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) 2013 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 (c) 2016 Actifio, Inc. All rights reserved.
32 * SPA: Storage Pool Allocator
34 * This file contains all the routines used when modifying on-disk SPA state.
35 * This includes opening, importing, destroying, exporting a pool, and syncing a
39 #include <sys/zfs_context.h>
40 #include <sys/fm/fs/zfs.h>
41 #include <sys/spa_impl.h>
43 #include <sys/zio_checksum.h>
45 #include <sys/dmu_tx.h>
49 #include <sys/vdev_impl.h>
50 #include <sys/vdev_disk.h>
51 #include <sys/metaslab.h>
52 #include <sys/metaslab_impl.h>
53 #include <sys/uberblock_impl.h>
56 #include <sys/dmu_traverse.h>
57 #include <sys/dmu_objset.h>
58 #include <sys/unique.h>
59 #include <sys/dsl_pool.h>
60 #include <sys/dsl_dataset.h>
61 #include <sys/dsl_dir.h>
62 #include <sys/dsl_prop.h>
63 #include <sys/dsl_synctask.h>
64 #include <sys/fs/zfs.h>
66 #include <sys/callb.h>
67 #include <sys/systeminfo.h>
68 #include <sys/spa_boot.h>
69 #include <sys/zfs_ioctl.h>
70 #include <sys/dsl_scan.h>
71 #include <sys/zfeature.h>
72 #include <sys/dsl_destroy.h>
76 #include <sys/bootprops.h>
77 #include <sys/callb.h>
78 #include <sys/cpupart.h>
80 #include <sys/sysdc.h>
85 #include "zfs_comutil.h"
88 * The interval, in seconds, at which failed configuration cache file writes
91 static int zfs_ccw_retry_interval
= 300;
93 typedef enum zti_modes
{
94 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
95 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
96 ZTI_MODE_NULL
, /* don't create a taskq */
100 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
101 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
102 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
103 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
105 #define ZTI_N(n) ZTI_P(n, 1)
106 #define ZTI_ONE ZTI_N(1)
108 typedef struct zio_taskq_info
{
109 zti_modes_t zti_mode
;
114 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
115 "iss", "iss_h", "int", "int_h"
119 * This table defines the taskq settings for each ZFS I/O type. When
120 * initializing a pool, we use this table to create an appropriately sized
121 * taskq. Some operations are low volume and therefore have a small, static
122 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
123 * macros. Other operations process a large amount of data; the ZTI_BATCH
124 * macro causes us to create a taskq oriented for throughput. Some operations
125 * are so high frequency and short-lived that the taskq itself can become a a
126 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
127 * additional degree of parallelism specified by the number of threads per-
128 * taskq and the number of taskqs; when dispatching an event in this case, the
129 * particular taskq is chosen at random.
131 * The different taskq priorities are to handle the different contexts (issue
132 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
133 * need to be handled with minimum delay.
135 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
136 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
137 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
138 { ZTI_N(8), ZTI_NULL
, ZTI_P(12, 8), ZTI_NULL
}, /* READ */
139 { ZTI_BATCH
, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */
140 { ZTI_P(12, 8), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
141 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
142 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
145 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
146 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
147 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
148 static inline int spa_load_impl(spa_t
*spa
, uint64_t, nvlist_t
*config
,
149 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
151 static void spa_vdev_resilver_done(spa_t
*spa
);
153 uint_t zio_taskq_batch_pct
= 75; /* 1 thread per cpu in pset */
154 id_t zio_taskq_psrset_bind
= PS_NONE
;
155 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
156 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
158 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
161 * This (illegal) pool name is used when temporarily importing a spa_t in order
162 * to get the vdev stats associated with the imported devices.
164 #define TRYIMPORT_NAME "$import"
167 * ==========================================================================
168 * SPA properties routines
169 * ==========================================================================
173 * Add a (source=src, propname=propval) list to an nvlist.
176 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
177 uint64_t intval
, zprop_source_t src
)
179 const char *propname
= zpool_prop_to_name(prop
);
182 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
183 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
186 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
188 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
190 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
191 nvlist_free(propval
);
195 * Get property values from the spa configuration.
198 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
200 vdev_t
*rvd
= spa
->spa_root_vdev
;
201 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
202 uint64_t size
, alloc
, cap
, version
;
203 zprop_source_t src
= ZPROP_SRC_NONE
;
204 spa_config_dirent_t
*dp
;
205 metaslab_class_t
*mc
= spa_normal_class(spa
);
207 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
210 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
211 size
= metaslab_class_get_space(spa_normal_class(spa
));
212 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
213 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
214 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
215 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
218 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
219 metaslab_class_fragmentation(mc
), src
);
220 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
221 metaslab_class_expandable_space(mc
), src
);
222 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
223 (spa_mode(spa
) == FREAD
), src
);
225 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
226 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
228 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
229 ddt_get_pool_dedup_ratio(spa
), src
);
231 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
232 rvd
->vdev_state
, src
);
234 version
= spa_version(spa
);
235 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
))
236 src
= ZPROP_SRC_DEFAULT
;
238 src
= ZPROP_SRC_LOCAL
;
239 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
244 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
245 * when opening pools before this version freedir will be NULL.
247 if (pool
->dp_free_dir
!= NULL
) {
248 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
249 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
252 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
256 if (pool
->dp_leak_dir
!= NULL
) {
257 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
258 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
261 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
266 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
268 if (spa
->spa_comment
!= NULL
) {
269 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
273 if (spa
->spa_root
!= NULL
)
274 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
277 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
278 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
279 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
281 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
282 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
285 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
)) {
286 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
287 DNODE_MAX_SIZE
, ZPROP_SRC_NONE
);
289 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
290 DNODE_MIN_SIZE
, ZPROP_SRC_NONE
);
293 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
294 if (dp
->scd_path
== NULL
) {
295 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
296 "none", 0, ZPROP_SRC_LOCAL
);
297 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
298 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
299 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
305 * Get zpool property values.
308 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
310 objset_t
*mos
= spa
->spa_meta_objset
;
315 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
319 mutex_enter(&spa
->spa_props_lock
);
322 * Get properties from the spa config.
324 spa_prop_get_config(spa
, nvp
);
326 /* If no pool property object, no more prop to get. */
327 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
328 mutex_exit(&spa
->spa_props_lock
);
333 * Get properties from the MOS pool property object.
335 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
336 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
337 zap_cursor_advance(&zc
)) {
340 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
343 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
346 switch (za
.za_integer_length
) {
348 /* integer property */
349 if (za
.za_first_integer
!=
350 zpool_prop_default_numeric(prop
))
351 src
= ZPROP_SRC_LOCAL
;
353 if (prop
== ZPOOL_PROP_BOOTFS
) {
355 dsl_dataset_t
*ds
= NULL
;
357 dp
= spa_get_dsl(spa
);
358 dsl_pool_config_enter(dp
, FTAG
);
359 if ((err
= dsl_dataset_hold_obj(dp
,
360 za
.za_first_integer
, FTAG
, &ds
))) {
361 dsl_pool_config_exit(dp
, FTAG
);
365 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
367 dsl_dataset_name(ds
, strval
);
368 dsl_dataset_rele(ds
, FTAG
);
369 dsl_pool_config_exit(dp
, FTAG
);
372 intval
= za
.za_first_integer
;
375 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
378 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
383 /* string property */
384 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
385 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
386 za
.za_name
, 1, za
.za_num_integers
, strval
);
388 kmem_free(strval
, za
.za_num_integers
);
391 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
392 kmem_free(strval
, za
.za_num_integers
);
399 zap_cursor_fini(&zc
);
400 mutex_exit(&spa
->spa_props_lock
);
402 if (err
&& err
!= ENOENT
) {
412 * Validate the given pool properties nvlist and modify the list
413 * for the property values to be set.
416 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
419 int error
= 0, reset_bootfs
= 0;
421 boolean_t has_feature
= B_FALSE
;
424 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
426 char *strval
, *slash
, *check
, *fname
;
427 const char *propname
= nvpair_name(elem
);
428 zpool_prop_t prop
= zpool_name_to_prop(propname
);
432 if (!zpool_prop_feature(propname
)) {
433 error
= SET_ERROR(EINVAL
);
438 * Sanitize the input.
440 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
441 error
= SET_ERROR(EINVAL
);
445 if (nvpair_value_uint64(elem
, &intval
) != 0) {
446 error
= SET_ERROR(EINVAL
);
451 error
= SET_ERROR(EINVAL
);
455 fname
= strchr(propname
, '@') + 1;
456 if (zfeature_lookup_name(fname
, NULL
) != 0) {
457 error
= SET_ERROR(EINVAL
);
461 has_feature
= B_TRUE
;
464 case ZPOOL_PROP_VERSION
:
465 error
= nvpair_value_uint64(elem
, &intval
);
467 (intval
< spa_version(spa
) ||
468 intval
> SPA_VERSION_BEFORE_FEATURES
||
470 error
= SET_ERROR(EINVAL
);
473 case ZPOOL_PROP_DELEGATION
:
474 case ZPOOL_PROP_AUTOREPLACE
:
475 case ZPOOL_PROP_LISTSNAPS
:
476 case ZPOOL_PROP_AUTOEXPAND
:
477 error
= nvpair_value_uint64(elem
, &intval
);
478 if (!error
&& intval
> 1)
479 error
= SET_ERROR(EINVAL
);
482 case ZPOOL_PROP_BOOTFS
:
484 * If the pool version is less than SPA_VERSION_BOOTFS,
485 * or the pool is still being created (version == 0),
486 * the bootfs property cannot be set.
488 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
489 error
= SET_ERROR(ENOTSUP
);
494 * Make sure the vdev config is bootable
496 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
497 error
= SET_ERROR(ENOTSUP
);
503 error
= nvpair_value_string(elem
, &strval
);
509 if (strval
== NULL
|| strval
[0] == '\0') {
510 objnum
= zpool_prop_default_numeric(
515 error
= dmu_objset_hold(strval
, FTAG
, &os
);
520 * Must be ZPL, and its property settings
521 * must be supported by GRUB (compression
522 * is not gzip, and large blocks or large
523 * dnodes are not used).
526 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
527 error
= SET_ERROR(ENOTSUP
);
529 dsl_prop_get_int_ds(dmu_objset_ds(os
),
530 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
532 !BOOTFS_COMPRESS_VALID(propval
)) {
533 error
= SET_ERROR(ENOTSUP
);
535 dsl_prop_get_int_ds(dmu_objset_ds(os
),
536 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
538 propval
> SPA_OLD_MAXBLOCKSIZE
) {
539 error
= SET_ERROR(ENOTSUP
);
541 dsl_prop_get_int_ds(dmu_objset_ds(os
),
542 zfs_prop_to_name(ZFS_PROP_DNODESIZE
),
544 propval
!= ZFS_DNSIZE_LEGACY
) {
545 error
= SET_ERROR(ENOTSUP
);
547 objnum
= dmu_objset_id(os
);
549 dmu_objset_rele(os
, FTAG
);
553 case ZPOOL_PROP_FAILUREMODE
:
554 error
= nvpair_value_uint64(elem
, &intval
);
555 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
556 intval
> ZIO_FAILURE_MODE_PANIC
))
557 error
= SET_ERROR(EINVAL
);
560 * This is a special case which only occurs when
561 * the pool has completely failed. This allows
562 * the user to change the in-core failmode property
563 * without syncing it out to disk (I/Os might
564 * currently be blocked). We do this by returning
565 * EIO to the caller (spa_prop_set) to trick it
566 * into thinking we encountered a property validation
569 if (!error
&& spa_suspended(spa
)) {
570 spa
->spa_failmode
= intval
;
571 error
= SET_ERROR(EIO
);
575 case ZPOOL_PROP_CACHEFILE
:
576 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
579 if (strval
[0] == '\0')
582 if (strcmp(strval
, "none") == 0)
585 if (strval
[0] != '/') {
586 error
= SET_ERROR(EINVAL
);
590 slash
= strrchr(strval
, '/');
591 ASSERT(slash
!= NULL
);
593 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
594 strcmp(slash
, "/..") == 0)
595 error
= SET_ERROR(EINVAL
);
598 case ZPOOL_PROP_COMMENT
:
599 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
601 for (check
= strval
; *check
!= '\0'; check
++) {
602 if (!isprint(*check
)) {
603 error
= SET_ERROR(EINVAL
);
607 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
608 error
= SET_ERROR(E2BIG
);
611 case ZPOOL_PROP_DEDUPDITTO
:
612 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
613 error
= SET_ERROR(ENOTSUP
);
615 error
= nvpair_value_uint64(elem
, &intval
);
617 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
618 error
= SET_ERROR(EINVAL
);
629 if (!error
&& reset_bootfs
) {
630 error
= nvlist_remove(props
,
631 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
634 error
= nvlist_add_uint64(props
,
635 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
643 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
646 spa_config_dirent_t
*dp
;
648 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
652 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
655 if (cachefile
[0] == '\0')
656 dp
->scd_path
= spa_strdup(spa_config_path
);
657 else if (strcmp(cachefile
, "none") == 0)
660 dp
->scd_path
= spa_strdup(cachefile
);
662 list_insert_head(&spa
->spa_config_list
, dp
);
664 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
668 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
671 nvpair_t
*elem
= NULL
;
672 boolean_t need_sync
= B_FALSE
;
674 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
677 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
678 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
680 if (prop
== ZPOOL_PROP_CACHEFILE
||
681 prop
== ZPOOL_PROP_ALTROOT
||
682 prop
== ZPOOL_PROP_READONLY
)
685 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPROP_INVAL
) {
688 if (prop
== ZPOOL_PROP_VERSION
) {
689 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
691 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
692 ver
= SPA_VERSION_FEATURES
;
696 /* Save time if the version is already set. */
697 if (ver
== spa_version(spa
))
701 * In addition to the pool directory object, we might
702 * create the pool properties object, the features for
703 * read object, the features for write object, or the
704 * feature descriptions object.
706 error
= dsl_sync_task(spa
->spa_name
, NULL
,
707 spa_sync_version
, &ver
,
708 6, ZFS_SPACE_CHECK_RESERVED
);
719 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
720 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
727 * If the bootfs property value is dsobj, clear it.
730 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
732 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
733 VERIFY(zap_remove(spa
->spa_meta_objset
,
734 spa
->spa_pool_props_object
,
735 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
742 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
744 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
745 vdev_t
*rvd
= spa
->spa_root_vdev
;
747 ASSERTV(uint64_t *newguid
= arg
);
749 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
750 vdev_state
= rvd
->vdev_state
;
751 spa_config_exit(spa
, SCL_STATE
, FTAG
);
753 if (vdev_state
!= VDEV_STATE_HEALTHY
)
754 return (SET_ERROR(ENXIO
));
756 ASSERT3U(spa_guid(spa
), !=, *newguid
);
762 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
764 uint64_t *newguid
= arg
;
765 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
767 vdev_t
*rvd
= spa
->spa_root_vdev
;
769 oldguid
= spa_guid(spa
);
771 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
772 rvd
->vdev_guid
= *newguid
;
773 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
774 vdev_config_dirty(rvd
);
775 spa_config_exit(spa
, SCL_STATE
, FTAG
);
777 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
782 * Change the GUID for the pool. This is done so that we can later
783 * re-import a pool built from a clone of our own vdevs. We will modify
784 * the root vdev's guid, our own pool guid, and then mark all of our
785 * vdevs dirty. Note that we must make sure that all our vdevs are
786 * online when we do this, or else any vdevs that weren't present
787 * would be orphaned from our pool. We are also going to issue a
788 * sysevent to update any watchers.
791 spa_change_guid(spa_t
*spa
)
796 mutex_enter(&spa
->spa_vdev_top_lock
);
797 mutex_enter(&spa_namespace_lock
);
798 guid
= spa_generate_guid(NULL
);
800 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
801 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
804 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
805 spa_event_notify(spa
, NULL
, ESC_ZFS_POOL_REGUID
);
808 mutex_exit(&spa_namespace_lock
);
809 mutex_exit(&spa
->spa_vdev_top_lock
);
815 * ==========================================================================
816 * SPA state manipulation (open/create/destroy/import/export)
817 * ==========================================================================
821 spa_error_entry_compare(const void *a
, const void *b
)
823 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
824 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
827 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
828 sizeof (zbookmark_phys_t
));
839 * Utility function which retrieves copies of the current logs and
840 * re-initializes them in the process.
843 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
845 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
847 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
848 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
850 avl_create(&spa
->spa_errlist_scrub
,
851 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
852 offsetof(spa_error_entry_t
, se_avl
));
853 avl_create(&spa
->spa_errlist_last
,
854 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
855 offsetof(spa_error_entry_t
, se_avl
));
859 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
861 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
862 enum zti_modes mode
= ztip
->zti_mode
;
863 uint_t value
= ztip
->zti_value
;
864 uint_t count
= ztip
->zti_count
;
865 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
867 uint_t i
, flags
= TASKQ_DYNAMIC
;
868 boolean_t batch
= B_FALSE
;
870 if (mode
== ZTI_MODE_NULL
) {
872 tqs
->stqs_taskq
= NULL
;
876 ASSERT3U(count
, >, 0);
878 tqs
->stqs_count
= count
;
879 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
883 ASSERT3U(value
, >=, 1);
884 value
= MAX(value
, 1);
889 flags
|= TASKQ_THREADS_CPU_PCT
;
890 value
= MIN(zio_taskq_batch_pct
, 100);
894 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
896 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
900 for (i
= 0; i
< count
; i
++) {
904 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
905 zio_type_name
[t
], zio_taskq_types
[q
], i
);
907 (void) snprintf(name
, sizeof (name
), "%s_%s",
908 zio_type_name
[t
], zio_taskq_types
[q
]);
911 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
913 flags
|= TASKQ_DC_BATCH
;
915 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
916 spa
->spa_proc
, zio_taskq_basedc
, flags
);
918 pri_t pri
= maxclsyspri
;
920 * The write issue taskq can be extremely CPU
921 * intensive. Run it at slightly less important
922 * priority than the other taskqs. Under Linux this
923 * means incrementing the priority value on platforms
924 * like illumos it should be decremented.
926 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
929 tq
= taskq_create_proc(name
, value
, pri
, 50,
930 INT_MAX
, spa
->spa_proc
, flags
);
933 tqs
->stqs_taskq
[i
] = tq
;
938 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
940 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
943 if (tqs
->stqs_taskq
== NULL
) {
944 ASSERT3U(tqs
->stqs_count
, ==, 0);
948 for (i
= 0; i
< tqs
->stqs_count
; i
++) {
949 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
950 taskq_destroy(tqs
->stqs_taskq
[i
]);
953 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
954 tqs
->stqs_taskq
= NULL
;
958 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
959 * Note that a type may have multiple discrete taskqs to avoid lock contention
960 * on the taskq itself. In that case we choose which taskq at random by using
961 * the low bits of gethrtime().
964 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
965 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
967 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
970 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
971 ASSERT3U(tqs
->stqs_count
, !=, 0);
973 if (tqs
->stqs_count
== 1) {
974 tq
= tqs
->stqs_taskq
[0];
976 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
979 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
983 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
986 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
987 task_func_t
*func
, void *arg
, uint_t flags
)
989 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
993 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
994 ASSERT3U(tqs
->stqs_count
, !=, 0);
996 if (tqs
->stqs_count
== 1) {
997 tq
= tqs
->stqs_taskq
[0];
999 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1002 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1004 taskq_wait_id(tq
, id
);
1008 spa_create_zio_taskqs(spa_t
*spa
)
1012 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1013 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1014 spa_taskqs_init(spa
, t
, q
);
1019 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1021 spa_thread(void *arg
)
1023 callb_cpr_t cprinfo
;
1026 user_t
*pu
= PTOU(curproc
);
1028 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1031 ASSERT(curproc
!= &p0
);
1032 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1033 "zpool-%s", spa
->spa_name
);
1034 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1036 /* bind this thread to the requested psrset */
1037 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1039 mutex_enter(&cpu_lock
);
1040 mutex_enter(&pidlock
);
1041 mutex_enter(&curproc
->p_lock
);
1043 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1044 0, NULL
, NULL
) == 0) {
1045 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1048 "Couldn't bind process for zfs pool \"%s\" to "
1049 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1052 mutex_exit(&curproc
->p_lock
);
1053 mutex_exit(&pidlock
);
1054 mutex_exit(&cpu_lock
);
1058 if (zio_taskq_sysdc
) {
1059 sysdc_thread_enter(curthread
, 100, 0);
1062 spa
->spa_proc
= curproc
;
1063 spa
->spa_did
= curthread
->t_did
;
1065 spa_create_zio_taskqs(spa
);
1067 mutex_enter(&spa
->spa_proc_lock
);
1068 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1070 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1071 cv_broadcast(&spa
->spa_proc_cv
);
1073 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1074 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1075 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1076 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1078 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1079 spa
->spa_proc_state
= SPA_PROC_GONE
;
1080 spa
->spa_proc
= &p0
;
1081 cv_broadcast(&spa
->spa_proc_cv
);
1082 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1084 mutex_enter(&curproc
->p_lock
);
1090 * Activate an uninitialized pool.
1093 spa_activate(spa_t
*spa
, int mode
)
1095 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1097 spa
->spa_state
= POOL_STATE_ACTIVE
;
1098 spa
->spa_mode
= mode
;
1100 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1101 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1103 /* Try to create a covering process */
1104 mutex_enter(&spa
->spa_proc_lock
);
1105 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1106 ASSERT(spa
->spa_proc
== &p0
);
1109 #ifdef HAVE_SPA_THREAD
1110 /* Only create a process if we're going to be around a while. */
1111 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1112 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1114 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1115 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1116 cv_wait(&spa
->spa_proc_cv
,
1117 &spa
->spa_proc_lock
);
1119 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1120 ASSERT(spa
->spa_proc
!= &p0
);
1121 ASSERT(spa
->spa_did
!= 0);
1125 "Couldn't create process for zfs pool \"%s\"\n",
1130 #endif /* HAVE_SPA_THREAD */
1131 mutex_exit(&spa
->spa_proc_lock
);
1133 /* If we didn't create a process, we need to create our taskqs. */
1134 if (spa
->spa_proc
== &p0
) {
1135 spa_create_zio_taskqs(spa
);
1138 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1139 offsetof(vdev_t
, vdev_config_dirty_node
));
1140 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1141 offsetof(objset_t
, os_evicting_node
));
1142 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1143 offsetof(vdev_t
, vdev_state_dirty_node
));
1145 txg_list_create(&spa
->spa_vdev_txg_list
,
1146 offsetof(struct vdev
, vdev_txg_node
));
1148 avl_create(&spa
->spa_errlist_scrub
,
1149 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1150 offsetof(spa_error_entry_t
, se_avl
));
1151 avl_create(&spa
->spa_errlist_last
,
1152 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1153 offsetof(spa_error_entry_t
, se_avl
));
1156 * This taskq is used to perform zvol-minor-related tasks
1157 * asynchronously. This has several advantages, including easy
1158 * resolution of various deadlocks (zfsonlinux bug #3681).
1160 * The taskq must be single threaded to ensure tasks are always
1161 * processed in the order in which they were dispatched.
1163 * A taskq per pool allows one to keep the pools independent.
1164 * This way if one pool is suspended, it will not impact another.
1166 * The preferred location to dispatch a zvol minor task is a sync
1167 * task. In this context, there is easy access to the spa_t and minimal
1168 * error handling is required because the sync task must succeed.
1170 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1175 * Opposite of spa_activate().
1178 spa_deactivate(spa_t
*spa
)
1182 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1183 ASSERT(spa
->spa_dsl_pool
== NULL
);
1184 ASSERT(spa
->spa_root_vdev
== NULL
);
1185 ASSERT(spa
->spa_async_zio_root
== NULL
);
1186 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1188 spa_evicting_os_wait(spa
);
1190 if (spa
->spa_zvol_taskq
) {
1191 taskq_destroy(spa
->spa_zvol_taskq
);
1192 spa
->spa_zvol_taskq
= NULL
;
1195 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1197 list_destroy(&spa
->spa_config_dirty_list
);
1198 list_destroy(&spa
->spa_evicting_os_list
);
1199 list_destroy(&spa
->spa_state_dirty_list
);
1201 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
1203 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1204 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1205 spa_taskqs_fini(spa
, t
, q
);
1209 metaslab_class_destroy(spa
->spa_normal_class
);
1210 spa
->spa_normal_class
= NULL
;
1212 metaslab_class_destroy(spa
->spa_log_class
);
1213 spa
->spa_log_class
= NULL
;
1216 * If this was part of an import or the open otherwise failed, we may
1217 * still have errors left in the queues. Empty them just in case.
1219 spa_errlog_drain(spa
);
1221 avl_destroy(&spa
->spa_errlist_scrub
);
1222 avl_destroy(&spa
->spa_errlist_last
);
1224 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1226 mutex_enter(&spa
->spa_proc_lock
);
1227 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1228 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1229 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1230 cv_broadcast(&spa
->spa_proc_cv
);
1231 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1232 ASSERT(spa
->spa_proc
!= &p0
);
1233 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1235 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1236 spa
->spa_proc_state
= SPA_PROC_NONE
;
1238 ASSERT(spa
->spa_proc
== &p0
);
1239 mutex_exit(&spa
->spa_proc_lock
);
1242 * We want to make sure spa_thread() has actually exited the ZFS
1243 * module, so that the module can't be unloaded out from underneath
1246 if (spa
->spa_did
!= 0) {
1247 thread_join(spa
->spa_did
);
1253 * Verify a pool configuration, and construct the vdev tree appropriately. This
1254 * will create all the necessary vdevs in the appropriate layout, with each vdev
1255 * in the CLOSED state. This will prep the pool before open/creation/import.
1256 * All vdev validation is done by the vdev_alloc() routine.
1259 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1260 uint_t id
, int atype
)
1267 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1270 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1273 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1276 if (error
== ENOENT
)
1282 return (SET_ERROR(EINVAL
));
1285 for (c
= 0; c
< children
; c
++) {
1287 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1295 ASSERT(*vdp
!= NULL
);
1301 * Opposite of spa_load().
1304 spa_unload(spa_t
*spa
)
1308 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1313 spa_async_suspend(spa
);
1318 if (spa
->spa_sync_on
) {
1319 txg_sync_stop(spa
->spa_dsl_pool
);
1320 spa
->spa_sync_on
= B_FALSE
;
1324 * Wait for any outstanding async I/O to complete.
1326 if (spa
->spa_async_zio_root
!= NULL
) {
1327 for (i
= 0; i
< max_ncpus
; i
++)
1328 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1329 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1330 spa
->spa_async_zio_root
= NULL
;
1333 bpobj_close(&spa
->spa_deferred_bpobj
);
1335 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1340 if (spa
->spa_root_vdev
)
1341 vdev_free(spa
->spa_root_vdev
);
1342 ASSERT(spa
->spa_root_vdev
== NULL
);
1345 * Close the dsl pool.
1347 if (spa
->spa_dsl_pool
) {
1348 dsl_pool_close(spa
->spa_dsl_pool
);
1349 spa
->spa_dsl_pool
= NULL
;
1350 spa
->spa_meta_objset
= NULL
;
1357 * Drop and purge level 2 cache
1359 spa_l2cache_drop(spa
);
1361 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1362 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1363 if (spa
->spa_spares
.sav_vdevs
) {
1364 kmem_free(spa
->spa_spares
.sav_vdevs
,
1365 spa
->spa_spares
.sav_count
* sizeof (void *));
1366 spa
->spa_spares
.sav_vdevs
= NULL
;
1368 if (spa
->spa_spares
.sav_config
) {
1369 nvlist_free(spa
->spa_spares
.sav_config
);
1370 spa
->spa_spares
.sav_config
= NULL
;
1372 spa
->spa_spares
.sav_count
= 0;
1374 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1375 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1376 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1378 if (spa
->spa_l2cache
.sav_vdevs
) {
1379 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1380 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1381 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1383 if (spa
->spa_l2cache
.sav_config
) {
1384 nvlist_free(spa
->spa_l2cache
.sav_config
);
1385 spa
->spa_l2cache
.sav_config
= NULL
;
1387 spa
->spa_l2cache
.sav_count
= 0;
1389 spa
->spa_async_suspended
= 0;
1391 if (spa
->spa_comment
!= NULL
) {
1392 spa_strfree(spa
->spa_comment
);
1393 spa
->spa_comment
= NULL
;
1396 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1400 * Load (or re-load) the current list of vdevs describing the active spares for
1401 * this pool. When this is called, we have some form of basic information in
1402 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1403 * then re-generate a more complete list including status information.
1406 spa_load_spares(spa_t
*spa
)
1413 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1416 * First, close and free any existing spare vdevs.
1418 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1419 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1421 /* Undo the call to spa_activate() below */
1422 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1423 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1424 spa_spare_remove(tvd
);
1429 if (spa
->spa_spares
.sav_vdevs
)
1430 kmem_free(spa
->spa_spares
.sav_vdevs
,
1431 spa
->spa_spares
.sav_count
* sizeof (void *));
1433 if (spa
->spa_spares
.sav_config
== NULL
)
1436 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1437 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1439 spa
->spa_spares
.sav_count
= (int)nspares
;
1440 spa
->spa_spares
.sav_vdevs
= NULL
;
1446 * Construct the array of vdevs, opening them to get status in the
1447 * process. For each spare, there is potentially two different vdev_t
1448 * structures associated with it: one in the list of spares (used only
1449 * for basic validation purposes) and one in the active vdev
1450 * configuration (if it's spared in). During this phase we open and
1451 * validate each vdev on the spare list. If the vdev also exists in the
1452 * active configuration, then we also mark this vdev as an active spare.
1454 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1456 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1457 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1458 VDEV_ALLOC_SPARE
) == 0);
1461 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1463 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1464 B_FALSE
)) != NULL
) {
1465 if (!tvd
->vdev_isspare
)
1469 * We only mark the spare active if we were successfully
1470 * able to load the vdev. Otherwise, importing a pool
1471 * with a bad active spare would result in strange
1472 * behavior, because multiple pool would think the spare
1473 * is actively in use.
1475 * There is a vulnerability here to an equally bizarre
1476 * circumstance, where a dead active spare is later
1477 * brought back to life (onlined or otherwise). Given
1478 * the rarity of this scenario, and the extra complexity
1479 * it adds, we ignore the possibility.
1481 if (!vdev_is_dead(tvd
))
1482 spa_spare_activate(tvd
);
1486 vd
->vdev_aux
= &spa
->spa_spares
;
1488 if (vdev_open(vd
) != 0)
1491 if (vdev_validate_aux(vd
) == 0)
1496 * Recompute the stashed list of spares, with status information
1499 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1500 DATA_TYPE_NVLIST_ARRAY
) == 0);
1502 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1504 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1505 spares
[i
] = vdev_config_generate(spa
,
1506 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1507 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1508 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1509 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1510 nvlist_free(spares
[i
]);
1511 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1515 * Load (or re-load) the current list of vdevs describing the active l2cache for
1516 * this pool. When this is called, we have some form of basic information in
1517 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1518 * then re-generate a more complete list including status information.
1519 * Devices which are already active have their details maintained, and are
1523 spa_load_l2cache(spa_t
*spa
)
1527 int i
, j
, oldnvdevs
;
1529 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1530 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1532 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1534 if (sav
->sav_config
!= NULL
) {
1535 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1536 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1537 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1543 oldvdevs
= sav
->sav_vdevs
;
1544 oldnvdevs
= sav
->sav_count
;
1545 sav
->sav_vdevs
= NULL
;
1549 * Process new nvlist of vdevs.
1551 for (i
= 0; i
< nl2cache
; i
++) {
1552 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1556 for (j
= 0; j
< oldnvdevs
; j
++) {
1558 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1560 * Retain previous vdev for add/remove ops.
1568 if (newvdevs
[i
] == NULL
) {
1572 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1573 VDEV_ALLOC_L2CACHE
) == 0);
1578 * Commit this vdev as an l2cache device,
1579 * even if it fails to open.
1581 spa_l2cache_add(vd
);
1586 spa_l2cache_activate(vd
);
1588 if (vdev_open(vd
) != 0)
1591 (void) vdev_validate_aux(vd
);
1593 if (!vdev_is_dead(vd
))
1594 l2arc_add_vdev(spa
, vd
);
1599 * Purge vdevs that were dropped
1601 for (i
= 0; i
< oldnvdevs
; i
++) {
1606 ASSERT(vd
->vdev_isl2cache
);
1608 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1609 pool
!= 0ULL && l2arc_vdev_present(vd
))
1610 l2arc_remove_vdev(vd
);
1611 vdev_clear_stats(vd
);
1617 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1619 if (sav
->sav_config
== NULL
)
1622 sav
->sav_vdevs
= newvdevs
;
1623 sav
->sav_count
= (int)nl2cache
;
1626 * Recompute the stashed list of l2cache devices, with status
1627 * information this time.
1629 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1630 DATA_TYPE_NVLIST_ARRAY
) == 0);
1632 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1633 for (i
= 0; i
< sav
->sav_count
; i
++)
1634 l2cache
[i
] = vdev_config_generate(spa
,
1635 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1636 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1637 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1639 for (i
= 0; i
< sav
->sav_count
; i
++)
1640 nvlist_free(l2cache
[i
]);
1642 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1646 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1649 char *packed
= NULL
;
1654 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1658 nvsize
= *(uint64_t *)db
->db_data
;
1659 dmu_buf_rele(db
, FTAG
);
1661 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1662 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1665 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1666 vmem_free(packed
, nvsize
);
1672 * Checks to see if the given vdev could not be opened, in which case we post a
1673 * sysevent to notify the autoreplace code that the device has been removed.
1676 spa_check_removed(vdev_t
*vd
)
1680 for (c
= 0; c
< vd
->vdev_children
; c
++)
1681 spa_check_removed(vd
->vdev_child
[c
]);
1683 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1685 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1686 spa_event_notify(vd
->vdev_spa
, vd
, ESC_ZFS_VDEV_CHECK
);
1691 spa_config_valid_zaps(vdev_t
*vd
, vdev_t
*mvd
)
1695 ASSERT3U(vd
->vdev_children
, ==, mvd
->vdev_children
);
1697 vd
->vdev_top_zap
= mvd
->vdev_top_zap
;
1698 vd
->vdev_leaf_zap
= mvd
->vdev_leaf_zap
;
1700 for (i
= 0; i
< vd
->vdev_children
; i
++) {
1701 spa_config_valid_zaps(vd
->vdev_child
[i
], mvd
->vdev_child
[i
]);
1706 * Validate the current config against the MOS config
1709 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1711 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1715 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1717 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1718 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1720 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1723 * If we're doing a normal import, then build up any additional
1724 * diagnostic information about missing devices in this config.
1725 * We'll pass this up to the user for further processing.
1727 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1728 nvlist_t
**child
, *nv
;
1731 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1733 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1735 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1736 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1737 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1739 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1740 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1742 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1747 VERIFY(nvlist_add_nvlist_array(nv
,
1748 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1749 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1750 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1752 for (i
= 0; i
< idx
; i
++)
1753 nvlist_free(child
[i
]);
1756 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1760 * Compare the root vdev tree with the information we have
1761 * from the MOS config (mrvd). Check each top-level vdev
1762 * with the corresponding MOS config top-level (mtvd).
1764 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1765 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1766 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1769 * Resolve any "missing" vdevs in the current configuration.
1770 * If we find that the MOS config has more accurate information
1771 * about the top-level vdev then use that vdev instead.
1773 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1774 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1776 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1780 * Device specific actions.
1782 if (mtvd
->vdev_islog
) {
1783 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1786 * XXX - once we have 'readonly' pool
1787 * support we should be able to handle
1788 * missing data devices by transitioning
1789 * the pool to readonly.
1795 * Swap the missing vdev with the data we were
1796 * able to obtain from the MOS config.
1798 vdev_remove_child(rvd
, tvd
);
1799 vdev_remove_child(mrvd
, mtvd
);
1801 vdev_add_child(rvd
, mtvd
);
1802 vdev_add_child(mrvd
, tvd
);
1804 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1806 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1810 if (mtvd
->vdev_islog
) {
1812 * Load the slog device's state from the MOS
1813 * config since it's possible that the label
1814 * does not contain the most up-to-date
1817 vdev_load_log_state(tvd
, mtvd
);
1822 * Per-vdev ZAP info is stored exclusively in the MOS.
1824 spa_config_valid_zaps(tvd
, mtvd
);
1829 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1832 * Ensure we were able to validate the config.
1834 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1838 * Check for missing log devices
1841 spa_check_logs(spa_t
*spa
)
1843 boolean_t rv
= B_FALSE
;
1844 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1846 switch (spa
->spa_log_state
) {
1849 case SPA_LOG_MISSING
:
1850 /* need to recheck in case slog has been restored */
1851 case SPA_LOG_UNKNOWN
:
1852 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1853 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1855 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1862 spa_passivate_log(spa_t
*spa
)
1864 vdev_t
*rvd
= spa
->spa_root_vdev
;
1865 boolean_t slog_found
= B_FALSE
;
1868 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1870 if (!spa_has_slogs(spa
))
1873 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1874 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1875 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1877 if (tvd
->vdev_islog
) {
1878 metaslab_group_passivate(mg
);
1879 slog_found
= B_TRUE
;
1883 return (slog_found
);
1887 spa_activate_log(spa_t
*spa
)
1889 vdev_t
*rvd
= spa
->spa_root_vdev
;
1892 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1894 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1895 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1896 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1898 if (tvd
->vdev_islog
)
1899 metaslab_group_activate(mg
);
1904 spa_offline_log(spa_t
*spa
)
1908 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1909 NULL
, DS_FIND_CHILDREN
);
1912 * We successfully offlined the log device, sync out the
1913 * current txg so that the "stubby" block can be removed
1916 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1922 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1926 for (i
= 0; i
< sav
->sav_count
; i
++)
1927 spa_check_removed(sav
->sav_vdevs
[i
]);
1931 spa_claim_notify(zio_t
*zio
)
1933 spa_t
*spa
= zio
->io_spa
;
1938 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1939 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1940 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1941 mutex_exit(&spa
->spa_props_lock
);
1944 typedef struct spa_load_error
{
1945 uint64_t sle_meta_count
;
1946 uint64_t sle_data_count
;
1950 spa_load_verify_done(zio_t
*zio
)
1952 blkptr_t
*bp
= zio
->io_bp
;
1953 spa_load_error_t
*sle
= zio
->io_private
;
1954 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1955 int error
= zio
->io_error
;
1956 spa_t
*spa
= zio
->io_spa
;
1959 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1960 type
!= DMU_OT_INTENT_LOG
)
1961 atomic_inc_64(&sle
->sle_meta_count
);
1963 atomic_inc_64(&sle
->sle_data_count
);
1965 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1967 mutex_enter(&spa
->spa_scrub_lock
);
1968 spa
->spa_scrub_inflight
--;
1969 cv_broadcast(&spa
->spa_scrub_io_cv
);
1970 mutex_exit(&spa
->spa_scrub_lock
);
1974 * Maximum number of concurrent scrub i/os to create while verifying
1975 * a pool while importing it.
1977 int spa_load_verify_maxinflight
= 10000;
1978 int spa_load_verify_metadata
= B_TRUE
;
1979 int spa_load_verify_data
= B_TRUE
;
1983 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1984 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1990 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
1993 * Note: normally this routine will not be called if
1994 * spa_load_verify_metadata is not set. However, it may be useful
1995 * to manually set the flag after the traversal has begun.
1997 if (!spa_load_verify_metadata
)
1999 if (BP_GET_BUFC_TYPE(bp
) == ARC_BUFC_DATA
&& !spa_load_verify_data
)
2003 size
= BP_GET_PSIZE(bp
);
2004 data
= zio_data_buf_alloc(size
);
2006 mutex_enter(&spa
->spa_scrub_lock
);
2007 while (spa
->spa_scrub_inflight
>= spa_load_verify_maxinflight
)
2008 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2009 spa
->spa_scrub_inflight
++;
2010 mutex_exit(&spa
->spa_scrub_lock
);
2012 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
2013 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2014 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2015 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2021 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2023 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2024 return (SET_ERROR(ENAMETOOLONG
));
2030 spa_load_verify(spa_t
*spa
)
2033 spa_load_error_t sle
= { 0 };
2034 zpool_rewind_policy_t policy
;
2035 boolean_t verify_ok
= B_FALSE
;
2038 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
2040 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
2043 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2044 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2045 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2047 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2051 rio
= zio_root(spa
, NULL
, &sle
,
2052 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2054 if (spa_load_verify_metadata
) {
2055 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2056 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
,
2057 spa_load_verify_cb
, rio
);
2060 (void) zio_wait(rio
);
2062 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2063 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2065 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
2066 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
2070 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2071 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2073 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2074 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2075 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2076 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2077 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2078 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2079 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2081 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2085 if (error
!= ENXIO
&& error
!= EIO
)
2086 error
= SET_ERROR(EIO
);
2090 return (verify_ok
? 0 : EIO
);
2094 * Find a value in the pool props object.
2097 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2099 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2100 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2104 * Find a value in the pool directory object.
2107 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2109 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2110 name
, sizeof (uint64_t), 1, val
));
2114 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2116 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2121 * Fix up config after a partly-completed split. This is done with the
2122 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2123 * pool have that entry in their config, but only the splitting one contains
2124 * a list of all the guids of the vdevs that are being split off.
2126 * This function determines what to do with that list: either rejoin
2127 * all the disks to the pool, or complete the splitting process. To attempt
2128 * the rejoin, each disk that is offlined is marked online again, and
2129 * we do a reopen() call. If the vdev label for every disk that was
2130 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2131 * then we call vdev_split() on each disk, and complete the split.
2133 * Otherwise we leave the config alone, with all the vdevs in place in
2134 * the original pool.
2137 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2144 boolean_t attempt_reopen
;
2146 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2149 /* check that the config is complete */
2150 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2151 &glist
, &gcount
) != 0)
2154 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2156 /* attempt to online all the vdevs & validate */
2157 attempt_reopen
= B_TRUE
;
2158 for (i
= 0; i
< gcount
; i
++) {
2159 if (glist
[i
] == 0) /* vdev is hole */
2162 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2163 if (vd
[i
] == NULL
) {
2165 * Don't bother attempting to reopen the disks;
2166 * just do the split.
2168 attempt_reopen
= B_FALSE
;
2170 /* attempt to re-online it */
2171 vd
[i
]->vdev_offline
= B_FALSE
;
2175 if (attempt_reopen
) {
2176 vdev_reopen(spa
->spa_root_vdev
);
2178 /* check each device to see what state it's in */
2179 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2180 if (vd
[i
] != NULL
&&
2181 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2188 * If every disk has been moved to the new pool, or if we never
2189 * even attempted to look at them, then we split them off for
2192 if (!attempt_reopen
|| gcount
== extracted
) {
2193 for (i
= 0; i
< gcount
; i
++)
2196 vdev_reopen(spa
->spa_root_vdev
);
2199 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2203 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2204 boolean_t mosconfig
)
2206 nvlist_t
*config
= spa
->spa_config
;
2207 char *ereport
= FM_EREPORT_ZFS_POOL
;
2213 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2214 return (SET_ERROR(EINVAL
));
2216 ASSERT(spa
->spa_comment
== NULL
);
2217 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2218 spa
->spa_comment
= spa_strdup(comment
);
2221 * Versioning wasn't explicitly added to the label until later, so if
2222 * it's not present treat it as the initial version.
2224 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2225 &spa
->spa_ubsync
.ub_version
) != 0)
2226 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2228 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2229 &spa
->spa_config_txg
);
2231 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2232 spa_guid_exists(pool_guid
, 0)) {
2233 error
= SET_ERROR(EEXIST
);
2235 spa
->spa_config_guid
= pool_guid
;
2237 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2239 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2243 nvlist_free(spa
->spa_load_info
);
2244 spa
->spa_load_info
= fnvlist_alloc();
2246 gethrestime(&spa
->spa_loaded_ts
);
2247 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2248 mosconfig
, &ereport
);
2252 * Don't count references from objsets that are already closed
2253 * and are making their way through the eviction process.
2255 spa_evicting_os_wait(spa
);
2256 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2258 if (error
!= EEXIST
) {
2259 spa
->spa_loaded_ts
.tv_sec
= 0;
2260 spa
->spa_loaded_ts
.tv_nsec
= 0;
2262 if (error
!= EBADF
) {
2263 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2266 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2274 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2275 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2276 * spa's per-vdev ZAP list.
2279 vdev_count_verify_zaps(vdev_t
*vd
)
2281 spa_t
*spa
= vd
->vdev_spa
;
2285 if (vd
->vdev_top_zap
!= 0) {
2287 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2288 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2290 if (vd
->vdev_leaf_zap
!= 0) {
2292 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2293 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2296 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2297 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2305 * Load an existing storage pool, using the pool's builtin spa_config as a
2306 * source of configuration information.
2308 __attribute__((always_inline
))
2310 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2311 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2315 nvlist_t
*nvroot
= NULL
;
2318 uberblock_t
*ub
= &spa
->spa_uberblock
;
2319 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2320 int orig_mode
= spa
->spa_mode
;
2323 boolean_t missing_feat_write
= B_FALSE
;
2324 nvlist_t
*mos_config
;
2327 * If this is an untrusted config, access the pool in read-only mode.
2328 * This prevents things like resilvering recently removed devices.
2331 spa
->spa_mode
= FREAD
;
2333 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2335 spa
->spa_load_state
= state
;
2337 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2338 return (SET_ERROR(EINVAL
));
2340 parse
= (type
== SPA_IMPORT_EXISTING
?
2341 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2344 * Create "The Godfather" zio to hold all async IOs
2346 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2348 for (i
= 0; i
< max_ncpus
; i
++) {
2349 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2350 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2351 ZIO_FLAG_GODFATHER
);
2355 * Parse the configuration into a vdev tree. We explicitly set the
2356 * value that will be returned by spa_version() since parsing the
2357 * configuration requires knowing the version number.
2359 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2360 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2361 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2366 ASSERT(spa
->spa_root_vdev
== rvd
);
2367 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2368 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2370 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2371 ASSERT(spa_guid(spa
) == pool_guid
);
2375 * Try to open all vdevs, loading each label in the process.
2377 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2378 error
= vdev_open(rvd
);
2379 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2384 * We need to validate the vdev labels against the configuration that
2385 * we have in hand, which is dependent on the setting of mosconfig. If
2386 * mosconfig is true then we're validating the vdev labels based on
2387 * that config. Otherwise, we're validating against the cached config
2388 * (zpool.cache) that was read when we loaded the zfs module, and then
2389 * later we will recursively call spa_load() and validate against
2392 * If we're assembling a new pool that's been split off from an
2393 * existing pool, the labels haven't yet been updated so we skip
2394 * validation for now.
2396 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2397 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2398 error
= vdev_validate(rvd
, mosconfig
);
2399 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2404 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2405 return (SET_ERROR(ENXIO
));
2409 * Find the best uberblock.
2411 vdev_uberblock_load(rvd
, ub
, &label
);
2414 * If we weren't able to find a single valid uberblock, return failure.
2416 if (ub
->ub_txg
== 0) {
2418 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2422 * If the pool has an unsupported version we can't open it.
2424 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2426 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2429 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2433 * If we weren't able to find what's necessary for reading the
2434 * MOS in the label, return failure.
2436 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2437 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2439 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2444 * Update our in-core representation with the definitive values
2447 nvlist_free(spa
->spa_label_features
);
2448 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2454 * Look through entries in the label nvlist's features_for_read. If
2455 * there is a feature listed there which we don't understand then we
2456 * cannot open a pool.
2458 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2459 nvlist_t
*unsup_feat
;
2462 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2465 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2467 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2468 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2469 VERIFY(nvlist_add_string(unsup_feat
,
2470 nvpair_name(nvp
), "") == 0);
2474 if (!nvlist_empty(unsup_feat
)) {
2475 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2476 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2477 nvlist_free(unsup_feat
);
2478 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2482 nvlist_free(unsup_feat
);
2486 * If the vdev guid sum doesn't match the uberblock, we have an
2487 * incomplete configuration. We first check to see if the pool
2488 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2489 * If it is, defer the vdev_guid_sum check till later so we
2490 * can handle missing vdevs.
2492 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2493 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2494 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2495 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2497 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2498 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2499 spa_try_repair(spa
, config
);
2500 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2501 nvlist_free(spa
->spa_config_splitting
);
2502 spa
->spa_config_splitting
= NULL
;
2506 * Initialize internal SPA structures.
2508 spa
->spa_state
= POOL_STATE_ACTIVE
;
2509 spa
->spa_ubsync
= spa
->spa_uberblock
;
2510 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2511 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2512 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2513 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2514 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2515 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2517 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2519 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2520 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2522 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2523 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2525 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2526 boolean_t missing_feat_read
= B_FALSE
;
2527 nvlist_t
*unsup_feat
, *enabled_feat
;
2530 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2531 &spa
->spa_feat_for_read_obj
) != 0) {
2532 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2535 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2536 &spa
->spa_feat_for_write_obj
) != 0) {
2537 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2540 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2541 &spa
->spa_feat_desc_obj
) != 0) {
2542 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2545 enabled_feat
= fnvlist_alloc();
2546 unsup_feat
= fnvlist_alloc();
2548 if (!spa_features_check(spa
, B_FALSE
,
2549 unsup_feat
, enabled_feat
))
2550 missing_feat_read
= B_TRUE
;
2552 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2553 if (!spa_features_check(spa
, B_TRUE
,
2554 unsup_feat
, enabled_feat
)) {
2555 missing_feat_write
= B_TRUE
;
2559 fnvlist_add_nvlist(spa
->spa_load_info
,
2560 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2562 if (!nvlist_empty(unsup_feat
)) {
2563 fnvlist_add_nvlist(spa
->spa_load_info
,
2564 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2567 fnvlist_free(enabled_feat
);
2568 fnvlist_free(unsup_feat
);
2570 if (!missing_feat_read
) {
2571 fnvlist_add_boolean(spa
->spa_load_info
,
2572 ZPOOL_CONFIG_CAN_RDONLY
);
2576 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2577 * twofold: to determine whether the pool is available for
2578 * import in read-write mode and (if it is not) whether the
2579 * pool is available for import in read-only mode. If the pool
2580 * is available for import in read-write mode, it is displayed
2581 * as available in userland; if it is not available for import
2582 * in read-only mode, it is displayed as unavailable in
2583 * userland. If the pool is available for import in read-only
2584 * mode but not read-write mode, it is displayed as unavailable
2585 * in userland with a special note that the pool is actually
2586 * available for open in read-only mode.
2588 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2589 * missing a feature for write, we must first determine whether
2590 * the pool can be opened read-only before returning to
2591 * userland in order to know whether to display the
2592 * abovementioned note.
2594 if (missing_feat_read
|| (missing_feat_write
&&
2595 spa_writeable(spa
))) {
2596 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2601 * Load refcounts for ZFS features from disk into an in-memory
2602 * cache during SPA initialization.
2604 for (i
= 0; i
< SPA_FEATURES
; i
++) {
2607 error
= feature_get_refcount_from_disk(spa
,
2608 &spa_feature_table
[i
], &refcount
);
2610 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2611 } else if (error
== ENOTSUP
) {
2612 spa
->spa_feat_refcount_cache
[i
] =
2613 SPA_FEATURE_DISABLED
;
2615 return (spa_vdev_err(rvd
,
2616 VDEV_AUX_CORRUPT_DATA
, EIO
));
2621 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2622 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2623 &spa
->spa_feat_enabled_txg_obj
) != 0)
2624 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2627 spa
->spa_is_initializing
= B_TRUE
;
2628 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2629 spa
->spa_is_initializing
= B_FALSE
;
2631 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2635 nvlist_t
*policy
= NULL
, *nvconfig
;
2637 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2638 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2640 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2641 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2643 unsigned long myhostid
= 0;
2645 VERIFY(nvlist_lookup_string(nvconfig
,
2646 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2649 myhostid
= zone_get_hostid(NULL
);
2652 * We're emulating the system's hostid in userland, so
2653 * we can't use zone_get_hostid().
2655 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2656 #endif /* _KERNEL */
2657 if (hostid
!= 0 && myhostid
!= 0 &&
2658 hostid
!= myhostid
) {
2659 nvlist_free(nvconfig
);
2660 cmn_err(CE_WARN
, "pool '%s' could not be "
2661 "loaded as it was last accessed by another "
2662 "system (host: %s hostid: 0x%lx). See: "
2663 "http://zfsonlinux.org/msg/ZFS-8000-EY",
2664 spa_name(spa
), hostname
,
2665 (unsigned long)hostid
);
2666 return (SET_ERROR(EBADF
));
2669 if (nvlist_lookup_nvlist(spa
->spa_config
,
2670 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2671 VERIFY(nvlist_add_nvlist(nvconfig
,
2672 ZPOOL_REWIND_POLICY
, policy
) == 0);
2674 spa_config_set(spa
, nvconfig
);
2676 spa_deactivate(spa
);
2677 spa_activate(spa
, orig_mode
);
2679 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2682 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2683 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2684 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2686 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2689 * Load the bit that tells us to use the new accounting function
2690 * (raid-z deflation). If we have an older pool, this will not
2693 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2694 if (error
!= 0 && error
!= ENOENT
)
2695 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2697 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2698 &spa
->spa_creation_version
);
2699 if (error
!= 0 && error
!= ENOENT
)
2700 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2703 * Load the persistent error log. If we have an older pool, this will
2706 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2707 if (error
!= 0 && error
!= ENOENT
)
2708 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2710 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2711 &spa
->spa_errlog_scrub
);
2712 if (error
!= 0 && error
!= ENOENT
)
2713 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2716 * Load the history object. If we have an older pool, this
2717 * will not be present.
2719 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2720 if (error
!= 0 && error
!= ENOENT
)
2721 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2724 * Load the per-vdev ZAP map. If we have an older pool, this will not
2725 * be present; in this case, defer its creation to a later time to
2726 * avoid dirtying the MOS this early / out of sync context. See
2727 * spa_sync_config_object.
2730 /* The sentinel is only available in the MOS config. */
2731 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
2732 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2734 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
2735 &spa
->spa_all_vdev_zaps
);
2737 if (error
!= ENOENT
&& error
!= 0) {
2738 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2739 } else if (error
== 0 && !nvlist_exists(mos_config
,
2740 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
2742 * An older version of ZFS overwrote the sentinel value, so
2743 * we have orphaned per-vdev ZAPs in the MOS. Defer their
2744 * destruction to later; see spa_sync_config_object.
2746 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
2748 * We're assuming that no vdevs have had their ZAPs created
2749 * before this. Better be sure of it.
2751 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
2753 nvlist_free(mos_config
);
2756 * If we're assembling the pool from the split-off vdevs of
2757 * an existing pool, we don't want to attach the spares & cache
2762 * Load any hot spares for this pool.
2764 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2765 if (error
!= 0 && error
!= ENOENT
)
2766 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2767 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2768 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2769 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2770 &spa
->spa_spares
.sav_config
) != 0)
2771 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2773 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2774 spa_load_spares(spa
);
2775 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2776 } else if (error
== 0) {
2777 spa
->spa_spares
.sav_sync
= B_TRUE
;
2781 * Load any level 2 ARC devices for this pool.
2783 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2784 &spa
->spa_l2cache
.sav_object
);
2785 if (error
!= 0 && error
!= ENOENT
)
2786 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2787 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2788 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2789 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2790 &spa
->spa_l2cache
.sav_config
) != 0)
2791 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2793 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2794 spa_load_l2cache(spa
);
2795 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2796 } else if (error
== 0) {
2797 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2800 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2802 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2803 if (error
&& error
!= ENOENT
)
2804 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2807 uint64_t autoreplace
= 0;
2809 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2810 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2811 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2812 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2813 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2814 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2815 &spa
->spa_dedup_ditto
);
2817 spa
->spa_autoreplace
= (autoreplace
!= 0);
2821 * If the 'autoreplace' property is set, then post a resource notifying
2822 * the ZFS DE that it should not issue any faults for unopenable
2823 * devices. We also iterate over the vdevs, and post a sysevent for any
2824 * unopenable vdevs so that the normal autoreplace handler can take
2827 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2828 spa_check_removed(spa
->spa_root_vdev
);
2830 * For the import case, this is done in spa_import(), because
2831 * at this point we're using the spare definitions from
2832 * the MOS config, not necessarily from the userland config.
2834 if (state
!= SPA_LOAD_IMPORT
) {
2835 spa_aux_check_removed(&spa
->spa_spares
);
2836 spa_aux_check_removed(&spa
->spa_l2cache
);
2841 * Load the vdev state for all toplevel vdevs.
2846 * Propagate the leaf DTLs we just loaded all the way up the tree.
2848 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2849 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2850 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2853 * Load the DDTs (dedup tables).
2855 error
= ddt_load(spa
);
2857 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2859 spa_update_dspace(spa
);
2862 * Validate the config, using the MOS config to fill in any
2863 * information which might be missing. If we fail to validate
2864 * the config then declare the pool unfit for use. If we're
2865 * assembling a pool from a split, the log is not transferred
2868 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2871 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2872 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2874 if (!spa_config_valid(spa
, nvconfig
)) {
2875 nvlist_free(nvconfig
);
2876 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2879 nvlist_free(nvconfig
);
2882 * Now that we've validated the config, check the state of the
2883 * root vdev. If it can't be opened, it indicates one or
2884 * more toplevel vdevs are faulted.
2886 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2887 return (SET_ERROR(ENXIO
));
2889 if (spa_writeable(spa
) && spa_check_logs(spa
)) {
2890 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2891 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2895 if (missing_feat_write
) {
2896 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2899 * At this point, we know that we can open the pool in
2900 * read-only mode but not read-write mode. We now have enough
2901 * information and can return to userland.
2903 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2907 * We've successfully opened the pool, verify that we're ready
2908 * to start pushing transactions.
2910 if (state
!= SPA_LOAD_TRYIMPORT
) {
2911 if ((error
= spa_load_verify(spa
)))
2912 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2916 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2917 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2919 int need_update
= B_FALSE
;
2920 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2923 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2926 * Claim log blocks that haven't been committed yet.
2927 * This must all happen in a single txg.
2928 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2929 * invoked from zil_claim_log_block()'s i/o done callback.
2930 * Price of rollback is that we abandon the log.
2932 spa
->spa_claiming
= B_TRUE
;
2934 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
2935 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
2936 zil_claim
, tx
, DS_FIND_CHILDREN
);
2939 spa
->spa_claiming
= B_FALSE
;
2941 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2942 spa
->spa_sync_on
= B_TRUE
;
2943 txg_sync_start(spa
->spa_dsl_pool
);
2946 * Wait for all claims to sync. We sync up to the highest
2947 * claimed log block birth time so that claimed log blocks
2948 * don't appear to be from the future. spa_claim_max_txg
2949 * will have been set for us by either zil_check_log_chain()
2950 * (invoked from spa_check_logs()) or zil_claim() above.
2952 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2955 * If the config cache is stale, or we have uninitialized
2956 * metaslabs (see spa_vdev_add()), then update the config.
2958 * If this is a verbatim import, trust the current
2959 * in-core spa_config and update the disk labels.
2961 if (config_cache_txg
!= spa
->spa_config_txg
||
2962 state
== SPA_LOAD_IMPORT
||
2963 state
== SPA_LOAD_RECOVER
||
2964 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2965 need_update
= B_TRUE
;
2967 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2968 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2969 need_update
= B_TRUE
;
2972 * Update the config cache asychronously in case we're the
2973 * root pool, in which case the config cache isn't writable yet.
2976 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2979 * Check all DTLs to see if anything needs resilvering.
2981 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2982 vdev_resilver_needed(rvd
, NULL
, NULL
))
2983 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2986 * Log the fact that we booted up (so that we can detect if
2987 * we rebooted in the middle of an operation).
2989 spa_history_log_version(spa
, "open");
2992 * Delete any inconsistent datasets.
2994 (void) dmu_objset_find(spa_name(spa
),
2995 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2998 * Clean up any stale temporary dataset userrefs.
3000 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3007 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
3009 int mode
= spa
->spa_mode
;
3012 spa_deactivate(spa
);
3014 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3016 spa_activate(spa
, mode
);
3017 spa_async_suspend(spa
);
3019 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
3023 * If spa_load() fails this function will try loading prior txg's. If
3024 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3025 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3026 * function will not rewind the pool and will return the same error as
3030 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
3031 uint64_t max_request
, int rewind_flags
)
3033 nvlist_t
*loadinfo
= NULL
;
3034 nvlist_t
*config
= NULL
;
3035 int load_error
, rewind_error
;
3036 uint64_t safe_rewind_txg
;
3039 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3040 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3041 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3043 spa
->spa_load_max_txg
= max_request
;
3044 if (max_request
!= UINT64_MAX
)
3045 spa
->spa_extreme_rewind
= B_TRUE
;
3048 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
3050 if (load_error
== 0)
3053 if (spa
->spa_root_vdev
!= NULL
)
3054 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3056 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3057 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3059 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3060 nvlist_free(config
);
3061 return (load_error
);
3064 if (state
== SPA_LOAD_RECOVER
) {
3065 /* Price of rolling back is discarding txgs, including log */
3066 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3069 * If we aren't rolling back save the load info from our first
3070 * import attempt so that we can restore it after attempting
3073 loadinfo
= spa
->spa_load_info
;
3074 spa
->spa_load_info
= fnvlist_alloc();
3077 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3078 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
3079 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
3080 TXG_INITIAL
: safe_rewind_txg
;
3083 * Continue as long as we're finding errors, we're still within
3084 * the acceptable rewind range, and we're still finding uberblocks
3086 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
3087 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
3088 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
3089 spa
->spa_extreme_rewind
= B_TRUE
;
3090 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
3093 spa
->spa_extreme_rewind
= B_FALSE
;
3094 spa
->spa_load_max_txg
= UINT64_MAX
;
3096 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
3097 spa_config_set(spa
, config
);
3099 if (state
== SPA_LOAD_RECOVER
) {
3100 ASSERT3P(loadinfo
, ==, NULL
);
3101 return (rewind_error
);
3103 /* Store the rewind info as part of the initial load info */
3104 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
3105 spa
->spa_load_info
);
3107 /* Restore the initial load info */
3108 fnvlist_free(spa
->spa_load_info
);
3109 spa
->spa_load_info
= loadinfo
;
3111 return (load_error
);
3118 * The import case is identical to an open except that the configuration is sent
3119 * down from userland, instead of grabbed from the configuration cache. For the
3120 * case of an open, the pool configuration will exist in the
3121 * POOL_STATE_UNINITIALIZED state.
3123 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3124 * the same time open the pool, without having to keep around the spa_t in some
3128 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
3132 spa_load_state_t state
= SPA_LOAD_OPEN
;
3134 int locked
= B_FALSE
;
3135 int firstopen
= B_FALSE
;
3140 * As disgusting as this is, we need to support recursive calls to this
3141 * function because dsl_dir_open() is called during spa_load(), and ends
3142 * up calling spa_open() again. The real fix is to figure out how to
3143 * avoid dsl_dir_open() calling this in the first place.
3145 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
3146 mutex_enter(&spa_namespace_lock
);
3150 if ((spa
= spa_lookup(pool
)) == NULL
) {
3152 mutex_exit(&spa_namespace_lock
);
3153 return (SET_ERROR(ENOENT
));
3156 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3157 zpool_rewind_policy_t policy
;
3161 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3163 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3164 state
= SPA_LOAD_RECOVER
;
3166 spa_activate(spa
, spa_mode_global
);
3168 if (state
!= SPA_LOAD_RECOVER
)
3169 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3171 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3172 policy
.zrp_request
);
3174 if (error
== EBADF
) {
3176 * If vdev_validate() returns failure (indicated by
3177 * EBADF), it indicates that one of the vdevs indicates
3178 * that the pool has been exported or destroyed. If
3179 * this is the case, the config cache is out of sync and
3180 * we should remove the pool from the namespace.
3183 spa_deactivate(spa
);
3184 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3187 mutex_exit(&spa_namespace_lock
);
3188 return (SET_ERROR(ENOENT
));
3193 * We can't open the pool, but we still have useful
3194 * information: the state of each vdev after the
3195 * attempted vdev_open(). Return this to the user.
3197 if (config
!= NULL
&& spa
->spa_config
) {
3198 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3200 VERIFY(nvlist_add_nvlist(*config
,
3201 ZPOOL_CONFIG_LOAD_INFO
,
3202 spa
->spa_load_info
) == 0);
3205 spa_deactivate(spa
);
3206 spa
->spa_last_open_failed
= error
;
3208 mutex_exit(&spa_namespace_lock
);
3214 spa_open_ref(spa
, tag
);
3217 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3220 * If we've recovered the pool, pass back any information we
3221 * gathered while doing the load.
3223 if (state
== SPA_LOAD_RECOVER
) {
3224 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3225 spa
->spa_load_info
) == 0);
3229 spa
->spa_last_open_failed
= 0;
3230 spa
->spa_last_ubsync_txg
= 0;
3231 spa
->spa_load_txg
= 0;
3232 mutex_exit(&spa_namespace_lock
);
3236 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
3244 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3247 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3251 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3253 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3257 * Lookup the given spa_t, incrementing the inject count in the process,
3258 * preventing it from being exported or destroyed.
3261 spa_inject_addref(char *name
)
3265 mutex_enter(&spa_namespace_lock
);
3266 if ((spa
= spa_lookup(name
)) == NULL
) {
3267 mutex_exit(&spa_namespace_lock
);
3270 spa
->spa_inject_ref
++;
3271 mutex_exit(&spa_namespace_lock
);
3277 spa_inject_delref(spa_t
*spa
)
3279 mutex_enter(&spa_namespace_lock
);
3280 spa
->spa_inject_ref
--;
3281 mutex_exit(&spa_namespace_lock
);
3285 * Add spares device information to the nvlist.
3288 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3298 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3300 if (spa
->spa_spares
.sav_count
== 0)
3303 VERIFY(nvlist_lookup_nvlist(config
,
3304 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3305 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3306 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3308 VERIFY(nvlist_add_nvlist_array(nvroot
,
3309 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3310 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3311 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3314 * Go through and find any spares which have since been
3315 * repurposed as an active spare. If this is the case, update
3316 * their status appropriately.
3318 for (i
= 0; i
< nspares
; i
++) {
3319 VERIFY(nvlist_lookup_uint64(spares
[i
],
3320 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3321 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3323 VERIFY(nvlist_lookup_uint64_array(
3324 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3325 (uint64_t **)&vs
, &vsc
) == 0);
3326 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3327 vs
->vs_aux
= VDEV_AUX_SPARED
;
3334 * Add l2cache device information to the nvlist, including vdev stats.
3337 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3340 uint_t i
, j
, nl2cache
;
3347 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3349 if (spa
->spa_l2cache
.sav_count
== 0)
3352 VERIFY(nvlist_lookup_nvlist(config
,
3353 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3354 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3355 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3356 if (nl2cache
!= 0) {
3357 VERIFY(nvlist_add_nvlist_array(nvroot
,
3358 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3359 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3360 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3363 * Update level 2 cache device stats.
3366 for (i
= 0; i
< nl2cache
; i
++) {
3367 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3368 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3371 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3373 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3374 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3380 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3381 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3383 vdev_get_stats(vd
, vs
);
3384 vdev_config_generate_stats(vd
, l2cache
[i
]);
3391 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3396 if (spa
->spa_feat_for_read_obj
!= 0) {
3397 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3398 spa
->spa_feat_for_read_obj
);
3399 zap_cursor_retrieve(&zc
, &za
) == 0;
3400 zap_cursor_advance(&zc
)) {
3401 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3402 za
.za_num_integers
== 1);
3403 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3404 za
.za_first_integer
));
3406 zap_cursor_fini(&zc
);
3409 if (spa
->spa_feat_for_write_obj
!= 0) {
3410 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3411 spa
->spa_feat_for_write_obj
);
3412 zap_cursor_retrieve(&zc
, &za
) == 0;
3413 zap_cursor_advance(&zc
)) {
3414 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3415 za
.za_num_integers
== 1);
3416 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3417 za
.za_first_integer
));
3419 zap_cursor_fini(&zc
);
3424 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3428 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3429 zfeature_info_t feature
= spa_feature_table
[i
];
3432 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3435 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3440 * Store a list of pool features and their reference counts in the
3443 * The first time this is called on a spa, allocate a new nvlist, fetch
3444 * the pool features and reference counts from disk, then save the list
3445 * in the spa. In subsequent calls on the same spa use the saved nvlist
3446 * and refresh its values from the cached reference counts. This
3447 * ensures we don't block here on I/O on a suspended pool so 'zpool
3448 * clear' can resume the pool.
3451 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3455 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3457 mutex_enter(&spa
->spa_feat_stats_lock
);
3458 features
= spa
->spa_feat_stats
;
3460 if (features
!= NULL
) {
3461 spa_feature_stats_from_cache(spa
, features
);
3463 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3464 spa
->spa_feat_stats
= features
;
3465 spa_feature_stats_from_disk(spa
, features
);
3468 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3471 mutex_exit(&spa
->spa_feat_stats_lock
);
3475 spa_get_stats(const char *name
, nvlist_t
**config
,
3476 char *altroot
, size_t buflen
)
3482 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3486 * This still leaves a window of inconsistency where the spares
3487 * or l2cache devices could change and the config would be
3488 * self-inconsistent.
3490 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3492 if (*config
!= NULL
) {
3493 uint64_t loadtimes
[2];
3495 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3496 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3497 VERIFY(nvlist_add_uint64_array(*config
,
3498 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3500 VERIFY(nvlist_add_uint64(*config
,
3501 ZPOOL_CONFIG_ERRCOUNT
,
3502 spa_get_errlog_size(spa
)) == 0);
3504 if (spa_suspended(spa
))
3505 VERIFY(nvlist_add_uint64(*config
,
3506 ZPOOL_CONFIG_SUSPENDED
,
3507 spa
->spa_failmode
) == 0);
3509 spa_add_spares(spa
, *config
);
3510 spa_add_l2cache(spa
, *config
);
3511 spa_add_feature_stats(spa
, *config
);
3516 * We want to get the alternate root even for faulted pools, so we cheat
3517 * and call spa_lookup() directly.
3521 mutex_enter(&spa_namespace_lock
);
3522 spa
= spa_lookup(name
);
3524 spa_altroot(spa
, altroot
, buflen
);
3528 mutex_exit(&spa_namespace_lock
);
3530 spa_altroot(spa
, altroot
, buflen
);
3535 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3536 spa_close(spa
, FTAG
);
3543 * Validate that the auxiliary device array is well formed. We must have an
3544 * array of nvlists, each which describes a valid leaf vdev. If this is an
3545 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3546 * specified, as long as they are well-formed.
3549 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3550 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3551 vdev_labeltype_t label
)
3558 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3561 * It's acceptable to have no devs specified.
3563 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3567 return (SET_ERROR(EINVAL
));
3570 * Make sure the pool is formatted with a version that supports this
3573 if (spa_version(spa
) < version
)
3574 return (SET_ERROR(ENOTSUP
));
3577 * Set the pending device list so we correctly handle device in-use
3580 sav
->sav_pending
= dev
;
3581 sav
->sav_npending
= ndev
;
3583 for (i
= 0; i
< ndev
; i
++) {
3584 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3588 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3590 error
= SET_ERROR(EINVAL
);
3595 * The L2ARC currently only supports disk devices in
3596 * kernel context. For user-level testing, we allow it.
3599 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3600 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3601 error
= SET_ERROR(ENOTBLK
);
3608 if ((error
= vdev_open(vd
)) == 0 &&
3609 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3610 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3611 vd
->vdev_guid
) == 0);
3617 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3624 sav
->sav_pending
= NULL
;
3625 sav
->sav_npending
= 0;
3630 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3634 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3636 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3637 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3638 VDEV_LABEL_SPARE
)) != 0) {
3642 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3643 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3644 VDEV_LABEL_L2CACHE
));
3648 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3653 if (sav
->sav_config
!= NULL
) {
3659 * Generate new dev list by concatentating with the
3662 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3663 &olddevs
, &oldndevs
) == 0);
3665 newdevs
= kmem_alloc(sizeof (void *) *
3666 (ndevs
+ oldndevs
), KM_SLEEP
);
3667 for (i
= 0; i
< oldndevs
; i
++)
3668 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3670 for (i
= 0; i
< ndevs
; i
++)
3671 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3674 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3675 DATA_TYPE_NVLIST_ARRAY
) == 0);
3677 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3678 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3679 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3680 nvlist_free(newdevs
[i
]);
3681 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3684 * Generate a new dev list.
3686 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3688 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3694 * Stop and drop level 2 ARC devices
3697 spa_l2cache_drop(spa_t
*spa
)
3701 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3703 for (i
= 0; i
< sav
->sav_count
; i
++) {
3706 vd
= sav
->sav_vdevs
[i
];
3709 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3710 pool
!= 0ULL && l2arc_vdev_present(vd
))
3711 l2arc_remove_vdev(vd
);
3719 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3723 char *altroot
= NULL
;
3728 uint64_t txg
= TXG_INITIAL
;
3729 nvlist_t
**spares
, **l2cache
;
3730 uint_t nspares
, nl2cache
;
3731 uint64_t version
, obj
;
3732 boolean_t has_features
;
3738 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
3739 poolname
= (char *)pool
;
3742 * If this pool already exists, return failure.
3744 mutex_enter(&spa_namespace_lock
);
3745 if (spa_lookup(poolname
) != NULL
) {
3746 mutex_exit(&spa_namespace_lock
);
3747 return (SET_ERROR(EEXIST
));
3751 * Allocate a new spa_t structure.
3753 nvl
= fnvlist_alloc();
3754 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
3755 (void) nvlist_lookup_string(props
,
3756 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3757 spa
= spa_add(poolname
, nvl
, altroot
);
3759 spa_activate(spa
, spa_mode_global
);
3761 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3762 spa_deactivate(spa
);
3764 mutex_exit(&spa_namespace_lock
);
3769 * Temporary pool names should never be written to disk.
3771 if (poolname
!= pool
)
3772 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
3774 has_features
= B_FALSE
;
3775 for (elem
= nvlist_next_nvpair(props
, NULL
);
3776 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3777 if (zpool_prop_feature(nvpair_name(elem
)))
3778 has_features
= B_TRUE
;
3781 if (has_features
|| nvlist_lookup_uint64(props
,
3782 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3783 version
= SPA_VERSION
;
3785 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3787 spa
->spa_first_txg
= txg
;
3788 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3789 spa
->spa_uberblock
.ub_version
= version
;
3790 spa
->spa_ubsync
= spa
->spa_uberblock
;
3793 * Create "The Godfather" zio to hold all async IOs
3795 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3797 for (i
= 0; i
< max_ncpus
; i
++) {
3798 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3799 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3800 ZIO_FLAG_GODFATHER
);
3804 * Create the root vdev.
3806 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3808 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3810 ASSERT(error
!= 0 || rvd
!= NULL
);
3811 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3813 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3814 error
= SET_ERROR(EINVAL
);
3817 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3818 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3819 VDEV_ALLOC_ADD
)) == 0) {
3820 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
3821 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3822 vdev_expand(rvd
->vdev_child
[c
], txg
);
3826 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3830 spa_deactivate(spa
);
3832 mutex_exit(&spa_namespace_lock
);
3837 * Get the list of spares, if specified.
3839 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3840 &spares
, &nspares
) == 0) {
3841 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3843 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3844 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3845 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3846 spa_load_spares(spa
);
3847 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3848 spa
->spa_spares
.sav_sync
= B_TRUE
;
3852 * Get the list of level 2 cache devices, if specified.
3854 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3855 &l2cache
, &nl2cache
) == 0) {
3856 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3857 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3858 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3859 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3860 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3861 spa_load_l2cache(spa
);
3862 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3863 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3866 spa
->spa_is_initializing
= B_TRUE
;
3867 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3868 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3869 spa
->spa_is_initializing
= B_FALSE
;
3872 * Create DDTs (dedup tables).
3876 spa_update_dspace(spa
);
3878 tx
= dmu_tx_create_assigned(dp
, txg
);
3881 * Create the pool config object.
3883 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3884 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3885 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3887 if (zap_add(spa
->spa_meta_objset
,
3888 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3889 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3890 cmn_err(CE_PANIC
, "failed to add pool config");
3893 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3894 spa_feature_create_zap_objects(spa
, tx
);
3896 if (zap_add(spa
->spa_meta_objset
,
3897 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3898 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3899 cmn_err(CE_PANIC
, "failed to add pool version");
3902 /* Newly created pools with the right version are always deflated. */
3903 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3904 spa
->spa_deflate
= TRUE
;
3905 if (zap_add(spa
->spa_meta_objset
,
3906 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3907 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3908 cmn_err(CE_PANIC
, "failed to add deflate");
3913 * Create the deferred-free bpobj. Turn off compression
3914 * because sync-to-convergence takes longer if the blocksize
3917 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3918 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3919 ZIO_COMPRESS_OFF
, tx
);
3920 if (zap_add(spa
->spa_meta_objset
,
3921 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3922 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3923 cmn_err(CE_PANIC
, "failed to add bpobj");
3925 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3926 spa
->spa_meta_objset
, obj
));
3929 * Create the pool's history object.
3931 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3932 spa_history_create_obj(spa
, tx
);
3935 * Set pool properties.
3937 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3938 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3939 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3940 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3942 if (props
!= NULL
) {
3943 spa_configfile_set(spa
, props
, B_FALSE
);
3944 spa_sync_props(props
, tx
);
3949 spa
->spa_sync_on
= B_TRUE
;
3950 txg_sync_start(spa
->spa_dsl_pool
);
3953 * We explicitly wait for the first transaction to complete so that our
3954 * bean counters are appropriately updated.
3956 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3958 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3959 spa_event_notify(spa
, NULL
, ESC_ZFS_POOL_CREATE
);
3961 spa_history_log_version(spa
, "create");
3964 * Don't count references from objsets that are already closed
3965 * and are making their way through the eviction process.
3967 spa_evicting_os_wait(spa
);
3968 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3970 mutex_exit(&spa_namespace_lock
);
3976 * Import a non-root pool into the system.
3979 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
3982 char *altroot
= NULL
;
3983 spa_load_state_t state
= SPA_LOAD_IMPORT
;
3984 zpool_rewind_policy_t policy
;
3985 uint64_t mode
= spa_mode_global
;
3986 uint64_t readonly
= B_FALSE
;
3989 nvlist_t
**spares
, **l2cache
;
3990 uint_t nspares
, nl2cache
;
3993 * If a pool with this name exists, return failure.
3995 mutex_enter(&spa_namespace_lock
);
3996 if (spa_lookup(pool
) != NULL
) {
3997 mutex_exit(&spa_namespace_lock
);
3998 return (SET_ERROR(EEXIST
));
4002 * Create and initialize the spa structure.
4004 (void) nvlist_lookup_string(props
,
4005 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4006 (void) nvlist_lookup_uint64(props
,
4007 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4010 spa
= spa_add(pool
, config
, altroot
);
4011 spa
->spa_import_flags
= flags
;
4014 * Verbatim import - Take a pool and insert it into the namespace
4015 * as if it had been loaded at boot.
4017 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4019 spa_configfile_set(spa
, props
, B_FALSE
);
4021 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4022 spa_event_notify(spa
, NULL
, ESC_ZFS_POOL_IMPORT
);
4024 mutex_exit(&spa_namespace_lock
);
4028 spa_activate(spa
, mode
);
4031 * Don't start async tasks until we know everything is healthy.
4033 spa_async_suspend(spa
);
4035 zpool_get_rewind_policy(config
, &policy
);
4036 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4037 state
= SPA_LOAD_RECOVER
;
4040 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4041 * because the user-supplied config is actually the one to trust when
4044 if (state
!= SPA_LOAD_RECOVER
)
4045 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4047 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4048 policy
.zrp_request
);
4051 * Propagate anything learned while loading the pool and pass it
4052 * back to caller (i.e. rewind info, missing devices, etc).
4054 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4055 spa
->spa_load_info
) == 0);
4057 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4059 * Toss any existing sparelist, as it doesn't have any validity
4060 * anymore, and conflicts with spa_has_spare().
4062 if (spa
->spa_spares
.sav_config
) {
4063 nvlist_free(spa
->spa_spares
.sav_config
);
4064 spa
->spa_spares
.sav_config
= NULL
;
4065 spa_load_spares(spa
);
4067 if (spa
->spa_l2cache
.sav_config
) {
4068 nvlist_free(spa
->spa_l2cache
.sav_config
);
4069 spa
->spa_l2cache
.sav_config
= NULL
;
4070 spa_load_l2cache(spa
);
4073 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4076 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4079 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4080 VDEV_ALLOC_L2CACHE
);
4081 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4084 spa_configfile_set(spa
, props
, B_FALSE
);
4086 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4087 (error
= spa_prop_set(spa
, props
)))) {
4089 spa_deactivate(spa
);
4091 mutex_exit(&spa_namespace_lock
);
4095 spa_async_resume(spa
);
4098 * Override any spares and level 2 cache devices as specified by
4099 * the user, as these may have correct device names/devids, etc.
4101 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4102 &spares
, &nspares
) == 0) {
4103 if (spa
->spa_spares
.sav_config
)
4104 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4105 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4107 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4108 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4109 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4110 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4111 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4112 spa_load_spares(spa
);
4113 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4114 spa
->spa_spares
.sav_sync
= B_TRUE
;
4116 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4117 &l2cache
, &nl2cache
) == 0) {
4118 if (spa
->spa_l2cache
.sav_config
)
4119 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4120 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4122 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4123 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4124 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4125 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4126 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4127 spa_load_l2cache(spa
);
4128 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4129 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4133 * Check for any removed devices.
4135 if (spa
->spa_autoreplace
) {
4136 spa_aux_check_removed(&spa
->spa_spares
);
4137 spa_aux_check_removed(&spa
->spa_l2cache
);
4140 if (spa_writeable(spa
)) {
4142 * Update the config cache to include the newly-imported pool.
4144 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4148 * It's possible that the pool was expanded while it was exported.
4149 * We kick off an async task to handle this for us.
4151 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4153 spa_history_log_version(spa
, "import");
4155 spa_event_notify(spa
, NULL
, ESC_ZFS_POOL_IMPORT
);
4157 zvol_create_minors(spa
, pool
, B_TRUE
);
4159 mutex_exit(&spa_namespace_lock
);
4165 spa_tryimport(nvlist_t
*tryconfig
)
4167 nvlist_t
*config
= NULL
;
4173 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4176 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4180 * Create and initialize the spa structure.
4182 mutex_enter(&spa_namespace_lock
);
4183 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4184 spa_activate(spa
, FREAD
);
4187 * Pass off the heavy lifting to spa_load().
4188 * Pass TRUE for mosconfig because the user-supplied config
4189 * is actually the one to trust when doing an import.
4191 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4194 * If 'tryconfig' was at least parsable, return the current config.
4196 if (spa
->spa_root_vdev
!= NULL
) {
4197 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4198 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4200 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4202 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4203 spa
->spa_uberblock
.ub_timestamp
) == 0);
4204 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4205 spa
->spa_load_info
) == 0);
4206 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4207 spa
->spa_errata
) == 0);
4210 * If the bootfs property exists on this pool then we
4211 * copy it out so that external consumers can tell which
4212 * pools are bootable.
4214 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4215 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4218 * We have to play games with the name since the
4219 * pool was opened as TRYIMPORT_NAME.
4221 if (dsl_dsobj_to_dsname(spa_name(spa
),
4222 spa
->spa_bootfs
, tmpname
) == 0) {
4226 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4228 cp
= strchr(tmpname
, '/');
4230 (void) strlcpy(dsname
, tmpname
,
4233 (void) snprintf(dsname
, MAXPATHLEN
,
4234 "%s/%s", poolname
, ++cp
);
4236 VERIFY(nvlist_add_string(config
,
4237 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4238 kmem_free(dsname
, MAXPATHLEN
);
4240 kmem_free(tmpname
, MAXPATHLEN
);
4244 * Add the list of hot spares and level 2 cache devices.
4246 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4247 spa_add_spares(spa
, config
);
4248 spa_add_l2cache(spa
, config
);
4249 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4253 spa_deactivate(spa
);
4255 mutex_exit(&spa_namespace_lock
);
4261 * Pool export/destroy
4263 * The act of destroying or exporting a pool is very simple. We make sure there
4264 * is no more pending I/O and any references to the pool are gone. Then, we
4265 * update the pool state and sync all the labels to disk, removing the
4266 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4267 * we don't sync the labels or remove the configuration cache.
4270 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4271 boolean_t force
, boolean_t hardforce
)
4278 if (!(spa_mode_global
& FWRITE
))
4279 return (SET_ERROR(EROFS
));
4281 mutex_enter(&spa_namespace_lock
);
4282 if ((spa
= spa_lookup(pool
)) == NULL
) {
4283 mutex_exit(&spa_namespace_lock
);
4284 return (SET_ERROR(ENOENT
));
4288 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4289 * reacquire the namespace lock, and see if we can export.
4291 spa_open_ref(spa
, FTAG
);
4292 mutex_exit(&spa_namespace_lock
);
4293 spa_async_suspend(spa
);
4294 if (spa
->spa_zvol_taskq
) {
4295 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
4296 taskq_wait(spa
->spa_zvol_taskq
);
4298 mutex_enter(&spa_namespace_lock
);
4299 spa_close(spa
, FTAG
);
4301 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4304 * The pool will be in core if it's openable, in which case we can
4305 * modify its state. Objsets may be open only because they're dirty,
4306 * so we have to force it to sync before checking spa_refcnt.
4308 if (spa
->spa_sync_on
) {
4309 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4310 spa_evicting_os_wait(spa
);
4314 * A pool cannot be exported or destroyed if there are active
4315 * references. If we are resetting a pool, allow references by
4316 * fault injection handlers.
4318 if (!spa_refcount_zero(spa
) ||
4319 (spa
->spa_inject_ref
!= 0 &&
4320 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4321 spa_async_resume(spa
);
4322 mutex_exit(&spa_namespace_lock
);
4323 return (SET_ERROR(EBUSY
));
4326 if (spa
->spa_sync_on
) {
4328 * A pool cannot be exported if it has an active shared spare.
4329 * This is to prevent other pools stealing the active spare
4330 * from an exported pool. At user's own will, such pool can
4331 * be forcedly exported.
4333 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4334 spa_has_active_shared_spare(spa
)) {
4335 spa_async_resume(spa
);
4336 mutex_exit(&spa_namespace_lock
);
4337 return (SET_ERROR(EXDEV
));
4341 * We want this to be reflected on every label,
4342 * so mark them all dirty. spa_unload() will do the
4343 * final sync that pushes these changes out.
4345 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4346 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4347 spa
->spa_state
= new_state
;
4348 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4350 vdev_config_dirty(spa
->spa_root_vdev
);
4351 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4356 spa_event_notify(spa
, NULL
, ESC_ZFS_POOL_DESTROY
);
4358 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4360 spa_deactivate(spa
);
4363 if (oldconfig
&& spa
->spa_config
)
4364 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4366 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4368 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4371 mutex_exit(&spa_namespace_lock
);
4377 * Destroy a storage pool.
4380 spa_destroy(char *pool
)
4382 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4387 * Export a storage pool.
4390 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4391 boolean_t hardforce
)
4393 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4398 * Similar to spa_export(), this unloads the spa_t without actually removing it
4399 * from the namespace in any way.
4402 spa_reset(char *pool
)
4404 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4409 * ==========================================================================
4410 * Device manipulation
4411 * ==========================================================================
4415 * Add a device to a storage pool.
4418 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4422 vdev_t
*rvd
= spa
->spa_root_vdev
;
4424 nvlist_t
**spares
, **l2cache
;
4425 uint_t nspares
, nl2cache
;
4428 ASSERT(spa_writeable(spa
));
4430 txg
= spa_vdev_enter(spa
);
4432 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4433 VDEV_ALLOC_ADD
)) != 0)
4434 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4436 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4438 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4442 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4446 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4447 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4449 if (vd
->vdev_children
!= 0 &&
4450 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4451 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4454 * We must validate the spares and l2cache devices after checking the
4455 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4457 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4458 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4461 * Transfer each new top-level vdev from vd to rvd.
4463 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4466 * Set the vdev id to the first hole, if one exists.
4468 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4469 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4470 vdev_free(rvd
->vdev_child
[id
]);
4474 tvd
= vd
->vdev_child
[c
];
4475 vdev_remove_child(vd
, tvd
);
4477 vdev_add_child(rvd
, tvd
);
4478 vdev_config_dirty(tvd
);
4482 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4483 ZPOOL_CONFIG_SPARES
);
4484 spa_load_spares(spa
);
4485 spa
->spa_spares
.sav_sync
= B_TRUE
;
4488 if (nl2cache
!= 0) {
4489 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4490 ZPOOL_CONFIG_L2CACHE
);
4491 spa_load_l2cache(spa
);
4492 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4496 * We have to be careful when adding new vdevs to an existing pool.
4497 * If other threads start allocating from these vdevs before we
4498 * sync the config cache, and we lose power, then upon reboot we may
4499 * fail to open the pool because there are DVAs that the config cache
4500 * can't translate. Therefore, we first add the vdevs without
4501 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4502 * and then let spa_config_update() initialize the new metaslabs.
4504 * spa_load() checks for added-but-not-initialized vdevs, so that
4505 * if we lose power at any point in this sequence, the remaining
4506 * steps will be completed the next time we load the pool.
4508 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4510 mutex_enter(&spa_namespace_lock
);
4511 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4512 spa_event_notify(spa
, NULL
, ESC_ZFS_VDEV_ADD
);
4513 mutex_exit(&spa_namespace_lock
);
4519 * Attach a device to a mirror. The arguments are the path to any device
4520 * in the mirror, and the nvroot for the new device. If the path specifies
4521 * a device that is not mirrored, we automatically insert the mirror vdev.
4523 * If 'replacing' is specified, the new device is intended to replace the
4524 * existing device; in this case the two devices are made into their own
4525 * mirror using the 'replacing' vdev, which is functionally identical to
4526 * the mirror vdev (it actually reuses all the same ops) but has a few
4527 * extra rules: you can't attach to it after it's been created, and upon
4528 * completion of resilvering, the first disk (the one being replaced)
4529 * is automatically detached.
4532 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4534 uint64_t txg
, dtl_max_txg
;
4535 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4537 char *oldvdpath
, *newvdpath
;
4540 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4542 ASSERT(spa_writeable(spa
));
4544 txg
= spa_vdev_enter(spa
);
4546 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4549 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4551 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4552 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4554 pvd
= oldvd
->vdev_parent
;
4556 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4557 VDEV_ALLOC_ATTACH
)) != 0)
4558 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4560 if (newrootvd
->vdev_children
!= 1)
4561 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4563 newvd
= newrootvd
->vdev_child
[0];
4565 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4566 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4568 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4569 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4572 * Spares can't replace logs
4574 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4575 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4579 * For attach, the only allowable parent is a mirror or the root
4582 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4583 pvd
->vdev_ops
!= &vdev_root_ops
)
4584 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4586 pvops
= &vdev_mirror_ops
;
4589 * Active hot spares can only be replaced by inactive hot
4592 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4593 oldvd
->vdev_isspare
&&
4594 !spa_has_spare(spa
, newvd
->vdev_guid
))
4595 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4598 * If the source is a hot spare, and the parent isn't already a
4599 * spare, then we want to create a new hot spare. Otherwise, we
4600 * want to create a replacing vdev. The user is not allowed to
4601 * attach to a spared vdev child unless the 'isspare' state is
4602 * the same (spare replaces spare, non-spare replaces
4605 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4606 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4607 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4608 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4609 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4610 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4613 if (newvd
->vdev_isspare
)
4614 pvops
= &vdev_spare_ops
;
4616 pvops
= &vdev_replacing_ops
;
4620 * Make sure the new device is big enough.
4622 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4623 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4626 * The new device cannot have a higher alignment requirement
4627 * than the top-level vdev.
4629 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4630 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4633 * If this is an in-place replacement, update oldvd's path and devid
4634 * to make it distinguishable from newvd, and unopenable from now on.
4636 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4637 spa_strfree(oldvd
->vdev_path
);
4638 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4640 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4641 newvd
->vdev_path
, "old");
4642 if (oldvd
->vdev_devid
!= NULL
) {
4643 spa_strfree(oldvd
->vdev_devid
);
4644 oldvd
->vdev_devid
= NULL
;
4648 /* mark the device being resilvered */
4649 newvd
->vdev_resilver_txg
= txg
;
4652 * If the parent is not a mirror, or if we're replacing, insert the new
4653 * mirror/replacing/spare vdev above oldvd.
4655 if (pvd
->vdev_ops
!= pvops
)
4656 pvd
= vdev_add_parent(oldvd
, pvops
);
4658 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4659 ASSERT(pvd
->vdev_ops
== pvops
);
4660 ASSERT(oldvd
->vdev_parent
== pvd
);
4663 * Extract the new device from its root and add it to pvd.
4665 vdev_remove_child(newrootvd
, newvd
);
4666 newvd
->vdev_id
= pvd
->vdev_children
;
4667 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4668 vdev_add_child(pvd
, newvd
);
4670 tvd
= newvd
->vdev_top
;
4671 ASSERT(pvd
->vdev_top
== tvd
);
4672 ASSERT(tvd
->vdev_parent
== rvd
);
4674 vdev_config_dirty(tvd
);
4677 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4678 * for any dmu_sync-ed blocks. It will propagate upward when
4679 * spa_vdev_exit() calls vdev_dtl_reassess().
4681 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4683 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4684 dtl_max_txg
- TXG_INITIAL
);
4686 if (newvd
->vdev_isspare
) {
4687 spa_spare_activate(newvd
);
4688 spa_event_notify(spa
, newvd
, ESC_ZFS_VDEV_SPARE
);
4691 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4692 newvdpath
= spa_strdup(newvd
->vdev_path
);
4693 newvd_isspare
= newvd
->vdev_isspare
;
4696 * Mark newvd's DTL dirty in this txg.
4698 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4701 * Schedule the resilver to restart in the future. We do this to
4702 * ensure that dmu_sync-ed blocks have been stitched into the
4703 * respective datasets.
4705 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4707 if (spa
->spa_bootfs
)
4708 spa_event_notify(spa
, newvd
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
4710 spa_event_notify(spa
, newvd
, ESC_ZFS_VDEV_ATTACH
);
4715 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4717 spa_history_log_internal(spa
, "vdev attach", NULL
,
4718 "%s vdev=%s %s vdev=%s",
4719 replacing
&& newvd_isspare
? "spare in" :
4720 replacing
? "replace" : "attach", newvdpath
,
4721 replacing
? "for" : "to", oldvdpath
);
4723 spa_strfree(oldvdpath
);
4724 spa_strfree(newvdpath
);
4730 * Detach a device from a mirror or replacing vdev.
4732 * If 'replace_done' is specified, only detach if the parent
4733 * is a replacing vdev.
4736 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4740 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4741 boolean_t unspare
= B_FALSE
;
4742 uint64_t unspare_guid
= 0;
4745 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4746 ASSERT(spa_writeable(spa
));
4748 txg
= spa_vdev_enter(spa
);
4750 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4753 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4755 if (!vd
->vdev_ops
->vdev_op_leaf
)
4756 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4758 pvd
= vd
->vdev_parent
;
4761 * If the parent/child relationship is not as expected, don't do it.
4762 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4763 * vdev that's replacing B with C. The user's intent in replacing
4764 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4765 * the replace by detaching C, the expected behavior is to end up
4766 * M(A,B). But suppose that right after deciding to detach C,
4767 * the replacement of B completes. We would have M(A,C), and then
4768 * ask to detach C, which would leave us with just A -- not what
4769 * the user wanted. To prevent this, we make sure that the
4770 * parent/child relationship hasn't changed -- in this example,
4771 * that C's parent is still the replacing vdev R.
4773 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4774 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4777 * Only 'replacing' or 'spare' vdevs can be replaced.
4779 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4780 pvd
->vdev_ops
!= &vdev_spare_ops
)
4781 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4783 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4784 spa_version(spa
) >= SPA_VERSION_SPARES
);
4787 * Only mirror, replacing, and spare vdevs support detach.
4789 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4790 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4791 pvd
->vdev_ops
!= &vdev_spare_ops
)
4792 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4795 * If this device has the only valid copy of some data,
4796 * we cannot safely detach it.
4798 if (vdev_dtl_required(vd
))
4799 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4801 ASSERT(pvd
->vdev_children
>= 2);
4804 * If we are detaching the second disk from a replacing vdev, then
4805 * check to see if we changed the original vdev's path to have "/old"
4806 * at the end in spa_vdev_attach(). If so, undo that change now.
4808 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4809 vd
->vdev_path
!= NULL
) {
4810 size_t len
= strlen(vd
->vdev_path
);
4812 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4813 cvd
= pvd
->vdev_child
[c
];
4815 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4818 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4819 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4820 spa_strfree(cvd
->vdev_path
);
4821 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4828 * If we are detaching the original disk from a spare, then it implies
4829 * that the spare should become a real disk, and be removed from the
4830 * active spare list for the pool.
4832 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4834 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4838 * Erase the disk labels so the disk can be used for other things.
4839 * This must be done after all other error cases are handled,
4840 * but before we disembowel vd (so we can still do I/O to it).
4841 * But if we can't do it, don't treat the error as fatal --
4842 * it may be that the unwritability of the disk is the reason
4843 * it's being detached!
4845 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4848 * Remove vd from its parent and compact the parent's children.
4850 vdev_remove_child(pvd
, vd
);
4851 vdev_compact_children(pvd
);
4854 * Remember one of the remaining children so we can get tvd below.
4856 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4859 * If we need to remove the remaining child from the list of hot spares,
4860 * do it now, marking the vdev as no longer a spare in the process.
4861 * We must do this before vdev_remove_parent(), because that can
4862 * change the GUID if it creates a new toplevel GUID. For a similar
4863 * reason, we must remove the spare now, in the same txg as the detach;
4864 * otherwise someone could attach a new sibling, change the GUID, and
4865 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4868 ASSERT(cvd
->vdev_isspare
);
4869 spa_spare_remove(cvd
);
4870 unspare_guid
= cvd
->vdev_guid
;
4871 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4872 cvd
->vdev_unspare
= B_TRUE
;
4876 * If the parent mirror/replacing vdev only has one child,
4877 * the parent is no longer needed. Remove it from the tree.
4879 if (pvd
->vdev_children
== 1) {
4880 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4881 cvd
->vdev_unspare
= B_FALSE
;
4882 vdev_remove_parent(cvd
);
4887 * We don't set tvd until now because the parent we just removed
4888 * may have been the previous top-level vdev.
4890 tvd
= cvd
->vdev_top
;
4891 ASSERT(tvd
->vdev_parent
== rvd
);
4894 * Reevaluate the parent vdev state.
4896 vdev_propagate_state(cvd
);
4899 * If the 'autoexpand' property is set on the pool then automatically
4900 * try to expand the size of the pool. For example if the device we
4901 * just detached was smaller than the others, it may be possible to
4902 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4903 * first so that we can obtain the updated sizes of the leaf vdevs.
4905 if (spa
->spa_autoexpand
) {
4907 vdev_expand(tvd
, txg
);
4910 vdev_config_dirty(tvd
);
4913 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4914 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4915 * But first make sure we're not on any *other* txg's DTL list, to
4916 * prevent vd from being accessed after it's freed.
4918 vdpath
= spa_strdup(vd
->vdev_path
);
4919 for (t
= 0; t
< TXG_SIZE
; t
++)
4920 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4921 vd
->vdev_detached
= B_TRUE
;
4922 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4924 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_REMOVE
);
4926 /* hang on to the spa before we release the lock */
4927 spa_open_ref(spa
, FTAG
);
4929 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4931 spa_history_log_internal(spa
, "detach", NULL
,
4933 spa_strfree(vdpath
);
4936 * If this was the removal of the original device in a hot spare vdev,
4937 * then we want to go through and remove the device from the hot spare
4938 * list of every other pool.
4941 spa_t
*altspa
= NULL
;
4943 mutex_enter(&spa_namespace_lock
);
4944 while ((altspa
= spa_next(altspa
)) != NULL
) {
4945 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4949 spa_open_ref(altspa
, FTAG
);
4950 mutex_exit(&spa_namespace_lock
);
4951 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4952 mutex_enter(&spa_namespace_lock
);
4953 spa_close(altspa
, FTAG
);
4955 mutex_exit(&spa_namespace_lock
);
4957 /* search the rest of the vdevs for spares to remove */
4958 spa_vdev_resilver_done(spa
);
4961 /* all done with the spa; OK to release */
4962 mutex_enter(&spa_namespace_lock
);
4963 spa_close(spa
, FTAG
);
4964 mutex_exit(&spa_namespace_lock
);
4970 * Split a set of devices from their mirrors, and create a new pool from them.
4973 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
4974 nvlist_t
*props
, boolean_t exp
)
4977 uint64_t txg
, *glist
;
4979 uint_t c
, children
, lastlog
;
4980 nvlist_t
**child
, *nvl
, *tmp
;
4982 char *altroot
= NULL
;
4983 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
4984 boolean_t activate_slog
;
4986 ASSERT(spa_writeable(spa
));
4988 txg
= spa_vdev_enter(spa
);
4990 /* clear the log and flush everything up to now */
4991 activate_slog
= spa_passivate_log(spa
);
4992 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4993 error
= spa_offline_log(spa
);
4994 txg
= spa_vdev_config_enter(spa
);
4997 spa_activate_log(spa
);
5000 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5002 /* check new spa name before going any further */
5003 if (spa_lookup(newname
) != NULL
)
5004 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5007 * scan through all the children to ensure they're all mirrors
5009 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5010 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5012 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5014 /* first, check to ensure we've got the right child count */
5015 rvd
= spa
->spa_root_vdev
;
5017 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5018 vdev_t
*vd
= rvd
->vdev_child
[c
];
5020 /* don't count the holes & logs as children */
5021 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5029 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5030 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5032 /* next, ensure no spare or cache devices are part of the split */
5033 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5034 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5035 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5037 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5038 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5040 /* then, loop over each vdev and validate it */
5041 for (c
= 0; c
< children
; c
++) {
5042 uint64_t is_hole
= 0;
5044 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5048 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5049 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5052 error
= SET_ERROR(EINVAL
);
5057 /* which disk is going to be split? */
5058 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5060 error
= SET_ERROR(EINVAL
);
5064 /* look it up in the spa */
5065 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5066 if (vml
[c
] == NULL
) {
5067 error
= SET_ERROR(ENODEV
);
5071 /* make sure there's nothing stopping the split */
5072 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5073 vml
[c
]->vdev_islog
||
5074 vml
[c
]->vdev_ishole
||
5075 vml
[c
]->vdev_isspare
||
5076 vml
[c
]->vdev_isl2cache
||
5077 !vdev_writeable(vml
[c
]) ||
5078 vml
[c
]->vdev_children
!= 0 ||
5079 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5080 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5081 error
= SET_ERROR(EINVAL
);
5085 if (vdev_dtl_required(vml
[c
])) {
5086 error
= SET_ERROR(EBUSY
);
5090 /* we need certain info from the top level */
5091 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5092 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5093 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5094 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5095 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5096 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5097 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5098 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5100 /* transfer per-vdev ZAPs */
5101 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
5102 VERIFY0(nvlist_add_uint64(child
[c
],
5103 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
5105 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
5106 VERIFY0(nvlist_add_uint64(child
[c
],
5107 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
5108 vml
[c
]->vdev_parent
->vdev_top_zap
));
5112 kmem_free(vml
, children
* sizeof (vdev_t
*));
5113 kmem_free(glist
, children
* sizeof (uint64_t));
5114 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5117 /* stop writers from using the disks */
5118 for (c
= 0; c
< children
; c
++) {
5120 vml
[c
]->vdev_offline
= B_TRUE
;
5122 vdev_reopen(spa
->spa_root_vdev
);
5125 * Temporarily record the splitting vdevs in the spa config. This
5126 * will disappear once the config is regenerated.
5128 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5129 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5130 glist
, children
) == 0);
5131 kmem_free(glist
, children
* sizeof (uint64_t));
5133 mutex_enter(&spa
->spa_props_lock
);
5134 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5136 mutex_exit(&spa
->spa_props_lock
);
5137 spa
->spa_config_splitting
= nvl
;
5138 vdev_config_dirty(spa
->spa_root_vdev
);
5140 /* configure and create the new pool */
5141 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5142 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5143 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5144 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5145 spa_version(spa
)) == 0);
5146 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5147 spa
->spa_config_txg
) == 0);
5148 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5149 spa_generate_guid(NULL
)) == 0);
5150 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
5151 (void) nvlist_lookup_string(props
,
5152 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5154 /* add the new pool to the namespace */
5155 newspa
= spa_add(newname
, config
, altroot
);
5156 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5157 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5158 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5160 /* release the spa config lock, retaining the namespace lock */
5161 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5163 if (zio_injection_enabled
)
5164 zio_handle_panic_injection(spa
, FTAG
, 1);
5166 spa_activate(newspa
, spa_mode_global
);
5167 spa_async_suspend(newspa
);
5169 /* create the new pool from the disks of the original pool */
5170 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5174 /* if that worked, generate a real config for the new pool */
5175 if (newspa
->spa_root_vdev
!= NULL
) {
5176 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5177 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5178 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5179 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5180 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5185 if (props
!= NULL
) {
5186 spa_configfile_set(newspa
, props
, B_FALSE
);
5187 error
= spa_prop_set(newspa
, props
);
5192 /* flush everything */
5193 txg
= spa_vdev_config_enter(newspa
);
5194 vdev_config_dirty(newspa
->spa_root_vdev
);
5195 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5197 if (zio_injection_enabled
)
5198 zio_handle_panic_injection(spa
, FTAG
, 2);
5200 spa_async_resume(newspa
);
5202 /* finally, update the original pool's config */
5203 txg
= spa_vdev_config_enter(spa
);
5204 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5205 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5208 for (c
= 0; c
< children
; c
++) {
5209 if (vml
[c
] != NULL
) {
5212 spa_history_log_internal(spa
, "detach", tx
,
5213 "vdev=%s", vml
[c
]->vdev_path
);
5218 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5219 vdev_config_dirty(spa
->spa_root_vdev
);
5220 spa
->spa_config_splitting
= NULL
;
5224 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5226 if (zio_injection_enabled
)
5227 zio_handle_panic_injection(spa
, FTAG
, 3);
5229 /* split is complete; log a history record */
5230 spa_history_log_internal(newspa
, "split", NULL
,
5231 "from pool %s", spa_name(spa
));
5233 kmem_free(vml
, children
* sizeof (vdev_t
*));
5235 /* if we're not going to mount the filesystems in userland, export */
5237 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5244 spa_deactivate(newspa
);
5247 txg
= spa_vdev_config_enter(spa
);
5249 /* re-online all offlined disks */
5250 for (c
= 0; c
< children
; c
++) {
5252 vml
[c
]->vdev_offline
= B_FALSE
;
5254 vdev_reopen(spa
->spa_root_vdev
);
5256 nvlist_free(spa
->spa_config_splitting
);
5257 spa
->spa_config_splitting
= NULL
;
5258 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5260 kmem_free(vml
, children
* sizeof (vdev_t
*));
5265 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5269 for (i
= 0; i
< count
; i
++) {
5272 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5275 if (guid
== target_guid
)
5283 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5284 nvlist_t
*dev_to_remove
)
5286 nvlist_t
**newdev
= NULL
;
5290 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5292 for (i
= 0, j
= 0; i
< count
; i
++) {
5293 if (dev
[i
] == dev_to_remove
)
5295 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5298 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5299 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5301 for (i
= 0; i
< count
- 1; i
++)
5302 nvlist_free(newdev
[i
]);
5305 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5309 * Evacuate the device.
5312 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5317 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5318 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5319 ASSERT(vd
== vd
->vdev_top
);
5322 * Evacuate the device. We don't hold the config lock as writer
5323 * since we need to do I/O but we do keep the
5324 * spa_namespace_lock held. Once this completes the device
5325 * should no longer have any blocks allocated on it.
5327 if (vd
->vdev_islog
) {
5328 if (vd
->vdev_stat
.vs_alloc
!= 0)
5329 error
= spa_offline_log(spa
);
5331 error
= SET_ERROR(ENOTSUP
);
5338 * The evacuation succeeded. Remove any remaining MOS metadata
5339 * associated with this vdev, and wait for these changes to sync.
5341 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5342 txg
= spa_vdev_config_enter(spa
);
5343 vd
->vdev_removing
= B_TRUE
;
5344 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5345 vdev_config_dirty(vd
);
5346 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5352 * Complete the removal by cleaning up the namespace.
5355 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5357 vdev_t
*rvd
= spa
->spa_root_vdev
;
5358 uint64_t id
= vd
->vdev_id
;
5359 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5361 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5362 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5363 ASSERT(vd
== vd
->vdev_top
);
5366 * Only remove any devices which are empty.
5368 if (vd
->vdev_stat
.vs_alloc
!= 0)
5371 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5373 if (list_link_active(&vd
->vdev_state_dirty_node
))
5374 vdev_state_clean(vd
);
5375 if (list_link_active(&vd
->vdev_config_dirty_node
))
5376 vdev_config_clean(vd
);
5381 vdev_compact_children(rvd
);
5383 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5384 vdev_add_child(rvd
, vd
);
5386 vdev_config_dirty(rvd
);
5389 * Reassess the health of our root vdev.
5395 * Remove a device from the pool -
5397 * Removing a device from the vdev namespace requires several steps
5398 * and can take a significant amount of time. As a result we use
5399 * the spa_vdev_config_[enter/exit] functions which allow us to
5400 * grab and release the spa_config_lock while still holding the namespace
5401 * lock. During each step the configuration is synced out.
5403 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5407 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5410 metaslab_group_t
*mg
;
5411 nvlist_t
**spares
, **l2cache
, *nv
;
5413 uint_t nspares
, nl2cache
;
5415 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5417 ASSERT(spa_writeable(spa
));
5420 txg
= spa_vdev_enter(spa
);
5422 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5424 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5425 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5426 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5427 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5429 * Only remove the hot spare if it's not currently in use
5432 if (vd
== NULL
|| unspare
) {
5433 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5434 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5435 spa_load_spares(spa
);
5436 spa
->spa_spares
.sav_sync
= B_TRUE
;
5438 error
= SET_ERROR(EBUSY
);
5440 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_REMOVE_AUX
);
5441 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5442 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5443 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5444 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5446 * Cache devices can always be removed.
5448 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5449 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5450 spa_load_l2cache(spa
);
5451 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5452 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_REMOVE_AUX
);
5453 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5455 ASSERT(vd
== vd
->vdev_top
);
5460 * Stop allocating from this vdev.
5462 metaslab_group_passivate(mg
);
5465 * Wait for the youngest allocations and frees to sync,
5466 * and then wait for the deferral of those frees to finish.
5468 spa_vdev_config_exit(spa
, NULL
,
5469 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5472 * Attempt to evacuate the vdev.
5474 error
= spa_vdev_remove_evacuate(spa
, vd
);
5476 txg
= spa_vdev_config_enter(spa
);
5479 * If we couldn't evacuate the vdev, unwind.
5482 metaslab_group_activate(mg
);
5483 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5487 * Clean up the vdev namespace.
5489 spa_vdev_remove_from_namespace(spa
, vd
);
5491 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_REMOVE_DEV
);
5492 } else if (vd
!= NULL
) {
5494 * Normal vdevs cannot be removed (yet).
5496 error
= SET_ERROR(ENOTSUP
);
5499 * There is no vdev of any kind with the specified guid.
5501 error
= SET_ERROR(ENOENT
);
5505 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5511 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5512 * currently spared, so we can detach it.
5515 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5517 vdev_t
*newvd
, *oldvd
;
5520 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5521 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5527 * Check for a completed replacement. We always consider the first
5528 * vdev in the list to be the oldest vdev, and the last one to be
5529 * the newest (see spa_vdev_attach() for how that works). In
5530 * the case where the newest vdev is faulted, we will not automatically
5531 * remove it after a resilver completes. This is OK as it will require
5532 * user intervention to determine which disk the admin wishes to keep.
5534 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5535 ASSERT(vd
->vdev_children
> 1);
5537 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5538 oldvd
= vd
->vdev_child
[0];
5540 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5541 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5542 !vdev_dtl_required(oldvd
))
5547 * Check for a completed resilver with the 'unspare' flag set.
5549 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5550 vdev_t
*first
= vd
->vdev_child
[0];
5551 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5553 if (last
->vdev_unspare
) {
5556 } else if (first
->vdev_unspare
) {
5563 if (oldvd
!= NULL
&&
5564 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5565 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5566 !vdev_dtl_required(oldvd
))
5570 * If there are more than two spares attached to a disk,
5571 * and those spares are not required, then we want to
5572 * attempt to free them up now so that they can be used
5573 * by other pools. Once we're back down to a single
5574 * disk+spare, we stop removing them.
5576 if (vd
->vdev_children
> 2) {
5577 newvd
= vd
->vdev_child
[1];
5579 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5580 vdev_dtl_empty(last
, DTL_MISSING
) &&
5581 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5582 !vdev_dtl_required(newvd
))
5591 spa_vdev_resilver_done(spa_t
*spa
)
5593 vdev_t
*vd
, *pvd
, *ppvd
;
5594 uint64_t guid
, sguid
, pguid
, ppguid
;
5596 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5598 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5599 pvd
= vd
->vdev_parent
;
5600 ppvd
= pvd
->vdev_parent
;
5601 guid
= vd
->vdev_guid
;
5602 pguid
= pvd
->vdev_guid
;
5603 ppguid
= ppvd
->vdev_guid
;
5606 * If we have just finished replacing a hot spared device, then
5607 * we need to detach the parent's first child (the original hot
5610 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5611 ppvd
->vdev_children
== 2) {
5612 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5613 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5615 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5617 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5618 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5620 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5622 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5625 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5629 * Update the stored path or FRU for this vdev.
5632 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5636 boolean_t sync
= B_FALSE
;
5638 ASSERT(spa_writeable(spa
));
5640 spa_vdev_state_enter(spa
, SCL_ALL
);
5642 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5643 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5645 if (!vd
->vdev_ops
->vdev_op_leaf
)
5646 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5649 if (strcmp(value
, vd
->vdev_path
) != 0) {
5650 spa_strfree(vd
->vdev_path
);
5651 vd
->vdev_path
= spa_strdup(value
);
5655 if (vd
->vdev_fru
== NULL
) {
5656 vd
->vdev_fru
= spa_strdup(value
);
5658 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5659 spa_strfree(vd
->vdev_fru
);
5660 vd
->vdev_fru
= spa_strdup(value
);
5665 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5669 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5671 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5675 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5677 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5681 * ==========================================================================
5683 * ==========================================================================
5687 spa_scan_stop(spa_t
*spa
)
5689 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5690 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5691 return (SET_ERROR(EBUSY
));
5692 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5696 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5698 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5700 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5701 return (SET_ERROR(ENOTSUP
));
5704 * If a resilver was requested, but there is no DTL on a
5705 * writeable leaf device, we have nothing to do.
5707 if (func
== POOL_SCAN_RESILVER
&&
5708 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5709 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5713 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5717 * ==========================================================================
5718 * SPA async task processing
5719 * ==========================================================================
5723 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5727 if (vd
->vdev_remove_wanted
) {
5728 vd
->vdev_remove_wanted
= B_FALSE
;
5729 vd
->vdev_delayed_close
= B_FALSE
;
5730 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5733 * We want to clear the stats, but we don't want to do a full
5734 * vdev_clear() as that will cause us to throw away
5735 * degraded/faulted state as well as attempt to reopen the
5736 * device, all of which is a waste.
5738 vd
->vdev_stat
.vs_read_errors
= 0;
5739 vd
->vdev_stat
.vs_write_errors
= 0;
5740 vd
->vdev_stat
.vs_checksum_errors
= 0;
5742 vdev_state_dirty(vd
->vdev_top
);
5745 for (c
= 0; c
< vd
->vdev_children
; c
++)
5746 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5750 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5754 if (vd
->vdev_probe_wanted
) {
5755 vd
->vdev_probe_wanted
= B_FALSE
;
5756 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5759 for (c
= 0; c
< vd
->vdev_children
; c
++)
5760 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5764 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5768 if (!spa
->spa_autoexpand
)
5771 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5772 vdev_t
*cvd
= vd
->vdev_child
[c
];
5773 spa_async_autoexpand(spa
, cvd
);
5776 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5779 spa_event_notify(vd
->vdev_spa
, vd
, ESC_ZFS_VDEV_AUTOEXPAND
);
5783 spa_async_thread(spa_t
*spa
)
5787 ASSERT(spa
->spa_sync_on
);
5789 mutex_enter(&spa
->spa_async_lock
);
5790 tasks
= spa
->spa_async_tasks
;
5791 spa
->spa_async_tasks
= 0;
5792 mutex_exit(&spa
->spa_async_lock
);
5795 * See if the config needs to be updated.
5797 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5798 uint64_t old_space
, new_space
;
5800 mutex_enter(&spa_namespace_lock
);
5801 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5802 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5803 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5804 mutex_exit(&spa_namespace_lock
);
5807 * If the pool grew as a result of the config update,
5808 * then log an internal history event.
5810 if (new_space
!= old_space
) {
5811 spa_history_log_internal(spa
, "vdev online", NULL
,
5812 "pool '%s' size: %llu(+%llu)",
5813 spa_name(spa
), new_space
, new_space
- old_space
);
5818 * See if any devices need to be marked REMOVED.
5820 if (tasks
& SPA_ASYNC_REMOVE
) {
5821 spa_vdev_state_enter(spa
, SCL_NONE
);
5822 spa_async_remove(spa
, spa
->spa_root_vdev
);
5823 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5824 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5825 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5826 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5827 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5830 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5831 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5832 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5833 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5837 * See if any devices need to be probed.
5839 if (tasks
& SPA_ASYNC_PROBE
) {
5840 spa_vdev_state_enter(spa
, SCL_NONE
);
5841 spa_async_probe(spa
, spa
->spa_root_vdev
);
5842 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5846 * If any devices are done replacing, detach them.
5848 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5849 spa_vdev_resilver_done(spa
);
5852 * Kick off a resilver.
5854 if (tasks
& SPA_ASYNC_RESILVER
)
5855 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5858 * Let the world know that we're done.
5860 mutex_enter(&spa
->spa_async_lock
);
5861 spa
->spa_async_thread
= NULL
;
5862 cv_broadcast(&spa
->spa_async_cv
);
5863 mutex_exit(&spa
->spa_async_lock
);
5868 spa_async_suspend(spa_t
*spa
)
5870 mutex_enter(&spa
->spa_async_lock
);
5871 spa
->spa_async_suspended
++;
5872 while (spa
->spa_async_thread
!= NULL
)
5873 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5874 mutex_exit(&spa
->spa_async_lock
);
5878 spa_async_resume(spa_t
*spa
)
5880 mutex_enter(&spa
->spa_async_lock
);
5881 ASSERT(spa
->spa_async_suspended
!= 0);
5882 spa
->spa_async_suspended
--;
5883 mutex_exit(&spa
->spa_async_lock
);
5887 spa_async_tasks_pending(spa_t
*spa
)
5889 uint_t non_config_tasks
;
5891 boolean_t config_task_suspended
;
5893 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
5894 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
5895 if (spa
->spa_ccw_fail_time
== 0) {
5896 config_task_suspended
= B_FALSE
;
5898 config_task_suspended
=
5899 (gethrtime() - spa
->spa_ccw_fail_time
) <
5900 (zfs_ccw_retry_interval
* NANOSEC
);
5903 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
5907 spa_async_dispatch(spa_t
*spa
)
5909 mutex_enter(&spa
->spa_async_lock
);
5910 if (spa_async_tasks_pending(spa
) &&
5911 !spa
->spa_async_suspended
&&
5912 spa
->spa_async_thread
== NULL
&&
5914 spa
->spa_async_thread
= thread_create(NULL
, 0,
5915 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5916 mutex_exit(&spa
->spa_async_lock
);
5920 spa_async_request(spa_t
*spa
, int task
)
5922 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5923 mutex_enter(&spa
->spa_async_lock
);
5924 spa
->spa_async_tasks
|= task
;
5925 mutex_exit(&spa
->spa_async_lock
);
5929 * ==========================================================================
5930 * SPA syncing routines
5931 * ==========================================================================
5935 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5938 bpobj_enqueue(bpo
, bp
, tx
);
5943 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5947 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5953 * Note: this simple function is not inlined to make it easier to dtrace the
5954 * amount of time spent syncing frees.
5957 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
5959 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5960 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
5961 VERIFY(zio_wait(zio
) == 0);
5965 * Note: this simple function is not inlined to make it easier to dtrace the
5966 * amount of time spent syncing deferred frees.
5969 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
5971 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5972 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
5973 spa_free_sync_cb
, zio
, tx
), ==, 0);
5974 VERIFY0(zio_wait(zio
));
5978 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5980 char *packed
= NULL
;
5985 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5988 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5989 * information. This avoids the dmu_buf_will_dirty() path and
5990 * saves us a pre-read to get data we don't actually care about.
5992 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
5993 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
5995 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
5997 bzero(packed
+ nvsize
, bufsize
- nvsize
);
5999 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6001 vmem_free(packed
, bufsize
);
6003 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6004 dmu_buf_will_dirty(db
, tx
);
6005 *(uint64_t *)db
->db_data
= nvsize
;
6006 dmu_buf_rele(db
, FTAG
);
6010 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6011 const char *config
, const char *entry
)
6021 * Update the MOS nvlist describing the list of available devices.
6022 * spa_validate_aux() will have already made sure this nvlist is
6023 * valid and the vdevs are labeled appropriately.
6025 if (sav
->sav_object
== 0) {
6026 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6027 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6028 sizeof (uint64_t), tx
);
6029 VERIFY(zap_update(spa
->spa_meta_objset
,
6030 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6031 &sav
->sav_object
, tx
) == 0);
6034 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6035 if (sav
->sav_count
== 0) {
6036 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6038 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6039 for (i
= 0; i
< sav
->sav_count
; i
++)
6040 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6041 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6042 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6043 sav
->sav_count
) == 0);
6044 for (i
= 0; i
< sav
->sav_count
; i
++)
6045 nvlist_free(list
[i
]);
6046 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6049 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6050 nvlist_free(nvroot
);
6052 sav
->sav_sync
= B_FALSE
;
6056 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6057 * The all-vdev ZAP must be empty.
6060 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6062 spa_t
*spa
= vd
->vdev_spa
;
6065 if (vd
->vdev_top_zap
!= 0) {
6066 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6067 vd
->vdev_top_zap
, tx
));
6069 if (vd
->vdev_leaf_zap
!= 0) {
6070 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6071 vd
->vdev_leaf_zap
, tx
));
6073 for (i
= 0; i
< vd
->vdev_children
; i
++) {
6074 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6079 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6084 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6085 * its config may not be dirty but we still need to build per-vdev ZAPs.
6086 * Similarly, if the pool is being assembled (e.g. after a split), we
6087 * need to rebuild the AVZ although the config may not be dirty.
6089 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6090 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6093 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6095 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6096 spa
->spa_all_vdev_zaps
!= 0);
6098 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6102 /* Make and build the new AVZ */
6103 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6104 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6105 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6107 /* Diff old AVZ with new one */
6108 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6109 spa
->spa_all_vdev_zaps
);
6110 zap_cursor_retrieve(&zc
, &za
) == 0;
6111 zap_cursor_advance(&zc
)) {
6112 uint64_t vdzap
= za
.za_first_integer
;
6113 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6116 * ZAP is listed in old AVZ but not in new one;
6119 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6124 zap_cursor_fini(&zc
);
6126 /* Destroy the old AVZ */
6127 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6128 spa
->spa_all_vdev_zaps
, tx
));
6130 /* Replace the old AVZ in the dir obj with the new one */
6131 VERIFY0(zap_update(spa
->spa_meta_objset
,
6132 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6133 sizeof (new_avz
), 1, &new_avz
, tx
));
6135 spa
->spa_all_vdev_zaps
= new_avz
;
6136 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6140 /* Walk through the AVZ and destroy all listed ZAPs */
6141 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6142 spa
->spa_all_vdev_zaps
);
6143 zap_cursor_retrieve(&zc
, &za
) == 0;
6144 zap_cursor_advance(&zc
)) {
6145 uint64_t zap
= za
.za_first_integer
;
6146 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6149 zap_cursor_fini(&zc
);
6151 /* Destroy and unlink the AVZ itself */
6152 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6153 spa
->spa_all_vdev_zaps
, tx
));
6154 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6155 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6156 spa
->spa_all_vdev_zaps
= 0;
6159 if (spa
->spa_all_vdev_zaps
== 0) {
6160 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6161 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6162 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6164 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6166 /* Create ZAPs for vdevs that don't have them. */
6167 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6169 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6170 dmu_tx_get_txg(tx
), B_FALSE
);
6173 * If we're upgrading the spa version then make sure that
6174 * the config object gets updated with the correct version.
6176 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6177 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6178 spa
->spa_uberblock
.ub_version
);
6180 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6182 nvlist_free(spa
->spa_config_syncing
);
6183 spa
->spa_config_syncing
= config
;
6185 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6189 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6191 uint64_t *versionp
= arg
;
6192 uint64_t version
= *versionp
;
6193 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6196 * Setting the version is special cased when first creating the pool.
6198 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6200 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6201 ASSERT(version
>= spa_version(spa
));
6203 spa
->spa_uberblock
.ub_version
= version
;
6204 vdev_config_dirty(spa
->spa_root_vdev
);
6205 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6209 * Set zpool properties.
6212 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6214 nvlist_t
*nvp
= arg
;
6215 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6216 objset_t
*mos
= spa
->spa_meta_objset
;
6217 nvpair_t
*elem
= NULL
;
6219 mutex_enter(&spa
->spa_props_lock
);
6221 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6223 char *strval
, *fname
;
6225 const char *propname
;
6226 zprop_type_t proptype
;
6229 prop
= zpool_name_to_prop(nvpair_name(elem
));
6230 switch ((int)prop
) {
6233 * We checked this earlier in spa_prop_validate().
6235 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6237 fname
= strchr(nvpair_name(elem
), '@') + 1;
6238 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6240 spa_feature_enable(spa
, fid
, tx
);
6241 spa_history_log_internal(spa
, "set", tx
,
6242 "%s=enabled", nvpair_name(elem
));
6245 case ZPOOL_PROP_VERSION
:
6246 intval
= fnvpair_value_uint64(elem
);
6248 * The version is synced seperatly before other
6249 * properties and should be correct by now.
6251 ASSERT3U(spa_version(spa
), >=, intval
);
6254 case ZPOOL_PROP_ALTROOT
:
6256 * 'altroot' is a non-persistent property. It should
6257 * have been set temporarily at creation or import time.
6259 ASSERT(spa
->spa_root
!= NULL
);
6262 case ZPOOL_PROP_READONLY
:
6263 case ZPOOL_PROP_CACHEFILE
:
6265 * 'readonly' and 'cachefile' are also non-persisitent
6269 case ZPOOL_PROP_COMMENT
:
6270 strval
= fnvpair_value_string(elem
);
6271 if (spa
->spa_comment
!= NULL
)
6272 spa_strfree(spa
->spa_comment
);
6273 spa
->spa_comment
= spa_strdup(strval
);
6275 * We need to dirty the configuration on all the vdevs
6276 * so that their labels get updated. It's unnecessary
6277 * to do this for pool creation since the vdev's
6278 * configuratoin has already been dirtied.
6280 if (tx
->tx_txg
!= TXG_INITIAL
)
6281 vdev_config_dirty(spa
->spa_root_vdev
);
6282 spa_history_log_internal(spa
, "set", tx
,
6283 "%s=%s", nvpair_name(elem
), strval
);
6287 * Set pool property values in the poolprops mos object.
6289 if (spa
->spa_pool_props_object
== 0) {
6290 spa
->spa_pool_props_object
=
6291 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6292 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6296 /* normalize the property name */
6297 propname
= zpool_prop_to_name(prop
);
6298 proptype
= zpool_prop_get_type(prop
);
6300 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6301 ASSERT(proptype
== PROP_TYPE_STRING
);
6302 strval
= fnvpair_value_string(elem
);
6303 VERIFY0(zap_update(mos
,
6304 spa
->spa_pool_props_object
, propname
,
6305 1, strlen(strval
) + 1, strval
, tx
));
6306 spa_history_log_internal(spa
, "set", tx
,
6307 "%s=%s", nvpair_name(elem
), strval
);
6308 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6309 intval
= fnvpair_value_uint64(elem
);
6311 if (proptype
== PROP_TYPE_INDEX
) {
6313 VERIFY0(zpool_prop_index_to_string(
6314 prop
, intval
, &unused
));
6316 VERIFY0(zap_update(mos
,
6317 spa
->spa_pool_props_object
, propname
,
6318 8, 1, &intval
, tx
));
6319 spa_history_log_internal(spa
, "set", tx
,
6320 "%s=%lld", nvpair_name(elem
), intval
);
6322 ASSERT(0); /* not allowed */
6326 case ZPOOL_PROP_DELEGATION
:
6327 spa
->spa_delegation
= intval
;
6329 case ZPOOL_PROP_BOOTFS
:
6330 spa
->spa_bootfs
= intval
;
6332 case ZPOOL_PROP_FAILUREMODE
:
6333 spa
->spa_failmode
= intval
;
6335 case ZPOOL_PROP_AUTOEXPAND
:
6336 spa
->spa_autoexpand
= intval
;
6337 if (tx
->tx_txg
!= TXG_INITIAL
)
6338 spa_async_request(spa
,
6339 SPA_ASYNC_AUTOEXPAND
);
6341 case ZPOOL_PROP_DEDUPDITTO
:
6342 spa
->spa_dedup_ditto
= intval
;
6351 mutex_exit(&spa
->spa_props_lock
);
6355 * Perform one-time upgrade on-disk changes. spa_version() does not
6356 * reflect the new version this txg, so there must be no changes this
6357 * txg to anything that the upgrade code depends on after it executes.
6358 * Therefore this must be called after dsl_pool_sync() does the sync
6362 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6364 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6366 ASSERT(spa
->spa_sync_pass
== 1);
6368 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6370 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6371 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6372 dsl_pool_create_origin(dp
, tx
);
6374 /* Keeping the origin open increases spa_minref */
6375 spa
->spa_minref
+= 3;
6378 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6379 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6380 dsl_pool_upgrade_clones(dp
, tx
);
6383 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6384 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6385 dsl_pool_upgrade_dir_clones(dp
, tx
);
6387 /* Keeping the freedir open increases spa_minref */
6388 spa
->spa_minref
+= 3;
6391 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6392 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6393 spa_feature_create_zap_objects(spa
, tx
);
6397 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6398 * when possibility to use lz4 compression for metadata was added
6399 * Old pools that have this feature enabled must be upgraded to have
6400 * this feature active
6402 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6403 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6404 SPA_FEATURE_LZ4_COMPRESS
);
6405 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6406 SPA_FEATURE_LZ4_COMPRESS
);
6408 if (lz4_en
&& !lz4_ac
)
6409 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6411 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6415 * Sync the specified transaction group. New blocks may be dirtied as
6416 * part of the process, so we iterate until it converges.
6419 spa_sync(spa_t
*spa
, uint64_t txg
)
6421 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6422 objset_t
*mos
= spa
->spa_meta_objset
;
6423 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6424 vdev_t
*rvd
= spa
->spa_root_vdev
;
6430 VERIFY(spa_writeable(spa
));
6433 * Lock out configuration changes.
6435 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6437 spa
->spa_syncing_txg
= txg
;
6438 spa
->spa_sync_pass
= 0;
6441 * If there are any pending vdev state changes, convert them
6442 * into config changes that go out with this transaction group.
6444 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6445 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6447 * We need the write lock here because, for aux vdevs,
6448 * calling vdev_config_dirty() modifies sav_config.
6449 * This is ugly and will become unnecessary when we
6450 * eliminate the aux vdev wart by integrating all vdevs
6451 * into the root vdev tree.
6453 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6454 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6455 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6456 vdev_state_clean(vd
);
6457 vdev_config_dirty(vd
);
6459 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6460 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6462 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6464 tx
= dmu_tx_create_assigned(dp
, txg
);
6466 spa
->spa_sync_starttime
= gethrtime();
6467 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6468 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
6469 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6470 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6473 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6474 * set spa_deflate if we have no raid-z vdevs.
6476 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6477 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6480 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6481 vd
= rvd
->vdev_child
[i
];
6482 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6485 if (i
== rvd
->vdev_children
) {
6486 spa
->spa_deflate
= TRUE
;
6487 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6488 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6489 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6494 * Iterate to convergence.
6497 int pass
= ++spa
->spa_sync_pass
;
6499 spa_sync_config_object(spa
, tx
);
6500 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6501 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6502 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6503 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6504 spa_errlog_sync(spa
, txg
);
6505 dsl_pool_sync(dp
, txg
);
6507 if (pass
< zfs_sync_pass_deferred_free
) {
6508 spa_sync_frees(spa
, free_bpl
, tx
);
6511 * We can not defer frees in pass 1, because
6512 * we sync the deferred frees later in pass 1.
6514 ASSERT3U(pass
, >, 1);
6515 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6516 &spa
->spa_deferred_bpobj
, tx
);
6520 dsl_scan_sync(dp
, tx
);
6522 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6526 spa_sync_upgrades(spa
, tx
);
6528 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
6530 * Note: We need to check if the MOS is dirty
6531 * because we could have marked the MOS dirty
6532 * without updating the uberblock (e.g. if we
6533 * have sync tasks but no dirty user data). We
6534 * need to check the uberblock's rootbp because
6535 * it is updated if we have synced out dirty
6536 * data (though in this case the MOS will most
6537 * likely also be dirty due to second order
6538 * effects, we don't want to rely on that here).
6540 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
6541 !dmu_objset_is_dirty(mos
, txg
)) {
6543 * Nothing changed on the first pass,
6544 * therefore this TXG is a no-op. Avoid
6545 * syncing deferred frees, so that we
6546 * can keep this TXG as a no-op.
6548 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
6550 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6551 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
6554 spa_sync_deferred_frees(spa
, tx
);
6557 } while (dmu_objset_is_dirty(mos
, txg
));
6560 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
6562 * Make sure that the number of ZAPs for all the vdevs matches
6563 * the number of ZAPs in the per-vdev ZAP list. This only gets
6564 * called if the config is dirty; otherwise there may be
6565 * outstanding AVZ operations that weren't completed in
6566 * spa_sync_config_object.
6568 uint64_t all_vdev_zap_entry_count
;
6569 ASSERT0(zap_count(spa
->spa_meta_objset
,
6570 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
6571 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
6572 all_vdev_zap_entry_count
);
6577 * Rewrite the vdev configuration (which includes the uberblock)
6578 * to commit the transaction group.
6580 * If there are no dirty vdevs, we sync the uberblock to a few
6581 * random top-level vdevs that are known to be visible in the
6582 * config cache (see spa_vdev_add() for a complete description).
6583 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6587 * We hold SCL_STATE to prevent vdev open/close/etc.
6588 * while we're attempting to write the vdev labels.
6590 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6592 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6593 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6595 int children
= rvd
->vdev_children
;
6596 int c0
= spa_get_random(children
);
6598 for (c
= 0; c
< children
; c
++) {
6599 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6600 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6602 svd
[svdcount
++] = vd
;
6603 if (svdcount
== SPA_DVAS_PER_BP
)
6606 error
= vdev_config_sync(svd
, svdcount
, txg
);
6608 error
= vdev_config_sync(rvd
->vdev_child
,
6609 rvd
->vdev_children
, txg
);
6613 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6615 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6619 zio_suspend(spa
, NULL
);
6620 zio_resume_wait(spa
);
6624 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6625 spa
->spa_deadman_tqid
= 0;
6628 * Clear the dirty config list.
6630 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6631 vdev_config_clean(vd
);
6634 * Now that the new config has synced transactionally,
6635 * let it become visible to the config cache.
6637 if (spa
->spa_config_syncing
!= NULL
) {
6638 spa_config_set(spa
, spa
->spa_config_syncing
);
6639 spa
->spa_config_txg
= txg
;
6640 spa
->spa_config_syncing
= NULL
;
6643 spa
->spa_ubsync
= spa
->spa_uberblock
;
6645 dsl_pool_sync_done(dp
, txg
);
6648 * Update usable space statistics.
6650 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6651 vdev_sync_done(vd
, txg
);
6653 spa_update_dspace(spa
);
6656 * It had better be the case that we didn't dirty anything
6657 * since vdev_config_sync().
6659 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6660 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6661 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6663 spa
->spa_sync_pass
= 0;
6665 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6667 spa_handle_ignored_writes(spa
);
6670 * If any async tasks have been requested, kick them off.
6672 spa_async_dispatch(spa
);
6676 * Sync all pools. We don't want to hold the namespace lock across these
6677 * operations, so we take a reference on the spa_t and drop the lock during the
6681 spa_sync_allpools(void)
6684 mutex_enter(&spa_namespace_lock
);
6685 while ((spa
= spa_next(spa
)) != NULL
) {
6686 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6687 !spa_writeable(spa
) || spa_suspended(spa
))
6689 spa_open_ref(spa
, FTAG
);
6690 mutex_exit(&spa_namespace_lock
);
6691 txg_wait_synced(spa_get_dsl(spa
), 0);
6692 mutex_enter(&spa_namespace_lock
);
6693 spa_close(spa
, FTAG
);
6695 mutex_exit(&spa_namespace_lock
);
6699 * ==========================================================================
6700 * Miscellaneous routines
6701 * ==========================================================================
6705 * Remove all pools in the system.
6713 * Remove all cached state. All pools should be closed now,
6714 * so every spa in the AVL tree should be unreferenced.
6716 mutex_enter(&spa_namespace_lock
);
6717 while ((spa
= spa_next(NULL
)) != NULL
) {
6719 * Stop async tasks. The async thread may need to detach
6720 * a device that's been replaced, which requires grabbing
6721 * spa_namespace_lock, so we must drop it here.
6723 spa_open_ref(spa
, FTAG
);
6724 mutex_exit(&spa_namespace_lock
);
6725 spa_async_suspend(spa
);
6726 mutex_enter(&spa_namespace_lock
);
6727 spa_close(spa
, FTAG
);
6729 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6731 spa_deactivate(spa
);
6735 mutex_exit(&spa_namespace_lock
);
6739 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6744 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6748 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6749 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6750 if (vd
->vdev_guid
== guid
)
6754 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6755 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6756 if (vd
->vdev_guid
== guid
)
6765 spa_upgrade(spa_t
*spa
, uint64_t version
)
6767 ASSERT(spa_writeable(spa
));
6769 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6772 * This should only be called for a non-faulted pool, and since a
6773 * future version would result in an unopenable pool, this shouldn't be
6776 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
6777 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
6779 spa
->spa_uberblock
.ub_version
= version
;
6780 vdev_config_dirty(spa
->spa_root_vdev
);
6782 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6784 txg_wait_synced(spa_get_dsl(spa
), 0);
6788 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6792 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6794 for (i
= 0; i
< sav
->sav_count
; i
++)
6795 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6798 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6799 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6800 &spareguid
) == 0 && spareguid
== guid
)
6808 * Check if a pool has an active shared spare device.
6809 * Note: reference count of an active spare is 2, as a spare and as a replace
6812 spa_has_active_shared_spare(spa_t
*spa
)
6816 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6818 for (i
= 0; i
< sav
->sav_count
; i
++) {
6819 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6820 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6829 * Post a zevent corresponding to the given sysevent. The 'name' must be one
6830 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
6831 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6832 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6833 * or zdb as real changes.
6836 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6838 zfs_post_sysevent(spa
, vd
, name
);
6841 #if defined(_KERNEL) && defined(HAVE_SPL)
6842 /* state manipulation functions */
6843 EXPORT_SYMBOL(spa_open
);
6844 EXPORT_SYMBOL(spa_open_rewind
);
6845 EXPORT_SYMBOL(spa_get_stats
);
6846 EXPORT_SYMBOL(spa_create
);
6847 EXPORT_SYMBOL(spa_import
);
6848 EXPORT_SYMBOL(spa_tryimport
);
6849 EXPORT_SYMBOL(spa_destroy
);
6850 EXPORT_SYMBOL(spa_export
);
6851 EXPORT_SYMBOL(spa_reset
);
6852 EXPORT_SYMBOL(spa_async_request
);
6853 EXPORT_SYMBOL(spa_async_suspend
);
6854 EXPORT_SYMBOL(spa_async_resume
);
6855 EXPORT_SYMBOL(spa_inject_addref
);
6856 EXPORT_SYMBOL(spa_inject_delref
);
6857 EXPORT_SYMBOL(spa_scan_stat_init
);
6858 EXPORT_SYMBOL(spa_scan_get_stats
);
6860 /* device maniion */
6861 EXPORT_SYMBOL(spa_vdev_add
);
6862 EXPORT_SYMBOL(spa_vdev_attach
);
6863 EXPORT_SYMBOL(spa_vdev_detach
);
6864 EXPORT_SYMBOL(spa_vdev_remove
);
6865 EXPORT_SYMBOL(spa_vdev_setpath
);
6866 EXPORT_SYMBOL(spa_vdev_setfru
);
6867 EXPORT_SYMBOL(spa_vdev_split_mirror
);
6869 /* spare statech is global across all pools) */
6870 EXPORT_SYMBOL(spa_spare_add
);
6871 EXPORT_SYMBOL(spa_spare_remove
);
6872 EXPORT_SYMBOL(spa_spare_exists
);
6873 EXPORT_SYMBOL(spa_spare_activate
);
6875 /* L2ARC statech is global across all pools) */
6876 EXPORT_SYMBOL(spa_l2cache_add
);
6877 EXPORT_SYMBOL(spa_l2cache_remove
);
6878 EXPORT_SYMBOL(spa_l2cache_exists
);
6879 EXPORT_SYMBOL(spa_l2cache_activate
);
6880 EXPORT_SYMBOL(spa_l2cache_drop
);
6883 EXPORT_SYMBOL(spa_scan
);
6884 EXPORT_SYMBOL(spa_scan_stop
);
6887 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
6888 EXPORT_SYMBOL(spa_sync_allpools
);
6891 EXPORT_SYMBOL(spa_prop_set
);
6892 EXPORT_SYMBOL(spa_prop_get
);
6893 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
6895 /* asynchronous event notification */
6896 EXPORT_SYMBOL(spa_event_notify
);
6899 #if defined(_KERNEL) && defined(HAVE_SPL)
6900 module_param(spa_load_verify_maxinflight
, int, 0644);
6901 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
6902 "Max concurrent traversal I/Os while verifying pool during import -X");
6904 module_param(spa_load_verify_metadata
, int, 0644);
6905 MODULE_PARM_DESC(spa_load_verify_metadata
,
6906 "Set to traverse metadata on pool import");
6908 module_param(spa_load_verify_data
, int, 0644);
6909 MODULE_PARM_DESC(spa_load_verify_data
,
6910 "Set to traverse data on pool import");
6912 module_param(zio_taskq_batch_pct
, uint
, 0444);
6913 MODULE_PARM_DESC(zio_taskq_batch_pct
,
6914 "Percentage of CPUs to run an IO worker thread");