4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
32 * Copyright (c) 2017 Datto Inc.
33 * Copyright 2017 Joyent, Inc.
37 * SPA: Storage Pool Allocator
39 * This file contains all the routines used when modifying on-disk SPA state.
40 * This includes opening, importing, destroying, exporting a pool, and syncing a
44 #include <sys/zfs_context.h>
45 #include <sys/fm/fs/zfs.h>
46 #include <sys/spa_impl.h>
48 #include <sys/zio_checksum.h>
50 #include <sys/dmu_tx.h>
54 #include <sys/vdev_impl.h>
55 #include <sys/vdev_removal.h>
56 #include <sys/vdev_indirect_mapping.h>
57 #include <sys/vdev_indirect_births.h>
58 #include <sys/vdev_disk.h>
59 #include <sys/metaslab.h>
60 #include <sys/metaslab_impl.h>
62 #include <sys/uberblock_impl.h>
65 #include <sys/bpobj.h>
66 #include <sys/dmu_traverse.h>
67 #include <sys/dmu_objset.h>
68 #include <sys/unique.h>
69 #include <sys/dsl_pool.h>
70 #include <sys/dsl_dataset.h>
71 #include <sys/dsl_dir.h>
72 #include <sys/dsl_prop.h>
73 #include <sys/dsl_synctask.h>
74 #include <sys/fs/zfs.h>
76 #include <sys/callb.h>
77 #include <sys/systeminfo.h>
78 #include <sys/spa_boot.h>
79 #include <sys/zfs_ioctl.h>
80 #include <sys/dsl_scan.h>
81 #include <sys/zfeature.h>
82 #include <sys/dsl_destroy.h>
86 #include <sys/fm/protocol.h>
87 #include <sys/fm/util.h>
88 #include <sys/bootprops.h>
89 #include <sys/callb.h>
90 #include <sys/cpupart.h>
92 #include <sys/sysdc.h>
97 #include "zfs_comutil.h"
100 * The interval, in seconds, at which failed configuration cache file writes
103 int zfs_ccw_retry_interval
= 300;
105 typedef enum zti_modes
{
106 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
107 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
108 ZTI_MODE_NULL
, /* don't create a taskq */
112 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
113 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
114 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
115 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
117 #define ZTI_N(n) ZTI_P(n, 1)
118 #define ZTI_ONE ZTI_N(1)
120 typedef struct zio_taskq_info
{
121 zti_modes_t zti_mode
;
126 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
127 "iss", "iss_h", "int", "int_h"
131 * This table defines the taskq settings for each ZFS I/O type. When
132 * initializing a pool, we use this table to create an appropriately sized
133 * taskq. Some operations are low volume and therefore have a small, static
134 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
135 * macros. Other operations process a large amount of data; the ZTI_BATCH
136 * macro causes us to create a taskq oriented for throughput. Some operations
137 * are so high frequency and short-lived that the taskq itself can become a a
138 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
139 * additional degree of parallelism specified by the number of threads per-
140 * taskq and the number of taskqs; when dispatching an event in this case, the
141 * particular taskq is chosen at random.
143 * The different taskq priorities are to handle the different contexts (issue
144 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
145 * need to be handled with minimum delay.
147 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
148 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
149 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
150 { ZTI_N(8), ZTI_NULL
, ZTI_P(12, 8), ZTI_NULL
}, /* READ */
151 { ZTI_BATCH
, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */
152 { ZTI_P(12, 8), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
153 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
154 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
157 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
158 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
159 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
160 static inline int spa_load_impl(spa_t
*spa
, uint64_t, nvlist_t
*config
,
161 spa_load_state_t state
, spa_import_type_t type
, boolean_t trust_config
,
163 static void spa_vdev_resilver_done(spa_t
*spa
);
165 uint_t zio_taskq_batch_pct
= 75; /* 1 thread per cpu in pset */
166 id_t zio_taskq_psrset_bind
= PS_NONE
;
167 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
168 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
170 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
173 * This (illegal) pool name is used when temporarily importing a spa_t in order
174 * to get the vdev stats associated with the imported devices.
176 #define TRYIMPORT_NAME "$import"
179 * ==========================================================================
180 * SPA properties routines
181 * ==========================================================================
185 * Add a (source=src, propname=propval) list to an nvlist.
188 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
189 uint64_t intval
, zprop_source_t src
)
191 const char *propname
= zpool_prop_to_name(prop
);
194 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
195 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
198 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
200 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
202 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
203 nvlist_free(propval
);
207 * Get property values from the spa configuration.
210 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
212 vdev_t
*rvd
= spa
->spa_root_vdev
;
213 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
214 uint64_t size
, alloc
, cap
, version
;
215 const zprop_source_t src
= ZPROP_SRC_NONE
;
216 spa_config_dirent_t
*dp
;
217 metaslab_class_t
*mc
= spa_normal_class(spa
);
219 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
222 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
223 size
= metaslab_class_get_space(spa_normal_class(spa
));
224 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
225 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
226 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
227 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
230 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
231 metaslab_class_fragmentation(mc
), src
);
232 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
233 metaslab_class_expandable_space(mc
), src
);
234 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
235 (spa_mode(spa
) == FREAD
), src
);
237 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
238 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
240 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
241 ddt_get_pool_dedup_ratio(spa
), src
);
243 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
244 rvd
->vdev_state
, src
);
246 version
= spa_version(spa
);
247 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
)) {
248 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
249 version
, ZPROP_SRC_DEFAULT
);
251 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
252 version
, ZPROP_SRC_LOCAL
);
258 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
259 * when opening pools before this version freedir will be NULL.
261 if (pool
->dp_free_dir
!= NULL
) {
262 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
263 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
266 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
270 if (pool
->dp_leak_dir
!= NULL
) {
271 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
272 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
275 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
280 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
282 if (spa
->spa_comment
!= NULL
) {
283 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
287 if (spa
->spa_root
!= NULL
)
288 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
291 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
292 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
293 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
295 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
296 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
299 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
)) {
300 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
301 DNODE_MAX_SIZE
, ZPROP_SRC_NONE
);
303 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
304 DNODE_MIN_SIZE
, ZPROP_SRC_NONE
);
307 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
308 if (dp
->scd_path
== NULL
) {
309 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
310 "none", 0, ZPROP_SRC_LOCAL
);
311 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
312 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
313 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
319 * Get zpool property values.
322 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
324 objset_t
*mos
= spa
->spa_meta_objset
;
329 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
333 mutex_enter(&spa
->spa_props_lock
);
336 * Get properties from the spa config.
338 spa_prop_get_config(spa
, nvp
);
340 /* If no pool property object, no more prop to get. */
341 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
342 mutex_exit(&spa
->spa_props_lock
);
347 * Get properties from the MOS pool property object.
349 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
350 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
351 zap_cursor_advance(&zc
)) {
354 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
357 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPOOL_PROP_INVAL
)
360 switch (za
.za_integer_length
) {
362 /* integer property */
363 if (za
.za_first_integer
!=
364 zpool_prop_default_numeric(prop
))
365 src
= ZPROP_SRC_LOCAL
;
367 if (prop
== ZPOOL_PROP_BOOTFS
) {
369 dsl_dataset_t
*ds
= NULL
;
371 dp
= spa_get_dsl(spa
);
372 dsl_pool_config_enter(dp
, FTAG
);
373 if ((err
= dsl_dataset_hold_obj(dp
,
374 za
.za_first_integer
, FTAG
, &ds
))) {
375 dsl_pool_config_exit(dp
, FTAG
);
379 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
381 dsl_dataset_name(ds
, strval
);
382 dsl_dataset_rele(ds
, FTAG
);
383 dsl_pool_config_exit(dp
, FTAG
);
386 intval
= za
.za_first_integer
;
389 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
392 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
397 /* string property */
398 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
399 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
400 za
.za_name
, 1, za
.za_num_integers
, strval
);
402 kmem_free(strval
, za
.za_num_integers
);
405 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
406 kmem_free(strval
, za
.za_num_integers
);
413 zap_cursor_fini(&zc
);
414 mutex_exit(&spa
->spa_props_lock
);
416 if (err
&& err
!= ENOENT
) {
426 * Validate the given pool properties nvlist and modify the list
427 * for the property values to be set.
430 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
433 int error
= 0, reset_bootfs
= 0;
435 boolean_t has_feature
= B_FALSE
;
438 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
440 char *strval
, *slash
, *check
, *fname
;
441 const char *propname
= nvpair_name(elem
);
442 zpool_prop_t prop
= zpool_name_to_prop(propname
);
445 case ZPOOL_PROP_INVAL
:
446 if (!zpool_prop_feature(propname
)) {
447 error
= SET_ERROR(EINVAL
);
452 * Sanitize the input.
454 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
455 error
= SET_ERROR(EINVAL
);
459 if (nvpair_value_uint64(elem
, &intval
) != 0) {
460 error
= SET_ERROR(EINVAL
);
465 error
= SET_ERROR(EINVAL
);
469 fname
= strchr(propname
, '@') + 1;
470 if (zfeature_lookup_name(fname
, NULL
) != 0) {
471 error
= SET_ERROR(EINVAL
);
475 has_feature
= B_TRUE
;
478 case ZPOOL_PROP_VERSION
:
479 error
= nvpair_value_uint64(elem
, &intval
);
481 (intval
< spa_version(spa
) ||
482 intval
> SPA_VERSION_BEFORE_FEATURES
||
484 error
= SET_ERROR(EINVAL
);
487 case ZPOOL_PROP_DELEGATION
:
488 case ZPOOL_PROP_AUTOREPLACE
:
489 case ZPOOL_PROP_LISTSNAPS
:
490 case ZPOOL_PROP_AUTOEXPAND
:
491 error
= nvpair_value_uint64(elem
, &intval
);
492 if (!error
&& intval
> 1)
493 error
= SET_ERROR(EINVAL
);
496 case ZPOOL_PROP_MULTIHOST
:
497 error
= nvpair_value_uint64(elem
, &intval
);
498 if (!error
&& intval
> 1)
499 error
= SET_ERROR(EINVAL
);
501 if (!error
&& !spa_get_hostid())
502 error
= SET_ERROR(ENOTSUP
);
506 case ZPOOL_PROP_BOOTFS
:
508 * If the pool version is less than SPA_VERSION_BOOTFS,
509 * or the pool is still being created (version == 0),
510 * the bootfs property cannot be set.
512 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
513 error
= SET_ERROR(ENOTSUP
);
518 * Make sure the vdev config is bootable
520 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
521 error
= SET_ERROR(ENOTSUP
);
527 error
= nvpair_value_string(elem
, &strval
);
533 if (strval
== NULL
|| strval
[0] == '\0') {
534 objnum
= zpool_prop_default_numeric(
539 error
= dmu_objset_hold(strval
, FTAG
, &os
);
544 * Must be ZPL, and its property settings
545 * must be supported by GRUB (compression
546 * is not gzip, and large blocks or large
547 * dnodes are not used).
550 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
551 error
= SET_ERROR(ENOTSUP
);
553 dsl_prop_get_int_ds(dmu_objset_ds(os
),
554 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
556 !BOOTFS_COMPRESS_VALID(propval
)) {
557 error
= SET_ERROR(ENOTSUP
);
559 dsl_prop_get_int_ds(dmu_objset_ds(os
),
560 zfs_prop_to_name(ZFS_PROP_DNODESIZE
),
562 propval
!= ZFS_DNSIZE_LEGACY
) {
563 error
= SET_ERROR(ENOTSUP
);
565 objnum
= dmu_objset_id(os
);
567 dmu_objset_rele(os
, FTAG
);
571 case ZPOOL_PROP_FAILUREMODE
:
572 error
= nvpair_value_uint64(elem
, &intval
);
573 if (!error
&& intval
> ZIO_FAILURE_MODE_PANIC
)
574 error
= SET_ERROR(EINVAL
);
577 * This is a special case which only occurs when
578 * the pool has completely failed. This allows
579 * the user to change the in-core failmode property
580 * without syncing it out to disk (I/Os might
581 * currently be blocked). We do this by returning
582 * EIO to the caller (spa_prop_set) to trick it
583 * into thinking we encountered a property validation
586 if (!error
&& spa_suspended(spa
)) {
587 spa
->spa_failmode
= intval
;
588 error
= SET_ERROR(EIO
);
592 case ZPOOL_PROP_CACHEFILE
:
593 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
596 if (strval
[0] == '\0')
599 if (strcmp(strval
, "none") == 0)
602 if (strval
[0] != '/') {
603 error
= SET_ERROR(EINVAL
);
607 slash
= strrchr(strval
, '/');
608 ASSERT(slash
!= NULL
);
610 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
611 strcmp(slash
, "/..") == 0)
612 error
= SET_ERROR(EINVAL
);
615 case ZPOOL_PROP_COMMENT
:
616 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
618 for (check
= strval
; *check
!= '\0'; check
++) {
619 if (!isprint(*check
)) {
620 error
= SET_ERROR(EINVAL
);
624 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
625 error
= SET_ERROR(E2BIG
);
628 case ZPOOL_PROP_DEDUPDITTO
:
629 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
630 error
= SET_ERROR(ENOTSUP
);
632 error
= nvpair_value_uint64(elem
, &intval
);
634 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
635 error
= SET_ERROR(EINVAL
);
646 if (!error
&& reset_bootfs
) {
647 error
= nvlist_remove(props
,
648 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
651 error
= nvlist_add_uint64(props
,
652 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
660 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
663 spa_config_dirent_t
*dp
;
665 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
669 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
672 if (cachefile
[0] == '\0')
673 dp
->scd_path
= spa_strdup(spa_config_path
);
674 else if (strcmp(cachefile
, "none") == 0)
677 dp
->scd_path
= spa_strdup(cachefile
);
679 list_insert_head(&spa
->spa_config_list
, dp
);
681 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
685 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
688 nvpair_t
*elem
= NULL
;
689 boolean_t need_sync
= B_FALSE
;
691 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
694 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
695 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
697 if (prop
== ZPOOL_PROP_CACHEFILE
||
698 prop
== ZPOOL_PROP_ALTROOT
||
699 prop
== ZPOOL_PROP_READONLY
)
702 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPOOL_PROP_INVAL
) {
705 if (prop
== ZPOOL_PROP_VERSION
) {
706 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
708 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
709 ver
= SPA_VERSION_FEATURES
;
713 /* Save time if the version is already set. */
714 if (ver
== spa_version(spa
))
718 * In addition to the pool directory object, we might
719 * create the pool properties object, the features for
720 * read object, the features for write object, or the
721 * feature descriptions object.
723 error
= dsl_sync_task(spa
->spa_name
, NULL
,
724 spa_sync_version
, &ver
,
725 6, ZFS_SPACE_CHECK_RESERVED
);
736 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
737 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
744 * If the bootfs property value is dsobj, clear it.
747 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
749 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
750 VERIFY(zap_remove(spa
->spa_meta_objset
,
751 spa
->spa_pool_props_object
,
752 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
759 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
761 ASSERTV(uint64_t *newguid
= arg
);
762 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
763 vdev_t
*rvd
= spa
->spa_root_vdev
;
766 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
767 vdev_state
= rvd
->vdev_state
;
768 spa_config_exit(spa
, SCL_STATE
, FTAG
);
770 if (vdev_state
!= VDEV_STATE_HEALTHY
)
771 return (SET_ERROR(ENXIO
));
773 ASSERT3U(spa_guid(spa
), !=, *newguid
);
779 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
781 uint64_t *newguid
= arg
;
782 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
784 vdev_t
*rvd
= spa
->spa_root_vdev
;
786 oldguid
= spa_guid(spa
);
788 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
789 rvd
->vdev_guid
= *newguid
;
790 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
791 vdev_config_dirty(rvd
);
792 spa_config_exit(spa
, SCL_STATE
, FTAG
);
794 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
799 * Change the GUID for the pool. This is done so that we can later
800 * re-import a pool built from a clone of our own vdevs. We will modify
801 * the root vdev's guid, our own pool guid, and then mark all of our
802 * vdevs dirty. Note that we must make sure that all our vdevs are
803 * online when we do this, or else any vdevs that weren't present
804 * would be orphaned from our pool. We are also going to issue a
805 * sysevent to update any watchers.
808 spa_change_guid(spa_t
*spa
)
813 mutex_enter(&spa
->spa_vdev_top_lock
);
814 mutex_enter(&spa_namespace_lock
);
815 guid
= spa_generate_guid(NULL
);
817 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
818 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
821 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
822 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
825 mutex_exit(&spa_namespace_lock
);
826 mutex_exit(&spa
->spa_vdev_top_lock
);
832 * ==========================================================================
833 * SPA state manipulation (open/create/destroy/import/export)
834 * ==========================================================================
838 spa_error_entry_compare(const void *a
, const void *b
)
840 const spa_error_entry_t
*sa
= (const spa_error_entry_t
*)a
;
841 const spa_error_entry_t
*sb
= (const spa_error_entry_t
*)b
;
844 ret
= memcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
845 sizeof (zbookmark_phys_t
));
847 return (AVL_ISIGN(ret
));
851 * Utility function which retrieves copies of the current logs and
852 * re-initializes them in the process.
855 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
857 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
859 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
860 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
862 avl_create(&spa
->spa_errlist_scrub
,
863 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
864 offsetof(spa_error_entry_t
, se_avl
));
865 avl_create(&spa
->spa_errlist_last
,
866 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
867 offsetof(spa_error_entry_t
, se_avl
));
871 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
873 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
874 enum zti_modes mode
= ztip
->zti_mode
;
875 uint_t value
= ztip
->zti_value
;
876 uint_t count
= ztip
->zti_count
;
877 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
880 boolean_t batch
= B_FALSE
;
882 if (mode
== ZTI_MODE_NULL
) {
884 tqs
->stqs_taskq
= NULL
;
888 ASSERT3U(count
, >, 0);
890 tqs
->stqs_count
= count
;
891 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
895 ASSERT3U(value
, >=, 1);
896 value
= MAX(value
, 1);
897 flags
|= TASKQ_DYNAMIC
;
902 flags
|= TASKQ_THREADS_CPU_PCT
;
903 value
= MIN(zio_taskq_batch_pct
, 100);
907 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
909 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
913 for (uint_t i
= 0; i
< count
; i
++) {
917 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
918 zio_type_name
[t
], zio_taskq_types
[q
], i
);
920 (void) snprintf(name
, sizeof (name
), "%s_%s",
921 zio_type_name
[t
], zio_taskq_types
[q
]);
924 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
926 flags
|= TASKQ_DC_BATCH
;
928 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
929 spa
->spa_proc
, zio_taskq_basedc
, flags
);
931 pri_t pri
= maxclsyspri
;
933 * The write issue taskq can be extremely CPU
934 * intensive. Run it at slightly less important
935 * priority than the other taskqs. Under Linux this
936 * means incrementing the priority value on platforms
937 * like illumos it should be decremented.
939 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
942 tq
= taskq_create_proc(name
, value
, pri
, 50,
943 INT_MAX
, spa
->spa_proc
, flags
);
946 tqs
->stqs_taskq
[i
] = tq
;
951 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
953 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
955 if (tqs
->stqs_taskq
== NULL
) {
956 ASSERT3U(tqs
->stqs_count
, ==, 0);
960 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
961 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
962 taskq_destroy(tqs
->stqs_taskq
[i
]);
965 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
966 tqs
->stqs_taskq
= NULL
;
970 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
971 * Note that a type may have multiple discrete taskqs to avoid lock contention
972 * on the taskq itself. In that case we choose which taskq at random by using
973 * the low bits of gethrtime().
976 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
977 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
979 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
982 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
983 ASSERT3U(tqs
->stqs_count
, !=, 0);
985 if (tqs
->stqs_count
== 1) {
986 tq
= tqs
->stqs_taskq
[0];
988 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
991 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
995 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
998 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
999 task_func_t
*func
, void *arg
, uint_t flags
)
1001 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1005 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1006 ASSERT3U(tqs
->stqs_count
, !=, 0);
1008 if (tqs
->stqs_count
== 1) {
1009 tq
= tqs
->stqs_taskq
[0];
1011 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1014 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1016 taskq_wait_id(tq
, id
);
1020 spa_create_zio_taskqs(spa_t
*spa
)
1022 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1023 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1024 spa_taskqs_init(spa
, t
, q
);
1030 * Disabled until spa_thread() can be adapted for Linux.
1032 #undef HAVE_SPA_THREAD
1034 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1036 spa_thread(void *arg
)
1038 callb_cpr_t cprinfo
;
1041 user_t
*pu
= PTOU(curproc
);
1043 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1046 ASSERT(curproc
!= &p0
);
1047 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1048 "zpool-%s", spa
->spa_name
);
1049 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1051 /* bind this thread to the requested psrset */
1052 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1054 mutex_enter(&cpu_lock
);
1055 mutex_enter(&pidlock
);
1056 mutex_enter(&curproc
->p_lock
);
1058 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1059 0, NULL
, NULL
) == 0) {
1060 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1063 "Couldn't bind process for zfs pool \"%s\" to "
1064 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1067 mutex_exit(&curproc
->p_lock
);
1068 mutex_exit(&pidlock
);
1069 mutex_exit(&cpu_lock
);
1073 if (zio_taskq_sysdc
) {
1074 sysdc_thread_enter(curthread
, 100, 0);
1077 spa
->spa_proc
= curproc
;
1078 spa
->spa_did
= curthread
->t_did
;
1080 spa_create_zio_taskqs(spa
);
1082 mutex_enter(&spa
->spa_proc_lock
);
1083 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1085 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1086 cv_broadcast(&spa
->spa_proc_cv
);
1088 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1089 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1090 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1091 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1093 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1094 spa
->spa_proc_state
= SPA_PROC_GONE
;
1095 spa
->spa_proc
= &p0
;
1096 cv_broadcast(&spa
->spa_proc_cv
);
1097 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1099 mutex_enter(&curproc
->p_lock
);
1105 * Activate an uninitialized pool.
1108 spa_activate(spa_t
*spa
, int mode
)
1110 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1112 spa
->spa_state
= POOL_STATE_ACTIVE
;
1113 spa
->spa_mode
= mode
;
1115 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1116 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1118 /* Try to create a covering process */
1119 mutex_enter(&spa
->spa_proc_lock
);
1120 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1121 ASSERT(spa
->spa_proc
== &p0
);
1124 #ifdef HAVE_SPA_THREAD
1125 /* Only create a process if we're going to be around a while. */
1126 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1127 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1129 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1130 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1131 cv_wait(&spa
->spa_proc_cv
,
1132 &spa
->spa_proc_lock
);
1134 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1135 ASSERT(spa
->spa_proc
!= &p0
);
1136 ASSERT(spa
->spa_did
!= 0);
1140 "Couldn't create process for zfs pool \"%s\"\n",
1145 #endif /* HAVE_SPA_THREAD */
1146 mutex_exit(&spa
->spa_proc_lock
);
1148 /* If we didn't create a process, we need to create our taskqs. */
1149 if (spa
->spa_proc
== &p0
) {
1150 spa_create_zio_taskqs(spa
);
1153 for (size_t i
= 0; i
< TXG_SIZE
; i
++)
1154 spa
->spa_txg_zio
[i
] = zio_root(spa
, NULL
, NULL
, 0);
1156 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1157 offsetof(vdev_t
, vdev_config_dirty_node
));
1158 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1159 offsetof(objset_t
, os_evicting_node
));
1160 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1161 offsetof(vdev_t
, vdev_state_dirty_node
));
1163 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1164 offsetof(struct vdev
, vdev_txg_node
));
1166 avl_create(&spa
->spa_errlist_scrub
,
1167 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1168 offsetof(spa_error_entry_t
, se_avl
));
1169 avl_create(&spa
->spa_errlist_last
,
1170 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1171 offsetof(spa_error_entry_t
, se_avl
));
1173 spa_keystore_init(&spa
->spa_keystore
);
1176 * This taskq is used to perform zvol-minor-related tasks
1177 * asynchronously. This has several advantages, including easy
1178 * resolution of various deadlocks (zfsonlinux bug #3681).
1180 * The taskq must be single threaded to ensure tasks are always
1181 * processed in the order in which they were dispatched.
1183 * A taskq per pool allows one to keep the pools independent.
1184 * This way if one pool is suspended, it will not impact another.
1186 * The preferred location to dispatch a zvol minor task is a sync
1187 * task. In this context, there is easy access to the spa_t and minimal
1188 * error handling is required because the sync task must succeed.
1190 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1194 * Taskq dedicated to prefetcher threads: this is used to prevent the
1195 * pool traverse code from monopolizing the global (and limited)
1196 * system_taskq by inappropriately scheduling long running tasks on it.
1198 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", boot_ncpus
,
1199 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1202 * The taskq to upgrade datasets in this pool. Currently used by
1203 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1205 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", boot_ncpus
,
1206 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1210 * Opposite of spa_activate().
1213 spa_deactivate(spa_t
*spa
)
1215 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1216 ASSERT(spa
->spa_dsl_pool
== NULL
);
1217 ASSERT(spa
->spa_root_vdev
== NULL
);
1218 ASSERT(spa
->spa_async_zio_root
== NULL
);
1219 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1221 spa_evicting_os_wait(spa
);
1223 if (spa
->spa_zvol_taskq
) {
1224 taskq_destroy(spa
->spa_zvol_taskq
);
1225 spa
->spa_zvol_taskq
= NULL
;
1228 if (spa
->spa_prefetch_taskq
) {
1229 taskq_destroy(spa
->spa_prefetch_taskq
);
1230 spa
->spa_prefetch_taskq
= NULL
;
1233 if (spa
->spa_upgrade_taskq
) {
1234 taskq_destroy(spa
->spa_upgrade_taskq
);
1235 spa
->spa_upgrade_taskq
= NULL
;
1238 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1240 list_destroy(&spa
->spa_config_dirty_list
);
1241 list_destroy(&spa
->spa_evicting_os_list
);
1242 list_destroy(&spa
->spa_state_dirty_list
);
1244 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1246 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1247 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1248 spa_taskqs_fini(spa
, t
, q
);
1252 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1253 ASSERT3P(spa
->spa_txg_zio
[i
], !=, NULL
);
1254 VERIFY0(zio_wait(spa
->spa_txg_zio
[i
]));
1255 spa
->spa_txg_zio
[i
] = NULL
;
1258 metaslab_class_destroy(spa
->spa_normal_class
);
1259 spa
->spa_normal_class
= NULL
;
1261 metaslab_class_destroy(spa
->spa_log_class
);
1262 spa
->spa_log_class
= NULL
;
1265 * If this was part of an import or the open otherwise failed, we may
1266 * still have errors left in the queues. Empty them just in case.
1268 spa_errlog_drain(spa
);
1269 avl_destroy(&spa
->spa_errlist_scrub
);
1270 avl_destroy(&spa
->spa_errlist_last
);
1272 spa_keystore_fini(&spa
->spa_keystore
);
1274 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1276 mutex_enter(&spa
->spa_proc_lock
);
1277 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1278 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1279 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1280 cv_broadcast(&spa
->spa_proc_cv
);
1281 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1282 ASSERT(spa
->spa_proc
!= &p0
);
1283 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1285 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1286 spa
->spa_proc_state
= SPA_PROC_NONE
;
1288 ASSERT(spa
->spa_proc
== &p0
);
1289 mutex_exit(&spa
->spa_proc_lock
);
1292 * We want to make sure spa_thread() has actually exited the ZFS
1293 * module, so that the module can't be unloaded out from underneath
1296 if (spa
->spa_did
!= 0) {
1297 thread_join(spa
->spa_did
);
1303 * Verify a pool configuration, and construct the vdev tree appropriately. This
1304 * will create all the necessary vdevs in the appropriate layout, with each vdev
1305 * in the CLOSED state. This will prep the pool before open/creation/import.
1306 * All vdev validation is done by the vdev_alloc() routine.
1309 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1310 uint_t id
, int atype
)
1316 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1319 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1322 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1325 if (error
== ENOENT
)
1331 return (SET_ERROR(EINVAL
));
1334 for (int c
= 0; c
< children
; c
++) {
1336 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1344 ASSERT(*vdp
!= NULL
);
1350 * Opposite of spa_load().
1353 spa_unload(spa_t
*spa
)
1357 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1362 spa_async_suspend(spa
);
1367 if (spa
->spa_sync_on
) {
1368 txg_sync_stop(spa
->spa_dsl_pool
);
1369 spa
->spa_sync_on
= B_FALSE
;
1373 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1374 * to call it earlier, before we wait for async i/o to complete.
1375 * This ensures that there is no async metaslab prefetching, by
1376 * calling taskq_wait(mg_taskq).
1378 if (spa
->spa_root_vdev
!= NULL
) {
1379 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1380 for (int c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++)
1381 vdev_metaslab_fini(spa
->spa_root_vdev
->vdev_child
[c
]);
1382 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1385 if (spa
->spa_mmp
.mmp_thread
)
1386 mmp_thread_stop(spa
);
1389 * Wait for any outstanding async I/O to complete.
1391 if (spa
->spa_async_zio_root
!= NULL
) {
1392 for (int i
= 0; i
< max_ncpus
; i
++)
1393 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1394 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1395 spa
->spa_async_zio_root
= NULL
;
1398 if (spa
->spa_vdev_removal
!= NULL
) {
1399 spa_vdev_removal_destroy(spa
->spa_vdev_removal
);
1400 spa
->spa_vdev_removal
= NULL
;
1403 if (spa
->spa_condense_zthr
!= NULL
) {
1404 ASSERT(!zthr_isrunning(spa
->spa_condense_zthr
));
1405 zthr_destroy(spa
->spa_condense_zthr
);
1406 spa
->spa_condense_zthr
= NULL
;
1409 spa_condense_fini(spa
);
1411 bpobj_close(&spa
->spa_deferred_bpobj
);
1413 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1418 if (spa
->spa_root_vdev
)
1419 vdev_free(spa
->spa_root_vdev
);
1420 ASSERT(spa
->spa_root_vdev
== NULL
);
1423 * Close the dsl pool.
1425 if (spa
->spa_dsl_pool
) {
1426 dsl_pool_close(spa
->spa_dsl_pool
);
1427 spa
->spa_dsl_pool
= NULL
;
1428 spa
->spa_meta_objset
= NULL
;
1434 * Drop and purge level 2 cache
1436 spa_l2cache_drop(spa
);
1438 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1439 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1440 if (spa
->spa_spares
.sav_vdevs
) {
1441 kmem_free(spa
->spa_spares
.sav_vdevs
,
1442 spa
->spa_spares
.sav_count
* sizeof (void *));
1443 spa
->spa_spares
.sav_vdevs
= NULL
;
1445 if (spa
->spa_spares
.sav_config
) {
1446 nvlist_free(spa
->spa_spares
.sav_config
);
1447 spa
->spa_spares
.sav_config
= NULL
;
1449 spa
->spa_spares
.sav_count
= 0;
1451 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1452 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1453 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1455 if (spa
->spa_l2cache
.sav_vdevs
) {
1456 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1457 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1458 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1460 if (spa
->spa_l2cache
.sav_config
) {
1461 nvlist_free(spa
->spa_l2cache
.sav_config
);
1462 spa
->spa_l2cache
.sav_config
= NULL
;
1464 spa
->spa_l2cache
.sav_count
= 0;
1466 spa
->spa_async_suspended
= 0;
1468 spa
->spa_indirect_vdevs_loaded
= B_FALSE
;
1470 if (spa
->spa_comment
!= NULL
) {
1471 spa_strfree(spa
->spa_comment
);
1472 spa
->spa_comment
= NULL
;
1475 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1479 * Load (or re-load) the current list of vdevs describing the active spares for
1480 * this pool. When this is called, we have some form of basic information in
1481 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1482 * then re-generate a more complete list including status information.
1485 spa_load_spares(spa_t
*spa
)
1492 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1495 * First, close and free any existing spare vdevs.
1497 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1498 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1500 /* Undo the call to spa_activate() below */
1501 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1502 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1503 spa_spare_remove(tvd
);
1508 if (spa
->spa_spares
.sav_vdevs
)
1509 kmem_free(spa
->spa_spares
.sav_vdevs
,
1510 spa
->spa_spares
.sav_count
* sizeof (void *));
1512 if (spa
->spa_spares
.sav_config
== NULL
)
1515 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1516 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1518 spa
->spa_spares
.sav_count
= (int)nspares
;
1519 spa
->spa_spares
.sav_vdevs
= NULL
;
1525 * Construct the array of vdevs, opening them to get status in the
1526 * process. For each spare, there is potentially two different vdev_t
1527 * structures associated with it: one in the list of spares (used only
1528 * for basic validation purposes) and one in the active vdev
1529 * configuration (if it's spared in). During this phase we open and
1530 * validate each vdev on the spare list. If the vdev also exists in the
1531 * active configuration, then we also mark this vdev as an active spare.
1533 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1535 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1536 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1537 VDEV_ALLOC_SPARE
) == 0);
1540 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1542 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1543 B_FALSE
)) != NULL
) {
1544 if (!tvd
->vdev_isspare
)
1548 * We only mark the spare active if we were successfully
1549 * able to load the vdev. Otherwise, importing a pool
1550 * with a bad active spare would result in strange
1551 * behavior, because multiple pool would think the spare
1552 * is actively in use.
1554 * There is a vulnerability here to an equally bizarre
1555 * circumstance, where a dead active spare is later
1556 * brought back to life (onlined or otherwise). Given
1557 * the rarity of this scenario, and the extra complexity
1558 * it adds, we ignore the possibility.
1560 if (!vdev_is_dead(tvd
))
1561 spa_spare_activate(tvd
);
1565 vd
->vdev_aux
= &spa
->spa_spares
;
1567 if (vdev_open(vd
) != 0)
1570 if (vdev_validate_aux(vd
) == 0)
1575 * Recompute the stashed list of spares, with status information
1578 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1579 DATA_TYPE_NVLIST_ARRAY
) == 0);
1581 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1583 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1584 spares
[i
] = vdev_config_generate(spa
,
1585 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1586 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1587 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1588 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1589 nvlist_free(spares
[i
]);
1590 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1594 * Load (or re-load) the current list of vdevs describing the active l2cache for
1595 * this pool. When this is called, we have some form of basic information in
1596 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1597 * then re-generate a more complete list including status information.
1598 * Devices which are already active have their details maintained, and are
1602 spa_load_l2cache(spa_t
*spa
)
1604 nvlist_t
**l2cache
= NULL
;
1606 int i
, j
, oldnvdevs
;
1608 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1609 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1611 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1613 oldvdevs
= sav
->sav_vdevs
;
1614 oldnvdevs
= sav
->sav_count
;
1615 sav
->sav_vdevs
= NULL
;
1618 if (sav
->sav_config
== NULL
) {
1624 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1625 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1626 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1629 * Process new nvlist of vdevs.
1631 for (i
= 0; i
< nl2cache
; i
++) {
1632 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1636 for (j
= 0; j
< oldnvdevs
; j
++) {
1638 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1640 * Retain previous vdev for add/remove ops.
1648 if (newvdevs
[i
] == NULL
) {
1652 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1653 VDEV_ALLOC_L2CACHE
) == 0);
1658 * Commit this vdev as an l2cache device,
1659 * even if it fails to open.
1661 spa_l2cache_add(vd
);
1666 spa_l2cache_activate(vd
);
1668 if (vdev_open(vd
) != 0)
1671 (void) vdev_validate_aux(vd
);
1673 if (!vdev_is_dead(vd
))
1674 l2arc_add_vdev(spa
, vd
);
1678 sav
->sav_vdevs
= newvdevs
;
1679 sav
->sav_count
= (int)nl2cache
;
1682 * Recompute the stashed list of l2cache devices, with status
1683 * information this time.
1685 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1686 DATA_TYPE_NVLIST_ARRAY
) == 0);
1688 if (sav
->sav_count
> 0)
1689 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *),
1691 for (i
= 0; i
< sav
->sav_count
; i
++)
1692 l2cache
[i
] = vdev_config_generate(spa
,
1693 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1694 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1695 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1699 * Purge vdevs that were dropped
1701 for (i
= 0; i
< oldnvdevs
; i
++) {
1706 ASSERT(vd
->vdev_isl2cache
);
1708 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1709 pool
!= 0ULL && l2arc_vdev_present(vd
))
1710 l2arc_remove_vdev(vd
);
1711 vdev_clear_stats(vd
);
1717 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1719 for (i
= 0; i
< sav
->sav_count
; i
++)
1720 nvlist_free(l2cache
[i
]);
1722 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1726 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1729 char *packed
= NULL
;
1734 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1738 nvsize
= *(uint64_t *)db
->db_data
;
1739 dmu_buf_rele(db
, FTAG
);
1741 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1742 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1745 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1746 vmem_free(packed
, nvsize
);
1752 * Checks to see if the given vdev could not be opened, in which case we post a
1753 * sysevent to notify the autoreplace code that the device has been removed.
1756 spa_check_removed(vdev_t
*vd
)
1758 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1759 spa_check_removed(vd
->vdev_child
[c
]);
1761 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1762 vdev_is_concrete(vd
)) {
1763 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1764 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
1769 spa_config_valid_zaps(vdev_t
*vd
, vdev_t
*mvd
)
1771 ASSERT3U(vd
->vdev_children
, ==, mvd
->vdev_children
);
1773 vd
->vdev_top_zap
= mvd
->vdev_top_zap
;
1774 vd
->vdev_leaf_zap
= mvd
->vdev_leaf_zap
;
1776 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
1777 spa_config_valid_zaps(vd
->vdev_child
[i
], mvd
->vdev_child
[i
]);
1782 * Validate the current config against the MOS config
1785 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1787 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1790 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1792 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1793 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1795 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1798 * If we're doing a normal import, then build up any additional
1799 * diagnostic information about missing devices in this config.
1800 * We'll pass this up to the user for further processing.
1802 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1803 nvlist_t
**child
, *nv
;
1806 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
1808 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1810 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1811 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1812 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1814 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1815 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1817 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1822 VERIFY(nvlist_add_nvlist_array(nv
,
1823 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1824 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1825 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1827 for (int i
= 0; i
< idx
; i
++)
1828 nvlist_free(child
[i
]);
1831 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1835 * Compare the root vdev tree with the information we have
1836 * from the MOS config (mrvd). Check each top-level vdev
1837 * with the corresponding MOS config top-level (mtvd).
1839 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1840 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1841 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1844 * Resolve any "missing" vdevs in the current configuration.
1845 * Also trust the MOS config about any "indirect" vdevs.
1846 * If we find that the MOS config has more accurate information
1847 * about the top-level vdev then use that vdev instead.
1849 if ((tvd
->vdev_ops
== &vdev_missing_ops
&&
1850 mtvd
->vdev_ops
!= &vdev_missing_ops
) ||
1851 (mtvd
->vdev_ops
== &vdev_indirect_ops
&&
1852 tvd
->vdev_ops
!= &vdev_indirect_ops
)) {
1855 * Device specific actions.
1857 if (mtvd
->vdev_islog
) {
1858 if (!(spa
->spa_import_flags
&
1859 ZFS_IMPORT_MISSING_LOG
)) {
1863 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1864 } else if (mtvd
->vdev_ops
!= &vdev_indirect_ops
) {
1869 * Swap the missing vdev with the data we were
1870 * able to obtain from the MOS config.
1872 vdev_remove_child(rvd
, tvd
);
1873 vdev_remove_child(mrvd
, mtvd
);
1875 vdev_add_child(rvd
, mtvd
);
1876 vdev_add_child(mrvd
, tvd
);
1880 if (mtvd
->vdev_islog
) {
1882 * Load the slog device's state from the MOS
1883 * config since it's possible that the label
1884 * does not contain the most up-to-date
1887 vdev_load_log_state(tvd
, mtvd
);
1892 * Per-vdev ZAP info is stored exclusively in the MOS.
1894 spa_config_valid_zaps(tvd
, mtvd
);
1898 * Never trust this info from userland; always use what's
1899 * in the MOS. This prevents it from getting out of sync
1900 * with the rest of the info in the MOS.
1902 tvd
->vdev_removing
= mtvd
->vdev_removing
;
1903 tvd
->vdev_indirect_config
= mtvd
->vdev_indirect_config
;
1907 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1910 * Ensure we were able to validate the config.
1912 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1916 * Check for missing log devices
1919 spa_check_logs(spa_t
*spa
)
1921 boolean_t rv
= B_FALSE
;
1922 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1924 switch (spa
->spa_log_state
) {
1927 case SPA_LOG_MISSING
:
1928 /* need to recheck in case slog has been restored */
1929 case SPA_LOG_UNKNOWN
:
1930 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1931 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1933 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1940 spa_passivate_log(spa_t
*spa
)
1942 vdev_t
*rvd
= spa
->spa_root_vdev
;
1943 boolean_t slog_found
= B_FALSE
;
1945 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1947 if (!spa_has_slogs(spa
))
1950 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1951 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1952 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1954 if (tvd
->vdev_islog
) {
1955 metaslab_group_passivate(mg
);
1956 slog_found
= B_TRUE
;
1960 return (slog_found
);
1964 spa_activate_log(spa_t
*spa
)
1966 vdev_t
*rvd
= spa
->spa_root_vdev
;
1968 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1970 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1971 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1972 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1974 if (tvd
->vdev_islog
)
1975 metaslab_group_activate(mg
);
1980 spa_reset_logs(spa_t
*spa
)
1984 error
= dmu_objset_find(spa_name(spa
), zil_reset
,
1985 NULL
, DS_FIND_CHILDREN
);
1988 * We successfully offlined the log device, sync out the
1989 * current txg so that the "stubby" block can be removed
1992 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1998 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
2000 for (int i
= 0; i
< sav
->sav_count
; i
++)
2001 spa_check_removed(sav
->sav_vdevs
[i
]);
2005 spa_claim_notify(zio_t
*zio
)
2007 spa_t
*spa
= zio
->io_spa
;
2012 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
2013 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
2014 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
2015 mutex_exit(&spa
->spa_props_lock
);
2018 typedef struct spa_load_error
{
2019 uint64_t sle_meta_count
;
2020 uint64_t sle_data_count
;
2024 spa_load_verify_done(zio_t
*zio
)
2026 blkptr_t
*bp
= zio
->io_bp
;
2027 spa_load_error_t
*sle
= zio
->io_private
;
2028 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2029 int error
= zio
->io_error
;
2030 spa_t
*spa
= zio
->io_spa
;
2032 abd_free(zio
->io_abd
);
2034 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2035 type
!= DMU_OT_INTENT_LOG
)
2036 atomic_inc_64(&sle
->sle_meta_count
);
2038 atomic_inc_64(&sle
->sle_data_count
);
2041 mutex_enter(&spa
->spa_scrub_lock
);
2042 spa
->spa_load_verify_ios
--;
2043 cv_broadcast(&spa
->spa_scrub_io_cv
);
2044 mutex_exit(&spa
->spa_scrub_lock
);
2048 * Maximum number of concurrent scrub i/os to create while verifying
2049 * a pool while importing it.
2051 int spa_load_verify_maxinflight
= 10000;
2052 int spa_load_verify_metadata
= B_TRUE
;
2053 int spa_load_verify_data
= B_TRUE
;
2057 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2058 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2060 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
2063 * Note: normally this routine will not be called if
2064 * spa_load_verify_metadata is not set. However, it may be useful
2065 * to manually set the flag after the traversal has begun.
2067 if (!spa_load_verify_metadata
)
2069 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2073 size_t size
= BP_GET_PSIZE(bp
);
2075 mutex_enter(&spa
->spa_scrub_lock
);
2076 while (spa
->spa_load_verify_ios
>= spa_load_verify_maxinflight
)
2077 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2078 spa
->spa_load_verify_ios
++;
2079 mutex_exit(&spa
->spa_scrub_lock
);
2081 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2082 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2083 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2084 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2090 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2092 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2093 return (SET_ERROR(ENAMETOOLONG
));
2099 spa_load_verify(spa_t
*spa
)
2102 spa_load_error_t sle
= { 0 };
2103 zpool_rewind_policy_t policy
;
2104 boolean_t verify_ok
= B_FALSE
;
2107 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
2109 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
2112 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2113 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2114 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2116 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2120 rio
= zio_root(spa
, NULL
, &sle
,
2121 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2123 if (spa_load_verify_metadata
) {
2124 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2125 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2126 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2129 (void) zio_wait(rio
);
2131 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2132 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2134 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
2135 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
2139 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2140 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2142 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2143 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2144 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2145 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2146 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2147 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2148 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2150 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2154 if (error
!= ENXIO
&& error
!= EIO
)
2155 error
= SET_ERROR(EIO
);
2159 return (verify_ok
? 0 : EIO
);
2163 * Find a value in the pool props object.
2166 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2168 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2169 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2173 * Find a value in the pool directory object.
2176 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2178 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2179 name
, sizeof (uint64_t), 1, val
));
2183 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2185 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2186 return (SET_ERROR(err
));
2190 spa_spawn_aux_threads(spa_t
*spa
)
2192 ASSERT(spa_writeable(spa
));
2194 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2196 spa_start_indirect_condensing_thread(spa
);
2200 * Fix up config after a partly-completed split. This is done with the
2201 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2202 * pool have that entry in their config, but only the splitting one contains
2203 * a list of all the guids of the vdevs that are being split off.
2205 * This function determines what to do with that list: either rejoin
2206 * all the disks to the pool, or complete the splitting process. To attempt
2207 * the rejoin, each disk that is offlined is marked online again, and
2208 * we do a reopen() call. If the vdev label for every disk that was
2209 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2210 * then we call vdev_split() on each disk, and complete the split.
2212 * Otherwise we leave the config alone, with all the vdevs in place in
2213 * the original pool.
2216 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2223 boolean_t attempt_reopen
;
2225 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2228 /* check that the config is complete */
2229 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2230 &glist
, &gcount
) != 0)
2233 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2235 /* attempt to online all the vdevs & validate */
2236 attempt_reopen
= B_TRUE
;
2237 for (i
= 0; i
< gcount
; i
++) {
2238 if (glist
[i
] == 0) /* vdev is hole */
2241 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2242 if (vd
[i
] == NULL
) {
2244 * Don't bother attempting to reopen the disks;
2245 * just do the split.
2247 attempt_reopen
= B_FALSE
;
2249 /* attempt to re-online it */
2250 vd
[i
]->vdev_offline
= B_FALSE
;
2254 if (attempt_reopen
) {
2255 vdev_reopen(spa
->spa_root_vdev
);
2257 /* check each device to see what state it's in */
2258 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2259 if (vd
[i
] != NULL
&&
2260 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2267 * If every disk has been moved to the new pool, or if we never
2268 * even attempted to look at them, then we split them off for
2271 if (!attempt_reopen
|| gcount
== extracted
) {
2272 for (i
= 0; i
< gcount
; i
++)
2275 vdev_reopen(spa
->spa_root_vdev
);
2278 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2282 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2283 boolean_t trust_config
)
2285 nvlist_t
*config
= spa
->spa_config
;
2286 char *ereport
= FM_EREPORT_ZFS_POOL
;
2292 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2293 return (SET_ERROR(EINVAL
));
2295 ASSERT(spa
->spa_comment
== NULL
);
2296 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2297 spa
->spa_comment
= spa_strdup(comment
);
2300 * Versioning wasn't explicitly added to the label until later, so if
2301 * it's not present treat it as the initial version.
2303 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2304 &spa
->spa_ubsync
.ub_version
) != 0)
2305 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2307 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2308 &spa
->spa_config_txg
);
2310 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2311 spa_guid_exists(pool_guid
, 0)) {
2312 error
= SET_ERROR(EEXIST
);
2314 spa
->spa_config_guid
= pool_guid
;
2316 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2318 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2322 nvlist_free(spa
->spa_load_info
);
2323 spa
->spa_load_info
= fnvlist_alloc();
2325 gethrestime(&spa
->spa_loaded_ts
);
2326 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2327 trust_config
, &ereport
);
2331 * Don't count references from objsets that are already closed
2332 * and are making their way through the eviction process.
2334 spa_evicting_os_wait(spa
);
2335 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2337 if (error
!= EEXIST
) {
2338 spa
->spa_loaded_ts
.tv_sec
= 0;
2339 spa
->spa_loaded_ts
.tv_nsec
= 0;
2341 if (error
!= EBADF
) {
2342 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, NULL
, 0, 0);
2345 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2353 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2354 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2355 * spa's per-vdev ZAP list.
2358 vdev_count_verify_zaps(vdev_t
*vd
)
2360 spa_t
*spa
= vd
->vdev_spa
;
2363 if (vd
->vdev_top_zap
!= 0) {
2365 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2366 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2368 if (vd
->vdev_leaf_zap
!= 0) {
2370 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2371 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2374 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2375 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2383 * Determine whether the activity check is required.
2386 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
2390 uint64_t hostid
= 0;
2391 uint64_t tryconfig_txg
= 0;
2392 uint64_t tryconfig_timestamp
= 0;
2395 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2396 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2397 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2399 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2400 &tryconfig_timestamp
);
2403 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2406 * Disable the MMP activity check - This is used by zdb which
2407 * is intended to be used on potentially active pools.
2409 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
2413 * Skip the activity check when the MMP feature is disabled.
2415 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
2418 * If the tryconfig_* values are nonzero, they are the results of an
2419 * earlier tryimport. If they match the uberblock we just found, then
2420 * the pool has not changed and we return false so we do not test a
2423 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
2424 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
)
2428 * Allow the activity check to be skipped when importing the pool
2429 * on the same host which last imported it. Since the hostid from
2430 * configuration may be stale use the one read from the label.
2432 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
2433 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
2435 if (hostid
== spa_get_hostid())
2439 * Skip the activity test when the pool was cleanly exported.
2441 if (state
!= POOL_STATE_ACTIVE
)
2448 * Perform the import activity check. If the user canceled the import or
2449 * we detected activity then fail.
2452 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2454 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2455 uint64_t txg
= ub
->ub_txg
;
2456 uint64_t timestamp
= ub
->ub_timestamp
;
2457 uint64_t import_delay
= NANOSEC
;
2458 hrtime_t import_expire
;
2459 nvlist_t
*mmp_label
= NULL
;
2460 vdev_t
*rvd
= spa
->spa_root_vdev
;
2465 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
2466 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2470 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2471 * during the earlier tryimport. If the txg recorded there is 0 then
2472 * the pool is known to be active on another host.
2474 * Otherwise, the pool might be in use on another node. Check for
2475 * changes in the uberblocks on disk if necessary.
2477 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2478 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
2479 ZPOOL_CONFIG_LOAD_INFO
);
2481 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
2482 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
2483 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2484 error
= SET_ERROR(EREMOTEIO
);
2490 * Preferentially use the zfs_multihost_interval from the node which
2491 * last imported the pool. This value is stored in an MMP uberblock as.
2493 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2495 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
)
2496 import_delay
= MAX(import_delay
, import_intervals
*
2497 ub
->ub_mmp_delay
* MAX(vdev_count_leaves(spa
), 1));
2499 /* Apply a floor using the local default values. */
2500 import_delay
= MAX(import_delay
, import_intervals
*
2501 MSEC2NSEC(MAX(zfs_multihost_interval
, MMP_MIN_INTERVAL
)));
2503 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu import_intervals=%u "
2504 "leaves=%u", import_delay
, ub
->ub_mmp_delay
, import_intervals
,
2505 vdev_count_leaves(spa
));
2507 /* Add a small random factor in case of simultaneous imports (0-25%) */
2508 import_expire
= gethrtime() + import_delay
+
2509 (import_delay
* spa_get_random(250) / 1000);
2511 while (gethrtime() < import_expire
) {
2512 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2514 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
) {
2515 error
= SET_ERROR(EREMOTEIO
);
2520 nvlist_free(mmp_label
);
2524 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
2526 error
= SET_ERROR(EINTR
);
2534 mutex_destroy(&mtx
);
2538 * If the pool is determined to be active store the status in the
2539 * spa->spa_load_info nvlist. If the remote hostname or hostid are
2540 * available from configuration read from disk store them as well.
2541 * This allows 'zpool import' to generate a more useful message.
2543 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
2544 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2545 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
2547 if (error
== EREMOTEIO
) {
2548 char *hostname
= "<unknown>";
2549 uint64_t hostid
= 0;
2552 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
2553 hostname
= fnvlist_lookup_string(mmp_label
,
2554 ZPOOL_CONFIG_HOSTNAME
);
2555 fnvlist_add_string(spa
->spa_load_info
,
2556 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
2559 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
2560 hostid
= fnvlist_lookup_uint64(mmp_label
,
2561 ZPOOL_CONFIG_HOSTID
);
2562 fnvlist_add_uint64(spa
->spa_load_info
,
2563 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
2567 fnvlist_add_uint64(spa
->spa_load_info
,
2568 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
2569 fnvlist_add_uint64(spa
->spa_load_info
,
2570 ZPOOL_CONFIG_MMP_TXG
, 0);
2572 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
2576 nvlist_free(mmp_label
);
2582 spa_ld_parse_config(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2583 spa_load_state_t state
, spa_import_type_t type
)
2586 nvlist_t
*nvtree
= NULL
;
2590 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2592 spa
->spa_load_state
= state
;
2594 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
))
2595 return (SET_ERROR(EINVAL
));
2597 parse
= (type
== SPA_IMPORT_EXISTING
?
2598 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2601 * Create "The Godfather" zio to hold all async IOs
2603 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2605 for (int i
= 0; i
< max_ncpus
; i
++) {
2606 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2607 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2608 ZIO_FLAG_GODFATHER
);
2612 * Parse the configuration into a vdev tree. We explicitly set the
2613 * value that will be returned by spa_version() since parsing the
2614 * configuration requires knowing the version number.
2616 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2617 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
2618 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2623 ASSERT(spa
->spa_root_vdev
== rvd
);
2624 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2625 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2627 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2628 ASSERT(spa_guid(spa
) == pool_guid
);
2635 spa_ld_open_vdevs(spa_t
*spa
)
2639 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2640 error
= vdev_open(spa
->spa_root_vdev
);
2641 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2647 spa_ld_validate_vdevs(spa_t
*spa
, spa_import_type_t type
,
2648 boolean_t trust_config
)
2651 vdev_t
*rvd
= spa
->spa_root_vdev
;
2654 * We need to validate the vdev labels against the configuration that
2655 * we have in hand, which is dependent on the setting of trust_config.
2656 * If trust_config is true then we're validating the vdev labels based
2657 * on that config. Otherwise, we're validating against the cached
2658 * config (zpool.cache) that was read when we loaded the zfs module, and
2659 * then later we will recursively call spa_load() and validate against
2662 * If we're assembling a new pool that's been split off from an
2663 * existing pool, the labels haven't yet been updated so we skip
2664 * validation for now.
2666 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2667 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2668 error
= vdev_validate(rvd
, trust_config
);
2669 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2674 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2675 return (SET_ERROR(ENXIO
));
2682 spa_ld_select_uberblock(spa_t
*spa
, nvlist_t
*config
, spa_import_type_t type
,
2683 boolean_t trust_config
)
2685 vdev_t
*rvd
= spa
->spa_root_vdev
;
2687 uberblock_t
*ub
= &spa
->spa_uberblock
;
2689 boolean_t activity_check
= B_FALSE
;
2692 * Find the best uberblock.
2694 vdev_uberblock_load(rvd
, ub
, &label
);
2697 * If we weren't able to find a single valid uberblock, return failure.
2699 if (ub
->ub_txg
== 0) {
2701 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2705 * For pools which have the multihost property on determine if the
2706 * pool is truly inactive and can be safely imported. Prevent
2707 * hosts which don't have a hostid set from importing the pool.
2709 activity_check
= spa_activity_check_required(spa
, ub
, label
, config
);
2710 if (activity_check
) {
2711 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
2712 spa_get_hostid() == 0) {
2714 fnvlist_add_uint64(spa
->spa_load_info
,
2715 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
2716 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
2719 int error
= spa_activity_check(spa
, ub
, config
);
2725 fnvlist_add_uint64(spa
->spa_load_info
,
2726 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
2727 fnvlist_add_uint64(spa
->spa_load_info
,
2728 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
2732 * If the pool has an unsupported version we can't open it.
2734 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2736 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2739 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2743 * If we weren't able to find what's necessary for reading the
2744 * MOS in the label, return failure.
2746 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2747 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2749 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2754 * Update our in-core representation with the definitive values
2757 nvlist_free(spa
->spa_label_features
);
2758 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2764 * Look through entries in the label nvlist's features_for_read. If
2765 * there is a feature listed there which we don't understand then we
2766 * cannot open a pool.
2768 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2769 nvlist_t
*unsup_feat
;
2771 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2774 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
2776 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2777 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2778 VERIFY(nvlist_add_string(unsup_feat
,
2779 nvpair_name(nvp
), "") == 0);
2783 if (!nvlist_empty(unsup_feat
)) {
2784 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2785 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2786 nvlist_free(unsup_feat
);
2787 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2791 nvlist_free(unsup_feat
);
2795 * If the vdev guid sum doesn't match the uberblock, we have an
2796 * incomplete configuration. We first check to see if the pool
2797 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2798 * If it is, defer the vdev_guid_sum check till later so we
2799 * can handle missing vdevs.
2801 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2802 &children
) != 0 && trust_config
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2803 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2804 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2806 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2807 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2808 spa_try_repair(spa
, config
);
2809 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2810 nvlist_free(spa
->spa_config_splitting
);
2811 spa
->spa_config_splitting
= NULL
;
2815 * Initialize internal SPA structures.
2817 spa
->spa_state
= POOL_STATE_ACTIVE
;
2818 spa
->spa_ubsync
= spa
->spa_uberblock
;
2819 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2820 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2821 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2822 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2823 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2824 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2830 spa_ld_open_rootbp(spa_t
*spa
)
2833 vdev_t
*rvd
= spa
->spa_root_vdev
;
2835 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2837 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2838 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2844 spa_ld_validate_config(spa_t
*spa
, spa_import_type_t type
)
2846 vdev_t
*rvd
= spa
->spa_root_vdev
;
2848 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2849 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2852 * Validate the config, using the MOS config to fill in any
2853 * information which might be missing. If we fail to validate
2854 * the config then declare the pool unfit for use. If we're
2855 * assembling a pool from a split, the log is not transferred
2858 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2859 nvlist_t
*mos_config
;
2860 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
2861 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2863 if (!spa_config_valid(spa
, mos_config
)) {
2864 nvlist_free(mos_config
);
2865 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2868 nvlist_free(mos_config
);
2871 * Now that we've validated the config, check the state of the
2872 * root vdev. If it can't be opened, it indicates one or
2873 * more toplevel vdevs are faulted.
2875 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2876 return (SET_ERROR(ENXIO
));
2883 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
2886 vdev_t
*rvd
= spa
->spa_root_vdev
;
2889 * Everything that we read before spa_remove_init() must be stored
2890 * on concreted vdevs. Therefore we do this as early as possible.
2892 if (spa_remove_init(spa
) != 0)
2893 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2896 * Retrieve information needed to condense indirect vdev mappings.
2898 error
= spa_condense_init(spa
);
2900 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
2907 spa_ld_check_features(spa_t
*spa
, spa_load_state_t state
,
2908 boolean_t
*missing_feat_writep
)
2911 vdev_t
*rvd
= spa
->spa_root_vdev
;
2913 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2914 boolean_t missing_feat_read
= B_FALSE
;
2915 nvlist_t
*unsup_feat
, *enabled_feat
;
2917 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2918 &spa
->spa_feat_for_read_obj
) != 0) {
2919 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2922 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2923 &spa
->spa_feat_for_write_obj
) != 0) {
2924 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2927 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2928 &spa
->spa_feat_desc_obj
) != 0) {
2929 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2932 enabled_feat
= fnvlist_alloc();
2933 unsup_feat
= fnvlist_alloc();
2935 if (!spa_features_check(spa
, B_FALSE
,
2936 unsup_feat
, enabled_feat
))
2937 missing_feat_read
= B_TRUE
;
2939 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2940 if (!spa_features_check(spa
, B_TRUE
,
2941 unsup_feat
, enabled_feat
)) {
2942 *missing_feat_writep
= B_TRUE
;
2946 fnvlist_add_nvlist(spa
->spa_load_info
,
2947 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2949 if (!nvlist_empty(unsup_feat
)) {
2950 fnvlist_add_nvlist(spa
->spa_load_info
,
2951 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2954 fnvlist_free(enabled_feat
);
2955 fnvlist_free(unsup_feat
);
2957 if (!missing_feat_read
) {
2958 fnvlist_add_boolean(spa
->spa_load_info
,
2959 ZPOOL_CONFIG_CAN_RDONLY
);
2963 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2964 * twofold: to determine whether the pool is available for
2965 * import in read-write mode and (if it is not) whether the
2966 * pool is available for import in read-only mode. If the pool
2967 * is available for import in read-write mode, it is displayed
2968 * as available in userland; if it is not available for import
2969 * in read-only mode, it is displayed as unavailable in
2970 * userland. If the pool is available for import in read-only
2971 * mode but not read-write mode, it is displayed as unavailable
2972 * in userland with a special note that the pool is actually
2973 * available for open in read-only mode.
2975 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2976 * missing a feature for write, we must first determine whether
2977 * the pool can be opened read-only before returning to
2978 * userland in order to know whether to display the
2979 * abovementioned note.
2981 if (missing_feat_read
|| (*missing_feat_writep
&&
2982 spa_writeable(spa
))) {
2983 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2988 * Load refcounts for ZFS features from disk into an in-memory
2989 * cache during SPA initialization.
2991 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
2994 error
= feature_get_refcount_from_disk(spa
,
2995 &spa_feature_table
[i
], &refcount
);
2997 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2998 } else if (error
== ENOTSUP
) {
2999 spa
->spa_feat_refcount_cache
[i
] =
3000 SPA_FEATURE_DISABLED
;
3002 return (spa_vdev_err(rvd
,
3003 VDEV_AUX_CORRUPT_DATA
, EIO
));
3008 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
3009 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
3010 &spa
->spa_feat_enabled_txg_obj
) != 0)
3011 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3018 spa_ld_load_special_directories(spa_t
*spa
)
3021 vdev_t
*rvd
= spa
->spa_root_vdev
;
3023 spa
->spa_is_initializing
= B_TRUE
;
3024 error
= dsl_pool_open(spa
->spa_dsl_pool
);
3025 spa
->spa_is_initializing
= B_FALSE
;
3027 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3033 spa_ld_prepare_for_reload(spa_t
*spa
, int orig_mode
)
3035 vdev_t
*rvd
= spa
->spa_root_vdev
;
3038 nvlist_t
*policy
= NULL
;
3039 nvlist_t
*mos_config
;
3041 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
3042 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3044 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
3045 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
3047 unsigned long myhostid
= 0;
3049 VERIFY(nvlist_lookup_string(mos_config
,
3050 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
3052 myhostid
= spa_get_hostid();
3053 if (hostid
&& myhostid
&& hostid
!= myhostid
) {
3054 nvlist_free(mos_config
);
3055 return (SET_ERROR(EBADF
));
3058 if (nvlist_lookup_nvlist(spa
->spa_config
,
3059 ZPOOL_REWIND_POLICY
, &policy
) == 0)
3060 VERIFY(nvlist_add_nvlist(mos_config
,
3061 ZPOOL_REWIND_POLICY
, policy
) == 0);
3063 spa_config_set(spa
, mos_config
);
3065 spa_deactivate(spa
);
3066 spa_activate(spa
, orig_mode
);
3072 spa_ld_get_props(spa_t
*spa
)
3076 vdev_t
*rvd
= spa
->spa_root_vdev
;
3078 /* Grab the checksum salt from the MOS. */
3079 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
3080 DMU_POOL_CHECKSUM_SALT
, 1,
3081 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
3082 spa
->spa_cksum_salt
.zcs_bytes
);
3083 if (error
== ENOENT
) {
3084 /* Generate a new salt for subsequent use */
3085 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
3086 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
3087 } else if (error
!= 0) {
3088 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3091 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
3092 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3093 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
3095 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3098 * Load the bit that tells us to use the new accounting function
3099 * (raid-z deflation). If we have an older pool, this will not
3102 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
3103 if (error
!= 0 && error
!= ENOENT
)
3104 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3106 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
3107 &spa
->spa_creation_version
);
3108 if (error
!= 0 && error
!= ENOENT
)
3109 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3112 * Load the persistent error log. If we have an older pool, this will
3115 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
3116 if (error
!= 0 && error
!= ENOENT
)
3117 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3119 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
3120 &spa
->spa_errlog_scrub
);
3121 if (error
!= 0 && error
!= ENOENT
)
3122 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3125 * Load the history object. If we have an older pool, this
3126 * will not be present.
3128 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
3129 if (error
!= 0 && error
!= ENOENT
)
3130 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3133 * Load the per-vdev ZAP map. If we have an older pool, this will not
3134 * be present; in this case, defer its creation to a later time to
3135 * avoid dirtying the MOS this early / out of sync context. See
3136 * spa_sync_config_object.
3139 /* The sentinel is only available in the MOS config. */
3140 nvlist_t
*mos_config
;
3141 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
3142 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3144 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
3145 &spa
->spa_all_vdev_zaps
);
3147 if (error
== ENOENT
) {
3148 VERIFY(!nvlist_exists(mos_config
,
3149 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
3150 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
3151 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3152 } else if (error
!= 0) {
3153 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3154 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
3156 * An older version of ZFS overwrote the sentinel value, so
3157 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3158 * destruction to later; see spa_sync_config_object.
3160 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
3162 * We're assuming that no vdevs have had their ZAPs created
3163 * before this. Better be sure of it.
3165 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3167 nvlist_free(mos_config
);
3169 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3171 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
3172 if (error
&& error
!= ENOENT
)
3173 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3176 uint64_t autoreplace
;
3178 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3179 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3180 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3181 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3182 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3183 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
3184 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3185 &spa
->spa_dedup_ditto
);
3187 spa
->spa_autoreplace
= (autoreplace
!= 0);
3194 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
3197 vdev_t
*rvd
= spa
->spa_root_vdev
;
3200 * If we're assembling the pool from the split-off vdevs of
3201 * an existing pool, we don't want to attach the spares & cache
3206 * Load any hot spares for this pool.
3208 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
3209 if (error
!= 0 && error
!= ENOENT
)
3210 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3211 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3212 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3213 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3214 &spa
->spa_spares
.sav_config
) != 0)
3215 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3217 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3218 spa_load_spares(spa
);
3219 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3220 } else if (error
== 0) {
3221 spa
->spa_spares
.sav_sync
= B_TRUE
;
3225 * Load any level 2 ARC devices for this pool.
3227 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3228 &spa
->spa_l2cache
.sav_object
);
3229 if (error
!= 0 && error
!= ENOENT
)
3230 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3231 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3232 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3233 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3234 &spa
->spa_l2cache
.sav_config
) != 0)
3235 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3237 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3238 spa_load_l2cache(spa
);
3239 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3240 } else if (error
== 0) {
3241 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3248 spa_ld_load_vdev_metadata(spa_t
*spa
, spa_load_state_t state
)
3251 vdev_t
*rvd
= spa
->spa_root_vdev
;
3254 * If the 'multihost' property is set, then never allow a pool to
3255 * be imported when the system hostid is zero. The exception to
3256 * this rule is zdb which is always allowed to access pools.
3258 if (spa_multihost(spa
) && spa_get_hostid() == 0 &&
3259 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
3260 fnvlist_add_uint64(spa
->spa_load_info
,
3261 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3262 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3266 * If the 'autoreplace' property is set, then post a resource notifying
3267 * the ZFS DE that it should not issue any faults for unopenable
3268 * devices. We also iterate over the vdevs, and post a sysevent for any
3269 * unopenable vdevs so that the normal autoreplace handler can take
3272 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
3273 spa_check_removed(spa
->spa_root_vdev
);
3275 * For the import case, this is done in spa_import(), because
3276 * at this point we're using the spare definitions from
3277 * the MOS config, not necessarily from the userland config.
3279 if (state
!= SPA_LOAD_IMPORT
) {
3280 spa_aux_check_removed(&spa
->spa_spares
);
3281 spa_aux_check_removed(&spa
->spa_l2cache
);
3286 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3288 error
= vdev_load(rvd
);
3290 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3294 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3296 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3297 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3298 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3304 spa_ld_load_dedup_tables(spa_t
*spa
)
3307 vdev_t
*rvd
= spa
->spa_root_vdev
;
3309 error
= ddt_load(spa
);
3311 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3317 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
3319 vdev_t
*rvd
= spa
->spa_root_vdev
;
3321 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
) &&
3322 spa_check_logs(spa
)) {
3323 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3324 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
3331 spa_ld_verify_pool_data(spa_t
*spa
, spa_load_state_t state
)
3334 vdev_t
*rvd
= spa
->spa_root_vdev
;
3337 * We've successfully opened the pool, verify that we're ready
3338 * to start pushing transactions.
3340 if (state
!= SPA_LOAD_TRYIMPORT
) {
3341 error
= spa_load_verify(spa
);
3343 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3352 spa_ld_claim_log_blocks(spa_t
*spa
)
3355 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3358 * Claim log blocks that haven't been committed yet.
3359 * This must all happen in a single txg.
3360 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3361 * invoked from zil_claim_log_block()'s i/o done callback.
3362 * Price of rollback is that we abandon the log.
3364 spa
->spa_claiming
= B_TRUE
;
3366 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3367 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3368 zil_claim
, tx
, DS_FIND_CHILDREN
);
3371 spa
->spa_claiming
= B_FALSE
;
3373 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3377 spa_ld_check_for_config_update(spa_t
*spa
, spa_load_state_t state
,
3378 int64_t config_cache_txg
)
3380 vdev_t
*rvd
= spa
->spa_root_vdev
;
3381 int need_update
= B_FALSE
;
3384 * If the config cache is stale, or we have uninitialized
3385 * metaslabs (see spa_vdev_add()), then update the config.
3387 * If this is a verbatim import, trust the current
3388 * in-core spa_config and update the disk labels.
3390 if (config_cache_txg
!= spa
->spa_config_txg
||
3391 state
== SPA_LOAD_IMPORT
||
3392 state
== SPA_LOAD_RECOVER
||
3393 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3394 need_update
= B_TRUE
;
3396 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
3397 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3398 need_update
= B_TRUE
;
3401 * Update the config cache asychronously in case we're the
3402 * root pool, in which case the config cache isn't writable yet.
3405 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3409 * Load an existing storage pool, using the config provided. This config
3410 * describes which vdevs are part of the pool and is later validated against
3411 * partial configs present in each vdev's label and an entire copy of the
3412 * config stored in the MOS.
3415 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
3416 spa_load_state_t state
, spa_import_type_t type
, boolean_t trust_config
,
3420 uint64_t config_cache_txg
= spa
->spa_config_txg
;
3421 int orig_mode
= spa
->spa_mode
;
3422 boolean_t missing_feat_write
= B_FALSE
;
3425 * If this is an untrusted config, first access the pool in read-only
3426 * mode. We will then retrieve a trusted copy of the config from the MOS
3427 * and use it to reopen the pool in read-write mode.
3430 spa
->spa_mode
= FREAD
;
3433 * Parse the config provided to create a vdev tree.
3435 error
= spa_ld_parse_config(spa
, pool_guid
, config
, state
, type
);
3440 * Now that we have the vdev tree, try to open each vdev. This involves
3441 * opening the underlying physical device, retrieving its geometry and
3442 * probing the vdev with a dummy I/O. The state of each vdev will be set
3443 * based on the success of those operations. After this we'll be ready
3444 * to read from the vdevs.
3446 error
= spa_ld_open_vdevs(spa
);
3451 * Read the label of each vdev and make sure that the GUIDs stored
3452 * there match the GUIDs in the config provided.
3454 error
= spa_ld_validate_vdevs(spa
, type
, trust_config
);
3459 * Read vdev labels to find the best uberblock (i.e. latest, unless
3460 * spa_load_max_txg is set) and store it in spa_uberblock. We get the
3461 * list of features required to read blkptrs in the MOS from the vdev
3462 * label with the best uberblock and verify that our version of zfs
3463 * supports them all.
3465 error
= spa_ld_select_uberblock(spa
, config
, type
, trust_config
);
3470 * Pass that uberblock to the dsl_pool layer which will open the root
3471 * blkptr. This blkptr points to the latest version of the MOS and will
3472 * allow us to read its contents.
3474 error
= spa_ld_open_rootbp(spa
);
3479 * Retrieve the config stored in the MOS and use it to validate the
3480 * config provided. Also extract some information from the MOS config
3481 * to update our vdev tree.
3483 error
= spa_ld_validate_config(spa
, type
);
3488 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3489 * from the pool and their contents were re-mapped to other vdevs. Note
3490 * that everything that we read before this step must have been
3491 * rewritten on concrete vdevs after the last device removal was
3492 * initiated. Otherwise we could be reading from indirect vdevs before
3493 * we have loaded their mappings.
3495 error
= spa_ld_open_indirect_vdev_metadata(spa
);
3500 * Retrieve the full list of active features from the MOS and check if
3501 * they are all supported.
3503 error
= spa_ld_check_features(spa
, state
, &missing_feat_write
);
3508 * Load several special directories from the MOS needed by the dsl_pool
3511 error
= spa_ld_load_special_directories(spa
);
3516 * If the config provided is not trusted, discard it and use the config
3517 * from the MOS to reload the pool.
3519 if (!trust_config
) {
3520 error
= spa_ld_prepare_for_reload(spa
, orig_mode
);
3523 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
3527 * Retrieve pool properties from the MOS.
3529 error
= spa_ld_get_props(spa
);
3534 * Retrieve the list of auxiliary devices - cache devices and spares -
3537 error
= spa_ld_open_aux_vdevs(spa
, type
);
3542 * Load the metadata for all vdevs. Also check if unopenable devices
3543 * should be autoreplaced.
3545 error
= spa_ld_load_vdev_metadata(spa
, state
);
3549 error
= spa_ld_load_dedup_tables(spa
);
3554 * Verify the logs now to make sure we don't have any unexpected errors
3555 * when we claim log blocks later.
3557 error
= spa_ld_verify_logs(spa
, type
, ereport
);
3561 if (missing_feat_write
) {
3562 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
3565 * At this point, we know that we can open the pool in
3566 * read-only mode but not read-write mode. We now have enough
3567 * information and can return to userland.
3569 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
3574 * Traverse the last txgs to make sure the pool was left off in a safe
3575 * state. When performing an extreme rewind, we verify the whole pool,
3576 * which can take a very long time.
3578 error
= spa_ld_verify_pool_data(spa
, state
);
3583 * Calculate the deflated space for the pool. This must be done before
3584 * we write anything to the pool because we'd need to update the space
3585 * accounting using the deflated sizes.
3587 spa_update_dspace(spa
);
3590 * We have now retrieved all the information we needed to open the
3591 * pool. If we are importing the pool in read-write mode, a few
3592 * additional steps must be performed to finish the import.
3594 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
3595 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3596 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
3599 * Traverse the ZIL and claim all blocks.
3601 spa_ld_claim_log_blocks(spa
);
3604 * Kick-off the syncing thread.
3606 spa
->spa_sync_on
= B_TRUE
;
3607 txg_sync_start(spa
->spa_dsl_pool
);
3608 mmp_thread_start(spa
);
3611 * Wait for all claims to sync. We sync up to the highest
3612 * claimed log block birth time so that claimed log blocks
3613 * don't appear to be from the future. spa_claim_max_txg
3614 * will have been set for us by ZIL traversal operations
3617 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3620 * Check if we need to request an update of the config. On the
3621 * next sync, we would update the config stored in vdev labels
3622 * and the cachefile (by default /etc/zfs/zpool.cache).
3624 spa_ld_check_for_config_update(spa
, state
, config_cache_txg
);
3627 * Check all DTLs to see if anything needs resilvering.
3629 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3630 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
))
3631 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3634 * Log the fact that we booted up (so that we can detect if
3635 * we rebooted in the middle of an operation).
3637 spa_history_log_version(spa
, "open", NULL
);
3640 * Delete any inconsistent datasets.
3642 (void) dmu_objset_find(spa_name(spa
),
3643 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3646 * Clean up any stale temporary dataset userrefs.
3648 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3650 spa_restart_removal(spa
);
3652 spa_spawn_aux_threads(spa
);
3659 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int trust_config
)
3661 int mode
= spa
->spa_mode
;
3664 spa_deactivate(spa
);
3666 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3668 spa_activate(spa
, mode
);
3669 spa_async_suspend(spa
);
3671 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, trust_config
));
3675 * If spa_load() fails this function will try loading prior txg's. If
3676 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3677 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3678 * function will not rewind the pool and will return the same error as
3682 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int trust_config
,
3683 uint64_t max_request
, int rewind_flags
)
3685 nvlist_t
*loadinfo
= NULL
;
3686 nvlist_t
*config
= NULL
;
3687 int load_error
, rewind_error
;
3688 uint64_t safe_rewind_txg
;
3691 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3692 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3693 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3695 spa
->spa_load_max_txg
= max_request
;
3696 if (max_request
!= UINT64_MAX
)
3697 spa
->spa_extreme_rewind
= B_TRUE
;
3700 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
3702 if (load_error
== 0)
3705 if (spa
->spa_root_vdev
!= NULL
)
3706 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3708 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3709 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3711 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3712 nvlist_free(config
);
3713 return (load_error
);
3716 if (state
== SPA_LOAD_RECOVER
) {
3717 /* Price of rolling back is discarding txgs, including log */
3718 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3721 * If we aren't rolling back save the load info from our first
3722 * import attempt so that we can restore it after attempting
3725 loadinfo
= spa
->spa_load_info
;
3726 spa
->spa_load_info
= fnvlist_alloc();
3729 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3730 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
3731 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
3732 TXG_INITIAL
: safe_rewind_txg
;
3735 * Continue as long as we're finding errors, we're still within
3736 * the acceptable rewind range, and we're still finding uberblocks
3738 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
3739 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
3740 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
3741 spa
->spa_extreme_rewind
= B_TRUE
;
3742 rewind_error
= spa_load_retry(spa
, state
, trust_config
);
3745 spa
->spa_extreme_rewind
= B_FALSE
;
3746 spa
->spa_load_max_txg
= UINT64_MAX
;
3748 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
3749 spa_config_set(spa
, config
);
3751 nvlist_free(config
);
3753 if (state
== SPA_LOAD_RECOVER
) {
3754 ASSERT3P(loadinfo
, ==, NULL
);
3755 return (rewind_error
);
3757 /* Store the rewind info as part of the initial load info */
3758 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
3759 spa
->spa_load_info
);
3761 /* Restore the initial load info */
3762 fnvlist_free(spa
->spa_load_info
);
3763 spa
->spa_load_info
= loadinfo
;
3765 return (load_error
);
3772 * The import case is identical to an open except that the configuration is sent
3773 * down from userland, instead of grabbed from the configuration cache. For the
3774 * case of an open, the pool configuration will exist in the
3775 * POOL_STATE_UNINITIALIZED state.
3777 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3778 * the same time open the pool, without having to keep around the spa_t in some
3782 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
3786 spa_load_state_t state
= SPA_LOAD_OPEN
;
3788 int locked
= B_FALSE
;
3789 int firstopen
= B_FALSE
;
3794 * As disgusting as this is, we need to support recursive calls to this
3795 * function because dsl_dir_open() is called during spa_load(), and ends
3796 * up calling spa_open() again. The real fix is to figure out how to
3797 * avoid dsl_dir_open() calling this in the first place.
3799 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
3800 mutex_enter(&spa_namespace_lock
);
3804 if ((spa
= spa_lookup(pool
)) == NULL
) {
3806 mutex_exit(&spa_namespace_lock
);
3807 return (SET_ERROR(ENOENT
));
3810 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3811 zpool_rewind_policy_t policy
;
3815 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3817 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3818 state
= SPA_LOAD_RECOVER
;
3820 spa_activate(spa
, spa_mode_global
);
3822 if (state
!= SPA_LOAD_RECOVER
)
3823 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3825 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3826 policy
.zrp_request
);
3828 if (error
== EBADF
) {
3830 * If vdev_validate() returns failure (indicated by
3831 * EBADF), it indicates that one of the vdevs indicates
3832 * that the pool has been exported or destroyed. If
3833 * this is the case, the config cache is out of sync and
3834 * we should remove the pool from the namespace.
3837 spa_deactivate(spa
);
3838 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
3841 mutex_exit(&spa_namespace_lock
);
3842 return (SET_ERROR(ENOENT
));
3847 * We can't open the pool, but we still have useful
3848 * information: the state of each vdev after the
3849 * attempted vdev_open(). Return this to the user.
3851 if (config
!= NULL
&& spa
->spa_config
) {
3852 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3854 VERIFY(nvlist_add_nvlist(*config
,
3855 ZPOOL_CONFIG_LOAD_INFO
,
3856 spa
->spa_load_info
) == 0);
3859 spa_deactivate(spa
);
3860 spa
->spa_last_open_failed
= error
;
3862 mutex_exit(&spa_namespace_lock
);
3868 spa_open_ref(spa
, tag
);
3871 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3874 * If we've recovered the pool, pass back any information we
3875 * gathered while doing the load.
3877 if (state
== SPA_LOAD_RECOVER
) {
3878 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3879 spa
->spa_load_info
) == 0);
3883 spa
->spa_last_open_failed
= 0;
3884 spa
->spa_last_ubsync_txg
= 0;
3885 spa
->spa_load_txg
= 0;
3886 mutex_exit(&spa_namespace_lock
);
3890 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
3898 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3901 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3905 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3907 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3911 * Lookup the given spa_t, incrementing the inject count in the process,
3912 * preventing it from being exported or destroyed.
3915 spa_inject_addref(char *name
)
3919 mutex_enter(&spa_namespace_lock
);
3920 if ((spa
= spa_lookup(name
)) == NULL
) {
3921 mutex_exit(&spa_namespace_lock
);
3924 spa
->spa_inject_ref
++;
3925 mutex_exit(&spa_namespace_lock
);
3931 spa_inject_delref(spa_t
*spa
)
3933 mutex_enter(&spa_namespace_lock
);
3934 spa
->spa_inject_ref
--;
3935 mutex_exit(&spa_namespace_lock
);
3939 * Add spares device information to the nvlist.
3942 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3952 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3954 if (spa
->spa_spares
.sav_count
== 0)
3957 VERIFY(nvlist_lookup_nvlist(config
,
3958 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3959 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3960 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3962 VERIFY(nvlist_add_nvlist_array(nvroot
,
3963 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3964 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3965 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3968 * Go through and find any spares which have since been
3969 * repurposed as an active spare. If this is the case, update
3970 * their status appropriately.
3972 for (i
= 0; i
< nspares
; i
++) {
3973 VERIFY(nvlist_lookup_uint64(spares
[i
],
3974 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3975 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3977 VERIFY(nvlist_lookup_uint64_array(
3978 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3979 (uint64_t **)&vs
, &vsc
) == 0);
3980 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3981 vs
->vs_aux
= VDEV_AUX_SPARED
;
3988 * Add l2cache device information to the nvlist, including vdev stats.
3991 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3994 uint_t i
, j
, nl2cache
;
4001 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
4003 if (spa
->spa_l2cache
.sav_count
== 0)
4006 VERIFY(nvlist_lookup_nvlist(config
,
4007 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
4008 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
4009 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
4010 if (nl2cache
!= 0) {
4011 VERIFY(nvlist_add_nvlist_array(nvroot
,
4012 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4013 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
4014 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
4017 * Update level 2 cache device stats.
4020 for (i
= 0; i
< nl2cache
; i
++) {
4021 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
4022 ZPOOL_CONFIG_GUID
, &guid
) == 0);
4025 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
4027 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
4028 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
4034 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
4035 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
4037 vdev_get_stats(vd
, vs
);
4038 vdev_config_generate_stats(vd
, l2cache
[i
]);
4045 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
4050 if (spa
->spa_feat_for_read_obj
!= 0) {
4051 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
4052 spa
->spa_feat_for_read_obj
);
4053 zap_cursor_retrieve(&zc
, &za
) == 0;
4054 zap_cursor_advance(&zc
)) {
4055 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
4056 za
.za_num_integers
== 1);
4057 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
4058 za
.za_first_integer
));
4060 zap_cursor_fini(&zc
);
4063 if (spa
->spa_feat_for_write_obj
!= 0) {
4064 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
4065 spa
->spa_feat_for_write_obj
);
4066 zap_cursor_retrieve(&zc
, &za
) == 0;
4067 zap_cursor_advance(&zc
)) {
4068 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
4069 za
.za_num_integers
== 1);
4070 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
4071 za
.za_first_integer
));
4073 zap_cursor_fini(&zc
);
4078 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
4082 for (i
= 0; i
< SPA_FEATURES
; i
++) {
4083 zfeature_info_t feature
= spa_feature_table
[i
];
4086 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
4089 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
4094 * Store a list of pool features and their reference counts in the
4097 * The first time this is called on a spa, allocate a new nvlist, fetch
4098 * the pool features and reference counts from disk, then save the list
4099 * in the spa. In subsequent calls on the same spa use the saved nvlist
4100 * and refresh its values from the cached reference counts. This
4101 * ensures we don't block here on I/O on a suspended pool so 'zpool
4102 * clear' can resume the pool.
4105 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
4109 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
4111 mutex_enter(&spa
->spa_feat_stats_lock
);
4112 features
= spa
->spa_feat_stats
;
4114 if (features
!= NULL
) {
4115 spa_feature_stats_from_cache(spa
, features
);
4117 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
4118 spa
->spa_feat_stats
= features
;
4119 spa_feature_stats_from_disk(spa
, features
);
4122 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
4125 mutex_exit(&spa
->spa_feat_stats_lock
);
4129 spa_get_stats(const char *name
, nvlist_t
**config
,
4130 char *altroot
, size_t buflen
)
4136 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
4140 * This still leaves a window of inconsistency where the spares
4141 * or l2cache devices could change and the config would be
4142 * self-inconsistent.
4144 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4146 if (*config
!= NULL
) {
4147 uint64_t loadtimes
[2];
4149 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
4150 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
4151 VERIFY(nvlist_add_uint64_array(*config
,
4152 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
4154 VERIFY(nvlist_add_uint64(*config
,
4155 ZPOOL_CONFIG_ERRCOUNT
,
4156 spa_get_errlog_size(spa
)) == 0);
4158 if (spa_suspended(spa
)) {
4159 VERIFY(nvlist_add_uint64(*config
,
4160 ZPOOL_CONFIG_SUSPENDED
,
4161 spa
->spa_failmode
) == 0);
4162 VERIFY(nvlist_add_uint64(*config
,
4163 ZPOOL_CONFIG_SUSPENDED_REASON
,
4164 spa
->spa_suspended
) == 0);
4167 spa_add_spares(spa
, *config
);
4168 spa_add_l2cache(spa
, *config
);
4169 spa_add_feature_stats(spa
, *config
);
4174 * We want to get the alternate root even for faulted pools, so we cheat
4175 * and call spa_lookup() directly.
4179 mutex_enter(&spa_namespace_lock
);
4180 spa
= spa_lookup(name
);
4182 spa_altroot(spa
, altroot
, buflen
);
4186 mutex_exit(&spa_namespace_lock
);
4188 spa_altroot(spa
, altroot
, buflen
);
4193 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4194 spa_close(spa
, FTAG
);
4201 * Validate that the auxiliary device array is well formed. We must have an
4202 * array of nvlists, each which describes a valid leaf vdev. If this is an
4203 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4204 * specified, as long as they are well-formed.
4207 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
4208 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
4209 vdev_labeltype_t label
)
4216 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4219 * It's acceptable to have no devs specified.
4221 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
4225 return (SET_ERROR(EINVAL
));
4228 * Make sure the pool is formatted with a version that supports this
4231 if (spa_version(spa
) < version
)
4232 return (SET_ERROR(ENOTSUP
));
4235 * Set the pending device list so we correctly handle device in-use
4238 sav
->sav_pending
= dev
;
4239 sav
->sav_npending
= ndev
;
4241 for (i
= 0; i
< ndev
; i
++) {
4242 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
4246 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4248 error
= SET_ERROR(EINVAL
);
4254 if ((error
= vdev_open(vd
)) == 0 &&
4255 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
4256 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
4257 vd
->vdev_guid
) == 0);
4263 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
4270 sav
->sav_pending
= NULL
;
4271 sav
->sav_npending
= 0;
4276 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
4280 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4282 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
4283 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
4284 VDEV_LABEL_SPARE
)) != 0) {
4288 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
4289 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
4290 VDEV_LABEL_L2CACHE
));
4294 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
4299 if (sav
->sav_config
!= NULL
) {
4305 * Generate new dev list by concatenating with the
4308 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
4309 &olddevs
, &oldndevs
) == 0);
4311 newdevs
= kmem_alloc(sizeof (void *) *
4312 (ndevs
+ oldndevs
), KM_SLEEP
);
4313 for (i
= 0; i
< oldndevs
; i
++)
4314 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
4316 for (i
= 0; i
< ndevs
; i
++)
4317 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
4320 VERIFY(nvlist_remove(sav
->sav_config
, config
,
4321 DATA_TYPE_NVLIST_ARRAY
) == 0);
4323 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
4324 config
, newdevs
, ndevs
+ oldndevs
) == 0);
4325 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
4326 nvlist_free(newdevs
[i
]);
4327 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
4330 * Generate a new dev list.
4332 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
4334 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
4340 * Stop and drop level 2 ARC devices
4343 spa_l2cache_drop(spa_t
*spa
)
4347 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4349 for (i
= 0; i
< sav
->sav_count
; i
++) {
4352 vd
= sav
->sav_vdevs
[i
];
4355 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
4356 pool
!= 0ULL && l2arc_vdev_present(vd
))
4357 l2arc_remove_vdev(vd
);
4362 * Verify encryption parameters for spa creation. If we are encrypting, we must
4363 * have the encryption feature flag enabled.
4366 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
4367 boolean_t has_encryption
)
4369 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
4370 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
4372 return (SET_ERROR(ENOTSUP
));
4374 return (dmu_objset_create_crypt_check(NULL
, dcp
));
4381 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
4382 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
4385 char *altroot
= NULL
;
4390 uint64_t txg
= TXG_INITIAL
;
4391 nvlist_t
**spares
, **l2cache
;
4392 uint_t nspares
, nl2cache
;
4393 uint64_t version
, obj
, root_dsobj
= 0;
4394 boolean_t has_features
;
4395 boolean_t has_encryption
;
4401 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
4402 poolname
= (char *)pool
;
4405 * If this pool already exists, return failure.
4407 mutex_enter(&spa_namespace_lock
);
4408 if (spa_lookup(poolname
) != NULL
) {
4409 mutex_exit(&spa_namespace_lock
);
4410 return (SET_ERROR(EEXIST
));
4414 * Allocate a new spa_t structure.
4416 nvl
= fnvlist_alloc();
4417 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
4418 (void) nvlist_lookup_string(props
,
4419 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4420 spa
= spa_add(poolname
, nvl
, altroot
);
4422 spa_activate(spa
, spa_mode_global
);
4424 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4425 spa_deactivate(spa
);
4427 mutex_exit(&spa_namespace_lock
);
4432 * Temporary pool names should never be written to disk.
4434 if (poolname
!= pool
)
4435 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
4437 has_features
= B_FALSE
;
4438 has_encryption
= B_FALSE
;
4439 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
4440 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4441 if (zpool_prop_feature(nvpair_name(elem
))) {
4442 has_features
= B_TRUE
;
4444 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
4445 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
4446 if (feat
== SPA_FEATURE_ENCRYPTION
)
4447 has_encryption
= B_TRUE
;
4451 /* verify encryption params, if they were provided */
4453 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
4455 spa_deactivate(spa
);
4457 mutex_exit(&spa_namespace_lock
);
4462 if (has_features
|| nvlist_lookup_uint64(props
,
4463 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4464 version
= SPA_VERSION
;
4466 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4468 spa
->spa_first_txg
= txg
;
4469 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4470 spa
->spa_uberblock
.ub_version
= version
;
4471 spa
->spa_ubsync
= spa
->spa_uberblock
;
4472 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4473 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
4474 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
4475 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
4478 * Create "The Godfather" zio to hold all async IOs
4480 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4482 for (int i
= 0; i
< max_ncpus
; i
++) {
4483 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4484 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4485 ZIO_FLAG_GODFATHER
);
4489 * Create the root vdev.
4491 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4493 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4495 ASSERT(error
!= 0 || rvd
!= NULL
);
4496 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4498 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4499 error
= SET_ERROR(EINVAL
);
4502 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4503 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4504 VDEV_ALLOC_ADD
)) == 0) {
4505 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
4506 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4507 vdev_expand(rvd
->vdev_child
[c
], txg
);
4511 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4515 spa_deactivate(spa
);
4517 mutex_exit(&spa_namespace_lock
);
4522 * Get the list of spares, if specified.
4524 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4525 &spares
, &nspares
) == 0) {
4526 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4528 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4529 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4530 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4531 spa_load_spares(spa
);
4532 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4533 spa
->spa_spares
.sav_sync
= B_TRUE
;
4537 * Get the list of level 2 cache devices, if specified.
4539 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4540 &l2cache
, &nl2cache
) == 0) {
4541 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4542 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4543 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4544 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4545 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4546 spa_load_l2cache(spa
);
4547 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4548 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4551 spa
->spa_is_initializing
= B_TRUE
;
4552 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
4553 spa
->spa_is_initializing
= B_FALSE
;
4556 * Create DDTs (dedup tables).
4560 spa_update_dspace(spa
);
4562 tx
= dmu_tx_create_assigned(dp
, txg
);
4565 * Create the pool's history object.
4567 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
4568 spa_history_create_obj(spa
, tx
);
4570 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4571 spa_history_log_version(spa
, "create", tx
);
4574 * Create the pool config object.
4576 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4577 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4578 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4580 if (zap_add(spa
->spa_meta_objset
,
4581 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4582 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4583 cmn_err(CE_PANIC
, "failed to add pool config");
4586 if (zap_add(spa
->spa_meta_objset
,
4587 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4588 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4589 cmn_err(CE_PANIC
, "failed to add pool version");
4592 /* Newly created pools with the right version are always deflated. */
4593 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4594 spa
->spa_deflate
= TRUE
;
4595 if (zap_add(spa
->spa_meta_objset
,
4596 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4597 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4598 cmn_err(CE_PANIC
, "failed to add deflate");
4603 * Create the deferred-free bpobj. Turn off compression
4604 * because sync-to-convergence takes longer if the blocksize
4607 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4608 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4609 ZIO_COMPRESS_OFF
, tx
);
4610 if (zap_add(spa
->spa_meta_objset
,
4611 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4612 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4613 cmn_err(CE_PANIC
, "failed to add bpobj");
4615 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4616 spa
->spa_meta_objset
, obj
));
4619 * Generate some random noise for salted checksums to operate on.
4621 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4622 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4625 * Set pool properties.
4627 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4628 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4629 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4630 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4631 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
4633 if (props
!= NULL
) {
4634 spa_configfile_set(spa
, props
, B_FALSE
);
4635 spa_sync_props(props
, tx
);
4641 * If the root dataset is encrypted we will need to create key mappings
4642 * for the zio layer before we start to write any data to disk and hold
4643 * them until after the first txg has been synced. Waiting for the first
4644 * transaction to complete also ensures that our bean counters are
4645 * appropriately updated.
4647 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0) {
4648 root_dsobj
= dsl_dir_phys(dp
->dp_root_dir
)->dd_head_dataset_obj
;
4649 VERIFY0(spa_keystore_create_mapping_impl(spa
, root_dsobj
,
4650 dp
->dp_root_dir
, FTAG
));
4653 spa
->spa_sync_on
= B_TRUE
;
4655 mmp_thread_start(spa
);
4656 txg_wait_synced(dp
, txg
);
4658 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0)
4659 VERIFY0(spa_keystore_remove_mapping(spa
, root_dsobj
, FTAG
));
4661 spa_spawn_aux_threads(spa
);
4663 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
4666 * Don't count references from objsets that are already closed
4667 * and are making their way through the eviction process.
4669 spa_evicting_os_wait(spa
);
4670 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4671 spa
->spa_load_state
= SPA_LOAD_NONE
;
4673 mutex_exit(&spa_namespace_lock
);
4679 * Import a non-root pool into the system.
4682 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4685 char *altroot
= NULL
;
4686 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4687 zpool_rewind_policy_t policy
;
4688 uint64_t mode
= spa_mode_global
;
4689 uint64_t readonly
= B_FALSE
;
4692 nvlist_t
**spares
, **l2cache
;
4693 uint_t nspares
, nl2cache
;
4696 * If a pool with this name exists, return failure.
4698 mutex_enter(&spa_namespace_lock
);
4699 if (spa_lookup(pool
) != NULL
) {
4700 mutex_exit(&spa_namespace_lock
);
4701 return (SET_ERROR(EEXIST
));
4705 * Create and initialize the spa structure.
4707 (void) nvlist_lookup_string(props
,
4708 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4709 (void) nvlist_lookup_uint64(props
,
4710 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4713 spa
= spa_add(pool
, config
, altroot
);
4714 spa
->spa_import_flags
= flags
;
4717 * Verbatim import - Take a pool and insert it into the namespace
4718 * as if it had been loaded at boot.
4720 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4722 spa_configfile_set(spa
, props
, B_FALSE
);
4724 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
4725 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4727 mutex_exit(&spa_namespace_lock
);
4731 spa_activate(spa
, mode
);
4734 * Don't start async tasks until we know everything is healthy.
4736 spa_async_suspend(spa
);
4738 zpool_get_rewind_policy(config
, &policy
);
4739 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4740 state
= SPA_LOAD_RECOVER
;
4743 * Pass off the heavy lifting to spa_load(). Pass TRUE for trust_config
4744 * because the user-supplied config is actually the one to trust when
4747 if (state
!= SPA_LOAD_RECOVER
)
4748 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4750 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4751 policy
.zrp_request
);
4754 * Propagate anything learned while loading the pool and pass it
4755 * back to caller (i.e. rewind info, missing devices, etc).
4757 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4758 spa
->spa_load_info
) == 0);
4760 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4762 * Toss any existing sparelist, as it doesn't have any validity
4763 * anymore, and conflicts with spa_has_spare().
4765 if (spa
->spa_spares
.sav_config
) {
4766 nvlist_free(spa
->spa_spares
.sav_config
);
4767 spa
->spa_spares
.sav_config
= NULL
;
4768 spa_load_spares(spa
);
4770 if (spa
->spa_l2cache
.sav_config
) {
4771 nvlist_free(spa
->spa_l2cache
.sav_config
);
4772 spa
->spa_l2cache
.sav_config
= NULL
;
4773 spa_load_l2cache(spa
);
4776 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4778 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4781 spa_configfile_set(spa
, props
, B_FALSE
);
4783 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4784 (error
= spa_prop_set(spa
, props
)))) {
4786 spa_deactivate(spa
);
4788 mutex_exit(&spa_namespace_lock
);
4792 spa_async_resume(spa
);
4795 * Override any spares and level 2 cache devices as specified by
4796 * the user, as these may have correct device names/devids, etc.
4798 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4799 &spares
, &nspares
) == 0) {
4800 if (spa
->spa_spares
.sav_config
)
4801 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4802 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4804 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4805 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4806 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4807 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4808 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4809 spa_load_spares(spa
);
4810 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4811 spa
->spa_spares
.sav_sync
= B_TRUE
;
4813 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4814 &l2cache
, &nl2cache
) == 0) {
4815 if (spa
->spa_l2cache
.sav_config
)
4816 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4817 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4819 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4820 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4821 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4822 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4823 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4824 spa_load_l2cache(spa
);
4825 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4826 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4830 * Check for any removed devices.
4832 if (spa
->spa_autoreplace
) {
4833 spa_aux_check_removed(&spa
->spa_spares
);
4834 spa_aux_check_removed(&spa
->spa_l2cache
);
4837 if (spa_writeable(spa
)) {
4839 * Update the config cache to include the newly-imported pool.
4841 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4845 * It's possible that the pool was expanded while it was exported.
4846 * We kick off an async task to handle this for us.
4848 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4850 spa_history_log_version(spa
, "import", NULL
);
4852 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4854 zvol_create_minors(spa
, pool
, B_TRUE
);
4856 mutex_exit(&spa_namespace_lock
);
4862 spa_tryimport(nvlist_t
*tryconfig
)
4864 nvlist_t
*config
= NULL
;
4870 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4873 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4877 * Create and initialize the spa structure.
4879 mutex_enter(&spa_namespace_lock
);
4880 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4881 spa_activate(spa
, FREAD
);
4884 * Pass off the heavy lifting to spa_load().
4885 * Pass TRUE for trust_config because the user-supplied config
4886 * is actually the one to trust when doing an import.
4888 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4891 * If 'tryconfig' was at least parsable, return the current config.
4893 if (spa
->spa_root_vdev
!= NULL
) {
4894 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4895 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4897 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4899 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4900 spa
->spa_uberblock
.ub_timestamp
) == 0);
4901 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4902 spa
->spa_load_info
) == 0);
4903 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4904 spa
->spa_errata
) == 0);
4907 * If the bootfs property exists on this pool then we
4908 * copy it out so that external consumers can tell which
4909 * pools are bootable.
4911 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4912 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4915 * We have to play games with the name since the
4916 * pool was opened as TRYIMPORT_NAME.
4918 if (dsl_dsobj_to_dsname(spa_name(spa
),
4919 spa
->spa_bootfs
, tmpname
) == 0) {
4923 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4925 cp
= strchr(tmpname
, '/');
4927 (void) strlcpy(dsname
, tmpname
,
4930 (void) snprintf(dsname
, MAXPATHLEN
,
4931 "%s/%s", poolname
, ++cp
);
4933 VERIFY(nvlist_add_string(config
,
4934 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4935 kmem_free(dsname
, MAXPATHLEN
);
4937 kmem_free(tmpname
, MAXPATHLEN
);
4941 * Add the list of hot spares and level 2 cache devices.
4943 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4944 spa_add_spares(spa
, config
);
4945 spa_add_l2cache(spa
, config
);
4946 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4950 spa_deactivate(spa
);
4952 mutex_exit(&spa_namespace_lock
);
4958 * Pool export/destroy
4960 * The act of destroying or exporting a pool is very simple. We make sure there
4961 * is no more pending I/O and any references to the pool are gone. Then, we
4962 * update the pool state and sync all the labels to disk, removing the
4963 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4964 * we don't sync the labels or remove the configuration cache.
4967 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4968 boolean_t force
, boolean_t hardforce
)
4975 if (!(spa_mode_global
& FWRITE
))
4976 return (SET_ERROR(EROFS
));
4978 mutex_enter(&spa_namespace_lock
);
4979 if ((spa
= spa_lookup(pool
)) == NULL
) {
4980 mutex_exit(&spa_namespace_lock
);
4981 return (SET_ERROR(ENOENT
));
4985 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4986 * reacquire the namespace lock, and see if we can export.
4988 spa_open_ref(spa
, FTAG
);
4989 mutex_exit(&spa_namespace_lock
);
4990 spa_async_suspend(spa
);
4991 if (spa
->spa_zvol_taskq
) {
4992 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
4993 taskq_wait(spa
->spa_zvol_taskq
);
4995 mutex_enter(&spa_namespace_lock
);
4996 spa_close(spa
, FTAG
);
4998 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
5001 * The pool will be in core if it's openable, in which case we can
5002 * modify its state. Objsets may be open only because they're dirty,
5003 * so we have to force it to sync before checking spa_refcnt.
5005 if (spa
->spa_sync_on
) {
5006 txg_wait_synced(spa
->spa_dsl_pool
, 0);
5007 spa_evicting_os_wait(spa
);
5011 * A pool cannot be exported or destroyed if there are active
5012 * references. If we are resetting a pool, allow references by
5013 * fault injection handlers.
5015 if (!spa_refcount_zero(spa
) ||
5016 (spa
->spa_inject_ref
!= 0 &&
5017 new_state
!= POOL_STATE_UNINITIALIZED
)) {
5018 spa_async_resume(spa
);
5019 mutex_exit(&spa_namespace_lock
);
5020 return (SET_ERROR(EBUSY
));
5023 if (spa
->spa_sync_on
) {
5025 * A pool cannot be exported if it has an active shared spare.
5026 * This is to prevent other pools stealing the active spare
5027 * from an exported pool. At user's own will, such pool can
5028 * be forcedly exported.
5030 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
5031 spa_has_active_shared_spare(spa
)) {
5032 spa_async_resume(spa
);
5033 mutex_exit(&spa_namespace_lock
);
5034 return (SET_ERROR(EXDEV
));
5038 * We want this to be reflected on every label,
5039 * so mark them all dirty. spa_unload() will do the
5040 * final sync that pushes these changes out.
5042 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
5043 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5044 spa
->spa_state
= new_state
;
5045 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
5047 vdev_config_dirty(spa
->spa_root_vdev
);
5048 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5053 if (new_state
== POOL_STATE_DESTROYED
)
5054 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
5055 else if (new_state
== POOL_STATE_EXPORTED
)
5056 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
5058 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
5060 spa_deactivate(spa
);
5063 if (oldconfig
&& spa
->spa_config
)
5064 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
5066 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
5068 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
5071 mutex_exit(&spa_namespace_lock
);
5077 * Destroy a storage pool.
5080 spa_destroy(char *pool
)
5082 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
5087 * Export a storage pool.
5090 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
5091 boolean_t hardforce
)
5093 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
5098 * Similar to spa_export(), this unloads the spa_t without actually removing it
5099 * from the namespace in any way.
5102 spa_reset(char *pool
)
5104 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
5109 * ==========================================================================
5110 * Device manipulation
5111 * ==========================================================================
5115 * Add a device to a storage pool.
5118 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
5122 vdev_t
*rvd
= spa
->spa_root_vdev
;
5124 nvlist_t
**spares
, **l2cache
;
5125 uint_t nspares
, nl2cache
;
5127 ASSERT(spa_writeable(spa
));
5129 txg
= spa_vdev_enter(spa
);
5131 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
5132 VDEV_ALLOC_ADD
)) != 0)
5133 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5135 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
5137 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
5141 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
5145 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
5146 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5148 if (vd
->vdev_children
!= 0 &&
5149 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
5150 return (spa_vdev_exit(spa
, vd
, txg
, error
));
5153 * We must validate the spares and l2cache devices after checking the
5154 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5156 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
5157 return (spa_vdev_exit(spa
, vd
, txg
, error
));
5160 * If we are in the middle of a device removal, we can only add
5161 * devices which match the existing devices in the pool.
5162 * If we are in the middle of a removal, or have some indirect
5163 * vdevs, we can not add raidz toplevels.
5165 if (spa
->spa_vdev_removal
!= NULL
||
5166 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
5167 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5168 tvd
= vd
->vdev_child
[c
];
5169 if (spa
->spa_vdev_removal
!= NULL
&&
5170 tvd
->vdev_ashift
!= spa
->spa_max_ashift
) {
5171 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5173 /* Fail if top level vdev is raidz */
5174 if (tvd
->vdev_ops
== &vdev_raidz_ops
) {
5175 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5178 * Need the top level mirror to be
5179 * a mirror of leaf vdevs only
5181 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
5182 for (uint64_t cid
= 0;
5183 cid
< tvd
->vdev_children
; cid
++) {
5184 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
5185 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
5186 return (spa_vdev_exit(spa
, vd
,
5194 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5197 * Set the vdev id to the first hole, if one exists.
5199 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
5200 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
5201 vdev_free(rvd
->vdev_child
[id
]);
5205 tvd
= vd
->vdev_child
[c
];
5206 vdev_remove_child(vd
, tvd
);
5208 vdev_add_child(rvd
, tvd
);
5209 vdev_config_dirty(tvd
);
5213 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
5214 ZPOOL_CONFIG_SPARES
);
5215 spa_load_spares(spa
);
5216 spa
->spa_spares
.sav_sync
= B_TRUE
;
5219 if (nl2cache
!= 0) {
5220 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
5221 ZPOOL_CONFIG_L2CACHE
);
5222 spa_load_l2cache(spa
);
5223 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5227 * We have to be careful when adding new vdevs to an existing pool.
5228 * If other threads start allocating from these vdevs before we
5229 * sync the config cache, and we lose power, then upon reboot we may
5230 * fail to open the pool because there are DVAs that the config cache
5231 * can't translate. Therefore, we first add the vdevs without
5232 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5233 * and then let spa_config_update() initialize the new metaslabs.
5235 * spa_load() checks for added-but-not-initialized vdevs, so that
5236 * if we lose power at any point in this sequence, the remaining
5237 * steps will be completed the next time we load the pool.
5239 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
5241 mutex_enter(&spa_namespace_lock
);
5242 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5243 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
5244 mutex_exit(&spa_namespace_lock
);
5250 * Attach a device to a mirror. The arguments are the path to any device
5251 * in the mirror, and the nvroot for the new device. If the path specifies
5252 * a device that is not mirrored, we automatically insert the mirror vdev.
5254 * If 'replacing' is specified, the new device is intended to replace the
5255 * existing device; in this case the two devices are made into their own
5256 * mirror using the 'replacing' vdev, which is functionally identical to
5257 * the mirror vdev (it actually reuses all the same ops) but has a few
5258 * extra rules: you can't attach to it after it's been created, and upon
5259 * completion of resilvering, the first disk (the one being replaced)
5260 * is automatically detached.
5263 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
5265 uint64_t txg
, dtl_max_txg
;
5266 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5267 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
5269 char *oldvdpath
, *newvdpath
;
5273 ASSERT(spa_writeable(spa
));
5275 txg
= spa_vdev_enter(spa
);
5277 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5279 if (spa
->spa_vdev_removal
!= NULL
)
5280 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5283 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5285 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
5286 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5288 pvd
= oldvd
->vdev_parent
;
5290 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
5291 VDEV_ALLOC_ATTACH
)) != 0)
5292 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5294 if (newrootvd
->vdev_children
!= 1)
5295 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
5297 newvd
= newrootvd
->vdev_child
[0];
5299 if (!newvd
->vdev_ops
->vdev_op_leaf
)
5300 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
5302 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
5303 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
5306 * Spares can't replace logs
5308 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
5309 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5313 * For attach, the only allowable parent is a mirror or the root
5316 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5317 pvd
->vdev_ops
!= &vdev_root_ops
)
5318 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5320 pvops
= &vdev_mirror_ops
;
5323 * Active hot spares can only be replaced by inactive hot
5326 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5327 oldvd
->vdev_isspare
&&
5328 !spa_has_spare(spa
, newvd
->vdev_guid
))
5329 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5332 * If the source is a hot spare, and the parent isn't already a
5333 * spare, then we want to create a new hot spare. Otherwise, we
5334 * want to create a replacing vdev. The user is not allowed to
5335 * attach to a spared vdev child unless the 'isspare' state is
5336 * the same (spare replaces spare, non-spare replaces
5339 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
5340 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
5341 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5342 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5343 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
5344 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5347 if (newvd
->vdev_isspare
)
5348 pvops
= &vdev_spare_ops
;
5350 pvops
= &vdev_replacing_ops
;
5354 * Make sure the new device is big enough.
5356 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
5357 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
5360 * The new device cannot have a higher alignment requirement
5361 * than the top-level vdev.
5363 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
5364 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
5367 * If this is an in-place replacement, update oldvd's path and devid
5368 * to make it distinguishable from newvd, and unopenable from now on.
5370 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
5371 spa_strfree(oldvd
->vdev_path
);
5372 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
5374 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
5375 newvd
->vdev_path
, "old");
5376 if (oldvd
->vdev_devid
!= NULL
) {
5377 spa_strfree(oldvd
->vdev_devid
);
5378 oldvd
->vdev_devid
= NULL
;
5382 /* mark the device being resilvered */
5383 newvd
->vdev_resilver_txg
= txg
;
5386 * If the parent is not a mirror, or if we're replacing, insert the new
5387 * mirror/replacing/spare vdev above oldvd.
5389 if (pvd
->vdev_ops
!= pvops
)
5390 pvd
= vdev_add_parent(oldvd
, pvops
);
5392 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
5393 ASSERT(pvd
->vdev_ops
== pvops
);
5394 ASSERT(oldvd
->vdev_parent
== pvd
);
5397 * Extract the new device from its root and add it to pvd.
5399 vdev_remove_child(newrootvd
, newvd
);
5400 newvd
->vdev_id
= pvd
->vdev_children
;
5401 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
5402 vdev_add_child(pvd
, newvd
);
5405 * Reevaluate the parent vdev state.
5407 vdev_propagate_state(pvd
);
5409 tvd
= newvd
->vdev_top
;
5410 ASSERT(pvd
->vdev_top
== tvd
);
5411 ASSERT(tvd
->vdev_parent
== rvd
);
5413 vdev_config_dirty(tvd
);
5416 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5417 * for any dmu_sync-ed blocks. It will propagate upward when
5418 * spa_vdev_exit() calls vdev_dtl_reassess().
5420 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
5422 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
5423 dtl_max_txg
- TXG_INITIAL
);
5425 if (newvd
->vdev_isspare
) {
5426 spa_spare_activate(newvd
);
5427 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
5430 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
5431 newvdpath
= spa_strdup(newvd
->vdev_path
);
5432 newvd_isspare
= newvd
->vdev_isspare
;
5435 * Mark newvd's DTL dirty in this txg.
5437 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
5440 * Schedule the resilver to restart in the future. We do this to
5441 * ensure that dmu_sync-ed blocks have been stitched into the
5442 * respective datasets.
5444 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
5446 if (spa
->spa_bootfs
)
5447 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
5449 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
5454 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
5456 spa_history_log_internal(spa
, "vdev attach", NULL
,
5457 "%s vdev=%s %s vdev=%s",
5458 replacing
&& newvd_isspare
? "spare in" :
5459 replacing
? "replace" : "attach", newvdpath
,
5460 replacing
? "for" : "to", oldvdpath
);
5462 spa_strfree(oldvdpath
);
5463 spa_strfree(newvdpath
);
5469 * Detach a device from a mirror or replacing vdev.
5471 * If 'replace_done' is specified, only detach if the parent
5472 * is a replacing vdev.
5475 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5479 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5480 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5481 boolean_t unspare
= B_FALSE
;
5482 uint64_t unspare_guid
= 0;
5485 ASSERT(spa_writeable(spa
));
5487 txg
= spa_vdev_enter(spa
);
5489 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5492 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5494 if (!vd
->vdev_ops
->vdev_op_leaf
)
5495 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5497 pvd
= vd
->vdev_parent
;
5500 * If the parent/child relationship is not as expected, don't do it.
5501 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5502 * vdev that's replacing B with C. The user's intent in replacing
5503 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5504 * the replace by detaching C, the expected behavior is to end up
5505 * M(A,B). But suppose that right after deciding to detach C,
5506 * the replacement of B completes. We would have M(A,C), and then
5507 * ask to detach C, which would leave us with just A -- not what
5508 * the user wanted. To prevent this, we make sure that the
5509 * parent/child relationship hasn't changed -- in this example,
5510 * that C's parent is still the replacing vdev R.
5512 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5513 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5516 * Only 'replacing' or 'spare' vdevs can be replaced.
5518 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5519 pvd
->vdev_ops
!= &vdev_spare_ops
)
5520 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5522 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5523 spa_version(spa
) >= SPA_VERSION_SPARES
);
5526 * Only mirror, replacing, and spare vdevs support detach.
5528 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5529 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5530 pvd
->vdev_ops
!= &vdev_spare_ops
)
5531 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5534 * If this device has the only valid copy of some data,
5535 * we cannot safely detach it.
5537 if (vdev_dtl_required(vd
))
5538 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5540 ASSERT(pvd
->vdev_children
>= 2);
5543 * If we are detaching the second disk from a replacing vdev, then
5544 * check to see if we changed the original vdev's path to have "/old"
5545 * at the end in spa_vdev_attach(). If so, undo that change now.
5547 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5548 vd
->vdev_path
!= NULL
) {
5549 size_t len
= strlen(vd
->vdev_path
);
5551 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
5552 cvd
= pvd
->vdev_child
[c
];
5554 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5557 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5558 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5559 spa_strfree(cvd
->vdev_path
);
5560 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5567 * If we are detaching the original disk from a spare, then it implies
5568 * that the spare should become a real disk, and be removed from the
5569 * active spare list for the pool.
5571 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5573 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5577 * Erase the disk labels so the disk can be used for other things.
5578 * This must be done after all other error cases are handled,
5579 * but before we disembowel vd (so we can still do I/O to it).
5580 * But if we can't do it, don't treat the error as fatal --
5581 * it may be that the unwritability of the disk is the reason
5582 * it's being detached!
5584 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5587 * Remove vd from its parent and compact the parent's children.
5589 vdev_remove_child(pvd
, vd
);
5590 vdev_compact_children(pvd
);
5593 * Remember one of the remaining children so we can get tvd below.
5595 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5598 * If we need to remove the remaining child from the list of hot spares,
5599 * do it now, marking the vdev as no longer a spare in the process.
5600 * We must do this before vdev_remove_parent(), because that can
5601 * change the GUID if it creates a new toplevel GUID. For a similar
5602 * reason, we must remove the spare now, in the same txg as the detach;
5603 * otherwise someone could attach a new sibling, change the GUID, and
5604 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5607 ASSERT(cvd
->vdev_isspare
);
5608 spa_spare_remove(cvd
);
5609 unspare_guid
= cvd
->vdev_guid
;
5610 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5611 cvd
->vdev_unspare
= B_TRUE
;
5615 * If the parent mirror/replacing vdev only has one child,
5616 * the parent is no longer needed. Remove it from the tree.
5618 if (pvd
->vdev_children
== 1) {
5619 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5620 cvd
->vdev_unspare
= B_FALSE
;
5621 vdev_remove_parent(cvd
);
5626 * We don't set tvd until now because the parent we just removed
5627 * may have been the previous top-level vdev.
5629 tvd
= cvd
->vdev_top
;
5630 ASSERT(tvd
->vdev_parent
== rvd
);
5633 * Reevaluate the parent vdev state.
5635 vdev_propagate_state(cvd
);
5638 * If the 'autoexpand' property is set on the pool then automatically
5639 * try to expand the size of the pool. For example if the device we
5640 * just detached was smaller than the others, it may be possible to
5641 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5642 * first so that we can obtain the updated sizes of the leaf vdevs.
5644 if (spa
->spa_autoexpand
) {
5646 vdev_expand(tvd
, txg
);
5649 vdev_config_dirty(tvd
);
5652 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5653 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5654 * But first make sure we're not on any *other* txg's DTL list, to
5655 * prevent vd from being accessed after it's freed.
5657 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
5658 for (int t
= 0; t
< TXG_SIZE
; t
++)
5659 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5660 vd
->vdev_detached
= B_TRUE
;
5661 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5663 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5665 /* hang on to the spa before we release the lock */
5666 spa_open_ref(spa
, FTAG
);
5668 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5670 spa_history_log_internal(spa
, "detach", NULL
,
5672 spa_strfree(vdpath
);
5675 * If this was the removal of the original device in a hot spare vdev,
5676 * then we want to go through and remove the device from the hot spare
5677 * list of every other pool.
5680 spa_t
*altspa
= NULL
;
5682 mutex_enter(&spa_namespace_lock
);
5683 while ((altspa
= spa_next(altspa
)) != NULL
) {
5684 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5688 spa_open_ref(altspa
, FTAG
);
5689 mutex_exit(&spa_namespace_lock
);
5690 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5691 mutex_enter(&spa_namespace_lock
);
5692 spa_close(altspa
, FTAG
);
5694 mutex_exit(&spa_namespace_lock
);
5696 /* search the rest of the vdevs for spares to remove */
5697 spa_vdev_resilver_done(spa
);
5700 /* all done with the spa; OK to release */
5701 mutex_enter(&spa_namespace_lock
);
5702 spa_close(spa
, FTAG
);
5703 mutex_exit(&spa_namespace_lock
);
5709 * Split a set of devices from their mirrors, and create a new pool from them.
5712 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5713 nvlist_t
*props
, boolean_t exp
)
5716 uint64_t txg
, *glist
;
5718 uint_t c
, children
, lastlog
;
5719 nvlist_t
**child
, *nvl
, *tmp
;
5721 char *altroot
= NULL
;
5722 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5723 boolean_t activate_slog
;
5725 ASSERT(spa_writeable(spa
));
5727 txg
= spa_vdev_enter(spa
);
5729 /* clear the log and flush everything up to now */
5730 activate_slog
= spa_passivate_log(spa
);
5731 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5732 error
= spa_reset_logs(spa
);
5733 txg
= spa_vdev_config_enter(spa
);
5736 spa_activate_log(spa
);
5739 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5741 /* check new spa name before going any further */
5742 if (spa_lookup(newname
) != NULL
)
5743 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5746 * scan through all the children to ensure they're all mirrors
5748 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5749 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5751 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5753 /* first, check to ensure we've got the right child count */
5754 rvd
= spa
->spa_root_vdev
;
5756 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5757 vdev_t
*vd
= rvd
->vdev_child
[c
];
5759 /* don't count the holes & logs as children */
5760 if (vd
->vdev_islog
|| !vdev_is_concrete(vd
)) {
5768 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5769 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5771 /* next, ensure no spare or cache devices are part of the split */
5772 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5773 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5774 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5776 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5777 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5779 /* then, loop over each vdev and validate it */
5780 for (c
= 0; c
< children
; c
++) {
5781 uint64_t is_hole
= 0;
5783 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5787 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5788 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5791 error
= SET_ERROR(EINVAL
);
5796 /* which disk is going to be split? */
5797 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5799 error
= SET_ERROR(EINVAL
);
5803 /* look it up in the spa */
5804 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5805 if (vml
[c
] == NULL
) {
5806 error
= SET_ERROR(ENODEV
);
5810 /* make sure there's nothing stopping the split */
5811 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5812 vml
[c
]->vdev_islog
||
5813 !vdev_is_concrete(vml
[c
]) ||
5814 vml
[c
]->vdev_isspare
||
5815 vml
[c
]->vdev_isl2cache
||
5816 !vdev_writeable(vml
[c
]) ||
5817 vml
[c
]->vdev_children
!= 0 ||
5818 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5819 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5820 error
= SET_ERROR(EINVAL
);
5824 if (vdev_dtl_required(vml
[c
])) {
5825 error
= SET_ERROR(EBUSY
);
5829 /* we need certain info from the top level */
5830 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5831 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5832 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5833 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5834 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5835 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5836 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5837 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5839 /* transfer per-vdev ZAPs */
5840 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
5841 VERIFY0(nvlist_add_uint64(child
[c
],
5842 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
5844 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
5845 VERIFY0(nvlist_add_uint64(child
[c
],
5846 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
5847 vml
[c
]->vdev_parent
->vdev_top_zap
));
5851 kmem_free(vml
, children
* sizeof (vdev_t
*));
5852 kmem_free(glist
, children
* sizeof (uint64_t));
5853 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5856 /* stop writers from using the disks */
5857 for (c
= 0; c
< children
; c
++) {
5859 vml
[c
]->vdev_offline
= B_TRUE
;
5861 vdev_reopen(spa
->spa_root_vdev
);
5864 * Temporarily record the splitting vdevs in the spa config. This
5865 * will disappear once the config is regenerated.
5867 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5868 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5869 glist
, children
) == 0);
5870 kmem_free(glist
, children
* sizeof (uint64_t));
5872 mutex_enter(&spa
->spa_props_lock
);
5873 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5875 mutex_exit(&spa
->spa_props_lock
);
5876 spa
->spa_config_splitting
= nvl
;
5877 vdev_config_dirty(spa
->spa_root_vdev
);
5879 /* configure and create the new pool */
5880 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5881 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5882 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5883 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5884 spa_version(spa
)) == 0);
5885 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5886 spa
->spa_config_txg
) == 0);
5887 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5888 spa_generate_guid(NULL
)) == 0);
5889 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
5890 (void) nvlist_lookup_string(props
,
5891 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5893 /* add the new pool to the namespace */
5894 newspa
= spa_add(newname
, config
, altroot
);
5895 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5896 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5897 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5899 /* release the spa config lock, retaining the namespace lock */
5900 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5902 if (zio_injection_enabled
)
5903 zio_handle_panic_injection(spa
, FTAG
, 1);
5905 spa_activate(newspa
, spa_mode_global
);
5906 spa_async_suspend(newspa
);
5908 /* create the new pool from the disks of the original pool */
5909 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5913 /* if that worked, generate a real config for the new pool */
5914 if (newspa
->spa_root_vdev
!= NULL
) {
5915 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5916 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5917 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5918 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5919 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5924 if (props
!= NULL
) {
5925 spa_configfile_set(newspa
, props
, B_FALSE
);
5926 error
= spa_prop_set(newspa
, props
);
5931 /* flush everything */
5932 txg
= spa_vdev_config_enter(newspa
);
5933 vdev_config_dirty(newspa
->spa_root_vdev
);
5934 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5936 if (zio_injection_enabled
)
5937 zio_handle_panic_injection(spa
, FTAG
, 2);
5939 spa_async_resume(newspa
);
5941 /* finally, update the original pool's config */
5942 txg
= spa_vdev_config_enter(spa
);
5943 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5944 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5947 for (c
= 0; c
< children
; c
++) {
5948 if (vml
[c
] != NULL
) {
5951 spa_history_log_internal(spa
, "detach", tx
,
5952 "vdev=%s", vml
[c
]->vdev_path
);
5957 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5958 vdev_config_dirty(spa
->spa_root_vdev
);
5959 spa
->spa_config_splitting
= NULL
;
5963 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5965 if (zio_injection_enabled
)
5966 zio_handle_panic_injection(spa
, FTAG
, 3);
5968 /* split is complete; log a history record */
5969 spa_history_log_internal(newspa
, "split", NULL
,
5970 "from pool %s", spa_name(spa
));
5972 kmem_free(vml
, children
* sizeof (vdev_t
*));
5974 /* if we're not going to mount the filesystems in userland, export */
5976 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5983 spa_deactivate(newspa
);
5986 txg
= spa_vdev_config_enter(spa
);
5988 /* re-online all offlined disks */
5989 for (c
= 0; c
< children
; c
++) {
5991 vml
[c
]->vdev_offline
= B_FALSE
;
5993 vdev_reopen(spa
->spa_root_vdev
);
5995 nvlist_free(spa
->spa_config_splitting
);
5996 spa
->spa_config_splitting
= NULL
;
5997 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5999 kmem_free(vml
, children
* sizeof (vdev_t
*));
6004 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6005 * currently spared, so we can detach it.
6008 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
6010 vdev_t
*newvd
, *oldvd
;
6012 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6013 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
6019 * Check for a completed replacement. We always consider the first
6020 * vdev in the list to be the oldest vdev, and the last one to be
6021 * the newest (see spa_vdev_attach() for how that works). In
6022 * the case where the newest vdev is faulted, we will not automatically
6023 * remove it after a resilver completes. This is OK as it will require
6024 * user intervention to determine which disk the admin wishes to keep.
6026 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
6027 ASSERT(vd
->vdev_children
> 1);
6029 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
6030 oldvd
= vd
->vdev_child
[0];
6032 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
6033 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
6034 !vdev_dtl_required(oldvd
))
6039 * Check for a completed resilver with the 'unspare' flag set.
6041 if (vd
->vdev_ops
== &vdev_spare_ops
) {
6042 vdev_t
*first
= vd
->vdev_child
[0];
6043 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
6045 if (last
->vdev_unspare
) {
6048 } else if (first
->vdev_unspare
) {
6055 if (oldvd
!= NULL
&&
6056 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
6057 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
6058 !vdev_dtl_required(oldvd
))
6062 * If there are more than two spares attached to a disk,
6063 * and those spares are not required, then we want to
6064 * attempt to free them up now so that they can be used
6065 * by other pools. Once we're back down to a single
6066 * disk+spare, we stop removing them.
6068 if (vd
->vdev_children
> 2) {
6069 newvd
= vd
->vdev_child
[1];
6071 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
6072 vdev_dtl_empty(last
, DTL_MISSING
) &&
6073 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
6074 !vdev_dtl_required(newvd
))
6083 spa_vdev_resilver_done(spa_t
*spa
)
6085 vdev_t
*vd
, *pvd
, *ppvd
;
6086 uint64_t guid
, sguid
, pguid
, ppguid
;
6088 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6090 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
6091 pvd
= vd
->vdev_parent
;
6092 ppvd
= pvd
->vdev_parent
;
6093 guid
= vd
->vdev_guid
;
6094 pguid
= pvd
->vdev_guid
;
6095 ppguid
= ppvd
->vdev_guid
;
6098 * If we have just finished replacing a hot spared device, then
6099 * we need to detach the parent's first child (the original hot
6102 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
6103 ppvd
->vdev_children
== 2) {
6104 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
6105 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
6107 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
6109 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6110 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
6112 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
6114 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6117 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6121 * Update the stored path or FRU for this vdev.
6124 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
6128 boolean_t sync
= B_FALSE
;
6130 ASSERT(spa_writeable(spa
));
6132 spa_vdev_state_enter(spa
, SCL_ALL
);
6134 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
6135 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
6137 if (!vd
->vdev_ops
->vdev_op_leaf
)
6138 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
6141 if (strcmp(value
, vd
->vdev_path
) != 0) {
6142 spa_strfree(vd
->vdev_path
);
6143 vd
->vdev_path
= spa_strdup(value
);
6147 if (vd
->vdev_fru
== NULL
) {
6148 vd
->vdev_fru
= spa_strdup(value
);
6150 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
6151 spa_strfree(vd
->vdev_fru
);
6152 vd
->vdev_fru
= spa_strdup(value
);
6157 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
6161 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
6163 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
6167 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
6169 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
6173 * ==========================================================================
6175 * ==========================================================================
6178 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
6180 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6182 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6183 return (SET_ERROR(EBUSY
));
6185 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
6189 spa_scan_stop(spa_t
*spa
)
6191 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6192 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6193 return (SET_ERROR(EBUSY
));
6194 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
6198 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
6200 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6202 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
6203 return (SET_ERROR(ENOTSUP
));
6206 * If a resilver was requested, but there is no DTL on a
6207 * writeable leaf device, we have nothing to do.
6209 if (func
== POOL_SCAN_RESILVER
&&
6210 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
6211 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6215 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6219 * ==========================================================================
6220 * SPA async task processing
6221 * ==========================================================================
6225 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6227 if (vd
->vdev_remove_wanted
) {
6228 vd
->vdev_remove_wanted
= B_FALSE
;
6229 vd
->vdev_delayed_close
= B_FALSE
;
6230 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6233 * We want to clear the stats, but we don't want to do a full
6234 * vdev_clear() as that will cause us to throw away
6235 * degraded/faulted state as well as attempt to reopen the
6236 * device, all of which is a waste.
6238 vd
->vdev_stat
.vs_read_errors
= 0;
6239 vd
->vdev_stat
.vs_write_errors
= 0;
6240 vd
->vdev_stat
.vs_checksum_errors
= 0;
6242 vdev_state_dirty(vd
->vdev_top
);
6245 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6246 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6250 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6252 if (vd
->vdev_probe_wanted
) {
6253 vd
->vdev_probe_wanted
= B_FALSE
;
6254 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6257 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6258 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6262 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6264 if (!spa
->spa_autoexpand
)
6267 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6268 vdev_t
*cvd
= vd
->vdev_child
[c
];
6269 spa_async_autoexpand(spa
, cvd
);
6272 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6275 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
6279 spa_async_thread(void *arg
)
6281 spa_t
*spa
= (spa_t
*)arg
;
6284 ASSERT(spa
->spa_sync_on
);
6286 mutex_enter(&spa
->spa_async_lock
);
6287 tasks
= spa
->spa_async_tasks
;
6288 spa
->spa_async_tasks
= 0;
6289 mutex_exit(&spa
->spa_async_lock
);
6292 * See if the config needs to be updated.
6294 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6295 uint64_t old_space
, new_space
;
6297 mutex_enter(&spa_namespace_lock
);
6298 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6299 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6300 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6301 mutex_exit(&spa_namespace_lock
);
6304 * If the pool grew as a result of the config update,
6305 * then log an internal history event.
6307 if (new_space
!= old_space
) {
6308 spa_history_log_internal(spa
, "vdev online", NULL
,
6309 "pool '%s' size: %llu(+%llu)",
6310 spa_name(spa
), new_space
, new_space
- old_space
);
6315 * See if any devices need to be marked REMOVED.
6317 if (tasks
& SPA_ASYNC_REMOVE
) {
6318 spa_vdev_state_enter(spa
, SCL_NONE
);
6319 spa_async_remove(spa
, spa
->spa_root_vdev
);
6320 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6321 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6322 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6323 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6324 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6327 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6328 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6329 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6330 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6334 * See if any devices need to be probed.
6336 if (tasks
& SPA_ASYNC_PROBE
) {
6337 spa_vdev_state_enter(spa
, SCL_NONE
);
6338 spa_async_probe(spa
, spa
->spa_root_vdev
);
6339 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6343 * If any devices are done replacing, detach them.
6345 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6346 spa_vdev_resilver_done(spa
);
6349 * Kick off a resilver.
6351 if (tasks
& SPA_ASYNC_RESILVER
)
6352 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6355 * Let the world know that we're done.
6357 mutex_enter(&spa
->spa_async_lock
);
6358 spa
->spa_async_thread
= NULL
;
6359 cv_broadcast(&spa
->spa_async_cv
);
6360 mutex_exit(&spa
->spa_async_lock
);
6365 spa_async_suspend(spa_t
*spa
)
6367 mutex_enter(&spa
->spa_async_lock
);
6368 spa
->spa_async_suspended
++;
6369 while (spa
->spa_async_thread
!= NULL
)
6370 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6371 mutex_exit(&spa
->spa_async_lock
);
6373 spa_vdev_remove_suspend(spa
);
6375 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
6376 if (condense_thread
!= NULL
&& zthr_isrunning(condense_thread
))
6377 VERIFY0(zthr_cancel(condense_thread
));
6381 spa_async_resume(spa_t
*spa
)
6383 mutex_enter(&spa
->spa_async_lock
);
6384 ASSERT(spa
->spa_async_suspended
!= 0);
6385 spa
->spa_async_suspended
--;
6386 mutex_exit(&spa
->spa_async_lock
);
6387 spa_restart_removal(spa
);
6389 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
6390 if (condense_thread
!= NULL
&& !zthr_isrunning(condense_thread
))
6391 zthr_resume(condense_thread
);
6395 spa_async_tasks_pending(spa_t
*spa
)
6397 uint_t non_config_tasks
;
6399 boolean_t config_task_suspended
;
6401 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6402 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6403 if (spa
->spa_ccw_fail_time
== 0) {
6404 config_task_suspended
= B_FALSE
;
6406 config_task_suspended
=
6407 (gethrtime() - spa
->spa_ccw_fail_time
) <
6408 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
6411 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6415 spa_async_dispatch(spa_t
*spa
)
6417 mutex_enter(&spa
->spa_async_lock
);
6418 if (spa_async_tasks_pending(spa
) &&
6419 !spa
->spa_async_suspended
&&
6420 spa
->spa_async_thread
== NULL
&&
6422 spa
->spa_async_thread
= thread_create(NULL
, 0,
6423 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6424 mutex_exit(&spa
->spa_async_lock
);
6428 spa_async_request(spa_t
*spa
, int task
)
6430 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6431 mutex_enter(&spa
->spa_async_lock
);
6432 spa
->spa_async_tasks
|= task
;
6433 mutex_exit(&spa
->spa_async_lock
);
6437 * ==========================================================================
6438 * SPA syncing routines
6439 * ==========================================================================
6443 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6446 bpobj_enqueue(bpo
, bp
, tx
);
6451 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6455 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6461 * Note: this simple function is not inlined to make it easier to dtrace the
6462 * amount of time spent syncing frees.
6465 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6467 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6468 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6469 VERIFY(zio_wait(zio
) == 0);
6473 * Note: this simple function is not inlined to make it easier to dtrace the
6474 * amount of time spent syncing deferred frees.
6477 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6479 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6480 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6481 spa_free_sync_cb
, zio
, tx
), ==, 0);
6482 VERIFY0(zio_wait(zio
));
6486 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6488 char *packed
= NULL
;
6493 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6496 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6497 * information. This avoids the dmu_buf_will_dirty() path and
6498 * saves us a pre-read to get data we don't actually care about.
6500 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6501 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6503 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6505 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6507 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6509 vmem_free(packed
, bufsize
);
6511 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6512 dmu_buf_will_dirty(db
, tx
);
6513 *(uint64_t *)db
->db_data
= nvsize
;
6514 dmu_buf_rele(db
, FTAG
);
6518 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6519 const char *config
, const char *entry
)
6529 * Update the MOS nvlist describing the list of available devices.
6530 * spa_validate_aux() will have already made sure this nvlist is
6531 * valid and the vdevs are labeled appropriately.
6533 if (sav
->sav_object
== 0) {
6534 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6535 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6536 sizeof (uint64_t), tx
);
6537 VERIFY(zap_update(spa
->spa_meta_objset
,
6538 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6539 &sav
->sav_object
, tx
) == 0);
6542 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6543 if (sav
->sav_count
== 0) {
6544 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6546 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6547 for (i
= 0; i
< sav
->sav_count
; i
++)
6548 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6549 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6550 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6551 sav
->sav_count
) == 0);
6552 for (i
= 0; i
< sav
->sav_count
; i
++)
6553 nvlist_free(list
[i
]);
6554 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6557 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6558 nvlist_free(nvroot
);
6560 sav
->sav_sync
= B_FALSE
;
6564 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6565 * The all-vdev ZAP must be empty.
6568 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6570 spa_t
*spa
= vd
->vdev_spa
;
6572 if (vd
->vdev_top_zap
!= 0) {
6573 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6574 vd
->vdev_top_zap
, tx
));
6576 if (vd
->vdev_leaf_zap
!= 0) {
6577 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6578 vd
->vdev_leaf_zap
, tx
));
6580 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
6581 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6586 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6591 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6592 * its config may not be dirty but we still need to build per-vdev ZAPs.
6593 * Similarly, if the pool is being assembled (e.g. after a split), we
6594 * need to rebuild the AVZ although the config may not be dirty.
6596 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6597 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6600 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6602 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6603 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6604 spa
->spa_all_vdev_zaps
!= 0);
6606 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6607 /* Make and build the new AVZ */
6608 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6609 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6610 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6612 /* Diff old AVZ with new one */
6616 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6617 spa
->spa_all_vdev_zaps
);
6618 zap_cursor_retrieve(&zc
, &za
) == 0;
6619 zap_cursor_advance(&zc
)) {
6620 uint64_t vdzap
= za
.za_first_integer
;
6621 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6624 * ZAP is listed in old AVZ but not in new one;
6627 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6632 zap_cursor_fini(&zc
);
6634 /* Destroy the old AVZ */
6635 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6636 spa
->spa_all_vdev_zaps
, tx
));
6638 /* Replace the old AVZ in the dir obj with the new one */
6639 VERIFY0(zap_update(spa
->spa_meta_objset
,
6640 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6641 sizeof (new_avz
), 1, &new_avz
, tx
));
6643 spa
->spa_all_vdev_zaps
= new_avz
;
6644 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6648 /* Walk through the AVZ and destroy all listed ZAPs */
6649 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6650 spa
->spa_all_vdev_zaps
);
6651 zap_cursor_retrieve(&zc
, &za
) == 0;
6652 zap_cursor_advance(&zc
)) {
6653 uint64_t zap
= za
.za_first_integer
;
6654 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6657 zap_cursor_fini(&zc
);
6659 /* Destroy and unlink the AVZ itself */
6660 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6661 spa
->spa_all_vdev_zaps
, tx
));
6662 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6663 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6664 spa
->spa_all_vdev_zaps
= 0;
6667 if (spa
->spa_all_vdev_zaps
== 0) {
6668 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6669 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6670 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6672 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6674 /* Create ZAPs for vdevs that don't have them. */
6675 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6677 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6678 dmu_tx_get_txg(tx
), B_FALSE
);
6681 * If we're upgrading the spa version then make sure that
6682 * the config object gets updated with the correct version.
6684 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6685 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6686 spa
->spa_uberblock
.ub_version
);
6688 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6690 nvlist_free(spa
->spa_config_syncing
);
6691 spa
->spa_config_syncing
= config
;
6693 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6697 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6699 uint64_t *versionp
= arg
;
6700 uint64_t version
= *versionp
;
6701 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6704 * Setting the version is special cased when first creating the pool.
6706 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6708 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6709 ASSERT(version
>= spa_version(spa
));
6711 spa
->spa_uberblock
.ub_version
= version
;
6712 vdev_config_dirty(spa
->spa_root_vdev
);
6713 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6717 * Set zpool properties.
6720 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6722 nvlist_t
*nvp
= arg
;
6723 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6724 objset_t
*mos
= spa
->spa_meta_objset
;
6725 nvpair_t
*elem
= NULL
;
6727 mutex_enter(&spa
->spa_props_lock
);
6729 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6731 char *strval
, *fname
;
6733 const char *propname
;
6734 zprop_type_t proptype
;
6737 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
6738 case ZPOOL_PROP_INVAL
:
6740 * We checked this earlier in spa_prop_validate().
6742 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6744 fname
= strchr(nvpair_name(elem
), '@') + 1;
6745 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6747 spa_feature_enable(spa
, fid
, tx
);
6748 spa_history_log_internal(spa
, "set", tx
,
6749 "%s=enabled", nvpair_name(elem
));
6752 case ZPOOL_PROP_VERSION
:
6753 intval
= fnvpair_value_uint64(elem
);
6755 * The version is synced separately before other
6756 * properties and should be correct by now.
6758 ASSERT3U(spa_version(spa
), >=, intval
);
6761 case ZPOOL_PROP_ALTROOT
:
6763 * 'altroot' is a non-persistent property. It should
6764 * have been set temporarily at creation or import time.
6766 ASSERT(spa
->spa_root
!= NULL
);
6769 case ZPOOL_PROP_READONLY
:
6770 case ZPOOL_PROP_CACHEFILE
:
6772 * 'readonly' and 'cachefile' are also non-persisitent
6776 case ZPOOL_PROP_COMMENT
:
6777 strval
= fnvpair_value_string(elem
);
6778 if (spa
->spa_comment
!= NULL
)
6779 spa_strfree(spa
->spa_comment
);
6780 spa
->spa_comment
= spa_strdup(strval
);
6782 * We need to dirty the configuration on all the vdevs
6783 * so that their labels get updated. It's unnecessary
6784 * to do this for pool creation since the vdev's
6785 * configuration has already been dirtied.
6787 if (tx
->tx_txg
!= TXG_INITIAL
)
6788 vdev_config_dirty(spa
->spa_root_vdev
);
6789 spa_history_log_internal(spa
, "set", tx
,
6790 "%s=%s", nvpair_name(elem
), strval
);
6794 * Set pool property values in the poolprops mos object.
6796 if (spa
->spa_pool_props_object
== 0) {
6797 spa
->spa_pool_props_object
=
6798 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6799 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6803 /* normalize the property name */
6804 propname
= zpool_prop_to_name(prop
);
6805 proptype
= zpool_prop_get_type(prop
);
6807 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6808 ASSERT(proptype
== PROP_TYPE_STRING
);
6809 strval
= fnvpair_value_string(elem
);
6810 VERIFY0(zap_update(mos
,
6811 spa
->spa_pool_props_object
, propname
,
6812 1, strlen(strval
) + 1, strval
, tx
));
6813 spa_history_log_internal(spa
, "set", tx
,
6814 "%s=%s", nvpair_name(elem
), strval
);
6815 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6816 intval
= fnvpair_value_uint64(elem
);
6818 if (proptype
== PROP_TYPE_INDEX
) {
6820 VERIFY0(zpool_prop_index_to_string(
6821 prop
, intval
, &unused
));
6823 VERIFY0(zap_update(mos
,
6824 spa
->spa_pool_props_object
, propname
,
6825 8, 1, &intval
, tx
));
6826 spa_history_log_internal(spa
, "set", tx
,
6827 "%s=%lld", nvpair_name(elem
), intval
);
6829 ASSERT(0); /* not allowed */
6833 case ZPOOL_PROP_DELEGATION
:
6834 spa
->spa_delegation
= intval
;
6836 case ZPOOL_PROP_BOOTFS
:
6837 spa
->spa_bootfs
= intval
;
6839 case ZPOOL_PROP_FAILUREMODE
:
6840 spa
->spa_failmode
= intval
;
6842 case ZPOOL_PROP_AUTOEXPAND
:
6843 spa
->spa_autoexpand
= intval
;
6844 if (tx
->tx_txg
!= TXG_INITIAL
)
6845 spa_async_request(spa
,
6846 SPA_ASYNC_AUTOEXPAND
);
6848 case ZPOOL_PROP_MULTIHOST
:
6849 spa
->spa_multihost
= intval
;
6851 case ZPOOL_PROP_DEDUPDITTO
:
6852 spa
->spa_dedup_ditto
= intval
;
6861 mutex_exit(&spa
->spa_props_lock
);
6865 * Perform one-time upgrade on-disk changes. spa_version() does not
6866 * reflect the new version this txg, so there must be no changes this
6867 * txg to anything that the upgrade code depends on after it executes.
6868 * Therefore this must be called after dsl_pool_sync() does the sync
6872 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6874 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6876 ASSERT(spa
->spa_sync_pass
== 1);
6878 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6880 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6881 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6882 dsl_pool_create_origin(dp
, tx
);
6884 /* Keeping the origin open increases spa_minref */
6885 spa
->spa_minref
+= 3;
6888 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6889 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6890 dsl_pool_upgrade_clones(dp
, tx
);
6893 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6894 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6895 dsl_pool_upgrade_dir_clones(dp
, tx
);
6897 /* Keeping the freedir open increases spa_minref */
6898 spa
->spa_minref
+= 3;
6901 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6902 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6903 spa_feature_create_zap_objects(spa
, tx
);
6907 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6908 * when possibility to use lz4 compression for metadata was added
6909 * Old pools that have this feature enabled must be upgraded to have
6910 * this feature active
6912 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6913 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6914 SPA_FEATURE_LZ4_COMPRESS
);
6915 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6916 SPA_FEATURE_LZ4_COMPRESS
);
6918 if (lz4_en
&& !lz4_ac
)
6919 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6923 * If we haven't written the salt, do so now. Note that the
6924 * feature may not be activated yet, but that's fine since
6925 * the presence of this ZAP entry is backwards compatible.
6927 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
6928 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
6929 VERIFY0(zap_add(spa
->spa_meta_objset
,
6930 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
6931 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
6932 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
6935 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6939 vdev_indirect_state_sync_verify(vdev_t
*vd
)
6941 ASSERTV(vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
);
6942 ASSERTV(vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
);
6944 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
6945 ASSERT(vim
!= NULL
);
6946 ASSERT(vib
!= NULL
);
6949 if (vdev_obsolete_sm_object(vd
) != 0) {
6950 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
6951 ASSERT(vd
->vdev_removing
||
6952 vd
->vdev_ops
== &vdev_indirect_ops
);
6953 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
6954 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
6956 ASSERT3U(vdev_obsolete_sm_object(vd
), ==,
6957 space_map_object(vd
->vdev_obsolete_sm
));
6958 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
6959 space_map_allocated(vd
->vdev_obsolete_sm
));
6961 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
6964 * Since frees / remaps to an indirect vdev can only
6965 * happen in syncing context, the obsolete segments
6966 * tree must be empty when we start syncing.
6968 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
6972 * Sync the specified transaction group. New blocks may be dirtied as
6973 * part of the process, so we iterate until it converges.
6976 spa_sync(spa_t
*spa
, uint64_t txg
)
6978 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6979 objset_t
*mos
= spa
->spa_meta_objset
;
6980 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6981 vdev_t
*rvd
= spa
->spa_root_vdev
;
6985 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
6986 zfs_vdev_queue_depth_pct
/ 100;
6988 VERIFY(spa_writeable(spa
));
6991 * Wait for i/os issued in open context that need to complete
6992 * before this txg syncs.
6994 VERIFY0(zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]));
6995 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
, 0);
6998 * Lock out configuration changes.
7000 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
7002 spa
->spa_syncing_txg
= txg
;
7003 spa
->spa_sync_pass
= 0;
7005 mutex_enter(&spa
->spa_alloc_lock
);
7006 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7007 mutex_exit(&spa
->spa_alloc_lock
);
7010 * If there are any pending vdev state changes, convert them
7011 * into config changes that go out with this transaction group.
7013 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
7014 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
7016 * We need the write lock here because, for aux vdevs,
7017 * calling vdev_config_dirty() modifies sav_config.
7018 * This is ugly and will become unnecessary when we
7019 * eliminate the aux vdev wart by integrating all vdevs
7020 * into the root vdev tree.
7022 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7023 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
7024 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
7025 vdev_state_clean(vd
);
7026 vdev_config_dirty(vd
);
7028 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7029 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7031 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7033 tx
= dmu_tx_create_assigned(dp
, txg
);
7035 spa
->spa_sync_starttime
= gethrtime();
7036 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
7037 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
7038 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
7039 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
7042 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
7043 * set spa_deflate if we have no raid-z vdevs.
7045 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
7046 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
7049 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
7050 vd
= rvd
->vdev_child
[i
];
7051 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
7054 if (i
== rvd
->vdev_children
) {
7055 spa
->spa_deflate
= TRUE
;
7056 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
7057 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
7058 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
7063 * Set the top-level vdev's max queue depth. Evaluate each
7064 * top-level's async write queue depth in case it changed.
7065 * The max queue depth will not change in the middle of syncing
7068 uint64_t queue_depth_total
= 0;
7069 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
7070 vdev_t
*tvd
= rvd
->vdev_child
[c
];
7071 metaslab_group_t
*mg
= tvd
->vdev_mg
;
7073 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
7074 !metaslab_group_initialized(mg
))
7078 * It is safe to do a lock-free check here because only async
7079 * allocations look at mg_max_alloc_queue_depth, and async
7080 * allocations all happen from spa_sync().
7082 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
7083 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
7084 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
7086 metaslab_class_t
*mc
= spa_normal_class(spa
);
7087 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
7088 mc
->mc_alloc_max_slots
= queue_depth_total
;
7089 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
7091 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
7092 max_queue_depth
* rvd
->vdev_children
);
7094 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
7095 vdev_t
*vd
= rvd
->vdev_child
[c
];
7096 vdev_indirect_state_sync_verify(vd
);
7098 if (vdev_indirect_should_condense(vd
)) {
7099 spa_condense_indirect_start_sync(vd
, tx
);
7105 * Iterate to convergence.
7108 int pass
= ++spa
->spa_sync_pass
;
7110 spa_sync_config_object(spa
, tx
);
7111 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
7112 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
7113 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
7114 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
7115 spa_errlog_sync(spa
, txg
);
7116 dsl_pool_sync(dp
, txg
);
7118 if (pass
< zfs_sync_pass_deferred_free
) {
7119 spa_sync_frees(spa
, free_bpl
, tx
);
7122 * We can not defer frees in pass 1, because
7123 * we sync the deferred frees later in pass 1.
7125 ASSERT3U(pass
, >, 1);
7126 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
7127 &spa
->spa_deferred_bpobj
, tx
);
7131 dsl_scan_sync(dp
, tx
);
7133 if (spa
->spa_vdev_removal
!= NULL
)
7136 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
7141 spa_sync_upgrades(spa
, tx
);
7143 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
7145 * Note: We need to check if the MOS is dirty
7146 * because we could have marked the MOS dirty
7147 * without updating the uberblock (e.g. if we
7148 * have sync tasks but no dirty user data). We
7149 * need to check the uberblock's rootbp because
7150 * it is updated if we have synced out dirty
7151 * data (though in this case the MOS will most
7152 * likely also be dirty due to second order
7153 * effects, we don't want to rely on that here).
7155 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
7156 !dmu_objset_is_dirty(mos
, txg
)) {
7158 * Nothing changed on the first pass,
7159 * therefore this TXG is a no-op. Avoid
7160 * syncing deferred frees, so that we
7161 * can keep this TXG as a no-op.
7163 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
7165 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7166 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
7169 spa_sync_deferred_frees(spa
, tx
);
7172 } while (dmu_objset_is_dirty(mos
, txg
));
7175 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
7177 * Make sure that the number of ZAPs for all the vdevs matches
7178 * the number of ZAPs in the per-vdev ZAP list. This only gets
7179 * called if the config is dirty; otherwise there may be
7180 * outstanding AVZ operations that weren't completed in
7181 * spa_sync_config_object.
7183 uint64_t all_vdev_zap_entry_count
;
7184 ASSERT0(zap_count(spa
->spa_meta_objset
,
7185 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
7186 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
7187 all_vdev_zap_entry_count
);
7191 if (spa
->spa_vdev_removal
!= NULL
) {
7192 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
7196 * Rewrite the vdev configuration (which includes the uberblock)
7197 * to commit the transaction group.
7199 * If there are no dirty vdevs, we sync the uberblock to a few
7200 * random top-level vdevs that are known to be visible in the
7201 * config cache (see spa_vdev_add() for a complete description).
7202 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7206 * We hold SCL_STATE to prevent vdev open/close/etc.
7207 * while we're attempting to write the vdev labels.
7209 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
7211 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
7212 vdev_t
*svd
[SPA_DVAS_PER_BP
];
7214 int children
= rvd
->vdev_children
;
7215 int c0
= spa_get_random(children
);
7217 for (int c
= 0; c
< children
; c
++) {
7218 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
7219 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
7220 !vdev_is_concrete(vd
))
7222 svd
[svdcount
++] = vd
;
7223 if (svdcount
== SPA_DVAS_PER_BP
)
7226 error
= vdev_config_sync(svd
, svdcount
, txg
);
7228 error
= vdev_config_sync(rvd
->vdev_child
,
7229 rvd
->vdev_children
, txg
);
7233 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
7235 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7239 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
7240 zio_resume_wait(spa
);
7244 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
7245 spa
->spa_deadman_tqid
= 0;
7248 * Clear the dirty config list.
7250 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
7251 vdev_config_clean(vd
);
7254 * Now that the new config has synced transactionally,
7255 * let it become visible to the config cache.
7257 if (spa
->spa_config_syncing
!= NULL
) {
7258 spa_config_set(spa
, spa
->spa_config_syncing
);
7259 spa
->spa_config_txg
= txg
;
7260 spa
->spa_config_syncing
= NULL
;
7263 dsl_pool_sync_done(dp
, txg
);
7265 mutex_enter(&spa
->spa_alloc_lock
);
7266 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7267 mutex_exit(&spa
->spa_alloc_lock
);
7270 * Update usable space statistics.
7272 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
7273 vdev_sync_done(vd
, txg
);
7275 spa_update_dspace(spa
);
7278 * It had better be the case that we didn't dirty anything
7279 * since vdev_config_sync().
7281 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7282 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7283 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7285 spa
->spa_sync_pass
= 0;
7288 * Update the last synced uberblock here. We want to do this at
7289 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7290 * will be guaranteed that all the processing associated with
7291 * that txg has been completed.
7293 spa
->spa_ubsync
= spa
->spa_uberblock
;
7294 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7296 spa_handle_ignored_writes(spa
);
7299 * If any async tasks have been requested, kick them off.
7301 spa_async_dispatch(spa
);
7305 * Sync all pools. We don't want to hold the namespace lock across these
7306 * operations, so we take a reference on the spa_t and drop the lock during the
7310 spa_sync_allpools(void)
7313 mutex_enter(&spa_namespace_lock
);
7314 while ((spa
= spa_next(spa
)) != NULL
) {
7315 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7316 !spa_writeable(spa
) || spa_suspended(spa
))
7318 spa_open_ref(spa
, FTAG
);
7319 mutex_exit(&spa_namespace_lock
);
7320 txg_wait_synced(spa_get_dsl(spa
), 0);
7321 mutex_enter(&spa_namespace_lock
);
7322 spa_close(spa
, FTAG
);
7324 mutex_exit(&spa_namespace_lock
);
7328 * ==========================================================================
7329 * Miscellaneous routines
7330 * ==========================================================================
7334 * Remove all pools in the system.
7342 * Remove all cached state. All pools should be closed now,
7343 * so every spa in the AVL tree should be unreferenced.
7345 mutex_enter(&spa_namespace_lock
);
7346 while ((spa
= spa_next(NULL
)) != NULL
) {
7348 * Stop async tasks. The async thread may need to detach
7349 * a device that's been replaced, which requires grabbing
7350 * spa_namespace_lock, so we must drop it here.
7352 spa_open_ref(spa
, FTAG
);
7353 mutex_exit(&spa_namespace_lock
);
7354 spa_async_suspend(spa
);
7355 mutex_enter(&spa_namespace_lock
);
7356 spa_close(spa
, FTAG
);
7358 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7360 spa_deactivate(spa
);
7364 mutex_exit(&spa_namespace_lock
);
7368 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7373 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7377 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7378 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7379 if (vd
->vdev_guid
== guid
)
7383 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7384 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7385 if (vd
->vdev_guid
== guid
)
7394 spa_upgrade(spa_t
*spa
, uint64_t version
)
7396 ASSERT(spa_writeable(spa
));
7398 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7401 * This should only be called for a non-faulted pool, and since a
7402 * future version would result in an unopenable pool, this shouldn't be
7405 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7406 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7408 spa
->spa_uberblock
.ub_version
= version
;
7409 vdev_config_dirty(spa
->spa_root_vdev
);
7411 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7413 txg_wait_synced(spa_get_dsl(spa
), 0);
7417 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7421 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7423 for (i
= 0; i
< sav
->sav_count
; i
++)
7424 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7427 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7428 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7429 &spareguid
) == 0 && spareguid
== guid
)
7437 * Check if a pool has an active shared spare device.
7438 * Note: reference count of an active spare is 2, as a spare and as a replace
7441 spa_has_active_shared_spare(spa_t
*spa
)
7445 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7447 for (i
= 0; i
< sav
->sav_count
; i
++) {
7448 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7449 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7458 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7460 sysevent_t
*ev
= NULL
;
7464 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
7466 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
7467 ev
->resource
= resource
;
7474 spa_event_post(sysevent_t
*ev
)
7478 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
7479 kmem_free(ev
, sizeof (*ev
));
7485 * Post a zevent corresponding to the given sysevent. The 'name' must be one
7486 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
7487 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7488 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7489 * or zdb as real changes.
7492 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7494 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
7497 #if defined(_KERNEL) && defined(HAVE_SPL)
7498 /* state manipulation functions */
7499 EXPORT_SYMBOL(spa_open
);
7500 EXPORT_SYMBOL(spa_open_rewind
);
7501 EXPORT_SYMBOL(spa_get_stats
);
7502 EXPORT_SYMBOL(spa_create
);
7503 EXPORT_SYMBOL(spa_import
);
7504 EXPORT_SYMBOL(spa_tryimport
);
7505 EXPORT_SYMBOL(spa_destroy
);
7506 EXPORT_SYMBOL(spa_export
);
7507 EXPORT_SYMBOL(spa_reset
);
7508 EXPORT_SYMBOL(spa_async_request
);
7509 EXPORT_SYMBOL(spa_async_suspend
);
7510 EXPORT_SYMBOL(spa_async_resume
);
7511 EXPORT_SYMBOL(spa_inject_addref
);
7512 EXPORT_SYMBOL(spa_inject_delref
);
7513 EXPORT_SYMBOL(spa_scan_stat_init
);
7514 EXPORT_SYMBOL(spa_scan_get_stats
);
7516 /* device maniion */
7517 EXPORT_SYMBOL(spa_vdev_add
);
7518 EXPORT_SYMBOL(spa_vdev_attach
);
7519 EXPORT_SYMBOL(spa_vdev_detach
);
7520 EXPORT_SYMBOL(spa_vdev_setpath
);
7521 EXPORT_SYMBOL(spa_vdev_setfru
);
7522 EXPORT_SYMBOL(spa_vdev_split_mirror
);
7524 /* spare statech is global across all pools) */
7525 EXPORT_SYMBOL(spa_spare_add
);
7526 EXPORT_SYMBOL(spa_spare_remove
);
7527 EXPORT_SYMBOL(spa_spare_exists
);
7528 EXPORT_SYMBOL(spa_spare_activate
);
7530 /* L2ARC statech is global across all pools) */
7531 EXPORT_SYMBOL(spa_l2cache_add
);
7532 EXPORT_SYMBOL(spa_l2cache_remove
);
7533 EXPORT_SYMBOL(spa_l2cache_exists
);
7534 EXPORT_SYMBOL(spa_l2cache_activate
);
7535 EXPORT_SYMBOL(spa_l2cache_drop
);
7538 EXPORT_SYMBOL(spa_scan
);
7539 EXPORT_SYMBOL(spa_scan_stop
);
7542 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
7543 EXPORT_SYMBOL(spa_sync_allpools
);
7546 EXPORT_SYMBOL(spa_prop_set
);
7547 EXPORT_SYMBOL(spa_prop_get
);
7548 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
7550 /* asynchronous event notification */
7551 EXPORT_SYMBOL(spa_event_notify
);
7554 #if defined(_KERNEL) && defined(HAVE_SPL)
7555 module_param(spa_load_verify_maxinflight
, int, 0644);
7556 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
7557 "Max concurrent traversal I/Os while verifying pool during import -X");
7559 module_param(spa_load_verify_metadata
, int, 0644);
7560 MODULE_PARM_DESC(spa_load_verify_metadata
,
7561 "Set to traverse metadata on pool import");
7563 module_param(spa_load_verify_data
, int, 0644);
7564 MODULE_PARM_DESC(spa_load_verify_data
,
7565 "Set to traverse data on pool import");
7568 module_param(zio_taskq_batch_pct
, uint
, 0444);
7569 MODULE_PARM_DESC(zio_taskq_batch_pct
,
7570 "Percentage of CPUs to run an IO worker thread");