET_ERROR(ENOENT
));
5450 * We can't open the pool, but we still have useful
5451 * information: the state of each vdev after the
5452 * attempted vdev_open(). Return this to the user.
5454 if (config
!= NULL
&& spa
->spa_config
) {
5455 *config
= fnvlist_dup(spa
->spa_config
);
5456 fnvlist_add_nvlist(*config
,
5457 ZPOOL_CONFIG_LOAD_INFO
,
5458 spa
->spa_load_info
);
5461 spa_deactivate(spa
);
5462 spa
->spa_last_open_failed
= error
;
5464 mutex_exit(&spa_namespace_lock
);
5470 spa_open_ref(spa
, tag
);
5473 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5476 * If we've recovered the pool, pass back any information we
5477 * gathered while doing the load.
5479 if (state
== SPA_LOAD_RECOVER
&& config
!= NULL
) {
5480 fnvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
5481 spa
->spa_load_info
);
5485 spa
->spa_last_open_failed
= 0;
5486 spa
->spa_last_ubsync_txg
= 0;
5487 spa
->spa_load_txg
= 0;
5488 mutex_exit(&spa_namespace_lock
);
5492 zvol_create_minors_recursive(spa_name(spa
));
5500 spa_open_rewind(const char *name
, spa_t
**spapp
, const void *tag
,
5501 nvlist_t
*policy
, nvlist_t
**config
)
5503 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
5507 spa_open(const char *name
, spa_t
**spapp
, const void *tag
)
5509 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
5513 * Lookup the given spa_t, incrementing the inject count in the process,
5514 * preventing it from being exported or destroyed.
5517 spa_inject_addref(char *name
)
5521 mutex_enter(&spa_namespace_lock
);
5522 if ((spa
= spa_lookup(name
)) == NULL
) {
5523 mutex_exit(&spa_namespace_lock
);
5526 spa
->spa_inject_ref
++;
5527 mutex_exit(&spa_namespace_lock
);
5533 spa_inject_delref(spa_t
*spa
)
5535 mutex_enter(&spa_namespace_lock
);
5536 spa
->spa_inject_ref
--;
5537 mutex_exit(&spa_namespace_lock
);
5541 * Add spares device information to the nvlist.
5544 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
5554 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5556 if (spa
->spa_spares
.sav_count
== 0)
5559 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5560 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5561 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
));
5563 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5564 (const nvlist_t
* const *)spares
, nspares
);
5565 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5566 &spares
, &nspares
));
5569 * Go through and find any spares which have since been
5570 * repurposed as an active spare. If this is the case, update
5571 * their status appropriately.
5573 for (i
= 0; i
< nspares
; i
++) {
5574 guid
= fnvlist_lookup_uint64(spares
[i
],
5576 VERIFY0(nvlist_lookup_uint64_array(spares
[i
],
5577 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
));
5578 if (spa_spare_exists(guid
, &pool
, NULL
) &&
5580 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
5581 vs
->vs_aux
= VDEV_AUX_SPARED
;
5584 spa
->spa_spares
.sav_vdevs
[i
]->vdev_state
;
5591 * Add l2cache device information to the nvlist, including vdev stats.
5594 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
5597 uint_t i
, j
, nl2cache
;
5604 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5606 if (spa
->spa_l2cache
.sav_count
== 0)
5609 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5610 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5611 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
));
5612 if (nl2cache
!= 0) {
5613 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5614 (const nvlist_t
* const *)l2cache
, nl2cache
);
5615 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5616 &l2cache
, &nl2cache
));
5619 * Update level 2 cache device stats.
5622 for (i
= 0; i
< nl2cache
; i
++) {
5623 guid
= fnvlist_lookup_uint64(l2cache
[i
],
5627 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
5629 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
5630 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
5636 VERIFY0(nvlist_lookup_uint64_array(l2cache
[i
],
5637 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
));
5638 vdev_get_stats(vd
, vs
);
5639 vdev_config_generate_stats(vd
, l2cache
[i
]);
5646 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
5651 if (spa
->spa_feat_for_read_obj
!= 0) {
5652 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5653 spa
->spa_feat_for_read_obj
);
5654 zap_cursor_retrieve(&zc
, &za
) == 0;
5655 zap_cursor_advance(&zc
)) {
5656 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5657 za
.za_num_integers
== 1);
5658 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5659 za
.za_first_integer
));
5661 zap_cursor_fini(&zc
);
5664 if (spa
->spa_feat_for_write_obj
!= 0) {
5665 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5666 spa
->spa_feat_for_write_obj
);
5667 zap_cursor_retrieve(&zc
, &za
) == 0;
5668 zap_cursor_advance(&zc
)) {
5669 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5670 za
.za_num_integers
== 1);
5671 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5672 za
.za_first_integer
));
5674 zap_cursor_fini(&zc
);
5679 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
5683 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5684 zfeature_info_t feature
= spa_feature_table
[i
];
5687 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
5690 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
5695 * Store a list of pool features and their reference counts in the
5698 * The first time this is called on a spa, allocate a new nvlist, fetch
5699 * the pool features and reference counts from disk, then save the list
5700 * in the spa. In subsequent calls on the same spa use the saved nvlist
5701 * and refresh its values from the cached reference counts. This
5702 * ensures we don't block here on I/O on a suspended pool so 'zpool
5703 * clear' can resume the pool.
5706 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
5710 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5712 mutex_enter(&spa
->spa_feat_stats_lock
);
5713 features
= spa
->spa_feat_stats
;
5715 if (features
!= NULL
) {
5716 spa_feature_stats_from_cache(spa
, features
);
5718 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
5719 spa
->spa_feat_stats
= features
;
5720 spa_feature_stats_from_disk(spa
, features
);
5723 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
5726 mutex_exit(&spa
->spa_feat_stats_lock
);
5730 spa_get_stats(const char *name
, nvlist_t
**config
,
5731 char *altroot
, size_t buflen
)
5737 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
5741 * This still leaves a window of inconsistency where the spares
5742 * or l2cache devices could change and the config would be
5743 * self-inconsistent.
5745 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5747 if (*config
!= NULL
) {
5748 uint64_t loadtimes
[2];
5750 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
5751 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
5752 fnvlist_add_uint64_array(*config
,
5753 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2);
5755 fnvlist_add_uint64(*config
,
5756 ZPOOL_CONFIG_ERRCOUNT
,
5757 spa_approx_errlog_size(spa
));
5759 if (spa_suspended(spa
)) {
5760 fnvlist_add_uint64(*config
,
5761 ZPOOL_CONFIG_SUSPENDED
,
5763 fnvlist_add_uint64(*config
,
5764 ZPOOL_CONFIG_SUSPENDED_REASON
,
5765 spa
->spa_suspended
);
5768 spa_add_spares(spa
, *config
);
5769 spa_add_l2cache(spa
, *config
);
5770 spa_add_feature_stats(spa
, *config
);
5775 * We want to get the alternate root even for faulted pools, so we cheat
5776 * and call spa_lookup() directly.
5780 mutex_enter(&spa_namespace_lock
);
5781 spa
= spa_lookup(name
);
5783 spa_altroot(spa
, altroot
, buflen
);
5787 mutex_exit(&spa_namespace_lock
);
5789 spa_altroot(spa
, altroot
, buflen
);
5794 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5795 spa_close(spa
, FTAG
);
5802 * Validate that the auxiliary device array is well formed. We must have an
5803 * array of nvlists, each which describes a valid leaf vdev. If this is an
5804 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
5805 * specified, as long as they are well-formed.
5808 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
5809 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
5810 vdev_labeltype_t label
)
5817 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5820 * It's acceptable to have no devs specified.
5822 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
5826 return (SET_ERROR(EINVAL
));
5829 * Make sure the pool is formatted with a version that supports this
5832 if (spa_version(spa
) < version
)
5833 return (SET_ERROR(ENOTSUP
));
5836 * Set the pending device list so we correctly handle device in-use
5839 sav
->sav_pending
= dev
;
5840 sav
->sav_npending
= ndev
;
5842 for (i
= 0; i
< ndev
; i
++) {
5843 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
5847 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5849 error
= SET_ERROR(EINVAL
);
5855 if ((error
= vdev_open(vd
)) == 0 &&
5856 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
5857 fnvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
5864 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
5871 sav
->sav_pending
= NULL
;
5872 sav
->sav_npending
= 0;
5877 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
5881 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5883 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5884 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
5885 VDEV_LABEL_SPARE
)) != 0) {
5889 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5890 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
5891 VDEV_LABEL_L2CACHE
));
5895 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
5900 if (sav
->sav_config
!= NULL
) {
5906 * Generate new dev list by concatenating with the
5909 VERIFY0(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
5910 &olddevs
, &oldndevs
));
5912 newdevs
= kmem_alloc(sizeof (void *) *
5913 (ndevs
+ oldndevs
), KM_SLEEP
);
5914 for (i
= 0; i
< oldndevs
; i
++)
5915 newdevs
[i
] = fnvlist_dup(olddevs
[i
]);
5916 for (i
= 0; i
< ndevs
; i
++)
5917 newdevs
[i
+ oldndevs
] = fnvlist_dup(devs
[i
]);
5919 fnvlist_remove(sav
->sav_config
, config
);
5921 fnvlist_add_nvlist_array(sav
->sav_config
, config
,
5922 (const nvlist_t
* const *)newdevs
, ndevs
+ oldndevs
);
5923 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
5924 nvlist_free(newdevs
[i
]);
5925 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
5928 * Generate a new dev list.
5930 sav
->sav_config
= fnvlist_alloc();
5931 fnvlist_add_nvlist_array(sav
->sav_config
, config
,
5932 (const nvlist_t
* const *)devs
, ndevs
);
5937 * Stop and drop level 2 ARC devices
5940 spa_l2cache_drop(spa_t
*spa
)
5944 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
5946 for (i
= 0; i
< sav
->sav_count
; i
++) {
5949 vd
= sav
->sav_vdevs
[i
];
5952 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
5953 pool
!= 0ULL && l2arc_vdev_present(vd
))
5954 l2arc_remove_vdev(vd
);
5959 * Verify encryption parameters for spa creation. If we are encrypting, we must
5960 * have the encryption feature flag enabled.
5963 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
5964 boolean_t has_encryption
)
5966 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
5967 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
5969 return (SET_ERROR(ENOTSUP
));
5971 return (dmu_objset_create_crypt_check(NULL
, dcp
, NULL
));
5978 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
5979 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
5982 const char *altroot
= NULL
;
5987 uint64_t txg
= TXG_INITIAL
;
5988 nvlist_t
**spares
, **l2cache
;
5989 uint_t nspares
, nl2cache
;
5990 uint64_t version
, obj
, ndraid
= 0;
5991 boolean_t has_features
;
5992 boolean_t has_encryption
;
5993 boolean_t has_allocclass
;
5995 const char *feat_name
;
5996 const char *poolname
;
5999 if (props
== NULL
||
6000 nvlist_lookup_string(props
, "tname", &poolname
) != 0)
6001 poolname
= (char *)pool
;
6004 * If this pool already exists, return failure.
6006 mutex_enter(&spa_namespace_lock
);
6007 if (spa_lookup(poolname
) != NULL
) {
6008 mutex_exit(&spa_namespace_lock
);
6009 return (SET_ERROR(EEXIST
));
6013 * Allocate a new spa_t structure.
6015 nvl
= fnvlist_alloc();
6016 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
6017 (void) nvlist_lookup_string(props
,
6018 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
6019 spa
= spa_add(poolname
, nvl
, altroot
);
6021 spa_activate(spa
, spa_mode_global
);
6023 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
6024 spa_deactivate(spa
);
6026 mutex_exit(&spa_namespace_lock
);
6031 * Temporary pool names should never be written to disk.
6033 if (poolname
!= pool
)
6034 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
6036 has_features
= B_FALSE
;
6037 has_encryption
= B_FALSE
;
6038 has_allocclass
= B_FALSE
;
6039 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
6040 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
6041 if (zpool_prop_feature(nvpair_name(elem
))) {
6042 has_features
= B_TRUE
;
6044 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
6045 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
6046 if (feat
== SPA_FEATURE_ENCRYPTION
)
6047 has_encryption
= B_TRUE
;
6048 if (feat
== SPA_FEATURE_ALLOCATION_CLASSES
)
6049 has_allocclass
= B_TRUE
;
6053 /* verify encryption params, if they were provided */
6055 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
6057 spa_deactivate(spa
);
6059 mutex_exit(&spa_namespace_lock
);
6063 if (!has_allocclass
&& zfs_special_devs(nvroot
, NULL
)) {
6064 spa_deactivate(spa
);
6066 mutex_exit(&spa_namespace_lock
);
6070 if (has_features
|| nvlist_lookup_uint64(props
,
6071 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
6072 version
= SPA_VERSION
;
6074 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6076 spa
->spa_first_txg
= txg
;
6077 spa
->spa_uberblock
.ub_txg
= txg
- 1;
6078 spa
->spa_uberblock
.ub_version
= version
;
6079 spa
->spa_ubsync
= spa
->spa_uberblock
;
6080 spa
->spa_load_state
= SPA_LOAD_CREATE
;
6081 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
6082 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
6083 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
6084 spa
->spa_indirect_vdevs_loaded
= B_TRUE
;
6087 * Create "The Godfather" zio to hold all async IOs
6089 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
6091 for (int i
= 0; i
< max_ncpus
; i
++) {
6092 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
6093 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
6094 ZIO_FLAG_GODFATHER
);
6098 * Create the root vdev.
6100 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6102 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
6104 ASSERT(error
!= 0 || rvd
!= NULL
);
6105 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
6107 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
6108 error
= SET_ERROR(EINVAL
);
6111 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
6112 (error
= vdev_draid_spare_create(nvroot
, rvd
, &ndraid
, 0)) == 0 &&
6113 (error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) == 0) {
6115 * instantiate the metaslab groups (this will dirty the vdevs)
6116 * we can no longer error exit past this point
6118 for (int c
= 0; error
== 0 && c
< rvd
->vdev_children
; c
++) {
6119 vdev_t
*vd
= rvd
->vdev_child
[c
];
6121 vdev_metaslab_set_size(vd
);
6122 vdev_expand(vd
, txg
);
6126 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6130 spa_deactivate(spa
);
6132 mutex_exit(&spa_namespace_lock
);
6137 * Get the list of spares, if specified.
6139 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
6140 &spares
, &nspares
) == 0) {
6141 spa
->spa_spares
.sav_config
= fnvlist_alloc();
6142 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
6143 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
6145 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6146 spa_load_spares(spa
);
6147 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6148 spa
->spa_spares
.sav_sync
= B_TRUE
;
6152 * Get the list of level 2 cache devices, if specified.
6154 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
6155 &l2cache
, &nl2cache
) == 0) {
6156 VERIFY0(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
6157 NV_UNIQUE_NAME
, KM_SLEEP
));
6158 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
6159 ZPOOL_CONFIG_L2CACHE
, (const nvlist_t
* const *)l2cache
,
6161 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6162 spa_load_l2cache(spa
);
6163 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6164 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6167 spa
->spa_is_initializing
= B_TRUE
;
6168 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
6169 spa
->spa_is_initializing
= B_FALSE
;
6172 * Create DDTs (dedup tables).
6176 * Create BRT table and BRT table object.
6180 spa_update_dspace(spa
);
6182 tx
= dmu_tx_create_assigned(dp
, txg
);
6185 * Create the pool's history object.
6187 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
6188 spa_history_create_obj(spa
, tx
);
6190 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
6191 spa_history_log_version(spa
, "create", tx
);
6194 * Create the pool config object.
6196 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6197 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
6198 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
6200 if (zap_add(spa
->spa_meta_objset
,
6201 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
6202 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
6203 cmn_err(CE_PANIC
, "failed to add pool config");
6206 if (zap_add(spa
->spa_meta_objset
,
6207 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
6208 sizeof (uint64_t), 1, &version
, tx
) != 0) {
6209 cmn_err(CE_PANIC
, "failed to add pool version");
6212 /* Newly created pools with the right version are always deflated. */
6213 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6214 spa
->spa_deflate
= TRUE
;
6215 if (zap_add(spa
->spa_meta_objset
,
6216 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6217 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
6218 cmn_err(CE_PANIC
, "failed to add deflate");
6223 * Create the deferred-free bpobj. Turn off compression
6224 * because sync-to-convergence takes longer if the blocksize
6227 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
6228 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
6229 ZIO_COMPRESS_OFF
, tx
);
6230 if (zap_add(spa
->spa_meta_objset
,
6231 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
6232 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
6233 cmn_err(CE_PANIC
, "failed to add bpobj");
6235 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
6236 spa
->spa_meta_objset
, obj
));
6239 * Generate some random noise for salted checksums to operate on.
6241 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
6242 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
6245 * Set pool properties.
6247 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
6248 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
6249 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
6250 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
6251 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
6252 spa
->spa_autotrim
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM
);
6254 if (props
!= NULL
) {
6255 spa_configfile_set(spa
, props
, B_FALSE
);
6256 spa_sync_props(props
, tx
);
6259 for (int i
= 0; i
< ndraid
; i
++)
6260 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
6264 spa
->spa_sync_on
= B_TRUE
;
6266 mmp_thread_start(spa
);
6267 txg_wait_synced(dp
, txg
);
6269 spa_spawn_aux_threads(spa
);
6271 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
, B_TRUE
);
6274 * Don't count references from objsets that are already closed
6275 * and are making their way through the eviction process.
6277 spa_evicting_os_wait(spa
);
6278 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
6279 spa
->spa_load_state
= SPA_LOAD_NONE
;
6283 mutex_exit(&spa_namespace_lock
);
6289 * Import a non-root pool into the system.
6292 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
6295 const char *altroot
= NULL
;
6296 spa_load_state_t state
= SPA_LOAD_IMPORT
;
6297 zpool_load_policy_t policy
;
6298 spa_mode_t mode
= spa_mode_global
;
6299 uint64_t readonly
= B_FALSE
;
6302 nvlist_t
**spares
, **l2cache
;
6303 uint_t nspares
, nl2cache
;
6306 * If a pool with this name exists, return failure.
6308 mutex_enter(&spa_namespace_lock
);
6309 if (spa_lookup(pool
) != NULL
) {
6310 mutex_exit(&spa_namespace_lock
);
6311 return (SET_ERROR(EEXIST
));
6315 * Create and initialize the spa structure.
6317 (void) nvlist_lookup_string(props
,
6318 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
6319 (void) nvlist_lookup_uint64(props
,
6320 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
6322 mode
= SPA_MODE_READ
;
6323 spa
= spa_add(pool
, config
, altroot
);
6324 spa
->spa_import_flags
= flags
;
6327 * Verbatim import - Take a pool and insert it into the namespace
6328 * as if it had been loaded at boot.
6330 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
6332 spa_configfile_set(spa
, props
, B_FALSE
);
6334 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
, B_FALSE
);
6335 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6336 zfs_dbgmsg("spa_import: verbatim import of %s", pool
);
6337 mutex_exit(&spa_namespace_lock
);
6341 spa_activate(spa
, mode
);
6344 * Don't start async tasks until we know everything is healthy.
6346 spa_async_suspend(spa
);
6348 zpool_get_load_policy(config
, &policy
);
6349 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
6350 state
= SPA_LOAD_RECOVER
;
6352 spa
->spa_config_source
= SPA_CONFIG_SRC_TRYIMPORT
;
6354 if (state
!= SPA_LOAD_RECOVER
) {
6355 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
6356 zfs_dbgmsg("spa_import: importing %s", pool
);
6358 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
6359 "(RECOVERY MODE)", pool
, (longlong_t
)policy
.zlp_txg
);
6361 error
= spa_load_best(spa
, state
, policy
.zlp_txg
, policy
.zlp_rewind
);
6364 * Propagate anything learned while loading the pool and pass it
6365 * back to caller (i.e. rewind info, missing devices, etc).
6367 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
, spa
->spa_load_info
);
6369 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6371 * Toss any existing sparelist, as it doesn't have any validity
6372 * anymore, and conflicts with spa_has_spare().
6374 if (spa
->spa_spares
.sav_config
) {
6375 nvlist_free(spa
->spa_spares
.sav_config
);
6376 spa
->spa_spares
.sav_config
= NULL
;
6377 spa_load_spares(spa
);
6379 if (spa
->spa_l2cache
.sav_config
) {
6380 nvlist_free(spa
->spa_l2cache
.sav_config
);
6381 spa
->spa_l2cache
.sav_config
= NULL
;
6382 spa_load_l2cache(spa
);
6385 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
6386 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6389 spa_configfile_set(spa
, props
, B_FALSE
);
6391 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
6392 (error
= spa_prop_set(spa
, props
)))) {
6394 spa_deactivate(spa
);
6396 mutex_exit(&spa_namespace_lock
);
6400 spa_async_resume(spa
);
6403 * Override any spares and level 2 cache devices as specified by
6404 * the user, as these may have correct device names/devids, etc.
6406 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
6407 &spares
, &nspares
) == 0) {
6408 if (spa
->spa_spares
.sav_config
)
6409 fnvlist_remove(spa
->spa_spares
.sav_config
,
6410 ZPOOL_CONFIG_SPARES
);
6412 spa
->spa_spares
.sav_config
= fnvlist_alloc();
6413 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
6414 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
6416 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6417 spa_load_spares(spa
);
6418 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6419 spa
->spa_spares
.sav_sync
= B_TRUE
;
6421 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
6422 &l2cache
, &nl2cache
) == 0) {
6423 if (spa
->spa_l2cache
.sav_config
)
6424 fnvlist_remove(spa
->spa_l2cache
.sav_config
,
6425 ZPOOL_CONFIG_L2CACHE
);
6427 spa
->spa_l2cache
.sav_config
= fnvlist_alloc();
6428 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
6429 ZPOOL_CONFIG_L2CACHE
, (const nvlist_t
* const *)l2cache
,
6431 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6432 spa_load_l2cache(spa
);
6433 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6434 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6438 * Check for any removed devices.
6440 if (spa
->spa_autoreplace
) {
6441 spa_aux_check_removed(&spa
->spa_spares
);
6442 spa_aux_check_removed(&spa
->spa_l2cache
);
6445 if (spa_writeable(spa
)) {
6447 * Update the config cache to include the newly-imported pool.
6449 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6453 * It's possible that the pool was expanded while it was exported.
6454 * We kick off an async task to handle this for us.
6456 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
6458 spa_history_log_version(spa
, "import", NULL
);
6460 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6462 mutex_exit(&spa_namespace_lock
);
6464 zvol_create_minors_recursive(pool
);
6472 spa_tryimport(nvlist_t
*tryconfig
)
6474 nvlist_t
*config
= NULL
;
6475 const char *poolname
, *cachefile
;
6479 zpool_load_policy_t policy
;
6481 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
6484 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
6488 * Create and initialize the spa structure.
6490 mutex_enter(&spa_namespace_lock
);
6491 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
6492 spa_activate(spa
, SPA_MODE_READ
);
6495 * Rewind pool if a max txg was provided.
6497 zpool_get_load_policy(spa
->spa_config
, &policy
);
6498 if (policy
.zlp_txg
!= UINT64_MAX
) {
6499 spa
->spa_load_max_txg
= policy
.zlp_txg
;
6500 spa
->spa_extreme_rewind
= B_TRUE
;
6501 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6502 poolname
, (longlong_t
)policy
.zlp_txg
);
6504 zfs_dbgmsg("spa_tryimport: importing %s", poolname
);
6507 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_CACHEFILE
, &cachefile
)
6509 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile
);
6510 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
6512 spa
->spa_config_source
= SPA_CONFIG_SRC_SCAN
;
6516 * spa_import() relies on a pool config fetched by spa_try_import()
6517 * for spare/cache devices. Import flags are not passed to
6518 * spa_tryimport(), which makes it return early due to a missing log
6519 * device and missing retrieving the cache device and spare eventually.
6520 * Passing ZFS_IMPORT_MISSING_LOG to spa_tryimport() makes it fetch
6521 * the correct configuration regardless of the missing log device.
6523 spa
->spa_import_flags
|= ZFS_IMPORT_MISSING_LOG
;
6525 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
);
6528 * If 'tryconfig' was at least parsable, return the current config.
6530 if (spa
->spa_root_vdev
!= NULL
) {
6531 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
6532 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, poolname
);
6533 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, state
);
6534 fnvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
6535 spa
->spa_uberblock
.ub_timestamp
);
6536 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
6537 spa
->spa_load_info
);
6538 fnvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
6542 * If the bootfs property exists on this pool then we
6543 * copy it out so that external consumers can tell which
6544 * pools are bootable.
6546 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
6547 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6550 * We have to play games with the name since the
6551 * pool was opened as TRYIMPORT_NAME.
6553 if (dsl_dsobj_to_dsname(spa_name(spa
),
6554 spa
->spa_bootfs
, tmpname
) == 0) {
6558 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6560 cp
= strchr(tmpname
, '/');
6562 (void) strlcpy(dsname
, tmpname
,
6565 (void) snprintf(dsname
, MAXPATHLEN
,
6566 "%s/%s", poolname
, ++cp
);
6568 fnvlist_add_string(config
, ZPOOL_CONFIG_BOOTFS
,
6570 kmem_free(dsname
, MAXPATHLEN
);
6572 kmem_free(tmpname
, MAXPATHLEN
);
6576 * Add the list of hot spares and level 2 cache devices.
6578 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6579 spa_add_spares(spa
, config
);
6580 spa_add_l2cache(spa
, config
);
6581 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6585 spa_deactivate(spa
);
6587 mutex_exit(&spa_namespace_lock
);
6593 * Pool export/destroy
6595 * The act of destroying or exporting a pool is very simple. We make sure there
6596 * is no more pending I/O and any references to the pool are gone. Then, we
6597 * update the pool state and sync all the labels to disk, removing the
6598 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6599 * we don't sync the labels or remove the configuration cache.
6602 spa_export_common(const char *pool
, int new_state
, nvlist_t
**oldconfig
,
6603 boolean_t force
, boolean_t hardforce
)
6611 if (!(spa_mode_global
& SPA_MODE_WRITE
))
6612 return (SET_ERROR(EROFS
));
6614 mutex_enter(&spa_namespace_lock
);
6615 if ((spa
= spa_lookup(pool
)) == NULL
) {
6616 mutex_exit(&spa_namespace_lock
);
6617 return (SET_ERROR(ENOENT
));
6620 if (spa
->spa_is_exporting
) {
6621 /* the pool is being exported by another thread */
6622 mutex_exit(&spa_namespace_lock
);
6623 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS
));
6625 spa
->spa_is_exporting
= B_TRUE
;
6628 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6629 * reacquire the namespace lock, and see if we can export.
6631 spa_open_ref(spa
, FTAG
);
6632 mutex_exit(&spa_namespace_lock
);
6633 spa_async_suspend(spa
);
6634 if (spa
->spa_zvol_taskq
) {
6635 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
6636 taskq_wait(spa
->spa_zvol_taskq
);
6638 mutex_enter(&spa_namespace_lock
);
6639 spa_close(spa
, FTAG
);
6641 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
6644 * The pool will be in core if it's openable, in which case we can
6645 * modify its state. Objsets may be open only because they're dirty,
6646 * so we have to force it to sync before checking spa_refcnt.
6648 if (spa
->spa_sync_on
) {
6649 txg_wait_synced(spa
->spa_dsl_pool
, 0);
6650 spa_evicting_os_wait(spa
);
6654 * A pool cannot be exported or destroyed if there are active
6655 * references. If we are resetting a pool, allow references by
6656 * fault injection handlers.
6658 if (!spa_refcount_zero(spa
) || (spa
->spa_inject_ref
!= 0)) {
6659 error
= SET_ERROR(EBUSY
);
6663 if (spa
->spa_sync_on
) {
6664 vdev_t
*rvd
= spa
->spa_root_vdev
;
6666 * A pool cannot be exported if it has an active shared spare.
6667 * This is to prevent other pools stealing the active spare
6668 * from an exported pool. At user's own will, such pool can
6669 * be forcedly exported.
6671 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
6672 spa_has_active_shared_spare(spa
)) {
6673 error
= SET_ERROR(EXDEV
);
6678 * We're about to export or destroy this pool. Make sure
6679 * we stop all initialization and trim activity here before
6680 * we set the spa_final_txg. This will ensure that all
6681 * dirty data resulting from the initialization is
6682 * committed to disk before we unload the pool.
6684 vdev_initialize_stop_all(rvd
, VDEV_INITIALIZE_ACTIVE
);
6685 vdev_trim_stop_all(rvd
, VDEV_TRIM_ACTIVE
);
6686 vdev_autotrim_stop_all(spa
);
6687 vdev_rebuild_stop_all(spa
);
6690 * We want this to be reflected on every label,
6691 * so mark them all dirty. spa_unload() will do the
6692 * final sync that pushes these changes out.
6694 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
6695 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6696 spa
->spa_state
= new_state
;
6697 vdev_config_dirty(rvd
);
6698 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6702 * If the log space map feature is enabled and the pool is
6703 * getting exported (but not destroyed), we want to spend some
6704 * time flushing as many metaslabs as we can in an attempt to
6705 * destroy log space maps and save import time. This has to be
6706 * done before we set the spa_final_txg, otherwise
6707 * spa_sync() -> spa_flush_metaslabs() may dirty the final TXGs.
6708 * spa_should_flush_logs_on_unload() should be called after
6709 * spa_state has been set to the new_state.
6711 if (spa_should_flush_logs_on_unload(spa
))
6712 spa_unload_log_sm_flush_all(spa
);
6714 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
6715 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6716 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
6718 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6725 if (new_state
== POOL_STATE_DESTROYED
)
6726 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
6727 else if (new_state
== POOL_STATE_EXPORTED
)
6728 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
6730 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6732 spa_deactivate(spa
);
6735 if (oldconfig
&& spa
->spa_config
)
6736 *oldconfig
= fnvlist_dup(spa
->spa_config
);
6738 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
6740 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
, B_FALSE
);
6744 * If spa_remove() is not called for this spa_t and
6745 * there is any possibility that it can be reused,
6746 * we make sure to reset the exporting flag.
6748 spa
->spa_is_exporting
= B_FALSE
;
6751 mutex_exit(&spa_namespace_lock
);
6755 spa
->spa_is_exporting
= B_FALSE
;
6756 spa_async_resume(spa
);
6757 mutex_exit(&spa_namespace_lock
);
6762 * Destroy a storage pool.
6765 spa_destroy(const char *pool
)
6767 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
6772 * Export a storage pool.
6775 spa_export(const char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
6776 boolean_t hardforce
)
6778 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
6783 * Similar to spa_export(), this unloads the spa_t without actually removing it
6784 * from the namespace in any way.
6787 spa_reset(const char *pool
)
6789 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
6794 * ==========================================================================
6795 * Device manipulation
6796 * ==========================================================================
6800 * This is called as a synctask to increment the draid feature flag
6803 spa_draid_feature_incr(void *arg
, dmu_tx_t
*tx
)
6805 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6806 int draid
= (int)(uintptr_t)arg
;
6808 for (int c
= 0; c
< draid
; c
++)
6809 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
6813 * Add a device to a storage pool.
6816 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
6818 uint64_t txg
, ndraid
= 0;
6820 vdev_t
*rvd
= spa
->spa_root_vdev
;
6822 nvlist_t
**spares
, **l2cache
;
6823 uint_t nspares
, nl2cache
;
6825 ASSERT(spa_writeable(spa
));
6827 txg
= spa_vdev_enter(spa
);
6829 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
6830 VDEV_ALLOC_ADD
)) != 0)
6831 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6833 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
6835 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
6839 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
6843 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
6844 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6846 if (vd
->vdev_children
!= 0 &&
6847 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0) {
6848 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6852 * The virtual dRAID spares must be added after vdev tree is created
6853 * and the vdev guids are generated. The guid of their associated
6854 * dRAID is stored in the config and used when opening the spare.
6856 if ((error
= vdev_draid_spare_create(nvroot
, vd
, &ndraid
,
6857 rvd
->vdev_children
)) == 0) {
6858 if (ndraid
> 0 && nvlist_lookup_nvlist_array(nvroot
,
6859 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) != 0)
6862 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6866 * We must validate the spares and l2cache devices after checking the
6867 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
6869 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
6870 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6873 * If we are in the middle of a device removal, we can only add
6874 * devices which match the existing devices in the pool.
6875 * If we are in the middle of a removal, or have some indirect
6876 * vdevs, we can not add raidz or dRAID top levels.
6878 if (spa
->spa_vdev_removal
!= NULL
||
6879 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
6880 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6881 tvd
= vd
->vdev_child
[c
];
6882 if (spa
->spa_vdev_removal
!= NULL
&&
6883 tvd
->vdev_ashift
!= spa
->spa_max_ashift
) {
6884 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6886 /* Fail if top level vdev is raidz or a dRAID */
6887 if (vdev_get_nparity(tvd
) != 0)
6888 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6891 * Need the top level mirror to be
6892 * a mirror of leaf vdevs only
6894 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
6895 for (uint64_t cid
= 0;
6896 cid
< tvd
->vdev_children
; cid
++) {
6897 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
6898 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
6899 return (spa_vdev_exit(spa
, vd
,
6907 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6908 tvd
= vd
->vdev_child
[c
];
6909 vdev_remove_child(vd
, tvd
);
6910 tvd
->vdev_id
= rvd
->vdev_children
;
6911 vdev_add_child(rvd
, tvd
);
6912 vdev_config_dirty(tvd
);
6916 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
6917 ZPOOL_CONFIG_SPARES
);
6918 spa_load_spares(spa
);
6919 spa
->spa_spares
.sav_sync
= B_TRUE
;
6922 if (nl2cache
!= 0) {
6923 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
6924 ZPOOL_CONFIG_L2CACHE
);
6925 spa_load_l2cache(spa
);
6926 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6930 * We can't increment a feature while holding spa_vdev so we
6931 * have to do it in a synctask.
6936 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
6937 dsl_sync_task_nowait(spa
->spa_dsl_pool
, spa_draid_feature_incr
,
6938 (void *)(uintptr_t)ndraid
, tx
);
6943 * We have to be careful when adding new vdevs to an existing pool.
6944 * If other threads start allocating from these vdevs before we
6945 * sync the config cache, and we lose power, then upon reboot we may
6946 * fail to open the pool because there are DVAs that the config cache
6947 * can't translate. Therefore, we first add the vdevs without
6948 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
6949 * and then let spa_config_update() initialize the new metaslabs.
6951 * spa_load() checks for added-but-not-initialized vdevs, so that
6952 * if we lose power at any point in this sequence, the remaining
6953 * steps will be completed the next time we load the pool.
6955 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
6957 mutex_enter(&spa_namespace_lock
);
6958 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6959 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
6960 mutex_exit(&spa_namespace_lock
);
6966 * Attach a device to a vdev specified by its guid. The vdev type can be
6967 * a mirror, a raidz, or a leaf device that is also a top-level (e.g. a
6968 * single device). When the vdev is a single device, a mirror vdev will be
6969 * automatically inserted.
6971 * If 'replacing' is specified, the new device is intended to replace the
6972 * existing device; in this case the two devices are made into their own
6973 * mirror using the 'replacing' vdev, which is functionally identical to
6974 * the mirror vdev (it actually reuses all the same ops) but has a few
6975 * extra rules: you can't attach to it after it's been created, and upon
6976 * completion of resilvering, the first disk (the one being replaced)
6977 * is automatically detached.
6979 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
6980 * should be performed instead of traditional healing reconstruction. From
6981 * an administrators perspective these are both resilver operations.
6984 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
,
6987 uint64_t txg
, dtl_max_txg
;
6988 vdev_t
*rvd
= spa
->spa_root_vdev
;
6989 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
6991 char *oldvdpath
, *newvdpath
;
6992 int newvd_isspare
= B_FALSE
;
6995 ASSERT(spa_writeable(spa
));
6997 txg
= spa_vdev_enter(spa
);
6999 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7001 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7002 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7003 error
= (spa_has_checkpoint(spa
)) ?
7004 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7005 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7009 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_DEVICE_REBUILD
))
7010 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7012 if (dsl_scan_resilvering(spa_get_dsl(spa
)) ||
7013 dsl_scan_resilver_scheduled(spa_get_dsl(spa
))) {
7014 return (spa_vdev_exit(spa
, NULL
, txg
,
7015 ZFS_ERR_RESILVER_IN_PROGRESS
));
7018 if (vdev_rebuild_active(rvd
))
7019 return (spa_vdev_exit(spa
, NULL
, txg
,
7020 ZFS_ERR_REBUILD_IN_PROGRESS
));
7023 if (spa
->spa_vdev_removal
!= NULL
) {
7024 return (spa_vdev_exit(spa
, NULL
, txg
,
7025 ZFS_ERR_DEVRM_IN_PROGRESS
));
7029 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
7031 boolean_t raidz
= oldvd
->vdev_ops
== &vdev_raidz_ops
;
7034 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_RAIDZ_EXPANSION
))
7035 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7038 * Can't expand a raidz while prior expand is in progress.
7040 if (spa
->spa_raidz_expand
!= NULL
) {
7041 return (spa_vdev_exit(spa
, NULL
, txg
,
7042 ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS
));
7044 } else if (!oldvd
->vdev_ops
->vdev_op_leaf
) {
7045 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7051 pvd
= oldvd
->vdev_parent
;
7053 if (spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
7054 VDEV_ALLOC_ATTACH
) != 0)
7055 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7057 if (newrootvd
->vdev_children
!= 1)
7058 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
7060 newvd
= newrootvd
->vdev_child
[0];
7062 if (!newvd
->vdev_ops
->vdev_op_leaf
)
7063 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
7065 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
7066 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
7069 * log, dedup and special vdevs should not be replaced by spares.
7071 if ((oldvd
->vdev_top
->vdev_alloc_bias
!= VDEV_BIAS_NONE
||
7072 oldvd
->vdev_top
->vdev_islog
) && newvd
->vdev_isspare
) {
7073 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7077 * A dRAID spare can only replace a child of its parent dRAID vdev.
7079 if (newvd
->vdev_ops
== &vdev_draid_spare_ops
&&
7080 oldvd
->vdev_top
!= vdev_draid_spare_get_parent(newvd
)) {
7081 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7086 * For rebuilds, the top vdev must support reconstruction
7087 * using only space maps. This means the only allowable
7088 * vdevs types are the root vdev, a mirror, or dRAID.
7091 if (pvd
->vdev_top
!= NULL
)
7092 tvd
= pvd
->vdev_top
;
7094 if (tvd
->vdev_ops
!= &vdev_mirror_ops
&&
7095 tvd
->vdev_ops
!= &vdev_root_ops
&&
7096 tvd
->vdev_ops
!= &vdev_draid_ops
) {
7097 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7103 * For attach, the only allowable parent is a mirror or
7104 * the root vdev. A raidz vdev can be attached to, but
7105 * you cannot attach to a raidz child.
7107 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
7108 pvd
->vdev_ops
!= &vdev_root_ops
&&
7110 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7112 pvops
= &vdev_mirror_ops
;
7115 * Active hot spares can only be replaced by inactive hot
7118 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
7119 oldvd
->vdev_isspare
&&
7120 !spa_has_spare(spa
, newvd
->vdev_guid
))
7121 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7124 * If the source is a hot spare, and the parent isn't already a
7125 * spare, then we want to create a new hot spare. Otherwise, we
7126 * want to create a replacing vdev. The user is not allowed to
7127 * attach to a spared vdev child unless the 'isspare' state is
7128 * the same (spare replaces spare, non-spare replaces
7131 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
7132 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
7133 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7134 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
7135 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
7136 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7139 if (newvd
->vdev_isspare
)
7140 pvops
= &vdev_spare_ops
;
7142 pvops
= &vdev_replacing_ops
;
7146 * Make sure the new device is big enough.
7148 vdev_t
*min_vdev
= raidz
? oldvd
->vdev_child
[0] : oldvd
;
7149 if (newvd
->vdev_asize
< vdev_get_min_asize(min_vdev
))
7150 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
7153 * The new device cannot have a higher alignment requirement
7154 * than the top-level vdev.
7156 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
7157 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7160 * RAIDZ-expansion-specific checks.
7163 if (vdev_raidz_attach_check(newvd
) != 0)
7164 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7167 * Fail early if a child is not healthy or being replaced
7169 for (int i
= 0; i
< oldvd
->vdev_children
; i
++) {
7170 if (vdev_is_dead(oldvd
->vdev_child
[i
]) ||
7171 !oldvd
->vdev_child
[i
]->vdev_ops
->vdev_op_leaf
) {
7172 return (spa_vdev_exit(spa
, newrootvd
, txg
,
7175 /* Also fail if reserved boot area is in-use */
7176 if (vdev_check_boot_reserve(spa
, oldvd
->vdev_child
[i
])
7178 return (spa_vdev_exit(spa
, newrootvd
, txg
,
7186 * Note: oldvdpath is freed by spa_strfree(), but
7187 * kmem_asprintf() is freed by kmem_strfree(), so we have to
7188 * move it to a spa_strdup-ed string.
7190 char *tmp
= kmem_asprintf("raidz%u-%u",
7191 (uint_t
)vdev_get_nparity(oldvd
), (uint_t
)oldvd
->vdev_id
);
7192 oldvdpath
= spa_strdup(tmp
);
7195 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
7197 newvdpath
= spa_strdup(newvd
->vdev_path
);
7200 * If this is an in-place replacement, update oldvd's path and devid
7201 * to make it distinguishable from newvd, and unopenable from now on.
7203 if (strcmp(oldvdpath
, newvdpath
) == 0) {
7204 spa_strfree(oldvd
->vdev_path
);
7205 oldvd
->vdev_path
= kmem_alloc(strlen(newvdpath
) + 5,
7207 (void) sprintf(oldvd
->vdev_path
, "%s/old",
7209 if (oldvd
->vdev_devid
!= NULL
) {
7210 spa_strfree(oldvd
->vdev_devid
);
7211 oldvd
->vdev_devid
= NULL
;
7213 spa_strfree(oldvdpath
);
7214 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
7218 * If the parent is not a mirror, or if we're replacing, insert the new
7219 * mirror/replacing/spare vdev above oldvd.
7221 if (!raidz
&& pvd
->vdev_ops
!= pvops
) {
7222 pvd
= vdev_add_parent(oldvd
, pvops
);
7223 ASSERT(pvd
->vdev_ops
== pvops
);
7224 ASSERT(oldvd
->vdev_parent
== pvd
);
7227 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
7230 * Extract the new device from its root and add it to pvd.
7232 vdev_remove_child(newrootvd
, newvd
);
7233 newvd
->vdev_id
= pvd
->vdev_children
;
7234 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
7235 vdev_add_child(pvd
, newvd
);
7238 * Reevaluate the parent vdev state.
7240 vdev_propagate_state(pvd
);
7242 tvd
= newvd
->vdev_top
;
7243 ASSERT(pvd
->vdev_top
== tvd
);
7244 ASSERT(tvd
->vdev_parent
== rvd
);
7246 vdev_config_dirty(tvd
);
7249 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
7250 * for any dmu_sync-ed blocks. It will propagate upward when
7251 * spa_vdev_exit() calls vdev_dtl_reassess().
7253 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
7257 * Wait for the youngest allocations and frees to sync,
7258 * and then wait for the deferral of those frees to finish.
7260 spa_vdev_config_exit(spa
, NULL
,
7261 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
7263 vdev_initialize_stop_all(tvd
, VDEV_INITIALIZE_ACTIVE
);
7264 vdev_trim_stop_all(tvd
, VDEV_TRIM_ACTIVE
);
7265 vdev_autotrim_stop_wait(tvd
);
7267 dtl_max_txg
= spa_vdev_config_enter(spa
);
7269 tvd
->vdev_rz_expanding
= B_TRUE
;
7271 vdev_dirty_leaves(tvd
, VDD_DTL
, dtl_max_txg
);
7272 vdev_config_dirty(tvd
);
7274 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
,
7276 dsl_sync_task_nowait(spa
->spa_dsl_pool
, vdev_raidz_attach_sync
,
7280 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
7281 dtl_max_txg
- TXG_INITIAL
);
7283 if (newvd
->vdev_isspare
) {
7284 spa_spare_activate(newvd
);
7285 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
7288 newvd_isspare
= newvd
->vdev_isspare
;
7291 * Mark newvd's DTL dirty in this txg.
7293 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
7296 * Schedule the resilver or rebuild to restart in the future.
7297 * We do this to ensure that dmu_sync-ed blocks have been
7298 * stitched into the respective datasets.
7301 newvd
->vdev_rebuild_txg
= txg
;
7305 newvd
->vdev_resilver_txg
= txg
;
7307 if (dsl_scan_resilvering(spa_get_dsl(spa
)) &&
7308 spa_feature_is_enabled(spa
,
7309 SPA_FEATURE_RESILVER_DEFER
)) {
7310 vdev_defer_resilver(newvd
);
7312 dsl_scan_restart_resilver(spa
->spa_dsl_pool
,
7318 if (spa
->spa_bootfs
)
7319 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
7321 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
7326 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
7328 spa_history_log_internal(spa
, "vdev attach", NULL
,
7329 "%s vdev=%s %s vdev=%s",
7330 replacing
&& newvd_isspare
? "spare in" :
7331 replacing
? "replace" : "attach", newvdpath
,
7332 replacing
? "for" : "to", oldvdpath
);
7334 spa_strfree(oldvdpath
);
7335 spa_strfree(newvdpath
);
7341 * Detach a device from a mirror or replacing vdev.
7343 * If 'replace_done' is specified, only detach if the parent
7344 * is a replacing or a spare vdev.
7347 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
7351 vdev_t
*rvd __maybe_unused
= spa
->spa_root_vdev
;
7352 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
7353 boolean_t unspare
= B_FALSE
;
7354 uint64_t unspare_guid
= 0;
7357 ASSERT(spa_writeable(spa
));
7359 txg
= spa_vdev_detach_enter(spa
, guid
);
7361 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7364 * Besides being called directly from the userland through the
7365 * ioctl interface, spa_vdev_detach() can be potentially called
7366 * at the end of spa_vdev_resilver_done().
7368 * In the regular case, when we have a checkpoint this shouldn't
7369 * happen as we never empty the DTLs of a vdev during the scrub
7370 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
7371 * should never get here when we have a checkpoint.
7373 * That said, even in a case when we checkpoint the pool exactly
7374 * as spa_vdev_resilver_done() calls this function everything
7375 * should be fine as the resilver will return right away.
7377 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7378 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7379 error
= (spa_has_checkpoint(spa
)) ?
7380 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7381 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7385 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
7387 if (!vd
->vdev_ops
->vdev_op_leaf
)
7388 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7390 pvd
= vd
->vdev_parent
;
7393 * If the parent/child relationship is not as expected, don't do it.
7394 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
7395 * vdev that's replacing B with C. The user's intent in replacing
7396 * is to go from M(A,B) to M(A,C). If the user decides to cancel
7397 * the replace by detaching C, the expected behavior is to end up
7398 * M(A,B). But suppose that right after deciding to detach C,
7399 * the replacement of B completes. We would have M(A,C), and then
7400 * ask to detach C, which would leave us with just A -- not what
7401 * the user wanted. To prevent this, we make sure that the
7402 * parent/child relationship hasn't changed -- in this example,
7403 * that C's parent is still the replacing vdev R.
7405 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
7406 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
7409 * Only 'replacing' or 'spare' vdevs can be replaced.
7411 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
7412 pvd
->vdev_ops
!= &vdev_spare_ops
)
7413 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7415 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
7416 spa_version(spa
) >= SPA_VERSION_SPARES
);
7419 * Only mirror, replacing, and spare vdevs support detach.
7421 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
7422 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
7423 pvd
->vdev_ops
!= &vdev_spare_ops
)
7424 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7427 * If this device has the only valid copy of some data,
7428 * we cannot safely detach it.
7430 if (vdev_dtl_required(vd
))
7431 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
7433 ASSERT(pvd
->vdev_children
>= 2);
7436 * If we are detaching the second disk from a replacing vdev, then
7437 * check to see if we changed the original vdev's path to have "/old"
7438 * at the end in spa_vdev_attach(). If so, undo that change now.
7440 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
7441 vd
->vdev_path
!= NULL
) {
7442 size_t len
= strlen(vd
->vdev_path
);
7444 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
7445 cvd
= pvd
->vdev_child
[c
];
7447 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
7450 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
7451 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
7452 spa_strfree(cvd
->vdev_path
);
7453 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
7460 * If we are detaching the original disk from a normal spare, then it
7461 * implies that the spare should become a real disk, and be removed
7462 * from the active spare list for the pool. dRAID spares on the
7463 * other hand are coupled to the pool and thus should never be removed
7464 * from the spares list.
7466 if (pvd
->vdev_ops
== &vdev_spare_ops
&& vd
->vdev_id
== 0) {
7467 vdev_t
*last_cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7469 if (last_cvd
->vdev_isspare
&&
7470 last_cvd
->vdev_ops
!= &vdev_draid_spare_ops
) {
7476 * Erase the disk labels so the disk can be used for other things.
7477 * This must be done after all other error cases are handled,
7478 * but before we disembowel vd (so we can still do I/O to it).
7479 * But if we can't do it, don't treat the error as fatal --
7480 * it may be that the unwritability of the disk is the reason
7481 * it's being detached!
7483 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
7486 * Remove vd from its parent and compact the parent's children.
7488 vdev_remove_child(pvd
, vd
);
7489 vdev_compact_children(pvd
);
7492 * Remember one of the remaining children so we can get tvd below.
7494 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7497 * If we need to remove the remaining child from the list of hot spares,
7498 * do it now, marking the vdev as no longer a spare in the process.
7499 * We must do this before vdev_remove_parent(), because that can
7500 * change the GUID if it creates a new toplevel GUID. For a similar
7501 * reason, we must remove the spare now, in the same txg as the detach;
7502 * otherwise someone could attach a new sibling, change the GUID, and
7503 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
7506 ASSERT(cvd
->vdev_isspare
);
7507 spa_spare_remove(cvd
);
7508 unspare_guid
= cvd
->vdev_guid
;
7509 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
7510 cvd
->vdev_unspare
= B_TRUE
;
7514 * If the parent mirror/replacing vdev only has one child,
7515 * the parent is no longer needed. Remove it from the tree.
7517 if (pvd
->vdev_children
== 1) {
7518 if (pvd
->vdev_ops
== &vdev_spare_ops
)
7519 cvd
->vdev_unspare
= B_FALSE
;
7520 vdev_remove_parent(cvd
);
7524 * We don't set tvd until now because the parent we just removed
7525 * may have been the previous top-level vdev.
7527 tvd
= cvd
->vdev_top
;
7528 ASSERT(tvd
->vdev_parent
== rvd
);
7531 * Reevaluate the parent vdev state.
7533 vdev_propagate_state(cvd
);
7536 * If the 'autoexpand' property is set on the pool then automatically
7537 * try to expand the size of the pool. For example if the device we
7538 * just detached was smaller than the others, it may be possible to
7539 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
7540 * first so that we can obtain the updated sizes of the leaf vdevs.
7542 if (spa
->spa_autoexpand
) {
7544 vdev_expand(tvd
, txg
);
7547 vdev_config_dirty(tvd
);
7550 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
7551 * vd->vdev_detached is set and free vd's DTL object in syncing context.
7552 * But first make sure we're not on any *other* txg's DTL list, to
7553 * prevent vd from being accessed after it's freed.
7555 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
7556 for (int t
= 0; t
< TXG_SIZE
; t
++)
7557 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
7558 vd
->vdev_detached
= B_TRUE
;
7559 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
7561 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
7562 spa_notify_waiters(spa
);
7564 /* hang on to the spa before we release the lock */
7565 spa_open_ref(spa
, FTAG
);
7567 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
7569 spa_history_log_internal(spa
, "detach", NULL
,
7571 spa_strfree(vdpath
);
7574 * If this was the removal of the original device in a hot spare vdev,
7575 * then we want to go through and remove the device from the hot spare
7576 * list of every other pool.
7579 spa_t
*altspa
= NULL
;
7581 mutex_enter(&spa_namespace_lock
);
7582 while ((altspa
= spa_next(altspa
)) != NULL
) {
7583 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
7587 spa_open_ref(altspa
, FTAG
);
7588 mutex_exit(&spa_namespace_lock
);
7589 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
7590 mutex_enter(&spa_namespace_lock
);
7591 spa_close(altspa
, FTAG
);
7593 mutex_exit(&spa_namespace_lock
);
7595 /* search the rest of the vdevs for spares to remove */
7596 spa_vdev_resilver_done(spa
);
7599 /* all done with the spa; OK to release */
7600 mutex_enter(&spa_namespace_lock
);
7601 spa_close(spa
, FTAG
);
7602 mutex_exit(&spa_namespace_lock
);
7608 spa_vdev_initialize_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7611 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7613 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7615 /* Look up vdev and ensure it's a leaf. */
7616 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7617 if (vd
== NULL
|| vd
->vdev_detached
) {
7618 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7619 return (SET_ERROR(ENODEV
));
7620 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7621 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7622 return (SET_ERROR(EINVAL
));
7623 } else if (!vdev_writeable(vd
)) {
7624 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7625 return (SET_ERROR(EROFS
));
7627 mutex_enter(&vd
->vdev_initialize_lock
);
7628 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7631 * When we activate an initialize action we check to see
7632 * if the vdev_initialize_thread is NULL. We do this instead
7633 * of using the vdev_initialize_state since there might be
7634 * a previous initialization process which has completed but
7635 * the thread is not exited.
7637 if (cmd_type
== POOL_INITIALIZE_START
&&
7638 (vd
->vdev_initialize_thread
!= NULL
||
7639 vd
->vdev_top
->vdev_removing
|| vd
->vdev_top
->vdev_rz_expanding
)) {
7640 mutex_exit(&vd
->vdev_initialize_lock
);
7641 return (SET_ERROR(EBUSY
));
7642 } else if (cmd_type
== POOL_INITIALIZE_CANCEL
&&
7643 (vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
&&
7644 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_SUSPENDED
)) {
7645 mutex_exit(&vd
->vdev_initialize_lock
);
7646 return (SET_ERROR(ESRCH
));
7647 } else if (cmd_type
== POOL_INITIALIZE_SUSPEND
&&
7648 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
) {
7649 mutex_exit(&vd
->vdev_initialize_lock
);
7650 return (SET_ERROR(ESRCH
));
7651 } else if (cmd_type
== POOL_INITIALIZE_UNINIT
&&
7652 vd
->vdev_initialize_thread
!= NULL
) {
7653 mutex_exit(&vd
->vdev_initialize_lock
);
7654 return (SET_ERROR(EBUSY
));
7658 case POOL_INITIALIZE_START
:
7659 vdev_initialize(vd
);
7661 case POOL_INITIALIZE_CANCEL
:
7662 vdev_initialize_stop(vd
, VDEV_INITIALIZE_CANCELED
, vd_list
);
7664 case POOL_INITIALIZE_SUSPEND
:
7665 vdev_initialize_stop(vd
, VDEV_INITIALIZE_SUSPENDED
, vd_list
);
7667 case POOL_INITIALIZE_UNINIT
:
7668 vdev_uninitialize(vd
);
7671 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7673 mutex_exit(&vd
->vdev_initialize_lock
);
7679 spa_vdev_initialize(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
,
7680 nvlist_t
*vdev_errlist
)
7682 int total_errors
= 0;
7685 list_create(&vd_list
, sizeof (vdev_t
),
7686 offsetof(vdev_t
, vdev_initialize_node
));
7689 * We hold the namespace lock through the whole function
7690 * to prevent any changes to the pool while we're starting or
7691 * stopping initialization. The config and state locks are held so that
7692 * we can properly assess the vdev state before we commit to
7693 * the initializing operation.
7695 mutex_enter(&spa_namespace_lock
);
7697 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7698 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7699 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7701 int error
= spa_vdev_initialize_impl(spa
, vdev_guid
, cmd_type
,
7704 char guid_as_str
[MAXNAMELEN
];
7706 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7707 "%llu", (unsigned long long)vdev_guid
);
7708 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7713 /* Wait for all initialize threads to stop. */
7714 vdev_initialize_stop_wait(spa
, &vd_list
);
7716 /* Sync out the initializing state */
7717 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7718 mutex_exit(&spa_namespace_lock
);
7720 list_destroy(&vd_list
);
7722 return (total_errors
);
7726 spa_vdev_trim_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7727 uint64_t rate
, boolean_t partial
, boolean_t secure
, list_t
*vd_list
)
7729 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7731 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7733 /* Look up vdev and ensure it's a leaf. */
7734 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7735 if (vd
== NULL
|| vd
->vdev_detached
) {
7736 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7737 return (SET_ERROR(ENODEV
));
7738 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7739 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7740 return (SET_ERROR(EINVAL
));
7741 } else if (!vdev_writeable(vd
)) {
7742 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7743 return (SET_ERROR(EROFS
));
7744 } else if (!vd
->vdev_has_trim
) {
7745 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7746 return (SET_ERROR(EOPNOTSUPP
));
7747 } else if (secure
&& !vd
->vdev_has_securetrim
) {
7748 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7749 return (SET_ERROR(EOPNOTSUPP
));
7751 mutex_enter(&vd
->vdev_trim_lock
);
7752 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7755 * When we activate a TRIM action we check to see if the
7756 * vdev_trim_thread is NULL. We do this instead of using the
7757 * vdev_trim_state since there might be a previous TRIM process
7758 * which has completed but the thread is not exited.
7760 if (cmd_type
== POOL_TRIM_START
&&
7761 (vd
->vdev_trim_thread
!= NULL
|| vd
->vdev_top
->vdev_removing
||
7762 vd
->vdev_top
->vdev_rz_expanding
)) {
7763 mutex_exit(&vd
->vdev_trim_lock
);
7764 return (SET_ERROR(EBUSY
));
7765 } else if (cmd_type
== POOL_TRIM_CANCEL
&&
7766 (vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
&&
7767 vd
->vdev_trim_state
!= VDEV_TRIM_SUSPENDED
)) {
7768 mutex_exit(&vd
->vdev_trim_lock
);
7769 return (SET_ERROR(ESRCH
));
7770 } else if (cmd_type
== POOL_TRIM_SUSPEND
&&
7771 vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
) {
7772 mutex_exit(&vd
->vdev_trim_lock
);
7773 return (SET_ERROR(ESRCH
));
7777 case POOL_TRIM_START
:
7778 vdev_trim(vd
, rate
, partial
, secure
);
7780 case POOL_TRIM_CANCEL
:
7781 vdev_trim_stop(vd
, VDEV_TRIM_CANCELED
, vd_list
);
7783 case POOL_TRIM_SUSPEND
:
7784 vdev_trim_stop(vd
, VDEV_TRIM_SUSPENDED
, vd_list
);
7787 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7789 mutex_exit(&vd
->vdev_trim_lock
);
7795 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
7796 * TRIM threads for each child vdev. These threads pass over all of the free
7797 * space in the vdev's metaslabs and issues TRIM commands for that space.
7800 spa_vdev_trim(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
, uint64_t rate
,
7801 boolean_t partial
, boolean_t secure
, nvlist_t
*vdev_errlist
)
7803 int total_errors
= 0;
7806 list_create(&vd_list
, sizeof (vdev_t
),
7807 offsetof(vdev_t
, vdev_trim_node
));
7810 * We hold the namespace lock through the whole function
7811 * to prevent any changes to the pool while we're starting or
7812 * stopping TRIM. The config and state locks are held so that
7813 * we can properly assess the vdev state before we commit to
7814 * the TRIM operation.
7816 mutex_enter(&spa_namespace_lock
);
7818 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7819 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7820 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7822 int error
= spa_vdev_trim_impl(spa
, vdev_guid
, cmd_type
,
7823 rate
, partial
, secure
, &vd_list
);
7825 char guid_as_str
[MAXNAMELEN
];
7827 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7828 "%llu", (unsigned long long)vdev_guid
);
7829 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7834 /* Wait for all TRIM threads to stop. */
7835 vdev_trim_stop_wait(spa
, &vd_list
);
7837 /* Sync out the TRIM state */
7838 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7839 mutex_exit(&spa_namespace_lock
);
7841 list_destroy(&vd_list
);
7843 return (total_errors
);
7847 * Split a set of devices from their mirrors, and create a new pool from them.
7850 spa_vdev_split_mirror(spa_t
*spa
, const char *newname
, nvlist_t
*config
,
7851 nvlist_t
*props
, boolean_t exp
)
7854 uint64_t txg
, *glist
;
7856 uint_t c
, children
, lastlog
;
7857 nvlist_t
**child
, *nvl
, *tmp
;
7859 const char *altroot
= NULL
;
7860 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
7861 boolean_t activate_slog
;
7863 ASSERT(spa_writeable(spa
));
7865 txg
= spa_vdev_enter(spa
);
7867 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7868 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7869 error
= (spa_has_checkpoint(spa
)) ?
7870 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7871 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7874 /* clear the log and flush everything up to now */
7875 activate_slog
= spa_passivate_log(spa
);
7876 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
7877 error
= spa_reset_logs(spa
);
7878 txg
= spa_vdev_config_enter(spa
);
7881 spa_activate_log(spa
);
7884 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7886 /* check new spa name before going any further */
7887 if (spa_lookup(newname
) != NULL
)
7888 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
7891 * scan through all the children to ensure they're all mirrors
7893 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
7894 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
7896 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7898 /* first, check to ensure we've got the right child count */
7899 rvd
= spa
->spa_root_vdev
;
7901 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
7902 vdev_t
*vd
= rvd
->vdev_child
[c
];
7904 /* don't count the holes & logs as children */
7905 if (vd
->vdev_islog
|| (vd
->vdev_ops
!= &vdev_indirect_ops
&&
7906 !vdev_is_concrete(vd
))) {
7914 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
7915 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7917 /* next, ensure no spare or cache devices are part of the split */
7918 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
7919 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
7920 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7922 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
7923 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
7925 /* then, loop over each vdev and validate it */
7926 for (c
= 0; c
< children
; c
++) {
7927 uint64_t is_hole
= 0;
7929 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
7933 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
7934 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
7937 error
= SET_ERROR(EINVAL
);
7942 /* deal with indirect vdevs */
7943 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ops
==
7947 /* which disk is going to be split? */
7948 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
7950 error
= SET_ERROR(EINVAL
);
7954 /* look it up in the spa */
7955 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
7956 if (vml
[c
] == NULL
) {
7957 error
= SET_ERROR(ENODEV
);
7961 /* make sure there's nothing stopping the split */
7962 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
7963 vml
[c
]->vdev_islog
||
7964 !vdev_is_concrete(vml
[c
]) ||
7965 vml
[c
]->vdev_isspare
||
7966 vml
[c
]->vdev_isl2cache
||
7967 !vdev_writeable(vml
[c
]) ||
7968 vml
[c
]->vdev_children
!= 0 ||
7969 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
7970 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
7971 error
= SET_ERROR(EINVAL
);
7975 if (vdev_dtl_required(vml
[c
]) ||
7976 vdev_resilver_needed(vml
[c
], NULL
, NULL
)) {
7977 error
= SET_ERROR(EBUSY
);
7981 /* we need certain info from the top level */
7982 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
7983 vml
[c
]->vdev_top
->vdev_ms_array
);
7984 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
7985 vml
[c
]->vdev_top
->vdev_ms_shift
);
7986 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
7987 vml
[c
]->vdev_top
->vdev_asize
);
7988 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
7989 vml
[c
]->vdev_top
->vdev_ashift
);
7991 /* transfer per-vdev ZAPs */
7992 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
7993 VERIFY0(nvlist_add_uint64(child
[c
],
7994 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
7996 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
7997 VERIFY0(nvlist_add_uint64(child
[c
],
7998 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
7999 vml
[c
]->vdev_parent
->vdev_top_zap
));
8003 kmem_free(vml
, children
* sizeof (vdev_t
*));
8004 kmem_free(glist
, children
* sizeof (uint64_t));
8005 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
8008 /* stop writers from using the disks */
8009 for (c
= 0; c
< children
; c
++) {
8011 vml
[c
]->vdev_offline
= B_TRUE
;
8013 vdev_reopen(spa
->spa_root_vdev
);
8016 * Temporarily record the splitting vdevs in the spa config. This
8017 * will disappear once the config is regenerated.
8019 nvl
= fnvlist_alloc();
8020 fnvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
, glist
, children
);
8021 kmem_free(glist
, children
* sizeof (uint64_t));
8023 mutex_enter(&spa
->spa_props_lock
);
8024 fnvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
, nvl
);
8025 mutex_exit(&spa
->spa_props_lock
);
8026 spa
->spa_config_splitting
= nvl
;
8027 vdev_config_dirty(spa
->spa_root_vdev
);
8029 /* configure and create the new pool */
8030 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
);
8031 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
8032 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
);
8033 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
, spa_version(spa
));
8034 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, spa
->spa_config_txg
);
8035 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
8036 spa_generate_guid(NULL
));
8037 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
8038 (void) nvlist_lookup_string(props
,
8039 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
8041 /* add the new pool to the namespace */
8042 newspa
= spa_add(newname
, config
, altroot
);
8043 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
8044 newspa
->spa_config_txg
= spa
->spa_config_txg
;
8045 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
8047 /* release the spa config lock, retaining the namespace lock */
8048 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
8050 if (zio_injection_enabled
)
8051 zio_handle_panic_injection(spa
, FTAG
, 1);
8053 spa_activate(newspa
, spa_mode_global
);
8054 spa_async_suspend(newspa
);
8057 * Temporarily stop the initializing and TRIM activity. We set the
8058 * state to ACTIVE so that we know to resume initializing or TRIM
8059 * once the split has completed.
8061 list_t vd_initialize_list
;
8062 list_create(&vd_initialize_list
, sizeof (vdev_t
),
8063 offsetof(vdev_t
, vdev_initialize_node
));
8065 list_t vd_trim_list
;
8066 list_create(&vd_trim_list
, sizeof (vdev_t
),
8067 offsetof(vdev_t
, vdev_trim_node
));
8069 for (c
= 0; c
< children
; c
++) {
8070 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
8071 mutex_enter(&vml
[c
]->vdev_initialize_lock
);
8072 vdev_initialize_stop(vml
[c
],
8073 VDEV_INITIALIZE_ACTIVE
, &vd_initialize_list
);
8074 mutex_exit(&vml
[c
]->vdev_initialize_lock
);
8076 mutex_enter(&vml
[c
]->vdev_trim_lock
);
8077 vdev_trim_stop(vml
[c
], VDEV_TRIM_ACTIVE
, &vd_trim_list
);
8078 mutex_exit(&vml
[c
]->vdev_trim_lock
);
8082 vdev_initialize_stop_wait(spa
, &vd_initialize_list
);
8083 vdev_trim_stop_wait(spa
, &vd_trim_list
);
8085 list_destroy(&vd_initialize_list
);
8086 list_destroy(&vd_trim_list
);
8088 newspa
->spa_config_source
= SPA_CONFIG_SRC_SPLIT
;
8089 newspa
->spa_is_splitting
= B_TRUE
;
8091 /* create the new pool from the disks of the original pool */
8092 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
);
8096 /* if that worked, generate a real config for the new pool */
8097 if (newspa
->spa_root_vdev
!= NULL
) {
8098 newspa
->spa_config_splitting
= fnvlist_alloc();
8099 fnvlist_add_uint64(newspa
->spa_config_splitting
,
8100 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
));
8101 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
8106 if (props
!= NULL
) {
8107 spa_configfile_set(newspa
, props
, B_FALSE
);
8108 error
= spa_prop_set(newspa
, props
);
8113 /* flush everything */
8114 txg
= spa_vdev_config_enter(newspa
);
8115 vdev_config_dirty(newspa
->spa_root_vdev
);
8116 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
8118 if (zio_injection_enabled
)
8119 zio_handle_panic_injection(spa
, FTAG
, 2);
8121 spa_async_resume(newspa
);
8123 /* finally, update the original pool's config */
8124 txg
= spa_vdev_config_enter(spa
);
8125 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
8126 error
= dmu_tx_assign(tx
, TXG_WAIT
);
8129 for (c
= 0; c
< children
; c
++) {
8130 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
8131 vdev_t
*tvd
= vml
[c
]->vdev_top
;
8134 * Need to be sure the detachable VDEV is not
8135 * on any *other* txg's DTL list to prevent it
8136 * from being accessed after it's freed.
8138 for (int t
= 0; t
< TXG_SIZE
; t
++) {
8139 (void) txg_list_remove_this(
8140 &tvd
->vdev_dtl_list
, vml
[c
], t
);
8145 spa_history_log_internal(spa
, "detach", tx
,
8146 "vdev=%s", vml
[c
]->vdev_path
);
8151 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
8152 vdev_config_dirty(spa
->spa_root_vdev
);
8153 spa
->spa_config_splitting
= NULL
;
8157 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
8159 if (zio_injection_enabled
)
8160 zio_handle_panic_injection(spa
, FTAG
, 3);
8162 /* split is complete; log a history record */
8163 spa_history_log_internal(newspa
, "split", NULL
,
8164 "from pool %s", spa_name(spa
));
8166 newspa
->spa_is_splitting
= B_FALSE
;
8167 kmem_free(vml
, children
* sizeof (vdev_t
*));
8169 /* if we're not going to mount the filesystems in userland, export */
8171 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
8178 spa_deactivate(newspa
);
8181 txg
= spa_vdev_config_enter(spa
);
8183 /* re-online all offlined disks */
8184 for (c
= 0; c
< children
; c
++) {
8186 vml
[c
]->vdev_offline
= B_FALSE
;
8189 /* restart initializing or trimming disks as necessary */
8190 spa_async_request(spa
, SPA_ASYNC_INITIALIZE_RESTART
);
8191 spa_async_request(spa
, SPA_ASYNC_TRIM_RESTART
);
8192 spa_async_request(spa
, SPA_ASYNC_AUTOTRIM_RESTART
);
8194 vdev_reopen(spa
->spa_root_vdev
);
8196 nvlist_free(spa
->spa_config_splitting
);
8197 spa
->spa_config_splitting
= NULL
;
8198 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
8200 kmem_free(vml
, children
* sizeof (vdev_t
*));
8205 * Find any device that's done replacing, or a vdev marked 'unspare' that's
8206 * currently spared, so we can detach it.
8209 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
8211 vdev_t
*newvd
, *oldvd
;
8213 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
8214 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
8220 * Check for a completed replacement. We always consider the first
8221 * vdev in the list to be the oldest vdev, and the last one to be
8222 * the newest (see spa_vdev_attach() for how that works). In
8223 * the case where the newest vdev is faulted, we will not automatically
8224 * remove it after a resilver completes. This is OK as it will require
8225 * user intervention to determine which disk the admin wishes to keep.
8227 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
8228 ASSERT(vd
->vdev_children
> 1);
8230 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
8231 oldvd
= vd
->vdev_child
[0];
8233 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
8234 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
8235 !vdev_dtl_required(oldvd
))
8240 * Check for a completed resilver with the 'unspare' flag set.
8241 * Also potentially update faulted state.
8243 if (vd
->vdev_ops
== &vdev_spare_ops
) {
8244 vdev_t
*first
= vd
->vdev_child
[0];
8245 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
8247 if (last
->vdev_unspare
) {
8250 } else if (first
->vdev_unspare
) {
8257 if (oldvd
!= NULL
&&
8258 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
8259 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
8260 !vdev_dtl_required(oldvd
))
8263 vdev_propagate_state(vd
);
8266 * If there are more than two spares attached to a disk,
8267 * and those spares are not required, then we want to
8268 * attempt to free them up now so that they can be used
8269 * by other pools. Once we're back down to a single
8270 * disk+spare, we stop removing them.
8272 if (vd
->vdev_children
> 2) {
8273 newvd
= vd
->vdev_child
[1];
8275 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
8276 vdev_dtl_empty(last
, DTL_MISSING
) &&
8277 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
8278 !vdev_dtl_required(newvd
))
8287 spa_vdev_resilver_done(spa_t
*spa
)
8289 vdev_t
*vd
, *pvd
, *ppvd
;
8290 uint64_t guid
, sguid
, pguid
, ppguid
;
8292 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
8294 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
8295 pvd
= vd
->vdev_parent
;
8296 ppvd
= pvd
->vdev_parent
;
8297 guid
= vd
->vdev_guid
;
8298 pguid
= pvd
->vdev_guid
;
8299 ppguid
= ppvd
->vdev_guid
;
8302 * If we have just finished replacing a hot spared device, then
8303 * we need to detach the parent's first child (the original hot
8306 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
8307 ppvd
->vdev_children
== 2) {
8308 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
8309 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
8311 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
8313 spa_config_exit(spa
, SCL_ALL
, FTAG
);
8314 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
8316 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
8318 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
8321 spa_config_exit(spa
, SCL_ALL
, FTAG
);
8324 * If a detach was not performed above replace waiters will not have
8325 * been notified. In which case we must do so now.
8327 spa_notify_waiters(spa
);
8331 * Update the stored path or FRU for this vdev.
8334 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
8338 boolean_t sync
= B_FALSE
;
8340 ASSERT(spa_writeable(spa
));
8342 spa_vdev_state_enter(spa
, SCL_ALL
);
8344 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
8345 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
8347 if (!vd
->vdev_ops
->vdev_op_leaf
)
8348 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
8351 if (strcmp(value
, vd
->vdev_path
) != 0) {
8352 spa_strfree(vd
->vdev_path
);
8353 vd
->vdev_path
= spa_strdup(value
);
8357 if (vd
->vdev_fru
== NULL
) {
8358 vd
->vdev_fru
= spa_strdup(value
);
8360 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
8361 spa_strfree(vd
->vdev_fru
);
8362 vd
->vdev_fru
= spa_strdup(value
);
8367 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
8371 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
8373 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
8377 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
8379 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
8383 * ==========================================================================
8385 * ==========================================================================
8388 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
8390 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8392 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
8393 return (SET_ERROR(EBUSY
));
8395 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
8399 spa_scan_stop(spa_t
*spa
)
8401 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8402 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
8403 return (SET_ERROR(EBUSY
));
8405 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
8409 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
8411 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8413 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
8414 return (SET_ERROR(ENOTSUP
));
8416 if (func
== POOL_SCAN_RESILVER
&&
8417 !spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
))
8418 return (SET_ERROR(ENOTSUP
));
8421 * If a resilver was requested, but there is no DTL on a
8422 * writeable leaf device, we have nothing to do.
8424 if (func
== POOL_SCAN_RESILVER
&&
8425 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
8426 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
8430 if (func
== POOL_SCAN_ERRORSCRUB
&&
8431 !spa_feature_is_enabled(spa
, SPA_FEATURE_HEAD_ERRLOG
))
8432 return (SET_ERROR(ENOTSUP
));
8434 return (dsl_scan(spa
->spa_dsl_pool
, func
));
8438 * ==========================================================================
8439 * SPA async task processing
8440 * ==========================================================================
8444 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
8446 if (vd
->vdev_remove_wanted
) {
8447 vd
->vdev_remove_wanted
= B_FALSE
;
8448 vd
->vdev_delayed_close
= B_FALSE
;
8449 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
8452 * We want to clear the stats, but we don't want to do a full
8453 * vdev_clear() as that will cause us to throw away
8454 * degraded/faulted state as well as attempt to reopen the
8455 * device, all of which is a waste.
8457 vd
->vdev_stat
.vs_read_errors
= 0;
8458 vd
->vdev_stat
.vs_write_errors
= 0;
8459 vd
->vdev_stat
.vs_checksum_errors
= 0;
8461 vdev_state_dirty(vd
->vdev_top
);
8463 /* Tell userspace that the vdev is gone. */
8464 zfs_post_remove(spa
, vd
);
8467 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8468 spa_async_remove(spa
, vd
->vdev_child
[c
]);
8472 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
8474 if (vd
->vdev_probe_wanted
) {
8475 vd
->vdev_probe_wanted
= B_FALSE
;
8476 vdev_reopen(vd
); /* vdev_open() does the actual probe */
8479 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8480 spa_async_probe(spa
, vd
->vdev_child
[c
]);
8484 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
8486 if (!spa
->spa_autoexpand
)
8489 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
8490 vdev_t
*cvd
= vd
->vdev_child
[c
];
8491 spa_async_autoexpand(spa
, cvd
);
8494 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
8497 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
8500 static __attribute__((noreturn
)) void
8501 spa_async_thread(void *arg
)
8503 spa_t
*spa
= (spa_t
*)arg
;
8504 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8507 ASSERT(spa
->spa_sync_on
);
8509 mutex_enter(&spa
->spa_async_lock
);
8510 tasks
= spa
->spa_async_tasks
;
8511 spa
->spa_async_tasks
= 0;
8512 mutex_exit(&spa
->spa_async_lock
);
8515 * See if the config needs to be updated.
8517 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
8518 uint64_t old_space
, new_space
;
8520 mutex_enter(&spa_namespace_lock
);
8521 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
8522 old_space
+= metaslab_class_get_space(spa_special_class(spa
));
8523 old_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8524 old_space
+= metaslab_class_get_space(
8525 spa_embedded_log_class(spa
));
8527 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
8529 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
8530 new_space
+= metaslab_class_get_space(spa_special_class(spa
));
8531 new_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8532 new_space
+= metaslab_class_get_space(
8533 spa_embedded_log_class(spa
));
8534 mutex_exit(&spa_namespace_lock
);
8537 * If the pool grew as a result of the config update,
8538 * then log an internal history event.
8540 if (new_space
!= old_space
) {
8541 spa_history_log_internal(spa
, "vdev online", NULL
,
8542 "pool '%s' size: %llu(+%llu)",
8543 spa_name(spa
), (u_longlong_t
)new_space
,
8544 (u_longlong_t
)(new_space
- old_space
));
8549 * See if any devices need to be marked REMOVED.
8551 if (tasks
& SPA_ASYNC_REMOVE
) {
8552 spa_vdev_state_enter(spa
, SCL_NONE
);
8553 spa_async_remove(spa
, spa
->spa_root_vdev
);
8554 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
8555 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
8556 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
8557 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
8558 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8561 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
8562 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8563 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
8564 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8568 * See if any devices need to be probed.
8570 if (tasks
& SPA_ASYNC_PROBE
) {
8571 spa_vdev_state_enter(spa
, SCL_NONE
);
8572 spa_async_probe(spa
, spa
->spa_root_vdev
);
8573 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8577 * If any devices are done replacing, detach them.
8579 if (tasks
& SPA_ASYNC_RESILVER_DONE
||
8580 tasks
& SPA_ASYNC_REBUILD_DONE
||
8581 tasks
& SPA_ASYNC_DETACH_SPARE
) {
8582 spa_vdev_resilver_done(spa
);
8586 * Kick off a resilver.
8588 if (tasks
& SPA_ASYNC_RESILVER
&&
8589 !vdev_rebuild_active(spa
->spa_root_vdev
) &&
8590 (!dsl_scan_resilvering(dp
) ||
8591 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_RESILVER_DEFER
)))
8592 dsl_scan_restart_resilver(dp
, 0);
8594 if (tasks
& SPA_ASYNC_INITIALIZE_RESTART
) {
8595 mutex_enter(&spa_namespace_lock
);
8596 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8597 vdev_initialize_restart(spa
->spa_root_vdev
);
8598 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8599 mutex_exit(&spa_namespace_lock
);
8602 if (tasks
& SPA_ASYNC_TRIM_RESTART
) {
8603 mutex_enter(&spa_namespace_lock
);
8604 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8605 vdev_trim_restart(spa
->spa_root_vdev
);
8606 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8607 mutex_exit(&spa_namespace_lock
);
8610 if (tasks
& SPA_ASYNC_AUTOTRIM_RESTART
) {
8611 mutex_enter(&spa_namespace_lock
);
8612 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8613 vdev_autotrim_restart(spa
);
8614 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8615 mutex_exit(&spa_namespace_lock
);
8619 * Kick off L2 cache whole device TRIM.
8621 if (tasks
& SPA_ASYNC_L2CACHE_TRIM
) {
8622 mutex_enter(&spa_namespace_lock
);
8623 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8624 vdev_trim_l2arc(spa
);
8625 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8626 mutex_exit(&spa_namespace_lock
);
8630 * Kick off L2 cache rebuilding.
8632 if (tasks
& SPA_ASYNC_L2CACHE_REBUILD
) {
8633 mutex_enter(&spa_namespace_lock
);
8634 spa_config_enter(spa
, SCL_L2ARC
, FTAG
, RW_READER
);
8635 l2arc_spa_rebuild_start(spa
);
8636 spa_config_exit(spa
, SCL_L2ARC
, FTAG
);
8637 mutex_exit(&spa_namespace_lock
);
8641 * Let the world know that we're done.
8643 mutex_enter(&spa
->spa_async_lock
);
8644 spa
->spa_async_thread
= NULL
;
8645 cv_broadcast(&spa
->spa_async_cv
);
8646 mutex_exit(&spa
->spa_async_lock
);
8651 spa_async_suspend(spa_t
*spa
)
8653 mutex_enter(&spa
->spa_async_lock
);
8654 spa
->spa_async_suspended
++;
8655 while (spa
->spa_async_thread
!= NULL
)
8656 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
8657 mutex_exit(&spa
->spa_async_lock
);
8659 spa_vdev_remove_suspend(spa
);
8661 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8662 if (condense_thread
!= NULL
)
8663 zthr_cancel(condense_thread
);
8665 zthr_t
*raidz_expand_thread
= spa
->spa_raidz_expand_zthr
;
8666 if (raidz_expand_thread
!= NULL
)
8667 zthr_cancel(raidz_expand_thread
);
8669 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8670 if (discard_thread
!= NULL
)
8671 zthr_cancel(discard_thread
);
8673 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8674 if (ll_delete_thread
!= NULL
)
8675 zthr_cancel(ll_delete_thread
);
8677 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8678 if (ll_condense_thread
!= NULL
)
8679 zthr_cancel(ll_condense_thread
);
8683 spa_async_resume(spa_t
*spa
)
8685 mutex_enter(&spa
->spa_async_lock
);
8686 ASSERT(spa
->spa_async_suspended
!= 0);
8687 spa
->spa_async_suspended
--;
8688 mutex_exit(&spa
->spa_async_lock
);
8689 spa_restart_removal(spa
);
8691 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8692 if (condense_thread
!= NULL
)
8693 zthr_resume(condense_thread
);
8695 zthr_t
*raidz_expand_thread
= spa
->spa_raidz_expand_zthr
;
8696 if (raidz_expand_thread
!= NULL
)
8697 zthr_resume(raidz_expand_thread
);
8699 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8700 if (discard_thread
!= NULL
)
8701 zthr_resume(discard_thread
);
8703 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8704 if (ll_delete_thread
!= NULL
)
8705 zthr_resume(ll_delete_thread
);
8707 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8708 if (ll_condense_thread
!= NULL
)
8709 zthr_resume(ll_condense_thread
);
8713 spa_async_tasks_pending(spa_t
*spa
)
8715 uint_t non_config_tasks
;
8717 boolean_t config_task_suspended
;
8719 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
8720 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
8721 if (spa
->spa_ccw_fail_time
== 0) {
8722 config_task_suspended
= B_FALSE
;
8724 config_task_suspended
=
8725 (gethrtime() - spa
->spa_ccw_fail_time
) <
8726 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
8729 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
8733 spa_async_dispatch(spa_t
*spa
)
8735 mutex_enter(&spa
->spa_async_lock
);
8736 if (spa_async_tasks_pending(spa
) &&
8737 !spa
->spa_async_suspended
&&
8738 spa
->spa_async_thread
== NULL
)
8739 spa
->spa_async_thread
= thread_create(NULL
, 0,
8740 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
8741 mutex_exit(&spa
->spa_async_lock
);
8745 spa_async_request(spa_t
*spa
, int task
)
8747 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
8748 mutex_enter(&spa
->spa_async_lock
);
8749 spa
->spa_async_tasks
|= task
;
8750 mutex_exit(&spa
->spa_async_lock
);
8754 spa_async_tasks(spa_t
*spa
)
8756 return (spa
->spa_async_tasks
);
8760 * ==========================================================================
8761 * SPA syncing routines
8762 * ==========================================================================
8767 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8771 bpobj_enqueue(bpo
, bp
, bp_freed
, tx
);
8776 bpobj_enqueue_alloc_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8778 return (bpobj_enqueue_cb(arg
, bp
, B_FALSE
, tx
));
8782 bpobj_enqueue_free_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8784 return (bpobj_enqueue_cb(arg
, bp
, B_TRUE
, tx
));
8788 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8792 zio_nowait(zio_free_sync(pio
, pio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
8798 bpobj_spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8802 return (spa_free_sync_cb(arg
, bp
, tx
));
8806 * Note: this simple function is not inlined to make it easier to dtrace the
8807 * amount of time spent syncing frees.
8810 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
8812 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8813 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
8814 VERIFY(zio_wait(zio
) == 0);
8818 * Note: this simple function is not inlined to make it easier to dtrace the
8819 * amount of time spent syncing deferred frees.
8822 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
8824 if (spa_sync_pass(spa
) != 1)
8829 * If the log space map feature is active, we stop deferring
8830 * frees to the next TXG and therefore running this function
8831 * would be considered a no-op as spa_deferred_bpobj should
8832 * not have any entries.
8834 * That said we run this function anyway (instead of returning
8835 * immediately) for the edge-case scenario where we just
8836 * activated the log space map feature in this TXG but we have
8837 * deferred frees from the previous TXG.
8839 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8840 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
8841 bpobj_spa_free_sync_cb
, zio
, tx
), ==, 0);
8842 VERIFY0(zio_wait(zio
));
8846 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
8848 char *packed
= NULL
;
8853 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
8856 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
8857 * information. This avoids the dmu_buf_will_dirty() path and
8858 * saves us a pre-read to get data we don't actually care about.
8860 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
8861 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
8863 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
8865 memset(packed
+ nvsize
, 0, bufsize
- nvsize
);
8867 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
8869 vmem_free(packed
, bufsize
);
8871 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
8872 dmu_buf_will_dirty(db
, tx
);
8873 *(uint64_t *)db
->db_data
= nvsize
;
8874 dmu_buf_rele(db
, FTAG
);
8878 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
8879 const char *config
, const char *entry
)
8889 * Update the MOS nvlist describing the list of available devices.
8890 * spa_validate_aux() will have already made sure this nvlist is
8891 * valid and the vdevs are labeled appropriately.
8893 if (sav
->sav_object
== 0) {
8894 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
8895 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
8896 sizeof (uint64_t), tx
);
8897 VERIFY(zap_update(spa
->spa_meta_objset
,
8898 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
8899 &sav
->sav_object
, tx
) == 0);
8902 nvroot
= fnvlist_alloc();
8903 if (sav
->sav_count
== 0) {
8904 fnvlist_add_nvlist_array(nvroot
, config
,
8905 (const nvlist_t
* const *)NULL
, 0);
8907 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
8908 for (i
= 0; i
< sav
->sav_count
; i
++)
8909 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
8910 B_FALSE
, VDEV_CONFIG_L2CACHE
);
8911 fnvlist_add_nvlist_array(nvroot
, config
,
8912 (const nvlist_t
* const *)list
, sav
->sav_count
);
8913 for (i
= 0; i
< sav
->sav_count
; i
++)
8914 nvlist_free(list
[i
]);
8915 kmem_free(list
, sav
->sav_count
* sizeof (void *));
8918 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
8919 nvlist_free(nvroot
);
8921 sav
->sav_sync
= B_FALSE
;
8925 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
8926 * The all-vdev ZAP must be empty.
8929 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
8931 spa_t
*spa
= vd
->vdev_spa
;
8933 if (vd
->vdev_root_zap
!= 0 &&
8934 spa_feature_is_active(spa
, SPA_FEATURE_AVZ_V2
)) {
8935 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8936 vd
->vdev_root_zap
, tx
));
8938 if (vd
->vdev_top_zap
!= 0) {
8939 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8940 vd
->vdev_top_zap
, tx
));
8942 if (vd
->vdev_leaf_zap
!= 0) {
8943 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8944 vd
->vdev_leaf_zap
, tx
));
8946 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
8947 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
8952 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
8957 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
8958 * its config may not be dirty but we still need to build per-vdev ZAPs.
8959 * Similarly, if the pool is being assembled (e.g. after a split), we
8960 * need to rebuild the AVZ although the config may not be dirty.
8962 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
8963 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
8966 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8968 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
8969 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
8970 spa
->spa_all_vdev_zaps
!= 0);
8972 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
8973 /* Make and build the new AVZ */
8974 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
8975 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
8976 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
8978 /* Diff old AVZ with new one */
8982 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
8983 spa
->spa_all_vdev_zaps
);
8984 zap_cursor_retrieve(&zc
, &za
) == 0;
8985 zap_cursor_advance(&zc
)) {
8986 uint64_t vdzap
= za
.za_first_integer
;
8987 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
8990 * ZAP is listed in old AVZ but not in new one;
8993 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
8998 zap_cursor_fini(&zc
);
9000 /* Destroy the old AVZ */
9001 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
9002 spa
->spa_all_vdev_zaps
, tx
));
9004 /* Replace the old AVZ in the dir obj with the new one */
9005 VERIFY0(zap_update(spa
->spa_meta_objset
,
9006 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
9007 sizeof (new_avz
), 1, &new_avz
, tx
));
9009 spa
->spa_all_vdev_zaps
= new_avz
;
9010 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
9014 /* Walk through the AVZ and destroy all listed ZAPs */
9015 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
9016 spa
->spa_all_vdev_zaps
);
9017 zap_cursor_retrieve(&zc
, &za
) == 0;
9018 zap_cursor_advance(&zc
)) {
9019 uint64_t zap
= za
.za_first_integer
;
9020 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
9023 zap_cursor_fini(&zc
);
9025 /* Destroy and unlink the AVZ itself */
9026 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
9027 spa
->spa_all_vdev_zaps
, tx
));
9028 VERIFY0(zap_remove(spa
->spa_meta_objset
,
9029 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
9030 spa
->spa_all_vdev_zaps
= 0;
9033 if (spa
->spa_all_vdev_zaps
== 0) {
9034 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
9035 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
9036 DMU_POOL_VDEV_ZAP_MAP
, tx
);
9038 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
9040 /* Create ZAPs for vdevs that don't have them. */
9041 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
9043 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
9044 dmu_tx_get_txg(tx
), B_FALSE
);
9047 * If we're upgrading the spa version then make sure that
9048 * the config object gets updated with the correct version.
9050 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
9051 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
9052 spa
->spa_uberblock
.ub_version
);
9054 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9056 nvlist_free(spa
->spa_config_syncing
);
9057 spa
->spa_config_syncing
= config
;
9059 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
9063 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
9065 uint64_t *versionp
= arg
;
9066 uint64_t version
= *versionp
;
9067 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
9070 * Setting the version is special cased when first creating the pool.
9072 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
9074 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
9075 ASSERT(version
>= spa_version(spa
));
9077 spa
->spa_uberblock
.ub_version
= version
;
9078 vdev_config_dirty(spa
->spa_root_vdev
);
9079 spa_history_log_internal(spa
, "set", tx
, "version=%lld",
9080 (longlong_t
)version
);
9084 * Set zpool properties.
9087 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
9089 nvlist_t
*nvp
= arg
;
9090 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
9091 objset_t
*mos
= spa
->spa_meta_objset
;
9092 nvpair_t
*elem
= NULL
;
9094 mutex_enter(&spa
->spa_props_lock
);
9096 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
9098 const char *strval
, *fname
;
9100 const char *propname
;
9101 const char *elemname
= nvpair_name(elem
);
9102 zprop_type_t proptype
;
9105 switch (prop
= zpool_name_to_prop(elemname
)) {
9106 case ZPOOL_PROP_VERSION
:
9107 intval
= fnvpair_value_uint64(elem
);
9109 * The version is synced separately before other
9110 * properties and should be correct by now.
9112 ASSERT3U(spa_version(spa
), >=, intval
);
9115 case ZPOOL_PROP_ALTROOT
:
9117 * 'altroot' is a non-persistent property. It should
9118 * have been set temporarily at creation or import time.
9120 ASSERT(spa
->spa_root
!= NULL
);
9123 case ZPOOL_PROP_READONLY
:
9124 case ZPOOL_PROP_CACHEFILE
:
9126 * 'readonly' and 'cachefile' are also non-persistent
9130 case ZPOOL_PROP_COMMENT
:
9131 strval
= fnvpair_value_string(elem
);
9132 if (spa
->spa_comment
!= NULL
)
9133 spa_strfree(spa
->spa_comment
);
9134 spa
->spa_comment
= spa_strdup(strval
);
9136 * We need to dirty the configuration on all the vdevs
9137 * so that their labels get updated. We also need to
9138 * update the cache file to keep it in sync with the
9139 * MOS version. It's unnecessary to do this for pool
9140 * creation since the vdev's configuration has already
9143 if (tx
->tx_txg
!= TXG_INITIAL
) {
9144 vdev_config_dirty(spa
->spa_root_vdev
);
9145 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
9147 spa_history_log_internal(spa
, "set", tx
,
9148 "%s=%s", elemname
, strval
);
9150 case ZPOOL_PROP_COMPATIBILITY
:
9151 strval
= fnvpair_value_string(elem
);
9152 if (spa
->spa_compatibility
!= NULL
)
9153 spa_strfree(spa
->spa_compatibility
);
9154 spa
->spa_compatibility
= spa_strdup(strval
);
9156 * Dirty the configuration on vdevs as above.
9158 if (tx
->tx_txg
!= TXG_INITIAL
) {
9159 vdev_config_dirty(spa
->spa_root_vdev
);
9160 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
9163 spa_history_log_internal(spa
, "set", tx
,
9164 "%s=%s", nvpair_name(elem
), strval
);
9167 case ZPOOL_PROP_INVAL
:
9168 if (zpool_prop_feature(elemname
)) {
9169 fname
= strchr(elemname
, '@') + 1;
9170 VERIFY0(zfeature_lookup_name(fname
, &fid
));
9172 spa_feature_enable(spa
, fid
, tx
);
9173 spa_history_log_internal(spa
, "set", tx
,
9174 "%s=enabled", elemname
);
9176 } else if (!zfs_prop_user(elemname
)) {
9177 ASSERT(zpool_prop_feature(elemname
));
9183 * Set pool property values in the poolprops mos object.
9185 if (spa
->spa_pool_props_object
== 0) {
9186 spa
->spa_pool_props_object
=
9187 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
9188 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
9192 /* normalize the property name */
9193 if (prop
== ZPOOL_PROP_INVAL
) {
9194 propname
= elemname
;
9195 proptype
= PROP_TYPE_STRING
;
9197 propname
= zpool_prop_to_name(prop
);
9198 proptype
= zpool_prop_get_type(prop
);
9201 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
9202 ASSERT(proptype
== PROP_TYPE_STRING
);
9203 strval
= fnvpair_value_string(elem
);
9204 VERIFY0(zap_update(mos
,
9205 spa
->spa_pool_props_object
, propname
,
9206 1, strlen(strval
) + 1, strval
, tx
));
9207 spa_history_log_internal(spa
, "set", tx
,
9208 "%s=%s", elemname
, strval
);
9209 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
9210 intval
= fnvpair_value_uint64(elem
);
9212 if (proptype
== PROP_TYPE_INDEX
) {
9214 VERIFY0(zpool_prop_index_to_string(
9215 prop
, intval
, &unused
));
9217 VERIFY0(zap_update(mos
,
9218 spa
->spa_pool_props_object
, propname
,
9219 8, 1, &intval
, tx
));
9220 spa_history_log_internal(spa
, "set", tx
,
9221 "%s=%lld", elemname
,
9222 (longlong_t
)intval
);
9225 case ZPOOL_PROP_DELEGATION
:
9226 spa
->spa_delegation
= intval
;
9228 case ZPOOL_PROP_BOOTFS
:
9229 spa
->spa_bootfs
= intval
;
9231 case ZPOOL_PROP_FAILUREMODE
:
9232 spa
->spa_failmode
= intval
;
9234 case ZPOOL_PROP_AUTOTRIM
:
9235 spa
->spa_autotrim
= intval
;
9236 spa_async_request(spa
,
9237 SPA_ASYNC_AUTOTRIM_RESTART
);
9239 case ZPOOL_PROP_AUTOEXPAND
:
9240 spa
->spa_autoexpand
= intval
;
9241 if (tx
->tx_txg
!= TXG_INITIAL
)
9242 spa_async_request(spa
,
9243 SPA_ASYNC_AUTOEXPAND
);
9245 case ZPOOL_PROP_MULTIHOST
:
9246 spa
->spa_multihost
= intval
;
9252 ASSERT(0); /* not allowed */
9258 mutex_exit(&spa
->spa_props_lock
);
9262 * Perform one-time upgrade on-disk changes. spa_version() does not
9263 * reflect the new version this txg, so there must be no changes this
9264 * txg to anything that the upgrade code depends on after it executes.
9265 * Therefore this must be called after dsl_pool_sync() does the sync
9269 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
9271 if (spa_sync_pass(spa
) != 1)
9274 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9275 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
9277 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
9278 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
9279 dsl_pool_create_origin(dp
, tx
);
9281 /* Keeping the origin open increases spa_minref */
9282 spa
->spa_minref
+= 3;
9285 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
9286 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
9287 dsl_pool_upgrade_clones(dp
, tx
);
9290 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
9291 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
9292 dsl_pool_upgrade_dir_clones(dp
, tx
);
9294 /* Keeping the freedir open increases spa_minref */
9295 spa
->spa_minref
+= 3;
9298 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
9299 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
9300 spa_feature_create_zap_objects(spa
, tx
);
9304 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
9305 * when possibility to use lz4 compression for metadata was added
9306 * Old pools that have this feature enabled must be upgraded to have
9307 * this feature active
9309 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
9310 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
9311 SPA_FEATURE_LZ4_COMPRESS
);
9312 boolean_t lz4_ac
= spa_feature_is_active(spa
,
9313 SPA_FEATURE_LZ4_COMPRESS
);
9315 if (lz4_en
&& !lz4_ac
)
9316 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
9320 * If we haven't written the salt, do so now. Note that the
9321 * feature may not be activated yet, but that's fine since
9322 * the presence of this ZAP entry is backwards compatible.
9324 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
9325 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
9326 VERIFY0(zap_add(spa
->spa_meta_objset
,
9327 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
9328 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
9329 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
9332 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
9336 vdev_indirect_state_sync_verify(vdev_t
*vd
)
9338 vdev_indirect_mapping_t
*vim __maybe_unused
= vd
->vdev_indirect_mapping
;
9339 vdev_indirect_births_t
*vib __maybe_unused
= vd
->vdev_indirect_births
;
9341 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
9342 ASSERT(vim
!= NULL
);
9343 ASSERT(vib
!= NULL
);
9346 uint64_t obsolete_sm_object
= 0;
9347 ASSERT0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
9348 if (obsolete_sm_object
!= 0) {
9349 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
9350 ASSERT(vd
->vdev_removing
||
9351 vd
->vdev_ops
== &vdev_indirect_ops
);
9352 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
9353 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
9354 ASSERT3U(obsolete_sm_object
, ==,
9355 space_map_object(vd
->vdev_obsolete_sm
));
9356 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
9357 space_map_allocated(vd
->vdev_obsolete_sm
));
9359 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
9362 * Since frees / remaps to an indirect vdev can only
9363 * happen in syncing context, the obsolete segments
9364 * tree must be empty when we start syncing.
9366 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
9370 * Set the top-level vdev's max queue depth. Evaluate each top-level's
9371 * async write queue depth in case it changed. The max queue depth will
9372 * not change in the middle of syncing out this txg.
9375 spa_sync_adjust_vdev_max_queue_depth(spa_t
*spa
)
9377 ASSERT(spa_writeable(spa
));
9379 vdev_t
*rvd
= spa
->spa_root_vdev
;
9380 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
9381 zfs_vdev_queue_depth_pct
/ 100;
9382 metaslab_class_t
*normal
= spa_normal_class(spa
);
9383 metaslab_class_t
*special
= spa_special_class(spa
);
9384 metaslab_class_t
*dedup
= spa_dedup_class(spa
);
9386 uint64_t slots_per_allocator
= 0;
9387 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
9388 vdev_t
*tvd
= rvd
->vdev_child
[c
];
9390 metaslab_group_t
*mg
= tvd
->vdev_mg
;
9391 if (mg
== NULL
|| !metaslab_group_initialized(mg
))
9394 metaslab_class_t
*mc
= mg
->mg_class
;
9395 if (mc
!= normal
&& mc
!= special
&& mc
!= dedup
)
9399 * It is safe to do a lock-free check here because only async
9400 * allocations look at mg_max_alloc_queue_depth, and async
9401 * allocations all happen from spa_sync().
9403 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
9404 ASSERT0(zfs_refcount_count(
9405 &(mg
->mg_allocator
[i
].mga_alloc_queue_depth
)));
9407 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
9409 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
9410 mg
->mg_allocator
[i
].mga_cur_max_alloc_queue_depth
=
9411 zfs_vdev_def_queue_depth
;
9413 slots_per_allocator
+= zfs_vdev_def_queue_depth
;
9416 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9417 ASSERT0(zfs_refcount_count(&normal
->mc_allocator
[i
].
9419 ASSERT0(zfs_refcount_count(&special
->mc_allocator
[i
].
9421 ASSERT0(zfs_refcount_count(&dedup
->mc_allocator
[i
].
9423 normal
->mc_allocator
[i
].mca_alloc_max_slots
=
9424 slots_per_allocator
;
9425 special
->mc_allocator
[i
].mca_alloc_max_slots
=
9426 slots_per_allocator
;
9427 dedup
->mc_allocator
[i
].mca_alloc_max_slots
=
9428 slots_per_allocator
;
9430 normal
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9431 special
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9432 dedup
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9436 spa_sync_condense_indirect(spa_t
*spa
, dmu_tx_t
*tx
)
9438 ASSERT(spa_writeable(spa
));
9440 vdev_t
*rvd
= spa
->spa_root_vdev
;
9441 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
9442 vdev_t
*vd
= rvd
->vdev_child
[c
];
9443 vdev_indirect_state_sync_verify(vd
);
9445 if (vdev_indirect_should_condense(vd
)) {
9446 spa_condense_indirect_start_sync(vd
, tx
);
9453 spa_sync_iterate_to_convergence(spa_t
*spa
, dmu_tx_t
*tx
)
9455 objset_t
*mos
= spa
->spa_meta_objset
;
9456 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9457 uint64_t txg
= tx
->tx_txg
;
9458 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
9461 int pass
= ++spa
->spa_sync_pass
;
9463 spa_sync_config_object(spa
, tx
);
9464 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
9465 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
9466 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
9467 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
9468 spa_errlog_sync(spa
, txg
);
9469 dsl_pool_sync(dp
, txg
);
9471 if (pass
< zfs_sync_pass_deferred_free
||
9472 spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
)) {
9474 * If the log space map feature is active we don't
9475 * care about deferred frees and the deferred bpobj
9476 * as the log space map should effectively have the
9477 * same results (i.e. appending only to one object).
9479 spa_sync_frees(spa
, free_bpl
, tx
);
9482 * We can not defer frees in pass 1, because
9483 * we sync the deferred frees later in pass 1.
9485 ASSERT3U(pass
, >, 1);
9486 bplist_iterate(free_bpl
, bpobj_enqueue_alloc_cb
,
9487 &spa
->spa_deferred_bpobj
, tx
);
9492 dsl_scan_sync(dp
, tx
);
9493 dsl_errorscrub_sync(dp
, tx
);
9495 spa_sync_upgrades(spa
, tx
);
9497 spa_flush_metaslabs(spa
, tx
);
9500 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
9506 * dsl_pool_sync() -> dp_sync_tasks may have dirtied
9507 * the config. If that happens, this txg should not
9508 * be a no-op. So we must sync the config to the MOS
9509 * before checking for no-op.
9511 * Note that when the config is dirty, it will
9512 * be written to the MOS (i.e. the MOS will be
9513 * dirtied) every time we call spa_sync_config_object()
9514 * in this txg. Therefore we can't call this after
9515 * dsl_pool_sync() every pass, because it would
9516 * prevent us from converging, since we'd dirty
9517 * the MOS every pass.
9519 * Sync tasks can only be processed in pass 1, so
9520 * there's no need to do this in later passes.
9522 spa_sync_config_object(spa
, tx
);
9526 * Note: We need to check if the MOS is dirty because we could
9527 * have marked the MOS dirty without updating the uberblock
9528 * (e.g. if we have sync tasks but no dirty user data). We need
9529 * to check the uberblock's rootbp because it is updated if we
9530 * have synced out dirty data (though in this case the MOS will
9531 * most likely also be dirty due to second order effects, we
9532 * don't want to rely on that here).
9535 spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
9536 !dmu_objset_is_dirty(mos
, txg
)) {
9538 * Nothing changed on the first pass, therefore this
9539 * TXG is a no-op. Avoid syncing deferred frees, so
9540 * that we can keep this TXG as a no-op.
9542 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9543 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9544 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
9545 ASSERT(txg_list_empty(&dp
->dp_early_sync_tasks
, txg
));
9549 spa_sync_deferred_frees(spa
, tx
);
9550 } while (dmu_objset_is_dirty(mos
, txg
));
9554 * Rewrite the vdev configuration (which includes the uberblock) to
9555 * commit the transaction group.
9557 * If there are no dirty vdevs, we sync the uberblock to a few random
9558 * top-level vdevs that are known to be visible in the config cache
9559 * (see spa_vdev_add() for a complete description). If there *are* dirty
9560 * vdevs, sync the uberblock to all vdevs.
9563 spa_sync_rewrite_vdev_config(spa_t
*spa
, dmu_tx_t
*tx
)
9565 vdev_t
*rvd
= spa
->spa_root_vdev
;
9566 uint64_t txg
= tx
->tx_txg
;
9572 * We hold SCL_STATE to prevent vdev open/close/etc.
9573 * while we're attempting to write the vdev labels.
9575 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9577 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
9578 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
9580 int children
= rvd
->vdev_children
;
9581 int c0
= random_in_range(children
);
9583 for (int c
= 0; c
< children
; c
++) {
9585 rvd
->vdev_child
[(c0
+ c
) % children
];
9587 /* Stop when revisiting the first vdev */
9588 if (c
> 0 && svd
[0] == vd
)
9591 if (vd
->vdev_ms_array
== 0 ||
9593 !vdev_is_concrete(vd
))
9596 svd
[svdcount
++] = vd
;
9597 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
9600 error
= vdev_config_sync(svd
, svdcount
, txg
);
9602 error
= vdev_config_sync(rvd
->vdev_child
,
9603 rvd
->vdev_children
, txg
);
9607 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
9609 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9613 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
9614 zio_resume_wait(spa
);
9619 * Sync the specified transaction group. New blocks may be dirtied as
9620 * part of the process, so we iterate until it converges.
9623 spa_sync(spa_t
*spa
, uint64_t txg
)
9627 VERIFY(spa_writeable(spa
));
9630 * Wait for i/os issued in open context that need to complete
9631 * before this txg syncs.
9633 (void) zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]);
9634 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
,
9638 * Now that there can be no more cloning in this transaction group,
9639 * but we are still before issuing frees, we can process pending BRT
9642 brt_pending_apply(spa
, txg
);
9645 * Lock out configuration changes.
9647 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
9649 spa
->spa_syncing_txg
= txg
;
9650 spa
->spa_sync_pass
= 0;
9652 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9653 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
9654 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
9655 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
9659 * If there are any pending vdev state changes, convert them
9660 * into config changes that go out with this transaction group.
9662 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9663 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
9664 /* Avoid holding the write lock unless actually necessary */
9665 if (vd
->vdev_aux
== NULL
) {
9666 vdev_state_clean(vd
);
9667 vdev_config_dirty(vd
);
9671 * We need the write lock here because, for aux vdevs,
9672 * calling vdev_config_dirty() modifies sav_config.
9673 * This is ugly and will become unnecessary when we
9674 * eliminate the aux vdev wart by integrating all vdevs
9675 * into the root vdev tree.
9677 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9678 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
9679 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
9680 vdev_state_clean(vd
);
9681 vdev_config_dirty(vd
);
9683 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9684 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9686 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9688 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9689 dmu_tx_t
*tx
= dmu_tx_create_assigned(dp
, txg
);
9691 spa
->spa_sync_starttime
= gethrtime();
9692 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9693 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
9694 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
9695 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
9698 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9699 * set spa_deflate if we have no raid-z vdevs.
9701 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
9702 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
9703 vdev_t
*rvd
= spa
->spa_root_vdev
;
9706 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
9707 vd
= rvd
->vdev_child
[i
];
9708 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
9711 if (i
== rvd
->vdev_children
) {
9712 spa
->spa_deflate
= TRUE
;
9713 VERIFY0(zap_add(spa
->spa_meta_objset
,
9714 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
9715 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
9719 spa_sync_adjust_vdev_max_queue_depth(spa
);
9721 spa_sync_condense_indirect(spa
, tx
);
9723 spa_sync_iterate_to_convergence(spa
, tx
);
9726 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
9728 * Make sure that the number of ZAPs for all the vdevs matches
9729 * the number of ZAPs in the per-vdev ZAP list. This only gets
9730 * called if the config is dirty; otherwise there may be
9731 * outstanding AVZ operations that weren't completed in
9732 * spa_sync_config_object.
9734 uint64_t all_vdev_zap_entry_count
;
9735 ASSERT0(zap_count(spa
->spa_meta_objset
,
9736 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
9737 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
9738 all_vdev_zap_entry_count
);
9742 if (spa
->spa_vdev_removal
!= NULL
) {
9743 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
9746 spa_sync_rewrite_vdev_config(spa
, tx
);
9749 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9750 spa
->spa_deadman_tqid
= 0;
9753 * Clear the dirty config list.
9755 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
9756 vdev_config_clean(vd
);
9759 * Now that the new config has synced transactionally,
9760 * let it become visible to the config cache.
9762 if (spa
->spa_config_syncing
!= NULL
) {
9763 spa_config_set(spa
, spa
->spa_config_syncing
);
9764 spa
->spa_config_txg
= txg
;
9765 spa
->spa_config_syncing
= NULL
;
9768 dsl_pool_sync_done(dp
, txg
);
9770 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9771 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
9772 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
9773 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
9777 * Update usable space statistics.
9779 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
9781 vdev_sync_done(vd
, txg
);
9783 metaslab_class_evict_old(spa
->spa_normal_class
, txg
);
9784 metaslab_class_evict_old(spa
->spa_log_class
, txg
);
9786 spa_sync_close_syncing_log_sm(spa
);
9788 spa_update_dspace(spa
);
9790 if (spa_get_autotrim(spa
) == SPA_AUTOTRIM_ON
)
9791 vdev_autotrim_kick(spa
);
9794 * It had better be the case that we didn't dirty anything
9795 * since vdev_config_sync().
9797 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9798 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9799 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
9801 while (zfs_pause_spa_sync
)
9804 spa
->spa_sync_pass
= 0;
9807 * Update the last synced uberblock here. We want to do this at
9808 * the end of spa_sync() so that consumers of spa_last_synced_txg()
9809 * will be guaranteed that all the processing associated with
9810 * that txg has been completed.
9812 spa
->spa_ubsync
= spa
->spa_uberblock
;
9813 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
9815 spa_handle_ignored_writes(spa
);
9818 * If any async tasks have been requested, kick them off.
9820 spa_async_dispatch(spa
);
9824 * Sync all pools. We don't want to hold the namespace lock across these
9825 * operations, so we take a reference on the spa_t and drop the lock during the
9829 spa_sync_allpools(void)
9832 mutex_enter(&spa_namespace_lock
);
9833 while ((spa
= spa_next(spa
)) != NULL
) {
9834 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
9835 !spa_writeable(spa
) || spa_suspended(spa
))
9837 spa_open_ref(spa
, FTAG
);
9838 mutex_exit(&spa_namespace_lock
);
9839 txg_wait_synced(spa_get_dsl(spa
), 0);
9840 mutex_enter(&spa_namespace_lock
);
9841 spa_close(spa
, FTAG
);
9843 mutex_exit(&spa_namespace_lock
);
9847 spa_sync_tq_create(spa_t
*spa
, const char *name
)
9849 kthread_t
**kthreads
;
9851 ASSERT(spa
->spa_sync_tq
== NULL
);
9852 ASSERT3S(spa
->spa_alloc_count
, <=, boot_ncpus
);
9855 * - do not allow more allocators than cpus.
9856 * - there may be more cpus than allocators.
9857 * - do not allow more sync taskq threads than allocators or cpus.
9859 int nthreads
= spa
->spa_alloc_count
;
9860 spa
->spa_syncthreads
= kmem_zalloc(sizeof (spa_syncthread_info_t
) *
9861 nthreads
, KM_SLEEP
);
9863 spa
->spa_sync_tq
= taskq_create_synced(name
, nthreads
, minclsyspri
,
9864 nthreads
, INT_MAX
, TASKQ_PREPOPULATE
, &kthreads
);
9865 VERIFY(spa
->spa_sync_tq
!= NULL
);
9866 VERIFY(kthreads
!= NULL
);
9869 &spa
->spa_zio_taskq
[ZIO_TYPE_WRITE
][ZIO_TASKQ_ISSUE
];
9871 spa_syncthread_info_t
*ti
= spa
->spa_syncthreads
;
9872 for (int i
= 0, w
= 0; i
< nthreads
; i
++, w
++, ti
++) {
9873 ti
->sti_thread
= kthreads
[i
];
9874 if (w
== tqs
->stqs_count
) {
9877 ti
->sti_wr_iss_tq
= tqs
->stqs_taskq
[w
];
9880 kmem_free(kthreads
, sizeof (*kthreads
) * nthreads
);
9881 return (spa
->spa_sync_tq
);
9885 spa_sync_tq_destroy(spa_t
*spa
)
9887 ASSERT(spa
->spa_sync_tq
!= NULL
);
9889 taskq_wait(spa
->spa_sync_tq
);
9890 taskq_destroy(spa
->spa_sync_tq
);
9891 kmem_free(spa
->spa_syncthreads
,
9892 sizeof (spa_syncthread_info_t
) * spa
->spa_alloc_count
);
9893 spa
->spa_sync_tq
= NULL
;
9897 spa_select_allocator(zio_t
*zio
)
9899 zbookmark_phys_t
*bm
= &zio
->io_bookmark
;
9900 spa_t
*spa
= zio
->io_spa
;
9902 ASSERT(zio
->io_type
== ZIO_TYPE_WRITE
);
9905 * A gang block (for example) may have inherited its parent's
9906 * allocator, in which case there is nothing further to do here.
9908 if (ZIO_HAS_ALLOCATOR(zio
))
9911 ASSERT(spa
!= NULL
);
9915 * First try to use an allocator assigned to the syncthread, and set
9916 * the corresponding write issue taskq for the allocator.
9917 * Note, we must have an open pool to do this.
9919 if (spa
->spa_sync_tq
!= NULL
) {
9920 spa_syncthread_info_t
*ti
= spa
->spa_syncthreads
;
9921 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++, ti
++) {
9922 if (ti
->sti_thread
== curthread
) {
9923 zio
->io_allocator
= i
;
9924 zio
->io_wr_iss_tq
= ti
->sti_wr_iss_tq
;
9931 * We want to try to use as many allocators as possible to help improve
9932 * performance, but we also want logically adjacent IOs to be physically
9933 * adjacent to improve sequential read performance. We chunk each object
9934 * into 2^20 block regions, and then hash based on the objset, object,
9935 * level, and region to accomplish both of these goals.
9937 uint64_t hv
= cityhash4(bm
->zb_objset
, bm
->zb_object
, bm
->zb_level
,
9938 bm
->zb_blkid
>> 20);
9940 zio
->io_allocator
= (uint_t
)hv
% spa
->spa_alloc_count
;
9941 zio
->io_wr_iss_tq
= NULL
;
9945 * ==========================================================================
9946 * Miscellaneous routines
9947 * ==========================================================================
9951 * Remove all pools in the system.
9959 * Remove all cached state. All pools should be closed now,
9960 * so every spa in the AVL tree should be unreferenced.
9962 mutex_enter(&spa_namespace_lock
);
9963 while ((spa
= spa_next(NULL
)) != NULL
) {
9965 * Stop async tasks. The async thread may need to detach
9966 * a device that's been replaced, which requires grabbing
9967 * spa_namespace_lock, so we must drop it here.
9969 spa_open_ref(spa
, FTAG
);
9970 mutex_exit(&spa_namespace_lock
);
9971 spa_async_suspend(spa
);
9972 mutex_enter(&spa_namespace_lock
);
9973 spa_close(spa
, FTAG
);
9975 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
9977 spa_deactivate(spa
);
9981 mutex_exit(&spa_namespace_lock
);
9985 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
9990 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
9994 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
9995 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
9996 if (vd
->vdev_guid
== guid
)
10000 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
10001 vd
= spa
->spa_spares
.sav_vdevs
[i
];
10002 if (vd
->vdev_guid
== guid
)
10011 spa_upgrade(spa_t
*spa
, uint64_t version
)
10013 ASSERT(spa_writeable(spa
));
10015 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
10018 * This should only be called for a non-faulted pool, and since a
10019 * future version would result in an unopenable pool, this shouldn't be
10022 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
10023 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
10025 spa
->spa_uberblock
.ub_version
= version
;
10026 vdev_config_dirty(spa
->spa_root_vdev
);
10028 spa_config_exit(spa
, SCL_ALL
, FTAG
);
10030 txg_wait_synced(spa_get_dsl(spa
), 0);
10034 spa_has_aux_vdev(spa_t
*spa
, uint64_t guid
, spa_aux_vdev_t
*sav
)
10038 uint64_t vdev_guid
;
10040 for (i
= 0; i
< sav
->sav_count
; i
++)
10041 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
10044 for (i
= 0; i
< sav
->sav_npending
; i
++) {
10045 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
10046 &vdev_guid
) == 0 && vdev_guid
== guid
)
10054 spa_has_l2cache(spa_t
*spa
, uint64_t guid
)
10056 return (spa_has_aux_vdev(spa
, guid
, &spa
->spa_l2cache
));
10060 spa_has_spare(spa_t
*spa
, uint64_t guid
)
10062 return (spa_has_aux_vdev(spa
, guid
, &spa
->spa_spares
));
10066 * Check if a pool has an active shared spare device.
10067 * Note: reference count of an active spare is 2, as a spare and as a replace
10070 spa_has_active_shared_spare(spa_t
*spa
)
10074 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
10076 for (i
= 0; i
< sav
->sav_count
; i
++) {
10077 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
10078 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
10087 spa_total_metaslabs(spa_t
*spa
)
10089 vdev_t
*rvd
= spa
->spa_root_vdev
;
10092 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
10093 vdev_t
*vd
= rvd
->vdev_child
[c
];
10094 if (!vdev_is_concrete(vd
))
10096 m
+= vd
->vdev_ms_count
;
10102 * Notify any waiting threads that some activity has switched from being in-
10103 * progress to not-in-progress so that the thread can wake up and determine
10104 * whether it is finished waiting.
10107 spa_notify_waiters(spa_t
*spa
)
10110 * Acquiring spa_activities_lock here prevents the cv_broadcast from
10111 * happening between the waiting thread's check and cv_wait.
10113 mutex_enter(&spa
->spa_activities_lock
);
10114 cv_broadcast(&spa
->spa_activities_cv
);
10115 mutex_exit(&spa
->spa_activities_lock
);
10119 * Notify any waiting threads that the pool is exporting, and then block until
10120 * they are finished using the spa_t.
10123 spa_wake_waiters(spa_t
*spa
)
10125 mutex_enter(&spa
->spa_activities_lock
);
10126 spa
->spa_waiters_cancel
= B_TRUE
;
10127 cv_broadcast(&spa
->spa_activities_cv
);
10128 while (spa
->spa_waiters
!= 0)
10129 cv_wait(&spa
->spa_waiters_cv
, &spa
->spa_activities_lock
);
10130 spa
->spa_waiters_cancel
= B_FALSE
;
10131 mutex_exit(&spa
->spa_activities_lock
);
10134 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
10136 spa_vdev_activity_in_progress_impl(vdev_t
*vd
, zpool_wait_activity_t activity
)
10138 spa_t
*spa
= vd
->vdev_spa
;
10140 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
));
10141 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
10142 ASSERT(activity
== ZPOOL_WAIT_INITIALIZE
||
10143 activity
== ZPOOL_WAIT_TRIM
);
10145 kmutex_t
*lock
= activity
== ZPOOL_WAIT_INITIALIZE
?
10146 &vd
->vdev_initialize_lock
: &vd
->vdev_trim_lock
;
10148 mutex_exit(&spa
->spa_activities_lock
);
10150 mutex_enter(&spa
->spa_activities_lock
);
10152 boolean_t in_progress
= (activity
== ZPOOL_WAIT_INITIALIZE
) ?
10153 (vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) :
10154 (vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
);
10160 for (int i
= 0; i
< vd
->vdev_children
; i
++) {
10161 if (spa_vdev_activity_in_progress_impl(vd
->vdev_child
[i
],
10170 * If use_guid is true, this checks whether the vdev specified by guid is
10171 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
10172 * is being initialized/trimmed. The caller must hold the config lock and
10173 * spa_activities_lock.
10176 spa_vdev_activity_in_progress(spa_t
*spa
, boolean_t use_guid
, uint64_t guid
,
10177 zpool_wait_activity_t activity
, boolean_t
*in_progress
)
10179 mutex_exit(&spa
->spa_activities_lock
);
10180 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
10181 mutex_enter(&spa
->spa_activities_lock
);
10185 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
10186 if (vd
== NULL
|| !vd
->vdev_ops
->vdev_op_leaf
) {
10187 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
10191 vd
= spa
->spa_root_vdev
;
10194 *in_progress
= spa_vdev_activity_in_progress_impl(vd
, activity
);
10196 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
10201 * Locking for waiting threads
10202 * ---------------------------
10204 * Waiting threads need a way to check whether a given activity is in progress,
10205 * and then, if it is, wait for it to complete. Each activity will have some
10206 * in-memory representation of the relevant on-disk state which can be used to
10207 * determine whether or not the activity is in progress. The in-memory state and
10208 * the locking used to protect it will be different for each activity, and may
10209 * not be suitable for use with a cvar (e.g., some state is protected by the
10210 * config lock). To allow waiting threads to wait without any races, another
10211 * lock, spa_activities_lock, is used.
10213 * When the state is checked, both the activity-specific lock (if there is one)
10214 * and spa_activities_lock are held. In some cases, the activity-specific lock
10215 * is acquired explicitly (e.g. the config lock). In others, the locking is
10216 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
10217 * thread releases the activity-specific lock and, if the activity is in
10218 * progress, then cv_waits using spa_activities_lock.
10220 * The waiting thread is woken when another thread, one completing some
10221 * activity, updates the state of the activity and then calls
10222 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
10223 * needs to hold its activity-specific lock when updating the state, and this
10224 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
10226 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
10227 * and because it is held when the waiting thread checks the state of the
10228 * activity, it can never be the case that the completing thread both updates
10229 * the activity state and cv_broadcasts in between the waiting thread's check
10230 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
10232 * In order to prevent deadlock, when the waiting thread does its check, in some
10233 * cases it will temporarily drop spa_activities_lock in order to acquire the
10234 * activity-specific lock. The order in which spa_activities_lock and the
10235 * activity specific lock are acquired in the waiting thread is determined by
10236 * the order in which they are acquired in the completing thread; if the
10237 * completing thread calls spa_notify_waiters with the activity-specific lock
10238 * held, then the waiting thread must also acquire the activity-specific lock
10243 spa_activity_in_progress(spa_t
*spa
, zpool_wait_activity_t activity
,
10244 boolean_t use_tag
, uint64_t tag
, boolean_t
*in_progress
)
10248 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
10250 switch (activity
) {
10251 case ZPOOL_WAIT_CKPT_DISCARD
:
10253 (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
) &&
10254 zap_contains(spa_meta_objset(spa
),
10255 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ZPOOL_CHECKPOINT
) ==
10258 case ZPOOL_WAIT_FREE
:
10259 *in_progress
= ((spa_version(spa
) >= SPA_VERSION_DEADLISTS
&&
10260 !bpobj_is_empty(&spa
->spa_dsl_pool
->dp_free_bpobj
)) ||
10261 spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
) ||
10262 spa_livelist_delete_check(spa
));
10264 case ZPOOL_WAIT_INITIALIZE
:
10265 case ZPOOL_WAIT_TRIM
:
10266 error
= spa_vdev_activity_in_progress(spa
, use_tag
, tag
,
10267 activity
, in_progress
);
10269 case ZPOOL_WAIT_REPLACE
:
10270 mutex_exit(&spa
->spa_activities_lock
);
10271 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
10272 mutex_enter(&spa
->spa_activities_lock
);
10274 *in_progress
= vdev_replace_in_progress(spa
->spa_root_vdev
);
10275 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
10277 case ZPOOL_WAIT_REMOVE
:
10278 *in_progress
= (spa
->spa_removing_phys
.sr_state
==
10281 case ZPOOL_WAIT_RESILVER
:
10282 *in_progress
= vdev_rebuild_active(spa
->spa_root_vdev
);
10286 case ZPOOL_WAIT_SCRUB
:
10288 boolean_t scanning
, paused
, is_scrub
;
10289 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
10291 is_scrub
= (scn
->scn_phys
.scn_func
== POOL_SCAN_SCRUB
);
10292 scanning
= (scn
->scn_phys
.scn_state
== DSS_SCANNING
);
10293 paused
= dsl_scan_is_paused_scrub(scn
);
10294 *in_progress
= (scanning
&& !paused
&&
10295 is_scrub
== (activity
== ZPOOL_WAIT_SCRUB
));
10298 case ZPOOL_WAIT_RAIDZ_EXPAND
:
10300 vdev_raidz_expand_t
*vre
= spa
->spa_raidz_expand
;
10301 *in_progress
= (vre
!= NULL
&& vre
->vre_state
== DSS_SCANNING
);
10305 panic("unrecognized value for activity %d", activity
);
10312 spa_wait_common(const char *pool
, zpool_wait_activity_t activity
,
10313 boolean_t use_tag
, uint64_t tag
, boolean_t
*waited
)
10316 * The tag is used to distinguish between instances of an activity.
10317 * 'initialize' and 'trim' are the only activities that we use this for.
10318 * The other activities can only have a single instance in progress in a
10319 * pool at one time, making the tag unnecessary.
10321 * There can be multiple devices being replaced at once, but since they
10322 * all finish once resilvering finishes, we don't bother keeping track
10323 * of them individually, we just wait for them all to finish.
10325 if (use_tag
&& activity
!= ZPOOL_WAIT_INITIALIZE
&&
10326 activity
!= ZPOOL_WAIT_TRIM
)
10329 if (activity
< 0 || activity
>= ZPOOL_WAIT_NUM_ACTIVITIES
)
10333 int error
= spa_open(pool
, &spa
, FTAG
);
10338 * Increment the spa's waiter count so that we can call spa_close and
10339 * still ensure that the spa_t doesn't get freed before this thread is
10340 * finished with it when the pool is exported. We want to call spa_close
10341 * before we start waiting because otherwise the additional ref would
10342 * prevent the pool from being exported or destroyed throughout the
10343 * potentially long wait.
10345 mutex_enter(&spa
->spa_activities_lock
);
10346 spa
->spa_waiters
++;
10347 spa_close(spa
, FTAG
);
10351 boolean_t in_progress
;
10352 error
= spa_activity_in_progress(spa
, activity
, use_tag
, tag
,
10355 if (error
|| !in_progress
|| spa
->spa_waiters_cancel
)
10360 if (cv_wait_sig(&spa
->spa_activities_cv
,
10361 &spa
->spa_activities_lock
) == 0) {
10367 spa
->spa_waiters
--;
10368 cv_signal(&spa
->spa_waiters_cv
);
10369 mutex_exit(&spa
->spa_activities_lock
);
10375 * Wait for a particular instance of the specified activity to complete, where
10376 * the instance is identified by 'tag'
10379 spa_wait_tag(const char *pool
, zpool_wait_activity_t activity
, uint64_t tag
,
10382 return (spa_wait_common(pool
, activity
, B_TRUE
, tag
, waited
));
10386 * Wait for all instances of the specified activity complete
10389 spa_wait(const char *pool
, zpool_wait_activity_t activity
, boolean_t
*waited
)
10392 return (spa_wait_common(pool
, activity
, B_FALSE
, 0, waited
));
10396 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
10398 sysevent_t
*ev
= NULL
;
10400 nvlist_t
*resource
;
10402 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
10404 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
10405 ev
->resource
= resource
;
10408 (void) spa
, (void) vd
, (void) hist_nvl
, (void) name
;
10414 spa_event_post(sysevent_t
*ev
)
10418 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
10419 kmem_free(ev
, sizeof (*ev
));
10427 * Post a zevent corresponding to the given sysevent. The 'name' must be one
10428 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
10429 * filled in from the spa and (optionally) the vdev. This doesn't do anything
10430 * in the userland libzpool, as we don't want consumers to misinterpret ztest
10431 * or zdb as real changes.
10434 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
10436 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
10439 /* state manipulation functions */
10440 EXPORT_SYMBOL(spa_open
);
10441 EXPORT_SYMBOL(spa_open_rewind
);
10442 EXPORT_SYMBOL(spa_get_stats
);
10443 EXPORT_SYMBOL(spa_create
);
10444 EXPORT_SYMBOL(spa_import
);
10445 EXPORT_SYMBOL(spa_tryimport
);
10446 EXPORT_SYMBOL(spa_destroy
);
10447 EXPORT_SYMBOL(spa_export
);
10448 EXPORT_SYMBOL(spa_reset
);
10449 EXPORT_SYMBOL(spa_async_request
);
10450 EXPORT_SYMBOL(spa_async_suspend
);
10451 EXPORT_SYMBOL(spa_async_resume
);
10452 EXPORT_SYMBOL(spa_inject_addref
);
10453 EXPORT_SYMBOL(spa_inject_delref
);
10454 EXPORT_SYMBOL(spa_scan_stat_init
);
10455 EXPORT_SYMBOL(spa_scan_get_stats
);
10457 /* device manipulation */
10458 EXPORT_SYMBOL(spa_vdev_add
);
10459 EXPORT_SYMBOL(spa_vdev_attach
);
10460 EXPORT_SYMBOL(spa_vdev_detach
);
10461 EXPORT_SYMBOL(spa_vdev_setpath
);
10462 EXPORT_SYMBOL(spa_vdev_setfru
);
10463 EXPORT_SYMBOL(spa_vdev_split_mirror
);
10465 /* spare statech is global across all pools) */
10466 EXPORT_SYMBOL(spa_spare_add
);
10467 EXPORT_SYMBOL(spa_spare_remove
);
10468 EXPORT_SYMBOL(spa_spare_exists
);
10469 EXPORT_SYMBOL(spa_spare_activate
);
10471 /* L2ARC statech is global across all pools) */
10472 EXPORT_SYMBOL(spa_l2cache_add
);
10473 EXPORT_SYMBOL(spa_l2cache_remove
);
10474 EXPORT_SYMBOL(spa_l2cache_exists
);
10475 EXPORT_SYMBOL(spa_l2cache_activate
);
10476 EXPORT_SYMBOL(spa_l2cache_drop
);
10479 EXPORT_SYMBOL(spa_scan
);
10480 EXPORT_SYMBOL(spa_scan_stop
);
10483 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
10484 EXPORT_SYMBOL(spa_sync_allpools
);
10487 EXPORT_SYMBOL(spa_prop_set
);
10488 EXPORT_SYMBOL(spa_prop_get
);
10489 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
10491 /* asynchronous event notification */
10492 EXPORT_SYMBOL(spa_event_notify
);
10494 ZFS_MODULE_PARAM(zfs_metaslab
, metaslab_
, preload_pct
, UINT
, ZMOD_RW
,
10495 "Percentage of CPUs to run a metaslab preload taskq");
10497 /* BEGIN CSTYLED */
10498 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_shift
, UINT
, ZMOD_RW
,
10499 "log2 fraction of arc that can be used by inflight I/Os when "
10500 "verifying pool during import");
10503 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_metadata
, INT
, ZMOD_RW
,
10504 "Set to traverse metadata on pool import");
10506 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_data
, INT
, ZMOD_RW
,
10507 "Set to traverse data on pool import");
10509 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_print_vdev_tree
, INT
, ZMOD_RW
,
10510 "Print vdev tree to zfs_dbgmsg during pool import");
10512 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_pct
, UINT
, ZMOD_RD
,
10513 "Percentage of CPUs to run an IO worker thread");
10515 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_tpq
, UINT
, ZMOD_RD
,
10516 "Number of threads per IO worker taskqueue");
10518 /* BEGIN CSTYLED */
10519 ZFS_MODULE_PARAM(zfs
, zfs_
, max_missing_tvds
, U64
, ZMOD_RW
,
10520 "Allow importing pool with up to this number of missing top-level "
10521 "vdevs (in read-only mode)");
10524 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_pause
, INT
,
10525 ZMOD_RW
, "Set the livelist condense zthr to pause");
10527 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_pause
, INT
,
10528 ZMOD_RW
, "Set the livelist condense synctask to pause");
10530 /* BEGIN CSTYLED */
10531 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_cancel
,
10533 "Whether livelist condensing was canceled in the synctask");
10535 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_cancel
,
10537 "Whether livelist condensing was canceled in the zthr function");
10539 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, new_alloc
, INT
,
10541 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
10542 "was being condensed");
10545 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_wr_iss_ncpus
, UINT
, ZMOD_RW
,
10546 "Number of CPUs to run write issue taskqs");