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
));
3113 gethrestime(&spa
->spa_loaded_ts
);
3114 error
= spa_load_impl(spa
, type
, &ereport
);
3117 * Don't count references from objsets that are already closed
3118 * and are making their way through the eviction process.
3120 spa_evicting_os_wait(spa
);
3121 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
3123 if (error
!= EEXIST
) {
3124 spa
->spa_loaded_ts
.tv_sec
= 0;
3125 spa
->spa_loaded_ts
.tv_nsec
= 0;
3127 if (error
!= EBADF
) {
3128 (void) zfs_ereport_post(ereport
, spa
,
3129 NULL
, NULL
, NULL
, 0);
3132 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
3135 (void) spa_import_progress_set_state(spa_guid(spa
),
3136 spa_load_state(spa
));
3143 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
3144 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
3145 * spa's per-vdev ZAP list.
3148 vdev_count_verify_zaps(vdev_t
*vd
)
3150 spa_t
*spa
= vd
->vdev_spa
;
3153 if (spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_AVZ_V2
) &&
3154 vd
->vdev_root_zap
!= 0) {
3156 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3157 spa
->spa_all_vdev_zaps
, vd
->vdev_root_zap
));
3159 if (vd
->vdev_top_zap
!= 0) {
3161 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3162 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
3164 if (vd
->vdev_leaf_zap
!= 0) {
3166 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3167 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
3170 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3171 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
3177 #define vdev_count_verify_zaps(vd) ((void) sizeof (vd), 0)
3181 * Determine whether the activity check is required.
3184 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
3188 uint64_t hostid
= 0;
3189 uint64_t tryconfig_txg
= 0;
3190 uint64_t tryconfig_timestamp
= 0;
3191 uint16_t tryconfig_mmp_seq
= 0;
3194 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3195 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
3196 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
3198 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
3199 &tryconfig_timestamp
);
3200 (void) nvlist_lookup_uint16(nvinfo
, ZPOOL_CONFIG_MMP_SEQ
,
3201 &tryconfig_mmp_seq
);
3204 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
3207 * Disable the MMP activity check - This is used by zdb which
3208 * is intended to be used on potentially active pools.
3210 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
3214 * Skip the activity check when the MMP feature is disabled.
3216 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
3220 * If the tryconfig_ values are nonzero, they are the results of an
3221 * earlier tryimport. If they all match the uberblock we just found,
3222 * then the pool has not changed and we return false so we do not test
3225 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
3226 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
&&
3227 tryconfig_mmp_seq
&& tryconfig_mmp_seq
==
3228 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0))
3232 * Allow the activity check to be skipped when importing the pool
3233 * on the same host which last imported it. Since the hostid from
3234 * configuration may be stale use the one read from the label.
3236 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
3237 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
3239 if (hostid
== spa_get_hostid(spa
))
3243 * Skip the activity test when the pool was cleanly exported.
3245 if (state
!= POOL_STATE_ACTIVE
)
3252 * Nanoseconds the activity check must watch for changes on-disk.
3255 spa_activity_check_duration(spa_t
*spa
, uberblock_t
*ub
)
3257 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
3258 uint64_t multihost_interval
= MSEC2NSEC(
3259 MMP_INTERVAL_OK(zfs_multihost_interval
));
3260 uint64_t import_delay
= MAX(NANOSEC
, import_intervals
*
3261 multihost_interval
);
3264 * Local tunables determine a minimum duration except for the case
3265 * where we know when the remote host will suspend the pool if MMP
3266 * writes do not land.
3268 * See Big Theory comment at the top of mmp.c for the reasoning behind
3269 * these cases and times.
3272 ASSERT(MMP_IMPORT_SAFETY_FACTOR
>= 100);
3274 if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3275 MMP_FAIL_INT(ub
) > 0) {
3277 /* MMP on remote host will suspend pool after failed writes */
3278 import_delay
= MMP_FAIL_INT(ub
) * MSEC2NSEC(MMP_INTERVAL(ub
)) *
3279 MMP_IMPORT_SAFETY_FACTOR
/ 100;
3281 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3282 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3283 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3284 (u_longlong_t
)MMP_FAIL_INT(ub
),
3285 (u_longlong_t
)MMP_INTERVAL(ub
),
3286 (u_longlong_t
)import_intervals
);
3288 } else if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3289 MMP_FAIL_INT(ub
) == 0) {
3291 /* MMP on remote host will never suspend pool */
3292 import_delay
= MAX(import_delay
, (MSEC2NSEC(MMP_INTERVAL(ub
)) +
3293 ub
->ub_mmp_delay
) * import_intervals
);
3295 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3296 "mmp_interval=%llu ub_mmp_delay=%llu "
3297 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3298 (u_longlong_t
)MMP_INTERVAL(ub
),
3299 (u_longlong_t
)ub
->ub_mmp_delay
,
3300 (u_longlong_t
)import_intervals
);
3302 } else if (MMP_VALID(ub
)) {
3304 * zfs-0.7 compatibility case
3307 import_delay
= MAX(import_delay
, (multihost_interval
+
3308 ub
->ub_mmp_delay
) * import_intervals
);
3310 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3311 "import_intervals=%llu leaves=%u",
3312 (u_longlong_t
)import_delay
,
3313 (u_longlong_t
)ub
->ub_mmp_delay
,
3314 (u_longlong_t
)import_intervals
,
3315 vdev_count_leaves(spa
));
3317 /* Using local tunings is the only reasonable option */
3318 zfs_dbgmsg("pool last imported on non-MMP aware "
3319 "host using import_delay=%llu multihost_interval=%llu "
3320 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3321 (u_longlong_t
)multihost_interval
,
3322 (u_longlong_t
)import_intervals
);
3325 return (import_delay
);
3329 * Perform the import activity check. If the user canceled the import or
3330 * we detected activity then fail.
3333 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
3335 uint64_t txg
= ub
->ub_txg
;
3336 uint64_t timestamp
= ub
->ub_timestamp
;
3337 uint64_t mmp_config
= ub
->ub_mmp_config
;
3338 uint16_t mmp_seq
= MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0;
3339 uint64_t import_delay
;
3340 hrtime_t import_expire
;
3341 nvlist_t
*mmp_label
= NULL
;
3342 vdev_t
*rvd
= spa
->spa_root_vdev
;
3347 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
3348 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3352 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3353 * during the earlier tryimport. If the txg recorded there is 0 then
3354 * the pool is known to be active on another host.
3356 * Otherwise, the pool might be in use on another host. Check for
3357 * changes in the uberblocks on disk if necessary.
3359 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3360 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
3361 ZPOOL_CONFIG_LOAD_INFO
);
3363 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
3364 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
3365 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3366 error
= SET_ERROR(EREMOTEIO
);
3371 import_delay
= spa_activity_check_duration(spa
, ub
);
3373 /* Add a small random factor in case of simultaneous imports (0-25%) */
3374 import_delay
+= import_delay
* random_in_range(250) / 1000;
3376 import_expire
= gethrtime() + import_delay
;
3378 while (gethrtime() < import_expire
) {
3379 (void) spa_import_progress_set_mmp_check(spa_guid(spa
),
3380 NSEC2SEC(import_expire
- gethrtime()));
3382 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3384 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
||
3385 mmp_seq
!= (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0)) {
3386 zfs_dbgmsg("multihost activity detected "
3387 "txg %llu ub_txg %llu "
3388 "timestamp %llu ub_timestamp %llu "
3389 "mmp_config %#llx ub_mmp_config %#llx",
3390 (u_longlong_t
)txg
, (u_longlong_t
)ub
->ub_txg
,
3391 (u_longlong_t
)timestamp
,
3392 (u_longlong_t
)ub
->ub_timestamp
,
3393 (u_longlong_t
)mmp_config
,
3394 (u_longlong_t
)ub
->ub_mmp_config
);
3396 error
= SET_ERROR(EREMOTEIO
);
3401 nvlist_free(mmp_label
);
3405 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
3407 error
= SET_ERROR(EINTR
);
3415 mutex_destroy(&mtx
);
3419 * If the pool is determined to be active store the status in the
3420 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3421 * available from configuration read from disk store them as well.
3422 * This allows 'zpool import' to generate a more useful message.
3424 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3425 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3426 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3428 if (error
== EREMOTEIO
) {
3429 const char *hostname
= "<unknown>";
3430 uint64_t hostid
= 0;
3433 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
3434 hostname
= fnvlist_lookup_string(mmp_label
,
3435 ZPOOL_CONFIG_HOSTNAME
);
3436 fnvlist_add_string(spa
->spa_load_info
,
3437 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
3440 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
3441 hostid
= fnvlist_lookup_uint64(mmp_label
,
3442 ZPOOL_CONFIG_HOSTID
);
3443 fnvlist_add_uint64(spa
->spa_load_info
,
3444 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
3448 fnvlist_add_uint64(spa
->spa_load_info
,
3449 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
3450 fnvlist_add_uint64(spa
->spa_load_info
,
3451 ZPOOL_CONFIG_MMP_TXG
, 0);
3453 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
3457 nvlist_free(mmp_label
);
3463 spa_verify_host(spa_t
*spa
, nvlist_t
*mos_config
)
3466 const char *hostname
;
3467 uint64_t myhostid
= 0;
3469 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
3470 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
3471 hostname
= fnvlist_lookup_string(mos_config
,
3472 ZPOOL_CONFIG_HOSTNAME
);
3474 myhostid
= zone_get_hostid(NULL
);
3476 if (hostid
!= 0 && myhostid
!= 0 && hostid
!= myhostid
) {
3477 cmn_err(CE_WARN
, "pool '%s' could not be "
3478 "loaded as it was last accessed by "
3479 "another system (host: %s hostid: 0x%llx). "
3480 "See: https://openzfs.github.io/openzfs-docs/msg/"
3482 spa_name(spa
), hostname
, (u_longlong_t
)hostid
);
3483 spa_load_failed(spa
, "hostid verification failed: pool "
3484 "last accessed by host: %s (hostid: 0x%llx)",
3485 hostname
, (u_longlong_t
)hostid
);
3486 return (SET_ERROR(EBADF
));
3494 spa_ld_parse_config(spa_t
*spa
, spa_import_type_t type
)
3497 nvlist_t
*nvtree
, *nvl
, *config
= spa
->spa_config
;
3501 const char *comment
;
3502 const char *compatibility
;
3505 * Versioning wasn't explicitly added to the label until later, so if
3506 * it's not present treat it as the initial version.
3508 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
3509 &spa
->spa_ubsync
.ub_version
) != 0)
3510 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
3512 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
3513 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3514 ZPOOL_CONFIG_POOL_GUID
);
3515 return (SET_ERROR(EINVAL
));
3519 * If we are doing an import, ensure that the pool is not already
3520 * imported by checking if its pool guid already exists in the
3523 * The only case that we allow an already imported pool to be
3524 * imported again, is when the pool is checkpointed and we want to
3525 * look at its checkpointed state from userland tools like zdb.
3528 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3529 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3530 spa_guid_exists(pool_guid
, 0)) {
3532 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3533 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3534 spa_guid_exists(pool_guid
, 0) &&
3535 !spa_importing_readonly_checkpoint(spa
)) {
3537 spa_load_failed(spa
, "a pool with guid %llu is already open",
3538 (u_longlong_t
)pool_guid
);
3539 return (SET_ERROR(EEXIST
));
3542 spa
->spa_config_guid
= pool_guid
;
3544 nvlist_free(spa
->spa_load_info
);
3545 spa
->spa_load_info
= fnvlist_alloc();
3547 ASSERT(spa
->spa_comment
== NULL
);
3548 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
3549 spa
->spa_comment
= spa_strdup(comment
);
3551 ASSERT(spa
->spa_compatibility
== NULL
);
3552 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMPATIBILITY
,
3553 &compatibility
) == 0)
3554 spa
->spa_compatibility
= spa_strdup(compatibility
);
3556 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
3557 &spa
->spa_config_txg
);
3559 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) == 0)
3560 spa
->spa_config_splitting
= fnvlist_dup(nvl
);
3562 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
)) {
3563 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3564 ZPOOL_CONFIG_VDEV_TREE
);
3565 return (SET_ERROR(EINVAL
));
3569 * Create "The Godfather" zio to hold all async IOs
3571 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3573 for (int i
= 0; i
< max_ncpus
; i
++) {
3574 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3575 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3576 ZIO_FLAG_GODFATHER
);
3580 * Parse the configuration into a vdev tree. We explicitly set the
3581 * value that will be returned by spa_version() since parsing the
3582 * configuration requires knowing the version number.
3584 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3585 parse
= (type
== SPA_IMPORT_EXISTING
?
3586 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
3587 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
3588 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3591 spa_load_failed(spa
, "unable to parse config [error=%d]",
3596 ASSERT(spa
->spa_root_vdev
== rvd
);
3597 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
3598 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
3600 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3601 ASSERT(spa_guid(spa
) == pool_guid
);
3608 * Recursively open all vdevs in the vdev tree. This function is called twice:
3609 * first with the untrusted config, then with the trusted config.
3612 spa_ld_open_vdevs(spa_t
*spa
)
3617 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3618 * missing/unopenable for the root vdev to be still considered openable.
3620 if (spa
->spa_trust_config
) {
3621 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds
;
3622 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_CACHEFILE
) {
3623 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_cachefile
;
3624 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_SCAN
) {
3625 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_scan
;
3627 spa
->spa_missing_tvds_allowed
= 0;
3630 spa
->spa_missing_tvds_allowed
=
3631 MAX(zfs_max_missing_tvds
, spa
->spa_missing_tvds_allowed
);
3633 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3634 error
= vdev_open(spa
->spa_root_vdev
);
3635 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3637 if (spa
->spa_missing_tvds
!= 0) {
3638 spa_load_note(spa
, "vdev tree has %lld missing top-level "
3639 "vdevs.", (u_longlong_t
)spa
->spa_missing_tvds
);
3640 if (spa
->spa_trust_config
&& (spa
->spa_mode
& SPA_MODE_WRITE
)) {
3642 * Although theoretically we could allow users to open
3643 * incomplete pools in RW mode, we'd need to add a lot
3644 * of extra logic (e.g. adjust pool space to account
3645 * for missing vdevs).
3646 * This limitation also prevents users from accidentally
3647 * opening the pool in RW mode during data recovery and
3648 * damaging it further.
3650 spa_load_note(spa
, "pools with missing top-level "
3651 "vdevs can only be opened in read-only mode.");
3652 error
= SET_ERROR(ENXIO
);
3654 spa_load_note(spa
, "current settings allow for maximum "
3655 "%lld missing top-level vdevs at this stage.",
3656 (u_longlong_t
)spa
->spa_missing_tvds_allowed
);
3660 spa_load_failed(spa
, "unable to open vdev tree [error=%d]",
3663 if (spa
->spa_missing_tvds
!= 0 || error
!= 0)
3664 vdev_dbgmsg_print_tree(spa
->spa_root_vdev
, 2);
3670 * We need to validate the vdev labels against the configuration that
3671 * we have in hand. This function is called twice: first with an untrusted
3672 * config, then with a trusted config. The validation is more strict when the
3673 * config is trusted.
3676 spa_ld_validate_vdevs(spa_t
*spa
)
3679 vdev_t
*rvd
= spa
->spa_root_vdev
;
3681 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3682 error
= vdev_validate(rvd
);
3683 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3686 spa_load_failed(spa
, "vdev_validate failed [error=%d]", error
);
3690 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
3691 spa_load_failed(spa
, "cannot open vdev tree after invalidating "
3693 vdev_dbgmsg_print_tree(rvd
, 2);
3694 return (SET_ERROR(ENXIO
));
3701 spa_ld_select_uberblock_done(spa_t
*spa
, uberblock_t
*ub
)
3703 spa
->spa_state
= POOL_STATE_ACTIVE
;
3704 spa
->spa_ubsync
= spa
->spa_uberblock
;
3705 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
3706 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
3707 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
3708 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
3709 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
3710 spa
->spa_prev_software_version
= ub
->ub_software_version
;
3714 spa_ld_select_uberblock(spa_t
*spa
, spa_import_type_t type
)
3716 vdev_t
*rvd
= spa
->spa_root_vdev
;
3718 uberblock_t
*ub
= &spa
->spa_uberblock
;
3719 boolean_t activity_check
= B_FALSE
;
3722 * If we are opening the checkpointed state of the pool by
3723 * rewinding to it, at this point we will have written the
3724 * checkpointed uberblock to the vdev labels, so searching
3725 * the labels will find the right uberblock. However, if
3726 * we are opening the checkpointed state read-only, we have
3727 * not modified the labels. Therefore, we must ignore the
3728 * labels and continue using the spa_uberblock that was set
3729 * by spa_ld_checkpoint_rewind.
3731 * Note that it would be fine to ignore the labels when
3732 * rewinding (opening writeable) as well. However, if we
3733 * crash just after writing the labels, we will end up
3734 * searching the labels. Doing so in the common case means
3735 * that this code path gets exercised normally, rather than
3736 * just in the edge case.
3738 if (ub
->ub_checkpoint_txg
!= 0 &&
3739 spa_importing_readonly_checkpoint(spa
)) {
3740 spa_ld_select_uberblock_done(spa
, ub
);
3745 * Find the best uberblock.
3747 vdev_uberblock_load(rvd
, ub
, &label
);
3750 * If we weren't able to find a single valid uberblock, return failure.
3752 if (ub
->ub_txg
== 0) {
3754 spa_load_failed(spa
, "no valid uberblock found");
3755 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
3758 if (spa
->spa_load_max_txg
!= UINT64_MAX
) {
3759 (void) spa_import_progress_set_max_txg(spa_guid(spa
),
3760 (u_longlong_t
)spa
->spa_load_max_txg
);
3762 spa_load_note(spa
, "using uberblock with txg=%llu",
3763 (u_longlong_t
)ub
->ub_txg
);
3764 if (ub
->ub_raidz_reflow_info
!= 0) {
3765 spa_load_note(spa
, "uberblock raidz_reflow_info: "
3766 "state=%u offset=%llu",
3767 (int)RRSS_GET_STATE(ub
),
3768 (u_longlong_t
)RRSS_GET_OFFSET(ub
));
3773 * For pools which have the multihost property on determine if the
3774 * pool is truly inactive and can be safely imported. Prevent
3775 * hosts which don't have a hostid set from importing the pool.
3777 activity_check
= spa_activity_check_required(spa
, ub
, label
,
3779 if (activity_check
) {
3780 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
3781 spa_get_hostid(spa
) == 0) {
3783 fnvlist_add_uint64(spa
->spa_load_info
,
3784 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3785 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3788 int error
= spa_activity_check(spa
, ub
, spa
->spa_config
);
3794 fnvlist_add_uint64(spa
->spa_load_info
,
3795 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
3796 fnvlist_add_uint64(spa
->spa_load_info
,
3797 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
3798 fnvlist_add_uint16(spa
->spa_load_info
,
3799 ZPOOL_CONFIG_MMP_SEQ
,
3800 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0));
3804 * If the pool has an unsupported version we can't open it.
3806 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
3808 spa_load_failed(spa
, "version %llu is not supported",
3809 (u_longlong_t
)ub
->ub_version
);
3810 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
3813 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3817 * If we weren't able to find what's necessary for reading the
3818 * MOS in the label, return failure.
3820 if (label
== NULL
) {
3821 spa_load_failed(spa
, "label config unavailable");
3822 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3826 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
3829 spa_load_failed(spa
, "invalid label: '%s' missing",
3830 ZPOOL_CONFIG_FEATURES_FOR_READ
);
3831 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3836 * Update our in-core representation with the definitive values
3839 nvlist_free(spa
->spa_label_features
);
3840 spa
->spa_label_features
= fnvlist_dup(features
);
3846 * Look through entries in the label nvlist's features_for_read. If
3847 * there is a feature listed there which we don't understand then we
3848 * cannot open a pool.
3850 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
3851 nvlist_t
*unsup_feat
;
3853 unsup_feat
= fnvlist_alloc();
3855 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
3857 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
3858 if (!zfeature_is_supported(nvpair_name(nvp
))) {
3859 fnvlist_add_string(unsup_feat
,
3860 nvpair_name(nvp
), "");
3864 if (!nvlist_empty(unsup_feat
)) {
3865 fnvlist_add_nvlist(spa
->spa_load_info
,
3866 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
3867 nvlist_free(unsup_feat
);
3868 spa_load_failed(spa
, "some features are unsupported");
3869 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
3873 nvlist_free(unsup_feat
);
3876 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
3877 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3878 spa_try_repair(spa
, spa
->spa_config
);
3879 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3880 nvlist_free(spa
->spa_config_splitting
);
3881 spa
->spa_config_splitting
= NULL
;
3885 * Initialize internal SPA structures.
3887 spa_ld_select_uberblock_done(spa
, ub
);
3893 spa_ld_open_rootbp(spa_t
*spa
)
3896 vdev_t
*rvd
= spa
->spa_root_vdev
;
3898 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
3900 spa_load_failed(spa
, "unable to open rootbp in dsl_pool_init "
3901 "[error=%d]", error
);
3902 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3904 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
3910 spa_ld_trusted_config(spa_t
*spa
, spa_import_type_t type
,
3911 boolean_t reloading
)
3913 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
3914 nvlist_t
*nv
, *mos_config
, *policy
;
3915 int error
= 0, copy_error
;
3916 uint64_t healthy_tvds
, healthy_tvds_mos
;
3917 uint64_t mos_config_txg
;
3919 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
, B_TRUE
)
3921 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3924 * If we're assembling a pool from a split, the config provided is
3925 * already trusted so there is nothing to do.
3927 if (type
== SPA_IMPORT_ASSEMBLE
)
3930 healthy_tvds
= spa_healthy_core_tvds(spa
);
3932 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
)
3934 spa_load_failed(spa
, "unable to retrieve MOS config");
3935 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3939 * If we are doing an open, pool owner wasn't verified yet, thus do
3940 * the verification here.
3942 if (spa
->spa_load_state
== SPA_LOAD_OPEN
) {
3943 error
= spa_verify_host(spa
, mos_config
);
3945 nvlist_free(mos_config
);
3950 nv
= fnvlist_lookup_nvlist(mos_config
, ZPOOL_CONFIG_VDEV_TREE
);
3952 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3955 * Build a new vdev tree from the trusted config
3957 error
= spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
);
3959 nvlist_free(mos_config
);
3960 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3961 spa_load_failed(spa
, "spa_config_parse failed [error=%d]",
3963 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3967 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3968 * obtained by scanning /dev/dsk, then it will have the right vdev
3969 * paths. We update the trusted MOS config with this information.
3970 * We first try to copy the paths with vdev_copy_path_strict, which
3971 * succeeds only when both configs have exactly the same vdev tree.
3972 * If that fails, we fall back to a more flexible method that has a
3973 * best effort policy.
3975 copy_error
= vdev_copy_path_strict(rvd
, mrvd
);
3976 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3977 spa_load_note(spa
, "provided vdev tree:");
3978 vdev_dbgmsg_print_tree(rvd
, 2);
3979 spa_load_note(spa
, "MOS vdev tree:");
3980 vdev_dbgmsg_print_tree(mrvd
, 2);
3982 if (copy_error
!= 0) {
3983 spa_load_note(spa
, "vdev_copy_path_strict failed, falling "
3984 "back to vdev_copy_path_relaxed");
3985 vdev_copy_path_relaxed(rvd
, mrvd
);
3990 spa
->spa_root_vdev
= mrvd
;
3992 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3995 * If 'zpool import' used a cached config, then the on-disk hostid and
3996 * hostname may be different to the cached config in ways that should
3997 * prevent import. Userspace can't discover this without a scan, but
3998 * we know, so we add these values to LOAD_INFO so the caller can know
4001 * Note that we have to do this before the config is regenerated,
4002 * because the new config will have the hostid and hostname for this
4003 * host, in readiness for import.
4005 if (nvlist_exists(mos_config
, ZPOOL_CONFIG_HOSTID
))
4006 fnvlist_add_uint64(spa
->spa_load_info
, ZPOOL_CONFIG_HOSTID
,
4007 fnvlist_lookup_uint64(mos_config
, ZPOOL_CONFIG_HOSTID
));
4008 if (nvlist_exists(mos_config
, ZPOOL_CONFIG_HOSTNAME
))
4009 fnvlist_add_string(spa
->spa_load_info
, ZPOOL_CONFIG_HOSTNAME
,
4010 fnvlist_lookup_string(mos_config
, ZPOOL_CONFIG_HOSTNAME
));
4013 * We will use spa_config if we decide to reload the spa or if spa_load
4014 * fails and we rewind. We must thus regenerate the config using the
4015 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
4016 * pass settings on how to load the pool and is not stored in the MOS.
4017 * We copy it over to our new, trusted config.
4019 mos_config_txg
= fnvlist_lookup_uint64(mos_config
,
4020 ZPOOL_CONFIG_POOL_TXG
);
4021 nvlist_free(mos_config
);
4022 mos_config
= spa_config_generate(spa
, NULL
, mos_config_txg
, B_FALSE
);
4023 if (nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_LOAD_POLICY
,
4025 fnvlist_add_nvlist(mos_config
, ZPOOL_LOAD_POLICY
, policy
);
4026 spa_config_set(spa
, mos_config
);
4027 spa
->spa_config_source
= SPA_CONFIG_SRC_MOS
;
4030 * Now that we got the config from the MOS, we should be more strict
4031 * in checking blkptrs and can make assumptions about the consistency
4032 * of the vdev tree. spa_trust_config must be set to true before opening
4033 * vdevs in order for them to be writeable.
4035 spa
->spa_trust_config
= B_TRUE
;
4038 * Open and validate the new vdev tree
4040 error
= spa_ld_open_vdevs(spa
);
4044 error
= spa_ld_validate_vdevs(spa
);
4048 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
4049 spa_load_note(spa
, "final vdev tree:");
4050 vdev_dbgmsg_print_tree(rvd
, 2);
4053 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
&&
4054 !spa
->spa_extreme_rewind
&& zfs_max_missing_tvds
== 0) {
4056 * Sanity check to make sure that we are indeed loading the
4057 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
4058 * in the config provided and they happened to be the only ones
4059 * to have the latest uberblock, we could involuntarily perform
4060 * an extreme rewind.
4062 healthy_tvds_mos
= spa_healthy_core_tvds(spa
);
4063 if (healthy_tvds_mos
- healthy_tvds
>=
4064 SPA_SYNC_MIN_VDEVS
) {
4065 spa_load_note(spa
, "config provided misses too many "
4066 "top-level vdevs compared to MOS (%lld vs %lld). ",
4067 (u_longlong_t
)healthy_tvds
,
4068 (u_longlong_t
)healthy_tvds_mos
);
4069 spa_load_note(spa
, "vdev tree:");
4070 vdev_dbgmsg_print_tree(rvd
, 2);
4072 spa_load_failed(spa
, "config was already "
4073 "provided from MOS. Aborting.");
4074 return (spa_vdev_err(rvd
,
4075 VDEV_AUX_CORRUPT_DATA
, EIO
));
4077 spa_load_note(spa
, "spa must be reloaded using MOS "
4079 return (SET_ERROR(EAGAIN
));
4083 error
= spa_check_for_missing_logs(spa
);
4085 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
4087 if (rvd
->vdev_guid_sum
!= spa
->spa_uberblock
.ub_guid_sum
) {
4088 spa_load_failed(spa
, "uberblock guid sum doesn't match MOS "
4089 "guid sum (%llu != %llu)",
4090 (u_longlong_t
)spa
->spa_uberblock
.ub_guid_sum
,
4091 (u_longlong_t
)rvd
->vdev_guid_sum
);
4092 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
4100 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
4103 vdev_t
*rvd
= spa
->spa_root_vdev
;
4106 * Everything that we read before spa_remove_init() must be stored
4107 * on concreted vdevs. Therefore we do this as early as possible.
4109 error
= spa_remove_init(spa
);
4111 spa_load_failed(spa
, "spa_remove_init failed [error=%d]",
4113 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4117 * Retrieve information needed to condense indirect vdev mappings.
4119 error
= spa_condense_init(spa
);
4121 spa_load_failed(spa
, "spa_condense_init failed [error=%d]",
4123 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4130 spa_ld_check_features(spa_t
*spa
, boolean_t
*missing_feat_writep
)
4133 vdev_t
*rvd
= spa
->spa_root_vdev
;
4135 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
4136 boolean_t missing_feat_read
= B_FALSE
;
4137 nvlist_t
*unsup_feat
, *enabled_feat
;
4139 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
4140 &spa
->spa_feat_for_read_obj
, B_TRUE
) != 0) {
4141 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4144 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
4145 &spa
->spa_feat_for_write_obj
, B_TRUE
) != 0) {
4146 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4149 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
4150 &spa
->spa_feat_desc_obj
, B_TRUE
) != 0) {
4151 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4154 enabled_feat
= fnvlist_alloc();
4155 unsup_feat
= fnvlist_alloc();
4157 if (!spa_features_check(spa
, B_FALSE
,
4158 unsup_feat
, enabled_feat
))
4159 missing_feat_read
= B_TRUE
;
4161 if (spa_writeable(spa
) ||
4162 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) {
4163 if (!spa_features_check(spa
, B_TRUE
,
4164 unsup_feat
, enabled_feat
)) {
4165 *missing_feat_writep
= B_TRUE
;
4169 fnvlist_add_nvlist(spa
->spa_load_info
,
4170 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
4172 if (!nvlist_empty(unsup_feat
)) {
4173 fnvlist_add_nvlist(spa
->spa_load_info
,
4174 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
4177 fnvlist_free(enabled_feat
);
4178 fnvlist_free(unsup_feat
);
4180 if (!missing_feat_read
) {
4181 fnvlist_add_boolean(spa
->spa_load_info
,
4182 ZPOOL_CONFIG_CAN_RDONLY
);
4186 * If the state is SPA_LOAD_TRYIMPORT, our objective is
4187 * twofold: to determine whether the pool is available for
4188 * import in read-write mode and (if it is not) whether the
4189 * pool is available for import in read-only mode. If the pool
4190 * is available for import in read-write mode, it is displayed
4191 * as available in userland; if it is not available for import
4192 * in read-only mode, it is displayed as unavailable in
4193 * userland. If the pool is available for import in read-only
4194 * mode but not read-write mode, it is displayed as unavailable
4195 * in userland with a special note that the pool is actually
4196 * available for open in read-only mode.
4198 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
4199 * missing a feature for write, we must first determine whether
4200 * the pool can be opened read-only before returning to
4201 * userland in order to know whether to display the
4202 * abovementioned note.
4204 if (missing_feat_read
|| (*missing_feat_writep
&&
4205 spa_writeable(spa
))) {
4206 spa_load_failed(spa
, "pool uses unsupported features");
4207 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
4212 * Load refcounts for ZFS features from disk into an in-memory
4213 * cache during SPA initialization.
4215 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
4218 error
= feature_get_refcount_from_disk(spa
,
4219 &spa_feature_table
[i
], &refcount
);
4221 spa
->spa_feat_refcount_cache
[i
] = refcount
;
4222 } else if (error
== ENOTSUP
) {
4223 spa
->spa_feat_refcount_cache
[i
] =
4224 SPA_FEATURE_DISABLED
;
4226 spa_load_failed(spa
, "error getting refcount "
4227 "for feature %s [error=%d]",
4228 spa_feature_table
[i
].fi_guid
, error
);
4229 return (spa_vdev_err(rvd
,
4230 VDEV_AUX_CORRUPT_DATA
, EIO
));
4235 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
4236 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
4237 &spa
->spa_feat_enabled_txg_obj
, B_TRUE
) != 0)
4238 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4242 * Encryption was added before bookmark_v2, even though bookmark_v2
4243 * is now a dependency. If this pool has encryption enabled without
4244 * bookmark_v2, trigger an errata message.
4246 if (spa_feature_is_enabled(spa
, SPA_FEATURE_ENCRYPTION
) &&
4247 !spa_feature_is_enabled(spa
, SPA_FEATURE_BOOKMARK_V2
)) {
4248 spa
->spa_errata
= ZPOOL_ERRATA_ZOL_8308_ENCRYPTION
;
4255 spa_ld_load_special_directories(spa_t
*spa
)
4258 vdev_t
*rvd
= spa
->spa_root_vdev
;
4260 spa
->spa_is_initializing
= B_TRUE
;
4261 error
= dsl_pool_open(spa
->spa_dsl_pool
);
4262 spa
->spa_is_initializing
= B_FALSE
;
4264 spa_load_failed(spa
, "dsl_pool_open failed [error=%d]", error
);
4265 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4272 spa_ld_get_props(spa_t
*spa
)
4276 vdev_t
*rvd
= spa
->spa_root_vdev
;
4278 /* Grab the checksum salt from the MOS. */
4279 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4280 DMU_POOL_CHECKSUM_SALT
, 1,
4281 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
4282 spa
->spa_cksum_salt
.zcs_bytes
);
4283 if (error
== ENOENT
) {
4284 /* Generate a new salt for subsequent use */
4285 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4286 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4287 } else if (error
!= 0) {
4288 spa_load_failed(spa
, "unable to retrieve checksum salt from "
4289 "MOS [error=%d]", error
);
4290 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4293 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
, B_TRUE
) != 0)
4294 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4295 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
4297 spa_load_failed(spa
, "error opening deferred-frees bpobj "
4298 "[error=%d]", error
);
4299 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4303 * Load the bit that tells us to use the new accounting function
4304 * (raid-z deflation). If we have an older pool, this will not
4307 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
, B_FALSE
);
4308 if (error
!= 0 && error
!= ENOENT
)
4309 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4311 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
4312 &spa
->spa_creation_version
, B_FALSE
);
4313 if (error
!= 0 && error
!= ENOENT
)
4314 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4317 * Load the persistent error log. If we have an older pool, this will
4320 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
,
4322 if (error
!= 0 && error
!= ENOENT
)
4323 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4325 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
4326 &spa
->spa_errlog_scrub
, B_FALSE
);
4327 if (error
!= 0 && error
!= ENOENT
)
4328 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4331 * Load the livelist deletion field. If a livelist is queued for
4332 * deletion, indicate that in the spa
4334 error
= spa_dir_prop(spa
, DMU_POOL_DELETED_CLONES
,
4335 &spa
->spa_livelists_to_delete
, B_FALSE
);
4336 if (error
!= 0 && error
!= ENOENT
)
4337 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4340 * Load the history object. If we have an older pool, this
4341 * will not be present.
4343 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
, B_FALSE
);
4344 if (error
!= 0 && error
!= ENOENT
)
4345 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4348 * Load the per-vdev ZAP map. If we have an older pool, this will not
4349 * be present; in this case, defer its creation to a later time to
4350 * avoid dirtying the MOS this early / out of sync context. See
4351 * spa_sync_config_object.
4354 /* The sentinel is only available in the MOS config. */
4355 nvlist_t
*mos_config
;
4356 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0) {
4357 spa_load_failed(spa
, "unable to retrieve MOS config");
4358 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4361 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
4362 &spa
->spa_all_vdev_zaps
, B_FALSE
);
4364 if (error
== ENOENT
) {
4365 VERIFY(!nvlist_exists(mos_config
,
4366 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
4367 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
4368 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4369 } else if (error
!= 0) {
4370 nvlist_free(mos_config
);
4371 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4372 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
4374 * An older version of ZFS overwrote the sentinel value, so
4375 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4376 * destruction to later; see spa_sync_config_object.
4378 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
4380 * We're assuming that no vdevs have had their ZAPs created
4381 * before this. Better be sure of it.
4383 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4385 nvlist_free(mos_config
);
4387 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4389 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
,
4391 if (error
&& error
!= ENOENT
)
4392 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4395 uint64_t autoreplace
= 0;
4397 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
4398 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
4399 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
4400 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
4401 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
4402 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
4403 spa_prop_find(spa
, ZPOOL_PROP_AUTOTRIM
, &spa
->spa_autotrim
);
4404 spa
->spa_autoreplace
= (autoreplace
!= 0);
4408 * If we are importing a pool with missing top-level vdevs,
4409 * we enforce that the pool doesn't panic or get suspended on
4410 * error since the likelihood of missing data is extremely high.
4412 if (spa
->spa_missing_tvds
> 0 &&
4413 spa
->spa_failmode
!= ZIO_FAILURE_MODE_CONTINUE
&&
4414 spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4415 spa_load_note(spa
, "forcing failmode to 'continue' "
4416 "as some top level vdevs are missing");
4417 spa
->spa_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
4424 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
4427 vdev_t
*rvd
= spa
->spa_root_vdev
;
4430 * If we're assembling the pool from the split-off vdevs of
4431 * an existing pool, we don't want to attach the spares & cache
4436 * Load any hot spares for this pool.
4438 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
,
4440 if (error
!= 0 && error
!= ENOENT
)
4441 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4442 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4443 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
4444 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
4445 &spa
->spa_spares
.sav_config
) != 0) {
4446 spa_load_failed(spa
, "error loading spares nvlist");
4447 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4450 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4451 spa_load_spares(spa
);
4452 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4453 } else if (error
== 0) {
4454 spa
->spa_spares
.sav_sync
= B_TRUE
;
4458 * Load any level 2 ARC devices for this pool.
4460 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
4461 &spa
->spa_l2cache
.sav_object
, B_FALSE
);
4462 if (error
!= 0 && error
!= ENOENT
)
4463 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4464 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4465 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
4466 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
4467 &spa
->spa_l2cache
.sav_config
) != 0) {
4468 spa_load_failed(spa
, "error loading l2cache nvlist");
4469 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4472 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4473 spa_load_l2cache(spa
);
4474 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4475 } else if (error
== 0) {
4476 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4483 spa_ld_load_vdev_metadata(spa_t
*spa
)
4486 vdev_t
*rvd
= spa
->spa_root_vdev
;
4489 * If the 'multihost' property is set, then never allow a pool to
4490 * be imported when the system hostid is zero. The exception to
4491 * this rule is zdb which is always allowed to access pools.
4493 if (spa_multihost(spa
) && spa_get_hostid(spa
) == 0 &&
4494 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
4495 fnvlist_add_uint64(spa
->spa_load_info
,
4496 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
4497 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
4501 * If the 'autoreplace' property is set, then post a resource notifying
4502 * the ZFS DE that it should not issue any faults for unopenable
4503 * devices. We also iterate over the vdevs, and post a sysevent for any
4504 * unopenable vdevs so that the normal autoreplace handler can take
4507 if (spa
->spa_autoreplace
&& spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4508 spa_check_removed(spa
->spa_root_vdev
);
4510 * For the import case, this is done in spa_import(), because
4511 * at this point we're using the spare definitions from
4512 * the MOS config, not necessarily from the userland config.
4514 if (spa
->spa_load_state
!= SPA_LOAD_IMPORT
) {
4515 spa_aux_check_removed(&spa
->spa_spares
);
4516 spa_aux_check_removed(&spa
->spa_l2cache
);
4521 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4523 error
= vdev_load(rvd
);
4525 spa_load_failed(spa
, "vdev_load failed [error=%d]", error
);
4526 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4529 error
= spa_ld_log_spacemaps(spa
);
4531 spa_load_failed(spa
, "spa_ld_log_spacemaps failed [error=%d]",
4533 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4537 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4539 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4540 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
, B_FALSE
);
4541 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4547 spa_ld_load_dedup_tables(spa_t
*spa
)
4550 vdev_t
*rvd
= spa
->spa_root_vdev
;
4552 error
= ddt_load(spa
);
4554 spa_load_failed(spa
, "ddt_load failed [error=%d]", error
);
4555 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4562 spa_ld_load_brt(spa_t
*spa
)
4565 vdev_t
*rvd
= spa
->spa_root_vdev
;
4567 error
= brt_load(spa
);
4569 spa_load_failed(spa
, "brt_load failed [error=%d]", error
);
4570 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4577 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, const char **ereport
)
4579 vdev_t
*rvd
= spa
->spa_root_vdev
;
4581 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
)) {
4582 boolean_t missing
= spa_check_logs(spa
);
4584 if (spa
->spa_missing_tvds
!= 0) {
4585 spa_load_note(spa
, "spa_check_logs failed "
4586 "so dropping the logs");
4588 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
4589 spa_load_failed(spa
, "spa_check_logs failed");
4590 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
,
4600 spa_ld_verify_pool_data(spa_t
*spa
)
4603 vdev_t
*rvd
= spa
->spa_root_vdev
;
4606 * We've successfully opened the pool, verify that we're ready
4607 * to start pushing transactions.
4609 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4610 error
= spa_load_verify(spa
);
4612 spa_load_failed(spa
, "spa_load_verify failed "
4613 "[error=%d]", error
);
4614 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
4623 spa_ld_claim_log_blocks(spa_t
*spa
)
4626 dsl_pool_t
*dp
= spa_get_dsl(spa
);
4629 * Claim log blocks that haven't been committed yet.
4630 * This must all happen in a single txg.
4631 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4632 * invoked from zil_claim_log_block()'s i/o done callback.
4633 * Price of rollback is that we abandon the log.
4635 spa
->spa_claiming
= B_TRUE
;
4637 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
4638 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
4639 zil_claim
, tx
, DS_FIND_CHILDREN
);
4642 spa
->spa_claiming
= B_FALSE
;
4644 spa_set_log_state(spa
, SPA_LOG_GOOD
);
4648 spa_ld_check_for_config_update(spa_t
*spa
, uint64_t config_cache_txg
,
4649 boolean_t update_config_cache
)
4651 vdev_t
*rvd
= spa
->spa_root_vdev
;
4652 int need_update
= B_FALSE
;
4655 * If the config cache is stale, or we have uninitialized
4656 * metaslabs (see spa_vdev_add()), then update the config.
4658 * If this is a verbatim import, trust the current
4659 * in-core spa_config and update the disk labels.
4661 if (update_config_cache
|| config_cache_txg
!= spa
->spa_config_txg
||
4662 spa
->spa_load_state
== SPA_LOAD_IMPORT
||
4663 spa
->spa_load_state
== SPA_LOAD_RECOVER
||
4664 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
4665 need_update
= B_TRUE
;
4667 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
4668 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
4669 need_update
= B_TRUE
;
4672 * Update the config cache asynchronously in case we're the
4673 * root pool, in which case the config cache isn't writable yet.
4676 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4680 spa_ld_prepare_for_reload(spa_t
*spa
)
4682 spa_mode_t mode
= spa
->spa_mode
;
4683 int async_suspended
= spa
->spa_async_suspended
;
4686 spa_deactivate(spa
);
4687 spa_activate(spa
, mode
);
4690 * We save the value of spa_async_suspended as it gets reset to 0 by
4691 * spa_unload(). We want to restore it back to the original value before
4692 * returning as we might be calling spa_async_resume() later.
4694 spa
->spa_async_suspended
= async_suspended
;
4698 spa_ld_read_checkpoint_txg(spa_t
*spa
)
4700 uberblock_t checkpoint
;
4703 ASSERT0(spa
->spa_checkpoint_txg
);
4704 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4706 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4707 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4708 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4710 if (error
== ENOENT
)
4716 ASSERT3U(checkpoint
.ub_txg
, !=, 0);
4717 ASSERT3U(checkpoint
.ub_checkpoint_txg
, !=, 0);
4718 ASSERT3U(checkpoint
.ub_timestamp
, !=, 0);
4719 spa
->spa_checkpoint_txg
= checkpoint
.ub_txg
;
4720 spa
->spa_checkpoint_info
.sci_timestamp
= checkpoint
.ub_timestamp
;
4726 spa_ld_mos_init(spa_t
*spa
, spa_import_type_t type
)
4730 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4731 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4734 * Never trust the config that is provided unless we are assembling
4735 * a pool following a split.
4736 * This means don't trust blkptrs and the vdev tree in general. This
4737 * also effectively puts the spa in read-only mode since
4738 * spa_writeable() checks for spa_trust_config to be true.
4739 * We will later load a trusted config from the MOS.
4741 if (type
!= SPA_IMPORT_ASSEMBLE
)
4742 spa
->spa_trust_config
= B_FALSE
;
4745 * Parse the config provided to create a vdev tree.
4747 error
= spa_ld_parse_config(spa
, type
);
4751 spa_import_progress_add(spa
);
4754 * Now that we have the vdev tree, try to open each vdev. This involves
4755 * opening the underlying physical device, retrieving its geometry and
4756 * probing the vdev with a dummy I/O. The state of each vdev will be set
4757 * based on the success of those operations. After this we'll be ready
4758 * to read from the vdevs.
4760 error
= spa_ld_open_vdevs(spa
);
4765 * Read the label of each vdev and make sure that the GUIDs stored
4766 * there match the GUIDs in the config provided.
4767 * If we're assembling a new pool that's been split off from an
4768 * existing pool, the labels haven't yet been updated so we skip
4769 * validation for now.
4771 if (type
!= SPA_IMPORT_ASSEMBLE
) {
4772 error
= spa_ld_validate_vdevs(spa
);
4778 * Read all vdev labels to find the best uberblock (i.e. latest,
4779 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
4780 * get the list of features required to read blkptrs in the MOS from
4781 * the vdev label with the best uberblock and verify that our version
4782 * of zfs supports them all.
4784 error
= spa_ld_select_uberblock(spa
, type
);
4789 * Pass that uberblock to the dsl_pool layer which will open the root
4790 * blkptr. This blkptr points to the latest version of the MOS and will
4791 * allow us to read its contents.
4793 error
= spa_ld_open_rootbp(spa
);
4801 spa_ld_checkpoint_rewind(spa_t
*spa
)
4803 uberblock_t checkpoint
;
4806 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4807 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4809 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4810 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4811 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4814 spa_load_failed(spa
, "unable to retrieve checkpointed "
4815 "uberblock from the MOS config [error=%d]", error
);
4817 if (error
== ENOENT
)
4818 error
= ZFS_ERR_NO_CHECKPOINT
;
4823 ASSERT3U(checkpoint
.ub_txg
, <, spa
->spa_uberblock
.ub_txg
);
4824 ASSERT3U(checkpoint
.ub_txg
, ==, checkpoint
.ub_checkpoint_txg
);
4827 * We need to update the txg and timestamp of the checkpointed
4828 * uberblock to be higher than the latest one. This ensures that
4829 * the checkpointed uberblock is selected if we were to close and
4830 * reopen the pool right after we've written it in the vdev labels.
4831 * (also see block comment in vdev_uberblock_compare)
4833 checkpoint
.ub_txg
= spa
->spa_uberblock
.ub_txg
+ 1;
4834 checkpoint
.ub_timestamp
= gethrestime_sec();
4837 * Set current uberblock to be the checkpointed uberblock.
4839 spa
->spa_uberblock
= checkpoint
;
4842 * If we are doing a normal rewind, then the pool is open for
4843 * writing and we sync the "updated" checkpointed uberblock to
4844 * disk. Once this is done, we've basically rewound the whole
4845 * pool and there is no way back.
4847 * There are cases when we don't want to attempt and sync the
4848 * checkpointed uberblock to disk because we are opening a
4849 * pool as read-only. Specifically, verifying the checkpointed
4850 * state with zdb, and importing the checkpointed state to get
4851 * a "preview" of its content.
4853 if (spa_writeable(spa
)) {
4854 vdev_t
*rvd
= spa
->spa_root_vdev
;
4856 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4857 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
4859 int children
= rvd
->vdev_children
;
4860 int c0
= random_in_range(children
);
4862 for (int c
= 0; c
< children
; c
++) {
4863 vdev_t
*vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
4865 /* Stop when revisiting the first vdev */
4866 if (c
> 0 && svd
[0] == vd
)
4869 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
4870 !vdev_is_concrete(vd
))
4873 svd
[svdcount
++] = vd
;
4874 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
4877 error
= vdev_config_sync(svd
, svdcount
, spa
->spa_first_txg
);
4879 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
4880 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4883 spa_load_failed(spa
, "failed to write checkpointed "
4884 "uberblock to the vdev labels [error=%d]", error
);
4893 spa_ld_mos_with_trusted_config(spa_t
*spa
, spa_import_type_t type
,
4894 boolean_t
*update_config_cache
)
4899 * Parse the config for pool, open and validate vdevs,
4900 * select an uberblock, and use that uberblock to open
4903 error
= spa_ld_mos_init(spa
, type
);
4908 * Retrieve the trusted config stored in the MOS and use it to create
4909 * a new, exact version of the vdev tree, then reopen all vdevs.
4911 error
= spa_ld_trusted_config(spa
, type
, B_FALSE
);
4912 if (error
== EAGAIN
) {
4913 if (update_config_cache
!= NULL
)
4914 *update_config_cache
= B_TRUE
;
4917 * Redo the loading process with the trusted config if it is
4918 * too different from the untrusted config.
4920 spa_ld_prepare_for_reload(spa
);
4921 spa_load_note(spa
, "RELOADING");
4922 error
= spa_ld_mos_init(spa
, type
);
4926 error
= spa_ld_trusted_config(spa
, type
, B_TRUE
);
4930 } else if (error
!= 0) {
4938 * Load an existing storage pool, using the config provided. This config
4939 * describes which vdevs are part of the pool and is later validated against
4940 * partial configs present in each vdev's label and an entire copy of the
4941 * config stored in the MOS.
4944 spa_load_impl(spa_t
*spa
, spa_import_type_t type
, const char **ereport
)
4947 boolean_t missing_feat_write
= B_FALSE
;
4948 boolean_t checkpoint_rewind
=
4949 (spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
4950 boolean_t update_config_cache
= B_FALSE
;
4952 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4953 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
4955 spa_load_note(spa
, "LOADING");
4957 error
= spa_ld_mos_with_trusted_config(spa
, type
, &update_config_cache
);
4962 * If we are rewinding to the checkpoint then we need to repeat
4963 * everything we've done so far in this function but this time
4964 * selecting the checkpointed uberblock and using that to open
4967 if (checkpoint_rewind
) {
4969 * If we are rewinding to the checkpoint update config cache
4972 update_config_cache
= B_TRUE
;
4975 * Extract the checkpointed uberblock from the current MOS
4976 * and use this as the pool's uberblock from now on. If the
4977 * pool is imported as writeable we also write the checkpoint
4978 * uberblock to the labels, making the rewind permanent.
4980 error
= spa_ld_checkpoint_rewind(spa
);
4985 * Redo the loading process again with the
4986 * checkpointed uberblock.
4988 spa_ld_prepare_for_reload(spa
);
4989 spa_load_note(spa
, "LOADING checkpointed uberblock");
4990 error
= spa_ld_mos_with_trusted_config(spa
, type
, NULL
);
4996 * Retrieve the checkpoint txg if the pool has a checkpoint.
4998 error
= spa_ld_read_checkpoint_txg(spa
);
5003 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
5004 * from the pool and their contents were re-mapped to other vdevs. Note
5005 * that everything that we read before this step must have been
5006 * rewritten on concrete vdevs after the last device removal was
5007 * initiated. Otherwise we could be reading from indirect vdevs before
5008 * we have loaded their mappings.
5010 error
= spa_ld_open_indirect_vdev_metadata(spa
);
5015 * Retrieve the full list of active features from the MOS and check if
5016 * they are all supported.
5018 error
= spa_ld_check_features(spa
, &missing_feat_write
);
5023 * Load several special directories from the MOS needed by the dsl_pool
5026 error
= spa_ld_load_special_directories(spa
);
5031 * Retrieve pool properties from the MOS.
5033 error
= spa_ld_get_props(spa
);
5038 * Retrieve the list of auxiliary devices - cache devices and spares -
5041 error
= spa_ld_open_aux_vdevs(spa
, type
);
5046 * Load the metadata for all vdevs. Also check if unopenable devices
5047 * should be autoreplaced.
5049 error
= spa_ld_load_vdev_metadata(spa
);
5053 error
= spa_ld_load_dedup_tables(spa
);
5057 error
= spa_ld_load_brt(spa
);
5062 * Verify the logs now to make sure we don't have any unexpected errors
5063 * when we claim log blocks later.
5065 error
= spa_ld_verify_logs(spa
, type
, ereport
);
5069 if (missing_feat_write
) {
5070 ASSERT(spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
);
5073 * At this point, we know that we can open the pool in
5074 * read-only mode but not read-write mode. We now have enough
5075 * information and can return to userland.
5077 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
5082 * Traverse the last txgs to make sure the pool was left off in a safe
5083 * state. When performing an extreme rewind, we verify the whole pool,
5084 * which can take a very long time.
5086 error
= spa_ld_verify_pool_data(spa
);
5091 * Calculate the deflated space for the pool. This must be done before
5092 * we write anything to the pool because we'd need to update the space
5093 * accounting using the deflated sizes.
5095 spa_update_dspace(spa
);
5098 * We have now retrieved all the information we needed to open the
5099 * pool. If we are importing the pool in read-write mode, a few
5100 * additional steps must be performed to finish the import.
5102 if (spa_writeable(spa
) && (spa
->spa_load_state
== SPA_LOAD_RECOVER
||
5103 spa
->spa_load_max_txg
== UINT64_MAX
)) {
5104 uint64_t config_cache_txg
= spa
->spa_config_txg
;
5106 ASSERT(spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
);
5109 * Before we do any zio_write's, complete the raidz expansion
5110 * scratch space copying, if necessary.
5112 if (RRSS_GET_STATE(&spa
->spa_uberblock
) == RRSS_SCRATCH_VALID
)
5113 vdev_raidz_reflow_copy_scratch(spa
);
5116 * In case of a checkpoint rewind, log the original txg
5117 * of the checkpointed uberblock.
5119 if (checkpoint_rewind
) {
5120 spa_history_log_internal(spa
, "checkpoint rewind",
5121 NULL
, "rewound state to txg=%llu",
5122 (u_longlong_t
)spa
->spa_uberblock
.ub_checkpoint_txg
);
5126 * Traverse the ZIL and claim all blocks.
5128 spa_ld_claim_log_blocks(spa
);
5131 * Kick-off the syncing thread.
5133 spa
->spa_sync_on
= B_TRUE
;
5134 txg_sync_start(spa
->spa_dsl_pool
);
5135 mmp_thread_start(spa
);
5138 * Wait for all claims to sync. We sync up to the highest
5139 * claimed log block birth time so that claimed log blocks
5140 * don't appear to be from the future. spa_claim_max_txg
5141 * will have been set for us by ZIL traversal operations
5144 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
5147 * Check if we need to request an update of the config. On the
5148 * next sync, we would update the config stored in vdev labels
5149 * and the cachefile (by default /etc/zfs/zpool.cache).
5151 spa_ld_check_for_config_update(spa
, config_cache_txg
,
5152 update_config_cache
);
5155 * Check if a rebuild was in progress and if so resume it.
5156 * Then check all DTLs to see if anything needs resilvering.
5157 * The resilver will be deferred if a rebuild was started.
5159 if (vdev_rebuild_active(spa
->spa_root_vdev
)) {
5160 vdev_rebuild_restart(spa
);
5161 } else if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
5162 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5163 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
5167 * Log the fact that we booted up (so that we can detect if
5168 * we rebooted in the middle of an operation).
5170 spa_history_log_version(spa
, "open", NULL
);
5172 spa_restart_removal(spa
);
5173 spa_spawn_aux_threads(spa
);
5176 * Delete any inconsistent datasets.
5179 * Since we may be issuing deletes for clones here,
5180 * we make sure to do so after we've spawned all the
5181 * auxiliary threads above (from which the livelist
5182 * deletion zthr is part of).
5184 (void) dmu_objset_find(spa_name(spa
),
5185 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
5188 * Clean up any stale temporary dataset userrefs.
5190 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
5192 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5193 vdev_initialize_restart(spa
->spa_root_vdev
);
5194 vdev_trim_restart(spa
->spa_root_vdev
);
5195 vdev_autotrim_restart(spa
);
5196 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5199 spa_import_progress_remove(spa_guid(spa
));
5200 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
5202 spa_load_note(spa
, "LOADED");
5208 spa_load_retry(spa_t
*spa
, spa_load_state_t state
)
5210 spa_mode_t mode
= spa
->spa_mode
;
5213 spa_deactivate(spa
);
5215 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
5217 spa_activate(spa
, mode
);
5218 spa_async_suspend(spa
);
5220 spa_load_note(spa
, "spa_load_retry: rewind, max txg: %llu",
5221 (u_longlong_t
)spa
->spa_load_max_txg
);
5223 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
));
5227 * If spa_load() fails this function will try loading prior txg's. If
5228 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
5229 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
5230 * function will not rewind the pool and will return the same error as
5234 spa_load_best(spa_t
*spa
, spa_load_state_t state
, uint64_t max_request
,
5237 nvlist_t
*loadinfo
= NULL
;
5238 nvlist_t
*config
= NULL
;
5239 int load_error
, rewind_error
;
5240 uint64_t safe_rewind_txg
;
5243 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
5244 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
5245 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5247 spa
->spa_load_max_txg
= max_request
;
5248 if (max_request
!= UINT64_MAX
)
5249 spa
->spa_extreme_rewind
= B_TRUE
;
5252 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
);
5253 if (load_error
== 0)
5255 if (load_error
== ZFS_ERR_NO_CHECKPOINT
) {
5257 * When attempting checkpoint-rewind on a pool with no
5258 * checkpoint, we should not attempt to load uberblocks
5259 * from previous txgs when spa_load fails.
5261 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
5262 spa_import_progress_remove(spa_guid(spa
));
5263 return (load_error
);
5266 if (spa
->spa_root_vdev
!= NULL
)
5267 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5269 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
5270 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
5272 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
5273 nvlist_free(config
);
5274 spa_import_progress_remove(spa_guid(spa
));
5275 return (load_error
);
5278 if (state
== SPA_LOAD_RECOVER
) {
5279 /* Price of rolling back is discarding txgs, including log */
5280 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5283 * If we aren't rolling back save the load info from our first
5284 * import attempt so that we can restore it after attempting
5287 loadinfo
= spa
->spa_load_info
;
5288 spa
->spa_load_info
= fnvlist_alloc();
5291 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
5292 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
5293 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
5294 TXG_INITIAL
: safe_rewind_txg
;
5297 * Continue as long as we're finding errors, we're still within
5298 * the acceptable rewind range, and we're still finding uberblocks
5300 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
5301 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
5302 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
5303 spa
->spa_extreme_rewind
= B_TRUE
;
5304 rewind_error
= spa_load_retry(spa
, state
);
5307 spa
->spa_extreme_rewind
= B_FALSE
;
5308 spa
->spa_load_max_txg
= UINT64_MAX
;
5310 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
5311 spa_config_set(spa
, config
);
5313 nvlist_free(config
);
5315 if (state
== SPA_LOAD_RECOVER
) {
5316 ASSERT3P(loadinfo
, ==, NULL
);
5317 spa_import_progress_remove(spa_guid(spa
));
5318 return (rewind_error
);
5320 /* Store the rewind info as part of the initial load info */
5321 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
5322 spa
->spa_load_info
);
5324 /* Restore the initial load info */
5325 fnvlist_free(spa
->spa_load_info
);
5326 spa
->spa_load_info
= loadinfo
;
5328 spa_import_progress_remove(spa_guid(spa
));
5329 return (load_error
);
5336 * The import case is identical to an open except that the configuration is sent
5337 * down from userland, instead of grabbed from the configuration cache. For the
5338 * case of an open, the pool configuration will exist in the
5339 * POOL_STATE_UNINITIALIZED state.
5341 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5342 * the same time open the pool, without having to keep around the spa_t in some
5346 spa_open_common(const char *pool
, spa_t
**spapp
, const void *tag
,
5347 nvlist_t
*nvpolicy
, nvlist_t
**config
)
5350 spa_load_state_t state
= SPA_LOAD_OPEN
;
5352 int locked
= B_FALSE
;
5353 int firstopen
= B_FALSE
;
5358 * As disgusting as this is, we need to support recursive calls to this
5359 * function because dsl_dir_open() is called during spa_load(), and ends
5360 * up calling spa_open() again. The real fix is to figure out how to
5361 * avoid dsl_dir_open() calling this in the first place.
5363 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
5364 mutex_enter(&spa_namespace_lock
);
5368 if ((spa
= spa_lookup(pool
)) == NULL
) {
5370 mutex_exit(&spa_namespace_lock
);
5371 return (SET_ERROR(ENOENT
));
5374 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
5375 zpool_load_policy_t policy
;
5379 zpool_get_load_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
5381 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
5382 state
= SPA_LOAD_RECOVER
;
5384 spa_activate(spa
, spa_mode_global
);
5386 if (state
!= SPA_LOAD_RECOVER
)
5387 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
5388 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
5390 zfs_dbgmsg("spa_open_common: opening %s", pool
);
5391 error
= spa_load_best(spa
, state
, policy
.zlp_txg
,
5394 if (error
== EBADF
) {
5396 * If vdev_validate() returns failure (indicated by
5397 * EBADF), it indicates that one of the vdevs indicates
5398 * that the pool has been exported or destroyed. If
5399 * this is the case, the config cache is out of sync and
5400 * we should remove the pool from the namespace.
5403 spa_deactivate(spa
);
5404 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
, B_FALSE
);
5407 mutex_exit(&spa_namespace_lock
);
5408 return (SET_ERROR(ENOENT
));
5413 * We can't open the pool, but we still have useful
5414 * information: the state of each vdev after the
5415 * attempted vdev_open(). Return this to the user.
5417 if (config
!= NULL
&& spa
->spa_config
) {
5418 *config
= fnvlist_dup(spa
->spa_config
);
5419 fnvlist_add_nvlist(*config
,
5420 ZPOOL_CONFIG_LOAD_INFO
,
5421 spa
->spa_load_info
);
5424 spa_deactivate(spa
);
5425 spa
->spa_last_open_failed
= error
;
5427 mutex_exit(&spa_namespace_lock
);
5433 spa_open_ref(spa
, tag
);
5436 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5439 * If we've recovered the pool, pass back any information we
5440 * gathered while doing the load.
5442 if (state
== SPA_LOAD_RECOVER
&& config
!= NULL
) {
5443 fnvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
5444 spa
->spa_load_info
);
5448 spa
->spa_last_open_failed
= 0;
5449 spa
->spa_last_ubsync_txg
= 0;
5450 spa
->spa_load_txg
= 0;
5451 mutex_exit(&spa_namespace_lock
);
5455 zvol_create_minors_recursive(spa_name(spa
));
5463 spa_open_rewind(const char *name
, spa_t
**spapp
, const void *tag
,
5464 nvlist_t
*policy
, nvlist_t
**config
)
5466 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
5470 spa_open(const char *name
, spa_t
**spapp
, const void *tag
)
5472 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
5476 * Lookup the given spa_t, incrementing the inject count in the process,
5477 * preventing it from being exported or destroyed.
5480 spa_inject_addref(char *name
)
5484 mutex_enter(&spa_namespace_lock
);
5485 if ((spa
= spa_lookup(name
)) == NULL
) {
5486 mutex_exit(&spa_namespace_lock
);
5489 spa
->spa_inject_ref
++;
5490 mutex_exit(&spa_namespace_lock
);
5496 spa_inject_delref(spa_t
*spa
)
5498 mutex_enter(&spa_namespace_lock
);
5499 spa
->spa_inject_ref
--;
5500 mutex_exit(&spa_namespace_lock
);
5504 * Add spares device information to the nvlist.
5507 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
5517 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5519 if (spa
->spa_spares
.sav_count
== 0)
5522 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5523 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5524 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
));
5526 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5527 (const nvlist_t
* const *)spares
, nspares
);
5528 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5529 &spares
, &nspares
));
5532 * Go through and find any spares which have since been
5533 * repurposed as an active spare. If this is the case, update
5534 * their status appropriately.
5536 for (i
= 0; i
< nspares
; i
++) {
5537 guid
= fnvlist_lookup_uint64(spares
[i
],
5539 VERIFY0(nvlist_lookup_uint64_array(spares
[i
],
5540 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
));
5541 if (spa_spare_exists(guid
, &pool
, NULL
) &&
5543 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
5544 vs
->vs_aux
= VDEV_AUX_SPARED
;
5547 spa
->spa_spares
.sav_vdevs
[i
]->vdev_state
;
5554 * Add l2cache device information to the nvlist, including vdev stats.
5557 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
5560 uint_t i
, j
, nl2cache
;
5567 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5569 if (spa
->spa_l2cache
.sav_count
== 0)
5572 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5573 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5574 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
));
5575 if (nl2cache
!= 0) {
5576 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5577 (const nvlist_t
* const *)l2cache
, nl2cache
);
5578 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5579 &l2cache
, &nl2cache
));
5582 * Update level 2 cache device stats.
5585 for (i
= 0; i
< nl2cache
; i
++) {
5586 guid
= fnvlist_lookup_uint64(l2cache
[i
],
5590 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
5592 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
5593 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
5599 VERIFY0(nvlist_lookup_uint64_array(l2cache
[i
],
5600 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
));
5601 vdev_get_stats(vd
, vs
);
5602 vdev_config_generate_stats(vd
, l2cache
[i
]);
5609 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
5614 if (spa
->spa_feat_for_read_obj
!= 0) {
5615 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5616 spa
->spa_feat_for_read_obj
);
5617 zap_cursor_retrieve(&zc
, &za
) == 0;
5618 zap_cursor_advance(&zc
)) {
5619 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5620 za
.za_num_integers
== 1);
5621 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5622 za
.za_first_integer
));
5624 zap_cursor_fini(&zc
);
5627 if (spa
->spa_feat_for_write_obj
!= 0) {
5628 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5629 spa
->spa_feat_for_write_obj
);
5630 zap_cursor_retrieve(&zc
, &za
) == 0;
5631 zap_cursor_advance(&zc
)) {
5632 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5633 za
.za_num_integers
== 1);
5634 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5635 za
.za_first_integer
));
5637 zap_cursor_fini(&zc
);
5642 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
5646 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5647 zfeature_info_t feature
= spa_feature_table
[i
];
5650 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
5653 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
5658 * Store a list of pool features and their reference counts in the
5661 * The first time this is called on a spa, allocate a new nvlist, fetch
5662 * the pool features and reference counts from disk, then save the list
5663 * in the spa. In subsequent calls on the same spa use the saved nvlist
5664 * and refresh its values from the cached reference counts. This
5665 * ensures we don't block here on I/O on a suspended pool so 'zpool
5666 * clear' can resume the pool.
5669 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
5673 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5675 mutex_enter(&spa
->spa_feat_stats_lock
);
5676 features
= spa
->spa_feat_stats
;
5678 if (features
!= NULL
) {
5679 spa_feature_stats_from_cache(spa
, features
);
5681 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
5682 spa
->spa_feat_stats
= features
;
5683 spa_feature_stats_from_disk(spa
, features
);
5686 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
5689 mutex_exit(&spa
->spa_feat_stats_lock
);
5693 spa_get_stats(const char *name
, nvlist_t
**config
,
5694 char *altroot
, size_t buflen
)
5700 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
5704 * This still leaves a window of inconsistency where the spares
5705 * or l2cache devices could change and the config would be
5706 * self-inconsistent.
5708 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5710 if (*config
!= NULL
) {
5711 uint64_t loadtimes
[2];
5713 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
5714 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
5715 fnvlist_add_uint64_array(*config
,
5716 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2);
5718 fnvlist_add_uint64(*config
,
5719 ZPOOL_CONFIG_ERRCOUNT
,
5720 spa_approx_errlog_size(spa
));
5722 if (spa_suspended(spa
)) {
5723 fnvlist_add_uint64(*config
,
5724 ZPOOL_CONFIG_SUSPENDED
,
5726 fnvlist_add_uint64(*config
,
5727 ZPOOL_CONFIG_SUSPENDED_REASON
,
5728 spa
->spa_suspended
);
5731 spa_add_spares(spa
, *config
);
5732 spa_add_l2cache(spa
, *config
);
5733 spa_add_feature_stats(spa
, *config
);
5738 * We want to get the alternate root even for faulted pools, so we cheat
5739 * and call spa_lookup() directly.
5743 mutex_enter(&spa_namespace_lock
);
5744 spa
= spa_lookup(name
);
5746 spa_altroot(spa
, altroot
, buflen
);
5750 mutex_exit(&spa_namespace_lock
);
5752 spa_altroot(spa
, altroot
, buflen
);
5757 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5758 spa_close(spa
, FTAG
);
5765 * Validate that the auxiliary device array is well formed. We must have an
5766 * array of nvlists, each which describes a valid leaf vdev. If this is an
5767 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
5768 * specified, as long as they are well-formed.
5771 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
5772 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
5773 vdev_labeltype_t label
)
5780 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5783 * It's acceptable to have no devs specified.
5785 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
5789 return (SET_ERROR(EINVAL
));
5792 * Make sure the pool is formatted with a version that supports this
5795 if (spa_version(spa
) < version
)
5796 return (SET_ERROR(ENOTSUP
));
5799 * Set the pending device list so we correctly handle device in-use
5802 sav
->sav_pending
= dev
;
5803 sav
->sav_npending
= ndev
;
5805 for (i
= 0; i
< ndev
; i
++) {
5806 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
5810 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5812 error
= SET_ERROR(EINVAL
);
5818 if ((error
= vdev_open(vd
)) == 0 &&
5819 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
5820 fnvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
5827 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
5834 sav
->sav_pending
= NULL
;
5835 sav
->sav_npending
= 0;
5840 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
5844 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5846 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5847 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
5848 VDEV_LABEL_SPARE
)) != 0) {
5852 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
5853 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
5854 VDEV_LABEL_L2CACHE
));
5858 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
5863 if (sav
->sav_config
!= NULL
) {
5869 * Generate new dev list by concatenating with the
5872 VERIFY0(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
5873 &olddevs
, &oldndevs
));
5875 newdevs
= kmem_alloc(sizeof (void *) *
5876 (ndevs
+ oldndevs
), KM_SLEEP
);
5877 for (i
= 0; i
< oldndevs
; i
++)
5878 newdevs
[i
] = fnvlist_dup(olddevs
[i
]);
5879 for (i
= 0; i
< ndevs
; i
++)
5880 newdevs
[i
+ oldndevs
] = fnvlist_dup(devs
[i
]);
5882 fnvlist_remove(sav
->sav_config
, config
);
5884 fnvlist_add_nvlist_array(sav
->sav_config
, config
,
5885 (const nvlist_t
* const *)newdevs
, ndevs
+ oldndevs
);
5886 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
5887 nvlist_free(newdevs
[i
]);
5888 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
5891 * Generate a new dev list.
5893 sav
->sav_config
= fnvlist_alloc();
5894 fnvlist_add_nvlist_array(sav
->sav_config
, config
,
5895 (const nvlist_t
* const *)devs
, ndevs
);
5900 * Stop and drop level 2 ARC devices
5903 spa_l2cache_drop(spa_t
*spa
)
5907 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
5909 for (i
= 0; i
< sav
->sav_count
; i
++) {
5912 vd
= sav
->sav_vdevs
[i
];
5915 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
5916 pool
!= 0ULL && l2arc_vdev_present(vd
))
5917 l2arc_remove_vdev(vd
);
5922 * Verify encryption parameters for spa creation. If we are encrypting, we must
5923 * have the encryption feature flag enabled.
5926 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
5927 boolean_t has_encryption
)
5929 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
5930 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
5932 return (SET_ERROR(ENOTSUP
));
5934 return (dmu_objset_create_crypt_check(NULL
, dcp
, NULL
));
5941 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
5942 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
5945 const char *altroot
= NULL
;
5950 uint64_t txg
= TXG_INITIAL
;
5951 nvlist_t
**spares
, **l2cache
;
5952 uint_t nspares
, nl2cache
;
5953 uint64_t version
, obj
, ndraid
= 0;
5954 boolean_t has_features
;
5955 boolean_t has_encryption
;
5956 boolean_t has_allocclass
;
5958 const char *feat_name
;
5959 const char *poolname
;
5962 if (props
== NULL
||
5963 nvlist_lookup_string(props
, "tname", &poolname
) != 0)
5964 poolname
= (char *)pool
;
5967 * If this pool already exists, return failure.
5969 mutex_enter(&spa_namespace_lock
);
5970 if (spa_lookup(poolname
) != NULL
) {
5971 mutex_exit(&spa_namespace_lock
);
5972 return (SET_ERROR(EEXIST
));
5976 * Allocate a new spa_t structure.
5978 nvl
= fnvlist_alloc();
5979 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
5980 (void) nvlist_lookup_string(props
,
5981 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5982 spa
= spa_add(poolname
, nvl
, altroot
);
5984 spa_activate(spa
, spa_mode_global
);
5986 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
5987 spa_deactivate(spa
);
5989 mutex_exit(&spa_namespace_lock
);
5994 * Temporary pool names should never be written to disk.
5996 if (poolname
!= pool
)
5997 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
5999 has_features
= B_FALSE
;
6000 has_encryption
= B_FALSE
;
6001 has_allocclass
= B_FALSE
;
6002 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
6003 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
6004 if (zpool_prop_feature(nvpair_name(elem
))) {
6005 has_features
= B_TRUE
;
6007 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
6008 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
6009 if (feat
== SPA_FEATURE_ENCRYPTION
)
6010 has_encryption
= B_TRUE
;
6011 if (feat
== SPA_FEATURE_ALLOCATION_CLASSES
)
6012 has_allocclass
= B_TRUE
;
6016 /* verify encryption params, if they were provided */
6018 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
6020 spa_deactivate(spa
);
6022 mutex_exit(&spa_namespace_lock
);
6026 if (!has_allocclass
&& zfs_special_devs(nvroot
, NULL
)) {
6027 spa_deactivate(spa
);
6029 mutex_exit(&spa_namespace_lock
);
6033 if (has_features
|| nvlist_lookup_uint64(props
,
6034 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
6035 version
= SPA_VERSION
;
6037 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6039 spa
->spa_first_txg
= txg
;
6040 spa
->spa_uberblock
.ub_txg
= txg
- 1;
6041 spa
->spa_uberblock
.ub_version
= version
;
6042 spa
->spa_ubsync
= spa
->spa_uberblock
;
6043 spa
->spa_load_state
= SPA_LOAD_CREATE
;
6044 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
6045 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
6046 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
6047 spa
->spa_indirect_vdevs_loaded
= B_TRUE
;
6050 * Create "The Godfather" zio to hold all async IOs
6052 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
6054 for (int i
= 0; i
< max_ncpus
; i
++) {
6055 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
6056 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
6057 ZIO_FLAG_GODFATHER
);
6061 * Create the root vdev.
6063 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6065 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
6067 ASSERT(error
!= 0 || rvd
!= NULL
);
6068 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
6070 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
6071 error
= SET_ERROR(EINVAL
);
6074 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
6075 (error
= vdev_draid_spare_create(nvroot
, rvd
, &ndraid
, 0)) == 0 &&
6076 (error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) == 0) {
6078 * instantiate the metaslab groups (this will dirty the vdevs)
6079 * we can no longer error exit past this point
6081 for (int c
= 0; error
== 0 && c
< rvd
->vdev_children
; c
++) {
6082 vdev_t
*vd
= rvd
->vdev_child
[c
];
6084 vdev_metaslab_set_size(vd
);
6085 vdev_expand(vd
, txg
);
6089 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6093 spa_deactivate(spa
);
6095 mutex_exit(&spa_namespace_lock
);
6100 * Get the list of spares, if specified.
6102 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
6103 &spares
, &nspares
) == 0) {
6104 spa
->spa_spares
.sav_config
= fnvlist_alloc();
6105 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
6106 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
6108 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6109 spa_load_spares(spa
);
6110 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6111 spa
->spa_spares
.sav_sync
= B_TRUE
;
6115 * Get the list of level 2 cache devices, if specified.
6117 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
6118 &l2cache
, &nl2cache
) == 0) {
6119 VERIFY0(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
6120 NV_UNIQUE_NAME
, KM_SLEEP
));
6121 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
6122 ZPOOL_CONFIG_L2CACHE
, (const nvlist_t
* const *)l2cache
,
6124 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6125 spa_load_l2cache(spa
);
6126 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6127 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6130 spa
->spa_is_initializing
= B_TRUE
;
6131 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
6132 spa
->spa_is_initializing
= B_FALSE
;
6135 * Create DDTs (dedup tables).
6139 * Create BRT table and BRT table object.
6143 spa_update_dspace(spa
);
6145 tx
= dmu_tx_create_assigned(dp
, txg
);
6148 * Create the pool's history object.
6150 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
6151 spa_history_create_obj(spa
, tx
);
6153 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
6154 spa_history_log_version(spa
, "create", tx
);
6157 * Create the pool config object.
6159 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6160 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
6161 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
6163 if (zap_add(spa
->spa_meta_objset
,
6164 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
6165 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
6166 cmn_err(CE_PANIC
, "failed to add pool config");
6169 if (zap_add(spa
->spa_meta_objset
,
6170 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
6171 sizeof (uint64_t), 1, &version
, tx
) != 0) {
6172 cmn_err(CE_PANIC
, "failed to add pool version");
6175 /* Newly created pools with the right version are always deflated. */
6176 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6177 spa
->spa_deflate
= TRUE
;
6178 if (zap_add(spa
->spa_meta_objset
,
6179 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6180 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
6181 cmn_err(CE_PANIC
, "failed to add deflate");
6186 * Create the deferred-free bpobj. Turn off compression
6187 * because sync-to-convergence takes longer if the blocksize
6190 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
6191 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
6192 ZIO_COMPRESS_OFF
, tx
);
6193 if (zap_add(spa
->spa_meta_objset
,
6194 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
6195 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
6196 cmn_err(CE_PANIC
, "failed to add bpobj");
6198 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
6199 spa
->spa_meta_objset
, obj
));
6202 * Generate some random noise for salted checksums to operate on.
6204 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
6205 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
6208 * Set pool properties.
6210 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
6211 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
6212 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
6213 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
6214 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
6215 spa
->spa_autotrim
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM
);
6217 if (props
!= NULL
) {
6218 spa_configfile_set(spa
, props
, B_FALSE
);
6219 spa_sync_props(props
, tx
);
6222 for (int i
= 0; i
< ndraid
; i
++)
6223 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
6227 spa
->spa_sync_on
= B_TRUE
;
6229 mmp_thread_start(spa
);
6230 txg_wait_synced(dp
, txg
);
6232 spa_spawn_aux_threads(spa
);
6234 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
, B_TRUE
);
6237 * Don't count references from objsets that are already closed
6238 * and are making their way through the eviction process.
6240 spa_evicting_os_wait(spa
);
6241 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
6242 spa
->spa_load_state
= SPA_LOAD_NONE
;
6246 mutex_exit(&spa_namespace_lock
);
6252 * Import a non-root pool into the system.
6255 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
6258 const char *altroot
= NULL
;
6259 spa_load_state_t state
= SPA_LOAD_IMPORT
;
6260 zpool_load_policy_t policy
;
6261 spa_mode_t mode
= spa_mode_global
;
6262 uint64_t readonly
= B_FALSE
;
6265 nvlist_t
**spares
, **l2cache
;
6266 uint_t nspares
, nl2cache
;
6269 * If a pool with this name exists, return failure.
6271 mutex_enter(&spa_namespace_lock
);
6272 if (spa_lookup(pool
) != NULL
) {
6273 mutex_exit(&spa_namespace_lock
);
6274 return (SET_ERROR(EEXIST
));
6278 * Create and initialize the spa structure.
6280 (void) nvlist_lookup_string(props
,
6281 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
6282 (void) nvlist_lookup_uint64(props
,
6283 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
6285 mode
= SPA_MODE_READ
;
6286 spa
= spa_add(pool
, config
, altroot
);
6287 spa
->spa_import_flags
= flags
;
6290 * Verbatim import - Take a pool and insert it into the namespace
6291 * as if it had been loaded at boot.
6293 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
6295 spa_configfile_set(spa
, props
, B_FALSE
);
6297 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
, B_FALSE
);
6298 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6299 zfs_dbgmsg("spa_import: verbatim import of %s", pool
);
6300 mutex_exit(&spa_namespace_lock
);
6304 spa_activate(spa
, mode
);
6307 * Don't start async tasks until we know everything is healthy.
6309 spa_async_suspend(spa
);
6311 zpool_get_load_policy(config
, &policy
);
6312 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
6313 state
= SPA_LOAD_RECOVER
;
6315 spa
->spa_config_source
= SPA_CONFIG_SRC_TRYIMPORT
;
6317 if (state
!= SPA_LOAD_RECOVER
) {
6318 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
6319 zfs_dbgmsg("spa_import: importing %s", pool
);
6321 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
6322 "(RECOVERY MODE)", pool
, (longlong_t
)policy
.zlp_txg
);
6324 error
= spa_load_best(spa
, state
, policy
.zlp_txg
, policy
.zlp_rewind
);
6327 * Propagate anything learned while loading the pool and pass it
6328 * back to caller (i.e. rewind info, missing devices, etc).
6330 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
, spa
->spa_load_info
);
6332 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6334 * Toss any existing sparelist, as it doesn't have any validity
6335 * anymore, and conflicts with spa_has_spare().
6337 if (spa
->spa_spares
.sav_config
) {
6338 nvlist_free(spa
->spa_spares
.sav_config
);
6339 spa
->spa_spares
.sav_config
= NULL
;
6340 spa_load_spares(spa
);
6342 if (spa
->spa_l2cache
.sav_config
) {
6343 nvlist_free(spa
->spa_l2cache
.sav_config
);
6344 spa
->spa_l2cache
.sav_config
= NULL
;
6345 spa_load_l2cache(spa
);
6348 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
6349 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6352 spa_configfile_set(spa
, props
, B_FALSE
);
6354 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
6355 (error
= spa_prop_set(spa
, props
)))) {
6357 spa_deactivate(spa
);
6359 mutex_exit(&spa_namespace_lock
);
6363 spa_async_resume(spa
);
6366 * Override any spares and level 2 cache devices as specified by
6367 * the user, as these may have correct device names/devids, etc.
6369 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
6370 &spares
, &nspares
) == 0) {
6371 if (spa
->spa_spares
.sav_config
)
6372 fnvlist_remove(spa
->spa_spares
.sav_config
,
6373 ZPOOL_CONFIG_SPARES
);
6375 spa
->spa_spares
.sav_config
= fnvlist_alloc();
6376 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
6377 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
6379 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6380 spa_load_spares(spa
);
6381 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6382 spa
->spa_spares
.sav_sync
= B_TRUE
;
6384 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
6385 &l2cache
, &nl2cache
) == 0) {
6386 if (spa
->spa_l2cache
.sav_config
)
6387 fnvlist_remove(spa
->spa_l2cache
.sav_config
,
6388 ZPOOL_CONFIG_L2CACHE
);
6390 spa
->spa_l2cache
.sav_config
= fnvlist_alloc();
6391 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
6392 ZPOOL_CONFIG_L2CACHE
, (const nvlist_t
* const *)l2cache
,
6394 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6395 spa_load_l2cache(spa
);
6396 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6397 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6401 * Check for any removed devices.
6403 if (spa
->spa_autoreplace
) {
6404 spa_aux_check_removed(&spa
->spa_spares
);
6405 spa_aux_check_removed(&spa
->spa_l2cache
);
6408 if (spa_writeable(spa
)) {
6410 * Update the config cache to include the newly-imported pool.
6412 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6416 * It's possible that the pool was expanded while it was exported.
6417 * We kick off an async task to handle this for us.
6419 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
6421 spa_history_log_version(spa
, "import", NULL
);
6423 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6425 mutex_exit(&spa_namespace_lock
);
6427 zvol_create_minors_recursive(pool
);
6435 spa_tryimport(nvlist_t
*tryconfig
)
6437 nvlist_t
*config
= NULL
;
6438 const char *poolname
, *cachefile
;
6442 zpool_load_policy_t policy
;
6444 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
6447 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
6451 * Create and initialize the spa structure.
6453 mutex_enter(&spa_namespace_lock
);
6454 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
6455 spa_activate(spa
, SPA_MODE_READ
);
6458 * Rewind pool if a max txg was provided.
6460 zpool_get_load_policy(spa
->spa_config
, &policy
);
6461 if (policy
.zlp_txg
!= UINT64_MAX
) {
6462 spa
->spa_load_max_txg
= policy
.zlp_txg
;
6463 spa
->spa_extreme_rewind
= B_TRUE
;
6464 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6465 poolname
, (longlong_t
)policy
.zlp_txg
);
6467 zfs_dbgmsg("spa_tryimport: importing %s", poolname
);
6470 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_CACHEFILE
, &cachefile
)
6472 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile
);
6473 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
6475 spa
->spa_config_source
= SPA_CONFIG_SRC_SCAN
;
6479 * spa_import() relies on a pool config fetched by spa_try_import()
6480 * for spare/cache devices. Import flags are not passed to
6481 * spa_tryimport(), which makes it return early due to a missing log
6482 * device and missing retrieving the cache device and spare eventually.
6483 * Passing ZFS_IMPORT_MISSING_LOG to spa_tryimport() makes it fetch
6484 * the correct configuration regardless of the missing log device.
6486 spa
->spa_import_flags
|= ZFS_IMPORT_MISSING_LOG
;
6488 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
);
6491 * If 'tryconfig' was at least parsable, return the current config.
6493 if (spa
->spa_root_vdev
!= NULL
) {
6494 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
6495 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, poolname
);
6496 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, state
);
6497 fnvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
6498 spa
->spa_uberblock
.ub_timestamp
);
6499 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
6500 spa
->spa_load_info
);
6501 fnvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
6505 * If the bootfs property exists on this pool then we
6506 * copy it out so that external consumers can tell which
6507 * pools are bootable.
6509 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
6510 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6513 * We have to play games with the name since the
6514 * pool was opened as TRYIMPORT_NAME.
6516 if (dsl_dsobj_to_dsname(spa_name(spa
),
6517 spa
->spa_bootfs
, tmpname
) == 0) {
6521 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6523 cp
= strchr(tmpname
, '/');
6525 (void) strlcpy(dsname
, tmpname
,
6528 (void) snprintf(dsname
, MAXPATHLEN
,
6529 "%s/%s", poolname
, ++cp
);
6531 fnvlist_add_string(config
, ZPOOL_CONFIG_BOOTFS
,
6533 kmem_free(dsname
, MAXPATHLEN
);
6535 kmem_free(tmpname
, MAXPATHLEN
);
6539 * Add the list of hot spares and level 2 cache devices.
6541 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6542 spa_add_spares(spa
, config
);
6543 spa_add_l2cache(spa
, config
);
6544 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6548 spa_deactivate(spa
);
6550 mutex_exit(&spa_namespace_lock
);
6556 * Pool export/destroy
6558 * The act of destroying or exporting a pool is very simple. We make sure there
6559 * is no more pending I/O and any references to the pool are gone. Then, we
6560 * update the pool state and sync all the labels to disk, removing the
6561 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6562 * we don't sync the labels or remove the configuration cache.
6565 spa_export_common(const char *pool
, int new_state
, nvlist_t
**oldconfig
,
6566 boolean_t force
, boolean_t hardforce
)
6574 if (!(spa_mode_global
& SPA_MODE_WRITE
))
6575 return (SET_ERROR(EROFS
));
6577 mutex_enter(&spa_namespace_lock
);
6578 if ((spa
= spa_lookup(pool
)) == NULL
) {
6579 mutex_exit(&spa_namespace_lock
);
6580 return (SET_ERROR(ENOENT
));
6583 if (spa
->spa_is_exporting
) {
6584 /* the pool is being exported by another thread */
6585 mutex_exit(&spa_namespace_lock
);
6586 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS
));
6588 spa
->spa_is_exporting
= B_TRUE
;
6591 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6592 * reacquire the namespace lock, and see if we can export.
6594 spa_open_ref(spa
, FTAG
);
6595 mutex_exit(&spa_namespace_lock
);
6596 spa_async_suspend(spa
);
6597 if (spa
->spa_zvol_taskq
) {
6598 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
6599 taskq_wait(spa
->spa_zvol_taskq
);
6601 mutex_enter(&spa_namespace_lock
);
6602 spa_close(spa
, FTAG
);
6604 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
6607 * The pool will be in core if it's openable, in which case we can
6608 * modify its state. Objsets may be open only because they're dirty,
6609 * so we have to force it to sync before checking spa_refcnt.
6611 if (spa
->spa_sync_on
) {
6612 txg_wait_synced(spa
->spa_dsl_pool
, 0);
6613 spa_evicting_os_wait(spa
);
6617 * A pool cannot be exported or destroyed if there are active
6618 * references. If we are resetting a pool, allow references by
6619 * fault injection handlers.
6621 if (!spa_refcount_zero(spa
) || (spa
->spa_inject_ref
!= 0)) {
6622 error
= SET_ERROR(EBUSY
);
6626 if (spa
->spa_sync_on
) {
6627 vdev_t
*rvd
= spa
->spa_root_vdev
;
6629 * A pool cannot be exported if it has an active shared spare.
6630 * This is to prevent other pools stealing the active spare
6631 * from an exported pool. At user's own will, such pool can
6632 * be forcedly exported.
6634 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
6635 spa_has_active_shared_spare(spa
)) {
6636 error
= SET_ERROR(EXDEV
);
6641 * We're about to export or destroy this pool. Make sure
6642 * we stop all initialization and trim activity here before
6643 * we set the spa_final_txg. This will ensure that all
6644 * dirty data resulting from the initialization is
6645 * committed to disk before we unload the pool.
6647 vdev_initialize_stop_all(rvd
, VDEV_INITIALIZE_ACTIVE
);
6648 vdev_trim_stop_all(rvd
, VDEV_TRIM_ACTIVE
);
6649 vdev_autotrim_stop_all(spa
);
6650 vdev_rebuild_stop_all(spa
);
6653 * We want this to be reflected on every label,
6654 * so mark them all dirty. spa_unload() will do the
6655 * final sync that pushes these changes out.
6657 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
6658 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6659 spa
->spa_state
= new_state
;
6660 vdev_config_dirty(rvd
);
6661 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6665 * If the log space map feature is enabled and the pool is
6666 * getting exported (but not destroyed), we want to spend some
6667 * time flushing as many metaslabs as we can in an attempt to
6668 * destroy log space maps and save import time. This has to be
6669 * done before we set the spa_final_txg, otherwise
6670 * spa_sync() -> spa_flush_metaslabs() may dirty the final TXGs.
6671 * spa_should_flush_logs_on_unload() should be called after
6672 * spa_state has been set to the new_state.
6674 if (spa_should_flush_logs_on_unload(spa
))
6675 spa_unload_log_sm_flush_all(spa
);
6677 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
6678 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6679 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
6681 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6688 if (new_state
== POOL_STATE_DESTROYED
)
6689 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
6690 else if (new_state
== POOL_STATE_EXPORTED
)
6691 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
6693 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6695 spa_deactivate(spa
);
6698 if (oldconfig
&& spa
->spa_config
)
6699 *oldconfig
= fnvlist_dup(spa
->spa_config
);
6701 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
6703 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
, B_FALSE
);
6707 * If spa_remove() is not called for this spa_t and
6708 * there is any possibility that it can be reused,
6709 * we make sure to reset the exporting flag.
6711 spa
->spa_is_exporting
= B_FALSE
;
6714 mutex_exit(&spa_namespace_lock
);
6718 spa
->spa_is_exporting
= B_FALSE
;
6719 spa_async_resume(spa
);
6720 mutex_exit(&spa_namespace_lock
);
6725 * Destroy a storage pool.
6728 spa_destroy(const char *pool
)
6730 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
6735 * Export a storage pool.
6738 spa_export(const char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
6739 boolean_t hardforce
)
6741 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
6746 * Similar to spa_export(), this unloads the spa_t without actually removing it
6747 * from the namespace in any way.
6750 spa_reset(const char *pool
)
6752 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
6757 * ==========================================================================
6758 * Device manipulation
6759 * ==========================================================================
6763 * This is called as a synctask to increment the draid feature flag
6766 spa_draid_feature_incr(void *arg
, dmu_tx_t
*tx
)
6768 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6769 int draid
= (int)(uintptr_t)arg
;
6771 for (int c
= 0; c
< draid
; c
++)
6772 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
6776 * Add a device to a storage pool.
6779 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
6781 uint64_t txg
, ndraid
= 0;
6783 vdev_t
*rvd
= spa
->spa_root_vdev
;
6785 nvlist_t
**spares
, **l2cache
;
6786 uint_t nspares
, nl2cache
;
6788 ASSERT(spa_writeable(spa
));
6790 txg
= spa_vdev_enter(spa
);
6792 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
6793 VDEV_ALLOC_ADD
)) != 0)
6794 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6796 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
6798 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
6802 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
6806 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
6807 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6809 if (vd
->vdev_children
!= 0 &&
6810 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0) {
6811 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6815 * The virtual dRAID spares must be added after vdev tree is created
6816 * and the vdev guids are generated. The guid of their associated
6817 * dRAID is stored in the config and used when opening the spare.
6819 if ((error
= vdev_draid_spare_create(nvroot
, vd
, &ndraid
,
6820 rvd
->vdev_children
)) == 0) {
6821 if (ndraid
> 0 && nvlist_lookup_nvlist_array(nvroot
,
6822 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) != 0)
6825 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6829 * We must validate the spares and l2cache devices after checking the
6830 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
6832 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
6833 return (spa_vdev_exit(spa
, vd
, txg
, error
));
6836 * If we are in the middle of a device removal, we can only add
6837 * devices which match the existing devices in the pool.
6838 * If we are in the middle of a removal, or have some indirect
6839 * vdevs, we can not add raidz or dRAID top levels.
6841 if (spa
->spa_vdev_removal
!= NULL
||
6842 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
6843 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6844 tvd
= vd
->vdev_child
[c
];
6845 if (spa
->spa_vdev_removal
!= NULL
&&
6846 tvd
->vdev_ashift
!= spa
->spa_max_ashift
) {
6847 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6849 /* Fail if top level vdev is raidz or a dRAID */
6850 if (vdev_get_nparity(tvd
) != 0)
6851 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
6854 * Need the top level mirror to be
6855 * a mirror of leaf vdevs only
6857 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
6858 for (uint64_t cid
= 0;
6859 cid
< tvd
->vdev_children
; cid
++) {
6860 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
6861 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
6862 return (spa_vdev_exit(spa
, vd
,
6870 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6871 tvd
= vd
->vdev_child
[c
];
6872 vdev_remove_child(vd
, tvd
);
6873 tvd
->vdev_id
= rvd
->vdev_children
;
6874 vdev_add_child(rvd
, tvd
);
6875 vdev_config_dirty(tvd
);
6879 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
6880 ZPOOL_CONFIG_SPARES
);
6881 spa_load_spares(spa
);
6882 spa
->spa_spares
.sav_sync
= B_TRUE
;
6885 if (nl2cache
!= 0) {
6886 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
6887 ZPOOL_CONFIG_L2CACHE
);
6888 spa_load_l2cache(spa
);
6889 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6893 * We can't increment a feature while holding spa_vdev so we
6894 * have to do it in a synctask.
6899 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
6900 dsl_sync_task_nowait(spa
->spa_dsl_pool
, spa_draid_feature_incr
,
6901 (void *)(uintptr_t)ndraid
, tx
);
6906 * We have to be careful when adding new vdevs to an existing pool.
6907 * If other threads start allocating from these vdevs before we
6908 * sync the config cache, and we lose power, then upon reboot we may
6909 * fail to open the pool because there are DVAs that the config cache
6910 * can't translate. Therefore, we first add the vdevs without
6911 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
6912 * and then let spa_config_update() initialize the new metaslabs.
6914 * spa_load() checks for added-but-not-initialized vdevs, so that
6915 * if we lose power at any point in this sequence, the remaining
6916 * steps will be completed the next time we load the pool.
6918 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
6920 mutex_enter(&spa_namespace_lock
);
6921 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6922 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
6923 mutex_exit(&spa_namespace_lock
);
6929 * Attach a device to a vdev specified by its guid. The vdev type can be
6930 * a mirror, a raidz, or a leaf device that is also a top-level (e.g. a
6931 * single device). When the vdev is a single device, a mirror vdev will be
6932 * automatically inserted.
6934 * If 'replacing' is specified, the new device is intended to replace the
6935 * existing device; in this case the two devices are made into their own
6936 * mirror using the 'replacing' vdev, which is functionally identical to
6937 * the mirror vdev (it actually reuses all the same ops) but has a few
6938 * extra rules: you can't attach to it after it's been created, and upon
6939 * completion of resilvering, the first disk (the one being replaced)
6940 * is automatically detached.
6942 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
6943 * should be performed instead of traditional healing reconstruction. From
6944 * an administrators perspective these are both resilver operations.
6947 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
,
6950 uint64_t txg
, dtl_max_txg
;
6951 vdev_t
*rvd
= spa
->spa_root_vdev
;
6952 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
6954 char *oldvdpath
, *newvdpath
;
6955 int newvd_isspare
= B_FALSE
;
6958 ASSERT(spa_writeable(spa
));
6960 txg
= spa_vdev_enter(spa
);
6962 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
6964 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
6965 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
6966 error
= (spa_has_checkpoint(spa
)) ?
6967 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
6968 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6972 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_DEVICE_REBUILD
))
6973 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
6975 if (dsl_scan_resilvering(spa_get_dsl(spa
)) ||
6976 dsl_scan_resilver_scheduled(spa_get_dsl(spa
))) {
6977 return (spa_vdev_exit(spa
, NULL
, txg
,
6978 ZFS_ERR_RESILVER_IN_PROGRESS
));
6981 if (vdev_rebuild_active(rvd
))
6982 return (spa_vdev_exit(spa
, NULL
, txg
,
6983 ZFS_ERR_REBUILD_IN_PROGRESS
));
6986 if (spa
->spa_vdev_removal
!= NULL
) {
6987 return (spa_vdev_exit(spa
, NULL
, txg
,
6988 ZFS_ERR_DEVRM_IN_PROGRESS
));
6992 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
6994 boolean_t raidz
= oldvd
->vdev_ops
== &vdev_raidz_ops
;
6997 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_RAIDZ_EXPANSION
))
6998 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7001 * Can't expand a raidz while prior expand is in progress.
7003 if (spa
->spa_raidz_expand
!= NULL
) {
7004 return (spa_vdev_exit(spa
, NULL
, txg
,
7005 ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS
));
7007 } else if (!oldvd
->vdev_ops
->vdev_op_leaf
) {
7008 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7014 pvd
= oldvd
->vdev_parent
;
7016 if (spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
7017 VDEV_ALLOC_ATTACH
) != 0)
7018 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7020 if (newrootvd
->vdev_children
!= 1)
7021 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
7023 newvd
= newrootvd
->vdev_child
[0];
7025 if (!newvd
->vdev_ops
->vdev_op_leaf
)
7026 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
7028 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
7029 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
7032 * log, dedup and special vdevs should not be replaced by spares.
7034 if ((oldvd
->vdev_top
->vdev_alloc_bias
!= VDEV_BIAS_NONE
||
7035 oldvd
->vdev_top
->vdev_islog
) && newvd
->vdev_isspare
) {
7036 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7040 * A dRAID spare can only replace a child of its parent dRAID vdev.
7042 if (newvd
->vdev_ops
== &vdev_draid_spare_ops
&&
7043 oldvd
->vdev_top
!= vdev_draid_spare_get_parent(newvd
)) {
7044 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7049 * For rebuilds, the top vdev must support reconstruction
7050 * using only space maps. This means the only allowable
7051 * vdevs types are the root vdev, a mirror, or dRAID.
7054 if (pvd
->vdev_top
!= NULL
)
7055 tvd
= pvd
->vdev_top
;
7057 if (tvd
->vdev_ops
!= &vdev_mirror_ops
&&
7058 tvd
->vdev_ops
!= &vdev_root_ops
&&
7059 tvd
->vdev_ops
!= &vdev_draid_ops
) {
7060 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7066 * For attach, the only allowable parent is a mirror or
7067 * the root vdev. A raidz vdev can be attached to, but
7068 * you cannot attach to a raidz child.
7070 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
7071 pvd
->vdev_ops
!= &vdev_root_ops
&&
7073 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7075 pvops
= &vdev_mirror_ops
;
7078 * Active hot spares can only be replaced by inactive hot
7081 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
7082 oldvd
->vdev_isspare
&&
7083 !spa_has_spare(spa
, newvd
->vdev_guid
))
7084 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7087 * If the source is a hot spare, and the parent isn't already a
7088 * spare, then we want to create a new hot spare. Otherwise, we
7089 * want to create a replacing vdev. The user is not allowed to
7090 * attach to a spared vdev child unless the 'isspare' state is
7091 * the same (spare replaces spare, non-spare replaces
7094 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
7095 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
7096 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7097 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
7098 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
7099 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7102 if (newvd
->vdev_isspare
)
7103 pvops
= &vdev_spare_ops
;
7105 pvops
= &vdev_replacing_ops
;
7109 * Make sure the new device is big enough.
7111 vdev_t
*min_vdev
= raidz
? oldvd
->vdev_child
[0] : oldvd
;
7112 if (newvd
->vdev_asize
< vdev_get_min_asize(min_vdev
))
7113 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
7116 * The new device cannot have a higher alignment requirement
7117 * than the top-level vdev.
7119 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
7120 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7123 * RAIDZ-expansion-specific checks.
7126 if (vdev_raidz_attach_check(newvd
) != 0)
7127 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7130 * Fail early if a child is not healthy or being replaced
7132 for (int i
= 0; i
< oldvd
->vdev_children
; i
++) {
7133 if (vdev_is_dead(oldvd
->vdev_child
[i
]) ||
7134 !oldvd
->vdev_child
[i
]->vdev_ops
->vdev_op_leaf
) {
7135 return (spa_vdev_exit(spa
, newrootvd
, txg
,
7138 /* Also fail if reserved boot area is in-use */
7139 if (vdev_check_boot_reserve(spa
, oldvd
->vdev_child
[i
])
7141 return (spa_vdev_exit(spa
, newrootvd
, txg
,
7149 * Note: oldvdpath is freed by spa_strfree(), but
7150 * kmem_asprintf() is freed by kmem_strfree(), so we have to
7151 * move it to a spa_strdup-ed string.
7153 char *tmp
= kmem_asprintf("raidz%u-%u",
7154 (uint_t
)vdev_get_nparity(oldvd
), (uint_t
)oldvd
->vdev_id
);
7155 oldvdpath
= spa_strdup(tmp
);
7158 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
7160 newvdpath
= spa_strdup(newvd
->vdev_path
);
7163 * If this is an in-place replacement, update oldvd's path and devid
7164 * to make it distinguishable from newvd, and unopenable from now on.
7166 if (strcmp(oldvdpath
, newvdpath
) == 0) {
7167 spa_strfree(oldvd
->vdev_path
);
7168 oldvd
->vdev_path
= kmem_alloc(strlen(newvdpath
) + 5,
7170 (void) sprintf(oldvd
->vdev_path
, "%s/old",
7172 if (oldvd
->vdev_devid
!= NULL
) {
7173 spa_strfree(oldvd
->vdev_devid
);
7174 oldvd
->vdev_devid
= NULL
;
7176 spa_strfree(oldvdpath
);
7177 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
7181 * If the parent is not a mirror, or if we're replacing, insert the new
7182 * mirror/replacing/spare vdev above oldvd.
7184 if (!raidz
&& pvd
->vdev_ops
!= pvops
) {
7185 pvd
= vdev_add_parent(oldvd
, pvops
);
7186 ASSERT(pvd
->vdev_ops
== pvops
);
7187 ASSERT(oldvd
->vdev_parent
== pvd
);
7190 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
7193 * Extract the new device from its root and add it to pvd.
7195 vdev_remove_child(newrootvd
, newvd
);
7196 newvd
->vdev_id
= pvd
->vdev_children
;
7197 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
7198 vdev_add_child(pvd
, newvd
);
7201 * Reevaluate the parent vdev state.
7203 vdev_propagate_state(pvd
);
7205 tvd
= newvd
->vdev_top
;
7206 ASSERT(pvd
->vdev_top
== tvd
);
7207 ASSERT(tvd
->vdev_parent
== rvd
);
7209 vdev_config_dirty(tvd
);
7212 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
7213 * for any dmu_sync-ed blocks. It will propagate upward when
7214 * spa_vdev_exit() calls vdev_dtl_reassess().
7216 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
7220 * Wait for the youngest allocations and frees to sync,
7221 * and then wait for the deferral of those frees to finish.
7223 spa_vdev_config_exit(spa
, NULL
,
7224 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
7226 vdev_initialize_stop_all(tvd
, VDEV_INITIALIZE_ACTIVE
);
7227 vdev_trim_stop_all(tvd
, VDEV_TRIM_ACTIVE
);
7228 vdev_autotrim_stop_wait(tvd
);
7230 dtl_max_txg
= spa_vdev_config_enter(spa
);
7232 tvd
->vdev_rz_expanding
= B_TRUE
;
7234 vdev_dirty_leaves(tvd
, VDD_DTL
, dtl_max_txg
);
7235 vdev_config_dirty(tvd
);
7237 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
,
7239 dsl_sync_task_nowait(spa
->spa_dsl_pool
, vdev_raidz_attach_sync
,
7243 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
7244 dtl_max_txg
- TXG_INITIAL
);
7246 if (newvd
->vdev_isspare
) {
7247 spa_spare_activate(newvd
);
7248 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
7251 newvd_isspare
= newvd
->vdev_isspare
;
7254 * Mark newvd's DTL dirty in this txg.
7256 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
7259 * Schedule the resilver or rebuild to restart in the future.
7260 * We do this to ensure that dmu_sync-ed blocks have been
7261 * stitched into the respective datasets.
7264 newvd
->vdev_rebuild_txg
= txg
;
7268 newvd
->vdev_resilver_txg
= txg
;
7270 if (dsl_scan_resilvering(spa_get_dsl(spa
)) &&
7271 spa_feature_is_enabled(spa
,
7272 SPA_FEATURE_RESILVER_DEFER
)) {
7273 vdev_defer_resilver(newvd
);
7275 dsl_scan_restart_resilver(spa
->spa_dsl_pool
,
7281 if (spa
->spa_bootfs
)
7282 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
7284 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
7289 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
7291 spa_history_log_internal(spa
, "vdev attach", NULL
,
7292 "%s vdev=%s %s vdev=%s",
7293 replacing
&& newvd_isspare
? "spare in" :
7294 replacing
? "replace" : "attach", newvdpath
,
7295 replacing
? "for" : "to", oldvdpath
);
7297 spa_strfree(oldvdpath
);
7298 spa_strfree(newvdpath
);
7304 * Detach a device from a mirror or replacing vdev.
7306 * If 'replace_done' is specified, only detach if the parent
7307 * is a replacing or a spare vdev.
7310 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
7314 vdev_t
*rvd __maybe_unused
= spa
->spa_root_vdev
;
7315 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
7316 boolean_t unspare
= B_FALSE
;
7317 uint64_t unspare_guid
= 0;
7320 ASSERT(spa_writeable(spa
));
7322 txg
= spa_vdev_detach_enter(spa
, guid
);
7324 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7327 * Besides being called directly from the userland through the
7328 * ioctl interface, spa_vdev_detach() can be potentially called
7329 * at the end of spa_vdev_resilver_done().
7331 * In the regular case, when we have a checkpoint this shouldn't
7332 * happen as we never empty the DTLs of a vdev during the scrub
7333 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
7334 * should never get here when we have a checkpoint.
7336 * That said, even in a case when we checkpoint the pool exactly
7337 * as spa_vdev_resilver_done() calls this function everything
7338 * should be fine as the resilver will return right away.
7340 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7341 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7342 error
= (spa_has_checkpoint(spa
)) ?
7343 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7344 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7348 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
7350 if (!vd
->vdev_ops
->vdev_op_leaf
)
7351 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7353 pvd
= vd
->vdev_parent
;
7356 * If the parent/child relationship is not as expected, don't do it.
7357 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
7358 * vdev that's replacing B with C. The user's intent in replacing
7359 * is to go from M(A,B) to M(A,C). If the user decides to cancel
7360 * the replace by detaching C, the expected behavior is to end up
7361 * M(A,B). But suppose that right after deciding to detach C,
7362 * the replacement of B completes. We would have M(A,C), and then
7363 * ask to detach C, which would leave us with just A -- not what
7364 * the user wanted. To prevent this, we make sure that the
7365 * parent/child relationship hasn't changed -- in this example,
7366 * that C's parent is still the replacing vdev R.
7368 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
7369 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
7372 * Only 'replacing' or 'spare' vdevs can be replaced.
7374 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
7375 pvd
->vdev_ops
!= &vdev_spare_ops
)
7376 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7378 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
7379 spa_version(spa
) >= SPA_VERSION_SPARES
);
7382 * Only mirror, replacing, and spare vdevs support detach.
7384 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
7385 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
7386 pvd
->vdev_ops
!= &vdev_spare_ops
)
7387 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7390 * If this device has the only valid copy of some data,
7391 * we cannot safely detach it.
7393 if (vdev_dtl_required(vd
))
7394 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
7396 ASSERT(pvd
->vdev_children
>= 2);
7399 * If we are detaching the second disk from a replacing vdev, then
7400 * check to see if we changed the original vdev's path to have "/old"
7401 * at the end in spa_vdev_attach(). If so, undo that change now.
7403 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
7404 vd
->vdev_path
!= NULL
) {
7405 size_t len
= strlen(vd
->vdev_path
);
7407 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
7408 cvd
= pvd
->vdev_child
[c
];
7410 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
7413 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
7414 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
7415 spa_strfree(cvd
->vdev_path
);
7416 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
7423 * If we are detaching the original disk from a normal spare, then it
7424 * implies that the spare should become a real disk, and be removed
7425 * from the active spare list for the pool. dRAID spares on the
7426 * other hand are coupled to the pool and thus should never be removed
7427 * from the spares list.
7429 if (pvd
->vdev_ops
== &vdev_spare_ops
&& vd
->vdev_id
== 0) {
7430 vdev_t
*last_cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7432 if (last_cvd
->vdev_isspare
&&
7433 last_cvd
->vdev_ops
!= &vdev_draid_spare_ops
) {
7439 * Erase the disk labels so the disk can be used for other things.
7440 * This must be done after all other error cases are handled,
7441 * but before we disembowel vd (so we can still do I/O to it).
7442 * But if we can't do it, don't treat the error as fatal --
7443 * it may be that the unwritability of the disk is the reason
7444 * it's being detached!
7446 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
7449 * Remove vd from its parent and compact the parent's children.
7451 vdev_remove_child(pvd
, vd
);
7452 vdev_compact_children(pvd
);
7455 * Remember one of the remaining children so we can get tvd below.
7457 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7460 * If we need to remove the remaining child from the list of hot spares,
7461 * do it now, marking the vdev as no longer a spare in the process.
7462 * We must do this before vdev_remove_parent(), because that can
7463 * change the GUID if it creates a new toplevel GUID. For a similar
7464 * reason, we must remove the spare now, in the same txg as the detach;
7465 * otherwise someone could attach a new sibling, change the GUID, and
7466 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
7469 ASSERT(cvd
->vdev_isspare
);
7470 spa_spare_remove(cvd
);
7471 unspare_guid
= cvd
->vdev_guid
;
7472 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
7473 cvd
->vdev_unspare
= B_TRUE
;
7477 * If the parent mirror/replacing vdev only has one child,
7478 * the parent is no longer needed. Remove it from the tree.
7480 if (pvd
->vdev_children
== 1) {
7481 if (pvd
->vdev_ops
== &vdev_spare_ops
)
7482 cvd
->vdev_unspare
= B_FALSE
;
7483 vdev_remove_parent(cvd
);
7487 * We don't set tvd until now because the parent we just removed
7488 * may have been the previous top-level vdev.
7490 tvd
= cvd
->vdev_top
;
7491 ASSERT(tvd
->vdev_parent
== rvd
);
7494 * Reevaluate the parent vdev state.
7496 vdev_propagate_state(cvd
);
7499 * If the 'autoexpand' property is set on the pool then automatically
7500 * try to expand the size of the pool. For example if the device we
7501 * just detached was smaller than the others, it may be possible to
7502 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
7503 * first so that we can obtain the updated sizes of the leaf vdevs.
7505 if (spa
->spa_autoexpand
) {
7507 vdev_expand(tvd
, txg
);
7510 vdev_config_dirty(tvd
);
7513 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
7514 * vd->vdev_detached is set and free vd's DTL object in syncing context.
7515 * But first make sure we're not on any *other* txg's DTL list, to
7516 * prevent vd from being accessed after it's freed.
7518 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
7519 for (int t
= 0; t
< TXG_SIZE
; t
++)
7520 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
7521 vd
->vdev_detached
= B_TRUE
;
7522 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
7524 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
7525 spa_notify_waiters(spa
);
7527 /* hang on to the spa before we release the lock */
7528 spa_open_ref(spa
, FTAG
);
7530 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
7532 spa_history_log_internal(spa
, "detach", NULL
,
7534 spa_strfree(vdpath
);
7537 * If this was the removal of the original device in a hot spare vdev,
7538 * then we want to go through and remove the device from the hot spare
7539 * list of every other pool.
7542 spa_t
*altspa
= NULL
;
7544 mutex_enter(&spa_namespace_lock
);
7545 while ((altspa
= spa_next(altspa
)) != NULL
) {
7546 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
7550 spa_open_ref(altspa
, FTAG
);
7551 mutex_exit(&spa_namespace_lock
);
7552 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
7553 mutex_enter(&spa_namespace_lock
);
7554 spa_close(altspa
, FTAG
);
7556 mutex_exit(&spa_namespace_lock
);
7558 /* search the rest of the vdevs for spares to remove */
7559 spa_vdev_resilver_done(spa
);
7562 /* all done with the spa; OK to release */
7563 mutex_enter(&spa_namespace_lock
);
7564 spa_close(spa
, FTAG
);
7565 mutex_exit(&spa_namespace_lock
);
7571 spa_vdev_initialize_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7574 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7576 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7578 /* Look up vdev and ensure it's a leaf. */
7579 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7580 if (vd
== NULL
|| vd
->vdev_detached
) {
7581 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7582 return (SET_ERROR(ENODEV
));
7583 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7584 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7585 return (SET_ERROR(EINVAL
));
7586 } else if (!vdev_writeable(vd
)) {
7587 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7588 return (SET_ERROR(EROFS
));
7590 mutex_enter(&vd
->vdev_initialize_lock
);
7591 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7594 * When we activate an initialize action we check to see
7595 * if the vdev_initialize_thread is NULL. We do this instead
7596 * of using the vdev_initialize_state since there might be
7597 * a previous initialization process which has completed but
7598 * the thread is not exited.
7600 if (cmd_type
== POOL_INITIALIZE_START
&&
7601 (vd
->vdev_initialize_thread
!= NULL
||
7602 vd
->vdev_top
->vdev_removing
|| vd
->vdev_top
->vdev_rz_expanding
)) {
7603 mutex_exit(&vd
->vdev_initialize_lock
);
7604 return (SET_ERROR(EBUSY
));
7605 } else if (cmd_type
== POOL_INITIALIZE_CANCEL
&&
7606 (vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
&&
7607 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_SUSPENDED
)) {
7608 mutex_exit(&vd
->vdev_initialize_lock
);
7609 return (SET_ERROR(ESRCH
));
7610 } else if (cmd_type
== POOL_INITIALIZE_SUSPEND
&&
7611 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
) {
7612 mutex_exit(&vd
->vdev_initialize_lock
);
7613 return (SET_ERROR(ESRCH
));
7614 } else if (cmd_type
== POOL_INITIALIZE_UNINIT
&&
7615 vd
->vdev_initialize_thread
!= NULL
) {
7616 mutex_exit(&vd
->vdev_initialize_lock
);
7617 return (SET_ERROR(EBUSY
));
7621 case POOL_INITIALIZE_START
:
7622 vdev_initialize(vd
);
7624 case POOL_INITIALIZE_CANCEL
:
7625 vdev_initialize_stop(vd
, VDEV_INITIALIZE_CANCELED
, vd_list
);
7627 case POOL_INITIALIZE_SUSPEND
:
7628 vdev_initialize_stop(vd
, VDEV_INITIALIZE_SUSPENDED
, vd_list
);
7630 case POOL_INITIALIZE_UNINIT
:
7631 vdev_uninitialize(vd
);
7634 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7636 mutex_exit(&vd
->vdev_initialize_lock
);
7642 spa_vdev_initialize(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
,
7643 nvlist_t
*vdev_errlist
)
7645 int total_errors
= 0;
7648 list_create(&vd_list
, sizeof (vdev_t
),
7649 offsetof(vdev_t
, vdev_initialize_node
));
7652 * We hold the namespace lock through the whole function
7653 * to prevent any changes to the pool while we're starting or
7654 * stopping initialization. The config and state locks are held so that
7655 * we can properly assess the vdev state before we commit to
7656 * the initializing operation.
7658 mutex_enter(&spa_namespace_lock
);
7660 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7661 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7662 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7664 int error
= spa_vdev_initialize_impl(spa
, vdev_guid
, cmd_type
,
7667 char guid_as_str
[MAXNAMELEN
];
7669 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7670 "%llu", (unsigned long long)vdev_guid
);
7671 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7676 /* Wait for all initialize threads to stop. */
7677 vdev_initialize_stop_wait(spa
, &vd_list
);
7679 /* Sync out the initializing state */
7680 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7681 mutex_exit(&spa_namespace_lock
);
7683 list_destroy(&vd_list
);
7685 return (total_errors
);
7689 spa_vdev_trim_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7690 uint64_t rate
, boolean_t partial
, boolean_t secure
, list_t
*vd_list
)
7692 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7694 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7696 /* Look up vdev and ensure it's a leaf. */
7697 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7698 if (vd
== NULL
|| vd
->vdev_detached
) {
7699 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7700 return (SET_ERROR(ENODEV
));
7701 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7702 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7703 return (SET_ERROR(EINVAL
));
7704 } else if (!vdev_writeable(vd
)) {
7705 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7706 return (SET_ERROR(EROFS
));
7707 } else if (!vd
->vdev_has_trim
) {
7708 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7709 return (SET_ERROR(EOPNOTSUPP
));
7710 } else if (secure
&& !vd
->vdev_has_securetrim
) {
7711 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7712 return (SET_ERROR(EOPNOTSUPP
));
7714 mutex_enter(&vd
->vdev_trim_lock
);
7715 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7718 * When we activate a TRIM action we check to see if the
7719 * vdev_trim_thread is NULL. We do this instead of using the
7720 * vdev_trim_state since there might be a previous TRIM process
7721 * which has completed but the thread is not exited.
7723 if (cmd_type
== POOL_TRIM_START
&&
7724 (vd
->vdev_trim_thread
!= NULL
|| vd
->vdev_top
->vdev_removing
||
7725 vd
->vdev_top
->vdev_rz_expanding
)) {
7726 mutex_exit(&vd
->vdev_trim_lock
);
7727 return (SET_ERROR(EBUSY
));
7728 } else if (cmd_type
== POOL_TRIM_CANCEL
&&
7729 (vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
&&
7730 vd
->vdev_trim_state
!= VDEV_TRIM_SUSPENDED
)) {
7731 mutex_exit(&vd
->vdev_trim_lock
);
7732 return (SET_ERROR(ESRCH
));
7733 } else if (cmd_type
== POOL_TRIM_SUSPEND
&&
7734 vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
) {
7735 mutex_exit(&vd
->vdev_trim_lock
);
7736 return (SET_ERROR(ESRCH
));
7740 case POOL_TRIM_START
:
7741 vdev_trim(vd
, rate
, partial
, secure
);
7743 case POOL_TRIM_CANCEL
:
7744 vdev_trim_stop(vd
, VDEV_TRIM_CANCELED
, vd_list
);
7746 case POOL_TRIM_SUSPEND
:
7747 vdev_trim_stop(vd
, VDEV_TRIM_SUSPENDED
, vd_list
);
7750 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7752 mutex_exit(&vd
->vdev_trim_lock
);
7758 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
7759 * TRIM threads for each child vdev. These threads pass over all of the free
7760 * space in the vdev's metaslabs and issues TRIM commands for that space.
7763 spa_vdev_trim(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
, uint64_t rate
,
7764 boolean_t partial
, boolean_t secure
, nvlist_t
*vdev_errlist
)
7766 int total_errors
= 0;
7769 list_create(&vd_list
, sizeof (vdev_t
),
7770 offsetof(vdev_t
, vdev_trim_node
));
7773 * We hold the namespace lock through the whole function
7774 * to prevent any changes to the pool while we're starting or
7775 * stopping TRIM. The config and state locks are held so that
7776 * we can properly assess the vdev state before we commit to
7777 * the TRIM operation.
7779 mutex_enter(&spa_namespace_lock
);
7781 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7782 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7783 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7785 int error
= spa_vdev_trim_impl(spa
, vdev_guid
, cmd_type
,
7786 rate
, partial
, secure
, &vd_list
);
7788 char guid_as_str
[MAXNAMELEN
];
7790 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7791 "%llu", (unsigned long long)vdev_guid
);
7792 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7797 /* Wait for all TRIM threads to stop. */
7798 vdev_trim_stop_wait(spa
, &vd_list
);
7800 /* Sync out the TRIM state */
7801 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7802 mutex_exit(&spa_namespace_lock
);
7804 list_destroy(&vd_list
);
7806 return (total_errors
);
7810 * Split a set of devices from their mirrors, and create a new pool from them.
7813 spa_vdev_split_mirror(spa_t
*spa
, const char *newname
, nvlist_t
*config
,
7814 nvlist_t
*props
, boolean_t exp
)
7817 uint64_t txg
, *glist
;
7819 uint_t c
, children
, lastlog
;
7820 nvlist_t
**child
, *nvl
, *tmp
;
7822 const char *altroot
= NULL
;
7823 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
7824 boolean_t activate_slog
;
7826 ASSERT(spa_writeable(spa
));
7828 txg
= spa_vdev_enter(spa
);
7830 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7831 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7832 error
= (spa_has_checkpoint(spa
)) ?
7833 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7834 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7837 /* clear the log and flush everything up to now */
7838 activate_slog
= spa_passivate_log(spa
);
7839 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
7840 error
= spa_reset_logs(spa
);
7841 txg
= spa_vdev_config_enter(spa
);
7844 spa_activate_log(spa
);
7847 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7849 /* check new spa name before going any further */
7850 if (spa_lookup(newname
) != NULL
)
7851 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
7854 * scan through all the children to ensure they're all mirrors
7856 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
7857 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
7859 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7861 /* first, check to ensure we've got the right child count */
7862 rvd
= spa
->spa_root_vdev
;
7864 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
7865 vdev_t
*vd
= rvd
->vdev_child
[c
];
7867 /* don't count the holes & logs as children */
7868 if (vd
->vdev_islog
|| (vd
->vdev_ops
!= &vdev_indirect_ops
&&
7869 !vdev_is_concrete(vd
))) {
7877 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
7878 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7880 /* next, ensure no spare or cache devices are part of the split */
7881 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
7882 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
7883 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7885 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
7886 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
7888 /* then, loop over each vdev and validate it */
7889 for (c
= 0; c
< children
; c
++) {
7890 uint64_t is_hole
= 0;
7892 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
7896 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
7897 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
7900 error
= SET_ERROR(EINVAL
);
7905 /* deal with indirect vdevs */
7906 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ops
==
7910 /* which disk is going to be split? */
7911 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
7913 error
= SET_ERROR(EINVAL
);
7917 /* look it up in the spa */
7918 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
7919 if (vml
[c
] == NULL
) {
7920 error
= SET_ERROR(ENODEV
);
7924 /* make sure there's nothing stopping the split */
7925 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
7926 vml
[c
]->vdev_islog
||
7927 !vdev_is_concrete(vml
[c
]) ||
7928 vml
[c
]->vdev_isspare
||
7929 vml
[c
]->vdev_isl2cache
||
7930 !vdev_writeable(vml
[c
]) ||
7931 vml
[c
]->vdev_children
!= 0 ||
7932 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
7933 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
7934 error
= SET_ERROR(EINVAL
);
7938 if (vdev_dtl_required(vml
[c
]) ||
7939 vdev_resilver_needed(vml
[c
], NULL
, NULL
)) {
7940 error
= SET_ERROR(EBUSY
);
7944 /* we need certain info from the top level */
7945 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
7946 vml
[c
]->vdev_top
->vdev_ms_array
);
7947 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
7948 vml
[c
]->vdev_top
->vdev_ms_shift
);
7949 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
7950 vml
[c
]->vdev_top
->vdev_asize
);
7951 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
7952 vml
[c
]->vdev_top
->vdev_ashift
);
7954 /* transfer per-vdev ZAPs */
7955 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
7956 VERIFY0(nvlist_add_uint64(child
[c
],
7957 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
7959 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
7960 VERIFY0(nvlist_add_uint64(child
[c
],
7961 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
7962 vml
[c
]->vdev_parent
->vdev_top_zap
));
7966 kmem_free(vml
, children
* sizeof (vdev_t
*));
7967 kmem_free(glist
, children
* sizeof (uint64_t));
7968 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7971 /* stop writers from using the disks */
7972 for (c
= 0; c
< children
; c
++) {
7974 vml
[c
]->vdev_offline
= B_TRUE
;
7976 vdev_reopen(spa
->spa_root_vdev
);
7979 * Temporarily record the splitting vdevs in the spa config. This
7980 * will disappear once the config is regenerated.
7982 nvl
= fnvlist_alloc();
7983 fnvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
, glist
, children
);
7984 kmem_free(glist
, children
* sizeof (uint64_t));
7986 mutex_enter(&spa
->spa_props_lock
);
7987 fnvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
, nvl
);
7988 mutex_exit(&spa
->spa_props_lock
);
7989 spa
->spa_config_splitting
= nvl
;
7990 vdev_config_dirty(spa
->spa_root_vdev
);
7992 /* configure and create the new pool */
7993 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
);
7994 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
7995 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
);
7996 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
, spa_version(spa
));
7997 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, spa
->spa_config_txg
);
7998 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
7999 spa_generate_guid(NULL
));
8000 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
8001 (void) nvlist_lookup_string(props
,
8002 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
8004 /* add the new pool to the namespace */
8005 newspa
= spa_add(newname
, config
, altroot
);
8006 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
8007 newspa
->spa_config_txg
= spa
->spa_config_txg
;
8008 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
8010 /* release the spa config lock, retaining the namespace lock */
8011 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
8013 if (zio_injection_enabled
)
8014 zio_handle_panic_injection(spa
, FTAG
, 1);
8016 spa_activate(newspa
, spa_mode_global
);
8017 spa_async_suspend(newspa
);
8020 * Temporarily stop the initializing and TRIM activity. We set the
8021 * state to ACTIVE so that we know to resume initializing or TRIM
8022 * once the split has completed.
8024 list_t vd_initialize_list
;
8025 list_create(&vd_initialize_list
, sizeof (vdev_t
),
8026 offsetof(vdev_t
, vdev_initialize_node
));
8028 list_t vd_trim_list
;
8029 list_create(&vd_trim_list
, sizeof (vdev_t
),
8030 offsetof(vdev_t
, vdev_trim_node
));
8032 for (c
= 0; c
< children
; c
++) {
8033 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
8034 mutex_enter(&vml
[c
]->vdev_initialize_lock
);
8035 vdev_initialize_stop(vml
[c
],
8036 VDEV_INITIALIZE_ACTIVE
, &vd_initialize_list
);
8037 mutex_exit(&vml
[c
]->vdev_initialize_lock
);
8039 mutex_enter(&vml
[c
]->vdev_trim_lock
);
8040 vdev_trim_stop(vml
[c
], VDEV_TRIM_ACTIVE
, &vd_trim_list
);
8041 mutex_exit(&vml
[c
]->vdev_trim_lock
);
8045 vdev_initialize_stop_wait(spa
, &vd_initialize_list
);
8046 vdev_trim_stop_wait(spa
, &vd_trim_list
);
8048 list_destroy(&vd_initialize_list
);
8049 list_destroy(&vd_trim_list
);
8051 newspa
->spa_config_source
= SPA_CONFIG_SRC_SPLIT
;
8052 newspa
->spa_is_splitting
= B_TRUE
;
8054 /* create the new pool from the disks of the original pool */
8055 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
);
8059 /* if that worked, generate a real config for the new pool */
8060 if (newspa
->spa_root_vdev
!= NULL
) {
8061 newspa
->spa_config_splitting
= fnvlist_alloc();
8062 fnvlist_add_uint64(newspa
->spa_config_splitting
,
8063 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
));
8064 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
8069 if (props
!= NULL
) {
8070 spa_configfile_set(newspa
, props
, B_FALSE
);
8071 error
= spa_prop_set(newspa
, props
);
8076 /* flush everything */
8077 txg
= spa_vdev_config_enter(newspa
);
8078 vdev_config_dirty(newspa
->spa_root_vdev
);
8079 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
8081 if (zio_injection_enabled
)
8082 zio_handle_panic_injection(spa
, FTAG
, 2);
8084 spa_async_resume(newspa
);
8086 /* finally, update the original pool's config */
8087 txg
= spa_vdev_config_enter(spa
);
8088 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
8089 error
= dmu_tx_assign(tx
, TXG_WAIT
);
8092 for (c
= 0; c
< children
; c
++) {
8093 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
8094 vdev_t
*tvd
= vml
[c
]->vdev_top
;
8097 * Need to be sure the detachable VDEV is not
8098 * on any *other* txg's DTL list to prevent it
8099 * from being accessed after it's freed.
8101 for (int t
= 0; t
< TXG_SIZE
; t
++) {
8102 (void) txg_list_remove_this(
8103 &tvd
->vdev_dtl_list
, vml
[c
], t
);
8108 spa_history_log_internal(spa
, "detach", tx
,
8109 "vdev=%s", vml
[c
]->vdev_path
);
8114 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
8115 vdev_config_dirty(spa
->spa_root_vdev
);
8116 spa
->spa_config_splitting
= NULL
;
8120 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
8122 if (zio_injection_enabled
)
8123 zio_handle_panic_injection(spa
, FTAG
, 3);
8125 /* split is complete; log a history record */
8126 spa_history_log_internal(newspa
, "split", NULL
,
8127 "from pool %s", spa_name(spa
));
8129 newspa
->spa_is_splitting
= B_FALSE
;
8130 kmem_free(vml
, children
* sizeof (vdev_t
*));
8132 /* if we're not going to mount the filesystems in userland, export */
8134 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
8141 spa_deactivate(newspa
);
8144 txg
= spa_vdev_config_enter(spa
);
8146 /* re-online all offlined disks */
8147 for (c
= 0; c
< children
; c
++) {
8149 vml
[c
]->vdev_offline
= B_FALSE
;
8152 /* restart initializing or trimming disks as necessary */
8153 spa_async_request(spa
, SPA_ASYNC_INITIALIZE_RESTART
);
8154 spa_async_request(spa
, SPA_ASYNC_TRIM_RESTART
);
8155 spa_async_request(spa
, SPA_ASYNC_AUTOTRIM_RESTART
);
8157 vdev_reopen(spa
->spa_root_vdev
);
8159 nvlist_free(spa
->spa_config_splitting
);
8160 spa
->spa_config_splitting
= NULL
;
8161 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
8163 kmem_free(vml
, children
* sizeof (vdev_t
*));
8168 * Find any device that's done replacing, or a vdev marked 'unspare' that's
8169 * currently spared, so we can detach it.
8172 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
8174 vdev_t
*newvd
, *oldvd
;
8176 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
8177 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
8183 * Check for a completed replacement. We always consider the first
8184 * vdev in the list to be the oldest vdev, and the last one to be
8185 * the newest (see spa_vdev_attach() for how that works). In
8186 * the case where the newest vdev is faulted, we will not automatically
8187 * remove it after a resilver completes. This is OK as it will require
8188 * user intervention to determine which disk the admin wishes to keep.
8190 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
8191 ASSERT(vd
->vdev_children
> 1);
8193 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
8194 oldvd
= vd
->vdev_child
[0];
8196 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
8197 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
8198 !vdev_dtl_required(oldvd
))
8203 * Check for a completed resilver with the 'unspare' flag set.
8204 * Also potentially update faulted state.
8206 if (vd
->vdev_ops
== &vdev_spare_ops
) {
8207 vdev_t
*first
= vd
->vdev_child
[0];
8208 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
8210 if (last
->vdev_unspare
) {
8213 } else if (first
->vdev_unspare
) {
8220 if (oldvd
!= NULL
&&
8221 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
8222 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
8223 !vdev_dtl_required(oldvd
))
8226 vdev_propagate_state(vd
);
8229 * If there are more than two spares attached to a disk,
8230 * and those spares are not required, then we want to
8231 * attempt to free them up now so that they can be used
8232 * by other pools. Once we're back down to a single
8233 * disk+spare, we stop removing them.
8235 if (vd
->vdev_children
> 2) {
8236 newvd
= vd
->vdev_child
[1];
8238 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
8239 vdev_dtl_empty(last
, DTL_MISSING
) &&
8240 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
8241 !vdev_dtl_required(newvd
))
8250 spa_vdev_resilver_done(spa_t
*spa
)
8252 vdev_t
*vd
, *pvd
, *ppvd
;
8253 uint64_t guid
, sguid
, pguid
, ppguid
;
8255 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
8257 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
8258 pvd
= vd
->vdev_parent
;
8259 ppvd
= pvd
->vdev_parent
;
8260 guid
= vd
->vdev_guid
;
8261 pguid
= pvd
->vdev_guid
;
8262 ppguid
= ppvd
->vdev_guid
;
8265 * If we have just finished replacing a hot spared device, then
8266 * we need to detach the parent's first child (the original hot
8269 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
8270 ppvd
->vdev_children
== 2) {
8271 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
8272 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
8274 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
8276 spa_config_exit(spa
, SCL_ALL
, FTAG
);
8277 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
8279 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
8281 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
8284 spa_config_exit(spa
, SCL_ALL
, FTAG
);
8287 * If a detach was not performed above replace waiters will not have
8288 * been notified. In which case we must do so now.
8290 spa_notify_waiters(spa
);
8294 * Update the stored path or FRU for this vdev.
8297 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
8301 boolean_t sync
= B_FALSE
;
8303 ASSERT(spa_writeable(spa
));
8305 spa_vdev_state_enter(spa
, SCL_ALL
);
8307 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
8308 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
8310 if (!vd
->vdev_ops
->vdev_op_leaf
)
8311 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
8314 if (strcmp(value
, vd
->vdev_path
) != 0) {
8315 spa_strfree(vd
->vdev_path
);
8316 vd
->vdev_path
= spa_strdup(value
);
8320 if (vd
->vdev_fru
== NULL
) {
8321 vd
->vdev_fru
= spa_strdup(value
);
8323 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
8324 spa_strfree(vd
->vdev_fru
);
8325 vd
->vdev_fru
= spa_strdup(value
);
8330 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
8334 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
8336 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
8340 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
8342 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
8346 * ==========================================================================
8348 * ==========================================================================
8351 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
8353 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8355 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
8356 return (SET_ERROR(EBUSY
));
8358 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
8362 spa_scan_stop(spa_t
*spa
)
8364 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8365 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
8366 return (SET_ERROR(EBUSY
));
8368 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
8372 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
8374 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8376 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
8377 return (SET_ERROR(ENOTSUP
));
8379 if (func
== POOL_SCAN_RESILVER
&&
8380 !spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
))
8381 return (SET_ERROR(ENOTSUP
));
8384 * If a resilver was requested, but there is no DTL on a
8385 * writeable leaf device, we have nothing to do.
8387 if (func
== POOL_SCAN_RESILVER
&&
8388 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
8389 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
8393 if (func
== POOL_SCAN_ERRORSCRUB
&&
8394 !spa_feature_is_enabled(spa
, SPA_FEATURE_HEAD_ERRLOG
))
8395 return (SET_ERROR(ENOTSUP
));
8397 return (dsl_scan(spa
->spa_dsl_pool
, func
));
8401 * ==========================================================================
8402 * SPA async task processing
8403 * ==========================================================================
8407 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
8409 if (vd
->vdev_remove_wanted
) {
8410 vd
->vdev_remove_wanted
= B_FALSE
;
8411 vd
->vdev_delayed_close
= B_FALSE
;
8412 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
8415 * We want to clear the stats, but we don't want to do a full
8416 * vdev_clear() as that will cause us to throw away
8417 * degraded/faulted state as well as attempt to reopen the
8418 * device, all of which is a waste.
8420 vd
->vdev_stat
.vs_read_errors
= 0;
8421 vd
->vdev_stat
.vs_write_errors
= 0;
8422 vd
->vdev_stat
.vs_checksum_errors
= 0;
8424 vdev_state_dirty(vd
->vdev_top
);
8426 /* Tell userspace that the vdev is gone. */
8427 zfs_post_remove(spa
, vd
);
8430 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8431 spa_async_remove(spa
, vd
->vdev_child
[c
]);
8435 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
8437 if (vd
->vdev_probe_wanted
) {
8438 vd
->vdev_probe_wanted
= B_FALSE
;
8439 vdev_reopen(vd
); /* vdev_open() does the actual probe */
8442 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8443 spa_async_probe(spa
, vd
->vdev_child
[c
]);
8447 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
8449 if (!spa
->spa_autoexpand
)
8452 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
8453 vdev_t
*cvd
= vd
->vdev_child
[c
];
8454 spa_async_autoexpand(spa
, cvd
);
8457 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
8460 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
8463 static __attribute__((noreturn
)) void
8464 spa_async_thread(void *arg
)
8466 spa_t
*spa
= (spa_t
*)arg
;
8467 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8470 ASSERT(spa
->spa_sync_on
);
8472 mutex_enter(&spa
->spa_async_lock
);
8473 tasks
= spa
->spa_async_tasks
;
8474 spa
->spa_async_tasks
= 0;
8475 mutex_exit(&spa
->spa_async_lock
);
8478 * See if the config needs to be updated.
8480 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
8481 uint64_t old_space
, new_space
;
8483 mutex_enter(&spa_namespace_lock
);
8484 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
8485 old_space
+= metaslab_class_get_space(spa_special_class(spa
));
8486 old_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8487 old_space
+= metaslab_class_get_space(
8488 spa_embedded_log_class(spa
));
8490 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
8492 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
8493 new_space
+= metaslab_class_get_space(spa_special_class(spa
));
8494 new_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8495 new_space
+= metaslab_class_get_space(
8496 spa_embedded_log_class(spa
));
8497 mutex_exit(&spa_namespace_lock
);
8500 * If the pool grew as a result of the config update,
8501 * then log an internal history event.
8503 if (new_space
!= old_space
) {
8504 spa_history_log_internal(spa
, "vdev online", NULL
,
8505 "pool '%s' size: %llu(+%llu)",
8506 spa_name(spa
), (u_longlong_t
)new_space
,
8507 (u_longlong_t
)(new_space
- old_space
));
8512 * See if any devices need to be marked REMOVED.
8514 if (tasks
& SPA_ASYNC_REMOVE
) {
8515 spa_vdev_state_enter(spa
, SCL_NONE
);
8516 spa_async_remove(spa
, spa
->spa_root_vdev
);
8517 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
8518 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
8519 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
8520 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
8521 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8524 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
8525 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8526 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
8527 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8531 * See if any devices need to be probed.
8533 if (tasks
& SPA_ASYNC_PROBE
) {
8534 spa_vdev_state_enter(spa
, SCL_NONE
);
8535 spa_async_probe(spa
, spa
->spa_root_vdev
);
8536 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8540 * If any devices are done replacing, detach them.
8542 if (tasks
& SPA_ASYNC_RESILVER_DONE
||
8543 tasks
& SPA_ASYNC_REBUILD_DONE
||
8544 tasks
& SPA_ASYNC_DETACH_SPARE
) {
8545 spa_vdev_resilver_done(spa
);
8549 * Kick off a resilver.
8551 if (tasks
& SPA_ASYNC_RESILVER
&&
8552 !vdev_rebuild_active(spa
->spa_root_vdev
) &&
8553 (!dsl_scan_resilvering(dp
) ||
8554 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_RESILVER_DEFER
)))
8555 dsl_scan_restart_resilver(dp
, 0);
8557 if (tasks
& SPA_ASYNC_INITIALIZE_RESTART
) {
8558 mutex_enter(&spa_namespace_lock
);
8559 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8560 vdev_initialize_restart(spa
->spa_root_vdev
);
8561 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8562 mutex_exit(&spa_namespace_lock
);
8565 if (tasks
& SPA_ASYNC_TRIM_RESTART
) {
8566 mutex_enter(&spa_namespace_lock
);
8567 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8568 vdev_trim_restart(spa
->spa_root_vdev
);
8569 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8570 mutex_exit(&spa_namespace_lock
);
8573 if (tasks
& SPA_ASYNC_AUTOTRIM_RESTART
) {
8574 mutex_enter(&spa_namespace_lock
);
8575 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8576 vdev_autotrim_restart(spa
);
8577 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8578 mutex_exit(&spa_namespace_lock
);
8582 * Kick off L2 cache whole device TRIM.
8584 if (tasks
& SPA_ASYNC_L2CACHE_TRIM
) {
8585 mutex_enter(&spa_namespace_lock
);
8586 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8587 vdev_trim_l2arc(spa
);
8588 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8589 mutex_exit(&spa_namespace_lock
);
8593 * Kick off L2 cache rebuilding.
8595 if (tasks
& SPA_ASYNC_L2CACHE_REBUILD
) {
8596 mutex_enter(&spa_namespace_lock
);
8597 spa_config_enter(spa
, SCL_L2ARC
, FTAG
, RW_READER
);
8598 l2arc_spa_rebuild_start(spa
);
8599 spa_config_exit(spa
, SCL_L2ARC
, FTAG
);
8600 mutex_exit(&spa_namespace_lock
);
8604 * Let the world know that we're done.
8606 mutex_enter(&spa
->spa_async_lock
);
8607 spa
->spa_async_thread
= NULL
;
8608 cv_broadcast(&spa
->spa_async_cv
);
8609 mutex_exit(&spa
->spa_async_lock
);
8614 spa_async_suspend(spa_t
*spa
)
8616 mutex_enter(&spa
->spa_async_lock
);
8617 spa
->spa_async_suspended
++;
8618 while (spa
->spa_async_thread
!= NULL
)
8619 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
8620 mutex_exit(&spa
->spa_async_lock
);
8622 spa_vdev_remove_suspend(spa
);
8624 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8625 if (condense_thread
!= NULL
)
8626 zthr_cancel(condense_thread
);
8628 zthr_t
*raidz_expand_thread
= spa
->spa_raidz_expand_zthr
;
8629 if (raidz_expand_thread
!= NULL
)
8630 zthr_cancel(raidz_expand_thread
);
8632 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8633 if (discard_thread
!= NULL
)
8634 zthr_cancel(discard_thread
);
8636 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8637 if (ll_delete_thread
!= NULL
)
8638 zthr_cancel(ll_delete_thread
);
8640 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8641 if (ll_condense_thread
!= NULL
)
8642 zthr_cancel(ll_condense_thread
);
8646 spa_async_resume(spa_t
*spa
)
8648 mutex_enter(&spa
->spa_async_lock
);
8649 ASSERT(spa
->spa_async_suspended
!= 0);
8650 spa
->spa_async_suspended
--;
8651 mutex_exit(&spa
->spa_async_lock
);
8652 spa_restart_removal(spa
);
8654 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8655 if (condense_thread
!= NULL
)
8656 zthr_resume(condense_thread
);
8658 zthr_t
*raidz_expand_thread
= spa
->spa_raidz_expand_zthr
;
8659 if (raidz_expand_thread
!= NULL
)
8660 zthr_resume(raidz_expand_thread
);
8662 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8663 if (discard_thread
!= NULL
)
8664 zthr_resume(discard_thread
);
8666 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8667 if (ll_delete_thread
!= NULL
)
8668 zthr_resume(ll_delete_thread
);
8670 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8671 if (ll_condense_thread
!= NULL
)
8672 zthr_resume(ll_condense_thread
);
8676 spa_async_tasks_pending(spa_t
*spa
)
8678 uint_t non_config_tasks
;
8680 boolean_t config_task_suspended
;
8682 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
8683 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
8684 if (spa
->spa_ccw_fail_time
== 0) {
8685 config_task_suspended
= B_FALSE
;
8687 config_task_suspended
=
8688 (gethrtime() - spa
->spa_ccw_fail_time
) <
8689 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
8692 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
8696 spa_async_dispatch(spa_t
*spa
)
8698 mutex_enter(&spa
->spa_async_lock
);
8699 if (spa_async_tasks_pending(spa
) &&
8700 !spa
->spa_async_suspended
&&
8701 spa
->spa_async_thread
== NULL
)
8702 spa
->spa_async_thread
= thread_create(NULL
, 0,
8703 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
8704 mutex_exit(&spa
->spa_async_lock
);
8708 spa_async_request(spa_t
*spa
, int task
)
8710 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
8711 mutex_enter(&spa
->spa_async_lock
);
8712 spa
->spa_async_tasks
|= task
;
8713 mutex_exit(&spa
->spa_async_lock
);
8717 spa_async_tasks(spa_t
*spa
)
8719 return (spa
->spa_async_tasks
);
8723 * ==========================================================================
8724 * SPA syncing routines
8725 * ==========================================================================
8730 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8734 bpobj_enqueue(bpo
, bp
, bp_freed
, tx
);
8739 bpobj_enqueue_alloc_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8741 return (bpobj_enqueue_cb(arg
, bp
, B_FALSE
, tx
));
8745 bpobj_enqueue_free_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8747 return (bpobj_enqueue_cb(arg
, bp
, B_TRUE
, tx
));
8751 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
8755 zio_nowait(zio_free_sync(pio
, pio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
8761 bpobj_spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
8765 return (spa_free_sync_cb(arg
, bp
, tx
));
8769 * Note: this simple function is not inlined to make it easier to dtrace the
8770 * amount of time spent syncing frees.
8773 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
8775 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8776 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
8777 VERIFY(zio_wait(zio
) == 0);
8781 * Note: this simple function is not inlined to make it easier to dtrace the
8782 * amount of time spent syncing deferred frees.
8785 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
8787 if (spa_sync_pass(spa
) != 1)
8792 * If the log space map feature is active, we stop deferring
8793 * frees to the next TXG and therefore running this function
8794 * would be considered a no-op as spa_deferred_bpobj should
8795 * not have any entries.
8797 * That said we run this function anyway (instead of returning
8798 * immediately) for the edge-case scenario where we just
8799 * activated the log space map feature in this TXG but we have
8800 * deferred frees from the previous TXG.
8802 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
8803 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
8804 bpobj_spa_free_sync_cb
, zio
, tx
), ==, 0);
8805 VERIFY0(zio_wait(zio
));
8809 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
8811 char *packed
= NULL
;
8816 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
8819 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
8820 * information. This avoids the dmu_buf_will_dirty() path and
8821 * saves us a pre-read to get data we don't actually care about.
8823 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
8824 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
8826 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
8828 memset(packed
+ nvsize
, 0, bufsize
- nvsize
);
8830 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
8832 vmem_free(packed
, bufsize
);
8834 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
8835 dmu_buf_will_dirty(db
, tx
);
8836 *(uint64_t *)db
->db_data
= nvsize
;
8837 dmu_buf_rele(db
, FTAG
);
8841 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
8842 const char *config
, const char *entry
)
8852 * Update the MOS nvlist describing the list of available devices.
8853 * spa_validate_aux() will have already made sure this nvlist is
8854 * valid and the vdevs are labeled appropriately.
8856 if (sav
->sav_object
== 0) {
8857 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
8858 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
8859 sizeof (uint64_t), tx
);
8860 VERIFY(zap_update(spa
->spa_meta_objset
,
8861 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
8862 &sav
->sav_object
, tx
) == 0);
8865 nvroot
= fnvlist_alloc();
8866 if (sav
->sav_count
== 0) {
8867 fnvlist_add_nvlist_array(nvroot
, config
,
8868 (const nvlist_t
* const *)NULL
, 0);
8870 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
8871 for (i
= 0; i
< sav
->sav_count
; i
++)
8872 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
8873 B_FALSE
, VDEV_CONFIG_L2CACHE
);
8874 fnvlist_add_nvlist_array(nvroot
, config
,
8875 (const nvlist_t
* const *)list
, sav
->sav_count
);
8876 for (i
= 0; i
< sav
->sav_count
; i
++)
8877 nvlist_free(list
[i
]);
8878 kmem_free(list
, sav
->sav_count
* sizeof (void *));
8881 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
8882 nvlist_free(nvroot
);
8884 sav
->sav_sync
= B_FALSE
;
8888 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
8889 * The all-vdev ZAP must be empty.
8892 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
8894 spa_t
*spa
= vd
->vdev_spa
;
8896 if (vd
->vdev_root_zap
!= 0 &&
8897 spa_feature_is_active(spa
, SPA_FEATURE_AVZ_V2
)) {
8898 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8899 vd
->vdev_root_zap
, tx
));
8901 if (vd
->vdev_top_zap
!= 0) {
8902 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8903 vd
->vdev_top_zap
, tx
));
8905 if (vd
->vdev_leaf_zap
!= 0) {
8906 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
8907 vd
->vdev_leaf_zap
, tx
));
8909 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
8910 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
8915 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
8920 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
8921 * its config may not be dirty but we still need to build per-vdev ZAPs.
8922 * Similarly, if the pool is being assembled (e.g. after a split), we
8923 * need to rebuild the AVZ although the config may not be dirty.
8925 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
8926 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
8929 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8931 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
8932 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
8933 spa
->spa_all_vdev_zaps
!= 0);
8935 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
8936 /* Make and build the new AVZ */
8937 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
8938 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
8939 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
8941 /* Diff old AVZ with new one */
8945 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
8946 spa
->spa_all_vdev_zaps
);
8947 zap_cursor_retrieve(&zc
, &za
) == 0;
8948 zap_cursor_advance(&zc
)) {
8949 uint64_t vdzap
= za
.za_first_integer
;
8950 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
8953 * ZAP is listed in old AVZ but not in new one;
8956 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
8961 zap_cursor_fini(&zc
);
8963 /* Destroy the old AVZ */
8964 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
8965 spa
->spa_all_vdev_zaps
, tx
));
8967 /* Replace the old AVZ in the dir obj with the new one */
8968 VERIFY0(zap_update(spa
->spa_meta_objset
,
8969 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
8970 sizeof (new_avz
), 1, &new_avz
, tx
));
8972 spa
->spa_all_vdev_zaps
= new_avz
;
8973 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
8977 /* Walk through the AVZ and destroy all listed ZAPs */
8978 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
8979 spa
->spa_all_vdev_zaps
);
8980 zap_cursor_retrieve(&zc
, &za
) == 0;
8981 zap_cursor_advance(&zc
)) {
8982 uint64_t zap
= za
.za_first_integer
;
8983 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
8986 zap_cursor_fini(&zc
);
8988 /* Destroy and unlink the AVZ itself */
8989 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
8990 spa
->spa_all_vdev_zaps
, tx
));
8991 VERIFY0(zap_remove(spa
->spa_meta_objset
,
8992 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
8993 spa
->spa_all_vdev_zaps
= 0;
8996 if (spa
->spa_all_vdev_zaps
== 0) {
8997 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
8998 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
8999 DMU_POOL_VDEV_ZAP_MAP
, tx
);
9001 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
9003 /* Create ZAPs for vdevs that don't have them. */
9004 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
9006 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
9007 dmu_tx_get_txg(tx
), B_FALSE
);
9010 * If we're upgrading the spa version then make sure that
9011 * the config object gets updated with the correct version.
9013 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
9014 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
9015 spa
->spa_uberblock
.ub_version
);
9017 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9019 nvlist_free(spa
->spa_config_syncing
);
9020 spa
->spa_config_syncing
= config
;
9022 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
9026 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
9028 uint64_t *versionp
= arg
;
9029 uint64_t version
= *versionp
;
9030 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
9033 * Setting the version is special cased when first creating the pool.
9035 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
9037 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
9038 ASSERT(version
>= spa_version(spa
));
9040 spa
->spa_uberblock
.ub_version
= version
;
9041 vdev_config_dirty(spa
->spa_root_vdev
);
9042 spa_history_log_internal(spa
, "set", tx
, "version=%lld",
9043 (longlong_t
)version
);
9047 * Set zpool properties.
9050 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
9052 nvlist_t
*nvp
= arg
;
9053 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
9054 objset_t
*mos
= spa
->spa_meta_objset
;
9055 nvpair_t
*elem
= NULL
;
9057 mutex_enter(&spa
->spa_props_lock
);
9059 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
9061 const char *strval
, *fname
;
9063 const char *propname
;
9064 const char *elemname
= nvpair_name(elem
);
9065 zprop_type_t proptype
;
9068 switch (prop
= zpool_name_to_prop(elemname
)) {
9069 case ZPOOL_PROP_VERSION
:
9070 intval
= fnvpair_value_uint64(elem
);
9072 * The version is synced separately before other
9073 * properties and should be correct by now.
9075 ASSERT3U(spa_version(spa
), >=, intval
);
9078 case ZPOOL_PROP_ALTROOT
:
9080 * 'altroot' is a non-persistent property. It should
9081 * have been set temporarily at creation or import time.
9083 ASSERT(spa
->spa_root
!= NULL
);
9086 case ZPOOL_PROP_READONLY
:
9087 case ZPOOL_PROP_CACHEFILE
:
9089 * 'readonly' and 'cachefile' are also non-persistent
9093 case ZPOOL_PROP_COMMENT
:
9094 strval
= fnvpair_value_string(elem
);
9095 if (spa
->spa_comment
!= NULL
)
9096 spa_strfree(spa
->spa_comment
);
9097 spa
->spa_comment
= spa_strdup(strval
);
9099 * We need to dirty the configuration on all the vdevs
9100 * so that their labels get updated. We also need to
9101 * update the cache file to keep it in sync with the
9102 * MOS version. It's unnecessary to do this for pool
9103 * creation since the vdev's configuration has already
9106 if (tx
->tx_txg
!= TXG_INITIAL
) {
9107 vdev_config_dirty(spa
->spa_root_vdev
);
9108 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
9110 spa_history_log_internal(spa
, "set", tx
,
9111 "%s=%s", elemname
, strval
);
9113 case ZPOOL_PROP_COMPATIBILITY
:
9114 strval
= fnvpair_value_string(elem
);
9115 if (spa
->spa_compatibility
!= NULL
)
9116 spa_strfree(spa
->spa_compatibility
);
9117 spa
->spa_compatibility
= spa_strdup(strval
);
9119 * Dirty the configuration on vdevs as above.
9121 if (tx
->tx_txg
!= TXG_INITIAL
) {
9122 vdev_config_dirty(spa
->spa_root_vdev
);
9123 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
9126 spa_history_log_internal(spa
, "set", tx
,
9127 "%s=%s", nvpair_name(elem
), strval
);
9130 case ZPOOL_PROP_INVAL
:
9131 if (zpool_prop_feature(elemname
)) {
9132 fname
= strchr(elemname
, '@') + 1;
9133 VERIFY0(zfeature_lookup_name(fname
, &fid
));
9135 spa_feature_enable(spa
, fid
, tx
);
9136 spa_history_log_internal(spa
, "set", tx
,
9137 "%s=enabled", elemname
);
9139 } else if (!zfs_prop_user(elemname
)) {
9140 ASSERT(zpool_prop_feature(elemname
));
9146 * Set pool property values in the poolprops mos object.
9148 if (spa
->spa_pool_props_object
== 0) {
9149 spa
->spa_pool_props_object
=
9150 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
9151 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
9155 /* normalize the property name */
9156 if (prop
== ZPOOL_PROP_INVAL
) {
9157 propname
= elemname
;
9158 proptype
= PROP_TYPE_STRING
;
9160 propname
= zpool_prop_to_name(prop
);
9161 proptype
= zpool_prop_get_type(prop
);
9164 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
9165 ASSERT(proptype
== PROP_TYPE_STRING
);
9166 strval
= fnvpair_value_string(elem
);
9167 VERIFY0(zap_update(mos
,
9168 spa
->spa_pool_props_object
, propname
,
9169 1, strlen(strval
) + 1, strval
, tx
));
9170 spa_history_log_internal(spa
, "set", tx
,
9171 "%s=%s", elemname
, strval
);
9172 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
9173 intval
= fnvpair_value_uint64(elem
);
9175 if (proptype
== PROP_TYPE_INDEX
) {
9177 VERIFY0(zpool_prop_index_to_string(
9178 prop
, intval
, &unused
));
9180 VERIFY0(zap_update(mos
,
9181 spa
->spa_pool_props_object
, propname
,
9182 8, 1, &intval
, tx
));
9183 spa_history_log_internal(spa
, "set", tx
,
9184 "%s=%lld", elemname
,
9185 (longlong_t
)intval
);
9188 case ZPOOL_PROP_DELEGATION
:
9189 spa
->spa_delegation
= intval
;
9191 case ZPOOL_PROP_BOOTFS
:
9192 spa
->spa_bootfs
= intval
;
9194 case ZPOOL_PROP_FAILUREMODE
:
9195 spa
->spa_failmode
= intval
;
9197 case ZPOOL_PROP_AUTOTRIM
:
9198 spa
->spa_autotrim
= intval
;
9199 spa_async_request(spa
,
9200 SPA_ASYNC_AUTOTRIM_RESTART
);
9202 case ZPOOL_PROP_AUTOEXPAND
:
9203 spa
->spa_autoexpand
= intval
;
9204 if (tx
->tx_txg
!= TXG_INITIAL
)
9205 spa_async_request(spa
,
9206 SPA_ASYNC_AUTOEXPAND
);
9208 case ZPOOL_PROP_MULTIHOST
:
9209 spa
->spa_multihost
= intval
;
9215 ASSERT(0); /* not allowed */
9221 mutex_exit(&spa
->spa_props_lock
);
9225 * Perform one-time upgrade on-disk changes. spa_version() does not
9226 * reflect the new version this txg, so there must be no changes this
9227 * txg to anything that the upgrade code depends on after it executes.
9228 * Therefore this must be called after dsl_pool_sync() does the sync
9232 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
9234 if (spa_sync_pass(spa
) != 1)
9237 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9238 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
9240 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
9241 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
9242 dsl_pool_create_origin(dp
, tx
);
9244 /* Keeping the origin open increases spa_minref */
9245 spa
->spa_minref
+= 3;
9248 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
9249 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
9250 dsl_pool_upgrade_clones(dp
, tx
);
9253 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
9254 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
9255 dsl_pool_upgrade_dir_clones(dp
, tx
);
9257 /* Keeping the freedir open increases spa_minref */
9258 spa
->spa_minref
+= 3;
9261 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
9262 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
9263 spa_feature_create_zap_objects(spa
, tx
);
9267 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
9268 * when possibility to use lz4 compression for metadata was added
9269 * Old pools that have this feature enabled must be upgraded to have
9270 * this feature active
9272 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
9273 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
9274 SPA_FEATURE_LZ4_COMPRESS
);
9275 boolean_t lz4_ac
= spa_feature_is_active(spa
,
9276 SPA_FEATURE_LZ4_COMPRESS
);
9278 if (lz4_en
&& !lz4_ac
)
9279 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
9283 * If we haven't written the salt, do so now. Note that the
9284 * feature may not be activated yet, but that's fine since
9285 * the presence of this ZAP entry is backwards compatible.
9287 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
9288 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
9289 VERIFY0(zap_add(spa
->spa_meta_objset
,
9290 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
9291 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
9292 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
9295 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
9299 vdev_indirect_state_sync_verify(vdev_t
*vd
)
9301 vdev_indirect_mapping_t
*vim __maybe_unused
= vd
->vdev_indirect_mapping
;
9302 vdev_indirect_births_t
*vib __maybe_unused
= vd
->vdev_indirect_births
;
9304 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
9305 ASSERT(vim
!= NULL
);
9306 ASSERT(vib
!= NULL
);
9309 uint64_t obsolete_sm_object
= 0;
9310 ASSERT0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
9311 if (obsolete_sm_object
!= 0) {
9312 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
9313 ASSERT(vd
->vdev_removing
||
9314 vd
->vdev_ops
== &vdev_indirect_ops
);
9315 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
9316 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
9317 ASSERT3U(obsolete_sm_object
, ==,
9318 space_map_object(vd
->vdev_obsolete_sm
));
9319 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
9320 space_map_allocated(vd
->vdev_obsolete_sm
));
9322 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
9325 * Since frees / remaps to an indirect vdev can only
9326 * happen in syncing context, the obsolete segments
9327 * tree must be empty when we start syncing.
9329 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
9333 * Set the top-level vdev's max queue depth. Evaluate each top-level's
9334 * async write queue depth in case it changed. The max queue depth will
9335 * not change in the middle of syncing out this txg.
9338 spa_sync_adjust_vdev_max_queue_depth(spa_t
*spa
)
9340 ASSERT(spa_writeable(spa
));
9342 vdev_t
*rvd
= spa
->spa_root_vdev
;
9343 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
9344 zfs_vdev_queue_depth_pct
/ 100;
9345 metaslab_class_t
*normal
= spa_normal_class(spa
);
9346 metaslab_class_t
*special
= spa_special_class(spa
);
9347 metaslab_class_t
*dedup
= spa_dedup_class(spa
);
9349 uint64_t slots_per_allocator
= 0;
9350 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
9351 vdev_t
*tvd
= rvd
->vdev_child
[c
];
9353 metaslab_group_t
*mg
= tvd
->vdev_mg
;
9354 if (mg
== NULL
|| !metaslab_group_initialized(mg
))
9357 metaslab_class_t
*mc
= mg
->mg_class
;
9358 if (mc
!= normal
&& mc
!= special
&& mc
!= dedup
)
9362 * It is safe to do a lock-free check here because only async
9363 * allocations look at mg_max_alloc_queue_depth, and async
9364 * allocations all happen from spa_sync().
9366 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
9367 ASSERT0(zfs_refcount_count(
9368 &(mg
->mg_allocator
[i
].mga_alloc_queue_depth
)));
9370 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
9372 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
9373 mg
->mg_allocator
[i
].mga_cur_max_alloc_queue_depth
=
9374 zfs_vdev_def_queue_depth
;
9376 slots_per_allocator
+= zfs_vdev_def_queue_depth
;
9379 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9380 ASSERT0(zfs_refcount_count(&normal
->mc_allocator
[i
].
9382 ASSERT0(zfs_refcount_count(&special
->mc_allocator
[i
].
9384 ASSERT0(zfs_refcount_count(&dedup
->mc_allocator
[i
].
9386 normal
->mc_allocator
[i
].mca_alloc_max_slots
=
9387 slots_per_allocator
;
9388 special
->mc_allocator
[i
].mca_alloc_max_slots
=
9389 slots_per_allocator
;
9390 dedup
->mc_allocator
[i
].mca_alloc_max_slots
=
9391 slots_per_allocator
;
9393 normal
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9394 special
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9395 dedup
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9399 spa_sync_condense_indirect(spa_t
*spa
, dmu_tx_t
*tx
)
9401 ASSERT(spa_writeable(spa
));
9403 vdev_t
*rvd
= spa
->spa_root_vdev
;
9404 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
9405 vdev_t
*vd
= rvd
->vdev_child
[c
];
9406 vdev_indirect_state_sync_verify(vd
);
9408 if (vdev_indirect_should_condense(vd
)) {
9409 spa_condense_indirect_start_sync(vd
, tx
);
9416 spa_sync_iterate_to_convergence(spa_t
*spa
, dmu_tx_t
*tx
)
9418 objset_t
*mos
= spa
->spa_meta_objset
;
9419 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9420 uint64_t txg
= tx
->tx_txg
;
9421 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
9424 int pass
= ++spa
->spa_sync_pass
;
9426 spa_sync_config_object(spa
, tx
);
9427 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
9428 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
9429 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
9430 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
9431 spa_errlog_sync(spa
, txg
);
9432 dsl_pool_sync(dp
, txg
);
9434 if (pass
< zfs_sync_pass_deferred_free
||
9435 spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
)) {
9437 * If the log space map feature is active we don't
9438 * care about deferred frees and the deferred bpobj
9439 * as the log space map should effectively have the
9440 * same results (i.e. appending only to one object).
9442 spa_sync_frees(spa
, free_bpl
, tx
);
9445 * We can not defer frees in pass 1, because
9446 * we sync the deferred frees later in pass 1.
9448 ASSERT3U(pass
, >, 1);
9449 bplist_iterate(free_bpl
, bpobj_enqueue_alloc_cb
,
9450 &spa
->spa_deferred_bpobj
, tx
);
9455 dsl_scan_sync(dp
, tx
);
9456 dsl_errorscrub_sync(dp
, tx
);
9458 spa_sync_upgrades(spa
, tx
);
9460 spa_flush_metaslabs(spa
, tx
);
9463 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
9469 * dsl_pool_sync() -> dp_sync_tasks may have dirtied
9470 * the config. If that happens, this txg should not
9471 * be a no-op. So we must sync the config to the MOS
9472 * before checking for no-op.
9474 * Note that when the config is dirty, it will
9475 * be written to the MOS (i.e. the MOS will be
9476 * dirtied) every time we call spa_sync_config_object()
9477 * in this txg. Therefore we can't call this after
9478 * dsl_pool_sync() every pass, because it would
9479 * prevent us from converging, since we'd dirty
9480 * the MOS every pass.
9482 * Sync tasks can only be processed in pass 1, so
9483 * there's no need to do this in later passes.
9485 spa_sync_config_object(spa
, tx
);
9489 * Note: We need to check if the MOS is dirty because we could
9490 * have marked the MOS dirty without updating the uberblock
9491 * (e.g. if we have sync tasks but no dirty user data). We need
9492 * to check the uberblock's rootbp because it is updated if we
9493 * have synced out dirty data (though in this case the MOS will
9494 * most likely also be dirty due to second order effects, we
9495 * don't want to rely on that here).
9498 spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
9499 !dmu_objset_is_dirty(mos
, txg
)) {
9501 * Nothing changed on the first pass, therefore this
9502 * TXG is a no-op. Avoid syncing deferred frees, so
9503 * that we can keep this TXG as a no-op.
9505 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9506 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9507 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
9508 ASSERT(txg_list_empty(&dp
->dp_early_sync_tasks
, txg
));
9512 spa_sync_deferred_frees(spa
, tx
);
9513 } while (dmu_objset_is_dirty(mos
, txg
));
9517 * Rewrite the vdev configuration (which includes the uberblock) to
9518 * commit the transaction group.
9520 * If there are no dirty vdevs, we sync the uberblock to a few random
9521 * top-level vdevs that are known to be visible in the config cache
9522 * (see spa_vdev_add() for a complete description). If there *are* dirty
9523 * vdevs, sync the uberblock to all vdevs.
9526 spa_sync_rewrite_vdev_config(spa_t
*spa
, dmu_tx_t
*tx
)
9528 vdev_t
*rvd
= spa
->spa_root_vdev
;
9529 uint64_t txg
= tx
->tx_txg
;
9535 * We hold SCL_STATE to prevent vdev open/close/etc.
9536 * while we're attempting to write the vdev labels.
9538 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9540 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
9541 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
9543 int children
= rvd
->vdev_children
;
9544 int c0
= random_in_range(children
);
9546 for (int c
= 0; c
< children
; c
++) {
9548 rvd
->vdev_child
[(c0
+ c
) % children
];
9550 /* Stop when revisiting the first vdev */
9551 if (c
> 0 && svd
[0] == vd
)
9554 if (vd
->vdev_ms_array
== 0 ||
9556 !vdev_is_concrete(vd
))
9559 svd
[svdcount
++] = vd
;
9560 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
9563 error
= vdev_config_sync(svd
, svdcount
, txg
);
9565 error
= vdev_config_sync(rvd
->vdev_child
,
9566 rvd
->vdev_children
, txg
);
9570 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
9572 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9576 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
9577 zio_resume_wait(spa
);
9582 * Sync the specified transaction group. New blocks may be dirtied as
9583 * part of the process, so we iterate until it converges.
9586 spa_sync(spa_t
*spa
, uint64_t txg
)
9590 VERIFY(spa_writeable(spa
));
9593 * Wait for i/os issued in open context that need to complete
9594 * before this txg syncs.
9596 (void) zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]);
9597 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
,
9601 * Now that there can be no more cloning in this transaction group,
9602 * but we are still before issuing frees, we can process pending BRT
9605 brt_pending_apply(spa
, txg
);
9608 * Lock out configuration changes.
9610 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
9612 spa
->spa_syncing_txg
= txg
;
9613 spa
->spa_sync_pass
= 0;
9615 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9616 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
9617 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
9618 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
9622 * If there are any pending vdev state changes, convert them
9623 * into config changes that go out with this transaction group.
9625 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9626 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
9627 /* Avoid holding the write lock unless actually necessary */
9628 if (vd
->vdev_aux
== NULL
) {
9629 vdev_state_clean(vd
);
9630 vdev_config_dirty(vd
);
9634 * We need the write lock here because, for aux vdevs,
9635 * calling vdev_config_dirty() modifies sav_config.
9636 * This is ugly and will become unnecessary when we
9637 * eliminate the aux vdev wart by integrating all vdevs
9638 * into the root vdev tree.
9640 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9641 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
9642 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
9643 vdev_state_clean(vd
);
9644 vdev_config_dirty(vd
);
9646 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9647 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9649 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9651 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9652 dmu_tx_t
*tx
= dmu_tx_create_assigned(dp
, txg
);
9654 spa
->spa_sync_starttime
= gethrtime();
9655 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9656 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
9657 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
9658 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
9661 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9662 * set spa_deflate if we have no raid-z vdevs.
9664 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
9665 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
9666 vdev_t
*rvd
= spa
->spa_root_vdev
;
9669 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
9670 vd
= rvd
->vdev_child
[i
];
9671 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
9674 if (i
== rvd
->vdev_children
) {
9675 spa
->spa_deflate
= TRUE
;
9676 VERIFY0(zap_add(spa
->spa_meta_objset
,
9677 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
9678 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
9682 spa_sync_adjust_vdev_max_queue_depth(spa
);
9684 spa_sync_condense_indirect(spa
, tx
);
9686 spa_sync_iterate_to_convergence(spa
, tx
);
9689 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
9691 * Make sure that the number of ZAPs for all the vdevs matches
9692 * the number of ZAPs in the per-vdev ZAP list. This only gets
9693 * called if the config is dirty; otherwise there may be
9694 * outstanding AVZ operations that weren't completed in
9695 * spa_sync_config_object.
9697 uint64_t all_vdev_zap_entry_count
;
9698 ASSERT0(zap_count(spa
->spa_meta_objset
,
9699 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
9700 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
9701 all_vdev_zap_entry_count
);
9705 if (spa
->spa_vdev_removal
!= NULL
) {
9706 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
9709 spa_sync_rewrite_vdev_config(spa
, tx
);
9712 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9713 spa
->spa_deadman_tqid
= 0;
9716 * Clear the dirty config list.
9718 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
9719 vdev_config_clean(vd
);
9722 * Now that the new config has synced transactionally,
9723 * let it become visible to the config cache.
9725 if (spa
->spa_config_syncing
!= NULL
) {
9726 spa_config_set(spa
, spa
->spa_config_syncing
);
9727 spa
->spa_config_txg
= txg
;
9728 spa
->spa_config_syncing
= NULL
;
9731 dsl_pool_sync_done(dp
, txg
);
9733 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9734 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
9735 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
9736 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
9740 * Update usable space statistics.
9742 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
9744 vdev_sync_done(vd
, txg
);
9746 metaslab_class_evict_old(spa
->spa_normal_class
, txg
);
9747 metaslab_class_evict_old(spa
->spa_log_class
, txg
);
9749 spa_sync_close_syncing_log_sm(spa
);
9751 spa_update_dspace(spa
);
9753 if (spa_get_autotrim(spa
) == SPA_AUTOTRIM_ON
)
9754 vdev_autotrim_kick(spa
);
9757 * It had better be the case that we didn't dirty anything
9758 * since vdev_config_sync().
9760 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9761 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9762 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
9764 while (zfs_pause_spa_sync
)
9767 spa
->spa_sync_pass
= 0;
9770 * Update the last synced uberblock here. We want to do this at
9771 * the end of spa_sync() so that consumers of spa_last_synced_txg()
9772 * will be guaranteed that all the processing associated with
9773 * that txg has been completed.
9775 spa
->spa_ubsync
= spa
->spa_uberblock
;
9776 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
9778 spa_handle_ignored_writes(spa
);
9781 * If any async tasks have been requested, kick them off.
9783 spa_async_dispatch(spa
);
9787 * Sync all pools. We don't want to hold the namespace lock across these
9788 * operations, so we take a reference on the spa_t and drop the lock during the
9792 spa_sync_allpools(void)
9795 mutex_enter(&spa_namespace_lock
);
9796 while ((spa
= spa_next(spa
)) != NULL
) {
9797 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
9798 !spa_writeable(spa
) || spa_suspended(spa
))
9800 spa_open_ref(spa
, FTAG
);
9801 mutex_exit(&spa_namespace_lock
);
9802 txg_wait_synced(spa_get_dsl(spa
), 0);
9803 mutex_enter(&spa_namespace_lock
);
9804 spa_close(spa
, FTAG
);
9806 mutex_exit(&spa_namespace_lock
);
9810 spa_sync_tq_create(spa_t
*spa
, const char *name
)
9812 kthread_t
**kthreads
;
9814 ASSERT(spa
->spa_sync_tq
== NULL
);
9815 ASSERT3S(spa
->spa_alloc_count
, <=, boot_ncpus
);
9818 * - do not allow more allocators than cpus.
9819 * - there may be more cpus than allocators.
9820 * - do not allow more sync taskq threads than allocators or cpus.
9822 int nthreads
= spa
->spa_alloc_count
;
9823 spa
->spa_syncthreads
= kmem_zalloc(sizeof (spa_syncthread_info_t
) *
9824 nthreads
, KM_SLEEP
);
9826 spa
->spa_sync_tq
= taskq_create_synced(name
, nthreads
, minclsyspri
,
9827 nthreads
, INT_MAX
, TASKQ_PREPOPULATE
, &kthreads
);
9828 VERIFY(spa
->spa_sync_tq
!= NULL
);
9829 VERIFY(kthreads
!= NULL
);
9832 &spa
->spa_zio_taskq
[ZIO_TYPE_WRITE
][ZIO_TASKQ_ISSUE
];
9834 spa_syncthread_info_t
*ti
= spa
->spa_syncthreads
;
9835 for (int i
= 0, w
= 0; i
< nthreads
; i
++, w
++, ti
++) {
9836 ti
->sti_thread
= kthreads
[i
];
9837 if (w
== tqs
->stqs_count
) {
9840 ti
->sti_wr_iss_tq
= tqs
->stqs_taskq
[w
];
9843 kmem_free(kthreads
, sizeof (*kthreads
) * nthreads
);
9844 return (spa
->spa_sync_tq
);
9848 spa_sync_tq_destroy(spa_t
*spa
)
9850 ASSERT(spa
->spa_sync_tq
!= NULL
);
9852 taskq_wait(spa
->spa_sync_tq
);
9853 taskq_destroy(spa
->spa_sync_tq
);
9854 kmem_free(spa
->spa_syncthreads
,
9855 sizeof (spa_syncthread_info_t
) * spa
->spa_alloc_count
);
9856 spa
->spa_sync_tq
= NULL
;
9860 spa_select_allocator(zio_t
*zio
)
9862 zbookmark_phys_t
*bm
= &zio
->io_bookmark
;
9863 spa_t
*spa
= zio
->io_spa
;
9865 ASSERT(zio
->io_type
== ZIO_TYPE_WRITE
);
9868 * A gang block (for example) may have inherited its parent's
9869 * allocator, in which case there is nothing further to do here.
9871 if (ZIO_HAS_ALLOCATOR(zio
))
9874 ASSERT(spa
!= NULL
);
9878 * First try to use an allocator assigned to the syncthread, and set
9879 * the corresponding write issue taskq for the allocator.
9880 * Note, we must have an open pool to do this.
9882 if (spa
->spa_sync_tq
!= NULL
) {
9883 spa_syncthread_info_t
*ti
= spa
->spa_syncthreads
;
9884 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++, ti
++) {
9885 if (ti
->sti_thread
== curthread
) {
9886 zio
->io_allocator
= i
;
9887 zio
->io_wr_iss_tq
= ti
->sti_wr_iss_tq
;
9894 * We want to try to use as many allocators as possible to help improve
9895 * performance, but we also want logically adjacent IOs to be physically
9896 * adjacent to improve sequential read performance. We chunk each object
9897 * into 2^20 block regions, and then hash based on the objset, object,
9898 * level, and region to accomplish both of these goals.
9900 uint64_t hv
= cityhash4(bm
->zb_objset
, bm
->zb_object
, bm
->zb_level
,
9901 bm
->zb_blkid
>> 20);
9903 zio
->io_allocator
= (uint_t
)hv
% spa
->spa_alloc_count
;
9904 zio
->io_wr_iss_tq
= NULL
;
9908 * ==========================================================================
9909 * Miscellaneous routines
9910 * ==========================================================================
9914 * Remove all pools in the system.
9922 * Remove all cached state. All pools should be closed now,
9923 * so every spa in the AVL tree should be unreferenced.
9925 mutex_enter(&spa_namespace_lock
);
9926 while ((spa
= spa_next(NULL
)) != NULL
) {
9928 * Stop async tasks. The async thread may need to detach
9929 * a device that's been replaced, which requires grabbing
9930 * spa_namespace_lock, so we must drop it here.
9932 spa_open_ref(spa
, FTAG
);
9933 mutex_exit(&spa_namespace_lock
);
9934 spa_async_suspend(spa
);
9935 mutex_enter(&spa_namespace_lock
);
9936 spa_close(spa
, FTAG
);
9938 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
9940 spa_deactivate(spa
);
9944 mutex_exit(&spa_namespace_lock
);
9948 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
9953 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
9957 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
9958 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
9959 if (vd
->vdev_guid
== guid
)
9963 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
9964 vd
= spa
->spa_spares
.sav_vdevs
[i
];
9965 if (vd
->vdev_guid
== guid
)
9974 spa_upgrade(spa_t
*spa
, uint64_t version
)
9976 ASSERT(spa_writeable(spa
));
9978 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
9981 * This should only be called for a non-faulted pool, and since a
9982 * future version would result in an unopenable pool, this shouldn't be
9985 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
9986 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
9988 spa
->spa_uberblock
.ub_version
= version
;
9989 vdev_config_dirty(spa
->spa_root_vdev
);
9991 spa_config_exit(spa
, SCL_ALL
, FTAG
);
9993 txg_wait_synced(spa_get_dsl(spa
), 0);
9997 spa_has_aux_vdev(spa_t
*spa
, uint64_t guid
, spa_aux_vdev_t
*sav
)
10001 uint64_t vdev_guid
;
10003 for (i
= 0; i
< sav
->sav_count
; i
++)
10004 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
10007 for (i
= 0; i
< sav
->sav_npending
; i
++) {
10008 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
10009 &vdev_guid
) == 0 && vdev_guid
== guid
)
10017 spa_has_l2cache(spa_t
*spa
, uint64_t guid
)
10019 return (spa_has_aux_vdev(spa
, guid
, &spa
->spa_l2cache
));
10023 spa_has_spare(spa_t
*spa
, uint64_t guid
)
10025 return (spa_has_aux_vdev(spa
, guid
, &spa
->spa_spares
));
10029 * Check if a pool has an active shared spare device.
10030 * Note: reference count of an active spare is 2, as a spare and as a replace
10033 spa_has_active_shared_spare(spa_t
*spa
)
10037 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
10039 for (i
= 0; i
< sav
->sav_count
; i
++) {
10040 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
10041 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
10050 spa_total_metaslabs(spa_t
*spa
)
10052 vdev_t
*rvd
= spa
->spa_root_vdev
;
10055 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
10056 vdev_t
*vd
= rvd
->vdev_child
[c
];
10057 if (!vdev_is_concrete(vd
))
10059 m
+= vd
->vdev_ms_count
;
10065 * Notify any waiting threads that some activity has switched from being in-
10066 * progress to not-in-progress so that the thread can wake up and determine
10067 * whether it is finished waiting.
10070 spa_notify_waiters(spa_t
*spa
)
10073 * Acquiring spa_activities_lock here prevents the cv_broadcast from
10074 * happening between the waiting thread's check and cv_wait.
10076 mutex_enter(&spa
->spa_activities_lock
);
10077 cv_broadcast(&spa
->spa_activities_cv
);
10078 mutex_exit(&spa
->spa_activities_lock
);
10082 * Notify any waiting threads that the pool is exporting, and then block until
10083 * they are finished using the spa_t.
10086 spa_wake_waiters(spa_t
*spa
)
10088 mutex_enter(&spa
->spa_activities_lock
);
10089 spa
->spa_waiters_cancel
= B_TRUE
;
10090 cv_broadcast(&spa
->spa_activities_cv
);
10091 while (spa
->spa_waiters
!= 0)
10092 cv_wait(&spa
->spa_waiters_cv
, &spa
->spa_activities_lock
);
10093 spa
->spa_waiters_cancel
= B_FALSE
;
10094 mutex_exit(&spa
->spa_activities_lock
);
10097 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
10099 spa_vdev_activity_in_progress_impl(vdev_t
*vd
, zpool_wait_activity_t activity
)
10101 spa_t
*spa
= vd
->vdev_spa
;
10103 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
));
10104 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
10105 ASSERT(activity
== ZPOOL_WAIT_INITIALIZE
||
10106 activity
== ZPOOL_WAIT_TRIM
);
10108 kmutex_t
*lock
= activity
== ZPOOL_WAIT_INITIALIZE
?
10109 &vd
->vdev_initialize_lock
: &vd
->vdev_trim_lock
;
10111 mutex_exit(&spa
->spa_activities_lock
);
10113 mutex_enter(&spa
->spa_activities_lock
);
10115 boolean_t in_progress
= (activity
== ZPOOL_WAIT_INITIALIZE
) ?
10116 (vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) :
10117 (vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
);
10123 for (int i
= 0; i
< vd
->vdev_children
; i
++) {
10124 if (spa_vdev_activity_in_progress_impl(vd
->vdev_child
[i
],
10133 * If use_guid is true, this checks whether the vdev specified by guid is
10134 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
10135 * is being initialized/trimmed. The caller must hold the config lock and
10136 * spa_activities_lock.
10139 spa_vdev_activity_in_progress(spa_t
*spa
, boolean_t use_guid
, uint64_t guid
,
10140 zpool_wait_activity_t activity
, boolean_t
*in_progress
)
10142 mutex_exit(&spa
->spa_activities_lock
);
10143 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
10144 mutex_enter(&spa
->spa_activities_lock
);
10148 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
10149 if (vd
== NULL
|| !vd
->vdev_ops
->vdev_op_leaf
) {
10150 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
10154 vd
= spa
->spa_root_vdev
;
10157 *in_progress
= spa_vdev_activity_in_progress_impl(vd
, activity
);
10159 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
10164 * Locking for waiting threads
10165 * ---------------------------
10167 * Waiting threads need a way to check whether a given activity is in progress,
10168 * and then, if it is, wait for it to complete. Each activity will have some
10169 * in-memory representation of the relevant on-disk state which can be used to
10170 * determine whether or not the activity is in progress. The in-memory state and
10171 * the locking used to protect it will be different for each activity, and may
10172 * not be suitable for use with a cvar (e.g., some state is protected by the
10173 * config lock). To allow waiting threads to wait without any races, another
10174 * lock, spa_activities_lock, is used.
10176 * When the state is checked, both the activity-specific lock (if there is one)
10177 * and spa_activities_lock are held. In some cases, the activity-specific lock
10178 * is acquired explicitly (e.g. the config lock). In others, the locking is
10179 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
10180 * thread releases the activity-specific lock and, if the activity is in
10181 * progress, then cv_waits using spa_activities_lock.
10183 * The waiting thread is woken when another thread, one completing some
10184 * activity, updates the state of the activity and then calls
10185 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
10186 * needs to hold its activity-specific lock when updating the state, and this
10187 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
10189 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
10190 * and because it is held when the waiting thread checks the state of the
10191 * activity, it can never be the case that the completing thread both updates
10192 * the activity state and cv_broadcasts in between the waiting thread's check
10193 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
10195 * In order to prevent deadlock, when the waiting thread does its check, in some
10196 * cases it will temporarily drop spa_activities_lock in order to acquire the
10197 * activity-specific lock. The order in which spa_activities_lock and the
10198 * activity specific lock are acquired in the waiting thread is determined by
10199 * the order in which they are acquired in the completing thread; if the
10200 * completing thread calls spa_notify_waiters with the activity-specific lock
10201 * held, then the waiting thread must also acquire the activity-specific lock
10206 spa_activity_in_progress(spa_t
*spa
, zpool_wait_activity_t activity
,
10207 boolean_t use_tag
, uint64_t tag
, boolean_t
*in_progress
)
10211 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
10213 switch (activity
) {
10214 case ZPOOL_WAIT_CKPT_DISCARD
:
10216 (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
) &&
10217 zap_contains(spa_meta_objset(spa
),
10218 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ZPOOL_CHECKPOINT
) ==
10221 case ZPOOL_WAIT_FREE
:
10222 *in_progress
= ((spa_version(spa
) >= SPA_VERSION_DEADLISTS
&&
10223 !bpobj_is_empty(&spa
->spa_dsl_pool
->dp_free_bpobj
)) ||
10224 spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
) ||
10225 spa_livelist_delete_check(spa
));
10227 case ZPOOL_WAIT_INITIALIZE
:
10228 case ZPOOL_WAIT_TRIM
:
10229 error
= spa_vdev_activity_in_progress(spa
, use_tag
, tag
,
10230 activity
, in_progress
);
10232 case ZPOOL_WAIT_REPLACE
:
10233 mutex_exit(&spa
->spa_activities_lock
);
10234 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
10235 mutex_enter(&spa
->spa_activities_lock
);
10237 *in_progress
= vdev_replace_in_progress(spa
->spa_root_vdev
);
10238 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
10240 case ZPOOL_WAIT_REMOVE
:
10241 *in_progress
= (spa
->spa_removing_phys
.sr_state
==
10244 case ZPOOL_WAIT_RESILVER
:
10245 *in_progress
= vdev_rebuild_active(spa
->spa_root_vdev
);
10249 case ZPOOL_WAIT_SCRUB
:
10251 boolean_t scanning
, paused
, is_scrub
;
10252 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
10254 is_scrub
= (scn
->scn_phys
.scn_func
== POOL_SCAN_SCRUB
);
10255 scanning
= (scn
->scn_phys
.scn_state
== DSS_SCANNING
);
10256 paused
= dsl_scan_is_paused_scrub(scn
);
10257 *in_progress
= (scanning
&& !paused
&&
10258 is_scrub
== (activity
== ZPOOL_WAIT_SCRUB
));
10261 case ZPOOL_WAIT_RAIDZ_EXPAND
:
10263 vdev_raidz_expand_t
*vre
= spa
->spa_raidz_expand
;
10264 *in_progress
= (vre
!= NULL
&& vre
->vre_state
== DSS_SCANNING
);
10268 panic("unrecognized value for activity %d", activity
);
10275 spa_wait_common(const char *pool
, zpool_wait_activity_t activity
,
10276 boolean_t use_tag
, uint64_t tag
, boolean_t
*waited
)
10279 * The tag is used to distinguish between instances of an activity.
10280 * 'initialize' and 'trim' are the only activities that we use this for.
10281 * The other activities can only have a single instance in progress in a
10282 * pool at one time, making the tag unnecessary.
10284 * There can be multiple devices being replaced at once, but since they
10285 * all finish once resilvering finishes, we don't bother keeping track
10286 * of them individually, we just wait for them all to finish.
10288 if (use_tag
&& activity
!= ZPOOL_WAIT_INITIALIZE
&&
10289 activity
!= ZPOOL_WAIT_TRIM
)
10292 if (activity
< 0 || activity
>= ZPOOL_WAIT_NUM_ACTIVITIES
)
10296 int error
= spa_open(pool
, &spa
, FTAG
);
10301 * Increment the spa's waiter count so that we can call spa_close and
10302 * still ensure that the spa_t doesn't get freed before this thread is
10303 * finished with it when the pool is exported. We want to call spa_close
10304 * before we start waiting because otherwise the additional ref would
10305 * prevent the pool from being exported or destroyed throughout the
10306 * potentially long wait.
10308 mutex_enter(&spa
->spa_activities_lock
);
10309 spa
->spa_waiters
++;
10310 spa_close(spa
, FTAG
);
10314 boolean_t in_progress
;
10315 error
= spa_activity_in_progress(spa
, activity
, use_tag
, tag
,
10318 if (error
|| !in_progress
|| spa
->spa_waiters_cancel
)
10323 if (cv_wait_sig(&spa
->spa_activities_cv
,
10324 &spa
->spa_activities_lock
) == 0) {
10330 spa
->spa_waiters
--;
10331 cv_signal(&spa
->spa_waiters_cv
);
10332 mutex_exit(&spa
->spa_activities_lock
);
10338 * Wait for a particular instance of the specified activity to complete, where
10339 * the instance is identified by 'tag'
10342 spa_wait_tag(const char *pool
, zpool_wait_activity_t activity
, uint64_t tag
,
10345 return (spa_wait_common(pool
, activity
, B_TRUE
, tag
, waited
));
10349 * Wait for all instances of the specified activity complete
10352 spa_wait(const char *pool
, zpool_wait_activity_t activity
, boolean_t
*waited
)
10355 return (spa_wait_common(pool
, activity
, B_FALSE
, 0, waited
));
10359 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
10361 sysevent_t
*ev
= NULL
;
10363 nvlist_t
*resource
;
10365 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
10367 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
10368 ev
->resource
= resource
;
10371 (void) spa
, (void) vd
, (void) hist_nvl
, (void) name
;
10377 spa_event_post(sysevent_t
*ev
)
10381 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
10382 kmem_free(ev
, sizeof (*ev
));
10390 * Post a zevent corresponding to the given sysevent. The 'name' must be one
10391 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
10392 * filled in from the spa and (optionally) the vdev. This doesn't do anything
10393 * in the userland libzpool, as we don't want consumers to misinterpret ztest
10394 * or zdb as real changes.
10397 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
10399 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
10402 /* state manipulation functions */
10403 EXPORT_SYMBOL(spa_open
);
10404 EXPORT_SYMBOL(spa_open_rewind
);
10405 EXPORT_SYMBOL(spa_get_stats
);
10406 EXPORT_SYMBOL(spa_create
);
10407 EXPORT_SYMBOL(spa_import
);
10408 EXPORT_SYMBOL(spa_tryimport
);
10409 EXPORT_SYMBOL(spa_destroy
);
10410 EXPORT_SYMBOL(spa_export
);
10411 EXPORT_SYMBOL(spa_reset
);
10412 EXPORT_SYMBOL(spa_async_request
);
10413 EXPORT_SYMBOL(spa_async_suspend
);
10414 EXPORT_SYMBOL(spa_async_resume
);
10415 EXPORT_SYMBOL(spa_inject_addref
);
10416 EXPORT_SYMBOL(spa_inject_delref
);
10417 EXPORT_SYMBOL(spa_scan_stat_init
);
10418 EXPORT_SYMBOL(spa_scan_get_stats
);
10420 /* device manipulation */
10421 EXPORT_SYMBOL(spa_vdev_add
);
10422 EXPORT_SYMBOL(spa_vdev_attach
);
10423 EXPORT_SYMBOL(spa_vdev_detach
);
10424 EXPORT_SYMBOL(spa_vdev_setpath
);
10425 EXPORT_SYMBOL(spa_vdev_setfru
);
10426 EXPORT_SYMBOL(spa_vdev_split_mirror
);
10428 /* spare statech is global across all pools) */
10429 EXPORT_SYMBOL(spa_spare_add
);
10430 EXPORT_SYMBOL(spa_spare_remove
);
10431 EXPORT_SYMBOL(spa_spare_exists
);
10432 EXPORT_SYMBOL(spa_spare_activate
);
10434 /* L2ARC statech is global across all pools) */
10435 EXPORT_SYMBOL(spa_l2cache_add
);
10436 EXPORT_SYMBOL(spa_l2cache_remove
);
10437 EXPORT_SYMBOL(spa_l2cache_exists
);
10438 EXPORT_SYMBOL(spa_l2cache_activate
);
10439 EXPORT_SYMBOL(spa_l2cache_drop
);
10442 EXPORT_SYMBOL(spa_scan
);
10443 EXPORT_SYMBOL(spa_scan_stop
);
10446 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
10447 EXPORT_SYMBOL(spa_sync_allpools
);
10450 EXPORT_SYMBOL(spa_prop_set
);
10451 EXPORT_SYMBOL(spa_prop_get
);
10452 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
10454 /* asynchronous event notification */
10455 EXPORT_SYMBOL(spa_event_notify
);
10457 ZFS_MODULE_PARAM(zfs_metaslab
, metaslab_
, preload_pct
, UINT
, ZMOD_RW
,
10458 "Percentage of CPUs to run a metaslab preload taskq");
10460 /* BEGIN CSTYLED */
10461 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_shift
, UINT
, ZMOD_RW
,
10462 "log2 fraction of arc that can be used by inflight I/Os when "
10463 "verifying pool during import");
10466 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_metadata
, INT
, ZMOD_RW
,
10467 "Set to traverse metadata on pool import");
10469 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_data
, INT
, ZMOD_RW
,
10470 "Set to traverse data on pool import");
10472 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_print_vdev_tree
, INT
, ZMOD_RW
,
10473 "Print vdev tree to zfs_dbgmsg during pool import");
10475 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_pct
, UINT
, ZMOD_RD
,
10476 "Percentage of CPUs to run an IO worker thread");
10478 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_tpq
, UINT
, ZMOD_RD
,
10479 "Number of threads per IO worker taskqueue");
10481 /* BEGIN CSTYLED */
10482 ZFS_MODULE_PARAM(zfs
, zfs_
, max_missing_tvds
, U64
, ZMOD_RW
,
10483 "Allow importing pool with up to this number of missing top-level "
10484 "vdevs (in read-only mode)");
10487 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_pause
, INT
,
10488 ZMOD_RW
, "Set the livelist condense zthr to pause");
10490 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_pause
, INT
,
10491 ZMOD_RW
, "Set the livelist condense synctask to pause");
10493 /* BEGIN CSTYLED */
10494 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_cancel
,
10496 "Whether livelist condensing was canceled in the synctask");
10498 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_cancel
,
10500 "Whether livelist condensing was canceled in the zthr function");
10502 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, new_alloc
, INT
,
10504 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
10505 "was being condensed");
10508 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_wr_iss_ncpus
, UINT
, ZMOD_RW
,
10509 "Number of CPUs to run write issue taskqs");