4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
32 * Copyright (c) 2017 Datto Inc.
33 * Copyright 2017 Joyent, Inc.
37 * SPA: Storage Pool Allocator
39 * This file contains all the routines used when modifying on-disk SPA state.
40 * This includes opening, importing, destroying, exporting a pool, and syncing a
44 #include <sys/zfs_context.h>
45 #include <sys/fm/fs/zfs.h>
46 #include <sys/spa_impl.h>
48 #include <sys/zio_checksum.h>
50 #include <sys/dmu_tx.h>
54 #include <sys/vdev_impl.h>
55 #include <sys/vdev_removal.h>
56 #include <sys/vdev_indirect_mapping.h>
57 #include <sys/vdev_indirect_births.h>
58 #include <sys/vdev_disk.h>
59 #include <sys/metaslab.h>
60 #include <sys/metaslab_impl.h>
62 #include <sys/uberblock_impl.h>
65 #include <sys/bpobj.h>
66 #include <sys/dmu_traverse.h>
67 #include <sys/dmu_objset.h>
68 #include <sys/unique.h>
69 #include <sys/dsl_pool.h>
70 #include <sys/dsl_dataset.h>
71 #include <sys/dsl_dir.h>
72 #include <sys/dsl_prop.h>
73 #include <sys/dsl_synctask.h>
74 #include <sys/fs/zfs.h>
76 #include <sys/callb.h>
77 #include <sys/systeminfo.h>
78 #include <sys/spa_boot.h>
79 #include <sys/zfs_ioctl.h>
80 #include <sys/dsl_scan.h>
81 #include <sys/zfeature.h>
82 #include <sys/dsl_destroy.h>
86 #include <sys/fm/protocol.h>
87 #include <sys/fm/util.h>
88 #include <sys/callb.h>
93 #include "zfs_comutil.h"
96 * The interval, in seconds, at which failed configuration cache file writes
99 int zfs_ccw_retry_interval
= 300;
101 typedef enum zti_modes
{
102 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
103 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
104 ZTI_MODE_NULL
, /* don't create a taskq */
108 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
109 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
110 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
111 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
113 #define ZTI_N(n) ZTI_P(n, 1)
114 #define ZTI_ONE ZTI_N(1)
116 typedef struct zio_taskq_info
{
117 zti_modes_t zti_mode
;
122 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
123 "iss", "iss_h", "int", "int_h"
127 * This table defines the taskq settings for each ZFS I/O type. When
128 * initializing a pool, we use this table to create an appropriately sized
129 * taskq. Some operations are low volume and therefore have a small, static
130 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
131 * macros. Other operations process a large amount of data; the ZTI_BATCH
132 * macro causes us to create a taskq oriented for throughput. Some operations
133 * are so high frequency and short-lived that the taskq itself can become a a
134 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
135 * additional degree of parallelism specified by the number of threads per-
136 * taskq and the number of taskqs; when dispatching an event in this case, the
137 * particular taskq is chosen at random.
139 * The different taskq priorities are to handle the different contexts (issue
140 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
141 * need to be handled with minimum delay.
143 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
144 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
145 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
146 { ZTI_N(8), ZTI_NULL
, ZTI_P(12, 8), ZTI_NULL
}, /* READ */
147 { ZTI_BATCH
, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */
148 { ZTI_P(12, 8), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
149 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
150 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
153 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
154 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
155 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
156 static int spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
,
157 boolean_t reloading
);
158 static void spa_vdev_resilver_done(spa_t
*spa
);
160 uint_t zio_taskq_batch_pct
= 75; /* 1 thread per cpu in pset */
161 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
162 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
164 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
167 * Report any spa_load_verify errors found, but do not fail spa_load.
168 * This is used by zdb to analyze non-idle pools.
170 boolean_t spa_load_verify_dryrun
= B_FALSE
;
173 * This (illegal) pool name is used when temporarily importing a spa_t in order
174 * to get the vdev stats associated with the imported devices.
176 #define TRYIMPORT_NAME "$import"
179 * For debugging purposes: print out vdev tree during pool import.
181 int spa_load_print_vdev_tree
= B_FALSE
;
184 * A non-zero value for zfs_max_missing_tvds means that we allow importing
185 * pools with missing top-level vdevs. This is strictly intended for advanced
186 * pool recovery cases since missing data is almost inevitable. Pools with
187 * missing devices can only be imported read-only for safety reasons, and their
188 * fail-mode will be automatically set to "continue".
190 * With 1 missing vdev we should be able to import the pool and mount all
191 * datasets. User data that was not modified after the missing device has been
192 * added should be recoverable. This means that snapshots created prior to the
193 * addition of that device should be completely intact.
195 * With 2 missing vdevs, some datasets may fail to mount since there are
196 * dataset statistics that are stored as regular metadata. Some data might be
197 * recoverable if those vdevs were added recently.
199 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
200 * may be missing entirely. Chances of data recovery are very low. Note that
201 * there are also risks of performing an inadvertent rewind as we might be
202 * missing all the vdevs with the latest uberblocks.
204 unsigned long zfs_max_missing_tvds
= 0;
207 * The parameters below are similar to zfs_max_missing_tvds but are only
208 * intended for a preliminary open of the pool with an untrusted config which
209 * might be incomplete or out-dated.
211 * We are more tolerant for pools opened from a cachefile since we could have
212 * an out-dated cachefile where a device removal was not registered.
213 * We could have set the limit arbitrarily high but in the case where devices
214 * are really missing we would want to return the proper error codes; we chose
215 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
216 * and we get a chance to retrieve the trusted config.
218 uint64_t zfs_max_missing_tvds_cachefile
= SPA_DVAS_PER_BP
- 1;
220 * In the case where config was assembled by scanning device paths (/dev/dsks
221 * by default) we are less tolerant since all the existing devices should have
222 * been detected and we want spa_load to return the right error codes.
224 uint64_t zfs_max_missing_tvds_scan
= 0;
227 * ==========================================================================
228 * SPA properties routines
229 * ==========================================================================
233 * Add a (source=src, propname=propval) list to an nvlist.
236 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
237 uint64_t intval
, zprop_source_t src
)
239 const char *propname
= zpool_prop_to_name(prop
);
242 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
243 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
246 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
248 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
250 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
251 nvlist_free(propval
);
255 * Get property values from the spa configuration.
258 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
260 vdev_t
*rvd
= spa
->spa_root_vdev
;
261 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
262 uint64_t size
, alloc
, cap
, version
;
263 const zprop_source_t src
= ZPROP_SRC_NONE
;
264 spa_config_dirent_t
*dp
;
265 metaslab_class_t
*mc
= spa_normal_class(spa
);
267 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
270 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
271 size
= metaslab_class_get_space(spa_normal_class(spa
));
272 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
273 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
274 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
275 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
278 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
279 metaslab_class_fragmentation(mc
), src
);
280 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
281 metaslab_class_expandable_space(mc
), src
);
282 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
283 (spa_mode(spa
) == FREAD
), src
);
285 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
286 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
288 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
289 ddt_get_pool_dedup_ratio(spa
), src
);
291 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
292 rvd
->vdev_state
, src
);
294 version
= spa_version(spa
);
295 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
)) {
296 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
297 version
, ZPROP_SRC_DEFAULT
);
299 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
300 version
, ZPROP_SRC_LOCAL
);
306 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
307 * when opening pools before this version freedir will be NULL.
309 if (pool
->dp_free_dir
!= NULL
) {
310 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
311 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
314 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
318 if (pool
->dp_leak_dir
!= NULL
) {
319 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
320 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
323 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
328 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
330 if (spa
->spa_comment
!= NULL
) {
331 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
335 if (spa
->spa_root
!= NULL
)
336 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
339 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
340 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
341 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
343 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
344 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
347 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
)) {
348 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
349 DNODE_MAX_SIZE
, ZPROP_SRC_NONE
);
351 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
352 DNODE_MIN_SIZE
, ZPROP_SRC_NONE
);
355 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
356 if (dp
->scd_path
== NULL
) {
357 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
358 "none", 0, ZPROP_SRC_LOCAL
);
359 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
360 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
361 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
367 * Get zpool property values.
370 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
372 objset_t
*mos
= spa
->spa_meta_objset
;
377 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
381 mutex_enter(&spa
->spa_props_lock
);
384 * Get properties from the spa config.
386 spa_prop_get_config(spa
, nvp
);
388 /* If no pool property object, no more prop to get. */
389 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
390 mutex_exit(&spa
->spa_props_lock
);
395 * Get properties from the MOS pool property object.
397 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
398 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
399 zap_cursor_advance(&zc
)) {
402 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
405 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPOOL_PROP_INVAL
)
408 switch (za
.za_integer_length
) {
410 /* integer property */
411 if (za
.za_first_integer
!=
412 zpool_prop_default_numeric(prop
))
413 src
= ZPROP_SRC_LOCAL
;
415 if (prop
== ZPOOL_PROP_BOOTFS
) {
417 dsl_dataset_t
*ds
= NULL
;
419 dp
= spa_get_dsl(spa
);
420 dsl_pool_config_enter(dp
, FTAG
);
421 if ((err
= dsl_dataset_hold_obj(dp
,
422 za
.za_first_integer
, FTAG
, &ds
))) {
423 dsl_pool_config_exit(dp
, FTAG
);
427 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
429 dsl_dataset_name(ds
, strval
);
430 dsl_dataset_rele(ds
, FTAG
);
431 dsl_pool_config_exit(dp
, FTAG
);
434 intval
= za
.za_first_integer
;
437 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
440 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
445 /* string property */
446 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
447 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
448 za
.za_name
, 1, za
.za_num_integers
, strval
);
450 kmem_free(strval
, za
.za_num_integers
);
453 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
454 kmem_free(strval
, za
.za_num_integers
);
461 zap_cursor_fini(&zc
);
462 mutex_exit(&spa
->spa_props_lock
);
464 if (err
&& err
!= ENOENT
) {
474 * Validate the given pool properties nvlist and modify the list
475 * for the property values to be set.
478 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
481 int error
= 0, reset_bootfs
= 0;
483 boolean_t has_feature
= B_FALSE
;
486 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
488 char *strval
, *slash
, *check
, *fname
;
489 const char *propname
= nvpair_name(elem
);
490 zpool_prop_t prop
= zpool_name_to_prop(propname
);
493 case ZPOOL_PROP_INVAL
:
494 if (!zpool_prop_feature(propname
)) {
495 error
= SET_ERROR(EINVAL
);
500 * Sanitize the input.
502 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
503 error
= SET_ERROR(EINVAL
);
507 if (nvpair_value_uint64(elem
, &intval
) != 0) {
508 error
= SET_ERROR(EINVAL
);
513 error
= SET_ERROR(EINVAL
);
517 fname
= strchr(propname
, '@') + 1;
518 if (zfeature_lookup_name(fname
, NULL
) != 0) {
519 error
= SET_ERROR(EINVAL
);
523 has_feature
= B_TRUE
;
526 case ZPOOL_PROP_VERSION
:
527 error
= nvpair_value_uint64(elem
, &intval
);
529 (intval
< spa_version(spa
) ||
530 intval
> SPA_VERSION_BEFORE_FEATURES
||
532 error
= SET_ERROR(EINVAL
);
535 case ZPOOL_PROP_DELEGATION
:
536 case ZPOOL_PROP_AUTOREPLACE
:
537 case ZPOOL_PROP_LISTSNAPS
:
538 case ZPOOL_PROP_AUTOEXPAND
:
539 error
= nvpair_value_uint64(elem
, &intval
);
540 if (!error
&& intval
> 1)
541 error
= SET_ERROR(EINVAL
);
544 case ZPOOL_PROP_MULTIHOST
:
545 error
= nvpair_value_uint64(elem
, &intval
);
546 if (!error
&& intval
> 1)
547 error
= SET_ERROR(EINVAL
);
549 if (!error
&& !spa_get_hostid())
550 error
= SET_ERROR(ENOTSUP
);
554 case ZPOOL_PROP_BOOTFS
:
556 * If the pool version is less than SPA_VERSION_BOOTFS,
557 * or the pool is still being created (version == 0),
558 * the bootfs property cannot be set.
560 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
561 error
= SET_ERROR(ENOTSUP
);
566 * Make sure the vdev config is bootable
568 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
569 error
= SET_ERROR(ENOTSUP
);
575 error
= nvpair_value_string(elem
, &strval
);
581 if (strval
== NULL
|| strval
[0] == '\0') {
582 objnum
= zpool_prop_default_numeric(
587 error
= dmu_objset_hold(strval
, FTAG
, &os
);
592 * Must be ZPL, and its property settings
593 * must be supported by GRUB (compression
594 * is not gzip, and large blocks or large
595 * dnodes are not used).
598 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
599 error
= SET_ERROR(ENOTSUP
);
601 dsl_prop_get_int_ds(dmu_objset_ds(os
),
602 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
604 !BOOTFS_COMPRESS_VALID(propval
)) {
605 error
= SET_ERROR(ENOTSUP
);
607 dsl_prop_get_int_ds(dmu_objset_ds(os
),
608 zfs_prop_to_name(ZFS_PROP_DNODESIZE
),
610 propval
!= ZFS_DNSIZE_LEGACY
) {
611 error
= SET_ERROR(ENOTSUP
);
613 objnum
= dmu_objset_id(os
);
615 dmu_objset_rele(os
, FTAG
);
619 case ZPOOL_PROP_FAILUREMODE
:
620 error
= nvpair_value_uint64(elem
, &intval
);
621 if (!error
&& intval
> ZIO_FAILURE_MODE_PANIC
)
622 error
= SET_ERROR(EINVAL
);
625 * This is a special case which only occurs when
626 * the pool has completely failed. This allows
627 * the user to change the in-core failmode property
628 * without syncing it out to disk (I/Os might
629 * currently be blocked). We do this by returning
630 * EIO to the caller (spa_prop_set) to trick it
631 * into thinking we encountered a property validation
634 if (!error
&& spa_suspended(spa
)) {
635 spa
->spa_failmode
= intval
;
636 error
= SET_ERROR(EIO
);
640 case ZPOOL_PROP_CACHEFILE
:
641 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
644 if (strval
[0] == '\0')
647 if (strcmp(strval
, "none") == 0)
650 if (strval
[0] != '/') {
651 error
= SET_ERROR(EINVAL
);
655 slash
= strrchr(strval
, '/');
656 ASSERT(slash
!= NULL
);
658 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
659 strcmp(slash
, "/..") == 0)
660 error
= SET_ERROR(EINVAL
);
663 case ZPOOL_PROP_COMMENT
:
664 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
666 for (check
= strval
; *check
!= '\0'; check
++) {
667 if (!isprint(*check
)) {
668 error
= SET_ERROR(EINVAL
);
672 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
673 error
= SET_ERROR(E2BIG
);
676 case ZPOOL_PROP_DEDUPDITTO
:
677 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
678 error
= SET_ERROR(ENOTSUP
);
680 error
= nvpair_value_uint64(elem
, &intval
);
682 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
683 error
= SET_ERROR(EINVAL
);
694 if (!error
&& reset_bootfs
) {
695 error
= nvlist_remove(props
,
696 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
699 error
= nvlist_add_uint64(props
,
700 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
708 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
711 spa_config_dirent_t
*dp
;
713 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
717 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
720 if (cachefile
[0] == '\0')
721 dp
->scd_path
= spa_strdup(spa_config_path
);
722 else if (strcmp(cachefile
, "none") == 0)
725 dp
->scd_path
= spa_strdup(cachefile
);
727 list_insert_head(&spa
->spa_config_list
, dp
);
729 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
733 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
736 nvpair_t
*elem
= NULL
;
737 boolean_t need_sync
= B_FALSE
;
739 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
742 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
743 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
745 if (prop
== ZPOOL_PROP_CACHEFILE
||
746 prop
== ZPOOL_PROP_ALTROOT
||
747 prop
== ZPOOL_PROP_READONLY
)
750 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPOOL_PROP_INVAL
) {
753 if (prop
== ZPOOL_PROP_VERSION
) {
754 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
756 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
757 ver
= SPA_VERSION_FEATURES
;
761 /* Save time if the version is already set. */
762 if (ver
== spa_version(spa
))
766 * In addition to the pool directory object, we might
767 * create the pool properties object, the features for
768 * read object, the features for write object, or the
769 * feature descriptions object.
771 error
= dsl_sync_task(spa
->spa_name
, NULL
,
772 spa_sync_version
, &ver
,
773 6, ZFS_SPACE_CHECK_RESERVED
);
784 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
785 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
792 * If the bootfs property value is dsobj, clear it.
795 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
797 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
798 VERIFY(zap_remove(spa
->spa_meta_objset
,
799 spa
->spa_pool_props_object
,
800 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
807 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
809 ASSERTV(uint64_t *newguid
= arg
);
810 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
811 vdev_t
*rvd
= spa
->spa_root_vdev
;
814 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
815 vdev_state
= rvd
->vdev_state
;
816 spa_config_exit(spa
, SCL_STATE
, FTAG
);
818 if (vdev_state
!= VDEV_STATE_HEALTHY
)
819 return (SET_ERROR(ENXIO
));
821 ASSERT3U(spa_guid(spa
), !=, *newguid
);
827 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
829 uint64_t *newguid
= arg
;
830 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
832 vdev_t
*rvd
= spa
->spa_root_vdev
;
834 oldguid
= spa_guid(spa
);
836 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
837 rvd
->vdev_guid
= *newguid
;
838 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
839 vdev_config_dirty(rvd
);
840 spa_config_exit(spa
, SCL_STATE
, FTAG
);
842 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
847 * Change the GUID for the pool. This is done so that we can later
848 * re-import a pool built from a clone of our own vdevs. We will modify
849 * the root vdev's guid, our own pool guid, and then mark all of our
850 * vdevs dirty. Note that we must make sure that all our vdevs are
851 * online when we do this, or else any vdevs that weren't present
852 * would be orphaned from our pool. We are also going to issue a
853 * sysevent to update any watchers.
856 spa_change_guid(spa_t
*spa
)
861 mutex_enter(&spa
->spa_vdev_top_lock
);
862 mutex_enter(&spa_namespace_lock
);
863 guid
= spa_generate_guid(NULL
);
865 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
866 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
869 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
870 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
873 mutex_exit(&spa_namespace_lock
);
874 mutex_exit(&spa
->spa_vdev_top_lock
);
880 * ==========================================================================
881 * SPA state manipulation (open/create/destroy/import/export)
882 * ==========================================================================
886 spa_error_entry_compare(const void *a
, const void *b
)
888 const spa_error_entry_t
*sa
= (const spa_error_entry_t
*)a
;
889 const spa_error_entry_t
*sb
= (const spa_error_entry_t
*)b
;
892 ret
= memcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
893 sizeof (zbookmark_phys_t
));
895 return (AVL_ISIGN(ret
));
899 * Utility function which retrieves copies of the current logs and
900 * re-initializes them in the process.
903 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
905 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
907 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
908 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
910 avl_create(&spa
->spa_errlist_scrub
,
911 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
912 offsetof(spa_error_entry_t
, se_avl
));
913 avl_create(&spa
->spa_errlist_last
,
914 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
915 offsetof(spa_error_entry_t
, se_avl
));
919 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
921 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
922 enum zti_modes mode
= ztip
->zti_mode
;
923 uint_t value
= ztip
->zti_value
;
924 uint_t count
= ztip
->zti_count
;
925 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
927 boolean_t batch
= B_FALSE
;
929 if (mode
== ZTI_MODE_NULL
) {
931 tqs
->stqs_taskq
= NULL
;
935 ASSERT3U(count
, >, 0);
937 tqs
->stqs_count
= count
;
938 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
942 ASSERT3U(value
, >=, 1);
943 value
= MAX(value
, 1);
944 flags
|= TASKQ_DYNAMIC
;
949 flags
|= TASKQ_THREADS_CPU_PCT
;
950 value
= MIN(zio_taskq_batch_pct
, 100);
954 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
956 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
960 for (uint_t i
= 0; i
< count
; i
++) {
964 (void) snprintf(name
, sizeof (name
), "%s_%s",
965 zio_type_name
[t
], zio_taskq_types
[q
]);
967 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
969 flags
|= TASKQ_DC_BATCH
;
971 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
972 spa
->spa_proc
, zio_taskq_basedc
, flags
);
974 pri_t pri
= maxclsyspri
;
976 * The write issue taskq can be extremely CPU
977 * intensive. Run it at slightly less important
978 * priority than the other taskqs. Under Linux this
979 * means incrementing the priority value on platforms
980 * like illumos it should be decremented.
982 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
985 tq
= taskq_create_proc(name
, value
, pri
, 50,
986 INT_MAX
, spa
->spa_proc
, flags
);
989 tqs
->stqs_taskq
[i
] = tq
;
994 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
996 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
998 if (tqs
->stqs_taskq
== NULL
) {
999 ASSERT3U(tqs
->stqs_count
, ==, 0);
1003 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
1004 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
1005 taskq_destroy(tqs
->stqs_taskq
[i
]);
1008 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
1009 tqs
->stqs_taskq
= NULL
;
1013 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1014 * Note that a type may have multiple discrete taskqs to avoid lock contention
1015 * on the taskq itself. In that case we choose which taskq at random by using
1016 * the low bits of gethrtime().
1019 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1020 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
1022 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1025 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1026 ASSERT3U(tqs
->stqs_count
, !=, 0);
1028 if (tqs
->stqs_count
== 1) {
1029 tq
= tqs
->stqs_taskq
[0];
1031 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1034 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
1038 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1041 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1042 task_func_t
*func
, void *arg
, uint_t flags
)
1044 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1048 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1049 ASSERT3U(tqs
->stqs_count
, !=, 0);
1051 if (tqs
->stqs_count
== 1) {
1052 tq
= tqs
->stqs_taskq
[0];
1054 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1057 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1059 taskq_wait_id(tq
, id
);
1063 spa_create_zio_taskqs(spa_t
*spa
)
1065 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1066 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1067 spa_taskqs_init(spa
, t
, q
);
1073 * Disabled until spa_thread() can be adapted for Linux.
1075 #undef HAVE_SPA_THREAD
1077 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1079 spa_thread(void *arg
)
1081 psetid_t zio_taskq_psrset_bind
= PS_NONE
;
1082 callb_cpr_t cprinfo
;
1085 user_t
*pu
= PTOU(curproc
);
1087 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1090 ASSERT(curproc
!= &p0
);
1091 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1092 "zpool-%s", spa
->spa_name
);
1093 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1095 /* bind this thread to the requested psrset */
1096 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1098 mutex_enter(&cpu_lock
);
1099 mutex_enter(&pidlock
);
1100 mutex_enter(&curproc
->p_lock
);
1102 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1103 0, NULL
, NULL
) == 0) {
1104 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1107 "Couldn't bind process for zfs pool \"%s\" to "
1108 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1111 mutex_exit(&curproc
->p_lock
);
1112 mutex_exit(&pidlock
);
1113 mutex_exit(&cpu_lock
);
1117 if (zio_taskq_sysdc
) {
1118 sysdc_thread_enter(curthread
, 100, 0);
1121 spa
->spa_proc
= curproc
;
1122 spa
->spa_did
= curthread
->t_did
;
1124 spa_create_zio_taskqs(spa
);
1126 mutex_enter(&spa
->spa_proc_lock
);
1127 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1129 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1130 cv_broadcast(&spa
->spa_proc_cv
);
1132 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1133 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1134 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1135 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1137 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1138 spa
->spa_proc_state
= SPA_PROC_GONE
;
1139 spa
->spa_proc
= &p0
;
1140 cv_broadcast(&spa
->spa_proc_cv
);
1141 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1143 mutex_enter(&curproc
->p_lock
);
1149 * Activate an uninitialized pool.
1152 spa_activate(spa_t
*spa
, int mode
)
1154 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1156 spa
->spa_state
= POOL_STATE_ACTIVE
;
1157 spa
->spa_mode
= mode
;
1159 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1160 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1162 /* Try to create a covering process */
1163 mutex_enter(&spa
->spa_proc_lock
);
1164 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1165 ASSERT(spa
->spa_proc
== &p0
);
1168 #ifdef HAVE_SPA_THREAD
1169 /* Only create a process if we're going to be around a while. */
1170 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1171 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1173 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1174 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1175 cv_wait(&spa
->spa_proc_cv
,
1176 &spa
->spa_proc_lock
);
1178 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1179 ASSERT(spa
->spa_proc
!= &p0
);
1180 ASSERT(spa
->spa_did
!= 0);
1184 "Couldn't create process for zfs pool \"%s\"\n",
1189 #endif /* HAVE_SPA_THREAD */
1190 mutex_exit(&spa
->spa_proc_lock
);
1192 /* If we didn't create a process, we need to create our taskqs. */
1193 if (spa
->spa_proc
== &p0
) {
1194 spa_create_zio_taskqs(spa
);
1197 for (size_t i
= 0; i
< TXG_SIZE
; i
++)
1198 spa
->spa_txg_zio
[i
] = zio_root(spa
, NULL
, NULL
, 0);
1200 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1201 offsetof(vdev_t
, vdev_config_dirty_node
));
1202 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1203 offsetof(objset_t
, os_evicting_node
));
1204 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1205 offsetof(vdev_t
, vdev_state_dirty_node
));
1207 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1208 offsetof(struct vdev
, vdev_txg_node
));
1210 avl_create(&spa
->spa_errlist_scrub
,
1211 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1212 offsetof(spa_error_entry_t
, se_avl
));
1213 avl_create(&spa
->spa_errlist_last
,
1214 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1215 offsetof(spa_error_entry_t
, se_avl
));
1217 spa_keystore_init(&spa
->spa_keystore
);
1220 * This taskq is used to perform zvol-minor-related tasks
1221 * asynchronously. This has several advantages, including easy
1222 * resolution of various deadlocks (zfsonlinux bug #3681).
1224 * The taskq must be single threaded to ensure tasks are always
1225 * processed in the order in which they were dispatched.
1227 * A taskq per pool allows one to keep the pools independent.
1228 * This way if one pool is suspended, it will not impact another.
1230 * The preferred location to dispatch a zvol minor task is a sync
1231 * task. In this context, there is easy access to the spa_t and minimal
1232 * error handling is required because the sync task must succeed.
1234 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1238 * Taskq dedicated to prefetcher threads: this is used to prevent the
1239 * pool traverse code from monopolizing the global (and limited)
1240 * system_taskq by inappropriately scheduling long running tasks on it.
1242 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", boot_ncpus
,
1243 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1246 * The taskq to upgrade datasets in this pool. Currently used by
1247 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1249 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", boot_ncpus
,
1250 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1254 * Opposite of spa_activate().
1257 spa_deactivate(spa_t
*spa
)
1259 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1260 ASSERT(spa
->spa_dsl_pool
== NULL
);
1261 ASSERT(spa
->spa_root_vdev
== NULL
);
1262 ASSERT(spa
->spa_async_zio_root
== NULL
);
1263 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1265 spa_evicting_os_wait(spa
);
1267 if (spa
->spa_zvol_taskq
) {
1268 taskq_destroy(spa
->spa_zvol_taskq
);
1269 spa
->spa_zvol_taskq
= NULL
;
1272 if (spa
->spa_prefetch_taskq
) {
1273 taskq_destroy(spa
->spa_prefetch_taskq
);
1274 spa
->spa_prefetch_taskq
= NULL
;
1277 if (spa
->spa_upgrade_taskq
) {
1278 taskq_destroy(spa
->spa_upgrade_taskq
);
1279 spa
->spa_upgrade_taskq
= NULL
;
1282 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1284 list_destroy(&spa
->spa_config_dirty_list
);
1285 list_destroy(&spa
->spa_evicting_os_list
);
1286 list_destroy(&spa
->spa_state_dirty_list
);
1288 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1290 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1291 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1292 spa_taskqs_fini(spa
, t
, q
);
1296 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1297 ASSERT3P(spa
->spa_txg_zio
[i
], !=, NULL
);
1298 VERIFY0(zio_wait(spa
->spa_txg_zio
[i
]));
1299 spa
->spa_txg_zio
[i
] = NULL
;
1302 metaslab_class_destroy(spa
->spa_normal_class
);
1303 spa
->spa_normal_class
= NULL
;
1305 metaslab_class_destroy(spa
->spa_log_class
);
1306 spa
->spa_log_class
= NULL
;
1309 * If this was part of an import or the open otherwise failed, we may
1310 * still have errors left in the queues. Empty them just in case.
1312 spa_errlog_drain(spa
);
1313 avl_destroy(&spa
->spa_errlist_scrub
);
1314 avl_destroy(&spa
->spa_errlist_last
);
1316 spa_keystore_fini(&spa
->spa_keystore
);
1318 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1320 mutex_enter(&spa
->spa_proc_lock
);
1321 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1322 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1323 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1324 cv_broadcast(&spa
->spa_proc_cv
);
1325 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1326 ASSERT(spa
->spa_proc
!= &p0
);
1327 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1329 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1330 spa
->spa_proc_state
= SPA_PROC_NONE
;
1332 ASSERT(spa
->spa_proc
== &p0
);
1333 mutex_exit(&spa
->spa_proc_lock
);
1336 * We want to make sure spa_thread() has actually exited the ZFS
1337 * module, so that the module can't be unloaded out from underneath
1340 if (spa
->spa_did
!= 0) {
1341 thread_join(spa
->spa_did
);
1347 * Verify a pool configuration, and construct the vdev tree appropriately. This
1348 * will create all the necessary vdevs in the appropriate layout, with each vdev
1349 * in the CLOSED state. This will prep the pool before open/creation/import.
1350 * All vdev validation is done by the vdev_alloc() routine.
1353 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1354 uint_t id
, int atype
)
1360 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1363 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1366 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1369 if (error
== ENOENT
)
1375 return (SET_ERROR(EINVAL
));
1378 for (int c
= 0; c
< children
; c
++) {
1380 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1388 ASSERT(*vdp
!= NULL
);
1394 * Opposite of spa_load().
1397 spa_unload(spa_t
*spa
)
1401 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1403 spa_load_note(spa
, "UNLOADING");
1408 spa_async_suspend(spa
);
1413 if (spa
->spa_sync_on
) {
1414 txg_sync_stop(spa
->spa_dsl_pool
);
1415 spa
->spa_sync_on
= B_FALSE
;
1419 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1420 * to call it earlier, before we wait for async i/o to complete.
1421 * This ensures that there is no async metaslab prefetching, by
1422 * calling taskq_wait(mg_taskq).
1424 if (spa
->spa_root_vdev
!= NULL
) {
1425 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1426 for (int c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++)
1427 vdev_metaslab_fini(spa
->spa_root_vdev
->vdev_child
[c
]);
1428 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1431 if (spa
->spa_mmp
.mmp_thread
)
1432 mmp_thread_stop(spa
);
1435 * Wait for any outstanding async I/O to complete.
1437 if (spa
->spa_async_zio_root
!= NULL
) {
1438 for (int i
= 0; i
< max_ncpus
; i
++)
1439 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1440 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1441 spa
->spa_async_zio_root
= NULL
;
1444 if (spa
->spa_vdev_removal
!= NULL
) {
1445 spa_vdev_removal_destroy(spa
->spa_vdev_removal
);
1446 spa
->spa_vdev_removal
= NULL
;
1449 if (spa
->spa_condense_zthr
!= NULL
) {
1450 ASSERT(!zthr_isrunning(spa
->spa_condense_zthr
));
1451 zthr_destroy(spa
->spa_condense_zthr
);
1452 spa
->spa_condense_zthr
= NULL
;
1455 spa_condense_fini(spa
);
1457 bpobj_close(&spa
->spa_deferred_bpobj
);
1459 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1464 if (spa
->spa_root_vdev
)
1465 vdev_free(spa
->spa_root_vdev
);
1466 ASSERT(spa
->spa_root_vdev
== NULL
);
1469 * Close the dsl pool.
1471 if (spa
->spa_dsl_pool
) {
1472 dsl_pool_close(spa
->spa_dsl_pool
);
1473 spa
->spa_dsl_pool
= NULL
;
1474 spa
->spa_meta_objset
= NULL
;
1480 * Drop and purge level 2 cache
1482 spa_l2cache_drop(spa
);
1484 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1485 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1486 if (spa
->spa_spares
.sav_vdevs
) {
1487 kmem_free(spa
->spa_spares
.sav_vdevs
,
1488 spa
->spa_spares
.sav_count
* sizeof (void *));
1489 spa
->spa_spares
.sav_vdevs
= NULL
;
1491 if (spa
->spa_spares
.sav_config
) {
1492 nvlist_free(spa
->spa_spares
.sav_config
);
1493 spa
->spa_spares
.sav_config
= NULL
;
1495 spa
->spa_spares
.sav_count
= 0;
1497 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1498 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1499 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1501 if (spa
->spa_l2cache
.sav_vdevs
) {
1502 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1503 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1504 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1506 if (spa
->spa_l2cache
.sav_config
) {
1507 nvlist_free(spa
->spa_l2cache
.sav_config
);
1508 spa
->spa_l2cache
.sav_config
= NULL
;
1510 spa
->spa_l2cache
.sav_count
= 0;
1512 spa
->spa_async_suspended
= 0;
1514 spa
->spa_indirect_vdevs_loaded
= B_FALSE
;
1516 if (spa
->spa_comment
!= NULL
) {
1517 spa_strfree(spa
->spa_comment
);
1518 spa
->spa_comment
= NULL
;
1521 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1525 * Load (or re-load) the current list of vdevs describing the active spares for
1526 * this pool. When this is called, we have some form of basic information in
1527 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1528 * then re-generate a more complete list including status information.
1531 spa_load_spares(spa_t
*spa
)
1538 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1541 * First, close and free any existing spare vdevs.
1543 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1544 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1546 /* Undo the call to spa_activate() below */
1547 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1548 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1549 spa_spare_remove(tvd
);
1554 if (spa
->spa_spares
.sav_vdevs
)
1555 kmem_free(spa
->spa_spares
.sav_vdevs
,
1556 spa
->spa_spares
.sav_count
* sizeof (void *));
1558 if (spa
->spa_spares
.sav_config
== NULL
)
1561 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1562 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1564 spa
->spa_spares
.sav_count
= (int)nspares
;
1565 spa
->spa_spares
.sav_vdevs
= NULL
;
1571 * Construct the array of vdevs, opening them to get status in the
1572 * process. For each spare, there is potentially two different vdev_t
1573 * structures associated with it: one in the list of spares (used only
1574 * for basic validation purposes) and one in the active vdev
1575 * configuration (if it's spared in). During this phase we open and
1576 * validate each vdev on the spare list. If the vdev also exists in the
1577 * active configuration, then we also mark this vdev as an active spare.
1579 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1581 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1582 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1583 VDEV_ALLOC_SPARE
) == 0);
1586 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1588 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1589 B_FALSE
)) != NULL
) {
1590 if (!tvd
->vdev_isspare
)
1594 * We only mark the spare active if we were successfully
1595 * able to load the vdev. Otherwise, importing a pool
1596 * with a bad active spare would result in strange
1597 * behavior, because multiple pool would think the spare
1598 * is actively in use.
1600 * There is a vulnerability here to an equally bizarre
1601 * circumstance, where a dead active spare is later
1602 * brought back to life (onlined or otherwise). Given
1603 * the rarity of this scenario, and the extra complexity
1604 * it adds, we ignore the possibility.
1606 if (!vdev_is_dead(tvd
))
1607 spa_spare_activate(tvd
);
1611 vd
->vdev_aux
= &spa
->spa_spares
;
1613 if (vdev_open(vd
) != 0)
1616 if (vdev_validate_aux(vd
) == 0)
1621 * Recompute the stashed list of spares, with status information
1624 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1625 DATA_TYPE_NVLIST_ARRAY
) == 0);
1627 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1629 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1630 spares
[i
] = vdev_config_generate(spa
,
1631 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1632 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1633 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1634 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1635 nvlist_free(spares
[i
]);
1636 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1640 * Load (or re-load) the current list of vdevs describing the active l2cache for
1641 * this pool. When this is called, we have some form of basic information in
1642 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1643 * then re-generate a more complete list including status information.
1644 * Devices which are already active have their details maintained, and are
1648 spa_load_l2cache(spa_t
*spa
)
1650 nvlist_t
**l2cache
= NULL
;
1652 int i
, j
, oldnvdevs
;
1654 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1655 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1657 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1659 oldvdevs
= sav
->sav_vdevs
;
1660 oldnvdevs
= sav
->sav_count
;
1661 sav
->sav_vdevs
= NULL
;
1664 if (sav
->sav_config
== NULL
) {
1670 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1671 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1672 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1675 * Process new nvlist of vdevs.
1677 for (i
= 0; i
< nl2cache
; i
++) {
1678 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1682 for (j
= 0; j
< oldnvdevs
; j
++) {
1684 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1686 * Retain previous vdev for add/remove ops.
1694 if (newvdevs
[i
] == NULL
) {
1698 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1699 VDEV_ALLOC_L2CACHE
) == 0);
1704 * Commit this vdev as an l2cache device,
1705 * even if it fails to open.
1707 spa_l2cache_add(vd
);
1712 spa_l2cache_activate(vd
);
1714 if (vdev_open(vd
) != 0)
1717 (void) vdev_validate_aux(vd
);
1719 if (!vdev_is_dead(vd
))
1720 l2arc_add_vdev(spa
, vd
);
1724 sav
->sav_vdevs
= newvdevs
;
1725 sav
->sav_count
= (int)nl2cache
;
1728 * Recompute the stashed list of l2cache devices, with status
1729 * information this time.
1731 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1732 DATA_TYPE_NVLIST_ARRAY
) == 0);
1734 if (sav
->sav_count
> 0)
1735 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *),
1737 for (i
= 0; i
< sav
->sav_count
; i
++)
1738 l2cache
[i
] = vdev_config_generate(spa
,
1739 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1740 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1741 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1745 * Purge vdevs that were dropped
1747 for (i
= 0; i
< oldnvdevs
; i
++) {
1752 ASSERT(vd
->vdev_isl2cache
);
1754 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1755 pool
!= 0ULL && l2arc_vdev_present(vd
))
1756 l2arc_remove_vdev(vd
);
1757 vdev_clear_stats(vd
);
1763 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1765 for (i
= 0; i
< sav
->sav_count
; i
++)
1766 nvlist_free(l2cache
[i
]);
1768 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1772 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1775 char *packed
= NULL
;
1780 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1784 nvsize
= *(uint64_t *)db
->db_data
;
1785 dmu_buf_rele(db
, FTAG
);
1787 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1788 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1791 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1792 vmem_free(packed
, nvsize
);
1798 * Concrete top-level vdevs that are not missing and are not logs. At every
1799 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1802 spa_healthy_core_tvds(spa_t
*spa
)
1804 vdev_t
*rvd
= spa
->spa_root_vdev
;
1807 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
1808 vdev_t
*vd
= rvd
->vdev_child
[i
];
1811 if (vdev_is_concrete(vd
) && !vdev_is_dead(vd
))
1819 * Checks to see if the given vdev could not be opened, in which case we post a
1820 * sysevent to notify the autoreplace code that the device has been removed.
1823 spa_check_removed(vdev_t
*vd
)
1825 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1826 spa_check_removed(vd
->vdev_child
[c
]);
1828 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1829 vdev_is_concrete(vd
)) {
1830 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1831 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
1836 spa_check_for_missing_logs(spa_t
*spa
)
1838 vdev_t
*rvd
= spa
->spa_root_vdev
;
1841 * If we're doing a normal import, then build up any additional
1842 * diagnostic information about missing log devices.
1843 * We'll pass this up to the user for further processing.
1845 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1846 nvlist_t
**child
, *nv
;
1849 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
1851 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1853 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
1854 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1857 * We consider a device as missing only if it failed
1858 * to open (i.e. offline or faulted is not considered
1861 if (tvd
->vdev_islog
&&
1862 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
1863 child
[idx
++] = vdev_config_generate(spa
, tvd
,
1864 B_FALSE
, VDEV_CONFIG_MISSING
);
1869 fnvlist_add_nvlist_array(nv
,
1870 ZPOOL_CONFIG_CHILDREN
, child
, idx
);
1871 fnvlist_add_nvlist(spa
->spa_load_info
,
1872 ZPOOL_CONFIG_MISSING_DEVICES
, nv
);
1874 for (uint64_t i
= 0; i
< idx
; i
++)
1875 nvlist_free(child
[i
]);
1878 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1881 spa_load_failed(spa
, "some log devices are missing");
1882 vdev_dbgmsg_print_tree(rvd
, 2);
1883 return (SET_ERROR(ENXIO
));
1886 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
1887 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1889 if (tvd
->vdev_islog
&&
1890 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
1891 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1892 spa_load_note(spa
, "some log devices are "
1893 "missing, ZIL is dropped.");
1894 vdev_dbgmsg_print_tree(rvd
, 2);
1904 * Check for missing log devices
1907 spa_check_logs(spa_t
*spa
)
1909 boolean_t rv
= B_FALSE
;
1910 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1912 switch (spa
->spa_log_state
) {
1915 case SPA_LOG_MISSING
:
1916 /* need to recheck in case slog has been restored */
1917 case SPA_LOG_UNKNOWN
:
1918 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1919 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1921 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1928 spa_passivate_log(spa_t
*spa
)
1930 vdev_t
*rvd
= spa
->spa_root_vdev
;
1931 boolean_t slog_found
= B_FALSE
;
1933 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1935 if (!spa_has_slogs(spa
))
1938 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1939 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1940 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1942 if (tvd
->vdev_islog
) {
1943 metaslab_group_passivate(mg
);
1944 slog_found
= B_TRUE
;
1948 return (slog_found
);
1952 spa_activate_log(spa_t
*spa
)
1954 vdev_t
*rvd
= spa
->spa_root_vdev
;
1956 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1958 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1959 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1960 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1962 if (tvd
->vdev_islog
)
1963 metaslab_group_activate(mg
);
1968 spa_reset_logs(spa_t
*spa
)
1972 error
= dmu_objset_find(spa_name(spa
), zil_reset
,
1973 NULL
, DS_FIND_CHILDREN
);
1976 * We successfully offlined the log device, sync out the
1977 * current txg so that the "stubby" block can be removed
1980 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1986 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1988 for (int i
= 0; i
< sav
->sav_count
; i
++)
1989 spa_check_removed(sav
->sav_vdevs
[i
]);
1993 spa_claim_notify(zio_t
*zio
)
1995 spa_t
*spa
= zio
->io_spa
;
2000 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
2001 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
2002 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
2003 mutex_exit(&spa
->spa_props_lock
);
2006 typedef struct spa_load_error
{
2007 uint64_t sle_meta_count
;
2008 uint64_t sle_data_count
;
2012 spa_load_verify_done(zio_t
*zio
)
2014 blkptr_t
*bp
= zio
->io_bp
;
2015 spa_load_error_t
*sle
= zio
->io_private
;
2016 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2017 int error
= zio
->io_error
;
2018 spa_t
*spa
= zio
->io_spa
;
2020 abd_free(zio
->io_abd
);
2022 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2023 type
!= DMU_OT_INTENT_LOG
)
2024 atomic_inc_64(&sle
->sle_meta_count
);
2026 atomic_inc_64(&sle
->sle_data_count
);
2029 mutex_enter(&spa
->spa_scrub_lock
);
2030 spa
->spa_load_verify_ios
--;
2031 cv_broadcast(&spa
->spa_scrub_io_cv
);
2032 mutex_exit(&spa
->spa_scrub_lock
);
2036 * Maximum number of concurrent scrub i/os to create while verifying
2037 * a pool while importing it.
2039 int spa_load_verify_maxinflight
= 10000;
2040 int spa_load_verify_metadata
= B_TRUE
;
2041 int spa_load_verify_data
= B_TRUE
;
2045 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2046 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2048 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
2051 * Note: normally this routine will not be called if
2052 * spa_load_verify_metadata is not set. However, it may be useful
2053 * to manually set the flag after the traversal has begun.
2055 if (!spa_load_verify_metadata
)
2057 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2061 size_t size
= BP_GET_PSIZE(bp
);
2063 mutex_enter(&spa
->spa_scrub_lock
);
2064 while (spa
->spa_load_verify_ios
>= spa_load_verify_maxinflight
)
2065 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2066 spa
->spa_load_verify_ios
++;
2067 mutex_exit(&spa
->spa_scrub_lock
);
2069 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2070 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2071 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2072 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2078 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2080 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2081 return (SET_ERROR(ENAMETOOLONG
));
2087 spa_load_verify(spa_t
*spa
)
2090 spa_load_error_t sle
= { 0 };
2091 zpool_load_policy_t policy
;
2092 boolean_t verify_ok
= B_FALSE
;
2095 zpool_get_load_policy(spa
->spa_config
, &policy
);
2097 if (policy
.zlp_rewind
& ZPOOL_NEVER_REWIND
)
2100 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2101 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2102 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2104 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2108 rio
= zio_root(spa
, NULL
, &sle
,
2109 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2111 if (spa_load_verify_metadata
) {
2112 if (spa
->spa_extreme_rewind
) {
2113 spa_load_note(spa
, "performing a complete scan of the "
2114 "pool since extreme rewind is on. This may take "
2115 "a very long time.\n (spa_load_verify_data=%u, "
2116 "spa_load_verify_metadata=%u)",
2117 spa_load_verify_data
, spa_load_verify_metadata
);
2119 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2120 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2121 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2124 (void) zio_wait(rio
);
2126 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2127 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2129 if (sle
.sle_meta_count
!= 0 || sle
.sle_data_count
!= 0) {
2130 spa_load_note(spa
, "spa_load_verify found %llu metadata errors "
2131 "and %llu data errors", (u_longlong_t
)sle
.sle_meta_count
,
2132 (u_longlong_t
)sle
.sle_data_count
);
2135 if (spa_load_verify_dryrun
||
2136 (!error
&& sle
.sle_meta_count
<= policy
.zlp_maxmeta
&&
2137 sle
.sle_data_count
<= policy
.zlp_maxdata
)) {
2141 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2142 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2144 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2145 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2146 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2147 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2148 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2149 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2150 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2152 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2155 if (spa_load_verify_dryrun
)
2159 if (error
!= ENXIO
&& error
!= EIO
)
2160 error
= SET_ERROR(EIO
);
2164 return (verify_ok
? 0 : EIO
);
2168 * Find a value in the pool props object.
2171 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2173 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2174 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2178 * Find a value in the pool directory object.
2181 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
, boolean_t log_enoent
)
2183 int error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2184 name
, sizeof (uint64_t), 1, val
);
2186 if (error
!= 0 && (error
!= ENOENT
|| log_enoent
)) {
2187 spa_load_failed(spa
, "couldn't get '%s' value in MOS directory "
2188 "[error=%d]", name
, error
);
2195 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2197 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2198 return (SET_ERROR(err
));
2202 spa_spawn_aux_threads(spa_t
*spa
)
2204 ASSERT(spa_writeable(spa
));
2206 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2208 spa_start_indirect_condensing_thread(spa
);
2212 * Fix up config after a partly-completed split. This is done with the
2213 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2214 * pool have that entry in their config, but only the splitting one contains
2215 * a list of all the guids of the vdevs that are being split off.
2217 * This function determines what to do with that list: either rejoin
2218 * all the disks to the pool, or complete the splitting process. To attempt
2219 * the rejoin, each disk that is offlined is marked online again, and
2220 * we do a reopen() call. If the vdev label for every disk that was
2221 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2222 * then we call vdev_split() on each disk, and complete the split.
2224 * Otherwise we leave the config alone, with all the vdevs in place in
2225 * the original pool.
2228 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2235 boolean_t attempt_reopen
;
2237 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2240 /* check that the config is complete */
2241 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2242 &glist
, &gcount
) != 0)
2245 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2247 /* attempt to online all the vdevs & validate */
2248 attempt_reopen
= B_TRUE
;
2249 for (i
= 0; i
< gcount
; i
++) {
2250 if (glist
[i
] == 0) /* vdev is hole */
2253 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2254 if (vd
[i
] == NULL
) {
2256 * Don't bother attempting to reopen the disks;
2257 * just do the split.
2259 attempt_reopen
= B_FALSE
;
2261 /* attempt to re-online it */
2262 vd
[i
]->vdev_offline
= B_FALSE
;
2266 if (attempt_reopen
) {
2267 vdev_reopen(spa
->spa_root_vdev
);
2269 /* check each device to see what state it's in */
2270 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2271 if (vd
[i
] != NULL
&&
2272 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2279 * If every disk has been moved to the new pool, or if we never
2280 * even attempted to look at them, then we split them off for
2283 if (!attempt_reopen
|| gcount
== extracted
) {
2284 for (i
= 0; i
< gcount
; i
++)
2287 vdev_reopen(spa
->spa_root_vdev
);
2290 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2294 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
)
2296 char *ereport
= FM_EREPORT_ZFS_POOL
;
2299 spa
->spa_load_state
= state
;
2301 gethrestime(&spa
->spa_loaded_ts
);
2302 error
= spa_load_impl(spa
, type
, &ereport
, B_FALSE
);
2305 * Don't count references from objsets that are already closed
2306 * and are making their way through the eviction process.
2308 spa_evicting_os_wait(spa
);
2309 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2311 if (error
!= EEXIST
) {
2312 spa
->spa_loaded_ts
.tv_sec
= 0;
2313 spa
->spa_loaded_ts
.tv_nsec
= 0;
2315 if (error
!= EBADF
) {
2316 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, NULL
, 0, 0);
2319 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2327 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2328 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2329 * spa's per-vdev ZAP list.
2332 vdev_count_verify_zaps(vdev_t
*vd
)
2334 spa_t
*spa
= vd
->vdev_spa
;
2337 if (vd
->vdev_top_zap
!= 0) {
2339 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2340 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2342 if (vd
->vdev_leaf_zap
!= 0) {
2344 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2345 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2348 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2349 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2357 * Determine whether the activity check is required.
2360 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
2364 uint64_t hostid
= 0;
2365 uint64_t tryconfig_txg
= 0;
2366 uint64_t tryconfig_timestamp
= 0;
2369 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2370 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2371 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2373 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2374 &tryconfig_timestamp
);
2377 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2380 * Disable the MMP activity check - This is used by zdb which
2381 * is intended to be used on potentially active pools.
2383 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
2387 * Skip the activity check when the MMP feature is disabled.
2389 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
2392 * If the tryconfig_* values are nonzero, they are the results of an
2393 * earlier tryimport. If they match the uberblock we just found, then
2394 * the pool has not changed and we return false so we do not test a
2397 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
2398 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
)
2402 * Allow the activity check to be skipped when importing the pool
2403 * on the same host which last imported it. Since the hostid from
2404 * configuration may be stale use the one read from the label.
2406 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
2407 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
2409 if (hostid
== spa_get_hostid())
2413 * Skip the activity test when the pool was cleanly exported.
2415 if (state
!= POOL_STATE_ACTIVE
)
2422 * Perform the import activity check. If the user canceled the import or
2423 * we detected activity then fail.
2426 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2428 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2429 uint64_t txg
= ub
->ub_txg
;
2430 uint64_t timestamp
= ub
->ub_timestamp
;
2431 uint64_t import_delay
= NANOSEC
;
2432 hrtime_t import_expire
;
2433 nvlist_t
*mmp_label
= NULL
;
2434 vdev_t
*rvd
= spa
->spa_root_vdev
;
2439 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
2440 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2444 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2445 * during the earlier tryimport. If the txg recorded there is 0 then
2446 * the pool is known to be active on another host.
2448 * Otherwise, the pool might be in use on another node. Check for
2449 * changes in the uberblocks on disk if necessary.
2451 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2452 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
2453 ZPOOL_CONFIG_LOAD_INFO
);
2455 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
2456 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
2457 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2458 error
= SET_ERROR(EREMOTEIO
);
2464 * Preferentially use the zfs_multihost_interval from the node which
2465 * last imported the pool. This value is stored in an MMP uberblock as.
2467 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2469 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
)
2470 import_delay
= MAX(import_delay
, import_intervals
*
2471 ub
->ub_mmp_delay
* MAX(vdev_count_leaves(spa
), 1));
2473 /* Apply a floor using the local default values. */
2474 import_delay
= MAX(import_delay
, import_intervals
*
2475 MSEC2NSEC(MAX(zfs_multihost_interval
, MMP_MIN_INTERVAL
)));
2477 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu import_intervals=%u "
2478 "leaves=%u", import_delay
, ub
->ub_mmp_delay
, import_intervals
,
2479 vdev_count_leaves(spa
));
2481 /* Add a small random factor in case of simultaneous imports (0-25%) */
2482 import_expire
= gethrtime() + import_delay
+
2483 (import_delay
* spa_get_random(250) / 1000);
2485 while (gethrtime() < import_expire
) {
2486 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2488 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
) {
2489 error
= SET_ERROR(EREMOTEIO
);
2494 nvlist_free(mmp_label
);
2498 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
2500 error
= SET_ERROR(EINTR
);
2508 mutex_destroy(&mtx
);
2512 * If the pool is determined to be active store the status in the
2513 * spa->spa_load_info nvlist. If the remote hostname or hostid are
2514 * available from configuration read from disk store them as well.
2515 * This allows 'zpool import' to generate a more useful message.
2517 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
2518 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2519 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
2521 if (error
== EREMOTEIO
) {
2522 char *hostname
= "<unknown>";
2523 uint64_t hostid
= 0;
2526 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
2527 hostname
= fnvlist_lookup_string(mmp_label
,
2528 ZPOOL_CONFIG_HOSTNAME
);
2529 fnvlist_add_string(spa
->spa_load_info
,
2530 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
2533 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
2534 hostid
= fnvlist_lookup_uint64(mmp_label
,
2535 ZPOOL_CONFIG_HOSTID
);
2536 fnvlist_add_uint64(spa
->spa_load_info
,
2537 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
2541 fnvlist_add_uint64(spa
->spa_load_info
,
2542 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
2543 fnvlist_add_uint64(spa
->spa_load_info
,
2544 ZPOOL_CONFIG_MMP_TXG
, 0);
2546 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
2550 nvlist_free(mmp_label
);
2556 spa_verify_host(spa_t
*spa
, nvlist_t
*mos_config
)
2560 uint64_t myhostid
= 0;
2562 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
2563 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2564 hostname
= fnvlist_lookup_string(mos_config
,
2565 ZPOOL_CONFIG_HOSTNAME
);
2567 myhostid
= zone_get_hostid(NULL
);
2569 if (hostid
!= 0 && myhostid
!= 0 && hostid
!= myhostid
) {
2570 cmn_err(CE_WARN
, "pool '%s' could not be "
2571 "loaded as it was last accessed by "
2572 "another system (host: %s hostid: 0x%llx). "
2573 "See: http://illumos.org/msg/ZFS-8000-EY",
2574 spa_name(spa
), hostname
, (u_longlong_t
)hostid
);
2575 spa_load_failed(spa
, "hostid verification failed: pool "
2576 "last accessed by host: %s (hostid: 0x%llx)",
2577 hostname
, (u_longlong_t
)hostid
);
2578 return (SET_ERROR(EBADF
));
2586 spa_ld_parse_config(spa_t
*spa
, spa_import_type_t type
)
2589 nvlist_t
*nvtree
, *nvl
, *config
= spa
->spa_config
;
2596 * Versioning wasn't explicitly added to the label until later, so if
2597 * it's not present treat it as the initial version.
2599 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2600 &spa
->spa_ubsync
.ub_version
) != 0)
2601 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2603 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
2604 spa_load_failed(spa
, "invalid config provided: '%s' missing",
2605 ZPOOL_CONFIG_POOL_GUID
);
2606 return (SET_ERROR(EINVAL
));
2609 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
|| spa
->spa_load_state
==
2610 SPA_LOAD_TRYIMPORT
) && spa_guid_exists(pool_guid
, 0)) {
2611 spa_load_failed(spa
, "a pool with guid %llu is already open",
2612 (u_longlong_t
)pool_guid
);
2613 return (SET_ERROR(EEXIST
));
2616 spa
->spa_config_guid
= pool_guid
;
2618 nvlist_free(spa
->spa_load_info
);
2619 spa
->spa_load_info
= fnvlist_alloc();
2621 ASSERT(spa
->spa_comment
== NULL
);
2622 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2623 spa
->spa_comment
= spa_strdup(comment
);
2625 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2626 &spa
->spa_config_txg
);
2628 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) == 0)
2629 spa
->spa_config_splitting
= fnvlist_dup(nvl
);
2631 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
)) {
2632 spa_load_failed(spa
, "invalid config provided: '%s' missing",
2633 ZPOOL_CONFIG_VDEV_TREE
);
2634 return (SET_ERROR(EINVAL
));
2638 * Create "The Godfather" zio to hold all async IOs
2640 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2642 for (int i
= 0; i
< max_ncpus
; i
++) {
2643 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2644 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2645 ZIO_FLAG_GODFATHER
);
2649 * Parse the configuration into a vdev tree. We explicitly set the
2650 * value that will be returned by spa_version() since parsing the
2651 * configuration requires knowing the version number.
2653 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2654 parse
= (type
== SPA_IMPORT_EXISTING
?
2655 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2656 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
2657 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2660 spa_load_failed(spa
, "unable to parse config [error=%d]",
2665 ASSERT(spa
->spa_root_vdev
== rvd
);
2666 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2667 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2669 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2670 ASSERT(spa_guid(spa
) == pool_guid
);
2677 * Recursively open all vdevs in the vdev tree. This function is called twice:
2678 * first with the untrusted config, then with the trusted config.
2681 spa_ld_open_vdevs(spa_t
*spa
)
2686 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2687 * missing/unopenable for the root vdev to be still considered openable.
2689 if (spa
->spa_trust_config
) {
2690 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds
;
2691 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_CACHEFILE
) {
2692 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_cachefile
;
2693 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_SCAN
) {
2694 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_scan
;
2696 spa
->spa_missing_tvds_allowed
= 0;
2699 spa
->spa_missing_tvds_allowed
=
2700 MAX(zfs_max_missing_tvds
, spa
->spa_missing_tvds_allowed
);
2702 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2703 error
= vdev_open(spa
->spa_root_vdev
);
2704 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2706 if (spa
->spa_missing_tvds
!= 0) {
2707 spa_load_note(spa
, "vdev tree has %lld missing top-level "
2708 "vdevs.", (u_longlong_t
)spa
->spa_missing_tvds
);
2709 if (spa
->spa_trust_config
&& (spa
->spa_mode
& FWRITE
)) {
2711 * Although theoretically we could allow users to open
2712 * incomplete pools in RW mode, we'd need to add a lot
2713 * of extra logic (e.g. adjust pool space to account
2714 * for missing vdevs).
2715 * This limitation also prevents users from accidentally
2716 * opening the pool in RW mode during data recovery and
2717 * damaging it further.
2719 spa_load_note(spa
, "pools with missing top-level "
2720 "vdevs can only be opened in read-only mode.");
2721 error
= SET_ERROR(ENXIO
);
2723 spa_load_note(spa
, "current settings allow for maximum "
2724 "%lld missing top-level vdevs at this stage.",
2725 (u_longlong_t
)spa
->spa_missing_tvds_allowed
);
2729 spa_load_failed(spa
, "unable to open vdev tree [error=%d]",
2732 if (spa
->spa_missing_tvds
!= 0 || error
!= 0)
2733 vdev_dbgmsg_print_tree(spa
->spa_root_vdev
, 2);
2739 * We need to validate the vdev labels against the configuration that
2740 * we have in hand. This function is called twice: first with an untrusted
2741 * config, then with a trusted config. The validation is more strict when the
2742 * config is trusted.
2745 spa_ld_validate_vdevs(spa_t
*spa
)
2748 vdev_t
*rvd
= spa
->spa_root_vdev
;
2750 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2751 error
= vdev_validate(rvd
);
2752 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2755 spa_load_failed(spa
, "vdev_validate failed [error=%d]", error
);
2759 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
2760 spa_load_failed(spa
, "cannot open vdev tree after invalidating "
2762 vdev_dbgmsg_print_tree(rvd
, 2);
2763 return (SET_ERROR(ENXIO
));
2770 spa_ld_select_uberblock(spa_t
*spa
, spa_import_type_t type
)
2772 vdev_t
*rvd
= spa
->spa_root_vdev
;
2774 uberblock_t
*ub
= &spa
->spa_uberblock
;
2775 boolean_t activity_check
= B_FALSE
;
2778 * Find the best uberblock.
2780 vdev_uberblock_load(rvd
, ub
, &label
);
2783 * If we weren't able to find a single valid uberblock, return failure.
2785 if (ub
->ub_txg
== 0) {
2787 spa_load_failed(spa
, "no valid uberblock found");
2788 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2791 spa_load_note(spa
, "using uberblock with txg=%llu",
2792 (u_longlong_t
)ub
->ub_txg
);
2796 * For pools which have the multihost property on determine if the
2797 * pool is truly inactive and can be safely imported. Prevent
2798 * hosts which don't have a hostid set from importing the pool.
2800 activity_check
= spa_activity_check_required(spa
, ub
, label
,
2802 if (activity_check
) {
2803 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
2804 spa_get_hostid() == 0) {
2806 fnvlist_add_uint64(spa
->spa_load_info
,
2807 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
2808 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
2811 int error
= spa_activity_check(spa
, ub
, spa
->spa_config
);
2817 fnvlist_add_uint64(spa
->spa_load_info
,
2818 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
2819 fnvlist_add_uint64(spa
->spa_load_info
,
2820 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
2824 * If the pool has an unsupported version we can't open it.
2826 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2828 spa_load_failed(spa
, "version %llu is not supported",
2829 (u_longlong_t
)ub
->ub_version
);
2830 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2833 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2837 * If we weren't able to find what's necessary for reading the
2838 * MOS in the label, return failure.
2840 if (label
== NULL
) {
2841 spa_load_failed(spa
, "label config unavailable");
2842 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2846 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
2849 spa_load_failed(spa
, "invalid label: '%s' missing",
2850 ZPOOL_CONFIG_FEATURES_FOR_READ
);
2851 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2856 * Update our in-core representation with the definitive values
2859 nvlist_free(spa
->spa_label_features
);
2860 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2866 * Look through entries in the label nvlist's features_for_read. If
2867 * there is a feature listed there which we don't understand then we
2868 * cannot open a pool.
2870 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2871 nvlist_t
*unsup_feat
;
2873 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2876 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
2878 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2879 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2880 VERIFY(nvlist_add_string(unsup_feat
,
2881 nvpair_name(nvp
), "") == 0);
2885 if (!nvlist_empty(unsup_feat
)) {
2886 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2887 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2888 nvlist_free(unsup_feat
);
2889 spa_load_failed(spa
, "some features are unsupported");
2890 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2894 nvlist_free(unsup_feat
);
2897 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2898 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2899 spa_try_repair(spa
, spa
->spa_config
);
2900 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2901 nvlist_free(spa
->spa_config_splitting
);
2902 spa
->spa_config_splitting
= NULL
;
2906 * Initialize internal SPA structures.
2908 spa
->spa_state
= POOL_STATE_ACTIVE
;
2909 spa
->spa_ubsync
= spa
->spa_uberblock
;
2910 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2911 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2912 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2913 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2914 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2915 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2921 spa_ld_open_rootbp(spa_t
*spa
)
2924 vdev_t
*rvd
= spa
->spa_root_vdev
;
2926 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2928 spa_load_failed(spa
, "unable to open rootbp in dsl_pool_init "
2929 "[error=%d]", error
);
2930 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2932 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2938 spa_ld_load_trusted_config(spa_t
*spa
, spa_import_type_t type
,
2939 boolean_t reloading
)
2941 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
2942 nvlist_t
*nv
, *mos_config
, *policy
;
2943 int error
= 0, copy_error
;
2944 uint64_t healthy_tvds
, healthy_tvds_mos
;
2945 uint64_t mos_config_txg
;
2947 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
, B_TRUE
)
2949 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2952 * If we're assembling a pool from a split, the config provided is
2953 * already trusted so there is nothing to do.
2955 if (type
== SPA_IMPORT_ASSEMBLE
)
2958 healthy_tvds
= spa_healthy_core_tvds(spa
);
2960 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
)
2962 spa_load_failed(spa
, "unable to retrieve MOS config");
2963 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2967 * If we are doing an open, pool owner wasn't verified yet, thus do
2968 * the verification here.
2970 if (spa
->spa_load_state
== SPA_LOAD_OPEN
) {
2971 error
= spa_verify_host(spa
, mos_config
);
2973 nvlist_free(mos_config
);
2978 nv
= fnvlist_lookup_nvlist(mos_config
, ZPOOL_CONFIG_VDEV_TREE
);
2980 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2983 * Build a new vdev tree from the trusted config
2985 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
2988 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2989 * obtained by scanning /dev/dsk, then it will have the right vdev
2990 * paths. We update the trusted MOS config with this information.
2991 * We first try to copy the paths with vdev_copy_path_strict, which
2992 * succeeds only when both configs have exactly the same vdev tree.
2993 * If that fails, we fall back to a more flexible method that has a
2994 * best effort policy.
2996 copy_error
= vdev_copy_path_strict(rvd
, mrvd
);
2997 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
2998 spa_load_note(spa
, "provided vdev tree:");
2999 vdev_dbgmsg_print_tree(rvd
, 2);
3000 spa_load_note(spa
, "MOS vdev tree:");
3001 vdev_dbgmsg_print_tree(mrvd
, 2);
3003 if (copy_error
!= 0) {
3004 spa_load_note(spa
, "vdev_copy_path_strict failed, falling "
3005 "back to vdev_copy_path_relaxed");
3006 vdev_copy_path_relaxed(rvd
, mrvd
);
3011 spa
->spa_root_vdev
= mrvd
;
3013 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3016 * We will use spa_config if we decide to reload the spa or if spa_load
3017 * fails and we rewind. We must thus regenerate the config using the
3018 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3019 * pass settings on how to load the pool and is not stored in the MOS.
3020 * We copy it over to our new, trusted config.
3022 mos_config_txg
= fnvlist_lookup_uint64(mos_config
,
3023 ZPOOL_CONFIG_POOL_TXG
);
3024 nvlist_free(mos_config
);
3025 mos_config
= spa_config_generate(spa
, NULL
, mos_config_txg
, B_FALSE
);
3026 if (nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_LOAD_POLICY
,
3028 fnvlist_add_nvlist(mos_config
, ZPOOL_LOAD_POLICY
, policy
);
3029 spa_config_set(spa
, mos_config
);
3030 spa
->spa_config_source
= SPA_CONFIG_SRC_MOS
;
3033 * Now that we got the config from the MOS, we should be more strict
3034 * in checking blkptrs and can make assumptions about the consistency
3035 * of the vdev tree. spa_trust_config must be set to true before opening
3036 * vdevs in order for them to be writeable.
3038 spa
->spa_trust_config
= B_TRUE
;
3041 * Open and validate the new vdev tree
3043 error
= spa_ld_open_vdevs(spa
);
3047 error
= spa_ld_validate_vdevs(spa
);
3051 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3052 spa_load_note(spa
, "final vdev tree:");
3053 vdev_dbgmsg_print_tree(rvd
, 2);
3056 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
&&
3057 !spa
->spa_extreme_rewind
&& zfs_max_missing_tvds
== 0) {
3059 * Sanity check to make sure that we are indeed loading the
3060 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3061 * in the config provided and they happened to be the only ones
3062 * to have the latest uberblock, we could involuntarily perform
3063 * an extreme rewind.
3065 healthy_tvds_mos
= spa_healthy_core_tvds(spa
);
3066 if (healthy_tvds_mos
- healthy_tvds
>=
3067 SPA_SYNC_MIN_VDEVS
) {
3068 spa_load_note(spa
, "config provided misses too many "
3069 "top-level vdevs compared to MOS (%lld vs %lld). ",
3070 (u_longlong_t
)healthy_tvds
,
3071 (u_longlong_t
)healthy_tvds_mos
);
3072 spa_load_note(spa
, "vdev tree:");
3073 vdev_dbgmsg_print_tree(rvd
, 2);
3075 spa_load_failed(spa
, "config was already "
3076 "provided from MOS. Aborting.");
3077 return (spa_vdev_err(rvd
,
3078 VDEV_AUX_CORRUPT_DATA
, EIO
));
3080 spa_load_note(spa
, "spa must be reloaded using MOS "
3082 return (SET_ERROR(EAGAIN
));
3086 error
= spa_check_for_missing_logs(spa
);
3088 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
3090 if (rvd
->vdev_guid_sum
!= spa
->spa_uberblock
.ub_guid_sum
) {
3091 spa_load_failed(spa
, "uberblock guid sum doesn't match MOS "
3092 "guid sum (%llu != %llu)",
3093 (u_longlong_t
)spa
->spa_uberblock
.ub_guid_sum
,
3094 (u_longlong_t
)rvd
->vdev_guid_sum
);
3095 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3103 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
3106 vdev_t
*rvd
= spa
->spa_root_vdev
;
3109 * Everything that we read before spa_remove_init() must be stored
3110 * on concreted vdevs. Therefore we do this as early as possible.
3112 error
= spa_remove_init(spa
);
3114 spa_load_failed(spa
, "spa_remove_init failed [error=%d]",
3116 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3120 * Retrieve information needed to condense indirect vdev mappings.
3122 error
= spa_condense_init(spa
);
3124 spa_load_failed(spa
, "spa_condense_init failed [error=%d]",
3126 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3133 spa_ld_check_features(spa_t
*spa
, boolean_t
*missing_feat_writep
)
3136 vdev_t
*rvd
= spa
->spa_root_vdev
;
3138 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
3139 boolean_t missing_feat_read
= B_FALSE
;
3140 nvlist_t
*unsup_feat
, *enabled_feat
;
3142 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
3143 &spa
->spa_feat_for_read_obj
, B_TRUE
) != 0) {
3144 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3147 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
3148 &spa
->spa_feat_for_write_obj
, B_TRUE
) != 0) {
3149 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3152 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
3153 &spa
->spa_feat_desc_obj
, B_TRUE
) != 0) {
3154 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3157 enabled_feat
= fnvlist_alloc();
3158 unsup_feat
= fnvlist_alloc();
3160 if (!spa_features_check(spa
, B_FALSE
,
3161 unsup_feat
, enabled_feat
))
3162 missing_feat_read
= B_TRUE
;
3164 if (spa_writeable(spa
) ||
3165 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) {
3166 if (!spa_features_check(spa
, B_TRUE
,
3167 unsup_feat
, enabled_feat
)) {
3168 *missing_feat_writep
= B_TRUE
;
3172 fnvlist_add_nvlist(spa
->spa_load_info
,
3173 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
3175 if (!nvlist_empty(unsup_feat
)) {
3176 fnvlist_add_nvlist(spa
->spa_load_info
,
3177 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
3180 fnvlist_free(enabled_feat
);
3181 fnvlist_free(unsup_feat
);
3183 if (!missing_feat_read
) {
3184 fnvlist_add_boolean(spa
->spa_load_info
,
3185 ZPOOL_CONFIG_CAN_RDONLY
);
3189 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3190 * twofold: to determine whether the pool is available for
3191 * import in read-write mode and (if it is not) whether the
3192 * pool is available for import in read-only mode. If the pool
3193 * is available for import in read-write mode, it is displayed
3194 * as available in userland; if it is not available for import
3195 * in read-only mode, it is displayed as unavailable in
3196 * userland. If the pool is available for import in read-only
3197 * mode but not read-write mode, it is displayed as unavailable
3198 * in userland with a special note that the pool is actually
3199 * available for open in read-only mode.
3201 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3202 * missing a feature for write, we must first determine whether
3203 * the pool can be opened read-only before returning to
3204 * userland in order to know whether to display the
3205 * abovementioned note.
3207 if (missing_feat_read
|| (*missing_feat_writep
&&
3208 spa_writeable(spa
))) {
3209 spa_load_failed(spa
, "pool uses unsupported features");
3210 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
3215 * Load refcounts for ZFS features from disk into an in-memory
3216 * cache during SPA initialization.
3218 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
3221 error
= feature_get_refcount_from_disk(spa
,
3222 &spa_feature_table
[i
], &refcount
);
3224 spa
->spa_feat_refcount_cache
[i
] = refcount
;
3225 } else if (error
== ENOTSUP
) {
3226 spa
->spa_feat_refcount_cache
[i
] =
3227 SPA_FEATURE_DISABLED
;
3229 spa_load_failed(spa
, "error getting refcount "
3230 "for feature %s [error=%d]",
3231 spa_feature_table
[i
].fi_guid
, error
);
3232 return (spa_vdev_err(rvd
,
3233 VDEV_AUX_CORRUPT_DATA
, EIO
));
3238 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
3239 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
3240 &spa
->spa_feat_enabled_txg_obj
, B_TRUE
) != 0)
3241 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3248 spa_ld_load_special_directories(spa_t
*spa
)
3251 vdev_t
*rvd
= spa
->spa_root_vdev
;
3253 spa
->spa_is_initializing
= B_TRUE
;
3254 error
= dsl_pool_open(spa
->spa_dsl_pool
);
3255 spa
->spa_is_initializing
= B_FALSE
;
3257 spa_load_failed(spa
, "dsl_pool_open failed [error=%d]", error
);
3258 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3265 spa_ld_get_props(spa_t
*spa
)
3269 vdev_t
*rvd
= spa
->spa_root_vdev
;
3271 /* Grab the checksum salt from the MOS. */
3272 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
3273 DMU_POOL_CHECKSUM_SALT
, 1,
3274 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
3275 spa
->spa_cksum_salt
.zcs_bytes
);
3276 if (error
== ENOENT
) {
3277 /* Generate a new salt for subsequent use */
3278 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
3279 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
3280 } else if (error
!= 0) {
3281 spa_load_failed(spa
, "unable to retrieve checksum salt from "
3282 "MOS [error=%d]", error
);
3283 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3286 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
, B_TRUE
) != 0)
3287 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3288 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
3290 spa_load_failed(spa
, "error opening deferred-frees bpobj "
3291 "[error=%d]", error
);
3292 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3296 * Load the bit that tells us to use the new accounting function
3297 * (raid-z deflation). If we have an older pool, this will not
3300 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
, B_FALSE
);
3301 if (error
!= 0 && error
!= ENOENT
)
3302 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3304 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
3305 &spa
->spa_creation_version
, B_FALSE
);
3306 if (error
!= 0 && error
!= ENOENT
)
3307 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3310 * Load the persistent error log. If we have an older pool, this will
3313 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
,
3315 if (error
!= 0 && error
!= ENOENT
)
3316 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3318 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
3319 &spa
->spa_errlog_scrub
, B_FALSE
);
3320 if (error
!= 0 && error
!= ENOENT
)
3321 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3324 * Load the history object. If we have an older pool, this
3325 * will not be present.
3327 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
, B_FALSE
);
3328 if (error
!= 0 && error
!= ENOENT
)
3329 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3332 * Load the per-vdev ZAP map. If we have an older pool, this will not
3333 * be present; in this case, defer its creation to a later time to
3334 * avoid dirtying the MOS this early / out of sync context. See
3335 * spa_sync_config_object.
3338 /* The sentinel is only available in the MOS config. */
3339 nvlist_t
*mos_config
;
3340 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0) {
3341 spa_load_failed(spa
, "unable to retrieve MOS config");
3342 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3345 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
3346 &spa
->spa_all_vdev_zaps
, B_FALSE
);
3348 if (error
== ENOENT
) {
3349 VERIFY(!nvlist_exists(mos_config
,
3350 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
3351 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
3352 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3353 } else if (error
!= 0) {
3354 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3355 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
3357 * An older version of ZFS overwrote the sentinel value, so
3358 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3359 * destruction to later; see spa_sync_config_object.
3361 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
3363 * We're assuming that no vdevs have had their ZAPs created
3364 * before this. Better be sure of it.
3366 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3368 nvlist_free(mos_config
);
3370 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3372 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
,
3374 if (error
&& error
!= ENOENT
)
3375 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3378 uint64_t autoreplace
;
3380 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3381 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3382 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3383 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3384 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3385 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
3386 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3387 &spa
->spa_dedup_ditto
);
3389 spa
->spa_autoreplace
= (autoreplace
!= 0);
3393 * If we are importing a pool with missing top-level vdevs,
3394 * we enforce that the pool doesn't panic or get suspended on
3395 * error since the likelihood of missing data is extremely high.
3397 if (spa
->spa_missing_tvds
> 0 &&
3398 spa
->spa_failmode
!= ZIO_FAILURE_MODE_CONTINUE
&&
3399 spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
3400 spa_load_note(spa
, "forcing failmode to 'continue' "
3401 "as some top level vdevs are missing");
3402 spa
->spa_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
3409 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
3412 vdev_t
*rvd
= spa
->spa_root_vdev
;
3415 * If we're assembling the pool from the split-off vdevs of
3416 * an existing pool, we don't want to attach the spares & cache
3421 * Load any hot spares for this pool.
3423 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
,
3425 if (error
!= 0 && error
!= ENOENT
)
3426 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3427 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3428 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3429 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3430 &spa
->spa_spares
.sav_config
) != 0) {
3431 spa_load_failed(spa
, "error loading spares nvlist");
3432 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3435 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3436 spa_load_spares(spa
);
3437 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3438 } else if (error
== 0) {
3439 spa
->spa_spares
.sav_sync
= B_TRUE
;
3443 * Load any level 2 ARC devices for this pool.
3445 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3446 &spa
->spa_l2cache
.sav_object
, B_FALSE
);
3447 if (error
!= 0 && error
!= ENOENT
)
3448 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3449 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3450 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3451 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3452 &spa
->spa_l2cache
.sav_config
) != 0) {
3453 spa_load_failed(spa
, "error loading l2cache nvlist");
3454 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3457 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3458 spa_load_l2cache(spa
);
3459 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3460 } else if (error
== 0) {
3461 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3468 spa_ld_load_vdev_metadata(spa_t
*spa
)
3471 vdev_t
*rvd
= spa
->spa_root_vdev
;
3474 * If the 'multihost' property is set, then never allow a pool to
3475 * be imported when the system hostid is zero. The exception to
3476 * this rule is zdb which is always allowed to access pools.
3478 if (spa_multihost(spa
) && spa_get_hostid() == 0 &&
3479 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
3480 fnvlist_add_uint64(spa
->spa_load_info
,
3481 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3482 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3486 * If the 'autoreplace' property is set, then post a resource notifying
3487 * the ZFS DE that it should not issue any faults for unopenable
3488 * devices. We also iterate over the vdevs, and post a sysevent for any
3489 * unopenable vdevs so that the normal autoreplace handler can take
3492 if (spa
->spa_autoreplace
&& spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
3493 spa_check_removed(spa
->spa_root_vdev
);
3495 * For the import case, this is done in spa_import(), because
3496 * at this point we're using the spare definitions from
3497 * the MOS config, not necessarily from the userland config.
3499 if (spa
->spa_load_state
!= SPA_LOAD_IMPORT
) {
3500 spa_aux_check_removed(&spa
->spa_spares
);
3501 spa_aux_check_removed(&spa
->spa_l2cache
);
3506 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3508 error
= vdev_load(rvd
);
3510 spa_load_failed(spa
, "vdev_load failed [error=%d]", error
);
3511 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3515 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3517 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3518 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3519 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3525 spa_ld_load_dedup_tables(spa_t
*spa
)
3528 vdev_t
*rvd
= spa
->spa_root_vdev
;
3530 error
= ddt_load(spa
);
3532 spa_load_failed(spa
, "ddt_load failed [error=%d]", error
);
3533 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3540 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
3542 vdev_t
*rvd
= spa
->spa_root_vdev
;
3544 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
)) {
3545 boolean_t missing
= spa_check_logs(spa
);
3547 if (spa
->spa_missing_tvds
!= 0) {
3548 spa_load_note(spa
, "spa_check_logs failed "
3549 "so dropping the logs");
3551 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3552 spa_load_failed(spa
, "spa_check_logs failed");
3553 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
,
3563 spa_ld_verify_pool_data(spa_t
*spa
)
3566 vdev_t
*rvd
= spa
->spa_root_vdev
;
3569 * We've successfully opened the pool, verify that we're ready
3570 * to start pushing transactions.
3572 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
3573 error
= spa_load_verify(spa
);
3575 spa_load_failed(spa
, "spa_load_verify failed "
3576 "[error=%d]", error
);
3577 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3586 spa_ld_claim_log_blocks(spa_t
*spa
)
3589 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3592 * Claim log blocks that haven't been committed yet.
3593 * This must all happen in a single txg.
3594 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3595 * invoked from zil_claim_log_block()'s i/o done callback.
3596 * Price of rollback is that we abandon the log.
3598 spa
->spa_claiming
= B_TRUE
;
3600 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3601 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3602 zil_claim
, tx
, DS_FIND_CHILDREN
);
3605 spa
->spa_claiming
= B_FALSE
;
3607 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3611 spa_ld_check_for_config_update(spa_t
*spa
, uint64_t config_cache_txg
,
3612 boolean_t reloading
)
3614 vdev_t
*rvd
= spa
->spa_root_vdev
;
3615 int need_update
= B_FALSE
;
3618 * If the config cache is stale, or we have uninitialized
3619 * metaslabs (see spa_vdev_add()), then update the config.
3621 * If this is a verbatim import, trust the current
3622 * in-core spa_config and update the disk labels.
3624 if (reloading
|| config_cache_txg
!= spa
->spa_config_txg
||
3625 spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3626 spa
->spa_load_state
== SPA_LOAD_RECOVER
||
3627 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3628 need_update
= B_TRUE
;
3630 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
3631 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3632 need_update
= B_TRUE
;
3635 * Update the config cache asychronously in case we're the
3636 * root pool, in which case the config cache isn't writable yet.
3639 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3643 spa_ld_prepare_for_reload(spa_t
*spa
)
3645 int mode
= spa
->spa_mode
;
3646 int async_suspended
= spa
->spa_async_suspended
;
3649 spa_deactivate(spa
);
3650 spa_activate(spa
, mode
);
3653 * We save the value of spa_async_suspended as it gets reset to 0 by
3654 * spa_unload(). We want to restore it back to the original value before
3655 * returning as we might be calling spa_async_resume() later.
3657 spa
->spa_async_suspended
= async_suspended
;
3661 * Load an existing storage pool, using the config provided. This config
3662 * describes which vdevs are part of the pool and is later validated against
3663 * partial configs present in each vdev's label and an entire copy of the
3664 * config stored in the MOS.
3667 spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
,
3668 boolean_t reloading
)
3671 boolean_t missing_feat_write
= B_FALSE
;
3673 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
3674 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
3677 * Never trust the config that is provided unless we are assembling
3678 * a pool following a split.
3679 * This means don't trust blkptrs and the vdev tree in general. This
3680 * also effectively puts the spa in read-only mode since
3681 * spa_writeable() checks for spa_trust_config to be true.
3682 * We will later load a trusted config from the MOS.
3684 if (type
!= SPA_IMPORT_ASSEMBLE
)
3685 spa
->spa_trust_config
= B_FALSE
;
3688 spa_load_note(spa
, "RELOADING");
3690 spa_load_note(spa
, "LOADING");
3693 * Parse the config provided to create a vdev tree.
3695 error
= spa_ld_parse_config(spa
, type
);
3700 * Now that we have the vdev tree, try to open each vdev. This involves
3701 * opening the underlying physical device, retrieving its geometry and
3702 * probing the vdev with a dummy I/O. The state of each vdev will be set
3703 * based on the success of those operations. After this we'll be ready
3704 * to read from the vdevs.
3706 error
= spa_ld_open_vdevs(spa
);
3711 * Read the label of each vdev and make sure that the GUIDs stored
3712 * there match the GUIDs in the config provided.
3713 * If we're assembling a new pool that's been split off from an
3714 * existing pool, the labels haven't yet been updated so we skip
3715 * validation for now.
3717 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3718 error
= spa_ld_validate_vdevs(spa
);
3724 * Read vdev labels to find the best uberblock (i.e. latest, unless
3725 * spa_load_max_txg is set) and store it in spa_uberblock. We get the
3726 * list of features required to read blkptrs in the MOS from the vdev
3727 * label with the best uberblock and verify that our version of zfs
3728 * supports them all.
3730 error
= spa_ld_select_uberblock(spa
, type
);
3735 * Pass that uberblock to the dsl_pool layer which will open the root
3736 * blkptr. This blkptr points to the latest version of the MOS and will
3737 * allow us to read its contents.
3739 error
= spa_ld_open_rootbp(spa
);
3744 * Retrieve the trusted config stored in the MOS and use it to create
3745 * a new, exact version of the vdev tree, then reopen all vdevs.
3747 error
= spa_ld_load_trusted_config(spa
, type
, reloading
);
3748 if (error
== EAGAIN
) {
3751 * Redo the loading process with the trusted config if it is
3752 * too different from the untrusted config.
3754 spa_ld_prepare_for_reload(spa
);
3755 return (spa_load_impl(spa
, type
, ereport
, B_TRUE
));
3756 } else if (error
!= 0) {
3761 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3762 * from the pool and their contents were re-mapped to other vdevs. Note
3763 * that everything that we read before this step must have been
3764 * rewritten on concrete vdevs after the last device removal was
3765 * initiated. Otherwise we could be reading from indirect vdevs before
3766 * we have loaded their mappings.
3768 error
= spa_ld_open_indirect_vdev_metadata(spa
);
3773 * Retrieve the full list of active features from the MOS and check if
3774 * they are all supported.
3776 error
= spa_ld_check_features(spa
, &missing_feat_write
);
3781 * Load several special directories from the MOS needed by the dsl_pool
3784 error
= spa_ld_load_special_directories(spa
);
3789 * Retrieve pool properties from the MOS.
3791 error
= spa_ld_get_props(spa
);
3796 * Retrieve the list of auxiliary devices - cache devices and spares -
3799 error
= spa_ld_open_aux_vdevs(spa
, type
);
3804 * Load the metadata for all vdevs. Also check if unopenable devices
3805 * should be autoreplaced.
3807 error
= spa_ld_load_vdev_metadata(spa
);
3811 error
= spa_ld_load_dedup_tables(spa
);
3816 * Verify the logs now to make sure we don't have any unexpected errors
3817 * when we claim log blocks later.
3819 error
= spa_ld_verify_logs(spa
, type
, ereport
);
3823 if (missing_feat_write
) {
3824 ASSERT(spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
);
3827 * At this point, we know that we can open the pool in
3828 * read-only mode but not read-write mode. We now have enough
3829 * information and can return to userland.
3831 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
3836 * Traverse the last txgs to make sure the pool was left off in a safe
3837 * state. When performing an extreme rewind, we verify the whole pool,
3838 * which can take a very long time.
3840 error
= spa_ld_verify_pool_data(spa
);
3845 * Calculate the deflated space for the pool. This must be done before
3846 * we write anything to the pool because we'd need to update the space
3847 * accounting using the deflated sizes.
3849 spa_update_dspace(spa
);
3852 * We have now retrieved all the information we needed to open the
3853 * pool. If we are importing the pool in read-write mode, a few
3854 * additional steps must be performed to finish the import.
3856 if (spa_writeable(spa
) && (spa
->spa_load_state
== SPA_LOAD_RECOVER
||
3857 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3858 uint64_t config_cache_txg
= spa
->spa_config_txg
;
3860 ASSERT(spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
);
3863 * Traverse the ZIL and claim all blocks.
3865 spa_ld_claim_log_blocks(spa
);
3868 * Kick-off the syncing thread.
3870 spa
->spa_sync_on
= B_TRUE
;
3871 txg_sync_start(spa
->spa_dsl_pool
);
3872 mmp_thread_start(spa
);
3875 * Wait for all claims to sync. We sync up to the highest
3876 * claimed log block birth time so that claimed log blocks
3877 * don't appear to be from the future. spa_claim_max_txg
3878 * will have been set for us by ZIL traversal operations
3881 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3884 * Check if we need to request an update of the config. On the
3885 * next sync, we would update the config stored in vdev labels
3886 * and the cachefile (by default /etc/zfs/zpool.cache).
3888 spa_ld_check_for_config_update(spa
, config_cache_txg
,
3892 * Check all DTLs to see if anything needs resilvering.
3894 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3895 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
))
3896 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3899 * Log the fact that we booted up (so that we can detect if
3900 * we rebooted in the middle of an operation).
3902 spa_history_log_version(spa
, "open", NULL
);
3905 * Delete any inconsistent datasets.
3907 (void) dmu_objset_find(spa_name(spa
),
3908 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3911 * Clean up any stale temporary dataset userrefs.
3913 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3915 spa_restart_removal(spa
);
3917 spa_spawn_aux_threads(spa
);
3920 spa_load_note(spa
, "LOADED");
3926 spa_load_retry(spa_t
*spa
, spa_load_state_t state
)
3928 int mode
= spa
->spa_mode
;
3931 spa_deactivate(spa
);
3933 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3935 spa_activate(spa
, mode
);
3936 spa_async_suspend(spa
);
3938 spa_load_note(spa
, "spa_load_retry: rewind, max txg: %llu",
3939 (u_longlong_t
)spa
->spa_load_max_txg
);
3941 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
));
3945 * If spa_load() fails this function will try loading prior txg's. If
3946 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3947 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3948 * function will not rewind the pool and will return the same error as
3952 spa_load_best(spa_t
*spa
, spa_load_state_t state
, uint64_t max_request
,
3955 nvlist_t
*loadinfo
= NULL
;
3956 nvlist_t
*config
= NULL
;
3957 int load_error
, rewind_error
;
3958 uint64_t safe_rewind_txg
;
3961 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3962 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3963 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3965 spa
->spa_load_max_txg
= max_request
;
3966 if (max_request
!= UINT64_MAX
)
3967 spa
->spa_extreme_rewind
= B_TRUE
;
3970 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
);
3971 if (load_error
== 0)
3974 if (spa
->spa_root_vdev
!= NULL
)
3975 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3977 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3978 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3980 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3981 nvlist_free(config
);
3982 return (load_error
);
3985 if (state
== SPA_LOAD_RECOVER
) {
3986 /* Price of rolling back is discarding txgs, including log */
3987 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3990 * If we aren't rolling back save the load info from our first
3991 * import attempt so that we can restore it after attempting
3994 loadinfo
= spa
->spa_load_info
;
3995 spa
->spa_load_info
= fnvlist_alloc();
3998 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3999 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
4000 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
4001 TXG_INITIAL
: safe_rewind_txg
;
4004 * Continue as long as we're finding errors, we're still within
4005 * the acceptable rewind range, and we're still finding uberblocks
4007 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
4008 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
4009 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
4010 spa
->spa_extreme_rewind
= B_TRUE
;
4011 rewind_error
= spa_load_retry(spa
, state
);
4014 spa
->spa_extreme_rewind
= B_FALSE
;
4015 spa
->spa_load_max_txg
= UINT64_MAX
;
4017 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
4018 spa_config_set(spa
, config
);
4020 nvlist_free(config
);
4022 if (state
== SPA_LOAD_RECOVER
) {
4023 ASSERT3P(loadinfo
, ==, NULL
);
4024 return (rewind_error
);
4026 /* Store the rewind info as part of the initial load info */
4027 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
4028 spa
->spa_load_info
);
4030 /* Restore the initial load info */
4031 fnvlist_free(spa
->spa_load_info
);
4032 spa
->spa_load_info
= loadinfo
;
4034 return (load_error
);
4041 * The import case is identical to an open except that the configuration is sent
4042 * down from userland, instead of grabbed from the configuration cache. For the
4043 * case of an open, the pool configuration will exist in the
4044 * POOL_STATE_UNINITIALIZED state.
4046 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4047 * the same time open the pool, without having to keep around the spa_t in some
4051 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
4055 spa_load_state_t state
= SPA_LOAD_OPEN
;
4057 int locked
= B_FALSE
;
4058 int firstopen
= B_FALSE
;
4063 * As disgusting as this is, we need to support recursive calls to this
4064 * function because dsl_dir_open() is called during spa_load(), and ends
4065 * up calling spa_open() again. The real fix is to figure out how to
4066 * avoid dsl_dir_open() calling this in the first place.
4068 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
4069 mutex_enter(&spa_namespace_lock
);
4073 if ((spa
= spa_lookup(pool
)) == NULL
) {
4075 mutex_exit(&spa_namespace_lock
);
4076 return (SET_ERROR(ENOENT
));
4079 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
4080 zpool_load_policy_t policy
;
4084 zpool_get_load_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
4086 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
4087 state
= SPA_LOAD_RECOVER
;
4089 spa_activate(spa
, spa_mode_global
);
4091 if (state
!= SPA_LOAD_RECOVER
)
4092 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4093 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
4095 zfs_dbgmsg("spa_open_common: opening %s", pool
);
4096 error
= spa_load_best(spa
, state
, policy
.zlp_txg
,
4099 if (error
== EBADF
) {
4101 * If vdev_validate() returns failure (indicated by
4102 * EBADF), it indicates that one of the vdevs indicates
4103 * that the pool has been exported or destroyed. If
4104 * this is the case, the config cache is out of sync and
4105 * we should remove the pool from the namespace.
4108 spa_deactivate(spa
);
4109 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
4112 mutex_exit(&spa_namespace_lock
);
4113 return (SET_ERROR(ENOENT
));
4118 * We can't open the pool, but we still have useful
4119 * information: the state of each vdev after the
4120 * attempted vdev_open(). Return this to the user.
4122 if (config
!= NULL
&& spa
->spa_config
) {
4123 VERIFY(nvlist_dup(spa
->spa_config
, config
,
4125 VERIFY(nvlist_add_nvlist(*config
,
4126 ZPOOL_CONFIG_LOAD_INFO
,
4127 spa
->spa_load_info
) == 0);
4130 spa_deactivate(spa
);
4131 spa
->spa_last_open_failed
= error
;
4133 mutex_exit(&spa_namespace_lock
);
4139 spa_open_ref(spa
, tag
);
4142 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4145 * If we've recovered the pool, pass back any information we
4146 * gathered while doing the load.
4148 if (state
== SPA_LOAD_RECOVER
) {
4149 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
4150 spa
->spa_load_info
) == 0);
4154 spa
->spa_last_open_failed
= 0;
4155 spa
->spa_last_ubsync_txg
= 0;
4156 spa
->spa_load_txg
= 0;
4157 mutex_exit(&spa_namespace_lock
);
4161 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
4169 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
4172 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
4176 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
4178 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
4182 * Lookup the given spa_t, incrementing the inject count in the process,
4183 * preventing it from being exported or destroyed.
4186 spa_inject_addref(char *name
)
4190 mutex_enter(&spa_namespace_lock
);
4191 if ((spa
= spa_lookup(name
)) == NULL
) {
4192 mutex_exit(&spa_namespace_lock
);
4195 spa
->spa_inject_ref
++;
4196 mutex_exit(&spa_namespace_lock
);
4202 spa_inject_delref(spa_t
*spa
)
4204 mutex_enter(&spa_namespace_lock
);
4205 spa
->spa_inject_ref
--;
4206 mutex_exit(&spa_namespace_lock
);
4210 * Add spares device information to the nvlist.
4213 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
4223 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
4225 if (spa
->spa_spares
.sav_count
== 0)
4228 VERIFY(nvlist_lookup_nvlist(config
,
4229 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
4230 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
4231 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
4233 VERIFY(nvlist_add_nvlist_array(nvroot
,
4234 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4235 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
4236 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
4239 * Go through and find any spares which have since been
4240 * repurposed as an active spare. If this is the case, update
4241 * their status appropriately.
4243 for (i
= 0; i
< nspares
; i
++) {
4244 VERIFY(nvlist_lookup_uint64(spares
[i
],
4245 ZPOOL_CONFIG_GUID
, &guid
) == 0);
4246 if (spa_spare_exists(guid
, &pool
, NULL
) &&
4248 VERIFY(nvlist_lookup_uint64_array(
4249 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
4250 (uint64_t **)&vs
, &vsc
) == 0);
4251 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
4252 vs
->vs_aux
= VDEV_AUX_SPARED
;
4259 * Add l2cache device information to the nvlist, including vdev stats.
4262 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
4265 uint_t i
, j
, nl2cache
;
4272 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
4274 if (spa
->spa_l2cache
.sav_count
== 0)
4277 VERIFY(nvlist_lookup_nvlist(config
,
4278 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
4279 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
4280 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
4281 if (nl2cache
!= 0) {
4282 VERIFY(nvlist_add_nvlist_array(nvroot
,
4283 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4284 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
4285 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
4288 * Update level 2 cache device stats.
4291 for (i
= 0; i
< nl2cache
; i
++) {
4292 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
4293 ZPOOL_CONFIG_GUID
, &guid
) == 0);
4296 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
4298 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
4299 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
4305 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
4306 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
4308 vdev_get_stats(vd
, vs
);
4309 vdev_config_generate_stats(vd
, l2cache
[i
]);
4316 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
4321 if (spa
->spa_feat_for_read_obj
!= 0) {
4322 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
4323 spa
->spa_feat_for_read_obj
);
4324 zap_cursor_retrieve(&zc
, &za
) == 0;
4325 zap_cursor_advance(&zc
)) {
4326 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
4327 za
.za_num_integers
== 1);
4328 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
4329 za
.za_first_integer
));
4331 zap_cursor_fini(&zc
);
4334 if (spa
->spa_feat_for_write_obj
!= 0) {
4335 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
4336 spa
->spa_feat_for_write_obj
);
4337 zap_cursor_retrieve(&zc
, &za
) == 0;
4338 zap_cursor_advance(&zc
)) {
4339 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
4340 za
.za_num_integers
== 1);
4341 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
4342 za
.za_first_integer
));
4344 zap_cursor_fini(&zc
);
4349 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
4353 for (i
= 0; i
< SPA_FEATURES
; i
++) {
4354 zfeature_info_t feature
= spa_feature_table
[i
];
4357 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
4360 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
4365 * Store a list of pool features and their reference counts in the
4368 * The first time this is called on a spa, allocate a new nvlist, fetch
4369 * the pool features and reference counts from disk, then save the list
4370 * in the spa. In subsequent calls on the same spa use the saved nvlist
4371 * and refresh its values from the cached reference counts. This
4372 * ensures we don't block here on I/O on a suspended pool so 'zpool
4373 * clear' can resume the pool.
4376 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
4380 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
4382 mutex_enter(&spa
->spa_feat_stats_lock
);
4383 features
= spa
->spa_feat_stats
;
4385 if (features
!= NULL
) {
4386 spa_feature_stats_from_cache(spa
, features
);
4388 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
4389 spa
->spa_feat_stats
= features
;
4390 spa_feature_stats_from_disk(spa
, features
);
4393 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
4396 mutex_exit(&spa
->spa_feat_stats_lock
);
4400 spa_get_stats(const char *name
, nvlist_t
**config
,
4401 char *altroot
, size_t buflen
)
4407 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
4411 * This still leaves a window of inconsistency where the spares
4412 * or l2cache devices could change and the config would be
4413 * self-inconsistent.
4415 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4417 if (*config
!= NULL
) {
4418 uint64_t loadtimes
[2];
4420 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
4421 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
4422 VERIFY(nvlist_add_uint64_array(*config
,
4423 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
4425 VERIFY(nvlist_add_uint64(*config
,
4426 ZPOOL_CONFIG_ERRCOUNT
,
4427 spa_get_errlog_size(spa
)) == 0);
4429 if (spa_suspended(spa
)) {
4430 VERIFY(nvlist_add_uint64(*config
,
4431 ZPOOL_CONFIG_SUSPENDED
,
4432 spa
->spa_failmode
) == 0);
4433 VERIFY(nvlist_add_uint64(*config
,
4434 ZPOOL_CONFIG_SUSPENDED_REASON
,
4435 spa
->spa_suspended
) == 0);
4438 spa_add_spares(spa
, *config
);
4439 spa_add_l2cache(spa
, *config
);
4440 spa_add_feature_stats(spa
, *config
);
4445 * We want to get the alternate root even for faulted pools, so we cheat
4446 * and call spa_lookup() directly.
4450 mutex_enter(&spa_namespace_lock
);
4451 spa
= spa_lookup(name
);
4453 spa_altroot(spa
, altroot
, buflen
);
4457 mutex_exit(&spa_namespace_lock
);
4459 spa_altroot(spa
, altroot
, buflen
);
4464 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4465 spa_close(spa
, FTAG
);
4472 * Validate that the auxiliary device array is well formed. We must have an
4473 * array of nvlists, each which describes a valid leaf vdev. If this is an
4474 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4475 * specified, as long as they are well-formed.
4478 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
4479 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
4480 vdev_labeltype_t label
)
4487 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4490 * It's acceptable to have no devs specified.
4492 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
4496 return (SET_ERROR(EINVAL
));
4499 * Make sure the pool is formatted with a version that supports this
4502 if (spa_version(spa
) < version
)
4503 return (SET_ERROR(ENOTSUP
));
4506 * Set the pending device list so we correctly handle device in-use
4509 sav
->sav_pending
= dev
;
4510 sav
->sav_npending
= ndev
;
4512 for (i
= 0; i
< ndev
; i
++) {
4513 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
4517 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4519 error
= SET_ERROR(EINVAL
);
4525 if ((error
= vdev_open(vd
)) == 0 &&
4526 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
4527 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
4528 vd
->vdev_guid
) == 0);
4534 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
4541 sav
->sav_pending
= NULL
;
4542 sav
->sav_npending
= 0;
4547 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
4551 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4553 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
4554 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
4555 VDEV_LABEL_SPARE
)) != 0) {
4559 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
4560 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
4561 VDEV_LABEL_L2CACHE
));
4565 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
4570 if (sav
->sav_config
!= NULL
) {
4576 * Generate new dev list by concatenating with the
4579 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
4580 &olddevs
, &oldndevs
) == 0);
4582 newdevs
= kmem_alloc(sizeof (void *) *
4583 (ndevs
+ oldndevs
), KM_SLEEP
);
4584 for (i
= 0; i
< oldndevs
; i
++)
4585 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
4587 for (i
= 0; i
< ndevs
; i
++)
4588 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
4591 VERIFY(nvlist_remove(sav
->sav_config
, config
,
4592 DATA_TYPE_NVLIST_ARRAY
) == 0);
4594 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
4595 config
, newdevs
, ndevs
+ oldndevs
) == 0);
4596 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
4597 nvlist_free(newdevs
[i
]);
4598 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
4601 * Generate a new dev list.
4603 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
4605 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
4611 * Stop and drop level 2 ARC devices
4614 spa_l2cache_drop(spa_t
*spa
)
4618 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4620 for (i
= 0; i
< sav
->sav_count
; i
++) {
4623 vd
= sav
->sav_vdevs
[i
];
4626 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
4627 pool
!= 0ULL && l2arc_vdev_present(vd
))
4628 l2arc_remove_vdev(vd
);
4633 * Verify encryption parameters for spa creation. If we are encrypting, we must
4634 * have the encryption feature flag enabled.
4637 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
4638 boolean_t has_encryption
)
4640 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
4641 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
4643 return (SET_ERROR(ENOTSUP
));
4645 return (dmu_objset_create_crypt_check(NULL
, dcp
));
4652 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
4653 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
4656 char *altroot
= NULL
;
4661 uint64_t txg
= TXG_INITIAL
;
4662 nvlist_t
**spares
, **l2cache
;
4663 uint_t nspares
, nl2cache
;
4664 uint64_t version
, obj
, root_dsobj
= 0;
4665 boolean_t has_features
;
4666 boolean_t has_encryption
;
4672 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
4673 poolname
= (char *)pool
;
4676 * If this pool already exists, return failure.
4678 mutex_enter(&spa_namespace_lock
);
4679 if (spa_lookup(poolname
) != NULL
) {
4680 mutex_exit(&spa_namespace_lock
);
4681 return (SET_ERROR(EEXIST
));
4685 * Allocate a new spa_t structure.
4687 nvl
= fnvlist_alloc();
4688 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
4689 (void) nvlist_lookup_string(props
,
4690 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4691 spa
= spa_add(poolname
, nvl
, altroot
);
4693 spa_activate(spa
, spa_mode_global
);
4695 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4696 spa_deactivate(spa
);
4698 mutex_exit(&spa_namespace_lock
);
4703 * Temporary pool names should never be written to disk.
4705 if (poolname
!= pool
)
4706 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
4708 has_features
= B_FALSE
;
4709 has_encryption
= B_FALSE
;
4710 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
4711 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4712 if (zpool_prop_feature(nvpair_name(elem
))) {
4713 has_features
= B_TRUE
;
4715 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
4716 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
4717 if (feat
== SPA_FEATURE_ENCRYPTION
)
4718 has_encryption
= B_TRUE
;
4722 /* verify encryption params, if they were provided */
4724 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
4726 spa_deactivate(spa
);
4728 mutex_exit(&spa_namespace_lock
);
4733 if (has_features
|| nvlist_lookup_uint64(props
,
4734 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4735 version
= SPA_VERSION
;
4737 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4739 spa
->spa_first_txg
= txg
;
4740 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4741 spa
->spa_uberblock
.ub_version
= version
;
4742 spa
->spa_ubsync
= spa
->spa_uberblock
;
4743 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4744 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
4745 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
4746 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
4749 * Create "The Godfather" zio to hold all async IOs
4751 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4753 for (int i
= 0; i
< max_ncpus
; i
++) {
4754 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4755 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4756 ZIO_FLAG_GODFATHER
);
4760 * Create the root vdev.
4762 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4764 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4766 ASSERT(error
!= 0 || rvd
!= NULL
);
4767 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4769 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4770 error
= SET_ERROR(EINVAL
);
4773 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4774 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4775 VDEV_ALLOC_ADD
)) == 0) {
4776 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
4777 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4778 vdev_expand(rvd
->vdev_child
[c
], txg
);
4782 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4786 spa_deactivate(spa
);
4788 mutex_exit(&spa_namespace_lock
);
4793 * Get the list of spares, if specified.
4795 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4796 &spares
, &nspares
) == 0) {
4797 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4799 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4800 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4801 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4802 spa_load_spares(spa
);
4803 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4804 spa
->spa_spares
.sav_sync
= B_TRUE
;
4808 * Get the list of level 2 cache devices, if specified.
4810 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4811 &l2cache
, &nl2cache
) == 0) {
4812 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4813 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4814 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4815 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4816 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4817 spa_load_l2cache(spa
);
4818 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4819 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4822 spa
->spa_is_initializing
= B_TRUE
;
4823 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
4824 spa
->spa_is_initializing
= B_FALSE
;
4827 * Create DDTs (dedup tables).
4831 spa_update_dspace(spa
);
4833 tx
= dmu_tx_create_assigned(dp
, txg
);
4836 * Create the pool's history object.
4838 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
4839 spa_history_create_obj(spa
, tx
);
4841 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4842 spa_history_log_version(spa
, "create", tx
);
4845 * Create the pool config object.
4847 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4848 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4849 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4851 if (zap_add(spa
->spa_meta_objset
,
4852 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4853 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4854 cmn_err(CE_PANIC
, "failed to add pool config");
4857 if (zap_add(spa
->spa_meta_objset
,
4858 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4859 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4860 cmn_err(CE_PANIC
, "failed to add pool version");
4863 /* Newly created pools with the right version are always deflated. */
4864 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4865 spa
->spa_deflate
= TRUE
;
4866 if (zap_add(spa
->spa_meta_objset
,
4867 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4868 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4869 cmn_err(CE_PANIC
, "failed to add deflate");
4874 * Create the deferred-free bpobj. Turn off compression
4875 * because sync-to-convergence takes longer if the blocksize
4878 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4879 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4880 ZIO_COMPRESS_OFF
, tx
);
4881 if (zap_add(spa
->spa_meta_objset
,
4882 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4883 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4884 cmn_err(CE_PANIC
, "failed to add bpobj");
4886 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4887 spa
->spa_meta_objset
, obj
));
4890 * Generate some random noise for salted checksums to operate on.
4892 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4893 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4896 * Set pool properties.
4898 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4899 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4900 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4901 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4902 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
4904 if (props
!= NULL
) {
4905 spa_configfile_set(spa
, props
, B_FALSE
);
4906 spa_sync_props(props
, tx
);
4912 * If the root dataset is encrypted we will need to create key mappings
4913 * for the zio layer before we start to write any data to disk and hold
4914 * them until after the first txg has been synced. Waiting for the first
4915 * transaction to complete also ensures that our bean counters are
4916 * appropriately updated.
4918 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0) {
4919 root_dsobj
= dsl_dir_phys(dp
->dp_root_dir
)->dd_head_dataset_obj
;
4920 VERIFY0(spa_keystore_create_mapping_impl(spa
, root_dsobj
,
4921 dp
->dp_root_dir
, FTAG
));
4924 spa
->spa_sync_on
= B_TRUE
;
4926 mmp_thread_start(spa
);
4927 txg_wait_synced(dp
, txg
);
4929 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0)
4930 VERIFY0(spa_keystore_remove_mapping(spa
, root_dsobj
, FTAG
));
4932 spa_spawn_aux_threads(spa
);
4934 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
4937 * Don't count references from objsets that are already closed
4938 * and are making their way through the eviction process.
4940 spa_evicting_os_wait(spa
);
4941 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4942 spa
->spa_load_state
= SPA_LOAD_NONE
;
4944 mutex_exit(&spa_namespace_lock
);
4950 * Import a non-root pool into the system.
4953 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4956 char *altroot
= NULL
;
4957 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4958 zpool_load_policy_t policy
;
4959 uint64_t mode
= spa_mode_global
;
4960 uint64_t readonly
= B_FALSE
;
4963 nvlist_t
**spares
, **l2cache
;
4964 uint_t nspares
, nl2cache
;
4967 * If a pool with this name exists, return failure.
4969 mutex_enter(&spa_namespace_lock
);
4970 if (spa_lookup(pool
) != NULL
) {
4971 mutex_exit(&spa_namespace_lock
);
4972 return (SET_ERROR(EEXIST
));
4976 * Create and initialize the spa structure.
4978 (void) nvlist_lookup_string(props
,
4979 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4980 (void) nvlist_lookup_uint64(props
,
4981 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4984 spa
= spa_add(pool
, config
, altroot
);
4985 spa
->spa_import_flags
= flags
;
4988 * Verbatim import - Take a pool and insert it into the namespace
4989 * as if it had been loaded at boot.
4991 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4993 spa_configfile_set(spa
, props
, B_FALSE
);
4995 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
4996 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4997 zfs_dbgmsg("spa_import: verbatim import of %s", pool
);
4998 mutex_exit(&spa_namespace_lock
);
5002 spa_activate(spa
, mode
);
5005 * Don't start async tasks until we know everything is healthy.
5007 spa_async_suspend(spa
);
5009 zpool_get_load_policy(config
, &policy
);
5010 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
5011 state
= SPA_LOAD_RECOVER
;
5013 spa
->spa_config_source
= SPA_CONFIG_SRC_TRYIMPORT
;
5015 if (state
!= SPA_LOAD_RECOVER
) {
5016 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
5017 zfs_dbgmsg("spa_import: importing %s", pool
);
5019 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5020 "(RECOVERY MODE)", pool
, (longlong_t
)policy
.zlp_txg
);
5022 error
= spa_load_best(spa
, state
, policy
.zlp_txg
, policy
.zlp_rewind
);
5025 * Propagate anything learned while loading the pool and pass it
5026 * back to caller (i.e. rewind info, missing devices, etc).
5028 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
5029 spa
->spa_load_info
) == 0);
5031 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5033 * Toss any existing sparelist, as it doesn't have any validity
5034 * anymore, and conflicts with spa_has_spare().
5036 if (spa
->spa_spares
.sav_config
) {
5037 nvlist_free(spa
->spa_spares
.sav_config
);
5038 spa
->spa_spares
.sav_config
= NULL
;
5039 spa_load_spares(spa
);
5041 if (spa
->spa_l2cache
.sav_config
) {
5042 nvlist_free(spa
->spa_l2cache
.sav_config
);
5043 spa
->spa_l2cache
.sav_config
= NULL
;
5044 spa_load_l2cache(spa
);
5047 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
5049 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5052 spa_configfile_set(spa
, props
, B_FALSE
);
5054 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
5055 (error
= spa_prop_set(spa
, props
)))) {
5057 spa_deactivate(spa
);
5059 mutex_exit(&spa_namespace_lock
);
5063 spa_async_resume(spa
);
5066 * Override any spares and level 2 cache devices as specified by
5067 * the user, as these may have correct device names/devids, etc.
5069 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5070 &spares
, &nspares
) == 0) {
5071 if (spa
->spa_spares
.sav_config
)
5072 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
5073 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5075 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
5076 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5077 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
5078 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
5079 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5080 spa_load_spares(spa
);
5081 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5082 spa
->spa_spares
.sav_sync
= B_TRUE
;
5084 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5085 &l2cache
, &nl2cache
) == 0) {
5086 if (spa
->spa_l2cache
.sav_config
)
5087 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
5088 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5090 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
5091 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5092 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
5093 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
5094 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5095 spa_load_l2cache(spa
);
5096 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5097 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5101 * Check for any removed devices.
5103 if (spa
->spa_autoreplace
) {
5104 spa_aux_check_removed(&spa
->spa_spares
);
5105 spa_aux_check_removed(&spa
->spa_l2cache
);
5108 if (spa_writeable(spa
)) {
5110 * Update the config cache to include the newly-imported pool.
5112 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5116 * It's possible that the pool was expanded while it was exported.
5117 * We kick off an async task to handle this for us.
5119 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
5121 spa_history_log_version(spa
, "import", NULL
);
5123 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
5125 zvol_create_minors(spa
, pool
, B_TRUE
);
5127 mutex_exit(&spa_namespace_lock
);
5133 spa_tryimport(nvlist_t
*tryconfig
)
5135 nvlist_t
*config
= NULL
;
5136 char *poolname
, *cachefile
;
5140 zpool_load_policy_t policy
;
5142 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
5145 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
5149 * Create and initialize the spa structure.
5151 mutex_enter(&spa_namespace_lock
);
5152 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
5153 spa_activate(spa
, FREAD
);
5156 * Rewind pool if a max txg was provided.
5158 zpool_get_load_policy(spa
->spa_config
, &policy
);
5159 if (policy
.zlp_txg
!= UINT64_MAX
) {
5160 spa
->spa_load_max_txg
= policy
.zlp_txg
;
5161 spa
->spa_extreme_rewind
= B_TRUE
;
5162 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
5163 poolname
, (longlong_t
)policy
.zlp_txg
);
5165 zfs_dbgmsg("spa_tryimport: importing %s", poolname
);
5168 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_CACHEFILE
, &cachefile
)
5170 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile
);
5171 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
5173 spa
->spa_config_source
= SPA_CONFIG_SRC_SCAN
;
5176 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
);
5179 * If 'tryconfig' was at least parsable, return the current config.
5181 if (spa
->spa_root_vdev
!= NULL
) {
5182 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5183 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
5185 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5187 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
5188 spa
->spa_uberblock
.ub_timestamp
) == 0);
5189 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
5190 spa
->spa_load_info
) == 0);
5191 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
5192 spa
->spa_errata
) == 0);
5195 * If the bootfs property exists on this pool then we
5196 * copy it out so that external consumers can tell which
5197 * pools are bootable.
5199 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
5200 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
5203 * We have to play games with the name since the
5204 * pool was opened as TRYIMPORT_NAME.
5206 if (dsl_dsobj_to_dsname(spa_name(spa
),
5207 spa
->spa_bootfs
, tmpname
) == 0) {
5211 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
5213 cp
= strchr(tmpname
, '/');
5215 (void) strlcpy(dsname
, tmpname
,
5218 (void) snprintf(dsname
, MAXPATHLEN
,
5219 "%s/%s", poolname
, ++cp
);
5221 VERIFY(nvlist_add_string(config
,
5222 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
5223 kmem_free(dsname
, MAXPATHLEN
);
5225 kmem_free(tmpname
, MAXPATHLEN
);
5229 * Add the list of hot spares and level 2 cache devices.
5231 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5232 spa_add_spares(spa
, config
);
5233 spa_add_l2cache(spa
, config
);
5234 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5238 spa_deactivate(spa
);
5240 mutex_exit(&spa_namespace_lock
);
5246 * Pool export/destroy
5248 * The act of destroying or exporting a pool is very simple. We make sure there
5249 * is no more pending I/O and any references to the pool are gone. Then, we
5250 * update the pool state and sync all the labels to disk, removing the
5251 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5252 * we don't sync the labels or remove the configuration cache.
5255 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
5256 boolean_t force
, boolean_t hardforce
)
5263 if (!(spa_mode_global
& FWRITE
))
5264 return (SET_ERROR(EROFS
));
5266 mutex_enter(&spa_namespace_lock
);
5267 if ((spa
= spa_lookup(pool
)) == NULL
) {
5268 mutex_exit(&spa_namespace_lock
);
5269 return (SET_ERROR(ENOENT
));
5273 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5274 * reacquire the namespace lock, and see if we can export.
5276 spa_open_ref(spa
, FTAG
);
5277 mutex_exit(&spa_namespace_lock
);
5278 spa_async_suspend(spa
);
5279 if (spa
->spa_zvol_taskq
) {
5280 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
5281 taskq_wait(spa
->spa_zvol_taskq
);
5283 mutex_enter(&spa_namespace_lock
);
5284 spa_close(spa
, FTAG
);
5286 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
5289 * The pool will be in core if it's openable, in which case we can
5290 * modify its state. Objsets may be open only because they're dirty,
5291 * so we have to force it to sync before checking spa_refcnt.
5293 if (spa
->spa_sync_on
) {
5294 txg_wait_synced(spa
->spa_dsl_pool
, 0);
5295 spa_evicting_os_wait(spa
);
5299 * A pool cannot be exported or destroyed if there are active
5300 * references. If we are resetting a pool, allow references by
5301 * fault injection handlers.
5303 if (!spa_refcount_zero(spa
) ||
5304 (spa
->spa_inject_ref
!= 0 &&
5305 new_state
!= POOL_STATE_UNINITIALIZED
)) {
5306 spa_async_resume(spa
);
5307 mutex_exit(&spa_namespace_lock
);
5308 return (SET_ERROR(EBUSY
));
5311 if (spa
->spa_sync_on
) {
5313 * A pool cannot be exported if it has an active shared spare.
5314 * This is to prevent other pools stealing the active spare
5315 * from an exported pool. At user's own will, such pool can
5316 * be forcedly exported.
5318 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
5319 spa_has_active_shared_spare(spa
)) {
5320 spa_async_resume(spa
);
5321 mutex_exit(&spa_namespace_lock
);
5322 return (SET_ERROR(EXDEV
));
5326 * We want this to be reflected on every label,
5327 * so mark them all dirty. spa_unload() will do the
5328 * final sync that pushes these changes out.
5330 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
5331 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5332 spa
->spa_state
= new_state
;
5333 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
5335 vdev_config_dirty(spa
->spa_root_vdev
);
5336 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5341 if (new_state
== POOL_STATE_DESTROYED
)
5342 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
5343 else if (new_state
== POOL_STATE_EXPORTED
)
5344 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
5346 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
5348 spa_deactivate(spa
);
5351 if (oldconfig
&& spa
->spa_config
)
5352 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
5354 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
5356 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
5359 mutex_exit(&spa_namespace_lock
);
5365 * Destroy a storage pool.
5368 spa_destroy(char *pool
)
5370 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
5375 * Export a storage pool.
5378 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
5379 boolean_t hardforce
)
5381 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
5386 * Similar to spa_export(), this unloads the spa_t without actually removing it
5387 * from the namespace in any way.
5390 spa_reset(char *pool
)
5392 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
5397 * ==========================================================================
5398 * Device manipulation
5399 * ==========================================================================
5403 * Add a device to a storage pool.
5406 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
5410 vdev_t
*rvd
= spa
->spa_root_vdev
;
5412 nvlist_t
**spares
, **l2cache
;
5413 uint_t nspares
, nl2cache
;
5415 ASSERT(spa_writeable(spa
));
5417 txg
= spa_vdev_enter(spa
);
5419 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
5420 VDEV_ALLOC_ADD
)) != 0)
5421 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5423 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
5425 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
5429 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
5433 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
5434 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5436 if (vd
->vdev_children
!= 0 &&
5437 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
5438 return (spa_vdev_exit(spa
, vd
, txg
, error
));
5441 * We must validate the spares and l2cache devices after checking the
5442 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5444 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
5445 return (spa_vdev_exit(spa
, vd
, txg
, error
));
5448 * If we are in the middle of a device removal, we can only add
5449 * devices which match the existing devices in the pool.
5450 * If we are in the middle of a removal, or have some indirect
5451 * vdevs, we can not add raidz toplevels.
5453 if (spa
->spa_vdev_removal
!= NULL
||
5454 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
5455 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5456 tvd
= vd
->vdev_child
[c
];
5457 if (spa
->spa_vdev_removal
!= NULL
&&
5458 tvd
->vdev_ashift
!= spa
->spa_max_ashift
) {
5459 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5461 /* Fail if top level vdev is raidz */
5462 if (tvd
->vdev_ops
== &vdev_raidz_ops
) {
5463 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5466 * Need the top level mirror to be
5467 * a mirror of leaf vdevs only
5469 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
5470 for (uint64_t cid
= 0;
5471 cid
< tvd
->vdev_children
; cid
++) {
5472 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
5473 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
5474 return (spa_vdev_exit(spa
, vd
,
5482 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5485 * Set the vdev id to the first hole, if one exists.
5487 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
5488 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
5489 vdev_free(rvd
->vdev_child
[id
]);
5493 tvd
= vd
->vdev_child
[c
];
5494 vdev_remove_child(vd
, tvd
);
5496 vdev_add_child(rvd
, tvd
);
5497 vdev_config_dirty(tvd
);
5501 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
5502 ZPOOL_CONFIG_SPARES
);
5503 spa_load_spares(spa
);
5504 spa
->spa_spares
.sav_sync
= B_TRUE
;
5507 if (nl2cache
!= 0) {
5508 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
5509 ZPOOL_CONFIG_L2CACHE
);
5510 spa_load_l2cache(spa
);
5511 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5515 * We have to be careful when adding new vdevs to an existing pool.
5516 * If other threads start allocating from these vdevs before we
5517 * sync the config cache, and we lose power, then upon reboot we may
5518 * fail to open the pool because there are DVAs that the config cache
5519 * can't translate. Therefore, we first add the vdevs without
5520 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5521 * and then let spa_config_update() initialize the new metaslabs.
5523 * spa_load() checks for added-but-not-initialized vdevs, so that
5524 * if we lose power at any point in this sequence, the remaining
5525 * steps will be completed the next time we load the pool.
5527 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
5529 mutex_enter(&spa_namespace_lock
);
5530 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5531 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
5532 mutex_exit(&spa_namespace_lock
);
5538 * Attach a device to a mirror. The arguments are the path to any device
5539 * in the mirror, and the nvroot for the new device. If the path specifies
5540 * a device that is not mirrored, we automatically insert the mirror vdev.
5542 * If 'replacing' is specified, the new device is intended to replace the
5543 * existing device; in this case the two devices are made into their own
5544 * mirror using the 'replacing' vdev, which is functionally identical to
5545 * the mirror vdev (it actually reuses all the same ops) but has a few
5546 * extra rules: you can't attach to it after it's been created, and upon
5547 * completion of resilvering, the first disk (the one being replaced)
5548 * is automatically detached.
5551 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
5553 uint64_t txg
, dtl_max_txg
;
5554 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5555 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
5557 char *oldvdpath
, *newvdpath
;
5561 ASSERT(spa_writeable(spa
));
5563 txg
= spa_vdev_enter(spa
);
5565 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5567 if (spa
->spa_vdev_removal
!= NULL
)
5568 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5571 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5573 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
5574 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5576 pvd
= oldvd
->vdev_parent
;
5578 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
5579 VDEV_ALLOC_ATTACH
)) != 0)
5580 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5582 if (newrootvd
->vdev_children
!= 1)
5583 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
5585 newvd
= newrootvd
->vdev_child
[0];
5587 if (!newvd
->vdev_ops
->vdev_op_leaf
)
5588 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
5590 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
5591 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
5594 * Spares can't replace logs
5596 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
5597 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5601 * For attach, the only allowable parent is a mirror or the root
5604 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5605 pvd
->vdev_ops
!= &vdev_root_ops
)
5606 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5608 pvops
= &vdev_mirror_ops
;
5611 * Active hot spares can only be replaced by inactive hot
5614 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5615 oldvd
->vdev_isspare
&&
5616 !spa_has_spare(spa
, newvd
->vdev_guid
))
5617 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5620 * If the source is a hot spare, and the parent isn't already a
5621 * spare, then we want to create a new hot spare. Otherwise, we
5622 * want to create a replacing vdev. The user is not allowed to
5623 * attach to a spared vdev child unless the 'isspare' state is
5624 * the same (spare replaces spare, non-spare replaces
5627 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
5628 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
5629 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5630 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5631 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
5632 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5635 if (newvd
->vdev_isspare
)
5636 pvops
= &vdev_spare_ops
;
5638 pvops
= &vdev_replacing_ops
;
5642 * Make sure the new device is big enough.
5644 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
5645 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
5648 * The new device cannot have a higher alignment requirement
5649 * than the top-level vdev.
5651 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
5652 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
5655 * If this is an in-place replacement, update oldvd's path and devid
5656 * to make it distinguishable from newvd, and unopenable from now on.
5658 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
5659 spa_strfree(oldvd
->vdev_path
);
5660 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
5662 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
5663 newvd
->vdev_path
, "old");
5664 if (oldvd
->vdev_devid
!= NULL
) {
5665 spa_strfree(oldvd
->vdev_devid
);
5666 oldvd
->vdev_devid
= NULL
;
5670 /* mark the device being resilvered */
5671 newvd
->vdev_resilver_txg
= txg
;
5674 * If the parent is not a mirror, or if we're replacing, insert the new
5675 * mirror/replacing/spare vdev above oldvd.
5677 if (pvd
->vdev_ops
!= pvops
)
5678 pvd
= vdev_add_parent(oldvd
, pvops
);
5680 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
5681 ASSERT(pvd
->vdev_ops
== pvops
);
5682 ASSERT(oldvd
->vdev_parent
== pvd
);
5685 * Extract the new device from its root and add it to pvd.
5687 vdev_remove_child(newrootvd
, newvd
);
5688 newvd
->vdev_id
= pvd
->vdev_children
;
5689 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
5690 vdev_add_child(pvd
, newvd
);
5693 * Reevaluate the parent vdev state.
5695 vdev_propagate_state(pvd
);
5697 tvd
= newvd
->vdev_top
;
5698 ASSERT(pvd
->vdev_top
== tvd
);
5699 ASSERT(tvd
->vdev_parent
== rvd
);
5701 vdev_config_dirty(tvd
);
5704 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5705 * for any dmu_sync-ed blocks. It will propagate upward when
5706 * spa_vdev_exit() calls vdev_dtl_reassess().
5708 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
5710 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
5711 dtl_max_txg
- TXG_INITIAL
);
5713 if (newvd
->vdev_isspare
) {
5714 spa_spare_activate(newvd
);
5715 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
5718 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
5719 newvdpath
= spa_strdup(newvd
->vdev_path
);
5720 newvd_isspare
= newvd
->vdev_isspare
;
5723 * Mark newvd's DTL dirty in this txg.
5725 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
5728 * Schedule the resilver to restart in the future. We do this to
5729 * ensure that dmu_sync-ed blocks have been stitched into the
5730 * respective datasets.
5732 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
5734 if (spa
->spa_bootfs
)
5735 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
5737 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
5742 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
5744 spa_history_log_internal(spa
, "vdev attach", NULL
,
5745 "%s vdev=%s %s vdev=%s",
5746 replacing
&& newvd_isspare
? "spare in" :
5747 replacing
? "replace" : "attach", newvdpath
,
5748 replacing
? "for" : "to", oldvdpath
);
5750 spa_strfree(oldvdpath
);
5751 spa_strfree(newvdpath
);
5757 * Detach a device from a mirror or replacing vdev.
5759 * If 'replace_done' is specified, only detach if the parent
5760 * is a replacing vdev.
5763 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5767 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5768 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5769 boolean_t unspare
= B_FALSE
;
5770 uint64_t unspare_guid
= 0;
5773 ASSERT(spa_writeable(spa
));
5775 txg
= spa_vdev_enter(spa
);
5777 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5780 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5782 if (!vd
->vdev_ops
->vdev_op_leaf
)
5783 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5785 pvd
= vd
->vdev_parent
;
5788 * If the parent/child relationship is not as expected, don't do it.
5789 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5790 * vdev that's replacing B with C. The user's intent in replacing
5791 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5792 * the replace by detaching C, the expected behavior is to end up
5793 * M(A,B). But suppose that right after deciding to detach C,
5794 * the replacement of B completes. We would have M(A,C), and then
5795 * ask to detach C, which would leave us with just A -- not what
5796 * the user wanted. To prevent this, we make sure that the
5797 * parent/child relationship hasn't changed -- in this example,
5798 * that C's parent is still the replacing vdev R.
5800 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5801 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5804 * Only 'replacing' or 'spare' vdevs can be replaced.
5806 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5807 pvd
->vdev_ops
!= &vdev_spare_ops
)
5808 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5810 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5811 spa_version(spa
) >= SPA_VERSION_SPARES
);
5814 * Only mirror, replacing, and spare vdevs support detach.
5816 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5817 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5818 pvd
->vdev_ops
!= &vdev_spare_ops
)
5819 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5822 * If this device has the only valid copy of some data,
5823 * we cannot safely detach it.
5825 if (vdev_dtl_required(vd
))
5826 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5828 ASSERT(pvd
->vdev_children
>= 2);
5831 * If we are detaching the second disk from a replacing vdev, then
5832 * check to see if we changed the original vdev's path to have "/old"
5833 * at the end in spa_vdev_attach(). If so, undo that change now.
5835 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5836 vd
->vdev_path
!= NULL
) {
5837 size_t len
= strlen(vd
->vdev_path
);
5839 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
5840 cvd
= pvd
->vdev_child
[c
];
5842 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5845 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5846 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5847 spa_strfree(cvd
->vdev_path
);
5848 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5855 * If we are detaching the original disk from a spare, then it implies
5856 * that the spare should become a real disk, and be removed from the
5857 * active spare list for the pool.
5859 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5861 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5865 * Erase the disk labels so the disk can be used for other things.
5866 * This must be done after all other error cases are handled,
5867 * but before we disembowel vd (so we can still do I/O to it).
5868 * But if we can't do it, don't treat the error as fatal --
5869 * it may be that the unwritability of the disk is the reason
5870 * it's being detached!
5872 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5875 * Remove vd from its parent and compact the parent's children.
5877 vdev_remove_child(pvd
, vd
);
5878 vdev_compact_children(pvd
);
5881 * Remember one of the remaining children so we can get tvd below.
5883 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5886 * If we need to remove the remaining child from the list of hot spares,
5887 * do it now, marking the vdev as no longer a spare in the process.
5888 * We must do this before vdev_remove_parent(), because that can
5889 * change the GUID if it creates a new toplevel GUID. For a similar
5890 * reason, we must remove the spare now, in the same txg as the detach;
5891 * otherwise someone could attach a new sibling, change the GUID, and
5892 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5895 ASSERT(cvd
->vdev_isspare
);
5896 spa_spare_remove(cvd
);
5897 unspare_guid
= cvd
->vdev_guid
;
5898 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5899 cvd
->vdev_unspare
= B_TRUE
;
5903 * If the parent mirror/replacing vdev only has one child,
5904 * the parent is no longer needed. Remove it from the tree.
5906 if (pvd
->vdev_children
== 1) {
5907 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5908 cvd
->vdev_unspare
= B_FALSE
;
5909 vdev_remove_parent(cvd
);
5914 * We don't set tvd until now because the parent we just removed
5915 * may have been the previous top-level vdev.
5917 tvd
= cvd
->vdev_top
;
5918 ASSERT(tvd
->vdev_parent
== rvd
);
5921 * Reevaluate the parent vdev state.
5923 vdev_propagate_state(cvd
);
5926 * If the 'autoexpand' property is set on the pool then automatically
5927 * try to expand the size of the pool. For example if the device we
5928 * just detached was smaller than the others, it may be possible to
5929 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5930 * first so that we can obtain the updated sizes of the leaf vdevs.
5932 if (spa
->spa_autoexpand
) {
5934 vdev_expand(tvd
, txg
);
5937 vdev_config_dirty(tvd
);
5940 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5941 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5942 * But first make sure we're not on any *other* txg's DTL list, to
5943 * prevent vd from being accessed after it's freed.
5945 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
5946 for (int t
= 0; t
< TXG_SIZE
; t
++)
5947 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5948 vd
->vdev_detached
= B_TRUE
;
5949 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5951 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5953 /* hang on to the spa before we release the lock */
5954 spa_open_ref(spa
, FTAG
);
5956 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5958 spa_history_log_internal(spa
, "detach", NULL
,
5960 spa_strfree(vdpath
);
5963 * If this was the removal of the original device in a hot spare vdev,
5964 * then we want to go through and remove the device from the hot spare
5965 * list of every other pool.
5968 spa_t
*altspa
= NULL
;
5970 mutex_enter(&spa_namespace_lock
);
5971 while ((altspa
= spa_next(altspa
)) != NULL
) {
5972 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5976 spa_open_ref(altspa
, FTAG
);
5977 mutex_exit(&spa_namespace_lock
);
5978 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5979 mutex_enter(&spa_namespace_lock
);
5980 spa_close(altspa
, FTAG
);
5982 mutex_exit(&spa_namespace_lock
);
5984 /* search the rest of the vdevs for spares to remove */
5985 spa_vdev_resilver_done(spa
);
5988 /* all done with the spa; OK to release */
5989 mutex_enter(&spa_namespace_lock
);
5990 spa_close(spa
, FTAG
);
5991 mutex_exit(&spa_namespace_lock
);
5997 * Split a set of devices from their mirrors, and create a new pool from them.
6000 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
6001 nvlist_t
*props
, boolean_t exp
)
6004 uint64_t txg
, *glist
;
6006 uint_t c
, children
, lastlog
;
6007 nvlist_t
**child
, *nvl
, *tmp
;
6009 char *altroot
= NULL
;
6010 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
6011 boolean_t activate_slog
;
6013 ASSERT(spa_writeable(spa
));
6015 txg
= spa_vdev_enter(spa
);
6017 /* clear the log and flush everything up to now */
6018 activate_slog
= spa_passivate_log(spa
);
6019 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
6020 error
= spa_reset_logs(spa
);
6021 txg
= spa_vdev_config_enter(spa
);
6024 spa_activate_log(spa
);
6027 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6029 /* check new spa name before going any further */
6030 if (spa_lookup(newname
) != NULL
)
6031 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
6034 * scan through all the children to ensure they're all mirrors
6036 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
6037 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
6039 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
6041 /* first, check to ensure we've got the right child count */
6042 rvd
= spa
->spa_root_vdev
;
6044 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
6045 vdev_t
*vd
= rvd
->vdev_child
[c
];
6047 /* don't count the holes & logs as children */
6048 if (vd
->vdev_islog
|| !vdev_is_concrete(vd
)) {
6056 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
6057 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
6059 /* next, ensure no spare or cache devices are part of the split */
6060 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
6061 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
6062 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
6064 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
6065 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
6067 /* then, loop over each vdev and validate it */
6068 for (c
= 0; c
< children
; c
++) {
6069 uint64_t is_hole
= 0;
6071 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
6075 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
6076 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
6079 error
= SET_ERROR(EINVAL
);
6084 /* which disk is going to be split? */
6085 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
6087 error
= SET_ERROR(EINVAL
);
6091 /* look it up in the spa */
6092 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
6093 if (vml
[c
] == NULL
) {
6094 error
= SET_ERROR(ENODEV
);
6098 /* make sure there's nothing stopping the split */
6099 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
6100 vml
[c
]->vdev_islog
||
6101 !vdev_is_concrete(vml
[c
]) ||
6102 vml
[c
]->vdev_isspare
||
6103 vml
[c
]->vdev_isl2cache
||
6104 !vdev_writeable(vml
[c
]) ||
6105 vml
[c
]->vdev_children
!= 0 ||
6106 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
6107 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
6108 error
= SET_ERROR(EINVAL
);
6112 if (vdev_dtl_required(vml
[c
])) {
6113 error
= SET_ERROR(EBUSY
);
6117 /* we need certain info from the top level */
6118 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
6119 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
6120 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
6121 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
6122 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
6123 vml
[c
]->vdev_top
->vdev_asize
) == 0);
6124 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
6125 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
6127 /* transfer per-vdev ZAPs */
6128 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
6129 VERIFY0(nvlist_add_uint64(child
[c
],
6130 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
6132 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
6133 VERIFY0(nvlist_add_uint64(child
[c
],
6134 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
6135 vml
[c
]->vdev_parent
->vdev_top_zap
));
6139 kmem_free(vml
, children
* sizeof (vdev_t
*));
6140 kmem_free(glist
, children
* sizeof (uint64_t));
6141 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6144 /* stop writers from using the disks */
6145 for (c
= 0; c
< children
; c
++) {
6147 vml
[c
]->vdev_offline
= B_TRUE
;
6149 vdev_reopen(spa
->spa_root_vdev
);
6152 * Temporarily record the splitting vdevs in the spa config. This
6153 * will disappear once the config is regenerated.
6155 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6156 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
6157 glist
, children
) == 0);
6158 kmem_free(glist
, children
* sizeof (uint64_t));
6160 mutex_enter(&spa
->spa_props_lock
);
6161 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
6163 mutex_exit(&spa
->spa_props_lock
);
6164 spa
->spa_config_splitting
= nvl
;
6165 vdev_config_dirty(spa
->spa_root_vdev
);
6167 /* configure and create the new pool */
6168 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
6169 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
6170 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
6171 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6172 spa_version(spa
)) == 0);
6173 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
6174 spa
->spa_config_txg
) == 0);
6175 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
6176 spa_generate_guid(NULL
)) == 0);
6177 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
6178 (void) nvlist_lookup_string(props
,
6179 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
6181 /* add the new pool to the namespace */
6182 newspa
= spa_add(newname
, config
, altroot
);
6183 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
6184 newspa
->spa_config_txg
= spa
->spa_config_txg
;
6185 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
6187 /* release the spa config lock, retaining the namespace lock */
6188 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
6190 if (zio_injection_enabled
)
6191 zio_handle_panic_injection(spa
, FTAG
, 1);
6193 spa_activate(newspa
, spa_mode_global
);
6194 spa_async_suspend(newspa
);
6196 newspa
->spa_config_source
= SPA_CONFIG_SRC_SPLIT
;
6198 /* create the new pool from the disks of the original pool */
6199 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
);
6203 /* if that worked, generate a real config for the new pool */
6204 if (newspa
->spa_root_vdev
!= NULL
) {
6205 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
6206 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6207 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
6208 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
6209 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
6214 if (props
!= NULL
) {
6215 spa_configfile_set(newspa
, props
, B_FALSE
);
6216 error
= spa_prop_set(newspa
, props
);
6221 /* flush everything */
6222 txg
= spa_vdev_config_enter(newspa
);
6223 vdev_config_dirty(newspa
->spa_root_vdev
);
6224 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
6226 if (zio_injection_enabled
)
6227 zio_handle_panic_injection(spa
, FTAG
, 2);
6229 spa_async_resume(newspa
);
6231 /* finally, update the original pool's config */
6232 txg
= spa_vdev_config_enter(spa
);
6233 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
6234 error
= dmu_tx_assign(tx
, TXG_WAIT
);
6237 for (c
= 0; c
< children
; c
++) {
6238 if (vml
[c
] != NULL
) {
6241 spa_history_log_internal(spa
, "detach", tx
,
6242 "vdev=%s", vml
[c
]->vdev_path
);
6247 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
6248 vdev_config_dirty(spa
->spa_root_vdev
);
6249 spa
->spa_config_splitting
= NULL
;
6253 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
6255 if (zio_injection_enabled
)
6256 zio_handle_panic_injection(spa
, FTAG
, 3);
6258 /* split is complete; log a history record */
6259 spa_history_log_internal(newspa
, "split", NULL
,
6260 "from pool %s", spa_name(spa
));
6262 kmem_free(vml
, children
* sizeof (vdev_t
*));
6264 /* if we're not going to mount the filesystems in userland, export */
6266 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
6273 spa_deactivate(newspa
);
6276 txg
= spa_vdev_config_enter(spa
);
6278 /* re-online all offlined disks */
6279 for (c
= 0; c
< children
; c
++) {
6281 vml
[c
]->vdev_offline
= B_FALSE
;
6283 vdev_reopen(spa
->spa_root_vdev
);
6285 nvlist_free(spa
->spa_config_splitting
);
6286 spa
->spa_config_splitting
= NULL
;
6287 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
6289 kmem_free(vml
, children
* sizeof (vdev_t
*));
6294 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6295 * currently spared, so we can detach it.
6298 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
6300 vdev_t
*newvd
, *oldvd
;
6302 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6303 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
6309 * Check for a completed replacement. We always consider the first
6310 * vdev in the list to be the oldest vdev, and the last one to be
6311 * the newest (see spa_vdev_attach() for how that works). In
6312 * the case where the newest vdev is faulted, we will not automatically
6313 * remove it after a resilver completes. This is OK as it will require
6314 * user intervention to determine which disk the admin wishes to keep.
6316 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
6317 ASSERT(vd
->vdev_children
> 1);
6319 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
6320 oldvd
= vd
->vdev_child
[0];
6322 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
6323 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
6324 !vdev_dtl_required(oldvd
))
6329 * Check for a completed resilver with the 'unspare' flag set.
6331 if (vd
->vdev_ops
== &vdev_spare_ops
) {
6332 vdev_t
*first
= vd
->vdev_child
[0];
6333 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
6335 if (last
->vdev_unspare
) {
6338 } else if (first
->vdev_unspare
) {
6345 if (oldvd
!= NULL
&&
6346 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
6347 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
6348 !vdev_dtl_required(oldvd
))
6352 * If there are more than two spares attached to a disk,
6353 * and those spares are not required, then we want to
6354 * attempt to free them up now so that they can be used
6355 * by other pools. Once we're back down to a single
6356 * disk+spare, we stop removing them.
6358 if (vd
->vdev_children
> 2) {
6359 newvd
= vd
->vdev_child
[1];
6361 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
6362 vdev_dtl_empty(last
, DTL_MISSING
) &&
6363 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
6364 !vdev_dtl_required(newvd
))
6373 spa_vdev_resilver_done(spa_t
*spa
)
6375 vdev_t
*vd
, *pvd
, *ppvd
;
6376 uint64_t guid
, sguid
, pguid
, ppguid
;
6378 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6380 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
6381 pvd
= vd
->vdev_parent
;
6382 ppvd
= pvd
->vdev_parent
;
6383 guid
= vd
->vdev_guid
;
6384 pguid
= pvd
->vdev_guid
;
6385 ppguid
= ppvd
->vdev_guid
;
6388 * If we have just finished replacing a hot spared device, then
6389 * we need to detach the parent's first child (the original hot
6392 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
6393 ppvd
->vdev_children
== 2) {
6394 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
6395 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
6397 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
6399 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6400 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
6402 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
6404 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6407 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6411 * Update the stored path or FRU for this vdev.
6414 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
6418 boolean_t sync
= B_FALSE
;
6420 ASSERT(spa_writeable(spa
));
6422 spa_vdev_state_enter(spa
, SCL_ALL
);
6424 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
6425 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
6427 if (!vd
->vdev_ops
->vdev_op_leaf
)
6428 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
6431 if (strcmp(value
, vd
->vdev_path
) != 0) {
6432 spa_strfree(vd
->vdev_path
);
6433 vd
->vdev_path
= spa_strdup(value
);
6437 if (vd
->vdev_fru
== NULL
) {
6438 vd
->vdev_fru
= spa_strdup(value
);
6440 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
6441 spa_strfree(vd
->vdev_fru
);
6442 vd
->vdev_fru
= spa_strdup(value
);
6447 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
6451 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
6453 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
6457 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
6459 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
6463 * ==========================================================================
6465 * ==========================================================================
6468 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
6470 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6472 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6473 return (SET_ERROR(EBUSY
));
6475 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
6479 spa_scan_stop(spa_t
*spa
)
6481 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6482 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6483 return (SET_ERROR(EBUSY
));
6484 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
6488 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
6490 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6492 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
6493 return (SET_ERROR(ENOTSUP
));
6496 * If a resilver was requested, but there is no DTL on a
6497 * writeable leaf device, we have nothing to do.
6499 if (func
== POOL_SCAN_RESILVER
&&
6500 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
6501 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6505 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6509 * ==========================================================================
6510 * SPA async task processing
6511 * ==========================================================================
6515 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6517 if (vd
->vdev_remove_wanted
) {
6518 vd
->vdev_remove_wanted
= B_FALSE
;
6519 vd
->vdev_delayed_close
= B_FALSE
;
6520 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6523 * We want to clear the stats, but we don't want to do a full
6524 * vdev_clear() as that will cause us to throw away
6525 * degraded/faulted state as well as attempt to reopen the
6526 * device, all of which is a waste.
6528 vd
->vdev_stat
.vs_read_errors
= 0;
6529 vd
->vdev_stat
.vs_write_errors
= 0;
6530 vd
->vdev_stat
.vs_checksum_errors
= 0;
6532 vdev_state_dirty(vd
->vdev_top
);
6535 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6536 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6540 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6542 if (vd
->vdev_probe_wanted
) {
6543 vd
->vdev_probe_wanted
= B_FALSE
;
6544 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6547 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6548 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6552 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6554 if (!spa
->spa_autoexpand
)
6557 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6558 vdev_t
*cvd
= vd
->vdev_child
[c
];
6559 spa_async_autoexpand(spa
, cvd
);
6562 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6565 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
6569 spa_async_thread(void *arg
)
6571 spa_t
*spa
= (spa_t
*)arg
;
6574 ASSERT(spa
->spa_sync_on
);
6576 mutex_enter(&spa
->spa_async_lock
);
6577 tasks
= spa
->spa_async_tasks
;
6578 spa
->spa_async_tasks
= 0;
6579 mutex_exit(&spa
->spa_async_lock
);
6582 * See if the config needs to be updated.
6584 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6585 uint64_t old_space
, new_space
;
6587 mutex_enter(&spa_namespace_lock
);
6588 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6589 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6590 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6591 mutex_exit(&spa_namespace_lock
);
6594 * If the pool grew as a result of the config update,
6595 * then log an internal history event.
6597 if (new_space
!= old_space
) {
6598 spa_history_log_internal(spa
, "vdev online", NULL
,
6599 "pool '%s' size: %llu(+%llu)",
6600 spa_name(spa
), new_space
, new_space
- old_space
);
6605 * See if any devices need to be marked REMOVED.
6607 if (tasks
& SPA_ASYNC_REMOVE
) {
6608 spa_vdev_state_enter(spa
, SCL_NONE
);
6609 spa_async_remove(spa
, spa
->spa_root_vdev
);
6610 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6611 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6612 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6613 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6614 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6617 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6618 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6619 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6620 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6624 * See if any devices need to be probed.
6626 if (tasks
& SPA_ASYNC_PROBE
) {
6627 spa_vdev_state_enter(spa
, SCL_NONE
);
6628 spa_async_probe(spa
, spa
->spa_root_vdev
);
6629 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6633 * If any devices are done replacing, detach them.
6635 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6636 spa_vdev_resilver_done(spa
);
6639 * Kick off a resilver.
6641 if (tasks
& SPA_ASYNC_RESILVER
)
6642 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6645 * Let the world know that we're done.
6647 mutex_enter(&spa
->spa_async_lock
);
6648 spa
->spa_async_thread
= NULL
;
6649 cv_broadcast(&spa
->spa_async_cv
);
6650 mutex_exit(&spa
->spa_async_lock
);
6655 spa_async_suspend(spa_t
*spa
)
6657 mutex_enter(&spa
->spa_async_lock
);
6658 spa
->spa_async_suspended
++;
6659 while (spa
->spa_async_thread
!= NULL
)
6660 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6661 mutex_exit(&spa
->spa_async_lock
);
6663 spa_vdev_remove_suspend(spa
);
6665 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
6666 if (condense_thread
!= NULL
&& zthr_isrunning(condense_thread
))
6667 VERIFY0(zthr_cancel(condense_thread
));
6671 spa_async_resume(spa_t
*spa
)
6673 mutex_enter(&spa
->spa_async_lock
);
6674 ASSERT(spa
->spa_async_suspended
!= 0);
6675 spa
->spa_async_suspended
--;
6676 mutex_exit(&spa
->spa_async_lock
);
6677 spa_restart_removal(spa
);
6679 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
6680 if (condense_thread
!= NULL
&& !zthr_isrunning(condense_thread
))
6681 zthr_resume(condense_thread
);
6685 spa_async_tasks_pending(spa_t
*spa
)
6687 uint_t non_config_tasks
;
6689 boolean_t config_task_suspended
;
6691 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6692 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6693 if (spa
->spa_ccw_fail_time
== 0) {
6694 config_task_suspended
= B_FALSE
;
6696 config_task_suspended
=
6697 (gethrtime() - spa
->spa_ccw_fail_time
) <
6698 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
6701 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6705 spa_async_dispatch(spa_t
*spa
)
6707 mutex_enter(&spa
->spa_async_lock
);
6708 if (spa_async_tasks_pending(spa
) &&
6709 !spa
->spa_async_suspended
&&
6710 spa
->spa_async_thread
== NULL
&&
6712 spa
->spa_async_thread
= thread_create(NULL
, 0,
6713 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6714 mutex_exit(&spa
->spa_async_lock
);
6718 spa_async_request(spa_t
*spa
, int task
)
6720 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6721 mutex_enter(&spa
->spa_async_lock
);
6722 spa
->spa_async_tasks
|= task
;
6723 mutex_exit(&spa
->spa_async_lock
);
6727 * ==========================================================================
6728 * SPA syncing routines
6729 * ==========================================================================
6733 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6736 bpobj_enqueue(bpo
, bp
, tx
);
6741 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6745 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6751 * Note: this simple function is not inlined to make it easier to dtrace the
6752 * amount of time spent syncing frees.
6755 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6757 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6758 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6759 VERIFY(zio_wait(zio
) == 0);
6763 * Note: this simple function is not inlined to make it easier to dtrace the
6764 * amount of time spent syncing deferred frees.
6767 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6769 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6770 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6771 spa_free_sync_cb
, zio
, tx
), ==, 0);
6772 VERIFY0(zio_wait(zio
));
6776 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6778 char *packed
= NULL
;
6783 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6786 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6787 * information. This avoids the dmu_buf_will_dirty() path and
6788 * saves us a pre-read to get data we don't actually care about.
6790 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6791 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6793 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6795 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6797 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6799 vmem_free(packed
, bufsize
);
6801 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6802 dmu_buf_will_dirty(db
, tx
);
6803 *(uint64_t *)db
->db_data
= nvsize
;
6804 dmu_buf_rele(db
, FTAG
);
6808 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6809 const char *config
, const char *entry
)
6819 * Update the MOS nvlist describing the list of available devices.
6820 * spa_validate_aux() will have already made sure this nvlist is
6821 * valid and the vdevs are labeled appropriately.
6823 if (sav
->sav_object
== 0) {
6824 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6825 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6826 sizeof (uint64_t), tx
);
6827 VERIFY(zap_update(spa
->spa_meta_objset
,
6828 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6829 &sav
->sav_object
, tx
) == 0);
6832 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6833 if (sav
->sav_count
== 0) {
6834 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6836 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6837 for (i
= 0; i
< sav
->sav_count
; i
++)
6838 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6839 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6840 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6841 sav
->sav_count
) == 0);
6842 for (i
= 0; i
< sav
->sav_count
; i
++)
6843 nvlist_free(list
[i
]);
6844 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6847 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6848 nvlist_free(nvroot
);
6850 sav
->sav_sync
= B_FALSE
;
6854 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6855 * The all-vdev ZAP must be empty.
6858 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6860 spa_t
*spa
= vd
->vdev_spa
;
6862 if (vd
->vdev_top_zap
!= 0) {
6863 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6864 vd
->vdev_top_zap
, tx
));
6866 if (vd
->vdev_leaf_zap
!= 0) {
6867 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6868 vd
->vdev_leaf_zap
, tx
));
6870 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
6871 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6876 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6881 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6882 * its config may not be dirty but we still need to build per-vdev ZAPs.
6883 * Similarly, if the pool is being assembled (e.g. after a split), we
6884 * need to rebuild the AVZ although the config may not be dirty.
6886 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6887 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6890 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6892 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6893 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6894 spa
->spa_all_vdev_zaps
!= 0);
6896 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6897 /* Make and build the new AVZ */
6898 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6899 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6900 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6902 /* Diff old AVZ with new one */
6906 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6907 spa
->spa_all_vdev_zaps
);
6908 zap_cursor_retrieve(&zc
, &za
) == 0;
6909 zap_cursor_advance(&zc
)) {
6910 uint64_t vdzap
= za
.za_first_integer
;
6911 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6914 * ZAP is listed in old AVZ but not in new one;
6917 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6922 zap_cursor_fini(&zc
);
6924 /* Destroy the old AVZ */
6925 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6926 spa
->spa_all_vdev_zaps
, tx
));
6928 /* Replace the old AVZ in the dir obj with the new one */
6929 VERIFY0(zap_update(spa
->spa_meta_objset
,
6930 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6931 sizeof (new_avz
), 1, &new_avz
, tx
));
6933 spa
->spa_all_vdev_zaps
= new_avz
;
6934 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6938 /* Walk through the AVZ and destroy all listed ZAPs */
6939 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6940 spa
->spa_all_vdev_zaps
);
6941 zap_cursor_retrieve(&zc
, &za
) == 0;
6942 zap_cursor_advance(&zc
)) {
6943 uint64_t zap
= za
.za_first_integer
;
6944 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6947 zap_cursor_fini(&zc
);
6949 /* Destroy and unlink the AVZ itself */
6950 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6951 spa
->spa_all_vdev_zaps
, tx
));
6952 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6953 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6954 spa
->spa_all_vdev_zaps
= 0;
6957 if (spa
->spa_all_vdev_zaps
== 0) {
6958 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6959 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6960 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6962 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6964 /* Create ZAPs for vdevs that don't have them. */
6965 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6967 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6968 dmu_tx_get_txg(tx
), B_FALSE
);
6971 * If we're upgrading the spa version then make sure that
6972 * the config object gets updated with the correct version.
6974 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6975 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6976 spa
->spa_uberblock
.ub_version
);
6978 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6980 nvlist_free(spa
->spa_config_syncing
);
6981 spa
->spa_config_syncing
= config
;
6983 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6987 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6989 uint64_t *versionp
= arg
;
6990 uint64_t version
= *versionp
;
6991 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6994 * Setting the version is special cased when first creating the pool.
6996 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6998 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6999 ASSERT(version
>= spa_version(spa
));
7001 spa
->spa_uberblock
.ub_version
= version
;
7002 vdev_config_dirty(spa
->spa_root_vdev
);
7003 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
7007 * Set zpool properties.
7010 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
7012 nvlist_t
*nvp
= arg
;
7013 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
7014 objset_t
*mos
= spa
->spa_meta_objset
;
7015 nvpair_t
*elem
= NULL
;
7017 mutex_enter(&spa
->spa_props_lock
);
7019 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
7021 char *strval
, *fname
;
7023 const char *propname
;
7024 zprop_type_t proptype
;
7027 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
7028 case ZPOOL_PROP_INVAL
:
7030 * We checked this earlier in spa_prop_validate().
7032 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
7034 fname
= strchr(nvpair_name(elem
), '@') + 1;
7035 VERIFY0(zfeature_lookup_name(fname
, &fid
));
7037 spa_feature_enable(spa
, fid
, tx
);
7038 spa_history_log_internal(spa
, "set", tx
,
7039 "%s=enabled", nvpair_name(elem
));
7042 case ZPOOL_PROP_VERSION
:
7043 intval
= fnvpair_value_uint64(elem
);
7045 * The version is synced separately before other
7046 * properties and should be correct by now.
7048 ASSERT3U(spa_version(spa
), >=, intval
);
7051 case ZPOOL_PROP_ALTROOT
:
7053 * 'altroot' is a non-persistent property. It should
7054 * have been set temporarily at creation or import time.
7056 ASSERT(spa
->spa_root
!= NULL
);
7059 case ZPOOL_PROP_READONLY
:
7060 case ZPOOL_PROP_CACHEFILE
:
7062 * 'readonly' and 'cachefile' are also non-persisitent
7066 case ZPOOL_PROP_COMMENT
:
7067 strval
= fnvpair_value_string(elem
);
7068 if (spa
->spa_comment
!= NULL
)
7069 spa_strfree(spa
->spa_comment
);
7070 spa
->spa_comment
= spa_strdup(strval
);
7072 * We need to dirty the configuration on all the vdevs
7073 * so that their labels get updated. It's unnecessary
7074 * to do this for pool creation since the vdev's
7075 * configuration has already been dirtied.
7077 if (tx
->tx_txg
!= TXG_INITIAL
)
7078 vdev_config_dirty(spa
->spa_root_vdev
);
7079 spa_history_log_internal(spa
, "set", tx
,
7080 "%s=%s", nvpair_name(elem
), strval
);
7084 * Set pool property values in the poolprops mos object.
7086 if (spa
->spa_pool_props_object
== 0) {
7087 spa
->spa_pool_props_object
=
7088 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
7089 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
7093 /* normalize the property name */
7094 propname
= zpool_prop_to_name(prop
);
7095 proptype
= zpool_prop_get_type(prop
);
7097 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
7098 ASSERT(proptype
== PROP_TYPE_STRING
);
7099 strval
= fnvpair_value_string(elem
);
7100 VERIFY0(zap_update(mos
,
7101 spa
->spa_pool_props_object
, propname
,
7102 1, strlen(strval
) + 1, strval
, tx
));
7103 spa_history_log_internal(spa
, "set", tx
,
7104 "%s=%s", nvpair_name(elem
), strval
);
7105 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
7106 intval
= fnvpair_value_uint64(elem
);
7108 if (proptype
== PROP_TYPE_INDEX
) {
7110 VERIFY0(zpool_prop_index_to_string(
7111 prop
, intval
, &unused
));
7113 VERIFY0(zap_update(mos
,
7114 spa
->spa_pool_props_object
, propname
,
7115 8, 1, &intval
, tx
));
7116 spa_history_log_internal(spa
, "set", tx
,
7117 "%s=%lld", nvpair_name(elem
), intval
);
7119 ASSERT(0); /* not allowed */
7123 case ZPOOL_PROP_DELEGATION
:
7124 spa
->spa_delegation
= intval
;
7126 case ZPOOL_PROP_BOOTFS
:
7127 spa
->spa_bootfs
= intval
;
7129 case ZPOOL_PROP_FAILUREMODE
:
7130 spa
->spa_failmode
= intval
;
7132 case ZPOOL_PROP_AUTOEXPAND
:
7133 spa
->spa_autoexpand
= intval
;
7134 if (tx
->tx_txg
!= TXG_INITIAL
)
7135 spa_async_request(spa
,
7136 SPA_ASYNC_AUTOEXPAND
);
7138 case ZPOOL_PROP_MULTIHOST
:
7139 spa
->spa_multihost
= intval
;
7141 case ZPOOL_PROP_DEDUPDITTO
:
7142 spa
->spa_dedup_ditto
= intval
;
7151 mutex_exit(&spa
->spa_props_lock
);
7155 * Perform one-time upgrade on-disk changes. spa_version() does not
7156 * reflect the new version this txg, so there must be no changes this
7157 * txg to anything that the upgrade code depends on after it executes.
7158 * Therefore this must be called after dsl_pool_sync() does the sync
7162 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
7164 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7166 ASSERT(spa
->spa_sync_pass
== 1);
7168 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
7170 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
7171 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
7172 dsl_pool_create_origin(dp
, tx
);
7174 /* Keeping the origin open increases spa_minref */
7175 spa
->spa_minref
+= 3;
7178 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
7179 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
7180 dsl_pool_upgrade_clones(dp
, tx
);
7183 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
7184 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
7185 dsl_pool_upgrade_dir_clones(dp
, tx
);
7187 /* Keeping the freedir open increases spa_minref */
7188 spa
->spa_minref
+= 3;
7191 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
7192 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
7193 spa_feature_create_zap_objects(spa
, tx
);
7197 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
7198 * when possibility to use lz4 compression for metadata was added
7199 * Old pools that have this feature enabled must be upgraded to have
7200 * this feature active
7202 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
7203 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
7204 SPA_FEATURE_LZ4_COMPRESS
);
7205 boolean_t lz4_ac
= spa_feature_is_active(spa
,
7206 SPA_FEATURE_LZ4_COMPRESS
);
7208 if (lz4_en
&& !lz4_ac
)
7209 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
7213 * If we haven't written the salt, do so now. Note that the
7214 * feature may not be activated yet, but that's fine since
7215 * the presence of this ZAP entry is backwards compatible.
7217 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
7218 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
7219 VERIFY0(zap_add(spa
->spa_meta_objset
,
7220 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
7221 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
7222 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
7225 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
7229 vdev_indirect_state_sync_verify(vdev_t
*vd
)
7231 ASSERTV(vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
);
7232 ASSERTV(vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
);
7234 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
7235 ASSERT(vim
!= NULL
);
7236 ASSERT(vib
!= NULL
);
7239 if (vdev_obsolete_sm_object(vd
) != 0) {
7240 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
7241 ASSERT(vd
->vdev_removing
||
7242 vd
->vdev_ops
== &vdev_indirect_ops
);
7243 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
7244 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
7246 ASSERT3U(vdev_obsolete_sm_object(vd
), ==,
7247 space_map_object(vd
->vdev_obsolete_sm
));
7248 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
7249 space_map_allocated(vd
->vdev_obsolete_sm
));
7251 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
7254 * Since frees / remaps to an indirect vdev can only
7255 * happen in syncing context, the obsolete segments
7256 * tree must be empty when we start syncing.
7258 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
7262 * Sync the specified transaction group. New blocks may be dirtied as
7263 * part of the process, so we iterate until it converges.
7266 spa_sync(spa_t
*spa
, uint64_t txg
)
7268 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7269 objset_t
*mos
= spa
->spa_meta_objset
;
7270 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
7271 vdev_t
*rvd
= spa
->spa_root_vdev
;
7275 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
7276 zfs_vdev_queue_depth_pct
/ 100;
7278 VERIFY(spa_writeable(spa
));
7281 * Wait for i/os issued in open context that need to complete
7282 * before this txg syncs.
7284 VERIFY0(zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]));
7285 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
, 0);
7288 * Lock out configuration changes.
7290 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
7292 spa
->spa_syncing_txg
= txg
;
7293 spa
->spa_sync_pass
= 0;
7295 mutex_enter(&spa
->spa_alloc_lock
);
7296 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7297 mutex_exit(&spa
->spa_alloc_lock
);
7300 * If there are any pending vdev state changes, convert them
7301 * into config changes that go out with this transaction group.
7303 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
7304 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
7306 * We need the write lock here because, for aux vdevs,
7307 * calling vdev_config_dirty() modifies sav_config.
7308 * This is ugly and will become unnecessary when we
7309 * eliminate the aux vdev wart by integrating all vdevs
7310 * into the root vdev tree.
7312 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7313 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
7314 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
7315 vdev_state_clean(vd
);
7316 vdev_config_dirty(vd
);
7318 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7319 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7321 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7323 tx
= dmu_tx_create_assigned(dp
, txg
);
7325 spa
->spa_sync_starttime
= gethrtime();
7326 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
7327 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
7328 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
7329 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
7332 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
7333 * set spa_deflate if we have no raid-z vdevs.
7335 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
7336 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
7339 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
7340 vd
= rvd
->vdev_child
[i
];
7341 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
7344 if (i
== rvd
->vdev_children
) {
7345 spa
->spa_deflate
= TRUE
;
7346 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
7347 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
7348 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
7353 * Set the top-level vdev's max queue depth. Evaluate each
7354 * top-level's async write queue depth in case it changed.
7355 * The max queue depth will not change in the middle of syncing
7358 uint64_t queue_depth_total
= 0;
7359 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
7360 vdev_t
*tvd
= rvd
->vdev_child
[c
];
7361 metaslab_group_t
*mg
= tvd
->vdev_mg
;
7363 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
7364 !metaslab_group_initialized(mg
))
7368 * It is safe to do a lock-free check here because only async
7369 * allocations look at mg_max_alloc_queue_depth, and async
7370 * allocations all happen from spa_sync().
7372 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
7373 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
7374 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
7376 metaslab_class_t
*mc
= spa_normal_class(spa
);
7377 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
7378 mc
->mc_alloc_max_slots
= queue_depth_total
;
7379 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
7381 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
7382 max_queue_depth
* rvd
->vdev_children
);
7384 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
7385 vdev_t
*vd
= rvd
->vdev_child
[c
];
7386 vdev_indirect_state_sync_verify(vd
);
7388 if (vdev_indirect_should_condense(vd
)) {
7389 spa_condense_indirect_start_sync(vd
, tx
);
7395 * Iterate to convergence.
7398 int pass
= ++spa
->spa_sync_pass
;
7400 spa_sync_config_object(spa
, tx
);
7401 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
7402 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
7403 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
7404 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
7405 spa_errlog_sync(spa
, txg
);
7406 dsl_pool_sync(dp
, txg
);
7408 if (pass
< zfs_sync_pass_deferred_free
) {
7409 spa_sync_frees(spa
, free_bpl
, tx
);
7412 * We can not defer frees in pass 1, because
7413 * we sync the deferred frees later in pass 1.
7415 ASSERT3U(pass
, >, 1);
7416 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
7417 &spa
->spa_deferred_bpobj
, tx
);
7421 dsl_scan_sync(dp
, tx
);
7423 if (spa
->spa_vdev_removal
!= NULL
)
7426 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
7431 spa_sync_upgrades(spa
, tx
);
7433 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
7435 * Note: We need to check if the MOS is dirty
7436 * because we could have marked the MOS dirty
7437 * without updating the uberblock (e.g. if we
7438 * have sync tasks but no dirty user data). We
7439 * need to check the uberblock's rootbp because
7440 * it is updated if we have synced out dirty
7441 * data (though in this case the MOS will most
7442 * likely also be dirty due to second order
7443 * effects, we don't want to rely on that here).
7445 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
7446 !dmu_objset_is_dirty(mos
, txg
)) {
7448 * Nothing changed on the first pass,
7449 * therefore this TXG is a no-op. Avoid
7450 * syncing deferred frees, so that we
7451 * can keep this TXG as a no-op.
7453 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
7455 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7456 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
7459 spa_sync_deferred_frees(spa
, tx
);
7462 } while (dmu_objset_is_dirty(mos
, txg
));
7465 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
7467 * Make sure that the number of ZAPs for all the vdevs matches
7468 * the number of ZAPs in the per-vdev ZAP list. This only gets
7469 * called if the config is dirty; otherwise there may be
7470 * outstanding AVZ operations that weren't completed in
7471 * spa_sync_config_object.
7473 uint64_t all_vdev_zap_entry_count
;
7474 ASSERT0(zap_count(spa
->spa_meta_objset
,
7475 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
7476 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
7477 all_vdev_zap_entry_count
);
7481 if (spa
->spa_vdev_removal
!= NULL
) {
7482 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
7486 * Rewrite the vdev configuration (which includes the uberblock)
7487 * to commit the transaction group.
7489 * If there are no dirty vdevs, we sync the uberblock to a few
7490 * random top-level vdevs that are known to be visible in the
7491 * config cache (see spa_vdev_add() for a complete description).
7492 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7496 * We hold SCL_STATE to prevent vdev open/close/etc.
7497 * while we're attempting to write the vdev labels.
7499 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
7501 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
7502 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
];
7504 int children
= rvd
->vdev_children
;
7505 int c0
= spa_get_random(children
);
7507 for (int c
= 0; c
< children
; c
++) {
7508 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
7509 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
7510 !vdev_is_concrete(vd
))
7512 svd
[svdcount
++] = vd
;
7513 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
7516 error
= vdev_config_sync(svd
, svdcount
, txg
);
7518 error
= vdev_config_sync(rvd
->vdev_child
,
7519 rvd
->vdev_children
, txg
);
7523 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
7525 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7529 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
7530 zio_resume_wait(spa
);
7534 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
7535 spa
->spa_deadman_tqid
= 0;
7538 * Clear the dirty config list.
7540 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
7541 vdev_config_clean(vd
);
7544 * Now that the new config has synced transactionally,
7545 * let it become visible to the config cache.
7547 if (spa
->spa_config_syncing
!= NULL
) {
7548 spa_config_set(spa
, spa
->spa_config_syncing
);
7549 spa
->spa_config_txg
= txg
;
7550 spa
->spa_config_syncing
= NULL
;
7553 dsl_pool_sync_done(dp
, txg
);
7555 mutex_enter(&spa
->spa_alloc_lock
);
7556 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7557 mutex_exit(&spa
->spa_alloc_lock
);
7560 * Update usable space statistics.
7562 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
7563 vdev_sync_done(vd
, txg
);
7565 spa_update_dspace(spa
);
7568 * It had better be the case that we didn't dirty anything
7569 * since vdev_config_sync().
7571 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7572 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7573 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7575 spa
->spa_sync_pass
= 0;
7578 * Update the last synced uberblock here. We want to do this at
7579 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7580 * will be guaranteed that all the processing associated with
7581 * that txg has been completed.
7583 spa
->spa_ubsync
= spa
->spa_uberblock
;
7584 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7586 spa_handle_ignored_writes(spa
);
7589 * If any async tasks have been requested, kick them off.
7591 spa_async_dispatch(spa
);
7595 * Sync all pools. We don't want to hold the namespace lock across these
7596 * operations, so we take a reference on the spa_t and drop the lock during the
7600 spa_sync_allpools(void)
7603 mutex_enter(&spa_namespace_lock
);
7604 while ((spa
= spa_next(spa
)) != NULL
) {
7605 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7606 !spa_writeable(spa
) || spa_suspended(spa
))
7608 spa_open_ref(spa
, FTAG
);
7609 mutex_exit(&spa_namespace_lock
);
7610 txg_wait_synced(spa_get_dsl(spa
), 0);
7611 mutex_enter(&spa_namespace_lock
);
7612 spa_close(spa
, FTAG
);
7614 mutex_exit(&spa_namespace_lock
);
7618 * ==========================================================================
7619 * Miscellaneous routines
7620 * ==========================================================================
7624 * Remove all pools in the system.
7632 * Remove all cached state. All pools should be closed now,
7633 * so every spa in the AVL tree should be unreferenced.
7635 mutex_enter(&spa_namespace_lock
);
7636 while ((spa
= spa_next(NULL
)) != NULL
) {
7638 * Stop async tasks. The async thread may need to detach
7639 * a device that's been replaced, which requires grabbing
7640 * spa_namespace_lock, so we must drop it here.
7642 spa_open_ref(spa
, FTAG
);
7643 mutex_exit(&spa_namespace_lock
);
7644 spa_async_suspend(spa
);
7645 mutex_enter(&spa_namespace_lock
);
7646 spa_close(spa
, FTAG
);
7648 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7650 spa_deactivate(spa
);
7654 mutex_exit(&spa_namespace_lock
);
7658 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7663 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7667 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7668 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7669 if (vd
->vdev_guid
== guid
)
7673 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7674 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7675 if (vd
->vdev_guid
== guid
)
7684 spa_upgrade(spa_t
*spa
, uint64_t version
)
7686 ASSERT(spa_writeable(spa
));
7688 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7691 * This should only be called for a non-faulted pool, and since a
7692 * future version would result in an unopenable pool, this shouldn't be
7695 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7696 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7698 spa
->spa_uberblock
.ub_version
= version
;
7699 vdev_config_dirty(spa
->spa_root_vdev
);
7701 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7703 txg_wait_synced(spa_get_dsl(spa
), 0);
7707 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7711 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7713 for (i
= 0; i
< sav
->sav_count
; i
++)
7714 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7717 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7718 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7719 &spareguid
) == 0 && spareguid
== guid
)
7727 * Check if a pool has an active shared spare device.
7728 * Note: reference count of an active spare is 2, as a spare and as a replace
7731 spa_has_active_shared_spare(spa_t
*spa
)
7735 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7737 for (i
= 0; i
< sav
->sav_count
; i
++) {
7738 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7739 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7748 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7750 sysevent_t
*ev
= NULL
;
7754 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
7756 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
7757 ev
->resource
= resource
;
7764 spa_event_post(sysevent_t
*ev
)
7768 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
7769 kmem_free(ev
, sizeof (*ev
));
7775 * Post a zevent corresponding to the given sysevent. The 'name' must be one
7776 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
7777 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7778 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7779 * or zdb as real changes.
7782 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7784 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
7787 #if defined(_KERNEL)
7788 /* state manipulation functions */
7789 EXPORT_SYMBOL(spa_open
);
7790 EXPORT_SYMBOL(spa_open_rewind
);
7791 EXPORT_SYMBOL(spa_get_stats
);
7792 EXPORT_SYMBOL(spa_create
);
7793 EXPORT_SYMBOL(spa_import
);
7794 EXPORT_SYMBOL(spa_tryimport
);
7795 EXPORT_SYMBOL(spa_destroy
);
7796 EXPORT_SYMBOL(spa_export
);
7797 EXPORT_SYMBOL(spa_reset
);
7798 EXPORT_SYMBOL(spa_async_request
);
7799 EXPORT_SYMBOL(spa_async_suspend
);
7800 EXPORT_SYMBOL(spa_async_resume
);
7801 EXPORT_SYMBOL(spa_inject_addref
);
7802 EXPORT_SYMBOL(spa_inject_delref
);
7803 EXPORT_SYMBOL(spa_scan_stat_init
);
7804 EXPORT_SYMBOL(spa_scan_get_stats
);
7806 /* device maniion */
7807 EXPORT_SYMBOL(spa_vdev_add
);
7808 EXPORT_SYMBOL(spa_vdev_attach
);
7809 EXPORT_SYMBOL(spa_vdev_detach
);
7810 EXPORT_SYMBOL(spa_vdev_setpath
);
7811 EXPORT_SYMBOL(spa_vdev_setfru
);
7812 EXPORT_SYMBOL(spa_vdev_split_mirror
);
7814 /* spare statech is global across all pools) */
7815 EXPORT_SYMBOL(spa_spare_add
);
7816 EXPORT_SYMBOL(spa_spare_remove
);
7817 EXPORT_SYMBOL(spa_spare_exists
);
7818 EXPORT_SYMBOL(spa_spare_activate
);
7820 /* L2ARC statech is global across all pools) */
7821 EXPORT_SYMBOL(spa_l2cache_add
);
7822 EXPORT_SYMBOL(spa_l2cache_remove
);
7823 EXPORT_SYMBOL(spa_l2cache_exists
);
7824 EXPORT_SYMBOL(spa_l2cache_activate
);
7825 EXPORT_SYMBOL(spa_l2cache_drop
);
7828 EXPORT_SYMBOL(spa_scan
);
7829 EXPORT_SYMBOL(spa_scan_stop
);
7832 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
7833 EXPORT_SYMBOL(spa_sync_allpools
);
7836 EXPORT_SYMBOL(spa_prop_set
);
7837 EXPORT_SYMBOL(spa_prop_get
);
7838 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
7840 /* asynchronous event notification */
7841 EXPORT_SYMBOL(spa_event_notify
);
7844 #if defined(_KERNEL)
7845 module_param(spa_load_verify_maxinflight
, int, 0644);
7846 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
7847 "Max concurrent traversal I/Os while verifying pool during import -X");
7849 module_param(spa_load_verify_metadata
, int, 0644);
7850 MODULE_PARM_DESC(spa_load_verify_metadata
,
7851 "Set to traverse metadata on pool import");
7853 module_param(spa_load_verify_data
, int, 0644);
7854 MODULE_PARM_DESC(spa_load_verify_data
,
7855 "Set to traverse data on pool import");
7857 module_param(spa_load_print_vdev_tree
, int, 0644);
7858 MODULE_PARM_DESC(spa_load_print_vdev_tree
,
7859 "Print vdev tree to zfs_dbgmsg during pool import");
7862 module_param(zio_taskq_batch_pct
, uint
, 0444);
7863 MODULE_PARM_DESC(zio_taskq_batch_pct
,
7864 "Percentage of CPUs to run an IO worker thread");
7867 module_param(zfs_max_missing_tvds
, ulong
, 0644);
7868 MODULE_PARM_DESC(zfs_max_missing_tvds
,
7869 "Allow importing pool with up to this number of missing top-level vdevs"
7870 " (in read-only mode)");