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 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
;
1216 * The READ and WRITE rows of zio_taskqs are configurable at module load time
1217 * by setting zio_taskq_read or zio_taskq_write.
1219 * Example (the defaults for READ and WRITE)
1220 * zio_taskq_read='fixed,1,8 null scale null'
1221 * zio_taskq_write='sync fixed,1,5 scale fixed,1,5'
1223 * Each sets the entire row at a time.
1225 * 'fixed' is parameterised: fixed,Q,T where Q is number of taskqs, T is number
1226 * of threads per taskq.
1228 * 'null' can only be set on the high-priority queues (queue selection for
1229 * high-priority queues will fall back to the regular queue if the high-pri
1232 static const char *const modes
[ZTI_NMODES
] = {
1233 "fixed", "scale", "sync", "null"
1236 /* Parse the incoming config string. Modifies cfg */
1238 spa_taskq_param_set(zio_type_t t
, char *cfg
)
1242 zio_taskq_info_t row
[ZIO_TASKQ_TYPES
] = {{0}};
1244 char *next
= cfg
, *tok
, *c
;
1247 * Parse out each element from the string and fill `row`. The entire
1248 * row has to be set at once, so any errors are flagged by just
1249 * breaking out of this loop early.
1252 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1253 /* `next` is the start of the config */
1257 /* Eat up leading space */
1258 while (isspace(*next
))
1263 /* Mode ends at space or end of string */
1265 next
= strchr(tok
, ' ');
1266 if (next
!= NULL
) *next
++ = '\0';
1268 /* Parameters start after a comma */
1269 c
= strchr(tok
, ',');
1270 if (c
!= NULL
) *c
++ = '\0';
1272 /* Match mode string */
1274 for (mode
= 0; mode
< ZTI_NMODES
; mode
++)
1275 if (strcmp(tok
, modes
[mode
]) == 0)
1277 if (mode
== ZTI_NMODES
)
1280 /* Invalid canary */
1281 row
[q
].zti_mode
= ZTI_NMODES
;
1283 /* Per-mode setup */
1287 * FIXED is parameterised: number of queues, and number of
1288 * threads per queue.
1290 case ZTI_MODE_FIXED
: {
1291 /* No parameters? */
1292 if (c
== NULL
|| *c
== '\0')
1295 /* Find next parameter */
1297 c
= strchr(tok
, ',');
1301 /* Take digits and convert */
1302 unsigned long long nq
;
1303 if (!(isdigit(*tok
)))
1305 err
= ddi_strtoull(tok
, &tok
, 10, &nq
);
1306 /* Must succeed and also end at the next param sep */
1307 if (err
!= 0 || tok
!= c
)
1310 /* Move past the comma */
1312 /* Need another number */
1313 if (!(isdigit(*tok
)))
1315 /* Remember start to make sure we moved */
1319 unsigned long long ntpq
;
1320 err
= ddi_strtoull(tok
, &tok
, 10, &ntpq
);
1321 /* Must succeed, and moved forward */
1322 if (err
!= 0 || tok
== c
|| *tok
!= '\0')
1326 * sanity; zero queues/threads make no sense, and
1327 * 16K is almost certainly more than anyone will ever
1328 * need and avoids silly numbers like UINT32_MAX
1330 if (nq
== 0 || nq
>= 16384 ||
1331 ntpq
== 0 || ntpq
>= 16384)
1334 const zio_taskq_info_t zti
= ZTI_P(ntpq
, nq
);
1339 case ZTI_MODE_SCALE
: {
1340 const zio_taskq_info_t zti
= ZTI_SCALE
;
1345 case ZTI_MODE_SYNC
: {
1346 const zio_taskq_info_t zti
= ZTI_SYNC
;
1351 case ZTI_MODE_NULL
: {
1353 * Can only null the high-priority queues; the general-
1354 * purpose ones have to exist.
1356 if (q
!= ZIO_TASKQ_ISSUE_HIGH
&&
1357 q
!= ZIO_TASKQ_INTERRUPT_HIGH
)
1360 const zio_taskq_info_t zti
= ZTI_NULL
;
1369 /* Ensure we set a mode */
1370 if (row
[q
].zti_mode
== ZTI_NMODES
)
1374 /* Didn't get a full row, fail */
1375 if (q
< ZIO_TASKQ_TYPES
)
1376 return (SET_ERROR(EINVAL
));
1378 /* Eat trailing space */
1380 while (isspace(*next
))
1383 /* If there's anything left over then fail */
1384 if (next
!= NULL
&& *next
!= '\0')
1385 return (SET_ERROR(EINVAL
));
1387 /* Success! Copy it into the real config */
1388 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++)
1389 zio_taskqs
[t
][q
] = row
[q
];
1395 spa_taskq_param_get(zio_type_t t
, char *buf
, boolean_t add_newline
)
1399 /* Build paramater string from live config */
1400 const char *sep
= "";
1401 for (uint_t q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1402 const zio_taskq_info_t
*zti
= &zio_taskqs
[t
][q
];
1403 if (zti
->zti_mode
== ZTI_MODE_FIXED
)
1404 pos
+= sprintf(&buf
[pos
], "%s%s,%u,%u", sep
,
1405 modes
[zti
->zti_mode
], zti
->zti_count
,
1408 pos
+= sprintf(&buf
[pos
], "%s%s", sep
,
1409 modes
[zti
->zti_mode
]);
1422 spa_taskq_read_param_set(const char *val
, zfs_kernel_param_t
*kp
)
1424 char *cfg
= kmem_strdup(val
);
1425 int err
= spa_taskq_param_set(ZIO_TYPE_READ
, cfg
);
1426 kmem_free(cfg
, strlen(val
)+1);
1430 spa_taskq_read_param_get(char *buf
, zfs_kernel_param_t
*kp
)
1432 return (spa_taskq_param_get(ZIO_TYPE_READ
, buf
, TRUE
));
1436 spa_taskq_write_param_set(const char *val
, zfs_kernel_param_t
*kp
)
1438 char *cfg
= kmem_strdup(val
);
1439 int err
= spa_taskq_param_set(ZIO_TYPE_WRITE
, cfg
);
1440 kmem_free(cfg
, strlen(val
)+1);
1444 spa_taskq_write_param_get(char *buf
, zfs_kernel_param_t
*kp
)
1446 return (spa_taskq_param_get(ZIO_TYPE_WRITE
, buf
, TRUE
));
1450 * On FreeBSD load-time parameters can be set up before malloc() is available,
1451 * so we have to do all the parsing work on the stack.
1453 #define SPA_TASKQ_PARAM_MAX (128)
1456 spa_taskq_read_param(ZFS_MODULE_PARAM_ARGS
)
1458 char buf
[SPA_TASKQ_PARAM_MAX
];
1461 (void) spa_taskq_param_get(ZIO_TYPE_READ
, buf
, FALSE
);
1462 err
= sysctl_handle_string(oidp
, buf
, sizeof (buf
), req
);
1463 if (err
|| req
->newptr
== NULL
)
1465 return (spa_taskq_param_set(ZIO_TYPE_READ
, buf
));
1469 spa_taskq_write_param(ZFS_MODULE_PARAM_ARGS
)
1471 char buf
[SPA_TASKQ_PARAM_MAX
];
1474 (void) spa_taskq_param_get(ZIO_TYPE_WRITE
, buf
, FALSE
);
1475 err
= sysctl_handle_string(oidp
, buf
, sizeof (buf
), req
);
1476 if (err
|| req
->newptr
== NULL
)
1478 return (spa_taskq_param_set(ZIO_TYPE_WRITE
, buf
));
1481 #endif /* _KERNEL */
1484 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1485 * Note that a type may have multiple discrete taskqs to avoid lock contention
1486 * on the taskq itself.
1489 spa_taskq_dispatch_select(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1492 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1495 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1496 ASSERT3U(tqs
->stqs_count
, !=, 0);
1498 if ((t
== ZIO_TYPE_WRITE
) && (q
== ZIO_TASKQ_ISSUE
) &&
1499 (zio
!= NULL
) && (zio
->io_wr_iss_tq
!= NULL
)) {
1500 /* dispatch to assigned write issue taskq */
1501 tq
= zio
->io_wr_iss_tq
;
1505 if (tqs
->stqs_count
== 1) {
1506 tq
= tqs
->stqs_taskq
[0];
1508 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1514 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1515 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
,
1518 taskq_t
*tq
= spa_taskq_dispatch_select(spa
, t
, q
, zio
);
1519 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
1523 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1526 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1527 task_func_t
*func
, void *arg
, uint_t flags
)
1529 taskq_t
*tq
= spa_taskq_dispatch_select(spa
, t
, q
, NULL
);
1530 taskqid_t id
= taskq_dispatch(tq
, func
, arg
, flags
);
1532 taskq_wait_id(tq
, id
);
1536 spa_create_zio_taskqs(spa_t
*spa
)
1538 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1539 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1540 spa_taskqs_init(spa
, t
, q
);
1545 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1547 spa_thread(void *arg
)
1549 psetid_t zio_taskq_psrset_bind
= PS_NONE
;
1550 callb_cpr_t cprinfo
;
1553 user_t
*pu
= PTOU(curproc
);
1555 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1558 ASSERT(curproc
!= &p0
);
1559 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1560 "zpool-%s", spa
->spa_name
);
1561 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1563 /* bind this thread to the requested psrset */
1564 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1566 mutex_enter(&cpu_lock
);
1567 mutex_enter(&pidlock
);
1568 mutex_enter(&curproc
->p_lock
);
1570 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1571 0, NULL
, NULL
) == 0) {
1572 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1575 "Couldn't bind process for zfs pool \"%s\" to "
1576 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1579 mutex_exit(&curproc
->p_lock
);
1580 mutex_exit(&pidlock
);
1581 mutex_exit(&cpu_lock
);
1586 if (zio_taskq_sysdc
) {
1587 sysdc_thread_enter(curthread
, 100, 0);
1591 spa
->spa_proc
= curproc
;
1592 spa
->spa_did
= curthread
->t_did
;
1594 spa_create_zio_taskqs(spa
);
1596 mutex_enter(&spa
->spa_proc_lock
);
1597 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1599 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1600 cv_broadcast(&spa
->spa_proc_cv
);
1602 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1603 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1604 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1605 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1607 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1608 spa
->spa_proc_state
= SPA_PROC_GONE
;
1609 spa
->spa_proc
= &p0
;
1610 cv_broadcast(&spa
->spa_proc_cv
);
1611 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1613 mutex_enter(&curproc
->p_lock
);
1618 extern metaslab_ops_t
*metaslab_allocator(spa_t
*spa
);
1621 * Activate an uninitialized pool.
1624 spa_activate(spa_t
*spa
, spa_mode_t mode
)
1626 metaslab_ops_t
*msp
= metaslab_allocator(spa
);
1627 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1629 spa
->spa_state
= POOL_STATE_ACTIVE
;
1630 spa
->spa_mode
= mode
;
1631 spa
->spa_read_spacemaps
= spa_mode_readable_spacemaps
;
1633 spa
->spa_normal_class
= metaslab_class_create(spa
, msp
);
1634 spa
->spa_log_class
= metaslab_class_create(spa
, msp
);
1635 spa
->spa_embedded_log_class
= metaslab_class_create(spa
, msp
);
1636 spa
->spa_special_class
= metaslab_class_create(spa
, msp
);
1637 spa
->spa_dedup_class
= metaslab_class_create(spa
, msp
);
1639 /* Try to create a covering process */
1640 mutex_enter(&spa
->spa_proc_lock
);
1641 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1642 ASSERT(spa
->spa_proc
== &p0
);
1645 #ifdef HAVE_SPA_THREAD
1646 /* Only create a process if we're going to be around a while. */
1647 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1648 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1650 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1651 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1652 cv_wait(&spa
->spa_proc_cv
,
1653 &spa
->spa_proc_lock
);
1655 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1656 ASSERT(spa
->spa_proc
!= &p0
);
1657 ASSERT(spa
->spa_did
!= 0);
1661 "Couldn't create process for zfs pool \"%s\"\n",
1666 #endif /* HAVE_SPA_THREAD */
1667 mutex_exit(&spa
->spa_proc_lock
);
1669 /* If we didn't create a process, we need to create our taskqs. */
1670 if (spa
->spa_proc
== &p0
) {
1671 spa_create_zio_taskqs(spa
);
1674 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1675 spa
->spa_txg_zio
[i
] = zio_root(spa
, NULL
, NULL
,
1679 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1680 offsetof(vdev_t
, vdev_config_dirty_node
));
1681 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1682 offsetof(objset_t
, os_evicting_node
));
1683 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1684 offsetof(vdev_t
, vdev_state_dirty_node
));
1686 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1687 offsetof(struct vdev
, vdev_txg_node
));
1689 avl_create(&spa
->spa_errlist_scrub
,
1690 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1691 offsetof(spa_error_entry_t
, se_avl
));
1692 avl_create(&spa
->spa_errlist_last
,
1693 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1694 offsetof(spa_error_entry_t
, se_avl
));
1695 avl_create(&spa
->spa_errlist_healed
,
1696 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1697 offsetof(spa_error_entry_t
, se_avl
));
1699 spa_activate_os(spa
);
1701 spa_keystore_init(&spa
->spa_keystore
);
1704 * This taskq is used to perform zvol-minor-related tasks
1705 * asynchronously. This has several advantages, including easy
1706 * resolution of various deadlocks.
1708 * The taskq must be single threaded to ensure tasks are always
1709 * processed in the order in which they were dispatched.
1711 * A taskq per pool allows one to keep the pools independent.
1712 * This way if one pool is suspended, it will not impact another.
1714 * The preferred location to dispatch a zvol minor task is a sync
1715 * task. In this context, there is easy access to the spa_t and minimal
1716 * error handling is required because the sync task must succeed.
1718 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1722 * The taskq to preload metaslabs.
1724 spa
->spa_metaslab_taskq
= taskq_create("z_metaslab",
1725 metaslab_preload_pct
, maxclsyspri
, 1, INT_MAX
,
1726 TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1729 * Taskq dedicated to prefetcher threads: this is used to prevent the
1730 * pool traverse code from monopolizing the global (and limited)
1731 * system_taskq by inappropriately scheduling long running tasks on it.
1733 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", 100,
1734 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1737 * The taskq to upgrade datasets in this pool. Currently used by
1738 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1740 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", 100,
1741 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
| TASKQ_THREADS_CPU_PCT
);
1745 * Opposite of spa_activate().
1748 spa_deactivate(spa_t
*spa
)
1750 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1751 ASSERT(spa
->spa_dsl_pool
== NULL
);
1752 ASSERT(spa
->spa_root_vdev
== NULL
);
1753 ASSERT(spa
->spa_async_zio_root
== NULL
);
1754 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1756 spa_evicting_os_wait(spa
);
1758 if (spa
->spa_zvol_taskq
) {
1759 taskq_destroy(spa
->spa_zvol_taskq
);
1760 spa
->spa_zvol_taskq
= NULL
;
1763 if (spa
->spa_metaslab_taskq
) {
1764 taskq_destroy(spa
->spa_metaslab_taskq
);
1765 spa
->spa_metaslab_taskq
= NULL
;
1768 if (spa
->spa_prefetch_taskq
) {
1769 taskq_destroy(spa
->spa_prefetch_taskq
);
1770 spa
->spa_prefetch_taskq
= NULL
;
1773 if (spa
->spa_upgrade_taskq
) {
1774 taskq_destroy(spa
->spa_upgrade_taskq
);
1775 spa
->spa_upgrade_taskq
= NULL
;
1778 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1780 list_destroy(&spa
->spa_config_dirty_list
);
1781 list_destroy(&spa
->spa_evicting_os_list
);
1782 list_destroy(&spa
->spa_state_dirty_list
);
1784 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1786 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1787 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1788 spa_taskqs_fini(spa
, t
, q
);
1792 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1793 ASSERT3P(spa
->spa_txg_zio
[i
], !=, NULL
);
1794 VERIFY0(zio_wait(spa
->spa_txg_zio
[i
]));
1795 spa
->spa_txg_zio
[i
] = NULL
;
1798 metaslab_class_destroy(spa
->spa_normal_class
);
1799 spa
->spa_normal_class
= NULL
;
1801 metaslab_class_destroy(spa
->spa_log_class
);
1802 spa
->spa_log_class
= NULL
;
1804 metaslab_class_destroy(spa
->spa_embedded_log_class
);
1805 spa
->spa_embedded_log_class
= NULL
;
1807 metaslab_class_destroy(spa
->spa_special_class
);
1808 spa
->spa_special_class
= NULL
;
1810 metaslab_class_destroy(spa
->spa_dedup_class
);
1811 spa
->spa_dedup_class
= NULL
;
1814 * If this was part of an import or the open otherwise failed, we may
1815 * still have errors left in the queues. Empty them just in case.
1817 spa_errlog_drain(spa
);
1818 avl_destroy(&spa
->spa_errlist_scrub
);
1819 avl_destroy(&spa
->spa_errlist_last
);
1820 avl_destroy(&spa
->spa_errlist_healed
);
1822 spa_keystore_fini(&spa
->spa_keystore
);
1824 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1826 mutex_enter(&spa
->spa_proc_lock
);
1827 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1828 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1829 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1830 cv_broadcast(&spa
->spa_proc_cv
);
1831 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1832 ASSERT(spa
->spa_proc
!= &p0
);
1833 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1835 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1836 spa
->spa_proc_state
= SPA_PROC_NONE
;
1838 ASSERT(spa
->spa_proc
== &p0
);
1839 mutex_exit(&spa
->spa_proc_lock
);
1842 * We want to make sure spa_thread() has actually exited the ZFS
1843 * module, so that the module can't be unloaded out from underneath
1846 if (spa
->spa_did
!= 0) {
1847 thread_join(spa
->spa_did
);
1851 spa_deactivate_os(spa
);
1856 * Verify a pool configuration, and construct the vdev tree appropriately. This
1857 * will create all the necessary vdevs in the appropriate layout, with each vdev
1858 * in the CLOSED state. This will prep the pool before open/creation/import.
1859 * All vdev validation is done by the vdev_alloc() routine.
1862 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1863 uint_t id
, int atype
)
1869 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1872 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1875 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1878 if (error
== ENOENT
)
1884 return (SET_ERROR(EINVAL
));
1887 for (int c
= 0; c
< children
; c
++) {
1889 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1897 ASSERT(*vdp
!= NULL
);
1903 spa_should_flush_logs_on_unload(spa_t
*spa
)
1905 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
1908 if (!spa_writeable(spa
))
1911 if (!spa
->spa_sync_on
)
1914 if (spa_state(spa
) != POOL_STATE_EXPORTED
)
1917 if (zfs_keep_log_spacemaps_at_export
)
1924 * Opens a transaction that will set the flag that will instruct
1925 * spa_sync to attempt to flush all the metaslabs for that txg.
1928 spa_unload_log_sm_flush_all(spa_t
*spa
)
1930 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
1931 VERIFY0(dmu_tx_assign(tx
, TXG_WAIT
));
1933 ASSERT3U(spa
->spa_log_flushall_txg
, ==, 0);
1934 spa
->spa_log_flushall_txg
= dmu_tx_get_txg(tx
);
1937 txg_wait_synced(spa_get_dsl(spa
), spa
->spa_log_flushall_txg
);
1941 spa_unload_log_sm_metadata(spa_t
*spa
)
1943 void *cookie
= NULL
;
1945 log_summary_entry_t
*e
;
1947 while ((sls
= avl_destroy_nodes(&spa
->spa_sm_logs_by_txg
,
1948 &cookie
)) != NULL
) {
1949 VERIFY0(sls
->sls_mscount
);
1950 kmem_free(sls
, sizeof (spa_log_sm_t
));
1953 while ((e
= list_remove_head(&spa
->spa_log_summary
)) != NULL
) {
1954 VERIFY0(e
->lse_mscount
);
1955 kmem_free(e
, sizeof (log_summary_entry_t
));
1958 spa
->spa_unflushed_stats
.sus_nblocks
= 0;
1959 spa
->spa_unflushed_stats
.sus_memused
= 0;
1960 spa
->spa_unflushed_stats
.sus_blocklimit
= 0;
1964 spa_destroy_aux_threads(spa_t
*spa
)
1966 if (spa
->spa_condense_zthr
!= NULL
) {
1967 zthr_destroy(spa
->spa_condense_zthr
);
1968 spa
->spa_condense_zthr
= NULL
;
1970 if (spa
->spa_checkpoint_discard_zthr
!= NULL
) {
1971 zthr_destroy(spa
->spa_checkpoint_discard_zthr
);
1972 spa
->spa_checkpoint_discard_zthr
= NULL
;
1974 if (spa
->spa_livelist_delete_zthr
!= NULL
) {
1975 zthr_destroy(spa
->spa_livelist_delete_zthr
);
1976 spa
->spa_livelist_delete_zthr
= NULL
;
1978 if (spa
->spa_livelist_condense_zthr
!= NULL
) {
1979 zthr_destroy(spa
->spa_livelist_condense_zthr
);
1980 spa
->spa_livelist_condense_zthr
= NULL
;
1982 if (spa
->spa_raidz_expand_zthr
!= NULL
) {
1983 zthr_destroy(spa
->spa_raidz_expand_zthr
);
1984 spa
->spa_raidz_expand_zthr
= NULL
;
1989 * Opposite of spa_load().
1992 spa_unload(spa_t
*spa
)
1994 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1995 ASSERT(spa_state(spa
) != POOL_STATE_UNINITIALIZED
);
1997 spa_import_progress_remove(spa_guid(spa
));
1998 spa_load_note(spa
, "UNLOADING");
2000 spa_wake_waiters(spa
);
2003 * If we have set the spa_final_txg, we have already performed the
2004 * tasks below in spa_export_common(). We should not redo it here since
2005 * we delay the final TXGs beyond what spa_final_txg is set at.
2007 if (spa
->spa_final_txg
== UINT64_MAX
) {
2009 * If the log space map feature is enabled and the pool is
2010 * getting exported (but not destroyed), we want to spend some
2011 * time flushing as many metaslabs as we can in an attempt to
2012 * destroy log space maps and save import time.
2014 if (spa_should_flush_logs_on_unload(spa
))
2015 spa_unload_log_sm_flush_all(spa
);
2020 spa_async_suspend(spa
);
2022 if (spa
->spa_root_vdev
) {
2023 vdev_t
*root_vdev
= spa
->spa_root_vdev
;
2024 vdev_initialize_stop_all(root_vdev
,
2025 VDEV_INITIALIZE_ACTIVE
);
2026 vdev_trim_stop_all(root_vdev
, VDEV_TRIM_ACTIVE
);
2027 vdev_autotrim_stop_all(spa
);
2028 vdev_rebuild_stop_all(spa
);
2035 if (spa
->spa_sync_on
) {
2036 txg_sync_stop(spa
->spa_dsl_pool
);
2037 spa
->spa_sync_on
= B_FALSE
;
2041 * This ensures that there is no async metaslab prefetching
2042 * while we attempt to unload the spa.
2044 taskq_wait(spa
->spa_metaslab_taskq
);
2046 if (spa
->spa_mmp
.mmp_thread
)
2047 mmp_thread_stop(spa
);
2050 * Wait for any outstanding async I/O to complete.
2052 if (spa
->spa_async_zio_root
!= NULL
) {
2053 for (int i
= 0; i
< max_ncpus
; i
++)
2054 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
2055 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
2056 spa
->spa_async_zio_root
= NULL
;
2059 if (spa
->spa_vdev_removal
!= NULL
) {
2060 spa_vdev_removal_destroy(spa
->spa_vdev_removal
);
2061 spa
->spa_vdev_removal
= NULL
;
2064 spa_destroy_aux_threads(spa
);
2066 spa_condense_fini(spa
);
2068 bpobj_close(&spa
->spa_deferred_bpobj
);
2070 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
2075 if (spa
->spa_root_vdev
)
2076 vdev_free(spa
->spa_root_vdev
);
2077 ASSERT(spa
->spa_root_vdev
== NULL
);
2080 * Close the dsl pool.
2082 if (spa
->spa_dsl_pool
) {
2083 dsl_pool_close(spa
->spa_dsl_pool
);
2084 spa
->spa_dsl_pool
= NULL
;
2085 spa
->spa_meta_objset
= NULL
;
2090 spa_unload_log_sm_metadata(spa
);
2093 * Drop and purge level 2 cache
2095 spa_l2cache_drop(spa
);
2097 if (spa
->spa_spares
.sav_vdevs
) {
2098 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
2099 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
2100 kmem_free(spa
->spa_spares
.sav_vdevs
,
2101 spa
->spa_spares
.sav_count
* sizeof (void *));
2102 spa
->spa_spares
.sav_vdevs
= NULL
;
2104 if (spa
->spa_spares
.sav_config
) {
2105 nvlist_free(spa
->spa_spares
.sav_config
);
2106 spa
->spa_spares
.sav_config
= NULL
;
2108 spa
->spa_spares
.sav_count
= 0;
2110 if (spa
->spa_l2cache
.sav_vdevs
) {
2111 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
2112 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
2113 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
2115 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
2116 spa
->spa_l2cache
.sav_count
* sizeof (void *));
2117 spa
->spa_l2cache
.sav_vdevs
= NULL
;
2119 if (spa
->spa_l2cache
.sav_config
) {
2120 nvlist_free(spa
->spa_l2cache
.sav_config
);
2121 spa
->spa_l2cache
.sav_config
= NULL
;
2123 spa
->spa_l2cache
.sav_count
= 0;
2125 spa
->spa_async_suspended
= 0;
2127 spa
->spa_indirect_vdevs_loaded
= B_FALSE
;
2129 if (spa
->spa_comment
!= NULL
) {
2130 spa_strfree(spa
->spa_comment
);
2131 spa
->spa_comment
= NULL
;
2133 if (spa
->spa_compatibility
!= NULL
) {
2134 spa_strfree(spa
->spa_compatibility
);
2135 spa
->spa_compatibility
= NULL
;
2138 spa
->spa_raidz_expand
= NULL
;
2140 spa_config_exit(spa
, SCL_ALL
, spa
);
2144 * Load (or re-load) the current list of vdevs describing the active spares for
2145 * this pool. When this is called, we have some form of basic information in
2146 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
2147 * then re-generate a more complete list including status information.
2150 spa_load_spares(spa_t
*spa
)
2159 * zdb opens both the current state of the pool and the
2160 * checkpointed state (if present), with a different spa_t.
2162 * As spare vdevs are shared among open pools, we skip loading
2163 * them when we load the checkpointed state of the pool.
2165 if (!spa_writeable(spa
))
2169 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
2172 * First, close and free any existing spare vdevs.
2174 if (spa
->spa_spares
.sav_vdevs
) {
2175 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
2176 vd
= spa
->spa_spares
.sav_vdevs
[i
];
2178 /* Undo the call to spa_activate() below */
2179 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
2180 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
2181 spa_spare_remove(tvd
);
2186 kmem_free(spa
->spa_spares
.sav_vdevs
,
2187 spa
->spa_spares
.sav_count
* sizeof (void *));
2190 if (spa
->spa_spares
.sav_config
== NULL
)
2193 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
2194 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
));
2196 spa
->spa_spares
.sav_count
= (int)nspares
;
2197 spa
->spa_spares
.sav_vdevs
= NULL
;
2203 * Construct the array of vdevs, opening them to get status in the
2204 * process. For each spare, there is potentially two different vdev_t
2205 * structures associated with it: one in the list of spares (used only
2206 * for basic validation purposes) and one in the active vdev
2207 * configuration (if it's spared in). During this phase we open and
2208 * validate each vdev on the spare list. If the vdev also exists in the
2209 * active configuration, then we also mark this vdev as an active spare.
2211 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
2213 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
2214 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
2215 VDEV_ALLOC_SPARE
) == 0);
2218 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
2220 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
2221 B_FALSE
)) != NULL
) {
2222 if (!tvd
->vdev_isspare
)
2226 * We only mark the spare active if we were successfully
2227 * able to load the vdev. Otherwise, importing a pool
2228 * with a bad active spare would result in strange
2229 * behavior, because multiple pool would think the spare
2230 * is actively in use.
2232 * There is a vulnerability here to an equally bizarre
2233 * circumstance, where a dead active spare is later
2234 * brought back to life (onlined or otherwise). Given
2235 * the rarity of this scenario, and the extra complexity
2236 * it adds, we ignore the possibility.
2238 if (!vdev_is_dead(tvd
))
2239 spa_spare_activate(tvd
);
2243 vd
->vdev_aux
= &spa
->spa_spares
;
2245 if (vdev_open(vd
) != 0)
2248 if (vdev_validate_aux(vd
) == 0)
2253 * Recompute the stashed list of spares, with status information
2256 fnvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
);
2258 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
2260 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
2261 spares
[i
] = vdev_config_generate(spa
,
2262 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
2263 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
2264 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
2265 spa
->spa_spares
.sav_count
);
2266 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
2267 nvlist_free(spares
[i
]);
2268 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
2272 * Load (or re-load) the current list of vdevs describing the active l2cache for
2273 * this pool. When this is called, we have some form of basic information in
2274 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
2275 * then re-generate a more complete list including status information.
2276 * Devices which are already active have their details maintained, and are
2280 spa_load_l2cache(spa_t
*spa
)
2282 nvlist_t
**l2cache
= NULL
;
2284 int i
, j
, oldnvdevs
;
2286 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
2287 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
2291 * zdb opens both the current state of the pool and the
2292 * checkpointed state (if present), with a different spa_t.
2294 * As L2 caches are part of the ARC which is shared among open
2295 * pools, we skip loading them when we load the checkpointed
2296 * state of the pool.
2298 if (!spa_writeable(spa
))
2302 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
2304 oldvdevs
= sav
->sav_vdevs
;
2305 oldnvdevs
= sav
->sav_count
;
2306 sav
->sav_vdevs
= NULL
;
2309 if (sav
->sav_config
== NULL
) {
2315 VERIFY0(nvlist_lookup_nvlist_array(sav
->sav_config
,
2316 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
));
2317 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
2320 * Process new nvlist of vdevs.
2322 for (i
= 0; i
< nl2cache
; i
++) {
2323 guid
= fnvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
);
2326 for (j
= 0; j
< oldnvdevs
; j
++) {
2328 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
2330 * Retain previous vdev for add/remove ops.
2338 if (newvdevs
[i
] == NULL
) {
2342 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
2343 VDEV_ALLOC_L2CACHE
) == 0);
2348 * Commit this vdev as an l2cache device,
2349 * even if it fails to open.
2351 spa_l2cache_add(vd
);
2356 spa_l2cache_activate(vd
);
2358 if (vdev_open(vd
) != 0)
2361 (void) vdev_validate_aux(vd
);
2363 if (!vdev_is_dead(vd
))
2364 l2arc_add_vdev(spa
, vd
);
2367 * Upon cache device addition to a pool or pool
2368 * creation with a cache device or if the header
2369 * of the device is invalid we issue an async
2370 * TRIM command for the whole device which will
2371 * execute if l2arc_trim_ahead > 0.
2373 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
2377 sav
->sav_vdevs
= newvdevs
;
2378 sav
->sav_count
= (int)nl2cache
;
2381 * Recompute the stashed list of l2cache devices, with status
2382 * information this time.
2384 fnvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
);
2386 if (sav
->sav_count
> 0)
2387 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *),
2389 for (i
= 0; i
< sav
->sav_count
; i
++)
2390 l2cache
[i
] = vdev_config_generate(spa
,
2391 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
2392 fnvlist_add_nvlist_array(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
2393 (const nvlist_t
* const *)l2cache
, sav
->sav_count
);
2397 * Purge vdevs that were dropped
2400 for (i
= 0; i
< oldnvdevs
; i
++) {
2405 ASSERT(vd
->vdev_isl2cache
);
2407 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
2408 pool
!= 0ULL && l2arc_vdev_present(vd
))
2409 l2arc_remove_vdev(vd
);
2410 vdev_clear_stats(vd
);
2415 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
2418 for (i
= 0; i
< sav
->sav_count
; i
++)
2419 nvlist_free(l2cache
[i
]);
2421 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
2425 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
2428 char *packed
= NULL
;
2433 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
2437 nvsize
= *(uint64_t *)db
->db_data
;
2438 dmu_buf_rele(db
, FTAG
);
2440 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
2441 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
2444 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
2445 vmem_free(packed
, nvsize
);
2451 * Concrete top-level vdevs that are not missing and are not logs. At every
2452 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
2455 spa_healthy_core_tvds(spa_t
*spa
)
2457 vdev_t
*rvd
= spa
->spa_root_vdev
;
2460 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
2461 vdev_t
*vd
= rvd
->vdev_child
[i
];
2464 if (vdev_is_concrete(vd
) && !vdev_is_dead(vd
))
2472 * Checks to see if the given vdev could not be opened, in which case we post a
2473 * sysevent to notify the autoreplace code that the device has been removed.
2476 spa_check_removed(vdev_t
*vd
)
2478 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
2479 spa_check_removed(vd
->vdev_child
[c
]);
2481 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
2482 vdev_is_concrete(vd
)) {
2483 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
2484 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
2489 spa_check_for_missing_logs(spa_t
*spa
)
2491 vdev_t
*rvd
= spa
->spa_root_vdev
;
2494 * If we're doing a normal import, then build up any additional
2495 * diagnostic information about missing log devices.
2496 * We'll pass this up to the user for further processing.
2498 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
2499 nvlist_t
**child
, *nv
;
2502 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
2504 nv
= fnvlist_alloc();
2506 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2507 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2510 * We consider a device as missing only if it failed
2511 * to open (i.e. offline or faulted is not considered
2514 if (tvd
->vdev_islog
&&
2515 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2516 child
[idx
++] = vdev_config_generate(spa
, tvd
,
2517 B_FALSE
, VDEV_CONFIG_MISSING
);
2522 fnvlist_add_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
2523 (const nvlist_t
* const *)child
, idx
);
2524 fnvlist_add_nvlist(spa
->spa_load_info
,
2525 ZPOOL_CONFIG_MISSING_DEVICES
, nv
);
2527 for (uint64_t i
= 0; i
< idx
; i
++)
2528 nvlist_free(child
[i
]);
2531 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
2534 spa_load_failed(spa
, "some log devices are missing");
2535 vdev_dbgmsg_print_tree(rvd
, 2);
2536 return (SET_ERROR(ENXIO
));
2539 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2540 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2542 if (tvd
->vdev_islog
&&
2543 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
2544 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2545 spa_load_note(spa
, "some log devices are "
2546 "missing, ZIL is dropped.");
2547 vdev_dbgmsg_print_tree(rvd
, 2);
2557 * Check for missing log devices
2560 spa_check_logs(spa_t
*spa
)
2562 boolean_t rv
= B_FALSE
;
2563 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2565 switch (spa
->spa_log_state
) {
2568 case SPA_LOG_MISSING
:
2569 /* need to recheck in case slog has been restored */
2570 case SPA_LOG_UNKNOWN
:
2571 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
2572 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
2574 spa_set_log_state(spa
, SPA_LOG_MISSING
);
2581 * Passivate any log vdevs (note, does not apply to embedded log metaslabs).
2584 spa_passivate_log(spa_t
*spa
)
2586 vdev_t
*rvd
= spa
->spa_root_vdev
;
2587 boolean_t slog_found
= B_FALSE
;
2589 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2591 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2592 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2594 if (tvd
->vdev_islog
) {
2595 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
2596 metaslab_group_passivate(tvd
->vdev_mg
);
2597 slog_found
= B_TRUE
;
2601 return (slog_found
);
2605 * Activate any log vdevs (note, does not apply to embedded log metaslabs).
2608 spa_activate_log(spa_t
*spa
)
2610 vdev_t
*rvd
= spa
->spa_root_vdev
;
2612 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
2614 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2615 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2617 if (tvd
->vdev_islog
) {
2618 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
2619 metaslab_group_activate(tvd
->vdev_mg
);
2625 spa_reset_logs(spa_t
*spa
)
2629 error
= dmu_objset_find(spa_name(spa
), zil_reset
,
2630 NULL
, DS_FIND_CHILDREN
);
2633 * We successfully offlined the log device, sync out the
2634 * current txg so that the "stubby" block can be removed
2637 txg_wait_synced(spa
->spa_dsl_pool
, 0);
2643 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
2645 for (int i
= 0; i
< sav
->sav_count
; i
++)
2646 spa_check_removed(sav
->sav_vdevs
[i
]);
2650 spa_claim_notify(zio_t
*zio
)
2652 spa_t
*spa
= zio
->io_spa
;
2657 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
2658 if (spa
->spa_claim_max_txg
< BP_GET_LOGICAL_BIRTH(zio
->io_bp
))
2659 spa
->spa_claim_max_txg
= BP_GET_LOGICAL_BIRTH(zio
->io_bp
);
2660 mutex_exit(&spa
->spa_props_lock
);
2663 typedef struct spa_load_error
{
2664 boolean_t sle_verify_data
;
2665 uint64_t sle_meta_count
;
2666 uint64_t sle_data_count
;
2670 spa_load_verify_done(zio_t
*zio
)
2672 blkptr_t
*bp
= zio
->io_bp
;
2673 spa_load_error_t
*sle
= zio
->io_private
;
2674 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2675 int error
= zio
->io_error
;
2676 spa_t
*spa
= zio
->io_spa
;
2678 abd_free(zio
->io_abd
);
2680 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2681 type
!= DMU_OT_INTENT_LOG
)
2682 atomic_inc_64(&sle
->sle_meta_count
);
2684 atomic_inc_64(&sle
->sle_data_count
);
2687 mutex_enter(&spa
->spa_scrub_lock
);
2688 spa
->spa_load_verify_bytes
-= BP_GET_PSIZE(bp
);
2689 cv_broadcast(&spa
->spa_scrub_io_cv
);
2690 mutex_exit(&spa
->spa_scrub_lock
);
2694 * Maximum number of inflight bytes is the log2 fraction of the arc size.
2695 * By default, we set it to 1/16th of the arc.
2697 static uint_t spa_load_verify_shift
= 4;
2698 static int spa_load_verify_metadata
= B_TRUE
;
2699 static int spa_load_verify_data
= B_TRUE
;
2702 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2703 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2706 spa_load_error_t
*sle
= rio
->io_private
;
2708 (void) zilog
, (void) dnp
;
2711 * Note: normally this routine will not be called if
2712 * spa_load_verify_metadata is not set. However, it may be useful
2713 * to manually set the flag after the traversal has begun.
2715 if (!spa_load_verify_metadata
)
2719 * Sanity check the block pointer in order to detect obvious damage
2720 * before using the contents in subsequent checks or in zio_read().
2721 * When damaged consider it to be a metadata error since we cannot
2722 * trust the BP_GET_TYPE and BP_GET_LEVEL values.
2724 if (!zfs_blkptr_verify(spa
, bp
, BLK_CONFIG_NEEDED
, BLK_VERIFY_LOG
)) {
2725 atomic_inc_64(&sle
->sle_meta_count
);
2729 if (zb
->zb_level
== ZB_DNODE_LEVEL
|| BP_IS_HOLE(bp
) ||
2730 BP_IS_EMBEDDED(bp
) || BP_IS_REDACTED(bp
))
2733 if (!BP_IS_METADATA(bp
) &&
2734 (!spa_load_verify_data
|| !sle
->sle_verify_data
))
2737 uint64_t maxinflight_bytes
=
2738 arc_target_bytes() >> spa_load_verify_shift
;
2739 size_t size
= BP_GET_PSIZE(bp
);
2741 mutex_enter(&spa
->spa_scrub_lock
);
2742 while (spa
->spa_load_verify_bytes
>= maxinflight_bytes
)
2743 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2744 spa
->spa_load_verify_bytes
+= size
;
2745 mutex_exit(&spa
->spa_scrub_lock
);
2747 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2748 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2749 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2750 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2755 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2757 (void) dp
, (void) arg
;
2759 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2760 return (SET_ERROR(ENAMETOOLONG
));
2766 spa_load_verify(spa_t
*spa
)
2769 spa_load_error_t sle
= { 0 };
2770 zpool_load_policy_t policy
;
2771 boolean_t verify_ok
= B_FALSE
;
2774 zpool_get_load_policy(spa
->spa_config
, &policy
);
2776 if (policy
.zlp_rewind
& ZPOOL_NEVER_REWIND
||
2777 policy
.zlp_maxmeta
== UINT64_MAX
)
2780 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2781 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2782 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2784 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2789 * Verify data only if we are rewinding or error limit was set.
2790 * Otherwise nothing except dbgmsg care about it to waste time.
2792 sle
.sle_verify_data
= (policy
.zlp_rewind
& ZPOOL_REWIND_MASK
) ||
2793 (policy
.zlp_maxdata
< UINT64_MAX
);
2795 rio
= zio_root(spa
, NULL
, &sle
,
2796 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2798 if (spa_load_verify_metadata
) {
2799 if (spa
->spa_extreme_rewind
) {
2800 spa_load_note(spa
, "performing a complete scan of the "
2801 "pool since extreme rewind is on. This may take "
2802 "a very long time.\n (spa_load_verify_data=%u, "
2803 "spa_load_verify_metadata=%u)",
2804 spa_load_verify_data
, spa_load_verify_metadata
);
2807 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2808 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2809 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2812 (void) zio_wait(rio
);
2813 ASSERT0(spa
->spa_load_verify_bytes
);
2815 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2816 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2818 if (sle
.sle_meta_count
!= 0 || sle
.sle_data_count
!= 0) {
2819 spa_load_note(spa
, "spa_load_verify found %llu metadata errors "
2820 "and %llu data errors", (u_longlong_t
)sle
.sle_meta_count
,
2821 (u_longlong_t
)sle
.sle_data_count
);
2824 if (spa_load_verify_dryrun
||
2825 (!error
&& sle
.sle_meta_count
<= policy
.zlp_maxmeta
&&
2826 sle
.sle_data_count
<= policy
.zlp_maxdata
)) {
2830 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2831 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2833 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2834 fnvlist_add_uint64(spa
->spa_load_info
, ZPOOL_CONFIG_LOAD_TIME
,
2835 spa
->spa_load_txg_ts
);
2836 fnvlist_add_int64(spa
->spa_load_info
, ZPOOL_CONFIG_REWIND_TIME
,
2838 fnvlist_add_uint64(spa
->spa_load_info
,
2839 ZPOOL_CONFIG_LOAD_META_ERRORS
, sle
.sle_meta_count
);
2840 fnvlist_add_uint64(spa
->spa_load_info
,
2841 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
);
2843 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2846 if (spa_load_verify_dryrun
)
2850 if (error
!= ENXIO
&& error
!= EIO
)
2851 error
= SET_ERROR(EIO
);
2855 return (verify_ok
? 0 : EIO
);
2859 * Find a value in the pool props object.
2862 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2864 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2865 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2869 * Find a value in the pool directory object.
2872 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
, boolean_t log_enoent
)
2874 int error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2875 name
, sizeof (uint64_t), 1, val
);
2877 if (error
!= 0 && (error
!= ENOENT
|| log_enoent
)) {
2878 spa_load_failed(spa
, "couldn't get '%s' value in MOS directory "
2879 "[error=%d]", name
, error
);
2886 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2888 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2889 return (SET_ERROR(err
));
2893 spa_livelist_delete_check(spa_t
*spa
)
2895 return (spa
->spa_livelists_to_delete
!= 0);
2899 spa_livelist_delete_cb_check(void *arg
, zthr_t
*z
)
2903 return (spa_livelist_delete_check(spa
));
2907 delete_blkptr_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
2910 zio_free(spa
, tx
->tx_txg
, bp
);
2911 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
, DD_USED_HEAD
,
2912 -bp_get_dsize_sync(spa
, bp
),
2913 -BP_GET_PSIZE(bp
), -BP_GET_UCSIZE(bp
), tx
);
2918 dsl_get_next_livelist_obj(objset_t
*os
, uint64_t zap_obj
, uint64_t *llp
)
2923 zap_cursor_init(&zc
, os
, zap_obj
);
2924 err
= zap_cursor_retrieve(&zc
, &za
);
2925 zap_cursor_fini(&zc
);
2927 *llp
= za
.za_first_integer
;
2932 * Components of livelist deletion that must be performed in syncing
2933 * context: freeing block pointers and updating the pool-wide data
2934 * structures to indicate how much work is left to do
2936 typedef struct sublist_delete_arg
{
2941 } sublist_delete_arg_t
;
2944 sublist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2946 sublist_delete_arg_t
*sda
= arg
;
2947 spa_t
*spa
= sda
->spa
;
2948 dsl_deadlist_t
*ll
= sda
->ll
;
2949 uint64_t key
= sda
->key
;
2950 bplist_t
*to_free
= sda
->to_free
;
2952 bplist_iterate(to_free
, delete_blkptr_cb
, spa
, tx
);
2953 dsl_deadlist_remove_entry(ll
, key
, tx
);
2956 typedef struct livelist_delete_arg
{
2960 } livelist_delete_arg_t
;
2963 livelist_delete_sync(void *arg
, dmu_tx_t
*tx
)
2965 livelist_delete_arg_t
*lda
= arg
;
2966 spa_t
*spa
= lda
->spa
;
2967 uint64_t ll_obj
= lda
->ll_obj
;
2968 uint64_t zap_obj
= lda
->zap_obj
;
2969 objset_t
*mos
= spa
->spa_meta_objset
;
2972 /* free the livelist and decrement the feature count */
2973 VERIFY0(zap_remove_int(mos
, zap_obj
, ll_obj
, tx
));
2974 dsl_deadlist_free(mos
, ll_obj
, tx
);
2975 spa_feature_decr(spa
, SPA_FEATURE_LIVELIST
, tx
);
2976 VERIFY0(zap_count(mos
, zap_obj
, &count
));
2978 /* no more livelists to delete */
2979 VERIFY0(zap_remove(mos
, DMU_POOL_DIRECTORY_OBJECT
,
2980 DMU_POOL_DELETED_CLONES
, tx
));
2981 VERIFY0(zap_destroy(mos
, zap_obj
, tx
));
2982 spa
->spa_livelists_to_delete
= 0;
2983 spa_notify_waiters(spa
);
2988 * Load in the value for the livelist to be removed and open it. Then,
2989 * load its first sublist and determine which block pointers should actually
2990 * be freed. Then, call a synctask which performs the actual frees and updates
2991 * the pool-wide livelist data.
2994 spa_livelist_delete_cb(void *arg
, zthr_t
*z
)
2997 uint64_t ll_obj
= 0, count
;
2998 objset_t
*mos
= spa
->spa_meta_objset
;
2999 uint64_t zap_obj
= spa
->spa_livelists_to_delete
;
3001 * Determine the next livelist to delete. This function should only
3002 * be called if there is at least one deleted clone.
3004 VERIFY0(dsl_get_next_livelist_obj(mos
, zap_obj
, &ll_obj
));
3005 VERIFY0(zap_count(mos
, ll_obj
, &count
));
3008 dsl_deadlist_entry_t
*dle
;
3010 ll
= kmem_zalloc(sizeof (dsl_deadlist_t
), KM_SLEEP
);
3011 dsl_deadlist_open(ll
, mos
, ll_obj
);
3012 dle
= dsl_deadlist_first(ll
);
3013 ASSERT3P(dle
, !=, NULL
);
3014 bplist_create(&to_free
);
3015 int err
= dsl_process_sub_livelist(&dle
->dle_bpobj
, &to_free
,
3018 sublist_delete_arg_t sync_arg
= {
3021 .key
= dle
->dle_mintxg
,
3024 zfs_dbgmsg("deleting sublist (id %llu) from"
3025 " livelist %llu, %lld remaining",
3026 (u_longlong_t
)dle
->dle_bpobj
.bpo_object
,
3027 (u_longlong_t
)ll_obj
, (longlong_t
)count
- 1);
3028 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
,
3029 sublist_delete_sync
, &sync_arg
, 0,
3030 ZFS_SPACE_CHECK_DESTROY
));
3032 VERIFY3U(err
, ==, EINTR
);
3034 bplist_clear(&to_free
);
3035 bplist_destroy(&to_free
);
3036 dsl_deadlist_close(ll
);
3037 kmem_free(ll
, sizeof (dsl_deadlist_t
));
3039 livelist_delete_arg_t sync_arg
= {
3044 zfs_dbgmsg("deletion of livelist %llu completed",
3045 (u_longlong_t
)ll_obj
);
3046 VERIFY0(dsl_sync_task(spa_name(spa
), NULL
, livelist_delete_sync
,
3047 &sync_arg
, 0, ZFS_SPACE_CHECK_DESTROY
));
3052 spa_start_livelist_destroy_thread(spa_t
*spa
)
3054 ASSERT3P(spa
->spa_livelist_delete_zthr
, ==, NULL
);
3055 spa
->spa_livelist_delete_zthr
=
3056 zthr_create("z_livelist_destroy",
3057 spa_livelist_delete_cb_check
, spa_livelist_delete_cb
, spa
,
3061 typedef struct livelist_new_arg
{
3064 } livelist_new_arg_t
;
3067 livelist_track_new_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
3071 livelist_new_arg_t
*lna
= arg
;
3073 bplist_append(lna
->frees
, bp
);
3075 bplist_append(lna
->allocs
, bp
);
3076 zfs_livelist_condense_new_alloc
++;
3081 typedef struct livelist_condense_arg
{
3084 uint64_t first_size
;
3086 } livelist_condense_arg_t
;
3089 spa_livelist_condense_sync(void *arg
, dmu_tx_t
*tx
)
3091 livelist_condense_arg_t
*lca
= arg
;
3092 spa_t
*spa
= lca
->spa
;
3094 dsl_dataset_t
*ds
= spa
->spa_to_condense
.ds
;
3096 /* Have we been cancelled? */
3097 if (spa
->spa_to_condense
.cancelled
) {
3098 zfs_livelist_condense_sync_cancel
++;
3102 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
3103 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
3104 dsl_deadlist_t
*ll
= &ds
->ds_dir
->dd_livelist
;
3107 * It's possible that the livelist was changed while the zthr was
3108 * running. Therefore, we need to check for new blkptrs in the two
3109 * entries being condensed and continue to track them in the livelist.
3110 * Because of the way we handle remapped blkptrs (see dbuf_remap_impl),
3111 * it's possible that the newly added blkptrs are FREEs or ALLOCs so
3112 * we need to sort them into two different bplists.
3114 uint64_t first_obj
= first
->dle_bpobj
.bpo_object
;
3115 uint64_t next_obj
= next
->dle_bpobj
.bpo_object
;
3116 uint64_t cur_first_size
= first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
3117 uint64_t cur_next_size
= next
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
;
3119 bplist_create(&new_frees
);
3120 livelist_new_arg_t new_bps
= {
3121 .allocs
= &lca
->to_keep
,
3122 .frees
= &new_frees
,
3125 if (cur_first_size
> lca
->first_size
) {
3126 VERIFY0(livelist_bpobj_iterate_from_nofree(&first
->dle_bpobj
,
3127 livelist_track_new_cb
, &new_bps
, lca
->first_size
));
3129 if (cur_next_size
> lca
->next_size
) {
3130 VERIFY0(livelist_bpobj_iterate_from_nofree(&next
->dle_bpobj
,
3131 livelist_track_new_cb
, &new_bps
, lca
->next_size
));
3134 dsl_deadlist_clear_entry(first
, ll
, tx
);
3135 ASSERT(bpobj_is_empty(&first
->dle_bpobj
));
3136 dsl_deadlist_remove_entry(ll
, next
->dle_mintxg
, tx
);
3138 bplist_iterate(&lca
->to_keep
, dsl_deadlist_insert_alloc_cb
, ll
, tx
);
3139 bplist_iterate(&new_frees
, dsl_deadlist_insert_free_cb
, ll
, tx
);
3140 bplist_destroy(&new_frees
);
3142 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
3143 dsl_dataset_name(ds
, dsname
);
3144 zfs_dbgmsg("txg %llu condensing livelist of %s (id %llu), bpobj %llu "
3145 "(%llu blkptrs) and bpobj %llu (%llu blkptrs) -> bpobj %llu "
3146 "(%llu blkptrs)", (u_longlong_t
)tx
->tx_txg
, dsname
,
3147 (u_longlong_t
)ds
->ds_object
, (u_longlong_t
)first_obj
,
3148 (u_longlong_t
)cur_first_size
, (u_longlong_t
)next_obj
,
3149 (u_longlong_t
)cur_next_size
,
3150 (u_longlong_t
)first
->dle_bpobj
.bpo_object
,
3151 (u_longlong_t
)first
->dle_bpobj
.bpo_phys
->bpo_num_blkptrs
);
3153 dmu_buf_rele(ds
->ds_dbuf
, spa
);
3154 spa
->spa_to_condense
.ds
= NULL
;
3155 bplist_clear(&lca
->to_keep
);
3156 bplist_destroy(&lca
->to_keep
);
3157 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
3158 spa
->spa_to_condense
.syncing
= B_FALSE
;
3162 spa_livelist_condense_cb(void *arg
, zthr_t
*t
)
3164 while (zfs_livelist_condense_zthr_pause
&&
3165 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
3169 dsl_deadlist_entry_t
*first
= spa
->spa_to_condense
.first
;
3170 dsl_deadlist_entry_t
*next
= spa
->spa_to_condense
.next
;
3171 uint64_t first_size
, next_size
;
3173 livelist_condense_arg_t
*lca
=
3174 kmem_alloc(sizeof (livelist_condense_arg_t
), KM_SLEEP
);
3175 bplist_create(&lca
->to_keep
);
3178 * Process the livelists (matching FREEs and ALLOCs) in open context
3179 * so we have minimal work in syncing context to condense.
3181 * We save bpobj sizes (first_size and next_size) to use later in
3182 * syncing context to determine if entries were added to these sublists
3183 * while in open context. This is possible because the clone is still
3184 * active and open for normal writes and we want to make sure the new,
3185 * unprocessed blockpointers are inserted into the livelist normally.
3187 * Note that dsl_process_sub_livelist() both stores the size number of
3188 * blockpointers and iterates over them while the bpobj's lock held, so
3189 * the sizes returned to us are consistent which what was actually
3192 int err
= dsl_process_sub_livelist(&first
->dle_bpobj
, &lca
->to_keep
, t
,
3195 err
= dsl_process_sub_livelist(&next
->dle_bpobj
, &lca
->to_keep
,
3199 while (zfs_livelist_condense_sync_pause
&&
3200 !(zthr_has_waiters(t
) || zthr_iscancelled(t
)))
3203 dmu_tx_t
*tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
3204 dmu_tx_mark_netfree(tx
);
3205 dmu_tx_hold_space(tx
, 1);
3206 err
= dmu_tx_assign(tx
, TXG_NOWAIT
| TXG_NOTHROTTLE
);
3209 * Prevent the condense zthr restarting before
3210 * the synctask completes.
3212 spa
->spa_to_condense
.syncing
= B_TRUE
;
3214 lca
->first_size
= first_size
;
3215 lca
->next_size
= next_size
;
3216 dsl_sync_task_nowait(spa_get_dsl(spa
),
3217 spa_livelist_condense_sync
, lca
, tx
);
3223 * Condensing can not continue: either it was externally stopped or
3224 * we were unable to assign to a tx because the pool has run out of
3225 * space. In the second case, we'll just end up trying to condense
3226 * again in a later txg.
3229 bplist_clear(&lca
->to_keep
);
3230 bplist_destroy(&lca
->to_keep
);
3231 kmem_free(lca
, sizeof (livelist_condense_arg_t
));
3232 dmu_buf_rele(spa
->spa_to_condense
.ds
->ds_dbuf
, spa
);
3233 spa
->spa_to_condense
.ds
= NULL
;
3235 zfs_livelist_condense_zthr_cancel
++;
3239 * Check that there is something to condense but that a condense is not
3240 * already in progress and that condensing has not been cancelled.
3243 spa_livelist_condense_cb_check(void *arg
, zthr_t
*z
)
3247 if ((spa
->spa_to_condense
.ds
!= NULL
) &&
3248 (spa
->spa_to_condense
.syncing
== B_FALSE
) &&
3249 (spa
->spa_to_condense
.cancelled
== B_FALSE
)) {
3256 spa_start_livelist_condensing_thread(spa_t
*spa
)
3258 spa
->spa_to_condense
.ds
= NULL
;
3259 spa
->spa_to_condense
.first
= NULL
;
3260 spa
->spa_to_condense
.next
= NULL
;
3261 spa
->spa_to_condense
.syncing
= B_FALSE
;
3262 spa
->spa_to_condense
.cancelled
= B_FALSE
;
3264 ASSERT3P(spa
->spa_livelist_condense_zthr
, ==, NULL
);
3265 spa
->spa_livelist_condense_zthr
=
3266 zthr_create("z_livelist_condense",
3267 spa_livelist_condense_cb_check
,
3268 spa_livelist_condense_cb
, spa
, minclsyspri
);
3272 spa_spawn_aux_threads(spa_t
*spa
)
3274 ASSERT(spa_writeable(spa
));
3276 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
3278 spa_start_raidz_expansion_thread(spa
);
3279 spa_start_indirect_condensing_thread(spa
);
3280 spa_start_livelist_destroy_thread(spa
);
3281 spa_start_livelist_condensing_thread(spa
);
3283 ASSERT3P(spa
->spa_checkpoint_discard_zthr
, ==, NULL
);
3284 spa
->spa_checkpoint_discard_zthr
=
3285 zthr_create("z_checkpoint_discard",
3286 spa_checkpoint_discard_thread_check
,
3287 spa_checkpoint_discard_thread
, spa
, minclsyspri
);
3291 * Fix up config after a partly-completed split. This is done with the
3292 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
3293 * pool have that entry in their config, but only the splitting one contains
3294 * a list of all the guids of the vdevs that are being split off.
3296 * This function determines what to do with that list: either rejoin
3297 * all the disks to the pool, or complete the splitting process. To attempt
3298 * the rejoin, each disk that is offlined is marked online again, and
3299 * we do a reopen() call. If the vdev label for every disk that was
3300 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
3301 * then we call vdev_split() on each disk, and complete the split.
3303 * Otherwise we leave the config alone, with all the vdevs in place in
3304 * the original pool.
3307 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
3314 boolean_t attempt_reopen
;
3316 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
3319 /* check that the config is complete */
3320 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
3321 &glist
, &gcount
) != 0)
3324 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
3326 /* attempt to online all the vdevs & validate */
3327 attempt_reopen
= B_TRUE
;
3328 for (i
= 0; i
< gcount
; i
++) {
3329 if (glist
[i
] == 0) /* vdev is hole */
3332 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
3333 if (vd
[i
] == NULL
) {
3335 * Don't bother attempting to reopen the disks;
3336 * just do the split.
3338 attempt_reopen
= B_FALSE
;
3340 /* attempt to re-online it */
3341 vd
[i
]->vdev_offline
= B_FALSE
;
3345 if (attempt_reopen
) {
3346 vdev_reopen(spa
->spa_root_vdev
);
3348 /* check each device to see what state it's in */
3349 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
3350 if (vd
[i
] != NULL
&&
3351 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
3358 * If every disk has been moved to the new pool, or if we never
3359 * even attempted to look at them, then we split them off for
3362 if (!attempt_reopen
|| gcount
== extracted
) {
3363 for (i
= 0; i
< gcount
; i
++)
3366 vdev_reopen(spa
->spa_root_vdev
);
3369 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
3373 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
)
3375 const char *ereport
= FM_EREPORT_ZFS_POOL
;
3378 spa
->spa_load_state
= state
;
3379 (void) spa_import_progress_set_state(spa_guid(spa
),
3380 spa_load_state(spa
));
3381 spa_import_progress_set_notes(spa
, "spa_load()");
3383 gethrestime(&spa
->spa_loaded_ts
);
3384 error
= spa_load_impl(spa
, type
, &ereport
);
3387 * Don't count references from objsets that are already closed
3388 * and are making their way through the eviction process.
3390 spa_evicting_os_wait(spa
);
3391 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
3393 if (error
!= EEXIST
) {
3394 spa
->spa_loaded_ts
.tv_sec
= 0;
3395 spa
->spa_loaded_ts
.tv_nsec
= 0;
3397 if (error
!= EBADF
) {
3398 (void) zfs_ereport_post(ereport
, spa
,
3399 NULL
, NULL
, NULL
, 0);
3402 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
3405 (void) spa_import_progress_set_state(spa_guid(spa
),
3406 spa_load_state(spa
));
3413 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
3414 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
3415 * spa's per-vdev ZAP list.
3418 vdev_count_verify_zaps(vdev_t
*vd
)
3420 spa_t
*spa
= vd
->vdev_spa
;
3423 if (spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_AVZ_V2
) &&
3424 vd
->vdev_root_zap
!= 0) {
3426 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3427 spa
->spa_all_vdev_zaps
, vd
->vdev_root_zap
));
3429 if (vd
->vdev_top_zap
!= 0) {
3431 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3432 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
3434 if (vd
->vdev_leaf_zap
!= 0) {
3436 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
3437 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
3440 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3441 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
3447 #define vdev_count_verify_zaps(vd) ((void) sizeof (vd), 0)
3451 * Determine whether the activity check is required.
3454 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
3458 uint64_t hostid
= 0;
3459 uint64_t tryconfig_txg
= 0;
3460 uint64_t tryconfig_timestamp
= 0;
3461 uint16_t tryconfig_mmp_seq
= 0;
3464 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3465 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
3466 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
3468 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
3469 &tryconfig_timestamp
);
3470 (void) nvlist_lookup_uint16(nvinfo
, ZPOOL_CONFIG_MMP_SEQ
,
3471 &tryconfig_mmp_seq
);
3474 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
3477 * Disable the MMP activity check - This is used by zdb which
3478 * is intended to be used on potentially active pools.
3480 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
3484 * Skip the activity check when the MMP feature is disabled.
3486 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
3490 * If the tryconfig_ values are nonzero, they are the results of an
3491 * earlier tryimport. If they all match the uberblock we just found,
3492 * then the pool has not changed and we return false so we do not test
3495 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
3496 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
&&
3497 tryconfig_mmp_seq
&& tryconfig_mmp_seq
==
3498 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0))
3502 * Allow the activity check to be skipped when importing the pool
3503 * on the same host which last imported it. Since the hostid from
3504 * configuration may be stale use the one read from the label.
3506 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
3507 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
3509 if (hostid
== spa_get_hostid(spa
))
3513 * Skip the activity test when the pool was cleanly exported.
3515 if (state
!= POOL_STATE_ACTIVE
)
3522 * Nanoseconds the activity check must watch for changes on-disk.
3525 spa_activity_check_duration(spa_t
*spa
, uberblock_t
*ub
)
3527 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
3528 uint64_t multihost_interval
= MSEC2NSEC(
3529 MMP_INTERVAL_OK(zfs_multihost_interval
));
3530 uint64_t import_delay
= MAX(NANOSEC
, import_intervals
*
3531 multihost_interval
);
3534 * Local tunables determine a minimum duration except for the case
3535 * where we know when the remote host will suspend the pool if MMP
3536 * writes do not land.
3538 * See Big Theory comment at the top of mmp.c for the reasoning behind
3539 * these cases and times.
3542 ASSERT(MMP_IMPORT_SAFETY_FACTOR
>= 100);
3544 if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3545 MMP_FAIL_INT(ub
) > 0) {
3547 /* MMP on remote host will suspend pool after failed writes */
3548 import_delay
= MMP_FAIL_INT(ub
) * MSEC2NSEC(MMP_INTERVAL(ub
)) *
3549 MMP_IMPORT_SAFETY_FACTOR
/ 100;
3551 zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
3552 "mmp_fails=%llu ub_mmp mmp_interval=%llu "
3553 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3554 (u_longlong_t
)MMP_FAIL_INT(ub
),
3555 (u_longlong_t
)MMP_INTERVAL(ub
),
3556 (u_longlong_t
)import_intervals
);
3558 } else if (MMP_INTERVAL_VALID(ub
) && MMP_FAIL_INT_VALID(ub
) &&
3559 MMP_FAIL_INT(ub
) == 0) {
3561 /* MMP on remote host will never suspend pool */
3562 import_delay
= MAX(import_delay
, (MSEC2NSEC(MMP_INTERVAL(ub
)) +
3563 ub
->ub_mmp_delay
) * import_intervals
);
3565 zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
3566 "mmp_interval=%llu ub_mmp_delay=%llu "
3567 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3568 (u_longlong_t
)MMP_INTERVAL(ub
),
3569 (u_longlong_t
)ub
->ub_mmp_delay
,
3570 (u_longlong_t
)import_intervals
);
3572 } else if (MMP_VALID(ub
)) {
3574 * zfs-0.7 compatibility case
3577 import_delay
= MAX(import_delay
, (multihost_interval
+
3578 ub
->ub_mmp_delay
) * import_intervals
);
3580 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
3581 "import_intervals=%llu leaves=%u",
3582 (u_longlong_t
)import_delay
,
3583 (u_longlong_t
)ub
->ub_mmp_delay
,
3584 (u_longlong_t
)import_intervals
,
3585 vdev_count_leaves(spa
));
3587 /* Using local tunings is the only reasonable option */
3588 zfs_dbgmsg("pool last imported on non-MMP aware "
3589 "host using import_delay=%llu multihost_interval=%llu "
3590 "import_intervals=%llu", (u_longlong_t
)import_delay
,
3591 (u_longlong_t
)multihost_interval
,
3592 (u_longlong_t
)import_intervals
);
3595 return (import_delay
);
3599 * Perform the import activity check. If the user canceled the import or
3600 * we detected activity then fail.
3603 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
3605 uint64_t txg
= ub
->ub_txg
;
3606 uint64_t timestamp
= ub
->ub_timestamp
;
3607 uint64_t mmp_config
= ub
->ub_mmp_config
;
3608 uint16_t mmp_seq
= MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0;
3609 uint64_t import_delay
;
3610 hrtime_t import_expire
, now
;
3611 nvlist_t
*mmp_label
= NULL
;
3612 vdev_t
*rvd
= spa
->spa_root_vdev
;
3617 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
3618 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
3622 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
3623 * during the earlier tryimport. If the txg recorded there is 0 then
3624 * the pool is known to be active on another host.
3626 * Otherwise, the pool might be in use on another host. Check for
3627 * changes in the uberblocks on disk if necessary.
3629 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
3630 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
3631 ZPOOL_CONFIG_LOAD_INFO
);
3633 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
3634 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
3635 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3636 error
= SET_ERROR(EREMOTEIO
);
3641 import_delay
= spa_activity_check_duration(spa
, ub
);
3643 /* Add a small random factor in case of simultaneous imports (0-25%) */
3644 import_delay
+= import_delay
* random_in_range(250) / 1000;
3646 import_expire
= gethrtime() + import_delay
;
3648 spa_import_progress_set_notes(spa
, "Checking MMP activity, waiting "
3649 "%llu ms", (u_longlong_t
)NSEC2MSEC(import_delay
));
3651 int interations
= 0;
3652 while ((now
= gethrtime()) < import_expire
) {
3653 if (interations
++ % 30 == 0) {
3654 spa_import_progress_set_notes(spa
, "Checking MMP "
3655 "activity, %llu ms remaining",
3656 (u_longlong_t
)NSEC2MSEC(import_expire
- now
));
3659 (void) spa_import_progress_set_mmp_check(spa_guid(spa
),
3660 NSEC2SEC(import_expire
- gethrtime()));
3662 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
3664 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
||
3665 mmp_seq
!= (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0)) {
3666 zfs_dbgmsg("multihost activity detected "
3667 "txg %llu ub_txg %llu "
3668 "timestamp %llu ub_timestamp %llu "
3669 "mmp_config %#llx ub_mmp_config %#llx",
3670 (u_longlong_t
)txg
, (u_longlong_t
)ub
->ub_txg
,
3671 (u_longlong_t
)timestamp
,
3672 (u_longlong_t
)ub
->ub_timestamp
,
3673 (u_longlong_t
)mmp_config
,
3674 (u_longlong_t
)ub
->ub_mmp_config
);
3676 error
= SET_ERROR(EREMOTEIO
);
3681 nvlist_free(mmp_label
);
3685 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
3687 error
= SET_ERROR(EINTR
);
3695 mutex_destroy(&mtx
);
3699 * If the pool is determined to be active store the status in the
3700 * spa->spa_load_info nvlist. If the remote hostname or hostid are
3701 * available from configuration read from disk store them as well.
3702 * This allows 'zpool import' to generate a more useful message.
3704 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
3705 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
3706 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
3708 if (error
== EREMOTEIO
) {
3709 const char *hostname
= "<unknown>";
3710 uint64_t hostid
= 0;
3713 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
3714 hostname
= fnvlist_lookup_string(mmp_label
,
3715 ZPOOL_CONFIG_HOSTNAME
);
3716 fnvlist_add_string(spa
->spa_load_info
,
3717 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
3720 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
3721 hostid
= fnvlist_lookup_uint64(mmp_label
,
3722 ZPOOL_CONFIG_HOSTID
);
3723 fnvlist_add_uint64(spa
->spa_load_info
,
3724 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
3728 fnvlist_add_uint64(spa
->spa_load_info
,
3729 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
3730 fnvlist_add_uint64(spa
->spa_load_info
,
3731 ZPOOL_CONFIG_MMP_TXG
, 0);
3733 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
3737 nvlist_free(mmp_label
);
3743 spa_verify_host(spa_t
*spa
, nvlist_t
*mos_config
)
3746 const char *hostname
;
3747 uint64_t myhostid
= 0;
3749 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
3750 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
3751 hostname
= fnvlist_lookup_string(mos_config
,
3752 ZPOOL_CONFIG_HOSTNAME
);
3754 myhostid
= zone_get_hostid(NULL
);
3756 if (hostid
!= 0 && myhostid
!= 0 && hostid
!= myhostid
) {
3757 cmn_err(CE_WARN
, "pool '%s' could not be "
3758 "loaded as it was last accessed by "
3759 "another system (host: %s hostid: 0x%llx). "
3760 "See: https://openzfs.github.io/openzfs-docs/msg/"
3762 spa_name(spa
), hostname
, (u_longlong_t
)hostid
);
3763 spa_load_failed(spa
, "hostid verification failed: pool "
3764 "last accessed by host: %s (hostid: 0x%llx)",
3765 hostname
, (u_longlong_t
)hostid
);
3766 return (SET_ERROR(EBADF
));
3774 spa_ld_parse_config(spa_t
*spa
, spa_import_type_t type
)
3777 nvlist_t
*nvtree
, *nvl
, *config
= spa
->spa_config
;
3781 const char *comment
;
3782 const char *compatibility
;
3785 * Versioning wasn't explicitly added to the label until later, so if
3786 * it's not present treat it as the initial version.
3788 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
3789 &spa
->spa_ubsync
.ub_version
) != 0)
3790 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
3792 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
3793 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3794 ZPOOL_CONFIG_POOL_GUID
);
3795 return (SET_ERROR(EINVAL
));
3799 * If we are doing an import, ensure that the pool is not already
3800 * imported by checking if its pool guid already exists in the
3803 * The only case that we allow an already imported pool to be
3804 * imported again, is when the pool is checkpointed and we want to
3805 * look at its checkpointed state from userland tools like zdb.
3808 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3809 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3810 spa_guid_exists(pool_guid
, 0)) {
3812 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3813 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) &&
3814 spa_guid_exists(pool_guid
, 0) &&
3815 !spa_importing_readonly_checkpoint(spa
)) {
3817 spa_load_failed(spa
, "a pool with guid %llu is already open",
3818 (u_longlong_t
)pool_guid
);
3819 return (SET_ERROR(EEXIST
));
3822 spa
->spa_config_guid
= pool_guid
;
3824 nvlist_free(spa
->spa_load_info
);
3825 spa
->spa_load_info
= fnvlist_alloc();
3827 ASSERT(spa
->spa_comment
== NULL
);
3828 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
3829 spa
->spa_comment
= spa_strdup(comment
);
3831 ASSERT(spa
->spa_compatibility
== NULL
);
3832 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMPATIBILITY
,
3833 &compatibility
) == 0)
3834 spa
->spa_compatibility
= spa_strdup(compatibility
);
3836 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
3837 &spa
->spa_config_txg
);
3839 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) == 0)
3840 spa
->spa_config_splitting
= fnvlist_dup(nvl
);
3842 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
)) {
3843 spa_load_failed(spa
, "invalid config provided: '%s' missing",
3844 ZPOOL_CONFIG_VDEV_TREE
);
3845 return (SET_ERROR(EINVAL
));
3849 * Create "The Godfather" zio to hold all async IOs
3851 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3853 for (int i
= 0; i
< max_ncpus
; i
++) {
3854 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3855 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3856 ZIO_FLAG_GODFATHER
);
3860 * Parse the configuration into a vdev tree. We explicitly set the
3861 * value that will be returned by spa_version() since parsing the
3862 * configuration requires knowing the version number.
3864 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3865 parse
= (type
== SPA_IMPORT_EXISTING
?
3866 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
3867 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
3868 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3871 spa_load_failed(spa
, "unable to parse config [error=%d]",
3876 ASSERT(spa
->spa_root_vdev
== rvd
);
3877 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
3878 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
3880 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3881 ASSERT(spa_guid(spa
) == pool_guid
);
3888 * Recursively open all vdevs in the vdev tree. This function is called twice:
3889 * first with the untrusted config, then with the trusted config.
3892 spa_ld_open_vdevs(spa_t
*spa
)
3897 * spa_missing_tvds_allowed defines how many top-level vdevs can be
3898 * missing/unopenable for the root vdev to be still considered openable.
3900 if (spa
->spa_trust_config
) {
3901 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds
;
3902 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_CACHEFILE
) {
3903 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_cachefile
;
3904 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_SCAN
) {
3905 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_scan
;
3907 spa
->spa_missing_tvds_allowed
= 0;
3910 spa
->spa_missing_tvds_allowed
=
3911 MAX(zfs_max_missing_tvds
, spa
->spa_missing_tvds_allowed
);
3913 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3914 error
= vdev_open(spa
->spa_root_vdev
);
3915 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3917 if (spa
->spa_missing_tvds
!= 0) {
3918 spa_load_note(spa
, "vdev tree has %lld missing top-level "
3919 "vdevs.", (u_longlong_t
)spa
->spa_missing_tvds
);
3920 if (spa
->spa_trust_config
&& (spa
->spa_mode
& SPA_MODE_WRITE
)) {
3922 * Although theoretically we could allow users to open
3923 * incomplete pools in RW mode, we'd need to add a lot
3924 * of extra logic (e.g. adjust pool space to account
3925 * for missing vdevs).
3926 * This limitation also prevents users from accidentally
3927 * opening the pool in RW mode during data recovery and
3928 * damaging it further.
3930 spa_load_note(spa
, "pools with missing top-level "
3931 "vdevs can only be opened in read-only mode.");
3932 error
= SET_ERROR(ENXIO
);
3934 spa_load_note(spa
, "current settings allow for maximum "
3935 "%lld missing top-level vdevs at this stage.",
3936 (u_longlong_t
)spa
->spa_missing_tvds_allowed
);
3940 spa_load_failed(spa
, "unable to open vdev tree [error=%d]",
3943 if (spa
->spa_missing_tvds
!= 0 || error
!= 0)
3944 vdev_dbgmsg_print_tree(spa
->spa_root_vdev
, 2);
3950 * We need to validate the vdev labels against the configuration that
3951 * we have in hand. This function is called twice: first with an untrusted
3952 * config, then with a trusted config. The validation is more strict when the
3953 * config is trusted.
3956 spa_ld_validate_vdevs(spa_t
*spa
)
3959 vdev_t
*rvd
= spa
->spa_root_vdev
;
3961 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3962 error
= vdev_validate(rvd
);
3963 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3966 spa_load_failed(spa
, "vdev_validate failed [error=%d]", error
);
3970 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
3971 spa_load_failed(spa
, "cannot open vdev tree after invalidating "
3973 vdev_dbgmsg_print_tree(rvd
, 2);
3974 return (SET_ERROR(ENXIO
));
3981 spa_ld_select_uberblock_done(spa_t
*spa
, uberblock_t
*ub
)
3983 spa
->spa_state
= POOL_STATE_ACTIVE
;
3984 spa
->spa_ubsync
= spa
->spa_uberblock
;
3985 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
3986 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
3987 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
3988 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
3989 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
3990 spa
->spa_prev_software_version
= ub
->ub_software_version
;
3994 spa_ld_select_uberblock(spa_t
*spa
, spa_import_type_t type
)
3996 vdev_t
*rvd
= spa
->spa_root_vdev
;
3998 uberblock_t
*ub
= &spa
->spa_uberblock
;
3999 boolean_t activity_check
= B_FALSE
;
4002 * If we are opening the checkpointed state of the pool by
4003 * rewinding to it, at this point we will have written the
4004 * checkpointed uberblock to the vdev labels, so searching
4005 * the labels will find the right uberblock. However, if
4006 * we are opening the checkpointed state read-only, we have
4007 * not modified the labels. Therefore, we must ignore the
4008 * labels and continue using the spa_uberblock that was set
4009 * by spa_ld_checkpoint_rewind.
4011 * Note that it would be fine to ignore the labels when
4012 * rewinding (opening writeable) as well. However, if we
4013 * crash just after writing the labels, we will end up
4014 * searching the labels. Doing so in the common case means
4015 * that this code path gets exercised normally, rather than
4016 * just in the edge case.
4018 if (ub
->ub_checkpoint_txg
!= 0 &&
4019 spa_importing_readonly_checkpoint(spa
)) {
4020 spa_ld_select_uberblock_done(spa
, ub
);
4025 * Find the best uberblock.
4027 vdev_uberblock_load(rvd
, ub
, &label
);
4030 * If we weren't able to find a single valid uberblock, return failure.
4032 if (ub
->ub_txg
== 0) {
4034 spa_load_failed(spa
, "no valid uberblock found");
4035 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
4038 if (spa
->spa_load_max_txg
!= UINT64_MAX
) {
4039 (void) spa_import_progress_set_max_txg(spa_guid(spa
),
4040 (u_longlong_t
)spa
->spa_load_max_txg
);
4042 spa_load_note(spa
, "using uberblock with txg=%llu",
4043 (u_longlong_t
)ub
->ub_txg
);
4044 if (ub
->ub_raidz_reflow_info
!= 0) {
4045 spa_load_note(spa
, "uberblock raidz_reflow_info: "
4046 "state=%u offset=%llu",
4047 (int)RRSS_GET_STATE(ub
),
4048 (u_longlong_t
)RRSS_GET_OFFSET(ub
));
4053 * For pools which have the multihost property on determine if the
4054 * pool is truly inactive and can be safely imported. Prevent
4055 * hosts which don't have a hostid set from importing the pool.
4057 activity_check
= spa_activity_check_required(spa
, ub
, label
,
4059 if (activity_check
) {
4060 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
4061 spa_get_hostid(spa
) == 0) {
4063 fnvlist_add_uint64(spa
->spa_load_info
,
4064 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
4065 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
4068 int error
= spa_activity_check(spa
, ub
, spa
->spa_config
);
4074 fnvlist_add_uint64(spa
->spa_load_info
,
4075 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
4076 fnvlist_add_uint64(spa
->spa_load_info
,
4077 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
4078 fnvlist_add_uint16(spa
->spa_load_info
,
4079 ZPOOL_CONFIG_MMP_SEQ
,
4080 (MMP_SEQ_VALID(ub
) ? MMP_SEQ(ub
) : 0));
4084 * If the pool has an unsupported version we can't open it.
4086 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
4088 spa_load_failed(spa
, "version %llu is not supported",
4089 (u_longlong_t
)ub
->ub_version
);
4090 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
4093 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
4097 * If we weren't able to find what's necessary for reading the
4098 * MOS in the label, return failure.
4100 if (label
== NULL
) {
4101 spa_load_failed(spa
, "label config unavailable");
4102 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
4106 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
4109 spa_load_failed(spa
, "invalid label: '%s' missing",
4110 ZPOOL_CONFIG_FEATURES_FOR_READ
);
4111 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
4116 * Update our in-core representation with the definitive values
4119 nvlist_free(spa
->spa_label_features
);
4120 spa
->spa_label_features
= fnvlist_dup(features
);
4126 * Look through entries in the label nvlist's features_for_read. If
4127 * there is a feature listed there which we don't understand then we
4128 * cannot open a pool.
4130 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
4131 nvlist_t
*unsup_feat
;
4133 unsup_feat
= fnvlist_alloc();
4135 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
4137 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
4138 if (!zfeature_is_supported(nvpair_name(nvp
))) {
4139 fnvlist_add_string(unsup_feat
,
4140 nvpair_name(nvp
), "");
4144 if (!nvlist_empty(unsup_feat
)) {
4145 fnvlist_add_nvlist(spa
->spa_load_info
,
4146 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
4147 nvlist_free(unsup_feat
);
4148 spa_load_failed(spa
, "some features are unsupported");
4149 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
4153 nvlist_free(unsup_feat
);
4156 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
4157 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4158 spa_try_repair(spa
, spa
->spa_config
);
4159 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4160 nvlist_free(spa
->spa_config_splitting
);
4161 spa
->spa_config_splitting
= NULL
;
4165 * Initialize internal SPA structures.
4167 spa_ld_select_uberblock_done(spa
, ub
);
4173 spa_ld_open_rootbp(spa_t
*spa
)
4176 vdev_t
*rvd
= spa
->spa_root_vdev
;
4178 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
4180 spa_load_failed(spa
, "unable to open rootbp in dsl_pool_init "
4181 "[error=%d]", error
);
4182 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4184 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
4190 spa_ld_trusted_config(spa_t
*spa
, spa_import_type_t type
,
4191 boolean_t reloading
)
4193 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
4194 nvlist_t
*nv
, *mos_config
, *policy
;
4195 int error
= 0, copy_error
;
4196 uint64_t healthy_tvds
, healthy_tvds_mos
;
4197 uint64_t mos_config_txg
;
4199 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
, B_TRUE
)
4201 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4204 * If we're assembling a pool from a split, the config provided is
4205 * already trusted so there is nothing to do.
4207 if (type
== SPA_IMPORT_ASSEMBLE
)
4210 healthy_tvds
= spa_healthy_core_tvds(spa
);
4212 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
)
4214 spa_load_failed(spa
, "unable to retrieve MOS config");
4215 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4219 * If we are doing an open, pool owner wasn't verified yet, thus do
4220 * the verification here.
4222 if (spa
->spa_load_state
== SPA_LOAD_OPEN
) {
4223 error
= spa_verify_host(spa
, mos_config
);
4225 nvlist_free(mos_config
);
4230 nv
= fnvlist_lookup_nvlist(mos_config
, ZPOOL_CONFIG_VDEV_TREE
);
4232 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4235 * Build a new vdev tree from the trusted config
4237 error
= spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
);
4239 nvlist_free(mos_config
);
4240 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4241 spa_load_failed(spa
, "spa_config_parse failed [error=%d]",
4243 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4247 * Vdev paths in the MOS may be obsolete. If the untrusted config was
4248 * obtained by scanning /dev/dsk, then it will have the right vdev
4249 * paths. We update the trusted MOS config with this information.
4250 * We first try to copy the paths with vdev_copy_path_strict, which
4251 * succeeds only when both configs have exactly the same vdev tree.
4252 * If that fails, we fall back to a more flexible method that has a
4253 * best effort policy.
4255 copy_error
= vdev_copy_path_strict(rvd
, mrvd
);
4256 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
4257 spa_load_note(spa
, "provided vdev tree:");
4258 vdev_dbgmsg_print_tree(rvd
, 2);
4259 spa_load_note(spa
, "MOS vdev tree:");
4260 vdev_dbgmsg_print_tree(mrvd
, 2);
4262 if (copy_error
!= 0) {
4263 spa_load_note(spa
, "vdev_copy_path_strict failed, falling "
4264 "back to vdev_copy_path_relaxed");
4265 vdev_copy_path_relaxed(rvd
, mrvd
);
4270 spa
->spa_root_vdev
= mrvd
;
4272 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4275 * If 'zpool import' used a cached config, then the on-disk hostid and
4276 * hostname may be different to the cached config in ways that should
4277 * prevent import. Userspace can't discover this without a scan, but
4278 * we know, so we add these values to LOAD_INFO so the caller can know
4281 * Note that we have to do this before the config is regenerated,
4282 * because the new config will have the hostid and hostname for this
4283 * host, in readiness for import.
4285 if (nvlist_exists(mos_config
, ZPOOL_CONFIG_HOSTID
))
4286 fnvlist_add_uint64(spa
->spa_load_info
, ZPOOL_CONFIG_HOSTID
,
4287 fnvlist_lookup_uint64(mos_config
, ZPOOL_CONFIG_HOSTID
));
4288 if (nvlist_exists(mos_config
, ZPOOL_CONFIG_HOSTNAME
))
4289 fnvlist_add_string(spa
->spa_load_info
, ZPOOL_CONFIG_HOSTNAME
,
4290 fnvlist_lookup_string(mos_config
, ZPOOL_CONFIG_HOSTNAME
));
4293 * We will use spa_config if we decide to reload the spa or if spa_load
4294 * fails and we rewind. We must thus regenerate the config using the
4295 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
4296 * pass settings on how to load the pool and is not stored in the MOS.
4297 * We copy it over to our new, trusted config.
4299 mos_config_txg
= fnvlist_lookup_uint64(mos_config
,
4300 ZPOOL_CONFIG_POOL_TXG
);
4301 nvlist_free(mos_config
);
4302 mos_config
= spa_config_generate(spa
, NULL
, mos_config_txg
, B_FALSE
);
4303 if (nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_LOAD_POLICY
,
4305 fnvlist_add_nvlist(mos_config
, ZPOOL_LOAD_POLICY
, policy
);
4306 spa_config_set(spa
, mos_config
);
4307 spa
->spa_config_source
= SPA_CONFIG_SRC_MOS
;
4310 * Now that we got the config from the MOS, we should be more strict
4311 * in checking blkptrs and can make assumptions about the consistency
4312 * of the vdev tree. spa_trust_config must be set to true before opening
4313 * vdevs in order for them to be writeable.
4315 spa
->spa_trust_config
= B_TRUE
;
4318 * Open and validate the new vdev tree
4320 error
= spa_ld_open_vdevs(spa
);
4324 error
= spa_ld_validate_vdevs(spa
);
4328 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
4329 spa_load_note(spa
, "final vdev tree:");
4330 vdev_dbgmsg_print_tree(rvd
, 2);
4333 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
&&
4334 !spa
->spa_extreme_rewind
&& zfs_max_missing_tvds
== 0) {
4336 * Sanity check to make sure that we are indeed loading the
4337 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
4338 * in the config provided and they happened to be the only ones
4339 * to have the latest uberblock, we could involuntarily perform
4340 * an extreme rewind.
4342 healthy_tvds_mos
= spa_healthy_core_tvds(spa
);
4343 if (healthy_tvds_mos
- healthy_tvds
>=
4344 SPA_SYNC_MIN_VDEVS
) {
4345 spa_load_note(spa
, "config provided misses too many "
4346 "top-level vdevs compared to MOS (%lld vs %lld). ",
4347 (u_longlong_t
)healthy_tvds
,
4348 (u_longlong_t
)healthy_tvds_mos
);
4349 spa_load_note(spa
, "vdev tree:");
4350 vdev_dbgmsg_print_tree(rvd
, 2);
4352 spa_load_failed(spa
, "config was already "
4353 "provided from MOS. Aborting.");
4354 return (spa_vdev_err(rvd
,
4355 VDEV_AUX_CORRUPT_DATA
, EIO
));
4357 spa_load_note(spa
, "spa must be reloaded using MOS "
4359 return (SET_ERROR(EAGAIN
));
4363 error
= spa_check_for_missing_logs(spa
);
4365 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
4367 if (rvd
->vdev_guid_sum
!= spa
->spa_uberblock
.ub_guid_sum
) {
4368 spa_load_failed(spa
, "uberblock guid sum doesn't match MOS "
4369 "guid sum (%llu != %llu)",
4370 (u_longlong_t
)spa
->spa_uberblock
.ub_guid_sum
,
4371 (u_longlong_t
)rvd
->vdev_guid_sum
);
4372 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
4380 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
4383 vdev_t
*rvd
= spa
->spa_root_vdev
;
4386 * Everything that we read before spa_remove_init() must be stored
4387 * on concreted vdevs. Therefore we do this as early as possible.
4389 error
= spa_remove_init(spa
);
4391 spa_load_failed(spa
, "spa_remove_init failed [error=%d]",
4393 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4397 * Retrieve information needed to condense indirect vdev mappings.
4399 error
= spa_condense_init(spa
);
4401 spa_load_failed(spa
, "spa_condense_init failed [error=%d]",
4403 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4410 spa_ld_check_features(spa_t
*spa
, boolean_t
*missing_feat_writep
)
4413 vdev_t
*rvd
= spa
->spa_root_vdev
;
4415 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
4416 boolean_t missing_feat_read
= B_FALSE
;
4417 nvlist_t
*unsup_feat
, *enabled_feat
;
4419 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
4420 &spa
->spa_feat_for_read_obj
, B_TRUE
) != 0) {
4421 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4424 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
4425 &spa
->spa_feat_for_write_obj
, B_TRUE
) != 0) {
4426 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4429 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
4430 &spa
->spa_feat_desc_obj
, B_TRUE
) != 0) {
4431 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4434 enabled_feat
= fnvlist_alloc();
4435 unsup_feat
= fnvlist_alloc();
4437 if (!spa_features_check(spa
, B_FALSE
,
4438 unsup_feat
, enabled_feat
))
4439 missing_feat_read
= B_TRUE
;
4441 if (spa_writeable(spa
) ||
4442 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) {
4443 if (!spa_features_check(spa
, B_TRUE
,
4444 unsup_feat
, enabled_feat
)) {
4445 *missing_feat_writep
= B_TRUE
;
4449 fnvlist_add_nvlist(spa
->spa_load_info
,
4450 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
4452 if (!nvlist_empty(unsup_feat
)) {
4453 fnvlist_add_nvlist(spa
->spa_load_info
,
4454 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
4457 fnvlist_free(enabled_feat
);
4458 fnvlist_free(unsup_feat
);
4460 if (!missing_feat_read
) {
4461 fnvlist_add_boolean(spa
->spa_load_info
,
4462 ZPOOL_CONFIG_CAN_RDONLY
);
4466 * If the state is SPA_LOAD_TRYIMPORT, our objective is
4467 * twofold: to determine whether the pool is available for
4468 * import in read-write mode and (if it is not) whether the
4469 * pool is available for import in read-only mode. If the pool
4470 * is available for import in read-write mode, it is displayed
4471 * as available in userland; if it is not available for import
4472 * in read-only mode, it is displayed as unavailable in
4473 * userland. If the pool is available for import in read-only
4474 * mode but not read-write mode, it is displayed as unavailable
4475 * in userland with a special note that the pool is actually
4476 * available for open in read-only mode.
4478 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
4479 * missing a feature for write, we must first determine whether
4480 * the pool can be opened read-only before returning to
4481 * userland in order to know whether to display the
4482 * abovementioned note.
4484 if (missing_feat_read
|| (*missing_feat_writep
&&
4485 spa_writeable(spa
))) {
4486 spa_load_failed(spa
, "pool uses unsupported features");
4487 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
4492 * Load refcounts for ZFS features from disk into an in-memory
4493 * cache during SPA initialization.
4495 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
4498 error
= feature_get_refcount_from_disk(spa
,
4499 &spa_feature_table
[i
], &refcount
);
4501 spa
->spa_feat_refcount_cache
[i
] = refcount
;
4502 } else if (error
== ENOTSUP
) {
4503 spa
->spa_feat_refcount_cache
[i
] =
4504 SPA_FEATURE_DISABLED
;
4506 spa_load_failed(spa
, "error getting refcount "
4507 "for feature %s [error=%d]",
4508 spa_feature_table
[i
].fi_guid
, error
);
4509 return (spa_vdev_err(rvd
,
4510 VDEV_AUX_CORRUPT_DATA
, EIO
));
4515 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
4516 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
4517 &spa
->spa_feat_enabled_txg_obj
, B_TRUE
) != 0)
4518 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4522 * Encryption was added before bookmark_v2, even though bookmark_v2
4523 * is now a dependency. If this pool has encryption enabled without
4524 * bookmark_v2, trigger an errata message.
4526 if (spa_feature_is_enabled(spa
, SPA_FEATURE_ENCRYPTION
) &&
4527 !spa_feature_is_enabled(spa
, SPA_FEATURE_BOOKMARK_V2
)) {
4528 spa
->spa_errata
= ZPOOL_ERRATA_ZOL_8308_ENCRYPTION
;
4535 spa_ld_load_special_directories(spa_t
*spa
)
4538 vdev_t
*rvd
= spa
->spa_root_vdev
;
4540 spa
->spa_is_initializing
= B_TRUE
;
4541 error
= dsl_pool_open(spa
->spa_dsl_pool
);
4542 spa
->spa_is_initializing
= B_FALSE
;
4544 spa_load_failed(spa
, "dsl_pool_open failed [error=%d]", error
);
4545 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4552 spa_ld_get_props(spa_t
*spa
)
4556 vdev_t
*rvd
= spa
->spa_root_vdev
;
4558 /* Grab the checksum salt from the MOS. */
4559 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4560 DMU_POOL_CHECKSUM_SALT
, 1,
4561 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
4562 spa
->spa_cksum_salt
.zcs_bytes
);
4563 if (error
== ENOENT
) {
4564 /* Generate a new salt for subsequent use */
4565 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4566 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4567 } else if (error
!= 0) {
4568 spa_load_failed(spa
, "unable to retrieve checksum salt from "
4569 "MOS [error=%d]", error
);
4570 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4573 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
, B_TRUE
) != 0)
4574 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4575 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
4577 spa_load_failed(spa
, "error opening deferred-frees bpobj "
4578 "[error=%d]", error
);
4579 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4583 * Load the bit that tells us to use the new accounting function
4584 * (raid-z deflation). If we have an older pool, this will not
4587 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
, B_FALSE
);
4588 if (error
!= 0 && error
!= ENOENT
)
4589 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4591 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
4592 &spa
->spa_creation_version
, B_FALSE
);
4593 if (error
!= 0 && error
!= ENOENT
)
4594 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4597 * Load the persistent error log. If we have an older pool, this will
4600 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
,
4602 if (error
!= 0 && error
!= ENOENT
)
4603 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4605 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
4606 &spa
->spa_errlog_scrub
, B_FALSE
);
4607 if (error
!= 0 && error
!= ENOENT
)
4608 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4611 * Load the livelist deletion field. If a livelist is queued for
4612 * deletion, indicate that in the spa
4614 error
= spa_dir_prop(spa
, DMU_POOL_DELETED_CLONES
,
4615 &spa
->spa_livelists_to_delete
, B_FALSE
);
4616 if (error
!= 0 && error
!= ENOENT
)
4617 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4620 * Load the history object. If we have an older pool, this
4621 * will not be present.
4623 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
, B_FALSE
);
4624 if (error
!= 0 && error
!= ENOENT
)
4625 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4628 * Load the per-vdev ZAP map. If we have an older pool, this will not
4629 * be present; in this case, defer its creation to a later time to
4630 * avoid dirtying the MOS this early / out of sync context. See
4631 * spa_sync_config_object.
4634 /* The sentinel is only available in the MOS config. */
4635 nvlist_t
*mos_config
;
4636 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0) {
4637 spa_load_failed(spa
, "unable to retrieve MOS config");
4638 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4641 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
4642 &spa
->spa_all_vdev_zaps
, B_FALSE
);
4644 if (error
== ENOENT
) {
4645 VERIFY(!nvlist_exists(mos_config
,
4646 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
4647 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
4648 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4649 } else if (error
!= 0) {
4650 nvlist_free(mos_config
);
4651 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4652 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
4654 * An older version of ZFS overwrote the sentinel value, so
4655 * we have orphaned per-vdev ZAPs in the MOS. Defer their
4656 * destruction to later; see spa_sync_config_object.
4658 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
4660 * We're assuming that no vdevs have had their ZAPs created
4661 * before this. Better be sure of it.
4663 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
4665 nvlist_free(mos_config
);
4667 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4669 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
,
4671 if (error
&& error
!= ENOENT
)
4672 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4675 uint64_t autoreplace
= 0;
4677 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
4678 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
4679 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
4680 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
4681 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
4682 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
4683 spa_prop_find(spa
, ZPOOL_PROP_AUTOTRIM
, &spa
->spa_autotrim
);
4684 spa
->spa_autoreplace
= (autoreplace
!= 0);
4688 * If we are importing a pool with missing top-level vdevs,
4689 * we enforce that the pool doesn't panic or get suspended on
4690 * error since the likelihood of missing data is extremely high.
4692 if (spa
->spa_missing_tvds
> 0 &&
4693 spa
->spa_failmode
!= ZIO_FAILURE_MODE_CONTINUE
&&
4694 spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4695 spa_load_note(spa
, "forcing failmode to 'continue' "
4696 "as some top level vdevs are missing");
4697 spa
->spa_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
4704 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
4707 vdev_t
*rvd
= spa
->spa_root_vdev
;
4710 * If we're assembling the pool from the split-off vdevs of
4711 * an existing pool, we don't want to attach the spares & cache
4716 * Load any hot spares for this pool.
4718 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
,
4720 if (error
!= 0 && error
!= ENOENT
)
4721 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4722 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4723 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
4724 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
4725 &spa
->spa_spares
.sav_config
) != 0) {
4726 spa_load_failed(spa
, "error loading spares nvlist");
4727 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4730 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4731 spa_load_spares(spa
);
4732 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4733 } else if (error
== 0) {
4734 spa
->spa_spares
.sav_sync
= B_TRUE
;
4738 * Load any level 2 ARC devices for this pool.
4740 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
4741 &spa
->spa_l2cache
.sav_object
, B_FALSE
);
4742 if (error
!= 0 && error
!= ENOENT
)
4743 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4744 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
4745 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
4746 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
4747 &spa
->spa_l2cache
.sav_config
) != 0) {
4748 spa_load_failed(spa
, "error loading l2cache nvlist");
4749 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4752 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4753 spa_load_l2cache(spa
);
4754 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4755 } else if (error
== 0) {
4756 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4763 spa_ld_load_vdev_metadata(spa_t
*spa
)
4766 vdev_t
*rvd
= spa
->spa_root_vdev
;
4769 * If the 'multihost' property is set, then never allow a pool to
4770 * be imported when the system hostid is zero. The exception to
4771 * this rule is zdb which is always allowed to access pools.
4773 if (spa_multihost(spa
) && spa_get_hostid(spa
) == 0 &&
4774 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
4775 fnvlist_add_uint64(spa
->spa_load_info
,
4776 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
4777 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
4781 * If the 'autoreplace' property is set, then post a resource notifying
4782 * the ZFS DE that it should not issue any faults for unopenable
4783 * devices. We also iterate over the vdevs, and post a sysevent for any
4784 * unopenable vdevs so that the normal autoreplace handler can take
4787 if (spa
->spa_autoreplace
&& spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4788 spa_check_removed(spa
->spa_root_vdev
);
4790 * For the import case, this is done in spa_import(), because
4791 * at this point we're using the spare definitions from
4792 * the MOS config, not necessarily from the userland config.
4794 if (spa
->spa_load_state
!= SPA_LOAD_IMPORT
) {
4795 spa_aux_check_removed(&spa
->spa_spares
);
4796 spa_aux_check_removed(&spa
->spa_l2cache
);
4801 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
4803 error
= vdev_load(rvd
);
4805 spa_load_failed(spa
, "vdev_load failed [error=%d]", error
);
4806 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4809 error
= spa_ld_log_spacemaps(spa
);
4811 spa_load_failed(spa
, "spa_ld_log_spacemaps failed [error=%d]",
4813 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
4817 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
4819 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4820 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
, B_FALSE
);
4821 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4827 spa_ld_load_dedup_tables(spa_t
*spa
)
4830 vdev_t
*rvd
= spa
->spa_root_vdev
;
4832 error
= ddt_load(spa
);
4834 spa_load_failed(spa
, "ddt_load failed [error=%d]", error
);
4835 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4842 spa_ld_load_brt(spa_t
*spa
)
4845 vdev_t
*rvd
= spa
->spa_root_vdev
;
4847 error
= brt_load(spa
);
4849 spa_load_failed(spa
, "brt_load failed [error=%d]", error
);
4850 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
4857 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, const char **ereport
)
4859 vdev_t
*rvd
= spa
->spa_root_vdev
;
4861 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
)) {
4862 boolean_t missing
= spa_check_logs(spa
);
4864 if (spa
->spa_missing_tvds
!= 0) {
4865 spa_load_note(spa
, "spa_check_logs failed "
4866 "so dropping the logs");
4868 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
4869 spa_load_failed(spa
, "spa_check_logs failed");
4870 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
,
4880 spa_ld_verify_pool_data(spa_t
*spa
)
4883 vdev_t
*rvd
= spa
->spa_root_vdev
;
4886 * We've successfully opened the pool, verify that we're ready
4887 * to start pushing transactions.
4889 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
4890 error
= spa_load_verify(spa
);
4892 spa_load_failed(spa
, "spa_load_verify failed "
4893 "[error=%d]", error
);
4894 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
4903 spa_ld_claim_log_blocks(spa_t
*spa
)
4906 dsl_pool_t
*dp
= spa_get_dsl(spa
);
4909 * Claim log blocks that haven't been committed yet.
4910 * This must all happen in a single txg.
4911 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
4912 * invoked from zil_claim_log_block()'s i/o done callback.
4913 * Price of rollback is that we abandon the log.
4915 spa
->spa_claiming
= B_TRUE
;
4917 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
4918 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
4919 zil_claim
, tx
, DS_FIND_CHILDREN
);
4922 spa
->spa_claiming
= B_FALSE
;
4924 spa_set_log_state(spa
, SPA_LOG_GOOD
);
4928 spa_ld_check_for_config_update(spa_t
*spa
, uint64_t config_cache_txg
,
4929 boolean_t update_config_cache
)
4931 vdev_t
*rvd
= spa
->spa_root_vdev
;
4932 int need_update
= B_FALSE
;
4935 * If the config cache is stale, or we have uninitialized
4936 * metaslabs (see spa_vdev_add()), then update the config.
4938 * If this is a verbatim import, trust the current
4939 * in-core spa_config and update the disk labels.
4941 if (update_config_cache
|| config_cache_txg
!= spa
->spa_config_txg
||
4942 spa
->spa_load_state
== SPA_LOAD_IMPORT
||
4943 spa
->spa_load_state
== SPA_LOAD_RECOVER
||
4944 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
4945 need_update
= B_TRUE
;
4947 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
4948 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
4949 need_update
= B_TRUE
;
4952 * Update the config cache asynchronously in case we're the
4953 * root pool, in which case the config cache isn't writable yet.
4956 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4960 spa_ld_prepare_for_reload(spa_t
*spa
)
4962 spa_mode_t mode
= spa
->spa_mode
;
4963 int async_suspended
= spa
->spa_async_suspended
;
4966 spa_deactivate(spa
);
4967 spa_activate(spa
, mode
);
4970 * We save the value of spa_async_suspended as it gets reset to 0 by
4971 * spa_unload(). We want to restore it back to the original value before
4972 * returning as we might be calling spa_async_resume() later.
4974 spa
->spa_async_suspended
= async_suspended
;
4978 spa_ld_read_checkpoint_txg(spa_t
*spa
)
4980 uberblock_t checkpoint
;
4983 ASSERT0(spa
->spa_checkpoint_txg
);
4984 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4986 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
4987 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
4988 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
4990 if (error
== ENOENT
)
4996 ASSERT3U(checkpoint
.ub_txg
, !=, 0);
4997 ASSERT3U(checkpoint
.ub_checkpoint_txg
, !=, 0);
4998 ASSERT3U(checkpoint
.ub_timestamp
, !=, 0);
4999 spa
->spa_checkpoint_txg
= checkpoint
.ub_txg
;
5000 spa
->spa_checkpoint_info
.sci_timestamp
= checkpoint
.ub_timestamp
;
5006 spa_ld_mos_init(spa_t
*spa
, spa_import_type_t type
)
5010 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5011 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
5014 * Never trust the config that is provided unless we are assembling
5015 * a pool following a split.
5016 * This means don't trust blkptrs and the vdev tree in general. This
5017 * also effectively puts the spa in read-only mode since
5018 * spa_writeable() checks for spa_trust_config to be true.
5019 * We will later load a trusted config from the MOS.
5021 if (type
!= SPA_IMPORT_ASSEMBLE
)
5022 spa
->spa_trust_config
= B_FALSE
;
5025 * Parse the config provided to create a vdev tree.
5027 error
= spa_ld_parse_config(spa
, type
);
5031 spa_import_progress_add(spa
);
5034 * Now that we have the vdev tree, try to open each vdev. This involves
5035 * opening the underlying physical device, retrieving its geometry and
5036 * probing the vdev with a dummy I/O. The state of each vdev will be set
5037 * based on the success of those operations. After this we'll be ready
5038 * to read from the vdevs.
5040 error
= spa_ld_open_vdevs(spa
);
5045 * Read the label of each vdev and make sure that the GUIDs stored
5046 * there match the GUIDs in the config provided.
5047 * If we're assembling a new pool that's been split off from an
5048 * existing pool, the labels haven't yet been updated so we skip
5049 * validation for now.
5051 if (type
!= SPA_IMPORT_ASSEMBLE
) {
5052 error
= spa_ld_validate_vdevs(spa
);
5058 * Read all vdev labels to find the best uberblock (i.e. latest,
5059 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
5060 * get the list of features required to read blkptrs in the MOS from
5061 * the vdev label with the best uberblock and verify that our version
5062 * of zfs supports them all.
5064 error
= spa_ld_select_uberblock(spa
, type
);
5069 * Pass that uberblock to the dsl_pool layer which will open the root
5070 * blkptr. This blkptr points to the latest version of the MOS and will
5071 * allow us to read its contents.
5073 error
= spa_ld_open_rootbp(spa
);
5081 spa_ld_checkpoint_rewind(spa_t
*spa
)
5083 uberblock_t checkpoint
;
5086 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5087 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
5089 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
5090 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
5091 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
5094 spa_load_failed(spa
, "unable to retrieve checkpointed "
5095 "uberblock from the MOS config [error=%d]", error
);
5097 if (error
== ENOENT
)
5098 error
= ZFS_ERR_NO_CHECKPOINT
;
5103 ASSERT3U(checkpoint
.ub_txg
, <, spa
->spa_uberblock
.ub_txg
);
5104 ASSERT3U(checkpoint
.ub_txg
, ==, checkpoint
.ub_checkpoint_txg
);
5107 * We need to update the txg and timestamp of the checkpointed
5108 * uberblock to be higher than the latest one. This ensures that
5109 * the checkpointed uberblock is selected if we were to close and
5110 * reopen the pool right after we've written it in the vdev labels.
5111 * (also see block comment in vdev_uberblock_compare)
5113 checkpoint
.ub_txg
= spa
->spa_uberblock
.ub_txg
+ 1;
5114 checkpoint
.ub_timestamp
= gethrestime_sec();
5117 * Set current uberblock to be the checkpointed uberblock.
5119 spa
->spa_uberblock
= checkpoint
;
5122 * If we are doing a normal rewind, then the pool is open for
5123 * writing and we sync the "updated" checkpointed uberblock to
5124 * disk. Once this is done, we've basically rewound the whole
5125 * pool and there is no way back.
5127 * There are cases when we don't want to attempt and sync the
5128 * checkpointed uberblock to disk because we are opening a
5129 * pool as read-only. Specifically, verifying the checkpointed
5130 * state with zdb, and importing the checkpointed state to get
5131 * a "preview" of its content.
5133 if (spa_writeable(spa
)) {
5134 vdev_t
*rvd
= spa
->spa_root_vdev
;
5136 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5137 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
5139 int children
= rvd
->vdev_children
;
5140 int c0
= random_in_range(children
);
5142 for (int c
= 0; c
< children
; c
++) {
5143 vdev_t
*vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
5145 /* Stop when revisiting the first vdev */
5146 if (c
> 0 && svd
[0] == vd
)
5149 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
5150 !vdev_is_concrete(vd
))
5153 svd
[svdcount
++] = vd
;
5154 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
5157 error
= vdev_config_sync(svd
, svdcount
, spa
->spa_first_txg
);
5159 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
5160 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5163 spa_load_failed(spa
, "failed to write checkpointed "
5164 "uberblock to the vdev labels [error=%d]", error
);
5173 spa_ld_mos_with_trusted_config(spa_t
*spa
, spa_import_type_t type
,
5174 boolean_t
*update_config_cache
)
5179 * Parse the config for pool, open and validate vdevs,
5180 * select an uberblock, and use that uberblock to open
5183 error
= spa_ld_mos_init(spa
, type
);
5188 * Retrieve the trusted config stored in the MOS and use it to create
5189 * a new, exact version of the vdev tree, then reopen all vdevs.
5191 error
= spa_ld_trusted_config(spa
, type
, B_FALSE
);
5192 if (error
== EAGAIN
) {
5193 if (update_config_cache
!= NULL
)
5194 *update_config_cache
= B_TRUE
;
5197 * Redo the loading process with the trusted config if it is
5198 * too different from the untrusted config.
5200 spa_ld_prepare_for_reload(spa
);
5201 spa_load_note(spa
, "RELOADING");
5202 error
= spa_ld_mos_init(spa
, type
);
5206 error
= spa_ld_trusted_config(spa
, type
, B_TRUE
);
5210 } else if (error
!= 0) {
5218 * Load an existing storage pool, using the config provided. This config
5219 * describes which vdevs are part of the pool and is later validated against
5220 * partial configs present in each vdev's label and an entire copy of the
5221 * config stored in the MOS.
5224 spa_load_impl(spa_t
*spa
, spa_import_type_t type
, const char **ereport
)
5227 boolean_t missing_feat_write
= B_FALSE
;
5228 boolean_t checkpoint_rewind
=
5229 (spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
5230 boolean_t update_config_cache
= B_FALSE
;
5232 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5233 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
5235 spa_load_note(spa
, "LOADING");
5237 error
= spa_ld_mos_with_trusted_config(spa
, type
, &update_config_cache
);
5242 * If we are rewinding to the checkpoint then we need to repeat
5243 * everything we've done so far in this function but this time
5244 * selecting the checkpointed uberblock and using that to open
5247 if (checkpoint_rewind
) {
5249 * If we are rewinding to the checkpoint update config cache
5252 update_config_cache
= B_TRUE
;
5255 * Extract the checkpointed uberblock from the current MOS
5256 * and use this as the pool's uberblock from now on. If the
5257 * pool is imported as writeable we also write the checkpoint
5258 * uberblock to the labels, making the rewind permanent.
5260 error
= spa_ld_checkpoint_rewind(spa
);
5265 * Redo the loading process again with the
5266 * checkpointed uberblock.
5268 spa_ld_prepare_for_reload(spa
);
5269 spa_load_note(spa
, "LOADING checkpointed uberblock");
5270 error
= spa_ld_mos_with_trusted_config(spa
, type
, NULL
);
5276 * Retrieve the checkpoint txg if the pool has a checkpoint.
5278 spa_import_progress_set_notes(spa
, "Loading checkpoint txg");
5279 error
= spa_ld_read_checkpoint_txg(spa
);
5284 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
5285 * from the pool and their contents were re-mapped to other vdevs. Note
5286 * that everything that we read before this step must have been
5287 * rewritten on concrete vdevs after the last device removal was
5288 * initiated. Otherwise we could be reading from indirect vdevs before
5289 * we have loaded their mappings.
5291 spa_import_progress_set_notes(spa
, "Loading indirect vdev metadata");
5292 error
= spa_ld_open_indirect_vdev_metadata(spa
);
5297 * Retrieve the full list of active features from the MOS and check if
5298 * they are all supported.
5300 spa_import_progress_set_notes(spa
, "Checking feature flags");
5301 error
= spa_ld_check_features(spa
, &missing_feat_write
);
5306 * Load several special directories from the MOS needed by the dsl_pool
5309 spa_import_progress_set_notes(spa
, "Loading special MOS directories");
5310 error
= spa_ld_load_special_directories(spa
);
5315 * Retrieve pool properties from the MOS.
5317 spa_import_progress_set_notes(spa
, "Loading properties");
5318 error
= spa_ld_get_props(spa
);
5323 * Retrieve the list of auxiliary devices - cache devices and spares -
5326 spa_import_progress_set_notes(spa
, "Loading AUX vdevs");
5327 error
= spa_ld_open_aux_vdevs(spa
, type
);
5332 * Load the metadata for all vdevs. Also check if unopenable devices
5333 * should be autoreplaced.
5335 spa_import_progress_set_notes(spa
, "Loading vdev metadata");
5336 error
= spa_ld_load_vdev_metadata(spa
);
5340 spa_import_progress_set_notes(spa
, "Loading dedup tables");
5341 error
= spa_ld_load_dedup_tables(spa
);
5345 spa_import_progress_set_notes(spa
, "Loading BRT");
5346 error
= spa_ld_load_brt(spa
);
5351 * Verify the logs now to make sure we don't have any unexpected errors
5352 * when we claim log blocks later.
5354 spa_import_progress_set_notes(spa
, "Verifying Log Devices");
5355 error
= spa_ld_verify_logs(spa
, type
, ereport
);
5359 if (missing_feat_write
) {
5360 ASSERT(spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
);
5363 * At this point, we know that we can open the pool in
5364 * read-only mode but not read-write mode. We now have enough
5365 * information and can return to userland.
5367 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
5372 * Traverse the last txgs to make sure the pool was left off in a safe
5373 * state. When performing an extreme rewind, we verify the whole pool,
5374 * which can take a very long time.
5376 spa_import_progress_set_notes(spa
, "Verifying pool data");
5377 error
= spa_ld_verify_pool_data(spa
);
5382 * Calculate the deflated space for the pool. This must be done before
5383 * we write anything to the pool because we'd need to update the space
5384 * accounting using the deflated sizes.
5386 spa_import_progress_set_notes(spa
, "Calculating deflated space");
5387 spa_update_dspace(spa
);
5390 * We have now retrieved all the information we needed to open the
5391 * pool. If we are importing the pool in read-write mode, a few
5392 * additional steps must be performed to finish the import.
5394 spa_import_progress_set_notes(spa
, "Starting import");
5395 if (spa_writeable(spa
) && (spa
->spa_load_state
== SPA_LOAD_RECOVER
||
5396 spa
->spa_load_max_txg
== UINT64_MAX
)) {
5397 uint64_t config_cache_txg
= spa
->spa_config_txg
;
5399 ASSERT(spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
);
5402 * Before we do any zio_write's, complete the raidz expansion
5403 * scratch space copying, if necessary.
5405 if (RRSS_GET_STATE(&spa
->spa_uberblock
) == RRSS_SCRATCH_VALID
)
5406 vdev_raidz_reflow_copy_scratch(spa
);
5409 * In case of a checkpoint rewind, log the original txg
5410 * of the checkpointed uberblock.
5412 if (checkpoint_rewind
) {
5413 spa_history_log_internal(spa
, "checkpoint rewind",
5414 NULL
, "rewound state to txg=%llu",
5415 (u_longlong_t
)spa
->spa_uberblock
.ub_checkpoint_txg
);
5418 spa_import_progress_set_notes(spa
, "Claiming ZIL blocks");
5420 * Traverse the ZIL and claim all blocks.
5422 spa_ld_claim_log_blocks(spa
);
5425 * Kick-off the syncing thread.
5427 spa
->spa_sync_on
= B_TRUE
;
5428 txg_sync_start(spa
->spa_dsl_pool
);
5429 mmp_thread_start(spa
);
5432 * Wait for all claims to sync. We sync up to the highest
5433 * claimed log block birth time so that claimed log blocks
5434 * don't appear to be from the future. spa_claim_max_txg
5435 * will have been set for us by ZIL traversal operations
5438 spa_import_progress_set_notes(spa
, "Syncing ZIL claims");
5439 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
5442 * Check if we need to request an update of the config. On the
5443 * next sync, we would update the config stored in vdev labels
5444 * and the cachefile (by default /etc/zfs/zpool.cache).
5446 spa_import_progress_set_notes(spa
, "Updating configs");
5447 spa_ld_check_for_config_update(spa
, config_cache_txg
,
5448 update_config_cache
);
5451 * Check if a rebuild was in progress and if so resume it.
5452 * Then check all DTLs to see if anything needs resilvering.
5453 * The resilver will be deferred if a rebuild was started.
5455 spa_import_progress_set_notes(spa
, "Starting resilvers");
5456 if (vdev_rebuild_active(spa
->spa_root_vdev
)) {
5457 vdev_rebuild_restart(spa
);
5458 } else if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
5459 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5460 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
5464 * Log the fact that we booted up (so that we can detect if
5465 * we rebooted in the middle of an operation).
5467 spa_history_log_version(spa
, "open", NULL
);
5469 spa_import_progress_set_notes(spa
,
5470 "Restarting device removals");
5471 spa_restart_removal(spa
);
5472 spa_spawn_aux_threads(spa
);
5475 * Delete any inconsistent datasets.
5478 * Since we may be issuing deletes for clones here,
5479 * we make sure to do so after we've spawned all the
5480 * auxiliary threads above (from which the livelist
5481 * deletion zthr is part of).
5483 spa_import_progress_set_notes(spa
,
5484 "Cleaning up inconsistent objsets");
5485 (void) dmu_objset_find(spa_name(spa
),
5486 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
5489 * Clean up any stale temporary dataset userrefs.
5491 spa_import_progress_set_notes(spa
,
5492 "Cleaning up temporary userrefs");
5493 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
5495 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5496 spa_import_progress_set_notes(spa
, "Restarting initialize");
5497 vdev_initialize_restart(spa
->spa_root_vdev
);
5498 spa_import_progress_set_notes(spa
, "Restarting TRIM");
5499 vdev_trim_restart(spa
->spa_root_vdev
);
5500 vdev_autotrim_restart(spa
);
5501 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5502 spa_import_progress_set_notes(spa
, "Finished importing");
5505 spa_import_progress_remove(spa_guid(spa
));
5506 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
5508 spa_load_note(spa
, "LOADED");
5514 spa_load_retry(spa_t
*spa
, spa_load_state_t state
)
5516 spa_mode_t mode
= spa
->spa_mode
;
5519 spa_deactivate(spa
);
5521 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
5523 spa_activate(spa
, mode
);
5524 spa_async_suspend(spa
);
5526 spa_load_note(spa
, "spa_load_retry: rewind, max txg: %llu",
5527 (u_longlong_t
)spa
->spa_load_max_txg
);
5529 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
));
5533 * If spa_load() fails this function will try loading prior txg's. If
5534 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
5535 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
5536 * function will not rewind the pool and will return the same error as
5540 spa_load_best(spa_t
*spa
, spa_load_state_t state
, uint64_t max_request
,
5543 nvlist_t
*loadinfo
= NULL
;
5544 nvlist_t
*config
= NULL
;
5545 int load_error
, rewind_error
;
5546 uint64_t safe_rewind_txg
;
5549 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
5550 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
5551 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5553 spa
->spa_load_max_txg
= max_request
;
5554 if (max_request
!= UINT64_MAX
)
5555 spa
->spa_extreme_rewind
= B_TRUE
;
5558 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
);
5559 if (load_error
== 0)
5561 if (load_error
== ZFS_ERR_NO_CHECKPOINT
) {
5563 * When attempting checkpoint-rewind on a pool with no
5564 * checkpoint, we should not attempt to load uberblocks
5565 * from previous txgs when spa_load fails.
5567 ASSERT(spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
);
5568 spa_import_progress_remove(spa_guid(spa
));
5569 return (load_error
);
5572 if (spa
->spa_root_vdev
!= NULL
)
5573 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5575 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
5576 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
5578 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
5579 nvlist_free(config
);
5580 spa_import_progress_remove(spa_guid(spa
));
5581 return (load_error
);
5584 if (state
== SPA_LOAD_RECOVER
) {
5585 /* Price of rolling back is discarding txgs, including log */
5586 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
5589 * If we aren't rolling back save the load info from our first
5590 * import attempt so that we can restore it after attempting
5593 loadinfo
= spa
->spa_load_info
;
5594 spa
->spa_load_info
= fnvlist_alloc();
5597 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
5598 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
5599 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
5600 TXG_INITIAL
: safe_rewind_txg
;
5603 * Continue as long as we're finding errors, we're still within
5604 * the acceptable rewind range, and we're still finding uberblocks
5606 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
5607 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
5608 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
5609 spa
->spa_extreme_rewind
= B_TRUE
;
5610 rewind_error
= spa_load_retry(spa
, state
);
5613 spa
->spa_extreme_rewind
= B_FALSE
;
5614 spa
->spa_load_max_txg
= UINT64_MAX
;
5616 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
5617 spa_config_set(spa
, config
);
5619 nvlist_free(config
);
5621 if (state
== SPA_LOAD_RECOVER
) {
5622 ASSERT3P(loadinfo
, ==, NULL
);
5623 spa_import_progress_remove(spa_guid(spa
));
5624 return (rewind_error
);
5626 /* Store the rewind info as part of the initial load info */
5627 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
5628 spa
->spa_load_info
);
5630 /* Restore the initial load info */
5631 fnvlist_free(spa
->spa_load_info
);
5632 spa
->spa_load_info
= loadinfo
;
5634 spa_import_progress_remove(spa_guid(spa
));
5635 return (load_error
);
5642 * The import case is identical to an open except that the configuration is sent
5643 * down from userland, instead of grabbed from the configuration cache. For the
5644 * case of an open, the pool configuration will exist in the
5645 * POOL_STATE_UNINITIALIZED state.
5647 * The stats information (gen/count/ustats) is used to gather vdev statistics at
5648 * the same time open the pool, without having to keep around the spa_t in some
5652 spa_open_common(const char *pool
, spa_t
**spapp
, const void *tag
,
5653 nvlist_t
*nvpolicy
, nvlist_t
**config
)
5656 spa_load_state_t state
= SPA_LOAD_OPEN
;
5658 int locked
= B_FALSE
;
5659 int firstopen
= B_FALSE
;
5664 * As disgusting as this is, we need to support recursive calls to this
5665 * function because dsl_dir_open() is called during spa_load(), and ends
5666 * up calling spa_open() again. The real fix is to figure out how to
5667 * avoid dsl_dir_open() calling this in the first place.
5669 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
5670 mutex_enter(&spa_namespace_lock
);
5674 if ((spa
= spa_lookup(pool
)) == NULL
) {
5676 mutex_exit(&spa_namespace_lock
);
5677 return (SET_ERROR(ENOENT
));
5680 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
5681 zpool_load_policy_t policy
;
5685 zpool_get_load_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
5687 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
5688 state
= SPA_LOAD_RECOVER
;
5690 spa_activate(spa
, spa_mode_global
);
5692 if (state
!= SPA_LOAD_RECOVER
)
5693 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
5694 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
5696 zfs_dbgmsg("spa_open_common: opening %s", pool
);
5697 error
= spa_load_best(spa
, state
, policy
.zlp_txg
,
5700 if (error
== EBADF
) {
5702 * If vdev_validate() returns failure (indicated by
5703 * EBADF), it indicates that one of the vdevs indicates
5704 * that the pool has been exported or destroyed. If
5705 * this is the case, the config cache is out of sync and
5706 * we should remove the pool from the namespace.
5709 spa_deactivate(spa
);
5710 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
, B_FALSE
);
5713 mutex_exit(&spa_namespace_lock
);
5714 return (SET_ERROR(ENOENT
));
5719 * We can't open the pool, but we still have useful
5720 * information: the state of each vdev after the
5721 * attempted vdev_open(). Return this to the user.
5723 if (config
!= NULL
&& spa
->spa_config
) {
5724 *config
= fnvlist_dup(spa
->spa_config
);
5725 fnvlist_add_nvlist(*config
,
5726 ZPOOL_CONFIG_LOAD_INFO
,
5727 spa
->spa_load_info
);
5730 spa_deactivate(spa
);
5731 spa
->spa_last_open_failed
= error
;
5733 mutex_exit(&spa_namespace_lock
);
5739 spa_open_ref(spa
, tag
);
5742 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5745 * If we've recovered the pool, pass back any information we
5746 * gathered while doing the load.
5748 if (state
== SPA_LOAD_RECOVER
&& config
!= NULL
) {
5749 fnvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
5750 spa
->spa_load_info
);
5754 spa
->spa_last_open_failed
= 0;
5755 spa
->spa_last_ubsync_txg
= 0;
5756 spa
->spa_load_txg
= 0;
5757 mutex_exit(&spa_namespace_lock
);
5761 zvol_create_minors_recursive(spa_name(spa
));
5769 spa_open_rewind(const char *name
, spa_t
**spapp
, const void *tag
,
5770 nvlist_t
*policy
, nvlist_t
**config
)
5772 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
5776 spa_open(const char *name
, spa_t
**spapp
, const void *tag
)
5778 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
5782 * Lookup the given spa_t, incrementing the inject count in the process,
5783 * preventing it from being exported or destroyed.
5786 spa_inject_addref(char *name
)
5790 mutex_enter(&spa_namespace_lock
);
5791 if ((spa
= spa_lookup(name
)) == NULL
) {
5792 mutex_exit(&spa_namespace_lock
);
5795 spa
->spa_inject_ref
++;
5796 mutex_exit(&spa_namespace_lock
);
5802 spa_inject_delref(spa_t
*spa
)
5804 mutex_enter(&spa_namespace_lock
);
5805 spa
->spa_inject_ref
--;
5806 mutex_exit(&spa_namespace_lock
);
5810 * Add spares device information to the nvlist.
5813 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
5823 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5825 if (spa
->spa_spares
.sav_count
== 0)
5828 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5829 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5830 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
));
5832 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5833 (const nvlist_t
* const *)spares
, nspares
);
5834 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5835 &spares
, &nspares
));
5838 * Go through and find any spares which have since been
5839 * repurposed as an active spare. If this is the case, update
5840 * their status appropriately.
5842 for (i
= 0; i
< nspares
; i
++) {
5843 guid
= fnvlist_lookup_uint64(spares
[i
],
5845 VERIFY0(nvlist_lookup_uint64_array(spares
[i
],
5846 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
));
5847 if (spa_spare_exists(guid
, &pool
, NULL
) &&
5849 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
5850 vs
->vs_aux
= VDEV_AUX_SPARED
;
5853 spa
->spa_spares
.sav_vdevs
[i
]->vdev_state
;
5860 * Add l2cache device information to the nvlist, including vdev stats.
5863 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
5866 uint_t i
, j
, nl2cache
;
5873 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5875 if (spa
->spa_l2cache
.sav_count
== 0)
5878 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
5879 VERIFY0(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5880 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
));
5881 if (nl2cache
!= 0) {
5882 fnvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5883 (const nvlist_t
* const *)l2cache
, nl2cache
);
5884 VERIFY0(nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5885 &l2cache
, &nl2cache
));
5888 * Update level 2 cache device stats.
5891 for (i
= 0; i
< nl2cache
; i
++) {
5892 guid
= fnvlist_lookup_uint64(l2cache
[i
],
5896 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
5898 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
5899 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
5905 VERIFY0(nvlist_lookup_uint64_array(l2cache
[i
],
5906 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
));
5907 vdev_get_stats(vd
, vs
);
5908 vdev_config_generate_stats(vd
, l2cache
[i
]);
5915 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
5920 if (spa
->spa_feat_for_read_obj
!= 0) {
5921 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5922 spa
->spa_feat_for_read_obj
);
5923 zap_cursor_retrieve(&zc
, &za
) == 0;
5924 zap_cursor_advance(&zc
)) {
5925 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5926 za
.za_num_integers
== 1);
5927 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5928 za
.za_first_integer
));
5930 zap_cursor_fini(&zc
);
5933 if (spa
->spa_feat_for_write_obj
!= 0) {
5934 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
5935 spa
->spa_feat_for_write_obj
);
5936 zap_cursor_retrieve(&zc
, &za
) == 0;
5937 zap_cursor_advance(&zc
)) {
5938 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
5939 za
.za_num_integers
== 1);
5940 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
5941 za
.za_first_integer
));
5943 zap_cursor_fini(&zc
);
5948 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
5952 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5953 zfeature_info_t feature
= spa_feature_table
[i
];
5956 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
5959 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
5964 * Store a list of pool features and their reference counts in the
5967 * The first time this is called on a spa, allocate a new nvlist, fetch
5968 * the pool features and reference counts from disk, then save the list
5969 * in the spa. In subsequent calls on the same spa use the saved nvlist
5970 * and refresh its values from the cached reference counts. This
5971 * ensures we don't block here on I/O on a suspended pool so 'zpool
5972 * clear' can resume the pool.
5975 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
5979 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
5981 mutex_enter(&spa
->spa_feat_stats_lock
);
5982 features
= spa
->spa_feat_stats
;
5984 if (features
!= NULL
) {
5985 spa_feature_stats_from_cache(spa
, features
);
5987 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
5988 spa
->spa_feat_stats
= features
;
5989 spa_feature_stats_from_disk(spa
, features
);
5992 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
5995 mutex_exit(&spa
->spa_feat_stats_lock
);
5999 spa_get_stats(const char *name
, nvlist_t
**config
,
6000 char *altroot
, size_t buflen
)
6006 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
6010 * This still leaves a window of inconsistency where the spares
6011 * or l2cache devices could change and the config would be
6012 * self-inconsistent.
6014 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6016 if (*config
!= NULL
) {
6017 uint64_t loadtimes
[2];
6019 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
6020 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
6021 fnvlist_add_uint64_array(*config
,
6022 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2);
6024 fnvlist_add_uint64(*config
,
6025 ZPOOL_CONFIG_ERRCOUNT
,
6026 spa_approx_errlog_size(spa
));
6028 if (spa_suspended(spa
)) {
6029 fnvlist_add_uint64(*config
,
6030 ZPOOL_CONFIG_SUSPENDED
,
6032 fnvlist_add_uint64(*config
,
6033 ZPOOL_CONFIG_SUSPENDED_REASON
,
6034 spa
->spa_suspended
);
6037 spa_add_spares(spa
, *config
);
6038 spa_add_l2cache(spa
, *config
);
6039 spa_add_feature_stats(spa
, *config
);
6044 * We want to get the alternate root even for faulted pools, so we cheat
6045 * and call spa_lookup() directly.
6049 mutex_enter(&spa_namespace_lock
);
6050 spa
= spa_lookup(name
);
6052 spa_altroot(spa
, altroot
, buflen
);
6056 mutex_exit(&spa_namespace_lock
);
6058 spa_altroot(spa
, altroot
, buflen
);
6063 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6064 spa_close(spa
, FTAG
);
6071 * Validate that the auxiliary device array is well formed. We must have an
6072 * array of nvlists, each which describes a valid leaf vdev. If this is an
6073 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
6074 * specified, as long as they are well-formed.
6077 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
6078 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
6079 vdev_labeltype_t label
)
6086 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
6089 * It's acceptable to have no devs specified.
6091 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
6095 return (SET_ERROR(EINVAL
));
6098 * Make sure the pool is formatted with a version that supports this
6101 if (spa_version(spa
) < version
)
6102 return (SET_ERROR(ENOTSUP
));
6105 * Set the pending device list so we correctly handle device in-use
6108 sav
->sav_pending
= dev
;
6109 sav
->sav_npending
= ndev
;
6111 for (i
= 0; i
< ndev
; i
++) {
6112 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
6116 if (!vd
->vdev_ops
->vdev_op_leaf
) {
6118 error
= SET_ERROR(EINVAL
);
6124 if ((error
= vdev_open(vd
)) == 0 &&
6125 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
6126 fnvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
6133 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
6140 sav
->sav_pending
= NULL
;
6141 sav
->sav_npending
= 0;
6146 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
6150 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
6152 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
6153 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
6154 VDEV_LABEL_SPARE
)) != 0) {
6158 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
6159 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
6160 VDEV_LABEL_L2CACHE
));
6164 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
6169 if (sav
->sav_config
!= NULL
) {
6175 * Generate new dev list by concatenating with the
6178 VERIFY0(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
6179 &olddevs
, &oldndevs
));
6181 newdevs
= kmem_alloc(sizeof (void *) *
6182 (ndevs
+ oldndevs
), KM_SLEEP
);
6183 for (i
= 0; i
< oldndevs
; i
++)
6184 newdevs
[i
] = fnvlist_dup(olddevs
[i
]);
6185 for (i
= 0; i
< ndevs
; i
++)
6186 newdevs
[i
+ oldndevs
] = fnvlist_dup(devs
[i
]);
6188 fnvlist_remove(sav
->sav_config
, config
);
6190 fnvlist_add_nvlist_array(sav
->sav_config
, config
,
6191 (const nvlist_t
* const *)newdevs
, ndevs
+ oldndevs
);
6192 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
6193 nvlist_free(newdevs
[i
]);
6194 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
6197 * Generate a new dev list.
6199 sav
->sav_config
= fnvlist_alloc();
6200 fnvlist_add_nvlist_array(sav
->sav_config
, config
,
6201 (const nvlist_t
* const *)devs
, ndevs
);
6206 * Stop and drop level 2 ARC devices
6209 spa_l2cache_drop(spa_t
*spa
)
6213 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
6215 for (i
= 0; i
< sav
->sav_count
; i
++) {
6218 vd
= sav
->sav_vdevs
[i
];
6221 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
6222 pool
!= 0ULL && l2arc_vdev_present(vd
))
6223 l2arc_remove_vdev(vd
);
6228 * Verify encryption parameters for spa creation. If we are encrypting, we must
6229 * have the encryption feature flag enabled.
6232 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
6233 boolean_t has_encryption
)
6235 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
6236 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
6238 return (SET_ERROR(ENOTSUP
));
6240 return (dmu_objset_create_crypt_check(NULL
, dcp
, NULL
));
6247 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
6248 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
6251 const char *altroot
= NULL
;
6256 uint64_t txg
= TXG_INITIAL
;
6257 nvlist_t
**spares
, **l2cache
;
6258 uint_t nspares
, nl2cache
;
6259 uint64_t version
, obj
, ndraid
= 0;
6260 boolean_t has_features
;
6261 boolean_t has_encryption
;
6262 boolean_t has_allocclass
;
6264 const char *feat_name
;
6265 const char *poolname
;
6268 if (props
== NULL
||
6269 nvlist_lookup_string(props
,
6270 zpool_prop_to_name(ZPOOL_PROP_TNAME
), &poolname
) != 0)
6271 poolname
= (char *)pool
;
6274 * If this pool already exists, return failure.
6276 mutex_enter(&spa_namespace_lock
);
6277 if (spa_lookup(poolname
) != NULL
) {
6278 mutex_exit(&spa_namespace_lock
);
6279 return (SET_ERROR(EEXIST
));
6283 * Allocate a new spa_t structure.
6285 nvl
= fnvlist_alloc();
6286 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
6287 (void) nvlist_lookup_string(props
,
6288 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
6289 spa
= spa_add(poolname
, nvl
, altroot
);
6291 spa_activate(spa
, spa_mode_global
);
6293 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
6294 spa_deactivate(spa
);
6296 mutex_exit(&spa_namespace_lock
);
6301 * Temporary pool names should never be written to disk.
6303 if (poolname
!= pool
)
6304 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
6306 has_features
= B_FALSE
;
6307 has_encryption
= B_FALSE
;
6308 has_allocclass
= B_FALSE
;
6309 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
6310 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
6311 if (zpool_prop_feature(nvpair_name(elem
))) {
6312 has_features
= B_TRUE
;
6314 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
6315 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
6316 if (feat
== SPA_FEATURE_ENCRYPTION
)
6317 has_encryption
= B_TRUE
;
6318 if (feat
== SPA_FEATURE_ALLOCATION_CLASSES
)
6319 has_allocclass
= B_TRUE
;
6323 /* verify encryption params, if they were provided */
6325 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
6327 spa_deactivate(spa
);
6329 mutex_exit(&spa_namespace_lock
);
6333 if (!has_allocclass
&& zfs_special_devs(nvroot
, NULL
)) {
6334 spa_deactivate(spa
);
6336 mutex_exit(&spa_namespace_lock
);
6340 if (has_features
|| nvlist_lookup_uint64(props
,
6341 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
6342 version
= SPA_VERSION
;
6344 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6346 spa
->spa_first_txg
= txg
;
6347 spa
->spa_uberblock
.ub_txg
= txg
- 1;
6348 spa
->spa_uberblock
.ub_version
= version
;
6349 spa
->spa_ubsync
= spa
->spa_uberblock
;
6350 spa
->spa_load_state
= SPA_LOAD_CREATE
;
6351 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
6352 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
6353 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
6354 spa
->spa_indirect_vdevs_loaded
= B_TRUE
;
6357 * Create "The Godfather" zio to hold all async IOs
6359 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
6361 for (int i
= 0; i
< max_ncpus
; i
++) {
6362 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
6363 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
6364 ZIO_FLAG_GODFATHER
);
6368 * Create the root vdev.
6370 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6372 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
6374 ASSERT(error
!= 0 || rvd
!= NULL
);
6375 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
6377 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
6378 error
= SET_ERROR(EINVAL
);
6381 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
6382 (error
= vdev_draid_spare_create(nvroot
, rvd
, &ndraid
, 0)) == 0 &&
6383 (error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) == 0) {
6385 * instantiate the metaslab groups (this will dirty the vdevs)
6386 * we can no longer error exit past this point
6388 for (int c
= 0; error
== 0 && c
< rvd
->vdev_children
; c
++) {
6389 vdev_t
*vd
= rvd
->vdev_child
[c
];
6391 vdev_metaslab_set_size(vd
);
6392 vdev_expand(vd
, txg
);
6396 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6400 spa_deactivate(spa
);
6402 mutex_exit(&spa_namespace_lock
);
6407 * Get the list of spares, if specified.
6409 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
6410 &spares
, &nspares
) == 0) {
6411 spa
->spa_spares
.sav_config
= fnvlist_alloc();
6412 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
6413 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
6415 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6416 spa_load_spares(spa
);
6417 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6418 spa
->spa_spares
.sav_sync
= B_TRUE
;
6422 * Get the list of level 2 cache devices, if specified.
6424 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
6425 &l2cache
, &nl2cache
) == 0) {
6426 VERIFY0(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
6427 NV_UNIQUE_NAME
, KM_SLEEP
));
6428 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
6429 ZPOOL_CONFIG_L2CACHE
, (const nvlist_t
* const *)l2cache
,
6431 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6432 spa_load_l2cache(spa
);
6433 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6434 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6437 spa
->spa_is_initializing
= B_TRUE
;
6438 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
6439 spa
->spa_is_initializing
= B_FALSE
;
6442 * Create DDTs (dedup tables).
6446 * Create BRT table and BRT table object.
6450 spa_update_dspace(spa
);
6452 tx
= dmu_tx_create_assigned(dp
, txg
);
6455 * Create the pool's history object.
6457 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
6458 spa_history_create_obj(spa
, tx
);
6460 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
6461 spa_history_log_version(spa
, "create", tx
);
6464 * Create the pool config object.
6466 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6467 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
6468 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
6470 if (zap_add(spa
->spa_meta_objset
,
6471 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
6472 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
6473 cmn_err(CE_PANIC
, "failed to add pool config");
6476 if (zap_add(spa
->spa_meta_objset
,
6477 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
6478 sizeof (uint64_t), 1, &version
, tx
) != 0) {
6479 cmn_err(CE_PANIC
, "failed to add pool version");
6482 /* Newly created pools with the right version are always deflated. */
6483 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6484 spa
->spa_deflate
= TRUE
;
6485 if (zap_add(spa
->spa_meta_objset
,
6486 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6487 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
6488 cmn_err(CE_PANIC
, "failed to add deflate");
6493 * Create the deferred-free bpobj. Turn off compression
6494 * because sync-to-convergence takes longer if the blocksize
6497 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
6498 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
6499 ZIO_COMPRESS_OFF
, tx
);
6500 if (zap_add(spa
->spa_meta_objset
,
6501 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
6502 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
6503 cmn_err(CE_PANIC
, "failed to add bpobj");
6505 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
6506 spa
->spa_meta_objset
, obj
));
6509 * Generate some random noise for salted checksums to operate on.
6511 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
6512 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
6515 * Set pool properties.
6517 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
6518 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
6519 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
6520 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
6521 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
6522 spa
->spa_autotrim
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOTRIM
);
6524 if (props
!= NULL
) {
6525 spa_configfile_set(spa
, props
, B_FALSE
);
6526 spa_sync_props(props
, tx
);
6529 for (int i
= 0; i
< ndraid
; i
++)
6530 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
6534 spa
->spa_sync_on
= B_TRUE
;
6536 mmp_thread_start(spa
);
6537 txg_wait_synced(dp
, txg
);
6539 spa_spawn_aux_threads(spa
);
6541 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
, B_TRUE
);
6544 * Don't count references from objsets that are already closed
6545 * and are making their way through the eviction process.
6547 spa_evicting_os_wait(spa
);
6548 spa
->spa_minref
= zfs_refcount_count(&spa
->spa_refcount
);
6549 spa
->spa_load_state
= SPA_LOAD_NONE
;
6553 mutex_exit(&spa_namespace_lock
);
6559 * Import a non-root pool into the system.
6562 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
6565 const char *altroot
= NULL
;
6566 spa_load_state_t state
= SPA_LOAD_IMPORT
;
6567 zpool_load_policy_t policy
;
6568 spa_mode_t mode
= spa_mode_global
;
6569 uint64_t readonly
= B_FALSE
;
6572 nvlist_t
**spares
, **l2cache
;
6573 uint_t nspares
, nl2cache
;
6576 * If a pool with this name exists, return failure.
6578 mutex_enter(&spa_namespace_lock
);
6579 if (spa_lookup(pool
) != NULL
) {
6580 mutex_exit(&spa_namespace_lock
);
6581 return (SET_ERROR(EEXIST
));
6585 * Create and initialize the spa structure.
6587 (void) nvlist_lookup_string(props
,
6588 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
6589 (void) nvlist_lookup_uint64(props
,
6590 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
6592 mode
= SPA_MODE_READ
;
6593 spa
= spa_add(pool
, config
, altroot
);
6594 spa
->spa_import_flags
= flags
;
6597 * Verbatim import - Take a pool and insert it into the namespace
6598 * as if it had been loaded at boot.
6600 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
6602 spa_configfile_set(spa
, props
, B_FALSE
);
6604 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
, B_FALSE
);
6605 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6606 zfs_dbgmsg("spa_import: verbatim import of %s", pool
);
6607 mutex_exit(&spa_namespace_lock
);
6611 spa_activate(spa
, mode
);
6614 * Don't start async tasks until we know everything is healthy.
6616 spa_async_suspend(spa
);
6618 zpool_get_load_policy(config
, &policy
);
6619 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
6620 state
= SPA_LOAD_RECOVER
;
6622 spa
->spa_config_source
= SPA_CONFIG_SRC_TRYIMPORT
;
6624 if (state
!= SPA_LOAD_RECOVER
) {
6625 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
6626 zfs_dbgmsg("spa_import: importing %s", pool
);
6628 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
6629 "(RECOVERY MODE)", pool
, (longlong_t
)policy
.zlp_txg
);
6631 error
= spa_load_best(spa
, state
, policy
.zlp_txg
, policy
.zlp_rewind
);
6634 * Propagate anything learned while loading the pool and pass it
6635 * back to caller (i.e. rewind info, missing devices, etc).
6637 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
, spa
->spa_load_info
);
6639 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6641 * Toss any existing sparelist, as it doesn't have any validity
6642 * anymore, and conflicts with spa_has_spare().
6644 if (spa
->spa_spares
.sav_config
) {
6645 nvlist_free(spa
->spa_spares
.sav_config
);
6646 spa
->spa_spares
.sav_config
= NULL
;
6647 spa_load_spares(spa
);
6649 if (spa
->spa_l2cache
.sav_config
) {
6650 nvlist_free(spa
->spa_l2cache
.sav_config
);
6651 spa
->spa_l2cache
.sav_config
= NULL
;
6652 spa_load_l2cache(spa
);
6655 nvroot
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
);
6656 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6659 spa_configfile_set(spa
, props
, B_FALSE
);
6661 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
6662 (error
= spa_prop_set(spa
, props
)))) {
6664 spa_deactivate(spa
);
6666 mutex_exit(&spa_namespace_lock
);
6670 spa_async_resume(spa
);
6673 * Override any spares and level 2 cache devices as specified by
6674 * the user, as these may have correct device names/devids, etc.
6676 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
6677 &spares
, &nspares
) == 0) {
6678 if (spa
->spa_spares
.sav_config
)
6679 fnvlist_remove(spa
->spa_spares
.sav_config
,
6680 ZPOOL_CONFIG_SPARES
);
6682 spa
->spa_spares
.sav_config
= fnvlist_alloc();
6683 fnvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
6684 ZPOOL_CONFIG_SPARES
, (const nvlist_t
* const *)spares
,
6686 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6687 spa_load_spares(spa
);
6688 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6689 spa
->spa_spares
.sav_sync
= B_TRUE
;
6691 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
6692 &l2cache
, &nl2cache
) == 0) {
6693 if (spa
->spa_l2cache
.sav_config
)
6694 fnvlist_remove(spa
->spa_l2cache
.sav_config
,
6695 ZPOOL_CONFIG_L2CACHE
);
6697 spa
->spa_l2cache
.sav_config
= fnvlist_alloc();
6698 fnvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
6699 ZPOOL_CONFIG_L2CACHE
, (const nvlist_t
* const *)l2cache
,
6701 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6702 spa_load_l2cache(spa
);
6703 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6704 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
6708 * Check for any removed devices.
6710 if (spa
->spa_autoreplace
) {
6711 spa_aux_check_removed(&spa
->spa_spares
);
6712 spa_aux_check_removed(&spa
->spa_l2cache
);
6715 if (spa_writeable(spa
)) {
6717 * Update the config cache to include the newly-imported pool.
6719 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6723 * It's possible that the pool was expanded while it was exported.
6724 * We kick off an async task to handle this for us.
6726 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
6728 spa_history_log_version(spa
, "import", NULL
);
6730 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
6732 mutex_exit(&spa_namespace_lock
);
6734 zvol_create_minors_recursive(pool
);
6742 spa_tryimport(nvlist_t
*tryconfig
)
6744 nvlist_t
*config
= NULL
;
6745 const char *poolname
, *cachefile
;
6749 zpool_load_policy_t policy
;
6751 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
6754 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
6758 * Create and initialize the spa structure.
6760 mutex_enter(&spa_namespace_lock
);
6761 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
6762 spa_activate(spa
, SPA_MODE_READ
);
6765 * Rewind pool if a max txg was provided.
6767 zpool_get_load_policy(spa
->spa_config
, &policy
);
6768 if (policy
.zlp_txg
!= UINT64_MAX
) {
6769 spa
->spa_load_max_txg
= policy
.zlp_txg
;
6770 spa
->spa_extreme_rewind
= B_TRUE
;
6771 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
6772 poolname
, (longlong_t
)policy
.zlp_txg
);
6774 zfs_dbgmsg("spa_tryimport: importing %s", poolname
);
6777 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_CACHEFILE
, &cachefile
)
6779 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile
);
6780 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
6782 spa
->spa_config_source
= SPA_CONFIG_SRC_SCAN
;
6786 * spa_import() relies on a pool config fetched by spa_try_import()
6787 * for spare/cache devices. Import flags are not passed to
6788 * spa_tryimport(), which makes it return early due to a missing log
6789 * device and missing retrieving the cache device and spare eventually.
6790 * Passing ZFS_IMPORT_MISSING_LOG to spa_tryimport() makes it fetch
6791 * the correct configuration regardless of the missing log device.
6793 spa
->spa_import_flags
|= ZFS_IMPORT_MISSING_LOG
;
6795 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
);
6798 * If 'tryconfig' was at least parsable, return the current config.
6800 if (spa
->spa_root_vdev
!= NULL
) {
6801 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
6802 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, poolname
);
6803 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, state
);
6804 fnvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
6805 spa
->spa_uberblock
.ub_timestamp
);
6806 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
6807 spa
->spa_load_info
);
6808 fnvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
6812 * If the bootfs property exists on this pool then we
6813 * copy it out so that external consumers can tell which
6814 * pools are bootable.
6816 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
6817 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6820 * We have to play games with the name since the
6821 * pool was opened as TRYIMPORT_NAME.
6823 if (dsl_dsobj_to_dsname(spa_name(spa
),
6824 spa
->spa_bootfs
, tmpname
) == 0) {
6828 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
6830 cp
= strchr(tmpname
, '/');
6832 (void) strlcpy(dsname
, tmpname
,
6835 (void) snprintf(dsname
, MAXPATHLEN
,
6836 "%s/%s", poolname
, ++cp
);
6838 fnvlist_add_string(config
, ZPOOL_CONFIG_BOOTFS
,
6840 kmem_free(dsname
, MAXPATHLEN
);
6842 kmem_free(tmpname
, MAXPATHLEN
);
6846 * Add the list of hot spares and level 2 cache devices.
6848 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6849 spa_add_spares(spa
, config
);
6850 spa_add_l2cache(spa
, config
);
6851 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6855 spa_deactivate(spa
);
6857 mutex_exit(&spa_namespace_lock
);
6863 * Pool export/destroy
6865 * The act of destroying or exporting a pool is very simple. We make sure there
6866 * is no more pending I/O and any references to the pool are gone. Then, we
6867 * update the pool state and sync all the labels to disk, removing the
6868 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
6869 * we don't sync the labels or remove the configuration cache.
6872 spa_export_common(const char *pool
, int new_state
, nvlist_t
**oldconfig
,
6873 boolean_t force
, boolean_t hardforce
)
6881 if (!(spa_mode_global
& SPA_MODE_WRITE
))
6882 return (SET_ERROR(EROFS
));
6884 mutex_enter(&spa_namespace_lock
);
6885 if ((spa
= spa_lookup(pool
)) == NULL
) {
6886 mutex_exit(&spa_namespace_lock
);
6887 return (SET_ERROR(ENOENT
));
6890 if (spa
->spa_is_exporting
) {
6891 /* the pool is being exported by another thread */
6892 mutex_exit(&spa_namespace_lock
);
6893 return (SET_ERROR(ZFS_ERR_EXPORT_IN_PROGRESS
));
6895 spa
->spa_is_exporting
= B_TRUE
;
6898 * Put a hold on the pool, drop the namespace lock, stop async tasks,
6899 * reacquire the namespace lock, and see if we can export.
6901 spa_open_ref(spa
, FTAG
);
6902 mutex_exit(&spa_namespace_lock
);
6903 spa_async_suspend(spa
);
6904 if (spa
->spa_zvol_taskq
) {
6905 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
6906 taskq_wait(spa
->spa_zvol_taskq
);
6908 mutex_enter(&spa_namespace_lock
);
6909 spa_close(spa
, FTAG
);
6911 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
6914 * The pool will be in core if it's openable, in which case we can
6915 * modify its state. Objsets may be open only because they're dirty,
6916 * so we have to force it to sync before checking spa_refcnt.
6918 if (spa
->spa_sync_on
) {
6919 txg_wait_synced(spa
->spa_dsl_pool
, 0);
6920 spa_evicting_os_wait(spa
);
6924 * A pool cannot be exported or destroyed if there are active
6925 * references. If we are resetting a pool, allow references by
6926 * fault injection handlers.
6928 if (!spa_refcount_zero(spa
) || (spa
->spa_inject_ref
!= 0)) {
6929 error
= SET_ERROR(EBUSY
);
6933 if (spa
->spa_sync_on
) {
6934 vdev_t
*rvd
= spa
->spa_root_vdev
;
6936 * A pool cannot be exported if it has an active shared spare.
6937 * This is to prevent other pools stealing the active spare
6938 * from an exported pool. At user's own will, such pool can
6939 * be forcedly exported.
6941 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
6942 spa_has_active_shared_spare(spa
)) {
6943 error
= SET_ERROR(EXDEV
);
6948 * We're about to export or destroy this pool. Make sure
6949 * we stop all initialization and trim activity here before
6950 * we set the spa_final_txg. This will ensure that all
6951 * dirty data resulting from the initialization is
6952 * committed to disk before we unload the pool.
6954 vdev_initialize_stop_all(rvd
, VDEV_INITIALIZE_ACTIVE
);
6955 vdev_trim_stop_all(rvd
, VDEV_TRIM_ACTIVE
);
6956 vdev_autotrim_stop_all(spa
);
6957 vdev_rebuild_stop_all(spa
);
6960 * We want this to be reflected on every label,
6961 * so mark them all dirty. spa_unload() will do the
6962 * final sync that pushes these changes out.
6964 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
6965 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6966 spa
->spa_state
= new_state
;
6967 vdev_config_dirty(rvd
);
6968 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6972 * If the log space map feature is enabled and the pool is
6973 * getting exported (but not destroyed), we want to spend some
6974 * time flushing as many metaslabs as we can in an attempt to
6975 * destroy log space maps and save import time. This has to be
6976 * done before we set the spa_final_txg, otherwise
6977 * spa_sync() -> spa_flush_metaslabs() may dirty the final TXGs.
6978 * spa_should_flush_logs_on_unload() should be called after
6979 * spa_state has been set to the new_state.
6981 if (spa_should_flush_logs_on_unload(spa
))
6982 spa_unload_log_sm_flush_all(spa
);
6984 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
6985 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6986 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
6988 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6995 if (new_state
== POOL_STATE_DESTROYED
)
6996 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
6997 else if (new_state
== POOL_STATE_EXPORTED
)
6998 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
7000 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7002 spa_deactivate(spa
);
7005 if (oldconfig
&& spa
->spa_config
)
7006 *oldconfig
= fnvlist_dup(spa
->spa_config
);
7008 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
7010 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
, B_FALSE
);
7014 * If spa_remove() is not called for this spa_t and
7015 * there is any possibility that it can be reused,
7016 * we make sure to reset the exporting flag.
7018 spa
->spa_is_exporting
= B_FALSE
;
7021 mutex_exit(&spa_namespace_lock
);
7025 spa
->spa_is_exporting
= B_FALSE
;
7026 spa_async_resume(spa
);
7027 mutex_exit(&spa_namespace_lock
);
7032 * Destroy a storage pool.
7035 spa_destroy(const char *pool
)
7037 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
7042 * Export a storage pool.
7045 spa_export(const char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
7046 boolean_t hardforce
)
7048 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
7053 * Similar to spa_export(), this unloads the spa_t without actually removing it
7054 * from the namespace in any way.
7057 spa_reset(const char *pool
)
7059 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
7064 * ==========================================================================
7065 * Device manipulation
7066 * ==========================================================================
7070 * This is called as a synctask to increment the draid feature flag
7073 spa_draid_feature_incr(void *arg
, dmu_tx_t
*tx
)
7075 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
7076 int draid
= (int)(uintptr_t)arg
;
7078 for (int c
= 0; c
< draid
; c
++)
7079 spa_feature_incr(spa
, SPA_FEATURE_DRAID
, tx
);
7083 * Add a device to a storage pool.
7086 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
7088 uint64_t txg
, ndraid
= 0;
7090 vdev_t
*rvd
= spa
->spa_root_vdev
;
7092 nvlist_t
**spares
, **l2cache
;
7093 uint_t nspares
, nl2cache
;
7095 ASSERT(spa_writeable(spa
));
7097 txg
= spa_vdev_enter(spa
);
7099 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
7100 VDEV_ALLOC_ADD
)) != 0)
7101 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7103 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
7105 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
7109 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
7113 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
7114 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
7116 if (vd
->vdev_children
!= 0 &&
7117 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0) {
7118 return (spa_vdev_exit(spa
, vd
, txg
, error
));
7122 * The virtual dRAID spares must be added after vdev tree is created
7123 * and the vdev guids are generated. The guid of their associated
7124 * dRAID is stored in the config and used when opening the spare.
7126 if ((error
= vdev_draid_spare_create(nvroot
, vd
, &ndraid
,
7127 rvd
->vdev_children
)) == 0) {
7128 if (ndraid
> 0 && nvlist_lookup_nvlist_array(nvroot
,
7129 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) != 0)
7132 return (spa_vdev_exit(spa
, vd
, txg
, error
));
7136 * We must validate the spares and l2cache devices after checking the
7137 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
7139 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
7140 return (spa_vdev_exit(spa
, vd
, txg
, error
));
7143 * If we are in the middle of a device removal, we can only add
7144 * devices which match the existing devices in the pool.
7145 * If we are in the middle of a removal, or have some indirect
7146 * vdevs, we can not add raidz or dRAID top levels.
7148 if (spa
->spa_vdev_removal
!= NULL
||
7149 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
7150 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
7151 tvd
= vd
->vdev_child
[c
];
7152 if (spa
->spa_vdev_removal
!= NULL
&&
7153 tvd
->vdev_ashift
!= spa
->spa_max_ashift
) {
7154 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
7156 /* Fail if top level vdev is raidz or a dRAID */
7157 if (vdev_get_nparity(tvd
) != 0)
7158 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
7161 * Need the top level mirror to be
7162 * a mirror of leaf vdevs only
7164 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
7165 for (uint64_t cid
= 0;
7166 cid
< tvd
->vdev_children
; cid
++) {
7167 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
7168 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
7169 return (spa_vdev_exit(spa
, vd
,
7177 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
7178 tvd
= vd
->vdev_child
[c
];
7179 vdev_remove_child(vd
, tvd
);
7180 tvd
->vdev_id
= rvd
->vdev_children
;
7181 vdev_add_child(rvd
, tvd
);
7182 vdev_config_dirty(tvd
);
7186 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
7187 ZPOOL_CONFIG_SPARES
);
7188 spa_load_spares(spa
);
7189 spa
->spa_spares
.sav_sync
= B_TRUE
;
7192 if (nl2cache
!= 0) {
7193 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
7194 ZPOOL_CONFIG_L2CACHE
);
7195 spa_load_l2cache(spa
);
7196 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
7200 * We can't increment a feature while holding spa_vdev so we
7201 * have to do it in a synctask.
7206 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
7207 dsl_sync_task_nowait(spa
->spa_dsl_pool
, spa_draid_feature_incr
,
7208 (void *)(uintptr_t)ndraid
, tx
);
7213 * We have to be careful when adding new vdevs to an existing pool.
7214 * If other threads start allocating from these vdevs before we
7215 * sync the config cache, and we lose power, then upon reboot we may
7216 * fail to open the pool because there are DVAs that the config cache
7217 * can't translate. Therefore, we first add the vdevs without
7218 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
7219 * and then let spa_config_update() initialize the new metaslabs.
7221 * spa_load() checks for added-but-not-initialized vdevs, so that
7222 * if we lose power at any point in this sequence, the remaining
7223 * steps will be completed the next time we load the pool.
7225 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
7227 mutex_enter(&spa_namespace_lock
);
7228 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
7229 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
7230 mutex_exit(&spa_namespace_lock
);
7236 * Attach a device to a vdev specified by its guid. The vdev type can be
7237 * a mirror, a raidz, or a leaf device that is also a top-level (e.g. a
7238 * single device). When the vdev is a single device, a mirror vdev will be
7239 * automatically inserted.
7241 * If 'replacing' is specified, the new device is intended to replace the
7242 * existing device; in this case the two devices are made into their own
7243 * mirror using the 'replacing' vdev, which is functionally identical to
7244 * the mirror vdev (it actually reuses all the same ops) but has a few
7245 * extra rules: you can't attach to it after it's been created, and upon
7246 * completion of resilvering, the first disk (the one being replaced)
7247 * is automatically detached.
7249 * If 'rebuild' is specified, then sequential reconstruction (a.ka. rebuild)
7250 * should be performed instead of traditional healing reconstruction. From
7251 * an administrators perspective these are both resilver operations.
7254 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
,
7257 uint64_t txg
, dtl_max_txg
;
7258 vdev_t
*rvd
= spa
->spa_root_vdev
;
7259 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
7261 char *oldvdpath
, *newvdpath
;
7262 int newvd_isspare
= B_FALSE
;
7265 ASSERT(spa_writeable(spa
));
7267 txg
= spa_vdev_enter(spa
);
7269 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7271 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7272 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7273 error
= (spa_has_checkpoint(spa
)) ?
7274 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7275 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7279 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_DEVICE_REBUILD
))
7280 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7282 if (dsl_scan_resilvering(spa_get_dsl(spa
)) ||
7283 dsl_scan_resilver_scheduled(spa_get_dsl(spa
))) {
7284 return (spa_vdev_exit(spa
, NULL
, txg
,
7285 ZFS_ERR_RESILVER_IN_PROGRESS
));
7288 if (vdev_rebuild_active(rvd
))
7289 return (spa_vdev_exit(spa
, NULL
, txg
,
7290 ZFS_ERR_REBUILD_IN_PROGRESS
));
7293 if (spa
->spa_vdev_removal
!= NULL
) {
7294 return (spa_vdev_exit(spa
, NULL
, txg
,
7295 ZFS_ERR_DEVRM_IN_PROGRESS
));
7299 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
7301 boolean_t raidz
= oldvd
->vdev_ops
== &vdev_raidz_ops
;
7304 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_RAIDZ_EXPANSION
))
7305 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7308 * Can't expand a raidz while prior expand is in progress.
7310 if (spa
->spa_raidz_expand
!= NULL
) {
7311 return (spa_vdev_exit(spa
, NULL
, txg
,
7312 ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS
));
7314 } else if (!oldvd
->vdev_ops
->vdev_op_leaf
) {
7315 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7321 pvd
= oldvd
->vdev_parent
;
7323 if (spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
7324 VDEV_ALLOC_ATTACH
) != 0)
7325 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
7327 if (newrootvd
->vdev_children
!= 1)
7328 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
7330 newvd
= newrootvd
->vdev_child
[0];
7332 if (!newvd
->vdev_ops
->vdev_op_leaf
)
7333 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
7335 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
7336 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
7339 * log, dedup and special vdevs should not be replaced by spares.
7341 if ((oldvd
->vdev_top
->vdev_alloc_bias
!= VDEV_BIAS_NONE
||
7342 oldvd
->vdev_top
->vdev_islog
) && newvd
->vdev_isspare
) {
7343 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7347 * A dRAID spare can only replace a child of its parent dRAID vdev.
7349 if (newvd
->vdev_ops
== &vdev_draid_spare_ops
&&
7350 oldvd
->vdev_top
!= vdev_draid_spare_get_parent(newvd
)) {
7351 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7356 * For rebuilds, the top vdev must support reconstruction
7357 * using only space maps. This means the only allowable
7358 * vdevs types are the root vdev, a mirror, or dRAID.
7361 if (pvd
->vdev_top
!= NULL
)
7362 tvd
= pvd
->vdev_top
;
7364 if (tvd
->vdev_ops
!= &vdev_mirror_ops
&&
7365 tvd
->vdev_ops
!= &vdev_root_ops
&&
7366 tvd
->vdev_ops
!= &vdev_draid_ops
) {
7367 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7373 * For attach, the only allowable parent is a mirror or
7374 * the root vdev. A raidz vdev can be attached to, but
7375 * you cannot attach to a raidz child.
7377 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
7378 pvd
->vdev_ops
!= &vdev_root_ops
&&
7380 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7382 pvops
= &vdev_mirror_ops
;
7385 * Active hot spares can only be replaced by inactive hot
7388 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
7389 oldvd
->vdev_isspare
&&
7390 !spa_has_spare(spa
, newvd
->vdev_guid
))
7391 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7394 * If the source is a hot spare, and the parent isn't already a
7395 * spare, then we want to create a new hot spare. Otherwise, we
7396 * want to create a replacing vdev. The user is not allowed to
7397 * attach to a spared vdev child unless the 'isspare' state is
7398 * the same (spare replaces spare, non-spare replaces
7401 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
7402 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
7403 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7404 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
7405 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
7406 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7409 if (newvd
->vdev_isspare
)
7410 pvops
= &vdev_spare_ops
;
7412 pvops
= &vdev_replacing_ops
;
7416 * Make sure the new device is big enough.
7418 vdev_t
*min_vdev
= raidz
? oldvd
->vdev_child
[0] : oldvd
;
7419 if (newvd
->vdev_asize
< vdev_get_min_asize(min_vdev
))
7420 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
7423 * The new device cannot have a higher alignment requirement
7424 * than the top-level vdev.
7426 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
7427 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7430 * RAIDZ-expansion-specific checks.
7433 if (vdev_raidz_attach_check(newvd
) != 0)
7434 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
7437 * Fail early if a child is not healthy or being replaced
7439 for (int i
= 0; i
< oldvd
->vdev_children
; i
++) {
7440 if (vdev_is_dead(oldvd
->vdev_child
[i
]) ||
7441 !oldvd
->vdev_child
[i
]->vdev_ops
->vdev_op_leaf
) {
7442 return (spa_vdev_exit(spa
, newrootvd
, txg
,
7445 /* Also fail if reserved boot area is in-use */
7446 if (vdev_check_boot_reserve(spa
, oldvd
->vdev_child
[i
])
7448 return (spa_vdev_exit(spa
, newrootvd
, txg
,
7456 * Note: oldvdpath is freed by spa_strfree(), but
7457 * kmem_asprintf() is freed by kmem_strfree(), so we have to
7458 * move it to a spa_strdup-ed string.
7460 char *tmp
= kmem_asprintf("raidz%u-%u",
7461 (uint_t
)vdev_get_nparity(oldvd
), (uint_t
)oldvd
->vdev_id
);
7462 oldvdpath
= spa_strdup(tmp
);
7465 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
7467 newvdpath
= spa_strdup(newvd
->vdev_path
);
7470 * If this is an in-place replacement, update oldvd's path and devid
7471 * to make it distinguishable from newvd, and unopenable from now on.
7473 if (strcmp(oldvdpath
, newvdpath
) == 0) {
7474 spa_strfree(oldvd
->vdev_path
);
7475 oldvd
->vdev_path
= kmem_alloc(strlen(newvdpath
) + 5,
7477 (void) sprintf(oldvd
->vdev_path
, "%s/old",
7479 if (oldvd
->vdev_devid
!= NULL
) {
7480 spa_strfree(oldvd
->vdev_devid
);
7481 oldvd
->vdev_devid
= NULL
;
7483 spa_strfree(oldvdpath
);
7484 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
7488 * If the parent is not a mirror, or if we're replacing, insert the new
7489 * mirror/replacing/spare vdev above oldvd.
7491 if (!raidz
&& pvd
->vdev_ops
!= pvops
) {
7492 pvd
= vdev_add_parent(oldvd
, pvops
);
7493 ASSERT(pvd
->vdev_ops
== pvops
);
7494 ASSERT(oldvd
->vdev_parent
== pvd
);
7497 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
7500 * Extract the new device from its root and add it to pvd.
7502 vdev_remove_child(newrootvd
, newvd
);
7503 newvd
->vdev_id
= pvd
->vdev_children
;
7504 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
7505 vdev_add_child(pvd
, newvd
);
7508 * Reevaluate the parent vdev state.
7510 vdev_propagate_state(pvd
);
7512 tvd
= newvd
->vdev_top
;
7513 ASSERT(pvd
->vdev_top
== tvd
);
7514 ASSERT(tvd
->vdev_parent
== rvd
);
7516 vdev_config_dirty(tvd
);
7519 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
7520 * for any dmu_sync-ed blocks. It will propagate upward when
7521 * spa_vdev_exit() calls vdev_dtl_reassess().
7523 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
7527 * Wait for the youngest allocations and frees to sync,
7528 * and then wait for the deferral of those frees to finish.
7530 spa_vdev_config_exit(spa
, NULL
,
7531 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
7533 vdev_initialize_stop_all(tvd
, VDEV_INITIALIZE_ACTIVE
);
7534 vdev_trim_stop_all(tvd
, VDEV_TRIM_ACTIVE
);
7535 vdev_autotrim_stop_wait(tvd
);
7537 dtl_max_txg
= spa_vdev_config_enter(spa
);
7539 tvd
->vdev_rz_expanding
= B_TRUE
;
7541 vdev_dirty_leaves(tvd
, VDD_DTL
, dtl_max_txg
);
7542 vdev_config_dirty(tvd
);
7544 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
,
7546 dsl_sync_task_nowait(spa
->spa_dsl_pool
, vdev_raidz_attach_sync
,
7550 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
7551 dtl_max_txg
- TXG_INITIAL
);
7553 if (newvd
->vdev_isspare
) {
7554 spa_spare_activate(newvd
);
7555 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
7558 newvd_isspare
= newvd
->vdev_isspare
;
7561 * Mark newvd's DTL dirty in this txg.
7563 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
7566 * Schedule the resilver or rebuild to restart in the future.
7567 * We do this to ensure that dmu_sync-ed blocks have been
7568 * stitched into the respective datasets.
7571 newvd
->vdev_rebuild_txg
= txg
;
7575 newvd
->vdev_resilver_txg
= txg
;
7577 if (dsl_scan_resilvering(spa_get_dsl(spa
)) &&
7578 spa_feature_is_enabled(spa
,
7579 SPA_FEATURE_RESILVER_DEFER
)) {
7580 vdev_defer_resilver(newvd
);
7582 dsl_scan_restart_resilver(spa
->spa_dsl_pool
,
7588 if (spa
->spa_bootfs
)
7589 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
7591 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
7596 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
7598 spa_history_log_internal(spa
, "vdev attach", NULL
,
7599 "%s vdev=%s %s vdev=%s",
7600 replacing
&& newvd_isspare
? "spare in" :
7601 replacing
? "replace" : "attach", newvdpath
,
7602 replacing
? "for" : "to", oldvdpath
);
7604 spa_strfree(oldvdpath
);
7605 spa_strfree(newvdpath
);
7611 * Detach a device from a mirror or replacing vdev.
7613 * If 'replace_done' is specified, only detach if the parent
7614 * is a replacing or a spare vdev.
7617 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
7621 vdev_t
*rvd __maybe_unused
= spa
->spa_root_vdev
;
7622 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
7623 boolean_t unspare
= B_FALSE
;
7624 uint64_t unspare_guid
= 0;
7627 ASSERT(spa_writeable(spa
));
7629 txg
= spa_vdev_detach_enter(spa
, guid
);
7631 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7634 * Besides being called directly from the userland through the
7635 * ioctl interface, spa_vdev_detach() can be potentially called
7636 * at the end of spa_vdev_resilver_done().
7638 * In the regular case, when we have a checkpoint this shouldn't
7639 * happen as we never empty the DTLs of a vdev during the scrub
7640 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
7641 * should never get here when we have a checkpoint.
7643 * That said, even in a case when we checkpoint the pool exactly
7644 * as spa_vdev_resilver_done() calls this function everything
7645 * should be fine as the resilver will return right away.
7647 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7648 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
7649 error
= (spa_has_checkpoint(spa
)) ?
7650 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
7651 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
7655 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
7657 if (!vd
->vdev_ops
->vdev_op_leaf
)
7658 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7660 pvd
= vd
->vdev_parent
;
7663 * If the parent/child relationship is not as expected, don't do it.
7664 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
7665 * vdev that's replacing B with C. The user's intent in replacing
7666 * is to go from M(A,B) to M(A,C). If the user decides to cancel
7667 * the replace by detaching C, the expected behavior is to end up
7668 * M(A,B). But suppose that right after deciding to detach C,
7669 * the replacement of B completes. We would have M(A,C), and then
7670 * ask to detach C, which would leave us with just A -- not what
7671 * the user wanted. To prevent this, we make sure that the
7672 * parent/child relationship hasn't changed -- in this example,
7673 * that C's parent is still the replacing vdev R.
7675 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
7676 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
7679 * Only 'replacing' or 'spare' vdevs can be replaced.
7681 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
7682 pvd
->vdev_ops
!= &vdev_spare_ops
)
7683 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7685 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
7686 spa_version(spa
) >= SPA_VERSION_SPARES
);
7689 * Only mirror, replacing, and spare vdevs support detach.
7691 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
7692 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
7693 pvd
->vdev_ops
!= &vdev_spare_ops
)
7694 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
7697 * If this device has the only valid copy of some data,
7698 * we cannot safely detach it.
7700 if (vdev_dtl_required(vd
))
7701 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
7703 ASSERT(pvd
->vdev_children
>= 2);
7706 * If we are detaching the second disk from a replacing vdev, then
7707 * check to see if we changed the original vdev's path to have "/old"
7708 * at the end in spa_vdev_attach(). If so, undo that change now.
7710 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
7711 vd
->vdev_path
!= NULL
) {
7712 size_t len
= strlen(vd
->vdev_path
);
7714 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
7715 cvd
= pvd
->vdev_child
[c
];
7717 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
7720 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
7721 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
7722 spa_strfree(cvd
->vdev_path
);
7723 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
7730 * If we are detaching the original disk from a normal spare, then it
7731 * implies that the spare should become a real disk, and be removed
7732 * from the active spare list for the pool. dRAID spares on the
7733 * other hand are coupled to the pool and thus should never be removed
7734 * from the spares list.
7736 if (pvd
->vdev_ops
== &vdev_spare_ops
&& vd
->vdev_id
== 0) {
7737 vdev_t
*last_cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7739 if (last_cvd
->vdev_isspare
&&
7740 last_cvd
->vdev_ops
!= &vdev_draid_spare_ops
) {
7746 * Erase the disk labels so the disk can be used for other things.
7747 * This must be done after all other error cases are handled,
7748 * but before we disembowel vd (so we can still do I/O to it).
7749 * But if we can't do it, don't treat the error as fatal --
7750 * it may be that the unwritability of the disk is the reason
7751 * it's being detached!
7753 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
7756 * Remove vd from its parent and compact the parent's children.
7758 vdev_remove_child(pvd
, vd
);
7759 vdev_compact_children(pvd
);
7762 * Remember one of the remaining children so we can get tvd below.
7764 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
7767 * If we need to remove the remaining child from the list of hot spares,
7768 * do it now, marking the vdev as no longer a spare in the process.
7769 * We must do this before vdev_remove_parent(), because that can
7770 * change the GUID if it creates a new toplevel GUID. For a similar
7771 * reason, we must remove the spare now, in the same txg as the detach;
7772 * otherwise someone could attach a new sibling, change the GUID, and
7773 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
7776 ASSERT(cvd
->vdev_isspare
);
7777 spa_spare_remove(cvd
);
7778 unspare_guid
= cvd
->vdev_guid
;
7779 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
7780 cvd
->vdev_unspare
= B_TRUE
;
7784 * If the parent mirror/replacing vdev only has one child,
7785 * the parent is no longer needed. Remove it from the tree.
7787 if (pvd
->vdev_children
== 1) {
7788 if (pvd
->vdev_ops
== &vdev_spare_ops
)
7789 cvd
->vdev_unspare
= B_FALSE
;
7790 vdev_remove_parent(cvd
);
7794 * We don't set tvd until now because the parent we just removed
7795 * may have been the previous top-level vdev.
7797 tvd
= cvd
->vdev_top
;
7798 ASSERT(tvd
->vdev_parent
== rvd
);
7801 * Reevaluate the parent vdev state.
7803 vdev_propagate_state(cvd
);
7806 * If the 'autoexpand' property is set on the pool then automatically
7807 * try to expand the size of the pool. For example if the device we
7808 * just detached was smaller than the others, it may be possible to
7809 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
7810 * first so that we can obtain the updated sizes of the leaf vdevs.
7812 if (spa
->spa_autoexpand
) {
7814 vdev_expand(tvd
, txg
);
7817 vdev_config_dirty(tvd
);
7820 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
7821 * vd->vdev_detached is set and free vd's DTL object in syncing context.
7822 * But first make sure we're not on any *other* txg's DTL list, to
7823 * prevent vd from being accessed after it's freed.
7825 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
7826 for (int t
= 0; t
< TXG_SIZE
; t
++)
7827 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
7828 vd
->vdev_detached
= B_TRUE
;
7829 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
7831 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
7832 spa_notify_waiters(spa
);
7834 /* hang on to the spa before we release the lock */
7835 spa_open_ref(spa
, FTAG
);
7837 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
7839 spa_history_log_internal(spa
, "detach", NULL
,
7841 spa_strfree(vdpath
);
7844 * If this was the removal of the original device in a hot spare vdev,
7845 * then we want to go through and remove the device from the hot spare
7846 * list of every other pool.
7849 spa_t
*altspa
= NULL
;
7851 mutex_enter(&spa_namespace_lock
);
7852 while ((altspa
= spa_next(altspa
)) != NULL
) {
7853 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
7857 spa_open_ref(altspa
, FTAG
);
7858 mutex_exit(&spa_namespace_lock
);
7859 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
7860 mutex_enter(&spa_namespace_lock
);
7861 spa_close(altspa
, FTAG
);
7863 mutex_exit(&spa_namespace_lock
);
7865 /* search the rest of the vdevs for spares to remove */
7866 spa_vdev_resilver_done(spa
);
7869 /* all done with the spa; OK to release */
7870 mutex_enter(&spa_namespace_lock
);
7871 spa_close(spa
, FTAG
);
7872 mutex_exit(&spa_namespace_lock
);
7878 spa_vdev_initialize_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7881 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
7883 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7885 /* Look up vdev and ensure it's a leaf. */
7886 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
7887 if (vd
== NULL
|| vd
->vdev_detached
) {
7888 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7889 return (SET_ERROR(ENODEV
));
7890 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
7891 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7892 return (SET_ERROR(EINVAL
));
7893 } else if (!vdev_writeable(vd
)) {
7894 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7895 return (SET_ERROR(EROFS
));
7897 mutex_enter(&vd
->vdev_initialize_lock
);
7898 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7901 * When we activate an initialize action we check to see
7902 * if the vdev_initialize_thread is NULL. We do this instead
7903 * of using the vdev_initialize_state since there might be
7904 * a previous initialization process which has completed but
7905 * the thread is not exited.
7907 if (cmd_type
== POOL_INITIALIZE_START
&&
7908 (vd
->vdev_initialize_thread
!= NULL
||
7909 vd
->vdev_top
->vdev_removing
|| vd
->vdev_top
->vdev_rz_expanding
)) {
7910 mutex_exit(&vd
->vdev_initialize_lock
);
7911 return (SET_ERROR(EBUSY
));
7912 } else if (cmd_type
== POOL_INITIALIZE_CANCEL
&&
7913 (vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
&&
7914 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_SUSPENDED
)) {
7915 mutex_exit(&vd
->vdev_initialize_lock
);
7916 return (SET_ERROR(ESRCH
));
7917 } else if (cmd_type
== POOL_INITIALIZE_SUSPEND
&&
7918 vd
->vdev_initialize_state
!= VDEV_INITIALIZE_ACTIVE
) {
7919 mutex_exit(&vd
->vdev_initialize_lock
);
7920 return (SET_ERROR(ESRCH
));
7921 } else if (cmd_type
== POOL_INITIALIZE_UNINIT
&&
7922 vd
->vdev_initialize_thread
!= NULL
) {
7923 mutex_exit(&vd
->vdev_initialize_lock
);
7924 return (SET_ERROR(EBUSY
));
7928 case POOL_INITIALIZE_START
:
7929 vdev_initialize(vd
);
7931 case POOL_INITIALIZE_CANCEL
:
7932 vdev_initialize_stop(vd
, VDEV_INITIALIZE_CANCELED
, vd_list
);
7934 case POOL_INITIALIZE_SUSPEND
:
7935 vdev_initialize_stop(vd
, VDEV_INITIALIZE_SUSPENDED
, vd_list
);
7937 case POOL_INITIALIZE_UNINIT
:
7938 vdev_uninitialize(vd
);
7941 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
7943 mutex_exit(&vd
->vdev_initialize_lock
);
7949 spa_vdev_initialize(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
,
7950 nvlist_t
*vdev_errlist
)
7952 int total_errors
= 0;
7955 list_create(&vd_list
, sizeof (vdev_t
),
7956 offsetof(vdev_t
, vdev_initialize_node
));
7959 * We hold the namespace lock through the whole function
7960 * to prevent any changes to the pool while we're starting or
7961 * stopping initialization. The config and state locks are held so that
7962 * we can properly assess the vdev state before we commit to
7963 * the initializing operation.
7965 mutex_enter(&spa_namespace_lock
);
7967 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
7968 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
7969 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
7971 int error
= spa_vdev_initialize_impl(spa
, vdev_guid
, cmd_type
,
7974 char guid_as_str
[MAXNAMELEN
];
7976 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
7977 "%llu", (unsigned long long)vdev_guid
);
7978 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
7983 /* Wait for all initialize threads to stop. */
7984 vdev_initialize_stop_wait(spa
, &vd_list
);
7986 /* Sync out the initializing state */
7987 txg_wait_synced(spa
->spa_dsl_pool
, 0);
7988 mutex_exit(&spa_namespace_lock
);
7990 list_destroy(&vd_list
);
7992 return (total_errors
);
7996 spa_vdev_trim_impl(spa_t
*spa
, uint64_t guid
, uint64_t cmd_type
,
7997 uint64_t rate
, boolean_t partial
, boolean_t secure
, list_t
*vd_list
)
7999 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
8001 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
8003 /* Look up vdev and ensure it's a leaf. */
8004 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
8005 if (vd
== NULL
|| vd
->vdev_detached
) {
8006 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
8007 return (SET_ERROR(ENODEV
));
8008 } else if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(vd
)) {
8009 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
8010 return (SET_ERROR(EINVAL
));
8011 } else if (!vdev_writeable(vd
)) {
8012 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
8013 return (SET_ERROR(EROFS
));
8014 } else if (!vd
->vdev_has_trim
) {
8015 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
8016 return (SET_ERROR(EOPNOTSUPP
));
8017 } else if (secure
&& !vd
->vdev_has_securetrim
) {
8018 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
8019 return (SET_ERROR(EOPNOTSUPP
));
8021 mutex_enter(&vd
->vdev_trim_lock
);
8022 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
8025 * When we activate a TRIM action we check to see if the
8026 * vdev_trim_thread is NULL. We do this instead of using the
8027 * vdev_trim_state since there might be a previous TRIM process
8028 * which has completed but the thread is not exited.
8030 if (cmd_type
== POOL_TRIM_START
&&
8031 (vd
->vdev_trim_thread
!= NULL
|| vd
->vdev_top
->vdev_removing
||
8032 vd
->vdev_top
->vdev_rz_expanding
)) {
8033 mutex_exit(&vd
->vdev_trim_lock
);
8034 return (SET_ERROR(EBUSY
));
8035 } else if (cmd_type
== POOL_TRIM_CANCEL
&&
8036 (vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
&&
8037 vd
->vdev_trim_state
!= VDEV_TRIM_SUSPENDED
)) {
8038 mutex_exit(&vd
->vdev_trim_lock
);
8039 return (SET_ERROR(ESRCH
));
8040 } else if (cmd_type
== POOL_TRIM_SUSPEND
&&
8041 vd
->vdev_trim_state
!= VDEV_TRIM_ACTIVE
) {
8042 mutex_exit(&vd
->vdev_trim_lock
);
8043 return (SET_ERROR(ESRCH
));
8047 case POOL_TRIM_START
:
8048 vdev_trim(vd
, rate
, partial
, secure
);
8050 case POOL_TRIM_CANCEL
:
8051 vdev_trim_stop(vd
, VDEV_TRIM_CANCELED
, vd_list
);
8053 case POOL_TRIM_SUSPEND
:
8054 vdev_trim_stop(vd
, VDEV_TRIM_SUSPENDED
, vd_list
);
8057 panic("invalid cmd_type %llu", (unsigned long long)cmd_type
);
8059 mutex_exit(&vd
->vdev_trim_lock
);
8065 * Initiates a manual TRIM for the requested vdevs. This kicks off individual
8066 * TRIM threads for each child vdev. These threads pass over all of the free
8067 * space in the vdev's metaslabs and issues TRIM commands for that space.
8070 spa_vdev_trim(spa_t
*spa
, nvlist_t
*nv
, uint64_t cmd_type
, uint64_t rate
,
8071 boolean_t partial
, boolean_t secure
, nvlist_t
*vdev_errlist
)
8073 int total_errors
= 0;
8076 list_create(&vd_list
, sizeof (vdev_t
),
8077 offsetof(vdev_t
, vdev_trim_node
));
8080 * We hold the namespace lock through the whole function
8081 * to prevent any changes to the pool while we're starting or
8082 * stopping TRIM. The config and state locks are held so that
8083 * we can properly assess the vdev state before we commit to
8084 * the TRIM operation.
8086 mutex_enter(&spa_namespace_lock
);
8088 for (nvpair_t
*pair
= nvlist_next_nvpair(nv
, NULL
);
8089 pair
!= NULL
; pair
= nvlist_next_nvpair(nv
, pair
)) {
8090 uint64_t vdev_guid
= fnvpair_value_uint64(pair
);
8092 int error
= spa_vdev_trim_impl(spa
, vdev_guid
, cmd_type
,
8093 rate
, partial
, secure
, &vd_list
);
8095 char guid_as_str
[MAXNAMELEN
];
8097 (void) snprintf(guid_as_str
, sizeof (guid_as_str
),
8098 "%llu", (unsigned long long)vdev_guid
);
8099 fnvlist_add_int64(vdev_errlist
, guid_as_str
, error
);
8104 /* Wait for all TRIM threads to stop. */
8105 vdev_trim_stop_wait(spa
, &vd_list
);
8107 /* Sync out the TRIM state */
8108 txg_wait_synced(spa
->spa_dsl_pool
, 0);
8109 mutex_exit(&spa_namespace_lock
);
8111 list_destroy(&vd_list
);
8113 return (total_errors
);
8117 * Split a set of devices from their mirrors, and create a new pool from them.
8120 spa_vdev_split_mirror(spa_t
*spa
, const char *newname
, nvlist_t
*config
,
8121 nvlist_t
*props
, boolean_t exp
)
8124 uint64_t txg
, *glist
;
8126 uint_t c
, children
, lastlog
;
8127 nvlist_t
**child
, *nvl
, *tmp
;
8129 const char *altroot
= NULL
;
8130 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
8131 boolean_t activate_slog
;
8133 ASSERT(spa_writeable(spa
));
8135 txg
= spa_vdev_enter(spa
);
8137 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
8138 if (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
)) {
8139 error
= (spa_has_checkpoint(spa
)) ?
8140 ZFS_ERR_CHECKPOINT_EXISTS
: ZFS_ERR_DISCARDING_CHECKPOINT
;
8141 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
8144 /* clear the log and flush everything up to now */
8145 activate_slog
= spa_passivate_log(spa
);
8146 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
8147 error
= spa_reset_logs(spa
);
8148 txg
= spa_vdev_config_enter(spa
);
8151 spa_activate_log(spa
);
8154 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
8156 /* check new spa name before going any further */
8157 if (spa_lookup(newname
) != NULL
)
8158 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
8161 * scan through all the children to ensure they're all mirrors
8163 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
8164 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
8166 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
8168 /* first, check to ensure we've got the right child count */
8169 rvd
= spa
->spa_root_vdev
;
8171 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
8172 vdev_t
*vd
= rvd
->vdev_child
[c
];
8174 /* don't count the holes & logs as children */
8175 if (vd
->vdev_islog
|| (vd
->vdev_ops
!= &vdev_indirect_ops
&&
8176 !vdev_is_concrete(vd
))) {
8184 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
8185 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
8187 /* next, ensure no spare or cache devices are part of the split */
8188 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
8189 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
8190 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
8192 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
8193 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
8195 /* then, loop over each vdev and validate it */
8196 for (c
= 0; c
< children
; c
++) {
8197 uint64_t is_hole
= 0;
8199 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
8203 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
8204 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
8207 error
= SET_ERROR(EINVAL
);
8212 /* deal with indirect vdevs */
8213 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ops
==
8217 /* which disk is going to be split? */
8218 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
8220 error
= SET_ERROR(EINVAL
);
8224 /* look it up in the spa */
8225 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
8226 if (vml
[c
] == NULL
) {
8227 error
= SET_ERROR(ENODEV
);
8231 /* make sure there's nothing stopping the split */
8232 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
8233 vml
[c
]->vdev_islog
||
8234 !vdev_is_concrete(vml
[c
]) ||
8235 vml
[c
]->vdev_isspare
||
8236 vml
[c
]->vdev_isl2cache
||
8237 !vdev_writeable(vml
[c
]) ||
8238 vml
[c
]->vdev_children
!= 0 ||
8239 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
8240 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
8241 error
= SET_ERROR(EINVAL
);
8245 if (vdev_dtl_required(vml
[c
]) ||
8246 vdev_resilver_needed(vml
[c
], NULL
, NULL
)) {
8247 error
= SET_ERROR(EBUSY
);
8251 /* we need certain info from the top level */
8252 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
8253 vml
[c
]->vdev_top
->vdev_ms_array
);
8254 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
8255 vml
[c
]->vdev_top
->vdev_ms_shift
);
8256 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
8257 vml
[c
]->vdev_top
->vdev_asize
);
8258 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
8259 vml
[c
]->vdev_top
->vdev_ashift
);
8261 /* transfer per-vdev ZAPs */
8262 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
8263 VERIFY0(nvlist_add_uint64(child
[c
],
8264 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
8266 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
8267 VERIFY0(nvlist_add_uint64(child
[c
],
8268 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
8269 vml
[c
]->vdev_parent
->vdev_top_zap
));
8273 kmem_free(vml
, children
* sizeof (vdev_t
*));
8274 kmem_free(glist
, children
* sizeof (uint64_t));
8275 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
8278 /* stop writers from using the disks */
8279 for (c
= 0; c
< children
; c
++) {
8281 vml
[c
]->vdev_offline
= B_TRUE
;
8283 vdev_reopen(spa
->spa_root_vdev
);
8286 * Temporarily record the splitting vdevs in the spa config. This
8287 * will disappear once the config is regenerated.
8289 nvl
= fnvlist_alloc();
8290 fnvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
, glist
, children
);
8291 kmem_free(glist
, children
* sizeof (uint64_t));
8293 mutex_enter(&spa
->spa_props_lock
);
8294 fnvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
, nvl
);
8295 mutex_exit(&spa
->spa_props_lock
);
8296 spa
->spa_config_splitting
= nvl
;
8297 vdev_config_dirty(spa
->spa_root_vdev
);
8299 /* configure and create the new pool */
8300 fnvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
);
8301 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
8302 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
);
8303 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
, spa_version(spa
));
8304 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, spa
->spa_config_txg
);
8305 fnvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
8306 spa_generate_guid(NULL
));
8307 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
8308 (void) nvlist_lookup_string(props
,
8309 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
8311 /* add the new pool to the namespace */
8312 newspa
= spa_add(newname
, config
, altroot
);
8313 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
8314 newspa
->spa_config_txg
= spa
->spa_config_txg
;
8315 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
8317 /* release the spa config lock, retaining the namespace lock */
8318 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
8320 if (zio_injection_enabled
)
8321 zio_handle_panic_injection(spa
, FTAG
, 1);
8323 spa_activate(newspa
, spa_mode_global
);
8324 spa_async_suspend(newspa
);
8327 * Temporarily stop the initializing and TRIM activity. We set the
8328 * state to ACTIVE so that we know to resume initializing or TRIM
8329 * once the split has completed.
8331 list_t vd_initialize_list
;
8332 list_create(&vd_initialize_list
, sizeof (vdev_t
),
8333 offsetof(vdev_t
, vdev_initialize_node
));
8335 list_t vd_trim_list
;
8336 list_create(&vd_trim_list
, sizeof (vdev_t
),
8337 offsetof(vdev_t
, vdev_trim_node
));
8339 for (c
= 0; c
< children
; c
++) {
8340 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
8341 mutex_enter(&vml
[c
]->vdev_initialize_lock
);
8342 vdev_initialize_stop(vml
[c
],
8343 VDEV_INITIALIZE_ACTIVE
, &vd_initialize_list
);
8344 mutex_exit(&vml
[c
]->vdev_initialize_lock
);
8346 mutex_enter(&vml
[c
]->vdev_trim_lock
);
8347 vdev_trim_stop(vml
[c
], VDEV_TRIM_ACTIVE
, &vd_trim_list
);
8348 mutex_exit(&vml
[c
]->vdev_trim_lock
);
8352 vdev_initialize_stop_wait(spa
, &vd_initialize_list
);
8353 vdev_trim_stop_wait(spa
, &vd_trim_list
);
8355 list_destroy(&vd_initialize_list
);
8356 list_destroy(&vd_trim_list
);
8358 newspa
->spa_config_source
= SPA_CONFIG_SRC_SPLIT
;
8359 newspa
->spa_is_splitting
= B_TRUE
;
8361 /* create the new pool from the disks of the original pool */
8362 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
);
8366 /* if that worked, generate a real config for the new pool */
8367 if (newspa
->spa_root_vdev
!= NULL
) {
8368 newspa
->spa_config_splitting
= fnvlist_alloc();
8369 fnvlist_add_uint64(newspa
->spa_config_splitting
,
8370 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
));
8371 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
8376 if (props
!= NULL
) {
8377 spa_configfile_set(newspa
, props
, B_FALSE
);
8378 error
= spa_prop_set(newspa
, props
);
8383 /* flush everything */
8384 txg
= spa_vdev_config_enter(newspa
);
8385 vdev_config_dirty(newspa
->spa_root_vdev
);
8386 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
8388 if (zio_injection_enabled
)
8389 zio_handle_panic_injection(spa
, FTAG
, 2);
8391 spa_async_resume(newspa
);
8393 /* finally, update the original pool's config */
8394 txg
= spa_vdev_config_enter(spa
);
8395 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
8396 error
= dmu_tx_assign(tx
, TXG_WAIT
);
8399 for (c
= 0; c
< children
; c
++) {
8400 if (vml
[c
] != NULL
&& vml
[c
]->vdev_ops
!= &vdev_indirect_ops
) {
8401 vdev_t
*tvd
= vml
[c
]->vdev_top
;
8404 * Need to be sure the detachable VDEV is not
8405 * on any *other* txg's DTL list to prevent it
8406 * from being accessed after it's freed.
8408 for (int t
= 0; t
< TXG_SIZE
; t
++) {
8409 (void) txg_list_remove_this(
8410 &tvd
->vdev_dtl_list
, vml
[c
], t
);
8415 spa_history_log_internal(spa
, "detach", tx
,
8416 "vdev=%s", vml
[c
]->vdev_path
);
8421 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
8422 vdev_config_dirty(spa
->spa_root_vdev
);
8423 spa
->spa_config_splitting
= NULL
;
8427 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
8429 if (zio_injection_enabled
)
8430 zio_handle_panic_injection(spa
, FTAG
, 3);
8432 /* split is complete; log a history record */
8433 spa_history_log_internal(newspa
, "split", NULL
,
8434 "from pool %s", spa_name(spa
));
8436 newspa
->spa_is_splitting
= B_FALSE
;
8437 kmem_free(vml
, children
* sizeof (vdev_t
*));
8439 /* if we're not going to mount the filesystems in userland, export */
8441 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
8448 spa_deactivate(newspa
);
8451 txg
= spa_vdev_config_enter(spa
);
8453 /* re-online all offlined disks */
8454 for (c
= 0; c
< children
; c
++) {
8456 vml
[c
]->vdev_offline
= B_FALSE
;
8459 /* restart initializing or trimming disks as necessary */
8460 spa_async_request(spa
, SPA_ASYNC_INITIALIZE_RESTART
);
8461 spa_async_request(spa
, SPA_ASYNC_TRIM_RESTART
);
8462 spa_async_request(spa
, SPA_ASYNC_AUTOTRIM_RESTART
);
8464 vdev_reopen(spa
->spa_root_vdev
);
8466 nvlist_free(spa
->spa_config_splitting
);
8467 spa
->spa_config_splitting
= NULL
;
8468 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
8470 kmem_free(vml
, children
* sizeof (vdev_t
*));
8475 * Find any device that's done replacing, or a vdev marked 'unspare' that's
8476 * currently spared, so we can detach it.
8479 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
8481 vdev_t
*newvd
, *oldvd
;
8483 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
8484 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
8490 * Check for a completed replacement. We always consider the first
8491 * vdev in the list to be the oldest vdev, and the last one to be
8492 * the newest (see spa_vdev_attach() for how that works). In
8493 * the case where the newest vdev is faulted, we will not automatically
8494 * remove it after a resilver completes. This is OK as it will require
8495 * user intervention to determine which disk the admin wishes to keep.
8497 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
8498 ASSERT(vd
->vdev_children
> 1);
8500 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
8501 oldvd
= vd
->vdev_child
[0];
8503 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
8504 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
8505 !vdev_dtl_required(oldvd
))
8510 * Check for a completed resilver with the 'unspare' flag set.
8511 * Also potentially update faulted state.
8513 if (vd
->vdev_ops
== &vdev_spare_ops
) {
8514 vdev_t
*first
= vd
->vdev_child
[0];
8515 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
8517 if (last
->vdev_unspare
) {
8520 } else if (first
->vdev_unspare
) {
8527 if (oldvd
!= NULL
&&
8528 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
8529 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
8530 !vdev_dtl_required(oldvd
))
8533 vdev_propagate_state(vd
);
8536 * If there are more than two spares attached to a disk,
8537 * and those spares are not required, then we want to
8538 * attempt to free them up now so that they can be used
8539 * by other pools. Once we're back down to a single
8540 * disk+spare, we stop removing them.
8542 if (vd
->vdev_children
> 2) {
8543 newvd
= vd
->vdev_child
[1];
8545 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
8546 vdev_dtl_empty(last
, DTL_MISSING
) &&
8547 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
8548 !vdev_dtl_required(newvd
))
8557 spa_vdev_resilver_done(spa_t
*spa
)
8559 vdev_t
*vd
, *pvd
, *ppvd
;
8560 uint64_t guid
, sguid
, pguid
, ppguid
;
8562 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
8564 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
8565 pvd
= vd
->vdev_parent
;
8566 ppvd
= pvd
->vdev_parent
;
8567 guid
= vd
->vdev_guid
;
8568 pguid
= pvd
->vdev_guid
;
8569 ppguid
= ppvd
->vdev_guid
;
8572 * If we have just finished replacing a hot spared device, then
8573 * we need to detach the parent's first child (the original hot
8576 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
8577 ppvd
->vdev_children
== 2) {
8578 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
8579 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
8581 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
8583 spa_config_exit(spa
, SCL_ALL
, FTAG
);
8584 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
8586 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
8588 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
8591 spa_config_exit(spa
, SCL_ALL
, FTAG
);
8594 * If a detach was not performed above replace waiters will not have
8595 * been notified. In which case we must do so now.
8597 spa_notify_waiters(spa
);
8601 * Update the stored path or FRU for this vdev.
8604 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
8608 boolean_t sync
= B_FALSE
;
8610 ASSERT(spa_writeable(spa
));
8612 spa_vdev_state_enter(spa
, SCL_ALL
);
8614 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
8615 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
8617 if (!vd
->vdev_ops
->vdev_op_leaf
)
8618 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
8621 if (strcmp(value
, vd
->vdev_path
) != 0) {
8622 spa_strfree(vd
->vdev_path
);
8623 vd
->vdev_path
= spa_strdup(value
);
8627 if (vd
->vdev_fru
== NULL
) {
8628 vd
->vdev_fru
= spa_strdup(value
);
8630 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
8631 spa_strfree(vd
->vdev_fru
);
8632 vd
->vdev_fru
= spa_strdup(value
);
8637 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
8641 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
8643 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
8647 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
8649 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
8653 * ==========================================================================
8655 * ==========================================================================
8658 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
8660 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8662 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
8663 return (SET_ERROR(EBUSY
));
8665 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
8669 spa_scan_stop(spa_t
*spa
)
8671 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8672 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
8673 return (SET_ERROR(EBUSY
));
8675 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
8679 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
8681 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
8683 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
8684 return (SET_ERROR(ENOTSUP
));
8686 if (func
== POOL_SCAN_RESILVER
&&
8687 !spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
))
8688 return (SET_ERROR(ENOTSUP
));
8691 * If a resilver was requested, but there is no DTL on a
8692 * writeable leaf device, we have nothing to do.
8694 if (func
== POOL_SCAN_RESILVER
&&
8695 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
8696 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
8700 if (func
== POOL_SCAN_ERRORSCRUB
&&
8701 !spa_feature_is_enabled(spa
, SPA_FEATURE_HEAD_ERRLOG
))
8702 return (SET_ERROR(ENOTSUP
));
8704 return (dsl_scan(spa
->spa_dsl_pool
, func
));
8708 * ==========================================================================
8709 * SPA async task processing
8710 * ==========================================================================
8714 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
8716 if (vd
->vdev_remove_wanted
) {
8717 vd
->vdev_remove_wanted
= B_FALSE
;
8718 vd
->vdev_delayed_close
= B_FALSE
;
8719 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
8722 * We want to clear the stats, but we don't want to do a full
8723 * vdev_clear() as that will cause us to throw away
8724 * degraded/faulted state as well as attempt to reopen the
8725 * device, all of which is a waste.
8727 vd
->vdev_stat
.vs_read_errors
= 0;
8728 vd
->vdev_stat
.vs_write_errors
= 0;
8729 vd
->vdev_stat
.vs_checksum_errors
= 0;
8731 vdev_state_dirty(vd
->vdev_top
);
8733 /* Tell userspace that the vdev is gone. */
8734 zfs_post_remove(spa
, vd
);
8737 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8738 spa_async_remove(spa
, vd
->vdev_child
[c
]);
8742 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
8744 if (vd
->vdev_probe_wanted
) {
8745 vd
->vdev_probe_wanted
= B_FALSE
;
8746 vdev_reopen(vd
); /* vdev_open() does the actual probe */
8749 for (int c
= 0; c
< vd
->vdev_children
; c
++)
8750 spa_async_probe(spa
, vd
->vdev_child
[c
]);
8754 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
8756 if (!spa
->spa_autoexpand
)
8759 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
8760 vdev_t
*cvd
= vd
->vdev_child
[c
];
8761 spa_async_autoexpand(spa
, cvd
);
8764 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
8767 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
8770 static __attribute__((noreturn
)) void
8771 spa_async_thread(void *arg
)
8773 spa_t
*spa
= (spa_t
*)arg
;
8774 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8777 ASSERT(spa
->spa_sync_on
);
8779 mutex_enter(&spa
->spa_async_lock
);
8780 tasks
= spa
->spa_async_tasks
;
8781 spa
->spa_async_tasks
= 0;
8782 mutex_exit(&spa
->spa_async_lock
);
8785 * See if the config needs to be updated.
8787 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
8788 uint64_t old_space
, new_space
;
8790 mutex_enter(&spa_namespace_lock
);
8791 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
8792 old_space
+= metaslab_class_get_space(spa_special_class(spa
));
8793 old_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8794 old_space
+= metaslab_class_get_space(
8795 spa_embedded_log_class(spa
));
8797 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
8799 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
8800 new_space
+= metaslab_class_get_space(spa_special_class(spa
));
8801 new_space
+= metaslab_class_get_space(spa_dedup_class(spa
));
8802 new_space
+= metaslab_class_get_space(
8803 spa_embedded_log_class(spa
));
8804 mutex_exit(&spa_namespace_lock
);
8807 * If the pool grew as a result of the config update,
8808 * then log an internal history event.
8810 if (new_space
!= old_space
) {
8811 spa_history_log_internal(spa
, "vdev online", NULL
,
8812 "pool '%s' size: %llu(+%llu)",
8813 spa_name(spa
), (u_longlong_t
)new_space
,
8814 (u_longlong_t
)(new_space
- old_space
));
8819 * See if any devices need to be marked REMOVED.
8821 if (tasks
& SPA_ASYNC_REMOVE
) {
8822 spa_vdev_state_enter(spa
, SCL_NONE
);
8823 spa_async_remove(spa
, spa
->spa_root_vdev
);
8824 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
8825 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
8826 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
8827 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
8828 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8831 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
8832 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8833 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
8834 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8838 * See if any devices need to be probed.
8840 if (tasks
& SPA_ASYNC_PROBE
) {
8841 spa_vdev_state_enter(spa
, SCL_NONE
);
8842 spa_async_probe(spa
, spa
->spa_root_vdev
);
8843 (void) spa_vdev_state_exit(spa
, NULL
, 0);
8847 * If any devices are done replacing, detach them.
8849 if (tasks
& SPA_ASYNC_RESILVER_DONE
||
8850 tasks
& SPA_ASYNC_REBUILD_DONE
||
8851 tasks
& SPA_ASYNC_DETACH_SPARE
) {
8852 spa_vdev_resilver_done(spa
);
8856 * Kick off a resilver.
8858 if (tasks
& SPA_ASYNC_RESILVER
&&
8859 !vdev_rebuild_active(spa
->spa_root_vdev
) &&
8860 (!dsl_scan_resilvering(dp
) ||
8861 !spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_RESILVER_DEFER
)))
8862 dsl_scan_restart_resilver(dp
, 0);
8864 if (tasks
& SPA_ASYNC_INITIALIZE_RESTART
) {
8865 mutex_enter(&spa_namespace_lock
);
8866 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8867 vdev_initialize_restart(spa
->spa_root_vdev
);
8868 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8869 mutex_exit(&spa_namespace_lock
);
8872 if (tasks
& SPA_ASYNC_TRIM_RESTART
) {
8873 mutex_enter(&spa_namespace_lock
);
8874 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8875 vdev_trim_restart(spa
->spa_root_vdev
);
8876 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8877 mutex_exit(&spa_namespace_lock
);
8880 if (tasks
& SPA_ASYNC_AUTOTRIM_RESTART
) {
8881 mutex_enter(&spa_namespace_lock
);
8882 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8883 vdev_autotrim_restart(spa
);
8884 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8885 mutex_exit(&spa_namespace_lock
);
8889 * Kick off L2 cache whole device TRIM.
8891 if (tasks
& SPA_ASYNC_L2CACHE_TRIM
) {
8892 mutex_enter(&spa_namespace_lock
);
8893 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
8894 vdev_trim_l2arc(spa
);
8895 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
8896 mutex_exit(&spa_namespace_lock
);
8900 * Kick off L2 cache rebuilding.
8902 if (tasks
& SPA_ASYNC_L2CACHE_REBUILD
) {
8903 mutex_enter(&spa_namespace_lock
);
8904 spa_config_enter(spa
, SCL_L2ARC
, FTAG
, RW_READER
);
8905 l2arc_spa_rebuild_start(spa
);
8906 spa_config_exit(spa
, SCL_L2ARC
, FTAG
);
8907 mutex_exit(&spa_namespace_lock
);
8911 * Let the world know that we're done.
8913 mutex_enter(&spa
->spa_async_lock
);
8914 spa
->spa_async_thread
= NULL
;
8915 cv_broadcast(&spa
->spa_async_cv
);
8916 mutex_exit(&spa
->spa_async_lock
);
8921 spa_async_suspend(spa_t
*spa
)
8923 mutex_enter(&spa
->spa_async_lock
);
8924 spa
->spa_async_suspended
++;
8925 while (spa
->spa_async_thread
!= NULL
)
8926 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
8927 mutex_exit(&spa
->spa_async_lock
);
8929 spa_vdev_remove_suspend(spa
);
8931 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8932 if (condense_thread
!= NULL
)
8933 zthr_cancel(condense_thread
);
8935 zthr_t
*raidz_expand_thread
= spa
->spa_raidz_expand_zthr
;
8936 if (raidz_expand_thread
!= NULL
)
8937 zthr_cancel(raidz_expand_thread
);
8939 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8940 if (discard_thread
!= NULL
)
8941 zthr_cancel(discard_thread
);
8943 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8944 if (ll_delete_thread
!= NULL
)
8945 zthr_cancel(ll_delete_thread
);
8947 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8948 if (ll_condense_thread
!= NULL
)
8949 zthr_cancel(ll_condense_thread
);
8953 spa_async_resume(spa_t
*spa
)
8955 mutex_enter(&spa
->spa_async_lock
);
8956 ASSERT(spa
->spa_async_suspended
!= 0);
8957 spa
->spa_async_suspended
--;
8958 mutex_exit(&spa
->spa_async_lock
);
8959 spa_restart_removal(spa
);
8961 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
8962 if (condense_thread
!= NULL
)
8963 zthr_resume(condense_thread
);
8965 zthr_t
*raidz_expand_thread
= spa
->spa_raidz_expand_zthr
;
8966 if (raidz_expand_thread
!= NULL
)
8967 zthr_resume(raidz_expand_thread
);
8969 zthr_t
*discard_thread
= spa
->spa_checkpoint_discard_zthr
;
8970 if (discard_thread
!= NULL
)
8971 zthr_resume(discard_thread
);
8973 zthr_t
*ll_delete_thread
= spa
->spa_livelist_delete_zthr
;
8974 if (ll_delete_thread
!= NULL
)
8975 zthr_resume(ll_delete_thread
);
8977 zthr_t
*ll_condense_thread
= spa
->spa_livelist_condense_zthr
;
8978 if (ll_condense_thread
!= NULL
)
8979 zthr_resume(ll_condense_thread
);
8983 spa_async_tasks_pending(spa_t
*spa
)
8985 uint_t non_config_tasks
;
8987 boolean_t config_task_suspended
;
8989 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
8990 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
8991 if (spa
->spa_ccw_fail_time
== 0) {
8992 config_task_suspended
= B_FALSE
;
8994 config_task_suspended
=
8995 (gethrtime() - spa
->spa_ccw_fail_time
) <
8996 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
8999 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
9003 spa_async_dispatch(spa_t
*spa
)
9005 mutex_enter(&spa
->spa_async_lock
);
9006 if (spa_async_tasks_pending(spa
) &&
9007 !spa
->spa_async_suspended
&&
9008 spa
->spa_async_thread
== NULL
)
9009 spa
->spa_async_thread
= thread_create(NULL
, 0,
9010 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
9011 mutex_exit(&spa
->spa_async_lock
);
9015 spa_async_request(spa_t
*spa
, int task
)
9017 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
9018 mutex_enter(&spa
->spa_async_lock
);
9019 spa
->spa_async_tasks
|= task
;
9020 mutex_exit(&spa
->spa_async_lock
);
9024 spa_async_tasks(spa_t
*spa
)
9026 return (spa
->spa_async_tasks
);
9030 * ==========================================================================
9031 * SPA syncing routines
9032 * ==========================================================================
9037 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
9041 bpobj_enqueue(bpo
, bp
, bp_freed
, tx
);
9046 bpobj_enqueue_alloc_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
9048 return (bpobj_enqueue_cb(arg
, bp
, B_FALSE
, tx
));
9052 bpobj_enqueue_free_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
9054 return (bpobj_enqueue_cb(arg
, bp
, B_TRUE
, tx
));
9058 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
9062 zio_nowait(zio_free_sync(pio
, pio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
9068 bpobj_spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
9072 return (spa_free_sync_cb(arg
, bp
, tx
));
9076 * Note: this simple function is not inlined to make it easier to dtrace the
9077 * amount of time spent syncing frees.
9080 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
9082 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
9083 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
9084 VERIFY(zio_wait(zio
) == 0);
9088 * Note: this simple function is not inlined to make it easier to dtrace the
9089 * amount of time spent syncing deferred frees.
9092 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
9094 if (spa_sync_pass(spa
) != 1)
9099 * If the log space map feature is active, we stop deferring
9100 * frees to the next TXG and therefore running this function
9101 * would be considered a no-op as spa_deferred_bpobj should
9102 * not have any entries.
9104 * That said we run this function anyway (instead of returning
9105 * immediately) for the edge-case scenario where we just
9106 * activated the log space map feature in this TXG but we have
9107 * deferred frees from the previous TXG.
9109 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
9110 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
9111 bpobj_spa_free_sync_cb
, zio
, tx
), ==, 0);
9112 VERIFY0(zio_wait(zio
));
9116 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
9118 char *packed
= NULL
;
9123 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
9126 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
9127 * information. This avoids the dmu_buf_will_dirty() path and
9128 * saves us a pre-read to get data we don't actually care about.
9130 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
9131 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
9133 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
9135 memset(packed
+ nvsize
, 0, bufsize
- nvsize
);
9137 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
9139 vmem_free(packed
, bufsize
);
9141 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
9142 dmu_buf_will_dirty(db
, tx
);
9143 *(uint64_t *)db
->db_data
= nvsize
;
9144 dmu_buf_rele(db
, FTAG
);
9148 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
9149 const char *config
, const char *entry
)
9159 * Update the MOS nvlist describing the list of available devices.
9160 * spa_validate_aux() will have already made sure this nvlist is
9161 * valid and the vdevs are labeled appropriately.
9163 if (sav
->sav_object
== 0) {
9164 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
9165 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
9166 sizeof (uint64_t), tx
);
9167 VERIFY(zap_update(spa
->spa_meta_objset
,
9168 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
9169 &sav
->sav_object
, tx
) == 0);
9172 nvroot
= fnvlist_alloc();
9173 if (sav
->sav_count
== 0) {
9174 fnvlist_add_nvlist_array(nvroot
, config
,
9175 (const nvlist_t
* const *)NULL
, 0);
9177 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
9178 for (i
= 0; i
< sav
->sav_count
; i
++)
9179 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
9180 B_FALSE
, VDEV_CONFIG_L2CACHE
);
9181 fnvlist_add_nvlist_array(nvroot
, config
,
9182 (const nvlist_t
* const *)list
, sav
->sav_count
);
9183 for (i
= 0; i
< sav
->sav_count
; i
++)
9184 nvlist_free(list
[i
]);
9185 kmem_free(list
, sav
->sav_count
* sizeof (void *));
9188 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
9189 nvlist_free(nvroot
);
9191 sav
->sav_sync
= B_FALSE
;
9195 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
9196 * The all-vdev ZAP must be empty.
9199 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
9201 spa_t
*spa
= vd
->vdev_spa
;
9203 if (vd
->vdev_root_zap
!= 0 &&
9204 spa_feature_is_active(spa
, SPA_FEATURE_AVZ_V2
)) {
9205 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
9206 vd
->vdev_root_zap
, tx
));
9208 if (vd
->vdev_top_zap
!= 0) {
9209 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
9210 vd
->vdev_top_zap
, tx
));
9212 if (vd
->vdev_leaf_zap
!= 0) {
9213 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
9214 vd
->vdev_leaf_zap
, tx
));
9216 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
9217 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
9222 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
9227 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
9228 * its config may not be dirty but we still need to build per-vdev ZAPs.
9229 * Similarly, if the pool is being assembled (e.g. after a split), we
9230 * need to rebuild the AVZ although the config may not be dirty.
9232 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
9233 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
9236 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9238 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
9239 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
9240 spa
->spa_all_vdev_zaps
!= 0);
9242 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
9243 /* Make and build the new AVZ */
9244 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
9245 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
9246 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
9248 /* Diff old AVZ with new one */
9252 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
9253 spa
->spa_all_vdev_zaps
);
9254 zap_cursor_retrieve(&zc
, &za
) == 0;
9255 zap_cursor_advance(&zc
)) {
9256 uint64_t vdzap
= za
.za_first_integer
;
9257 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
9260 * ZAP is listed in old AVZ but not in new one;
9263 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
9268 zap_cursor_fini(&zc
);
9270 /* Destroy the old AVZ */
9271 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
9272 spa
->spa_all_vdev_zaps
, tx
));
9274 /* Replace the old AVZ in the dir obj with the new one */
9275 VERIFY0(zap_update(spa
->spa_meta_objset
,
9276 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
9277 sizeof (new_avz
), 1, &new_avz
, tx
));
9279 spa
->spa_all_vdev_zaps
= new_avz
;
9280 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
9284 /* Walk through the AVZ and destroy all listed ZAPs */
9285 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
9286 spa
->spa_all_vdev_zaps
);
9287 zap_cursor_retrieve(&zc
, &za
) == 0;
9288 zap_cursor_advance(&zc
)) {
9289 uint64_t zap
= za
.za_first_integer
;
9290 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
9293 zap_cursor_fini(&zc
);
9295 /* Destroy and unlink the AVZ itself */
9296 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
9297 spa
->spa_all_vdev_zaps
, tx
));
9298 VERIFY0(zap_remove(spa
->spa_meta_objset
,
9299 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
9300 spa
->spa_all_vdev_zaps
= 0;
9303 if (spa
->spa_all_vdev_zaps
== 0) {
9304 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
9305 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
9306 DMU_POOL_VDEV_ZAP_MAP
, tx
);
9308 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
9310 /* Create ZAPs for vdevs that don't have them. */
9311 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
9313 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
9314 dmu_tx_get_txg(tx
), B_FALSE
);
9317 * If we're upgrading the spa version then make sure that
9318 * the config object gets updated with the correct version.
9320 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
9321 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
9322 spa
->spa_uberblock
.ub_version
);
9324 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9326 nvlist_free(spa
->spa_config_syncing
);
9327 spa
->spa_config_syncing
= config
;
9329 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
9333 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
9335 uint64_t *versionp
= arg
;
9336 uint64_t version
= *versionp
;
9337 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
9340 * Setting the version is special cased when first creating the pool.
9342 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
9344 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
9345 ASSERT(version
>= spa_version(spa
));
9347 spa
->spa_uberblock
.ub_version
= version
;
9348 vdev_config_dirty(spa
->spa_root_vdev
);
9349 spa_history_log_internal(spa
, "set", tx
, "version=%lld",
9350 (longlong_t
)version
);
9354 * Set zpool properties.
9357 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
9359 nvlist_t
*nvp
= arg
;
9360 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
9361 objset_t
*mos
= spa
->spa_meta_objset
;
9362 nvpair_t
*elem
= NULL
;
9364 mutex_enter(&spa
->spa_props_lock
);
9366 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
9368 const char *strval
, *fname
;
9370 const char *propname
;
9371 const char *elemname
= nvpair_name(elem
);
9372 zprop_type_t proptype
;
9375 switch (prop
= zpool_name_to_prop(elemname
)) {
9376 case ZPOOL_PROP_VERSION
:
9377 intval
= fnvpair_value_uint64(elem
);
9379 * The version is synced separately before other
9380 * properties and should be correct by now.
9382 ASSERT3U(spa_version(spa
), >=, intval
);
9385 case ZPOOL_PROP_ALTROOT
:
9387 * 'altroot' is a non-persistent property. It should
9388 * have been set temporarily at creation or import time.
9390 ASSERT(spa
->spa_root
!= NULL
);
9393 case ZPOOL_PROP_READONLY
:
9394 case ZPOOL_PROP_CACHEFILE
:
9396 * 'readonly' and 'cachefile' are also non-persistent
9400 case ZPOOL_PROP_COMMENT
:
9401 strval
= fnvpair_value_string(elem
);
9402 if (spa
->spa_comment
!= NULL
)
9403 spa_strfree(spa
->spa_comment
);
9404 spa
->spa_comment
= spa_strdup(strval
);
9406 * We need to dirty the configuration on all the vdevs
9407 * so that their labels get updated. We also need to
9408 * update the cache file to keep it in sync with the
9409 * MOS version. It's unnecessary to do this for pool
9410 * creation since the vdev's configuration has already
9413 if (tx
->tx_txg
!= TXG_INITIAL
) {
9414 vdev_config_dirty(spa
->spa_root_vdev
);
9415 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
9417 spa_history_log_internal(spa
, "set", tx
,
9418 "%s=%s", elemname
, strval
);
9420 case ZPOOL_PROP_COMPATIBILITY
:
9421 strval
= fnvpair_value_string(elem
);
9422 if (spa
->spa_compatibility
!= NULL
)
9423 spa_strfree(spa
->spa_compatibility
);
9424 spa
->spa_compatibility
= spa_strdup(strval
);
9426 * Dirty the configuration on vdevs as above.
9428 if (tx
->tx_txg
!= TXG_INITIAL
) {
9429 vdev_config_dirty(spa
->spa_root_vdev
);
9430 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
9433 spa_history_log_internal(spa
, "set", tx
,
9434 "%s=%s", nvpair_name(elem
), strval
);
9437 case ZPOOL_PROP_INVAL
:
9438 if (zpool_prop_feature(elemname
)) {
9439 fname
= strchr(elemname
, '@') + 1;
9440 VERIFY0(zfeature_lookup_name(fname
, &fid
));
9442 spa_feature_enable(spa
, fid
, tx
);
9443 spa_history_log_internal(spa
, "set", tx
,
9444 "%s=enabled", elemname
);
9446 } else if (!zfs_prop_user(elemname
)) {
9447 ASSERT(zpool_prop_feature(elemname
));
9453 * Set pool property values in the poolprops mos object.
9455 if (spa
->spa_pool_props_object
== 0) {
9456 spa
->spa_pool_props_object
=
9457 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
9458 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
9462 /* normalize the property name */
9463 if (prop
== ZPOOL_PROP_INVAL
) {
9464 propname
= elemname
;
9465 proptype
= PROP_TYPE_STRING
;
9467 propname
= zpool_prop_to_name(prop
);
9468 proptype
= zpool_prop_get_type(prop
);
9471 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
9472 ASSERT(proptype
== PROP_TYPE_STRING
);
9473 strval
= fnvpair_value_string(elem
);
9474 VERIFY0(zap_update(mos
,
9475 spa
->spa_pool_props_object
, propname
,
9476 1, strlen(strval
) + 1, strval
, tx
));
9477 spa_history_log_internal(spa
, "set", tx
,
9478 "%s=%s", elemname
, strval
);
9479 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
9480 intval
= fnvpair_value_uint64(elem
);
9482 if (proptype
== PROP_TYPE_INDEX
) {
9484 VERIFY0(zpool_prop_index_to_string(
9485 prop
, intval
, &unused
));
9487 VERIFY0(zap_update(mos
,
9488 spa
->spa_pool_props_object
, propname
,
9489 8, 1, &intval
, tx
));
9490 spa_history_log_internal(spa
, "set", tx
,
9491 "%s=%lld", elemname
,
9492 (longlong_t
)intval
);
9495 case ZPOOL_PROP_DELEGATION
:
9496 spa
->spa_delegation
= intval
;
9498 case ZPOOL_PROP_BOOTFS
:
9499 spa
->spa_bootfs
= intval
;
9501 case ZPOOL_PROP_FAILUREMODE
:
9502 spa
->spa_failmode
= intval
;
9504 case ZPOOL_PROP_AUTOTRIM
:
9505 spa
->spa_autotrim
= intval
;
9506 spa_async_request(spa
,
9507 SPA_ASYNC_AUTOTRIM_RESTART
);
9509 case ZPOOL_PROP_AUTOEXPAND
:
9510 spa
->spa_autoexpand
= intval
;
9511 if (tx
->tx_txg
!= TXG_INITIAL
)
9512 spa_async_request(spa
,
9513 SPA_ASYNC_AUTOEXPAND
);
9515 case ZPOOL_PROP_MULTIHOST
:
9516 spa
->spa_multihost
= intval
;
9522 ASSERT(0); /* not allowed */
9528 mutex_exit(&spa
->spa_props_lock
);
9532 * Perform one-time upgrade on-disk changes. spa_version() does not
9533 * reflect the new version this txg, so there must be no changes this
9534 * txg to anything that the upgrade code depends on after it executes.
9535 * Therefore this must be called after dsl_pool_sync() does the sync
9539 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
9541 if (spa_sync_pass(spa
) != 1)
9544 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9545 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
9547 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
9548 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
9549 dsl_pool_create_origin(dp
, tx
);
9551 /* Keeping the origin open increases spa_minref */
9552 spa
->spa_minref
+= 3;
9555 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
9556 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
9557 dsl_pool_upgrade_clones(dp
, tx
);
9560 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
9561 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
9562 dsl_pool_upgrade_dir_clones(dp
, tx
);
9564 /* Keeping the freedir open increases spa_minref */
9565 spa
->spa_minref
+= 3;
9568 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
9569 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
9570 spa_feature_create_zap_objects(spa
, tx
);
9574 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
9575 * when possibility to use lz4 compression for metadata was added
9576 * Old pools that have this feature enabled must be upgraded to have
9577 * this feature active
9579 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
9580 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
9581 SPA_FEATURE_LZ4_COMPRESS
);
9582 boolean_t lz4_ac
= spa_feature_is_active(spa
,
9583 SPA_FEATURE_LZ4_COMPRESS
);
9585 if (lz4_en
&& !lz4_ac
)
9586 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
9590 * If we haven't written the salt, do so now. Note that the
9591 * feature may not be activated yet, but that's fine since
9592 * the presence of this ZAP entry is backwards compatible.
9594 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
9595 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
9596 VERIFY0(zap_add(spa
->spa_meta_objset
,
9597 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
9598 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
9599 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
9602 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
9606 vdev_indirect_state_sync_verify(vdev_t
*vd
)
9608 vdev_indirect_mapping_t
*vim __maybe_unused
= vd
->vdev_indirect_mapping
;
9609 vdev_indirect_births_t
*vib __maybe_unused
= vd
->vdev_indirect_births
;
9611 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
9612 ASSERT(vim
!= NULL
);
9613 ASSERT(vib
!= NULL
);
9616 uint64_t obsolete_sm_object
= 0;
9617 ASSERT0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
9618 if (obsolete_sm_object
!= 0) {
9619 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
9620 ASSERT(vd
->vdev_removing
||
9621 vd
->vdev_ops
== &vdev_indirect_ops
);
9622 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
9623 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
9624 ASSERT3U(obsolete_sm_object
, ==,
9625 space_map_object(vd
->vdev_obsolete_sm
));
9626 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
9627 space_map_allocated(vd
->vdev_obsolete_sm
));
9629 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
9632 * Since frees / remaps to an indirect vdev can only
9633 * happen in syncing context, the obsolete segments
9634 * tree must be empty when we start syncing.
9636 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
9640 * Set the top-level vdev's max queue depth. Evaluate each top-level's
9641 * async write queue depth in case it changed. The max queue depth will
9642 * not change in the middle of syncing out this txg.
9645 spa_sync_adjust_vdev_max_queue_depth(spa_t
*spa
)
9647 ASSERT(spa_writeable(spa
));
9649 vdev_t
*rvd
= spa
->spa_root_vdev
;
9650 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
9651 zfs_vdev_queue_depth_pct
/ 100;
9652 metaslab_class_t
*normal
= spa_normal_class(spa
);
9653 metaslab_class_t
*special
= spa_special_class(spa
);
9654 metaslab_class_t
*dedup
= spa_dedup_class(spa
);
9656 uint64_t slots_per_allocator
= 0;
9657 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
9658 vdev_t
*tvd
= rvd
->vdev_child
[c
];
9660 metaslab_group_t
*mg
= tvd
->vdev_mg
;
9661 if (mg
== NULL
|| !metaslab_group_initialized(mg
))
9664 metaslab_class_t
*mc
= mg
->mg_class
;
9665 if (mc
!= normal
&& mc
!= special
&& mc
!= dedup
)
9669 * It is safe to do a lock-free check here because only async
9670 * allocations look at mg_max_alloc_queue_depth, and async
9671 * allocations all happen from spa_sync().
9673 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
9674 ASSERT0(zfs_refcount_count(
9675 &(mg
->mg_allocator
[i
].mga_alloc_queue_depth
)));
9677 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
9679 for (int i
= 0; i
< mg
->mg_allocators
; i
++) {
9680 mg
->mg_allocator
[i
].mga_cur_max_alloc_queue_depth
=
9681 zfs_vdev_def_queue_depth
;
9683 slots_per_allocator
+= zfs_vdev_def_queue_depth
;
9686 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9687 ASSERT0(zfs_refcount_count(&normal
->mc_allocator
[i
].
9689 ASSERT0(zfs_refcount_count(&special
->mc_allocator
[i
].
9691 ASSERT0(zfs_refcount_count(&dedup
->mc_allocator
[i
].
9693 normal
->mc_allocator
[i
].mca_alloc_max_slots
=
9694 slots_per_allocator
;
9695 special
->mc_allocator
[i
].mca_alloc_max_slots
=
9696 slots_per_allocator
;
9697 dedup
->mc_allocator
[i
].mca_alloc_max_slots
=
9698 slots_per_allocator
;
9700 normal
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9701 special
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9702 dedup
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
9706 spa_sync_condense_indirect(spa_t
*spa
, dmu_tx_t
*tx
)
9708 ASSERT(spa_writeable(spa
));
9710 vdev_t
*rvd
= spa
->spa_root_vdev
;
9711 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
9712 vdev_t
*vd
= rvd
->vdev_child
[c
];
9713 vdev_indirect_state_sync_verify(vd
);
9715 if (vdev_indirect_should_condense(vd
)) {
9716 spa_condense_indirect_start_sync(vd
, tx
);
9723 spa_sync_iterate_to_convergence(spa_t
*spa
, dmu_tx_t
*tx
)
9725 objset_t
*mos
= spa
->spa_meta_objset
;
9726 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9727 uint64_t txg
= tx
->tx_txg
;
9728 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
9731 int pass
= ++spa
->spa_sync_pass
;
9733 spa_sync_config_object(spa
, tx
);
9734 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
9735 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
9736 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
9737 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
9738 spa_errlog_sync(spa
, txg
);
9739 dsl_pool_sync(dp
, txg
);
9741 if (pass
< zfs_sync_pass_deferred_free
||
9742 spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
)) {
9744 * If the log space map feature is active we don't
9745 * care about deferred frees and the deferred bpobj
9746 * as the log space map should effectively have the
9747 * same results (i.e. appending only to one object).
9749 spa_sync_frees(spa
, free_bpl
, tx
);
9752 * We can not defer frees in pass 1, because
9753 * we sync the deferred frees later in pass 1.
9755 ASSERT3U(pass
, >, 1);
9756 bplist_iterate(free_bpl
, bpobj_enqueue_alloc_cb
,
9757 &spa
->spa_deferred_bpobj
, tx
);
9762 dsl_scan_sync(dp
, tx
);
9763 dsl_errorscrub_sync(dp
, tx
);
9765 spa_sync_upgrades(spa
, tx
);
9767 spa_flush_metaslabs(spa
, tx
);
9770 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
9776 * dsl_pool_sync() -> dp_sync_tasks may have dirtied
9777 * the config. If that happens, this txg should not
9778 * be a no-op. So we must sync the config to the MOS
9779 * before checking for no-op.
9781 * Note that when the config is dirty, it will
9782 * be written to the MOS (i.e. the MOS will be
9783 * dirtied) every time we call spa_sync_config_object()
9784 * in this txg. Therefore we can't call this after
9785 * dsl_pool_sync() every pass, because it would
9786 * prevent us from converging, since we'd dirty
9787 * the MOS every pass.
9789 * Sync tasks can only be processed in pass 1, so
9790 * there's no need to do this in later passes.
9792 spa_sync_config_object(spa
, tx
);
9796 * Note: We need to check if the MOS is dirty because we could
9797 * have marked the MOS dirty without updating the uberblock
9798 * (e.g. if we have sync tasks but no dirty user data). We need
9799 * to check the uberblock's rootbp because it is updated if we
9800 * have synced out dirty data (though in this case the MOS will
9801 * most likely also be dirty due to second order effects, we
9802 * don't want to rely on that here).
9805 BP_GET_LOGICAL_BIRTH(&spa
->spa_uberblock
.ub_rootbp
) < txg
&&
9806 !dmu_objset_is_dirty(mos
, txg
)) {
9808 * Nothing changed on the first pass, therefore this
9809 * TXG is a no-op. Avoid syncing deferred frees, so
9810 * that we can keep this TXG as a no-op.
9812 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
9813 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
9814 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
9815 ASSERT(txg_list_empty(&dp
->dp_early_sync_tasks
, txg
));
9819 spa_sync_deferred_frees(spa
, tx
);
9820 } while (dmu_objset_is_dirty(mos
, txg
));
9824 * Rewrite the vdev configuration (which includes the uberblock) to
9825 * commit the transaction group.
9827 * If there are no dirty vdevs, we sync the uberblock to a few random
9828 * top-level vdevs that are known to be visible in the config cache
9829 * (see spa_vdev_add() for a complete description). If there *are* dirty
9830 * vdevs, sync the uberblock to all vdevs.
9833 spa_sync_rewrite_vdev_config(spa_t
*spa
, dmu_tx_t
*tx
)
9835 vdev_t
*rvd
= spa
->spa_root_vdev
;
9836 uint64_t txg
= tx
->tx_txg
;
9842 * We hold SCL_STATE to prevent vdev open/close/etc.
9843 * while we're attempting to write the vdev labels.
9845 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9847 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
9848 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
] = { NULL
};
9850 int children
= rvd
->vdev_children
;
9851 int c0
= random_in_range(children
);
9853 for (int c
= 0; c
< children
; c
++) {
9855 rvd
->vdev_child
[(c0
+ c
) % children
];
9857 /* Stop when revisiting the first vdev */
9858 if (c
> 0 && svd
[0] == vd
)
9861 if (vd
->vdev_ms_array
== 0 ||
9863 !vdev_is_concrete(vd
))
9866 svd
[svdcount
++] = vd
;
9867 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
9870 error
= vdev_config_sync(svd
, svdcount
, txg
);
9872 error
= vdev_config_sync(rvd
->vdev_child
,
9873 rvd
->vdev_children
, txg
);
9877 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
9879 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9883 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
9884 zio_resume_wait(spa
);
9889 * Sync the specified transaction group. New blocks may be dirtied as
9890 * part of the process, so we iterate until it converges.
9893 spa_sync(spa_t
*spa
, uint64_t txg
)
9897 VERIFY(spa_writeable(spa
));
9900 * Wait for i/os issued in open context that need to complete
9901 * before this txg syncs.
9903 (void) zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]);
9904 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
,
9908 * Now that there can be no more cloning in this transaction group,
9909 * but we are still before issuing frees, we can process pending BRT
9912 brt_pending_apply(spa
, txg
);
9915 * Lock out configuration changes.
9917 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
9919 spa
->spa_syncing_txg
= txg
;
9920 spa
->spa_sync_pass
= 0;
9922 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
9923 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
9924 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
9925 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
9929 * If there are any pending vdev state changes, convert them
9930 * into config changes that go out with this transaction group.
9932 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
9933 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
9934 /* Avoid holding the write lock unless actually necessary */
9935 if (vd
->vdev_aux
== NULL
) {
9936 vdev_state_clean(vd
);
9937 vdev_config_dirty(vd
);
9941 * We need the write lock here because, for aux vdevs,
9942 * calling vdev_config_dirty() modifies sav_config.
9943 * This is ugly and will become unnecessary when we
9944 * eliminate the aux vdev wart by integrating all vdevs
9945 * into the root vdev tree.
9947 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9948 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
9949 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
9950 vdev_state_clean(vd
);
9951 vdev_config_dirty(vd
);
9953 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
9954 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
9956 spa_config_exit(spa
, SCL_STATE
, FTAG
);
9958 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
9959 dmu_tx_t
*tx
= dmu_tx_create_assigned(dp
, txg
);
9961 spa
->spa_sync_starttime
= gethrtime();
9962 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
9963 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
9964 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
9965 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
9968 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
9969 * set spa_deflate if we have no raid-z vdevs.
9971 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
9972 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
9973 vdev_t
*rvd
= spa
->spa_root_vdev
;
9976 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
9977 vd
= rvd
->vdev_child
[i
];
9978 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
9981 if (i
== rvd
->vdev_children
) {
9982 spa
->spa_deflate
= TRUE
;
9983 VERIFY0(zap_add(spa
->spa_meta_objset
,
9984 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
9985 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
9989 spa_sync_adjust_vdev_max_queue_depth(spa
);
9991 spa_sync_condense_indirect(spa
, tx
);
9993 spa_sync_iterate_to_convergence(spa
, tx
);
9996 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
9998 * Make sure that the number of ZAPs for all the vdevs matches
9999 * the number of ZAPs in the per-vdev ZAP list. This only gets
10000 * called if the config is dirty; otherwise there may be
10001 * outstanding AVZ operations that weren't completed in
10002 * spa_sync_config_object.
10004 uint64_t all_vdev_zap_entry_count
;
10005 ASSERT0(zap_count(spa
->spa_meta_objset
,
10006 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
10007 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
10008 all_vdev_zap_entry_count
);
10012 if (spa
->spa_vdev_removal
!= NULL
) {
10013 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
10016 spa_sync_rewrite_vdev_config(spa
, tx
);
10019 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
10020 spa
->spa_deadman_tqid
= 0;
10023 * Clear the dirty config list.
10025 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
10026 vdev_config_clean(vd
);
10029 * Now that the new config has synced transactionally,
10030 * let it become visible to the config cache.
10032 if (spa
->spa_config_syncing
!= NULL
) {
10033 spa_config_set(spa
, spa
->spa_config_syncing
);
10034 spa
->spa_config_txg
= txg
;
10035 spa
->spa_config_syncing
= NULL
;
10038 dsl_pool_sync_done(dp
, txg
);
10040 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
10041 mutex_enter(&spa
->spa_allocs
[i
].spaa_lock
);
10042 VERIFY0(avl_numnodes(&spa
->spa_allocs
[i
].spaa_tree
));
10043 mutex_exit(&spa
->spa_allocs
[i
].spaa_lock
);
10047 * Update usable space statistics.
10049 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
10051 vdev_sync_done(vd
, txg
);
10053 metaslab_class_evict_old(spa
->spa_normal_class
, txg
);
10054 metaslab_class_evict_old(spa
->spa_log_class
, txg
);
10056 spa_sync_close_syncing_log_sm(spa
);
10058 spa_update_dspace(spa
);
10060 if (spa_get_autotrim(spa
) == SPA_AUTOTRIM_ON
)
10061 vdev_autotrim_kick(spa
);
10064 * It had better be the case that we didn't dirty anything
10065 * since vdev_config_sync().
10067 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
10068 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
10069 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
10071 while (zfs_pause_spa_sync
)
10074 spa
->spa_sync_pass
= 0;
10077 * Update the last synced uberblock here. We want to do this at
10078 * the end of spa_sync() so that consumers of spa_last_synced_txg()
10079 * will be guaranteed that all the processing associated with
10080 * that txg has been completed.
10082 spa
->spa_ubsync
= spa
->spa_uberblock
;
10083 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
10085 spa_handle_ignored_writes(spa
);
10088 * If any async tasks have been requested, kick them off.
10090 spa_async_dispatch(spa
);
10094 * Sync all pools. We don't want to hold the namespace lock across these
10095 * operations, so we take a reference on the spa_t and drop the lock during the
10099 spa_sync_allpools(void)
10102 mutex_enter(&spa_namespace_lock
);
10103 while ((spa
= spa_next(spa
)) != NULL
) {
10104 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
10105 !spa_writeable(spa
) || spa_suspended(spa
))
10107 spa_open_ref(spa
, FTAG
);
10108 mutex_exit(&spa_namespace_lock
);
10109 txg_wait_synced(spa_get_dsl(spa
), 0);
10110 mutex_enter(&spa_namespace_lock
);
10111 spa_close(spa
, FTAG
);
10113 mutex_exit(&spa_namespace_lock
);
10117 spa_sync_tq_create(spa_t
*spa
, const char *name
)
10119 kthread_t
**kthreads
;
10121 ASSERT(spa
->spa_sync_tq
== NULL
);
10122 ASSERT3S(spa
->spa_alloc_count
, <=, boot_ncpus
);
10125 * - do not allow more allocators than cpus.
10126 * - there may be more cpus than allocators.
10127 * - do not allow more sync taskq threads than allocators or cpus.
10129 int nthreads
= spa
->spa_alloc_count
;
10130 spa
->spa_syncthreads
= kmem_zalloc(sizeof (spa_syncthread_info_t
) *
10131 nthreads
, KM_SLEEP
);
10133 spa
->spa_sync_tq
= taskq_create_synced(name
, nthreads
, minclsyspri
,
10134 nthreads
, INT_MAX
, TASKQ_PREPOPULATE
, &kthreads
);
10135 VERIFY(spa
->spa_sync_tq
!= NULL
);
10136 VERIFY(kthreads
!= NULL
);
10138 spa_taskqs_t
*tqs
=
10139 &spa
->spa_zio_taskq
[ZIO_TYPE_WRITE
][ZIO_TASKQ_ISSUE
];
10141 spa_syncthread_info_t
*ti
= spa
->spa_syncthreads
;
10142 for (int i
= 0, w
= 0; i
< nthreads
; i
++, w
++, ti
++) {
10143 ti
->sti_thread
= kthreads
[i
];
10144 if (w
== tqs
->stqs_count
) {
10147 ti
->sti_wr_iss_tq
= tqs
->stqs_taskq
[w
];
10150 kmem_free(kthreads
, sizeof (*kthreads
) * nthreads
);
10151 return (spa
->spa_sync_tq
);
10155 spa_sync_tq_destroy(spa_t
*spa
)
10157 ASSERT(spa
->spa_sync_tq
!= NULL
);
10159 taskq_wait(spa
->spa_sync_tq
);
10160 taskq_destroy(spa
->spa_sync_tq
);
10161 kmem_free(spa
->spa_syncthreads
,
10162 sizeof (spa_syncthread_info_t
) * spa
->spa_alloc_count
);
10163 spa
->spa_sync_tq
= NULL
;
10167 spa_select_allocator(zio_t
*zio
)
10169 zbookmark_phys_t
*bm
= &zio
->io_bookmark
;
10170 spa_t
*spa
= zio
->io_spa
;
10172 ASSERT(zio
->io_type
== ZIO_TYPE_WRITE
);
10175 * A gang block (for example) may have inherited its parent's
10176 * allocator, in which case there is nothing further to do here.
10178 if (ZIO_HAS_ALLOCATOR(zio
))
10181 ASSERT(spa
!= NULL
);
10182 ASSERT(bm
!= NULL
);
10185 * First try to use an allocator assigned to the syncthread, and set
10186 * the corresponding write issue taskq for the allocator.
10187 * Note, we must have an open pool to do this.
10189 if (spa
->spa_sync_tq
!= NULL
) {
10190 spa_syncthread_info_t
*ti
= spa
->spa_syncthreads
;
10191 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++, ti
++) {
10192 if (ti
->sti_thread
== curthread
) {
10193 zio
->io_allocator
= i
;
10194 zio
->io_wr_iss_tq
= ti
->sti_wr_iss_tq
;
10201 * We want to try to use as many allocators as possible to help improve
10202 * performance, but we also want logically adjacent IOs to be physically
10203 * adjacent to improve sequential read performance. We chunk each object
10204 * into 2^20 block regions, and then hash based on the objset, object,
10205 * level, and region to accomplish both of these goals.
10207 uint64_t hv
= cityhash4(bm
->zb_objset
, bm
->zb_object
, bm
->zb_level
,
10208 bm
->zb_blkid
>> 20);
10210 zio
->io_allocator
= (uint_t
)hv
% spa
->spa_alloc_count
;
10211 zio
->io_wr_iss_tq
= NULL
;
10215 * ==========================================================================
10216 * Miscellaneous routines
10217 * ==========================================================================
10221 * Remove all pools in the system.
10224 spa_evict_all(void)
10229 * Remove all cached state. All pools should be closed now,
10230 * so every spa in the AVL tree should be unreferenced.
10232 mutex_enter(&spa_namespace_lock
);
10233 while ((spa
= spa_next(NULL
)) != NULL
) {
10235 * Stop async tasks. The async thread may need to detach
10236 * a device that's been replaced, which requires grabbing
10237 * spa_namespace_lock, so we must drop it here.
10239 spa_open_ref(spa
, FTAG
);
10240 mutex_exit(&spa_namespace_lock
);
10241 spa_async_suspend(spa
);
10242 mutex_enter(&spa_namespace_lock
);
10243 spa_close(spa
, FTAG
);
10245 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
10247 spa_deactivate(spa
);
10251 mutex_exit(&spa_namespace_lock
);
10255 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
10260 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
10264 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
10265 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
10266 if (vd
->vdev_guid
== guid
)
10270 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
10271 vd
= spa
->spa_spares
.sav_vdevs
[i
];
10272 if (vd
->vdev_guid
== guid
)
10281 spa_upgrade(spa_t
*spa
, uint64_t version
)
10283 ASSERT(spa_writeable(spa
));
10285 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
10288 * This should only be called for a non-faulted pool, and since a
10289 * future version would result in an unopenable pool, this shouldn't be
10292 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
10293 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
10295 spa
->spa_uberblock
.ub_version
= version
;
10296 vdev_config_dirty(spa
->spa_root_vdev
);
10298 spa_config_exit(spa
, SCL_ALL
, FTAG
);
10300 txg_wait_synced(spa_get_dsl(spa
), 0);
10304 spa_has_aux_vdev(spa_t
*spa
, uint64_t guid
, spa_aux_vdev_t
*sav
)
10308 uint64_t vdev_guid
;
10310 for (i
= 0; i
< sav
->sav_count
; i
++)
10311 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
10314 for (i
= 0; i
< sav
->sav_npending
; i
++) {
10315 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
10316 &vdev_guid
) == 0 && vdev_guid
== guid
)
10324 spa_has_l2cache(spa_t
*spa
, uint64_t guid
)
10326 return (spa_has_aux_vdev(spa
, guid
, &spa
->spa_l2cache
));
10330 spa_has_spare(spa_t
*spa
, uint64_t guid
)
10332 return (spa_has_aux_vdev(spa
, guid
, &spa
->spa_spares
));
10336 * Check if a pool has an active shared spare device.
10337 * Note: reference count of an active spare is 2, as a spare and as a replace
10340 spa_has_active_shared_spare(spa_t
*spa
)
10344 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
10346 for (i
= 0; i
< sav
->sav_count
; i
++) {
10347 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
10348 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
10357 spa_total_metaslabs(spa_t
*spa
)
10359 vdev_t
*rvd
= spa
->spa_root_vdev
;
10362 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
10363 vdev_t
*vd
= rvd
->vdev_child
[c
];
10364 if (!vdev_is_concrete(vd
))
10366 m
+= vd
->vdev_ms_count
;
10372 * Notify any waiting threads that some activity has switched from being in-
10373 * progress to not-in-progress so that the thread can wake up and determine
10374 * whether it is finished waiting.
10377 spa_notify_waiters(spa_t
*spa
)
10380 * Acquiring spa_activities_lock here prevents the cv_broadcast from
10381 * happening between the waiting thread's check and cv_wait.
10383 mutex_enter(&spa
->spa_activities_lock
);
10384 cv_broadcast(&spa
->spa_activities_cv
);
10385 mutex_exit(&spa
->spa_activities_lock
);
10389 * Notify any waiting threads that the pool is exporting, and then block until
10390 * they are finished using the spa_t.
10393 spa_wake_waiters(spa_t
*spa
)
10395 mutex_enter(&spa
->spa_activities_lock
);
10396 spa
->spa_waiters_cancel
= B_TRUE
;
10397 cv_broadcast(&spa
->spa_activities_cv
);
10398 while (spa
->spa_waiters
!= 0)
10399 cv_wait(&spa
->spa_waiters_cv
, &spa
->spa_activities_lock
);
10400 spa
->spa_waiters_cancel
= B_FALSE
;
10401 mutex_exit(&spa
->spa_activities_lock
);
10404 /* Whether the vdev or any of its descendants are being initialized/trimmed. */
10406 spa_vdev_activity_in_progress_impl(vdev_t
*vd
, zpool_wait_activity_t activity
)
10408 spa_t
*spa
= vd
->vdev_spa
;
10410 ASSERT(spa_config_held(spa
, SCL_CONFIG
| SCL_STATE
, RW_READER
));
10411 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
10412 ASSERT(activity
== ZPOOL_WAIT_INITIALIZE
||
10413 activity
== ZPOOL_WAIT_TRIM
);
10415 kmutex_t
*lock
= activity
== ZPOOL_WAIT_INITIALIZE
?
10416 &vd
->vdev_initialize_lock
: &vd
->vdev_trim_lock
;
10418 mutex_exit(&spa
->spa_activities_lock
);
10420 mutex_enter(&spa
->spa_activities_lock
);
10422 boolean_t in_progress
= (activity
== ZPOOL_WAIT_INITIALIZE
) ?
10423 (vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) :
10424 (vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
);
10430 for (int i
= 0; i
< vd
->vdev_children
; i
++) {
10431 if (spa_vdev_activity_in_progress_impl(vd
->vdev_child
[i
],
10440 * If use_guid is true, this checks whether the vdev specified by guid is
10441 * being initialized/trimmed. Otherwise, it checks whether any vdev in the pool
10442 * is being initialized/trimmed. The caller must hold the config lock and
10443 * spa_activities_lock.
10446 spa_vdev_activity_in_progress(spa_t
*spa
, boolean_t use_guid
, uint64_t guid
,
10447 zpool_wait_activity_t activity
, boolean_t
*in_progress
)
10449 mutex_exit(&spa
->spa_activities_lock
);
10450 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
10451 mutex_enter(&spa
->spa_activities_lock
);
10455 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
10456 if (vd
== NULL
|| !vd
->vdev_ops
->vdev_op_leaf
) {
10457 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
10461 vd
= spa
->spa_root_vdev
;
10464 *in_progress
= spa_vdev_activity_in_progress_impl(vd
, activity
);
10466 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
10471 * Locking for waiting threads
10472 * ---------------------------
10474 * Waiting threads need a way to check whether a given activity is in progress,
10475 * and then, if it is, wait for it to complete. Each activity will have some
10476 * in-memory representation of the relevant on-disk state which can be used to
10477 * determine whether or not the activity is in progress. The in-memory state and
10478 * the locking used to protect it will be different for each activity, and may
10479 * not be suitable for use with a cvar (e.g., some state is protected by the
10480 * config lock). To allow waiting threads to wait without any races, another
10481 * lock, spa_activities_lock, is used.
10483 * When the state is checked, both the activity-specific lock (if there is one)
10484 * and spa_activities_lock are held. In some cases, the activity-specific lock
10485 * is acquired explicitly (e.g. the config lock). In others, the locking is
10486 * internal to some check (e.g. bpobj_is_empty). After checking, the waiting
10487 * thread releases the activity-specific lock and, if the activity is in
10488 * progress, then cv_waits using spa_activities_lock.
10490 * The waiting thread is woken when another thread, one completing some
10491 * activity, updates the state of the activity and then calls
10492 * spa_notify_waiters, which will cv_broadcast. This 'completing' thread only
10493 * needs to hold its activity-specific lock when updating the state, and this
10494 * lock can (but doesn't have to) be dropped before calling spa_notify_waiters.
10496 * Because spa_notify_waiters acquires spa_activities_lock before broadcasting,
10497 * and because it is held when the waiting thread checks the state of the
10498 * activity, it can never be the case that the completing thread both updates
10499 * the activity state and cv_broadcasts in between the waiting thread's check
10500 * and cv_wait. Thus, a waiting thread can never miss a wakeup.
10502 * In order to prevent deadlock, when the waiting thread does its check, in some
10503 * cases it will temporarily drop spa_activities_lock in order to acquire the
10504 * activity-specific lock. The order in which spa_activities_lock and the
10505 * activity specific lock are acquired in the waiting thread is determined by
10506 * the order in which they are acquired in the completing thread; if the
10507 * completing thread calls spa_notify_waiters with the activity-specific lock
10508 * held, then the waiting thread must also acquire the activity-specific lock
10513 spa_activity_in_progress(spa_t
*spa
, zpool_wait_activity_t activity
,
10514 boolean_t use_tag
, uint64_t tag
, boolean_t
*in_progress
)
10518 ASSERT(MUTEX_HELD(&spa
->spa_activities_lock
));
10520 switch (activity
) {
10521 case ZPOOL_WAIT_CKPT_DISCARD
:
10523 (spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
) &&
10524 zap_contains(spa_meta_objset(spa
),
10525 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ZPOOL_CHECKPOINT
) ==
10528 case ZPOOL_WAIT_FREE
:
10529 *in_progress
= ((spa_version(spa
) >= SPA_VERSION_DEADLISTS
&&
10530 !bpobj_is_empty(&spa
->spa_dsl_pool
->dp_free_bpobj
)) ||
10531 spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
) ||
10532 spa_livelist_delete_check(spa
));
10534 case ZPOOL_WAIT_INITIALIZE
:
10535 case ZPOOL_WAIT_TRIM
:
10536 error
= spa_vdev_activity_in_progress(spa
, use_tag
, tag
,
10537 activity
, in_progress
);
10539 case ZPOOL_WAIT_REPLACE
:
10540 mutex_exit(&spa
->spa_activities_lock
);
10541 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
10542 mutex_enter(&spa
->spa_activities_lock
);
10544 *in_progress
= vdev_replace_in_progress(spa
->spa_root_vdev
);
10545 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
10547 case ZPOOL_WAIT_REMOVE
:
10548 *in_progress
= (spa
->spa_removing_phys
.sr_state
==
10551 case ZPOOL_WAIT_RESILVER
:
10552 *in_progress
= vdev_rebuild_active(spa
->spa_root_vdev
);
10556 case ZPOOL_WAIT_SCRUB
:
10558 boolean_t scanning
, paused
, is_scrub
;
10559 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
10561 is_scrub
= (scn
->scn_phys
.scn_func
== POOL_SCAN_SCRUB
);
10562 scanning
= (scn
->scn_phys
.scn_state
== DSS_SCANNING
);
10563 paused
= dsl_scan_is_paused_scrub(scn
);
10564 *in_progress
= (scanning
&& !paused
&&
10565 is_scrub
== (activity
== ZPOOL_WAIT_SCRUB
));
10568 case ZPOOL_WAIT_RAIDZ_EXPAND
:
10570 vdev_raidz_expand_t
*vre
= spa
->spa_raidz_expand
;
10571 *in_progress
= (vre
!= NULL
&& vre
->vre_state
== DSS_SCANNING
);
10575 panic("unrecognized value for activity %d", activity
);
10582 spa_wait_common(const char *pool
, zpool_wait_activity_t activity
,
10583 boolean_t use_tag
, uint64_t tag
, boolean_t
*waited
)
10586 * The tag is used to distinguish between instances of an activity.
10587 * 'initialize' and 'trim' are the only activities that we use this for.
10588 * The other activities can only have a single instance in progress in a
10589 * pool at one time, making the tag unnecessary.
10591 * There can be multiple devices being replaced at once, but since they
10592 * all finish once resilvering finishes, we don't bother keeping track
10593 * of them individually, we just wait for them all to finish.
10595 if (use_tag
&& activity
!= ZPOOL_WAIT_INITIALIZE
&&
10596 activity
!= ZPOOL_WAIT_TRIM
)
10599 if (activity
< 0 || activity
>= ZPOOL_WAIT_NUM_ACTIVITIES
)
10603 int error
= spa_open(pool
, &spa
, FTAG
);
10608 * Increment the spa's waiter count so that we can call spa_close and
10609 * still ensure that the spa_t doesn't get freed before this thread is
10610 * finished with it when the pool is exported. We want to call spa_close
10611 * before we start waiting because otherwise the additional ref would
10612 * prevent the pool from being exported or destroyed throughout the
10613 * potentially long wait.
10615 mutex_enter(&spa
->spa_activities_lock
);
10616 spa
->spa_waiters
++;
10617 spa_close(spa
, FTAG
);
10621 boolean_t in_progress
;
10622 error
= spa_activity_in_progress(spa
, activity
, use_tag
, tag
,
10625 if (error
|| !in_progress
|| spa
->spa_waiters_cancel
)
10630 if (cv_wait_sig(&spa
->spa_activities_cv
,
10631 &spa
->spa_activities_lock
) == 0) {
10637 spa
->spa_waiters
--;
10638 cv_signal(&spa
->spa_waiters_cv
);
10639 mutex_exit(&spa
->spa_activities_lock
);
10645 * Wait for a particular instance of the specified activity to complete, where
10646 * the instance is identified by 'tag'
10649 spa_wait_tag(const char *pool
, zpool_wait_activity_t activity
, uint64_t tag
,
10652 return (spa_wait_common(pool
, activity
, B_TRUE
, tag
, waited
));
10656 * Wait for all instances of the specified activity complete
10659 spa_wait(const char *pool
, zpool_wait_activity_t activity
, boolean_t
*waited
)
10662 return (spa_wait_common(pool
, activity
, B_FALSE
, 0, waited
));
10666 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
10668 sysevent_t
*ev
= NULL
;
10670 nvlist_t
*resource
;
10672 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
10674 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
10675 ev
->resource
= resource
;
10678 (void) spa
, (void) vd
, (void) hist_nvl
, (void) name
;
10684 spa_event_post(sysevent_t
*ev
)
10688 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
10689 kmem_free(ev
, sizeof (*ev
));
10697 * Post a zevent corresponding to the given sysevent. The 'name' must be one
10698 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
10699 * filled in from the spa and (optionally) the vdev. This doesn't do anything
10700 * in the userland libzpool, as we don't want consumers to misinterpret ztest
10701 * or zdb as real changes.
10704 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
10706 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
10709 /* state manipulation functions */
10710 EXPORT_SYMBOL(spa_open
);
10711 EXPORT_SYMBOL(spa_open_rewind
);
10712 EXPORT_SYMBOL(spa_get_stats
);
10713 EXPORT_SYMBOL(spa_create
);
10714 EXPORT_SYMBOL(spa_import
);
10715 EXPORT_SYMBOL(spa_tryimport
);
10716 EXPORT_SYMBOL(spa_destroy
);
10717 EXPORT_SYMBOL(spa_export
);
10718 EXPORT_SYMBOL(spa_reset
);
10719 EXPORT_SYMBOL(spa_async_request
);
10720 EXPORT_SYMBOL(spa_async_suspend
);
10721 EXPORT_SYMBOL(spa_async_resume
);
10722 EXPORT_SYMBOL(spa_inject_addref
);
10723 EXPORT_SYMBOL(spa_inject_delref
);
10724 EXPORT_SYMBOL(spa_scan_stat_init
);
10725 EXPORT_SYMBOL(spa_scan_get_stats
);
10727 /* device manipulation */
10728 EXPORT_SYMBOL(spa_vdev_add
);
10729 EXPORT_SYMBOL(spa_vdev_attach
);
10730 EXPORT_SYMBOL(spa_vdev_detach
);
10731 EXPORT_SYMBOL(spa_vdev_setpath
);
10732 EXPORT_SYMBOL(spa_vdev_setfru
);
10733 EXPORT_SYMBOL(spa_vdev_split_mirror
);
10735 /* spare statech is global across all pools) */
10736 EXPORT_SYMBOL(spa_spare_add
);
10737 EXPORT_SYMBOL(spa_spare_remove
);
10738 EXPORT_SYMBOL(spa_spare_exists
);
10739 EXPORT_SYMBOL(spa_spare_activate
);
10741 /* L2ARC statech is global across all pools) */
10742 EXPORT_SYMBOL(spa_l2cache_add
);
10743 EXPORT_SYMBOL(spa_l2cache_remove
);
10744 EXPORT_SYMBOL(spa_l2cache_exists
);
10745 EXPORT_SYMBOL(spa_l2cache_activate
);
10746 EXPORT_SYMBOL(spa_l2cache_drop
);
10749 EXPORT_SYMBOL(spa_scan
);
10750 EXPORT_SYMBOL(spa_scan_stop
);
10753 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
10754 EXPORT_SYMBOL(spa_sync_allpools
);
10757 EXPORT_SYMBOL(spa_prop_set
);
10758 EXPORT_SYMBOL(spa_prop_get
);
10759 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
10761 /* asynchronous event notification */
10762 EXPORT_SYMBOL(spa_event_notify
);
10764 ZFS_MODULE_PARAM(zfs_metaslab
, metaslab_
, preload_pct
, UINT
, ZMOD_RW
,
10765 "Percentage of CPUs to run a metaslab preload taskq");
10767 /* BEGIN CSTYLED */
10768 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_shift
, UINT
, ZMOD_RW
,
10769 "log2 fraction of arc that can be used by inflight I/Os when "
10770 "verifying pool during import");
10773 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_metadata
, INT
, ZMOD_RW
,
10774 "Set to traverse metadata on pool import");
10776 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_verify_data
, INT
, ZMOD_RW
,
10777 "Set to traverse data on pool import");
10779 ZFS_MODULE_PARAM(zfs_spa
, spa_
, load_print_vdev_tree
, INT
, ZMOD_RW
,
10780 "Print vdev tree to zfs_dbgmsg during pool import");
10782 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_pct
, UINT
, ZMOD_RD
,
10783 "Percentage of CPUs to run an IO worker thread");
10785 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_batch_tpq
, UINT
, ZMOD_RD
,
10786 "Number of threads per IO worker taskqueue");
10788 /* BEGIN CSTYLED */
10789 ZFS_MODULE_PARAM(zfs
, zfs_
, max_missing_tvds
, U64
, ZMOD_RW
,
10790 "Allow importing pool with up to this number of missing top-level "
10791 "vdevs (in read-only mode)");
10794 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_pause
, INT
,
10795 ZMOD_RW
, "Set the livelist condense zthr to pause");
10797 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_pause
, INT
,
10798 ZMOD_RW
, "Set the livelist condense synctask to pause");
10800 /* BEGIN CSTYLED */
10801 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, sync_cancel
,
10803 "Whether livelist condensing was canceled in the synctask");
10805 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, zthr_cancel
,
10807 "Whether livelist condensing was canceled in the zthr function");
10809 ZFS_MODULE_PARAM(zfs_livelist_condense
, zfs_livelist_condense_
, new_alloc
, INT
,
10811 "Whether extra ALLOC blkptrs were added to a livelist entry while it "
10812 "was being condensed");
10815 ZFS_MODULE_VIRTUAL_PARAM_CALL(zfs_zio
, zio_
, taskq_read
,
10816 spa_taskq_read_param_set
, spa_taskq_read_param_get
, ZMOD_RD
,
10817 "Configure IO queues for read IO");
10818 ZFS_MODULE_VIRTUAL_PARAM_CALL(zfs_zio
, zio_
, taskq_write
,
10819 spa_taskq_write_param_set
, spa_taskq_write_param_get
, ZMOD_RD
,
10820 "Configure IO queues for write IO");
10824 ZFS_MODULE_PARAM(zfs_zio
, zio_
, taskq_wr_iss_ncpus
, UINT
, ZMOD_RW
,
10825 "Number of CPUs to run write issue taskqs");