4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
29 * SPA: Storage Pool Allocator
31 * This file contains all the routines used when modifying on-disk SPA state.
32 * This includes opening, importing, destroying, exporting a pool, and syncing a
36 #include <sys/zfs_context.h>
37 #include <sys/fm/fs/zfs.h>
38 #include <sys/spa_impl.h>
40 #include <sys/zio_checksum.h>
42 #include <sys/dmu_tx.h>
46 #include <sys/vdev_impl.h>
47 #include <sys/vdev_disk.h>
48 #include <sys/metaslab.h>
49 #include <sys/metaslab_impl.h>
50 #include <sys/uberblock_impl.h>
53 #include <sys/dmu_traverse.h>
54 #include <sys/dmu_objset.h>
55 #include <sys/unique.h>
56 #include <sys/dsl_pool.h>
57 #include <sys/dsl_dataset.h>
58 #include <sys/dsl_dir.h>
59 #include <sys/dsl_prop.h>
60 #include <sys/dsl_synctask.h>
61 #include <sys/fs/zfs.h>
63 #include <sys/callb.h>
64 #include <sys/systeminfo.h>
65 #include <sys/spa_boot.h>
66 #include <sys/zfs_ioctl.h>
67 #include <sys/dsl_scan.h>
68 #include <sys/zfeature.h>
69 #include <sys/dsl_destroy.h>
73 #include <sys/bootprops.h>
74 #include <sys/callb.h>
75 #include <sys/cpupart.h>
77 #include <sys/sysdc.h>
82 #include "zfs_comutil.h"
84 typedef enum zti_modes
{
85 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
86 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
87 ZTI_MODE_NULL
, /* don't create a taskq */
91 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
92 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
93 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
94 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
96 #define ZTI_N(n) ZTI_P(n, 1)
97 #define ZTI_ONE ZTI_N(1)
99 typedef struct zio_taskq_info
{
100 zti_modes_t zti_mode
;
105 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
106 "iss", "iss_h", "int", "int_h"
110 * This table defines the taskq settings for each ZFS I/O type. When
111 * initializing a pool, we use this table to create an appropriately sized
112 * taskq. Some operations are low volume and therefore have a small, static
113 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
114 * macros. Other operations process a large amount of data; the ZTI_BATCH
115 * macro causes us to create a taskq oriented for throughput. Some operations
116 * are so high frequency and short-lived that the taskq itself can become a a
117 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
118 * additional degree of parallelism specified by the number of threads per-
119 * taskq and the number of taskqs; when dispatching an event in this case, the
120 * particular taskq is chosen at random.
122 * The different taskq priorities are to handle the different contexts (issue
123 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
124 * need to be handled with minimum delay.
126 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
127 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
128 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
129 { ZTI_N(8), ZTI_NULL
, ZTI_BATCH
, ZTI_NULL
}, /* READ */
130 { ZTI_BATCH
, ZTI_N(5), ZTI_N(16), ZTI_N(5) }, /* WRITE */
131 { ZTI_P(4, 8), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
132 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
133 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
136 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
137 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
138 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
139 static inline int spa_load_impl(spa_t
*spa
, uint64_t, nvlist_t
*config
,
140 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
142 static void spa_vdev_resilver_done(spa_t
*spa
);
144 uint_t zio_taskq_batch_pct
= 75; /* 1 thread per cpu in pset */
145 id_t zio_taskq_psrset_bind
= PS_NONE
;
146 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
147 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
149 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
152 * This (illegal) pool name is used when temporarily importing a spa_t in order
153 * to get the vdev stats associated with the imported devices.
155 #define TRYIMPORT_NAME "$import"
158 * ==========================================================================
159 * SPA properties routines
160 * ==========================================================================
164 * Add a (source=src, propname=propval) list to an nvlist.
167 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
168 uint64_t intval
, zprop_source_t src
)
170 const char *propname
= zpool_prop_to_name(prop
);
173 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
174 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
177 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
179 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
181 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
182 nvlist_free(propval
);
186 * Get property values from the spa configuration.
189 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
191 vdev_t
*rvd
= spa
->spa_root_vdev
;
192 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
196 uint64_t cap
, version
;
197 zprop_source_t src
= ZPROP_SRC_NONE
;
198 spa_config_dirent_t
*dp
;
201 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
204 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
205 size
= metaslab_class_get_space(spa_normal_class(spa
));
206 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
207 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
208 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
209 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
213 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
214 vdev_t
*tvd
= rvd
->vdev_child
[c
];
215 space
+= tvd
->vdev_max_asize
- tvd
->vdev_asize
;
217 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
, space
,
220 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
221 (spa_mode(spa
) == FREAD
), src
);
223 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
224 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
226 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
227 ddt_get_pool_dedup_ratio(spa
), src
);
229 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
230 rvd
->vdev_state
, src
);
232 version
= spa_version(spa
);
233 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
))
234 src
= ZPROP_SRC_DEFAULT
;
236 src
= ZPROP_SRC_LOCAL
;
237 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
241 dsl_dir_t
*freedir
= pool
->dp_free_dir
;
244 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
245 * when opening pools before this version freedir will be NULL.
247 if (freedir
!= NULL
) {
248 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
249 freedir
->dd_phys
->dd_used_bytes
, src
);
251 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
256 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
258 if (spa
->spa_comment
!= NULL
) {
259 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
263 if (spa
->spa_root
!= NULL
)
264 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
267 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
268 if (dp
->scd_path
== NULL
) {
269 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
270 "none", 0, ZPROP_SRC_LOCAL
);
271 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
272 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
273 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
279 * Get zpool property values.
282 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
284 objset_t
*mos
= spa
->spa_meta_objset
;
289 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_PUSHPAGE
);
293 mutex_enter(&spa
->spa_props_lock
);
296 * Get properties from the spa config.
298 spa_prop_get_config(spa
, nvp
);
300 /* If no pool property object, no more prop to get. */
301 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
302 mutex_exit(&spa
->spa_props_lock
);
307 * Get properties from the MOS pool property object.
309 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
310 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
311 zap_cursor_advance(&zc
)) {
314 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
317 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
320 switch (za
.za_integer_length
) {
322 /* integer property */
323 if (za
.za_first_integer
!=
324 zpool_prop_default_numeric(prop
))
325 src
= ZPROP_SRC_LOCAL
;
327 if (prop
== ZPOOL_PROP_BOOTFS
) {
329 dsl_dataset_t
*ds
= NULL
;
331 dp
= spa_get_dsl(spa
);
332 dsl_pool_config_enter(dp
, FTAG
);
333 if ((err
= dsl_dataset_hold_obj(dp
,
334 za
.za_first_integer
, FTAG
, &ds
))) {
335 dsl_pool_config_exit(dp
, FTAG
);
340 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1,
342 dsl_dataset_name(ds
, strval
);
343 dsl_dataset_rele(ds
, FTAG
);
344 dsl_pool_config_exit(dp
, FTAG
);
347 intval
= za
.za_first_integer
;
350 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
354 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1);
359 /* string property */
360 strval
= kmem_alloc(za
.za_num_integers
, KM_PUSHPAGE
);
361 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
362 za
.za_name
, 1, za
.za_num_integers
, strval
);
364 kmem_free(strval
, za
.za_num_integers
);
367 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
368 kmem_free(strval
, za
.za_num_integers
);
375 zap_cursor_fini(&zc
);
376 mutex_exit(&spa
->spa_props_lock
);
378 if (err
&& err
!= ENOENT
) {
388 * Validate the given pool properties nvlist and modify the list
389 * for the property values to be set.
392 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
395 int error
= 0, reset_bootfs
= 0;
397 boolean_t has_feature
= B_FALSE
;
400 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
402 char *strval
, *slash
, *check
, *fname
;
403 const char *propname
= nvpair_name(elem
);
404 zpool_prop_t prop
= zpool_name_to_prop(propname
);
408 if (!zpool_prop_feature(propname
)) {
409 error
= SET_ERROR(EINVAL
);
414 * Sanitize the input.
416 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
417 error
= SET_ERROR(EINVAL
);
421 if (nvpair_value_uint64(elem
, &intval
) != 0) {
422 error
= SET_ERROR(EINVAL
);
427 error
= SET_ERROR(EINVAL
);
431 fname
= strchr(propname
, '@') + 1;
432 if (zfeature_lookup_name(fname
, NULL
) != 0) {
433 error
= SET_ERROR(EINVAL
);
437 has_feature
= B_TRUE
;
440 case ZPOOL_PROP_VERSION
:
441 error
= nvpair_value_uint64(elem
, &intval
);
443 (intval
< spa_version(spa
) ||
444 intval
> SPA_VERSION_BEFORE_FEATURES
||
446 error
= SET_ERROR(EINVAL
);
449 case ZPOOL_PROP_DELEGATION
:
450 case ZPOOL_PROP_AUTOREPLACE
:
451 case ZPOOL_PROP_LISTSNAPS
:
452 case ZPOOL_PROP_AUTOEXPAND
:
453 error
= nvpair_value_uint64(elem
, &intval
);
454 if (!error
&& intval
> 1)
455 error
= SET_ERROR(EINVAL
);
458 case ZPOOL_PROP_BOOTFS
:
460 * If the pool version is less than SPA_VERSION_BOOTFS,
461 * or the pool is still being created (version == 0),
462 * the bootfs property cannot be set.
464 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
465 error
= SET_ERROR(ENOTSUP
);
470 * Make sure the vdev config is bootable
472 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
473 error
= SET_ERROR(ENOTSUP
);
479 error
= nvpair_value_string(elem
, &strval
);
485 if (strval
== NULL
|| strval
[0] == '\0') {
486 objnum
= zpool_prop_default_numeric(
491 error
= dmu_objset_hold(strval
, FTAG
, &os
);
495 /* Must be ZPL and not gzip compressed. */
497 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
498 error
= SET_ERROR(ENOTSUP
);
500 dsl_prop_get_int_ds(dmu_objset_ds(os
),
501 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
503 !BOOTFS_COMPRESS_VALID(compress
)) {
504 error
= SET_ERROR(ENOTSUP
);
506 objnum
= dmu_objset_id(os
);
508 dmu_objset_rele(os
, FTAG
);
512 case ZPOOL_PROP_FAILUREMODE
:
513 error
= nvpair_value_uint64(elem
, &intval
);
514 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
515 intval
> ZIO_FAILURE_MODE_PANIC
))
516 error
= SET_ERROR(EINVAL
);
519 * This is a special case which only occurs when
520 * the pool has completely failed. This allows
521 * the user to change the in-core failmode property
522 * without syncing it out to disk (I/Os might
523 * currently be blocked). We do this by returning
524 * EIO to the caller (spa_prop_set) to trick it
525 * into thinking we encountered a property validation
528 if (!error
&& spa_suspended(spa
)) {
529 spa
->spa_failmode
= intval
;
530 error
= SET_ERROR(EIO
);
534 case ZPOOL_PROP_CACHEFILE
:
535 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
538 if (strval
[0] == '\0')
541 if (strcmp(strval
, "none") == 0)
544 if (strval
[0] != '/') {
545 error
= SET_ERROR(EINVAL
);
549 slash
= strrchr(strval
, '/');
550 ASSERT(slash
!= NULL
);
552 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
553 strcmp(slash
, "/..") == 0)
554 error
= SET_ERROR(EINVAL
);
557 case ZPOOL_PROP_COMMENT
:
558 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
560 for (check
= strval
; *check
!= '\0'; check
++) {
561 if (!isprint(*check
)) {
562 error
= SET_ERROR(EINVAL
);
567 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
568 error
= SET_ERROR(E2BIG
);
571 case ZPOOL_PROP_DEDUPDITTO
:
572 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
573 error
= SET_ERROR(ENOTSUP
);
575 error
= nvpair_value_uint64(elem
, &intval
);
577 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
578 error
= SET_ERROR(EINVAL
);
589 if (!error
&& reset_bootfs
) {
590 error
= nvlist_remove(props
,
591 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
594 error
= nvlist_add_uint64(props
,
595 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
603 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
606 spa_config_dirent_t
*dp
;
608 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
612 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
615 if (cachefile
[0] == '\0')
616 dp
->scd_path
= spa_strdup(spa_config_path
);
617 else if (strcmp(cachefile
, "none") == 0)
620 dp
->scd_path
= spa_strdup(cachefile
);
622 list_insert_head(&spa
->spa_config_list
, dp
);
624 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
628 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
631 nvpair_t
*elem
= NULL
;
632 boolean_t need_sync
= B_FALSE
;
634 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
637 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
638 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
640 if (prop
== ZPOOL_PROP_CACHEFILE
||
641 prop
== ZPOOL_PROP_ALTROOT
||
642 prop
== ZPOOL_PROP_READONLY
)
645 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPROP_INVAL
) {
648 if (prop
== ZPOOL_PROP_VERSION
) {
649 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
651 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
652 ver
= SPA_VERSION_FEATURES
;
656 /* Save time if the version is already set. */
657 if (ver
== spa_version(spa
))
661 * In addition to the pool directory object, we might
662 * create the pool properties object, the features for
663 * read object, the features for write object, or the
664 * feature descriptions object.
666 error
= dsl_sync_task(spa
->spa_name
, NULL
,
667 spa_sync_version
, &ver
, 6);
678 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
686 * If the bootfs property value is dsobj, clear it.
689 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
691 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
692 VERIFY(zap_remove(spa
->spa_meta_objset
,
693 spa
->spa_pool_props_object
,
694 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
701 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
703 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
704 vdev_t
*rvd
= spa
->spa_root_vdev
;
706 ASSERTV(uint64_t *newguid
= arg
);
708 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
709 vdev_state
= rvd
->vdev_state
;
710 spa_config_exit(spa
, SCL_STATE
, FTAG
);
712 if (vdev_state
!= VDEV_STATE_HEALTHY
)
713 return (SET_ERROR(ENXIO
));
715 ASSERT3U(spa_guid(spa
), !=, *newguid
);
721 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
723 uint64_t *newguid
= arg
;
724 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
726 vdev_t
*rvd
= spa
->spa_root_vdev
;
728 oldguid
= spa_guid(spa
);
730 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
731 rvd
->vdev_guid
= *newguid
;
732 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
733 vdev_config_dirty(rvd
);
734 spa_config_exit(spa
, SCL_STATE
, FTAG
);
736 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
741 * Change the GUID for the pool. This is done so that we can later
742 * re-import a pool built from a clone of our own vdevs. We will modify
743 * the root vdev's guid, our own pool guid, and then mark all of our
744 * vdevs dirty. Note that we must make sure that all our vdevs are
745 * online when we do this, or else any vdevs that weren't present
746 * would be orphaned from our pool. We are also going to issue a
747 * sysevent to update any watchers.
750 spa_change_guid(spa_t
*spa
)
755 mutex_enter(&spa
->spa_vdev_top_lock
);
756 mutex_enter(&spa_namespace_lock
);
757 guid
= spa_generate_guid(NULL
);
759 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
760 spa_change_guid_sync
, &guid
, 5);
763 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
764 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_REGUID
);
767 mutex_exit(&spa_namespace_lock
);
768 mutex_exit(&spa
->spa_vdev_top_lock
);
774 * ==========================================================================
775 * SPA state manipulation (open/create/destroy/import/export)
776 * ==========================================================================
780 spa_error_entry_compare(const void *a
, const void *b
)
782 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
783 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
786 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
787 sizeof (zbookmark_t
));
798 * Utility function which retrieves copies of the current logs and
799 * re-initializes them in the process.
802 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
804 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
806 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
807 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
809 avl_create(&spa
->spa_errlist_scrub
,
810 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
811 offsetof(spa_error_entry_t
, se_avl
));
812 avl_create(&spa
->spa_errlist_last
,
813 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
814 offsetof(spa_error_entry_t
, se_avl
));
818 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
820 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
821 enum zti_modes mode
= ztip
->zti_mode
;
822 uint_t value
= ztip
->zti_value
;
823 uint_t count
= ztip
->zti_count
;
824 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
827 boolean_t batch
= B_FALSE
;
829 if (mode
== ZTI_MODE_NULL
) {
831 tqs
->stqs_taskq
= NULL
;
835 ASSERT3U(count
, >, 0);
837 tqs
->stqs_count
= count
;
838 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
842 ASSERT3U(value
, >=, 1);
843 value
= MAX(value
, 1);
848 flags
|= TASKQ_THREADS_CPU_PCT
;
849 value
= zio_taskq_batch_pct
;
853 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
855 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
859 for (i
= 0; i
< count
; i
++) {
863 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
864 zio_type_name
[t
], zio_taskq_types
[q
], i
);
866 (void) snprintf(name
, sizeof (name
), "%s_%s",
867 zio_type_name
[t
], zio_taskq_types
[q
]);
870 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
872 flags
|= TASKQ_DC_BATCH
;
874 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
875 spa
->spa_proc
, zio_taskq_basedc
, flags
);
877 pri_t pri
= maxclsyspri
;
879 * The write issue taskq can be extremely CPU
880 * intensive. Run it at slightly lower priority
881 * than the other taskqs.
883 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
886 tq
= taskq_create_proc(name
, value
, pri
, 50,
887 INT_MAX
, spa
->spa_proc
, flags
);
890 tqs
->stqs_taskq
[i
] = tq
;
895 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
897 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
900 if (tqs
->stqs_taskq
== NULL
) {
901 ASSERT3U(tqs
->stqs_count
, ==, 0);
905 for (i
= 0; i
< tqs
->stqs_count
; i
++) {
906 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
907 taskq_destroy(tqs
->stqs_taskq
[i
]);
910 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
911 tqs
->stqs_taskq
= NULL
;
915 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
916 * Note that a type may have multiple discrete taskqs to avoid lock contention
917 * on the taskq itself. In that case we choose which taskq at random by using
918 * the low bits of gethrtime().
921 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
922 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
924 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
927 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
928 ASSERT3U(tqs
->stqs_count
, !=, 0);
930 if (tqs
->stqs_count
== 1) {
931 tq
= tqs
->stqs_taskq
[0];
933 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
936 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
940 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
943 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
944 task_func_t
*func
, void *arg
, uint_t flags
)
946 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
950 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
951 ASSERT3U(tqs
->stqs_count
, !=, 0);
953 if (tqs
->stqs_count
== 1) {
954 tq
= tqs
->stqs_taskq
[0];
956 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
959 id
= taskq_dispatch(tq
, func
, arg
, flags
);
961 taskq_wait_id(tq
, id
);
965 spa_create_zio_taskqs(spa_t
*spa
)
969 for (t
= 0; t
< ZIO_TYPES
; t
++) {
970 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
971 spa_taskqs_init(spa
, t
, q
);
976 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
978 spa_thread(void *arg
)
983 user_t
*pu
= PTOU(curproc
);
985 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
988 ASSERT(curproc
!= &p0
);
989 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
990 "zpool-%s", spa
->spa_name
);
991 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
993 /* bind this thread to the requested psrset */
994 if (zio_taskq_psrset_bind
!= PS_NONE
) {
996 mutex_enter(&cpu_lock
);
997 mutex_enter(&pidlock
);
998 mutex_enter(&curproc
->p_lock
);
1000 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1001 0, NULL
, NULL
) == 0) {
1002 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1005 "Couldn't bind process for zfs pool \"%s\" to "
1006 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1009 mutex_exit(&curproc
->p_lock
);
1010 mutex_exit(&pidlock
);
1011 mutex_exit(&cpu_lock
);
1015 if (zio_taskq_sysdc
) {
1016 sysdc_thread_enter(curthread
, 100, 0);
1019 spa
->spa_proc
= curproc
;
1020 spa
->spa_did
= curthread
->t_did
;
1022 spa_create_zio_taskqs(spa
);
1024 mutex_enter(&spa
->spa_proc_lock
);
1025 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1027 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1028 cv_broadcast(&spa
->spa_proc_cv
);
1030 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1031 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1032 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1033 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1035 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1036 spa
->spa_proc_state
= SPA_PROC_GONE
;
1037 spa
->spa_proc
= &p0
;
1038 cv_broadcast(&spa
->spa_proc_cv
);
1039 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1041 mutex_enter(&curproc
->p_lock
);
1047 * Activate an uninitialized pool.
1050 spa_activate(spa_t
*spa
, int mode
)
1052 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1054 spa
->spa_state
= POOL_STATE_ACTIVE
;
1055 spa
->spa_mode
= mode
;
1057 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1058 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1060 /* Try to create a covering process */
1061 mutex_enter(&spa
->spa_proc_lock
);
1062 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1063 ASSERT(spa
->spa_proc
== &p0
);
1066 #ifdef HAVE_SPA_THREAD
1067 /* Only create a process if we're going to be around a while. */
1068 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1069 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1071 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1072 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1073 cv_wait(&spa
->spa_proc_cv
,
1074 &spa
->spa_proc_lock
);
1076 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1077 ASSERT(spa
->spa_proc
!= &p0
);
1078 ASSERT(spa
->spa_did
!= 0);
1082 "Couldn't create process for zfs pool \"%s\"\n",
1087 #endif /* HAVE_SPA_THREAD */
1088 mutex_exit(&spa
->spa_proc_lock
);
1090 /* If we didn't create a process, we need to create our taskqs. */
1091 if (spa
->spa_proc
== &p0
) {
1092 spa_create_zio_taskqs(spa
);
1095 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1096 offsetof(vdev_t
, vdev_config_dirty_node
));
1097 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1098 offsetof(vdev_t
, vdev_state_dirty_node
));
1100 txg_list_create(&spa
->spa_vdev_txg_list
,
1101 offsetof(struct vdev
, vdev_txg_node
));
1103 avl_create(&spa
->spa_errlist_scrub
,
1104 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1105 offsetof(spa_error_entry_t
, se_avl
));
1106 avl_create(&spa
->spa_errlist_last
,
1107 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1108 offsetof(spa_error_entry_t
, se_avl
));
1112 * Opposite of spa_activate().
1115 spa_deactivate(spa_t
*spa
)
1119 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1120 ASSERT(spa
->spa_dsl_pool
== NULL
);
1121 ASSERT(spa
->spa_root_vdev
== NULL
);
1122 ASSERT(spa
->spa_async_zio_root
== NULL
);
1123 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1125 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1127 list_destroy(&spa
->spa_config_dirty_list
);
1128 list_destroy(&spa
->spa_state_dirty_list
);
1130 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
1132 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1133 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1134 spa_taskqs_fini(spa
, t
, q
);
1138 metaslab_class_destroy(spa
->spa_normal_class
);
1139 spa
->spa_normal_class
= NULL
;
1141 metaslab_class_destroy(spa
->spa_log_class
);
1142 spa
->spa_log_class
= NULL
;
1145 * If this was part of an import or the open otherwise failed, we may
1146 * still have errors left in the queues. Empty them just in case.
1148 spa_errlog_drain(spa
);
1150 avl_destroy(&spa
->spa_errlist_scrub
);
1151 avl_destroy(&spa
->spa_errlist_last
);
1153 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1155 mutex_enter(&spa
->spa_proc_lock
);
1156 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1157 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1158 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1159 cv_broadcast(&spa
->spa_proc_cv
);
1160 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1161 ASSERT(spa
->spa_proc
!= &p0
);
1162 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1164 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1165 spa
->spa_proc_state
= SPA_PROC_NONE
;
1167 ASSERT(spa
->spa_proc
== &p0
);
1168 mutex_exit(&spa
->spa_proc_lock
);
1171 * We want to make sure spa_thread() has actually exited the ZFS
1172 * module, so that the module can't be unloaded out from underneath
1175 if (spa
->spa_did
!= 0) {
1176 thread_join(spa
->spa_did
);
1182 * Verify a pool configuration, and construct the vdev tree appropriately. This
1183 * will create all the necessary vdevs in the appropriate layout, with each vdev
1184 * in the CLOSED state. This will prep the pool before open/creation/import.
1185 * All vdev validation is done by the vdev_alloc() routine.
1188 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1189 uint_t id
, int atype
)
1196 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1199 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1202 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1205 if (error
== ENOENT
)
1211 return (SET_ERROR(EINVAL
));
1214 for (c
= 0; c
< children
; c
++) {
1216 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1224 ASSERT(*vdp
!= NULL
);
1230 * Opposite of spa_load().
1233 spa_unload(spa_t
*spa
)
1237 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1242 spa_async_suspend(spa
);
1247 if (spa
->spa_sync_on
) {
1248 txg_sync_stop(spa
->spa_dsl_pool
);
1249 spa
->spa_sync_on
= B_FALSE
;
1253 * Wait for any outstanding async I/O to complete.
1255 if (spa
->spa_async_zio_root
!= NULL
) {
1256 (void) zio_wait(spa
->spa_async_zio_root
);
1257 spa
->spa_async_zio_root
= NULL
;
1260 bpobj_close(&spa
->spa_deferred_bpobj
);
1262 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1267 if (spa
->spa_root_vdev
)
1268 vdev_free(spa
->spa_root_vdev
);
1269 ASSERT(spa
->spa_root_vdev
== NULL
);
1272 * Close the dsl pool.
1274 if (spa
->spa_dsl_pool
) {
1275 dsl_pool_close(spa
->spa_dsl_pool
);
1276 spa
->spa_dsl_pool
= NULL
;
1277 spa
->spa_meta_objset
= NULL
;
1284 * Drop and purge level 2 cache
1286 spa_l2cache_drop(spa
);
1288 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1289 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1290 if (spa
->spa_spares
.sav_vdevs
) {
1291 kmem_free(spa
->spa_spares
.sav_vdevs
,
1292 spa
->spa_spares
.sav_count
* sizeof (void *));
1293 spa
->spa_spares
.sav_vdevs
= NULL
;
1295 if (spa
->spa_spares
.sav_config
) {
1296 nvlist_free(spa
->spa_spares
.sav_config
);
1297 spa
->spa_spares
.sav_config
= NULL
;
1299 spa
->spa_spares
.sav_count
= 0;
1301 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1302 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1303 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1305 if (spa
->spa_l2cache
.sav_vdevs
) {
1306 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1307 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1308 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1310 if (spa
->spa_l2cache
.sav_config
) {
1311 nvlist_free(spa
->spa_l2cache
.sav_config
);
1312 spa
->spa_l2cache
.sav_config
= NULL
;
1314 spa
->spa_l2cache
.sav_count
= 0;
1316 spa
->spa_async_suspended
= 0;
1318 if (spa
->spa_comment
!= NULL
) {
1319 spa_strfree(spa
->spa_comment
);
1320 spa
->spa_comment
= NULL
;
1323 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1327 * Load (or re-load) the current list of vdevs describing the active spares for
1328 * this pool. When this is called, we have some form of basic information in
1329 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1330 * then re-generate a more complete list including status information.
1333 spa_load_spares(spa_t
*spa
)
1340 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1343 * First, close and free any existing spare vdevs.
1345 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1346 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1348 /* Undo the call to spa_activate() below */
1349 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1350 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1351 spa_spare_remove(tvd
);
1356 if (spa
->spa_spares
.sav_vdevs
)
1357 kmem_free(spa
->spa_spares
.sav_vdevs
,
1358 spa
->spa_spares
.sav_count
* sizeof (void *));
1360 if (spa
->spa_spares
.sav_config
== NULL
)
1363 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1364 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1366 spa
->spa_spares
.sav_count
= (int)nspares
;
1367 spa
->spa_spares
.sav_vdevs
= NULL
;
1373 * Construct the array of vdevs, opening them to get status in the
1374 * process. For each spare, there is potentially two different vdev_t
1375 * structures associated with it: one in the list of spares (used only
1376 * for basic validation purposes) and one in the active vdev
1377 * configuration (if it's spared in). During this phase we open and
1378 * validate each vdev on the spare list. If the vdev also exists in the
1379 * active configuration, then we also mark this vdev as an active spare.
1381 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1383 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1384 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1385 VDEV_ALLOC_SPARE
) == 0);
1388 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1390 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1391 B_FALSE
)) != NULL
) {
1392 if (!tvd
->vdev_isspare
)
1396 * We only mark the spare active if we were successfully
1397 * able to load the vdev. Otherwise, importing a pool
1398 * with a bad active spare would result in strange
1399 * behavior, because multiple pool would think the spare
1400 * is actively in use.
1402 * There is a vulnerability here to an equally bizarre
1403 * circumstance, where a dead active spare is later
1404 * brought back to life (onlined or otherwise). Given
1405 * the rarity of this scenario, and the extra complexity
1406 * it adds, we ignore the possibility.
1408 if (!vdev_is_dead(tvd
))
1409 spa_spare_activate(tvd
);
1413 vd
->vdev_aux
= &spa
->spa_spares
;
1415 if (vdev_open(vd
) != 0)
1418 if (vdev_validate_aux(vd
) == 0)
1423 * Recompute the stashed list of spares, with status information
1426 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1427 DATA_TYPE_NVLIST_ARRAY
) == 0);
1429 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1431 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1432 spares
[i
] = vdev_config_generate(spa
,
1433 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1434 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1435 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1436 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1437 nvlist_free(spares
[i
]);
1438 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1442 * Load (or re-load) the current list of vdevs describing the active l2cache for
1443 * this pool. When this is called, we have some form of basic information in
1444 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1445 * then re-generate a more complete list including status information.
1446 * Devices which are already active have their details maintained, and are
1450 spa_load_l2cache(spa_t
*spa
)
1454 int i
, j
, oldnvdevs
;
1456 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1457 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1459 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1461 if (sav
->sav_config
!= NULL
) {
1462 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1463 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1464 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_PUSHPAGE
);
1470 oldvdevs
= sav
->sav_vdevs
;
1471 oldnvdevs
= sav
->sav_count
;
1472 sav
->sav_vdevs
= NULL
;
1476 * Process new nvlist of vdevs.
1478 for (i
= 0; i
< nl2cache
; i
++) {
1479 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1483 for (j
= 0; j
< oldnvdevs
; j
++) {
1485 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1487 * Retain previous vdev for add/remove ops.
1495 if (newvdevs
[i
] == NULL
) {
1499 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1500 VDEV_ALLOC_L2CACHE
) == 0);
1505 * Commit this vdev as an l2cache device,
1506 * even if it fails to open.
1508 spa_l2cache_add(vd
);
1513 spa_l2cache_activate(vd
);
1515 if (vdev_open(vd
) != 0)
1518 (void) vdev_validate_aux(vd
);
1520 if (!vdev_is_dead(vd
))
1521 l2arc_add_vdev(spa
, vd
);
1526 * Purge vdevs that were dropped
1528 for (i
= 0; i
< oldnvdevs
; i
++) {
1533 ASSERT(vd
->vdev_isl2cache
);
1535 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1536 pool
!= 0ULL && l2arc_vdev_present(vd
))
1537 l2arc_remove_vdev(vd
);
1538 vdev_clear_stats(vd
);
1544 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1546 if (sav
->sav_config
== NULL
)
1549 sav
->sav_vdevs
= newvdevs
;
1550 sav
->sav_count
= (int)nl2cache
;
1553 * Recompute the stashed list of l2cache devices, with status
1554 * information this time.
1556 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1557 DATA_TYPE_NVLIST_ARRAY
) == 0);
1559 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_PUSHPAGE
);
1560 for (i
= 0; i
< sav
->sav_count
; i
++)
1561 l2cache
[i
] = vdev_config_generate(spa
,
1562 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1563 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1564 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1566 for (i
= 0; i
< sav
->sav_count
; i
++)
1567 nvlist_free(l2cache
[i
]);
1569 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1573 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1576 char *packed
= NULL
;
1581 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1585 nvsize
= *(uint64_t *)db
->db_data
;
1586 dmu_buf_rele(db
, FTAG
);
1588 packed
= kmem_alloc(nvsize
, KM_PUSHPAGE
| KM_NODEBUG
);
1589 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1592 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1593 kmem_free(packed
, nvsize
);
1599 * Checks to see if the given vdev could not be opened, in which case we post a
1600 * sysevent to notify the autoreplace code that the device has been removed.
1603 spa_check_removed(vdev_t
*vd
)
1607 for (c
= 0; c
< vd
->vdev_children
; c
++)
1608 spa_check_removed(vd
->vdev_child
[c
]);
1610 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1612 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE
,
1613 vd
->vdev_spa
, vd
, NULL
, 0, 0);
1614 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_CHECK
);
1619 * Validate the current config against the MOS config
1622 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1624 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1628 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1630 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1631 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1633 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1636 * If we're doing a normal import, then build up any additional
1637 * diagnostic information about missing devices in this config.
1638 * We'll pass this up to the user for further processing.
1640 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1641 nvlist_t
**child
, *nv
;
1644 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1646 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
1648 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1649 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1650 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1652 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1653 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1655 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1660 VERIFY(nvlist_add_nvlist_array(nv
,
1661 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1662 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1663 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1665 for (i
= 0; i
< idx
; i
++)
1666 nvlist_free(child
[i
]);
1669 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1673 * Compare the root vdev tree with the information we have
1674 * from the MOS config (mrvd). Check each top-level vdev
1675 * with the corresponding MOS config top-level (mtvd).
1677 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1678 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1679 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1682 * Resolve any "missing" vdevs in the current configuration.
1683 * If we find that the MOS config has more accurate information
1684 * about the top-level vdev then use that vdev instead.
1686 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1687 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1689 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1693 * Device specific actions.
1695 if (mtvd
->vdev_islog
) {
1696 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1699 * XXX - once we have 'readonly' pool
1700 * support we should be able to handle
1701 * missing data devices by transitioning
1702 * the pool to readonly.
1708 * Swap the missing vdev with the data we were
1709 * able to obtain from the MOS config.
1711 vdev_remove_child(rvd
, tvd
);
1712 vdev_remove_child(mrvd
, mtvd
);
1714 vdev_add_child(rvd
, mtvd
);
1715 vdev_add_child(mrvd
, tvd
);
1717 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1719 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1722 } else if (mtvd
->vdev_islog
) {
1724 * Load the slog device's state from the MOS config
1725 * since it's possible that the label does not
1726 * contain the most up-to-date information.
1728 vdev_load_log_state(tvd
, mtvd
);
1733 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1736 * Ensure we were able to validate the config.
1738 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1742 * Check for missing log devices
1745 spa_check_logs(spa_t
*spa
)
1747 boolean_t rv
= B_FALSE
;
1749 switch (spa
->spa_log_state
) {
1752 case SPA_LOG_MISSING
:
1753 /* need to recheck in case slog has been restored */
1754 case SPA_LOG_UNKNOWN
:
1755 rv
= (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
,
1756 NULL
, DS_FIND_CHILDREN
) != 0);
1758 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1765 spa_passivate_log(spa_t
*spa
)
1767 vdev_t
*rvd
= spa
->spa_root_vdev
;
1768 boolean_t slog_found
= B_FALSE
;
1771 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1773 if (!spa_has_slogs(spa
))
1776 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1777 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1778 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1780 if (tvd
->vdev_islog
) {
1781 metaslab_group_passivate(mg
);
1782 slog_found
= B_TRUE
;
1786 return (slog_found
);
1790 spa_activate_log(spa_t
*spa
)
1792 vdev_t
*rvd
= spa
->spa_root_vdev
;
1795 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1797 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1798 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1799 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1801 if (tvd
->vdev_islog
)
1802 metaslab_group_activate(mg
);
1807 spa_offline_log(spa_t
*spa
)
1811 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1812 NULL
, DS_FIND_CHILDREN
);
1815 * We successfully offlined the log device, sync out the
1816 * current txg so that the "stubby" block can be removed
1819 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1825 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1829 for (i
= 0; i
< sav
->sav_count
; i
++)
1830 spa_check_removed(sav
->sav_vdevs
[i
]);
1834 spa_claim_notify(zio_t
*zio
)
1836 spa_t
*spa
= zio
->io_spa
;
1841 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1842 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1843 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1844 mutex_exit(&spa
->spa_props_lock
);
1847 typedef struct spa_load_error
{
1848 uint64_t sle_meta_count
;
1849 uint64_t sle_data_count
;
1853 spa_load_verify_done(zio_t
*zio
)
1855 blkptr_t
*bp
= zio
->io_bp
;
1856 spa_load_error_t
*sle
= zio
->io_private
;
1857 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1858 int error
= zio
->io_error
;
1861 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1862 type
!= DMU_OT_INTENT_LOG
)
1863 atomic_add_64(&sle
->sle_meta_count
, 1);
1865 atomic_add_64(&sle
->sle_data_count
, 1);
1867 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1872 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1873 const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1875 if (!BP_IS_HOLE(bp
)) {
1877 size_t size
= BP_GET_PSIZE(bp
);
1878 void *data
= zio_data_buf_alloc(size
);
1880 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1881 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1882 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1883 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1889 spa_load_verify(spa_t
*spa
)
1892 spa_load_error_t sle
= { 0 };
1893 zpool_rewind_policy_t policy
;
1894 boolean_t verify_ok
= B_FALSE
;
1897 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1899 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1902 rio
= zio_root(spa
, NULL
, &sle
,
1903 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1905 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1906 TRAVERSE_PRE
| TRAVERSE_PREFETCH
, spa_load_verify_cb
, rio
);
1908 (void) zio_wait(rio
);
1910 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1911 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1913 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1914 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1918 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1919 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1921 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1922 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1923 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1924 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1925 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1926 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1927 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1929 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1933 if (error
!= ENXIO
&& error
!= EIO
)
1934 error
= SET_ERROR(EIO
);
1938 return (verify_ok
? 0 : EIO
);
1942 * Find a value in the pool props object.
1945 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
1947 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
1948 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
1952 * Find a value in the pool directory object.
1955 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
1957 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1958 name
, sizeof (uint64_t), 1, val
));
1962 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
1964 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
1969 * Fix up config after a partly-completed split. This is done with the
1970 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1971 * pool have that entry in their config, but only the splitting one contains
1972 * a list of all the guids of the vdevs that are being split off.
1974 * This function determines what to do with that list: either rejoin
1975 * all the disks to the pool, or complete the splitting process. To attempt
1976 * the rejoin, each disk that is offlined is marked online again, and
1977 * we do a reopen() call. If the vdev label for every disk that was
1978 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1979 * then we call vdev_split() on each disk, and complete the split.
1981 * Otherwise we leave the config alone, with all the vdevs in place in
1982 * the original pool.
1985 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
1992 boolean_t attempt_reopen
;
1994 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
1997 /* check that the config is complete */
1998 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
1999 &glist
, &gcount
) != 0)
2002 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_PUSHPAGE
);
2004 /* attempt to online all the vdevs & validate */
2005 attempt_reopen
= B_TRUE
;
2006 for (i
= 0; i
< gcount
; i
++) {
2007 if (glist
[i
] == 0) /* vdev is hole */
2010 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2011 if (vd
[i
] == NULL
) {
2013 * Don't bother attempting to reopen the disks;
2014 * just do the split.
2016 attempt_reopen
= B_FALSE
;
2018 /* attempt to re-online it */
2019 vd
[i
]->vdev_offline
= B_FALSE
;
2023 if (attempt_reopen
) {
2024 vdev_reopen(spa
->spa_root_vdev
);
2026 /* check each device to see what state it's in */
2027 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2028 if (vd
[i
] != NULL
&&
2029 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2036 * If every disk has been moved to the new pool, or if we never
2037 * even attempted to look at them, then we split them off for
2040 if (!attempt_reopen
|| gcount
== extracted
) {
2041 for (i
= 0; i
< gcount
; i
++)
2044 vdev_reopen(spa
->spa_root_vdev
);
2047 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2051 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2052 boolean_t mosconfig
)
2054 nvlist_t
*config
= spa
->spa_config
;
2055 char *ereport
= FM_EREPORT_ZFS_POOL
;
2061 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2062 return (SET_ERROR(EINVAL
));
2064 ASSERT(spa
->spa_comment
== NULL
);
2065 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2066 spa
->spa_comment
= spa_strdup(comment
);
2069 * Versioning wasn't explicitly added to the label until later, so if
2070 * it's not present treat it as the initial version.
2072 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2073 &spa
->spa_ubsync
.ub_version
) != 0)
2074 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2076 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2077 &spa
->spa_config_txg
);
2079 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2080 spa_guid_exists(pool_guid
, 0)) {
2081 error
= SET_ERROR(EEXIST
);
2083 spa
->spa_config_guid
= pool_guid
;
2085 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2087 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2091 nvlist_free(spa
->spa_load_info
);
2092 spa
->spa_load_info
= fnvlist_alloc();
2094 gethrestime(&spa
->spa_loaded_ts
);
2095 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2096 mosconfig
, &ereport
);
2099 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2101 if (error
!= EEXIST
) {
2102 spa
->spa_loaded_ts
.tv_sec
= 0;
2103 spa
->spa_loaded_ts
.tv_nsec
= 0;
2105 if (error
!= EBADF
) {
2106 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2109 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2116 * Load an existing storage pool, using the pool's builtin spa_config as a
2117 * source of configuration information.
2119 __attribute__((always_inline
))
2121 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2122 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2126 nvlist_t
*nvroot
= NULL
;
2129 uberblock_t
*ub
= &spa
->spa_uberblock
;
2130 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2131 int orig_mode
= spa
->spa_mode
;
2134 boolean_t missing_feat_write
= B_FALSE
;
2137 * If this is an untrusted config, access the pool in read-only mode.
2138 * This prevents things like resilvering recently removed devices.
2141 spa
->spa_mode
= FREAD
;
2143 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2145 spa
->spa_load_state
= state
;
2147 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2148 return (SET_ERROR(EINVAL
));
2150 parse
= (type
== SPA_IMPORT_EXISTING
?
2151 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2154 * Create "The Godfather" zio to hold all async IOs
2156 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
2157 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
2160 * Parse the configuration into a vdev tree. We explicitly set the
2161 * value that will be returned by spa_version() since parsing the
2162 * configuration requires knowing the version number.
2164 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2165 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2166 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2171 ASSERT(spa
->spa_root_vdev
== rvd
);
2173 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2174 ASSERT(spa_guid(spa
) == pool_guid
);
2178 * Try to open all vdevs, loading each label in the process.
2180 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2181 error
= vdev_open(rvd
);
2182 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2187 * We need to validate the vdev labels against the configuration that
2188 * we have in hand, which is dependent on the setting of mosconfig. If
2189 * mosconfig is true then we're validating the vdev labels based on
2190 * that config. Otherwise, we're validating against the cached config
2191 * (zpool.cache) that was read when we loaded the zfs module, and then
2192 * later we will recursively call spa_load() and validate against
2195 * If we're assembling a new pool that's been split off from an
2196 * existing pool, the labels haven't yet been updated so we skip
2197 * validation for now.
2199 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2200 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2201 error
= vdev_validate(rvd
, mosconfig
);
2202 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2207 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2208 return (SET_ERROR(ENXIO
));
2212 * Find the best uberblock.
2214 vdev_uberblock_load(rvd
, ub
, &label
);
2217 * If we weren't able to find a single valid uberblock, return failure.
2219 if (ub
->ub_txg
== 0) {
2221 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2225 * If the pool has an unsupported version we can't open it.
2227 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2229 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2232 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2236 * If we weren't able to find what's necessary for reading the
2237 * MOS in the label, return failure.
2239 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2240 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2242 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2247 * Update our in-core representation with the definitive values
2250 nvlist_free(spa
->spa_label_features
);
2251 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2257 * Look through entries in the label nvlist's features_for_read. If
2258 * there is a feature listed there which we don't understand then we
2259 * cannot open a pool.
2261 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2262 nvlist_t
*unsup_feat
;
2265 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2268 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2270 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2271 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2272 VERIFY(nvlist_add_string(unsup_feat
,
2273 nvpair_name(nvp
), "") == 0);
2277 if (!nvlist_empty(unsup_feat
)) {
2278 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2279 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2280 nvlist_free(unsup_feat
);
2281 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2285 nvlist_free(unsup_feat
);
2289 * If the vdev guid sum doesn't match the uberblock, we have an
2290 * incomplete configuration. We first check to see if the pool
2291 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2292 * If it is, defer the vdev_guid_sum check till later so we
2293 * can handle missing vdevs.
2295 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2296 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2297 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2298 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2300 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2301 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2302 spa_try_repair(spa
, config
);
2303 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2304 nvlist_free(spa
->spa_config_splitting
);
2305 spa
->spa_config_splitting
= NULL
;
2309 * Initialize internal SPA structures.
2311 spa
->spa_state
= POOL_STATE_ACTIVE
;
2312 spa
->spa_ubsync
= spa
->spa_uberblock
;
2313 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2314 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2315 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2316 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2317 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2318 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2320 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2322 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2323 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2325 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2326 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2328 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2329 boolean_t missing_feat_read
= B_FALSE
;
2330 nvlist_t
*unsup_feat
, *enabled_feat
;
2333 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2334 &spa
->spa_feat_for_read_obj
) != 0) {
2335 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2338 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2339 &spa
->spa_feat_for_write_obj
) != 0) {
2340 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2343 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2344 &spa
->spa_feat_desc_obj
) != 0) {
2345 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2348 enabled_feat
= fnvlist_alloc();
2349 unsup_feat
= fnvlist_alloc();
2351 if (!spa_features_check(spa
, B_FALSE
,
2352 unsup_feat
, enabled_feat
))
2353 missing_feat_read
= B_TRUE
;
2355 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2356 if (!spa_features_check(spa
, B_TRUE
,
2357 unsup_feat
, enabled_feat
)) {
2358 missing_feat_write
= B_TRUE
;
2362 fnvlist_add_nvlist(spa
->spa_load_info
,
2363 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2365 if (!nvlist_empty(unsup_feat
)) {
2366 fnvlist_add_nvlist(spa
->spa_load_info
,
2367 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2370 fnvlist_free(enabled_feat
);
2371 fnvlist_free(unsup_feat
);
2373 if (!missing_feat_read
) {
2374 fnvlist_add_boolean(spa
->spa_load_info
,
2375 ZPOOL_CONFIG_CAN_RDONLY
);
2379 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2380 * twofold: to determine whether the pool is available for
2381 * import in read-write mode and (if it is not) whether the
2382 * pool is available for import in read-only mode. If the pool
2383 * is available for import in read-write mode, it is displayed
2384 * as available in userland; if it is not available for import
2385 * in read-only mode, it is displayed as unavailable in
2386 * userland. If the pool is available for import in read-only
2387 * mode but not read-write mode, it is displayed as unavailable
2388 * in userland with a special note that the pool is actually
2389 * available for open in read-only mode.
2391 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2392 * missing a feature for write, we must first determine whether
2393 * the pool can be opened read-only before returning to
2394 * userland in order to know whether to display the
2395 * abovementioned note.
2397 if (missing_feat_read
|| (missing_feat_write
&&
2398 spa_writeable(spa
))) {
2399 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2404 * Load refcounts for ZFS features from disk into an in-memory
2405 * cache during SPA initialization.
2407 for (i
= 0; i
< SPA_FEATURES
; i
++) {
2410 error
= feature_get_refcount_from_disk(spa
,
2411 &spa_feature_table
[i
], &refcount
);
2413 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2414 } else if (error
== ENOTSUP
) {
2415 spa
->spa_feat_refcount_cache
[i
] =
2416 SPA_FEATURE_DISABLED
;
2418 return (spa_vdev_err(rvd
,
2419 VDEV_AUX_CORRUPT_DATA
, EIO
));
2424 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2425 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2426 &spa
->spa_feat_enabled_txg_obj
) != 0) {
2427 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2431 spa
->spa_is_initializing
= B_TRUE
;
2432 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2433 spa
->spa_is_initializing
= B_FALSE
;
2435 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2439 nvlist_t
*policy
= NULL
, *nvconfig
;
2441 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2442 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2444 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2445 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2447 unsigned long myhostid
= 0;
2449 VERIFY(nvlist_lookup_string(nvconfig
,
2450 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2453 myhostid
= zone_get_hostid(NULL
);
2456 * We're emulating the system's hostid in userland, so
2457 * we can't use zone_get_hostid().
2459 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2460 #endif /* _KERNEL */
2461 if (hostid
!= 0 && myhostid
!= 0 &&
2462 hostid
!= myhostid
) {
2463 nvlist_free(nvconfig
);
2464 cmn_err(CE_WARN
, "pool '%s' could not be "
2465 "loaded as it was last accessed by another "
2466 "system (host: %s hostid: 0x%lx). See: "
2467 "http://zfsonlinux.org/msg/ZFS-8000-EY",
2468 spa_name(spa
), hostname
,
2469 (unsigned long)hostid
);
2470 return (SET_ERROR(EBADF
));
2473 if (nvlist_lookup_nvlist(spa
->spa_config
,
2474 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2475 VERIFY(nvlist_add_nvlist(nvconfig
,
2476 ZPOOL_REWIND_POLICY
, policy
) == 0);
2478 spa_config_set(spa
, nvconfig
);
2480 spa_deactivate(spa
);
2481 spa_activate(spa
, orig_mode
);
2483 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2486 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2487 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2488 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2490 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2493 * Load the bit that tells us to use the new accounting function
2494 * (raid-z deflation). If we have an older pool, this will not
2497 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2498 if (error
!= 0 && error
!= ENOENT
)
2499 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2501 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2502 &spa
->spa_creation_version
);
2503 if (error
!= 0 && error
!= ENOENT
)
2504 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2507 * Load the persistent error log. If we have an older pool, this will
2510 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2511 if (error
!= 0 && error
!= ENOENT
)
2512 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2514 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2515 &spa
->spa_errlog_scrub
);
2516 if (error
!= 0 && error
!= ENOENT
)
2517 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2520 * Load the history object. If we have an older pool, this
2521 * will not be present.
2523 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2524 if (error
!= 0 && error
!= ENOENT
)
2525 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2528 * If we're assembling the pool from the split-off vdevs of
2529 * an existing pool, we don't want to attach the spares & cache
2534 * Load any hot spares for this pool.
2536 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2537 if (error
!= 0 && error
!= ENOENT
)
2538 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2539 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2540 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2541 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2542 &spa
->spa_spares
.sav_config
) != 0)
2543 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2545 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2546 spa_load_spares(spa
);
2547 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2548 } else if (error
== 0) {
2549 spa
->spa_spares
.sav_sync
= B_TRUE
;
2553 * Load any level 2 ARC devices for this pool.
2555 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2556 &spa
->spa_l2cache
.sav_object
);
2557 if (error
!= 0 && error
!= ENOENT
)
2558 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2559 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2560 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2561 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2562 &spa
->spa_l2cache
.sav_config
) != 0)
2563 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2565 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2566 spa_load_l2cache(spa
);
2567 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2568 } else if (error
== 0) {
2569 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2572 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2574 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2575 if (error
&& error
!= ENOENT
)
2576 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2579 uint64_t autoreplace
= 0;
2581 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2582 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2583 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2584 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2585 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2586 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2587 &spa
->spa_dedup_ditto
);
2589 spa
->spa_autoreplace
= (autoreplace
!= 0);
2593 * If the 'autoreplace' property is set, then post a resource notifying
2594 * the ZFS DE that it should not issue any faults for unopenable
2595 * devices. We also iterate over the vdevs, and post a sysevent for any
2596 * unopenable vdevs so that the normal autoreplace handler can take
2599 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2600 spa_check_removed(spa
->spa_root_vdev
);
2602 * For the import case, this is done in spa_import(), because
2603 * at this point we're using the spare definitions from
2604 * the MOS config, not necessarily from the userland config.
2606 if (state
!= SPA_LOAD_IMPORT
) {
2607 spa_aux_check_removed(&spa
->spa_spares
);
2608 spa_aux_check_removed(&spa
->spa_l2cache
);
2613 * Load the vdev state for all toplevel vdevs.
2618 * Propagate the leaf DTLs we just loaded all the way up the tree.
2620 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2621 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2622 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2625 * Load the DDTs (dedup tables).
2627 error
= ddt_load(spa
);
2629 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2631 spa_update_dspace(spa
);
2634 * Validate the config, using the MOS config to fill in any
2635 * information which might be missing. If we fail to validate
2636 * the config then declare the pool unfit for use. If we're
2637 * assembling a pool from a split, the log is not transferred
2640 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2643 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2644 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2646 if (!spa_config_valid(spa
, nvconfig
)) {
2647 nvlist_free(nvconfig
);
2648 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2651 nvlist_free(nvconfig
);
2654 * Now that we've validated the config, check the state of the
2655 * root vdev. If it can't be opened, it indicates one or
2656 * more toplevel vdevs are faulted.
2658 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2659 return (SET_ERROR(ENXIO
));
2661 if (spa_check_logs(spa
)) {
2662 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2663 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2667 if (missing_feat_write
) {
2668 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2671 * At this point, we know that we can open the pool in
2672 * read-only mode but not read-write mode. We now have enough
2673 * information and can return to userland.
2675 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2679 * We've successfully opened the pool, verify that we're ready
2680 * to start pushing transactions.
2682 if (state
!= SPA_LOAD_TRYIMPORT
) {
2683 if ((error
= spa_load_verify(spa
)))
2684 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2688 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2689 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2691 int need_update
= B_FALSE
;
2694 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2697 * Claim log blocks that haven't been committed yet.
2698 * This must all happen in a single txg.
2699 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2700 * invoked from zil_claim_log_block()'s i/o done callback.
2701 * Price of rollback is that we abandon the log.
2703 spa
->spa_claiming
= B_TRUE
;
2705 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
2706 spa_first_txg(spa
));
2707 (void) dmu_objset_find(spa_name(spa
),
2708 zil_claim
, tx
, DS_FIND_CHILDREN
);
2711 spa
->spa_claiming
= B_FALSE
;
2713 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2714 spa
->spa_sync_on
= B_TRUE
;
2715 txg_sync_start(spa
->spa_dsl_pool
);
2718 * Wait for all claims to sync. We sync up to the highest
2719 * claimed log block birth time so that claimed log blocks
2720 * don't appear to be from the future. spa_claim_max_txg
2721 * will have been set for us by either zil_check_log_chain()
2722 * (invoked from spa_check_logs()) or zil_claim() above.
2724 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2727 * If the config cache is stale, or we have uninitialized
2728 * metaslabs (see spa_vdev_add()), then update the config.
2730 * If this is a verbatim import, trust the current
2731 * in-core spa_config and update the disk labels.
2733 if (config_cache_txg
!= spa
->spa_config_txg
||
2734 state
== SPA_LOAD_IMPORT
||
2735 state
== SPA_LOAD_RECOVER
||
2736 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2737 need_update
= B_TRUE
;
2739 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2740 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2741 need_update
= B_TRUE
;
2744 * Update the config cache asychronously in case we're the
2745 * root pool, in which case the config cache isn't writable yet.
2748 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2751 * Check all DTLs to see if anything needs resilvering.
2753 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2754 vdev_resilver_needed(rvd
, NULL
, NULL
))
2755 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2758 * Log the fact that we booted up (so that we can detect if
2759 * we rebooted in the middle of an operation).
2761 spa_history_log_version(spa
, "open");
2764 * Delete any inconsistent datasets.
2766 (void) dmu_objset_find(spa_name(spa
),
2767 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2770 * Clean up any stale temporary dataset userrefs.
2772 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2779 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2781 int mode
= spa
->spa_mode
;
2784 spa_deactivate(spa
);
2786 spa
->spa_load_max_txg
--;
2788 spa_activate(spa
, mode
);
2789 spa_async_suspend(spa
);
2791 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2795 * If spa_load() fails this function will try loading prior txg's. If
2796 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2797 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2798 * function will not rewind the pool and will return the same error as
2802 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2803 uint64_t max_request
, int rewind_flags
)
2805 nvlist_t
*loadinfo
= NULL
;
2806 nvlist_t
*config
= NULL
;
2807 int load_error
, rewind_error
;
2808 uint64_t safe_rewind_txg
;
2811 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2812 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2813 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2815 spa
->spa_load_max_txg
= max_request
;
2818 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2820 if (load_error
== 0)
2823 if (spa
->spa_root_vdev
!= NULL
)
2824 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2826 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2827 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2829 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2830 nvlist_free(config
);
2831 return (load_error
);
2834 if (state
== SPA_LOAD_RECOVER
) {
2835 /* Price of rolling back is discarding txgs, including log */
2836 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2839 * If we aren't rolling back save the load info from our first
2840 * import attempt so that we can restore it after attempting
2843 loadinfo
= spa
->spa_load_info
;
2844 spa
->spa_load_info
= fnvlist_alloc();
2847 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2848 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2849 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2850 TXG_INITIAL
: safe_rewind_txg
;
2853 * Continue as long as we're finding errors, we're still within
2854 * the acceptable rewind range, and we're still finding uberblocks
2856 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2857 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2858 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2859 spa
->spa_extreme_rewind
= B_TRUE
;
2860 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2863 spa
->spa_extreme_rewind
= B_FALSE
;
2864 spa
->spa_load_max_txg
= UINT64_MAX
;
2866 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2867 spa_config_set(spa
, config
);
2869 if (state
== SPA_LOAD_RECOVER
) {
2870 ASSERT3P(loadinfo
, ==, NULL
);
2871 return (rewind_error
);
2873 /* Store the rewind info as part of the initial load info */
2874 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
2875 spa
->spa_load_info
);
2877 /* Restore the initial load info */
2878 fnvlist_free(spa
->spa_load_info
);
2879 spa
->spa_load_info
= loadinfo
;
2881 return (load_error
);
2888 * The import case is identical to an open except that the configuration is sent
2889 * down from userland, instead of grabbed from the configuration cache. For the
2890 * case of an open, the pool configuration will exist in the
2891 * POOL_STATE_UNINITIALIZED state.
2893 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2894 * the same time open the pool, without having to keep around the spa_t in some
2898 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2902 spa_load_state_t state
= SPA_LOAD_OPEN
;
2904 int locked
= B_FALSE
;
2905 int firstopen
= B_FALSE
;
2910 * As disgusting as this is, we need to support recursive calls to this
2911 * function because dsl_dir_open() is called during spa_load(), and ends
2912 * up calling spa_open() again. The real fix is to figure out how to
2913 * avoid dsl_dir_open() calling this in the first place.
2915 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2916 mutex_enter(&spa_namespace_lock
);
2920 if ((spa
= spa_lookup(pool
)) == NULL
) {
2922 mutex_exit(&spa_namespace_lock
);
2923 return (SET_ERROR(ENOENT
));
2926 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
2927 zpool_rewind_policy_t policy
;
2931 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
2933 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
2934 state
= SPA_LOAD_RECOVER
;
2936 spa_activate(spa
, spa_mode_global
);
2938 if (state
!= SPA_LOAD_RECOVER
)
2939 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
2941 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
2942 policy
.zrp_request
);
2944 if (error
== EBADF
) {
2946 * If vdev_validate() returns failure (indicated by
2947 * EBADF), it indicates that one of the vdevs indicates
2948 * that the pool has been exported or destroyed. If
2949 * this is the case, the config cache is out of sync and
2950 * we should remove the pool from the namespace.
2953 spa_deactivate(spa
);
2954 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
2957 mutex_exit(&spa_namespace_lock
);
2958 return (SET_ERROR(ENOENT
));
2963 * We can't open the pool, but we still have useful
2964 * information: the state of each vdev after the
2965 * attempted vdev_open(). Return this to the user.
2967 if (config
!= NULL
&& spa
->spa_config
) {
2968 VERIFY(nvlist_dup(spa
->spa_config
, config
,
2970 VERIFY(nvlist_add_nvlist(*config
,
2971 ZPOOL_CONFIG_LOAD_INFO
,
2972 spa
->spa_load_info
) == 0);
2975 spa_deactivate(spa
);
2976 spa
->spa_last_open_failed
= error
;
2978 mutex_exit(&spa_namespace_lock
);
2984 spa_open_ref(spa
, tag
);
2987 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2990 * If we've recovered the pool, pass back any information we
2991 * gathered while doing the load.
2993 if (state
== SPA_LOAD_RECOVER
) {
2994 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
2995 spa
->spa_load_info
) == 0);
2999 spa
->spa_last_open_failed
= 0;
3000 spa
->spa_last_ubsync_txg
= 0;
3001 spa
->spa_load_txg
= 0;
3002 mutex_exit(&spa_namespace_lock
);
3007 zvol_create_minors(spa
->spa_name
);
3016 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3019 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3023 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3025 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3029 * Lookup the given spa_t, incrementing the inject count in the process,
3030 * preventing it from being exported or destroyed.
3033 spa_inject_addref(char *name
)
3037 mutex_enter(&spa_namespace_lock
);
3038 if ((spa
= spa_lookup(name
)) == NULL
) {
3039 mutex_exit(&spa_namespace_lock
);
3042 spa
->spa_inject_ref
++;
3043 mutex_exit(&spa_namespace_lock
);
3049 spa_inject_delref(spa_t
*spa
)
3051 mutex_enter(&spa_namespace_lock
);
3052 spa
->spa_inject_ref
--;
3053 mutex_exit(&spa_namespace_lock
);
3057 * Add spares device information to the nvlist.
3060 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3070 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3072 if (spa
->spa_spares
.sav_count
== 0)
3075 VERIFY(nvlist_lookup_nvlist(config
,
3076 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3077 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3078 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3080 VERIFY(nvlist_add_nvlist_array(nvroot
,
3081 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3082 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3083 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3086 * Go through and find any spares which have since been
3087 * repurposed as an active spare. If this is the case, update
3088 * their status appropriately.
3090 for (i
= 0; i
< nspares
; i
++) {
3091 VERIFY(nvlist_lookup_uint64(spares
[i
],
3092 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3093 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3095 VERIFY(nvlist_lookup_uint64_array(
3096 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3097 (uint64_t **)&vs
, &vsc
) == 0);
3098 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3099 vs
->vs_aux
= VDEV_AUX_SPARED
;
3106 * Add l2cache device information to the nvlist, including vdev stats.
3109 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3112 uint_t i
, j
, nl2cache
;
3119 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3121 if (spa
->spa_l2cache
.sav_count
== 0)
3124 VERIFY(nvlist_lookup_nvlist(config
,
3125 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3126 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3127 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3128 if (nl2cache
!= 0) {
3129 VERIFY(nvlist_add_nvlist_array(nvroot
,
3130 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3131 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3132 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3135 * Update level 2 cache device stats.
3138 for (i
= 0; i
< nl2cache
; i
++) {
3139 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3140 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3143 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3145 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3146 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3152 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3153 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3155 vdev_get_stats(vd
, vs
);
3161 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3167 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3168 VERIFY(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3170 if (spa
->spa_feat_for_read_obj
!= 0) {
3171 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3172 spa
->spa_feat_for_read_obj
);
3173 zap_cursor_retrieve(&zc
, &za
) == 0;
3174 zap_cursor_advance(&zc
)) {
3175 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3176 za
.za_num_integers
== 1);
3177 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
3178 za
.za_first_integer
));
3180 zap_cursor_fini(&zc
);
3183 if (spa
->spa_feat_for_write_obj
!= 0) {
3184 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3185 spa
->spa_feat_for_write_obj
);
3186 zap_cursor_retrieve(&zc
, &za
) == 0;
3187 zap_cursor_advance(&zc
)) {
3188 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3189 za
.za_num_integers
== 1);
3190 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
3191 za
.za_first_integer
));
3193 zap_cursor_fini(&zc
);
3196 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3198 nvlist_free(features
);
3202 spa_get_stats(const char *name
, nvlist_t
**config
,
3203 char *altroot
, size_t buflen
)
3209 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3213 * This still leaves a window of inconsistency where the spares
3214 * or l2cache devices could change and the config would be
3215 * self-inconsistent.
3217 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3219 if (*config
!= NULL
) {
3220 uint64_t loadtimes
[2];
3222 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3223 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3224 VERIFY(nvlist_add_uint64_array(*config
,
3225 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3227 VERIFY(nvlist_add_uint64(*config
,
3228 ZPOOL_CONFIG_ERRCOUNT
,
3229 spa_get_errlog_size(spa
)) == 0);
3231 if (spa_suspended(spa
))
3232 VERIFY(nvlist_add_uint64(*config
,
3233 ZPOOL_CONFIG_SUSPENDED
,
3234 spa
->spa_failmode
) == 0);
3236 spa_add_spares(spa
, *config
);
3237 spa_add_l2cache(spa
, *config
);
3238 spa_add_feature_stats(spa
, *config
);
3243 * We want to get the alternate root even for faulted pools, so we cheat
3244 * and call spa_lookup() directly.
3248 mutex_enter(&spa_namespace_lock
);
3249 spa
= spa_lookup(name
);
3251 spa_altroot(spa
, altroot
, buflen
);
3255 mutex_exit(&spa_namespace_lock
);
3257 spa_altroot(spa
, altroot
, buflen
);
3262 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3263 spa_close(spa
, FTAG
);
3270 * Validate that the auxiliary device array is well formed. We must have an
3271 * array of nvlists, each which describes a valid leaf vdev. If this is an
3272 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3273 * specified, as long as they are well-formed.
3276 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3277 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3278 vdev_labeltype_t label
)
3285 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3288 * It's acceptable to have no devs specified.
3290 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3294 return (SET_ERROR(EINVAL
));
3297 * Make sure the pool is formatted with a version that supports this
3300 if (spa_version(spa
) < version
)
3301 return (SET_ERROR(ENOTSUP
));
3304 * Set the pending device list so we correctly handle device in-use
3307 sav
->sav_pending
= dev
;
3308 sav
->sav_npending
= ndev
;
3310 for (i
= 0; i
< ndev
; i
++) {
3311 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3315 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3317 error
= SET_ERROR(EINVAL
);
3322 * The L2ARC currently only supports disk devices in
3323 * kernel context. For user-level testing, we allow it.
3326 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3327 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3328 error
= SET_ERROR(ENOTBLK
);
3335 if ((error
= vdev_open(vd
)) == 0 &&
3336 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3337 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3338 vd
->vdev_guid
) == 0);
3344 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3351 sav
->sav_pending
= NULL
;
3352 sav
->sav_npending
= 0;
3357 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3361 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3363 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3364 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3365 VDEV_LABEL_SPARE
)) != 0) {
3369 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3370 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3371 VDEV_LABEL_L2CACHE
));
3375 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3380 if (sav
->sav_config
!= NULL
) {
3386 * Generate new dev list by concatentating with the
3389 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3390 &olddevs
, &oldndevs
) == 0);
3392 newdevs
= kmem_alloc(sizeof (void *) *
3393 (ndevs
+ oldndevs
), KM_PUSHPAGE
);
3394 for (i
= 0; i
< oldndevs
; i
++)
3395 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3397 for (i
= 0; i
< ndevs
; i
++)
3398 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3401 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3402 DATA_TYPE_NVLIST_ARRAY
) == 0);
3404 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3405 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3406 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3407 nvlist_free(newdevs
[i
]);
3408 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3411 * Generate a new dev list.
3413 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3415 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3421 * Stop and drop level 2 ARC devices
3424 spa_l2cache_drop(spa_t
*spa
)
3428 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3430 for (i
= 0; i
< sav
->sav_count
; i
++) {
3433 vd
= sav
->sav_vdevs
[i
];
3436 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3437 pool
!= 0ULL && l2arc_vdev_present(vd
))
3438 l2arc_remove_vdev(vd
);
3446 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3450 char *altroot
= NULL
;
3455 uint64_t txg
= TXG_INITIAL
;
3456 nvlist_t
**spares
, **l2cache
;
3457 uint_t nspares
, nl2cache
;
3458 uint64_t version
, obj
;
3459 boolean_t has_features
;
3464 * If this pool already exists, return failure.
3466 mutex_enter(&spa_namespace_lock
);
3467 if (spa_lookup(pool
) != NULL
) {
3468 mutex_exit(&spa_namespace_lock
);
3469 return (SET_ERROR(EEXIST
));
3473 * Allocate a new spa_t structure.
3475 (void) nvlist_lookup_string(props
,
3476 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3477 spa
= spa_add(pool
, NULL
, altroot
);
3478 spa_activate(spa
, spa_mode_global
);
3480 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3481 spa_deactivate(spa
);
3483 mutex_exit(&spa_namespace_lock
);
3487 has_features
= B_FALSE
;
3488 for (elem
= nvlist_next_nvpair(props
, NULL
);
3489 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3490 if (zpool_prop_feature(nvpair_name(elem
)))
3491 has_features
= B_TRUE
;
3494 if (has_features
|| nvlist_lookup_uint64(props
,
3495 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3496 version
= SPA_VERSION
;
3498 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3500 spa
->spa_first_txg
= txg
;
3501 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3502 spa
->spa_uberblock
.ub_version
= version
;
3503 spa
->spa_ubsync
= spa
->spa_uberblock
;
3506 * Create "The Godfather" zio to hold all async IOs
3508 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
3509 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
3512 * Create the root vdev.
3514 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3516 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3518 ASSERT(error
!= 0 || rvd
!= NULL
);
3519 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3521 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3522 error
= SET_ERROR(EINVAL
);
3525 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3526 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3527 VDEV_ALLOC_ADD
)) == 0) {
3528 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
3529 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3530 vdev_expand(rvd
->vdev_child
[c
], txg
);
3534 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3538 spa_deactivate(spa
);
3540 mutex_exit(&spa_namespace_lock
);
3545 * Get the list of spares, if specified.
3547 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3548 &spares
, &nspares
) == 0) {
3549 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3551 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3552 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3553 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3554 spa_load_spares(spa
);
3555 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3556 spa
->spa_spares
.sav_sync
= B_TRUE
;
3560 * Get the list of level 2 cache devices, if specified.
3562 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3563 &l2cache
, &nl2cache
) == 0) {
3564 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3565 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3566 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3567 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3568 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3569 spa_load_l2cache(spa
);
3570 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3571 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3574 spa
->spa_is_initializing
= B_TRUE
;
3575 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3576 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3577 spa
->spa_is_initializing
= B_FALSE
;
3580 * Create DDTs (dedup tables).
3584 spa_update_dspace(spa
);
3586 tx
= dmu_tx_create_assigned(dp
, txg
);
3589 * Create the pool config object.
3591 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3592 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3593 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3595 if (zap_add(spa
->spa_meta_objset
,
3596 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3597 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3598 cmn_err(CE_PANIC
, "failed to add pool config");
3601 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3602 spa_feature_create_zap_objects(spa
, tx
);
3604 if (zap_add(spa
->spa_meta_objset
,
3605 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3606 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3607 cmn_err(CE_PANIC
, "failed to add pool version");
3610 /* Newly created pools with the right version are always deflated. */
3611 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3612 spa
->spa_deflate
= TRUE
;
3613 if (zap_add(spa
->spa_meta_objset
,
3614 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3615 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3616 cmn_err(CE_PANIC
, "failed to add deflate");
3621 * Create the deferred-free bpobj. Turn off compression
3622 * because sync-to-convergence takes longer if the blocksize
3625 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3626 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3627 ZIO_COMPRESS_OFF
, tx
);
3628 if (zap_add(spa
->spa_meta_objset
,
3629 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3630 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3631 cmn_err(CE_PANIC
, "failed to add bpobj");
3633 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3634 spa
->spa_meta_objset
, obj
));
3637 * Create the pool's history object.
3639 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3640 spa_history_create_obj(spa
, tx
);
3643 * Set pool properties.
3645 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3646 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3647 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3648 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3650 if (props
!= NULL
) {
3651 spa_configfile_set(spa
, props
, B_FALSE
);
3652 spa_sync_props(props
, tx
);
3657 spa
->spa_sync_on
= B_TRUE
;
3658 txg_sync_start(spa
->spa_dsl_pool
);
3661 * We explicitly wait for the first transaction to complete so that our
3662 * bean counters are appropriately updated.
3664 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3666 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3668 spa_history_log_version(spa
, "create");
3670 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3672 mutex_exit(&spa_namespace_lock
);
3679 * Get the root pool information from the root disk, then import the root pool
3680 * during the system boot up time.
3682 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3685 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3688 nvlist_t
*nvtop
, *nvroot
;
3691 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3695 * Add this top-level vdev to the child array.
3697 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3699 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3701 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3704 * Put this pool's top-level vdevs into a root vdev.
3706 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3707 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3708 VDEV_TYPE_ROOT
) == 0);
3709 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3710 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3711 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3715 * Replace the existing vdev_tree with the new root vdev in
3716 * this pool's configuration (remove the old, add the new).
3718 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3719 nvlist_free(nvroot
);
3724 * Walk the vdev tree and see if we can find a device with "better"
3725 * configuration. A configuration is "better" if the label on that
3726 * device has a more recent txg.
3729 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3733 for (c
= 0; c
< vd
->vdev_children
; c
++)
3734 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3736 if (vd
->vdev_ops
->vdev_op_leaf
) {
3740 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3744 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3748 * Do we have a better boot device?
3750 if (label_txg
> *txg
) {
3759 * Import a root pool.
3761 * For x86. devpath_list will consist of devid and/or physpath name of
3762 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3763 * The GRUB "findroot" command will return the vdev we should boot.
3765 * For Sparc, devpath_list consists the physpath name of the booting device
3766 * no matter the rootpool is a single device pool or a mirrored pool.
3768 * "/pci@1f,0/ide@d/disk@0,0:a"
3771 spa_import_rootpool(char *devpath
, char *devid
)
3774 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3775 nvlist_t
*config
, *nvtop
;
3781 * Read the label from the boot device and generate a configuration.
3783 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3784 #if defined(_OBP) && defined(_KERNEL)
3785 if (config
== NULL
) {
3786 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3788 get_iscsi_bootpath_phy(devpath
);
3789 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3793 if (config
== NULL
) {
3794 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
3796 return (SET_ERROR(EIO
));
3799 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3801 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3803 mutex_enter(&spa_namespace_lock
);
3804 if ((spa
= spa_lookup(pname
)) != NULL
) {
3806 * Remove the existing root pool from the namespace so that we
3807 * can replace it with the correct config we just read in.
3812 spa
= spa_add(pname
, config
, NULL
);
3813 spa
->spa_is_root
= B_TRUE
;
3814 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3817 * Build up a vdev tree based on the boot device's label config.
3819 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3821 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3822 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3823 VDEV_ALLOC_ROOTPOOL
);
3824 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3826 mutex_exit(&spa_namespace_lock
);
3827 nvlist_free(config
);
3828 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3834 * Get the boot vdev.
3836 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3837 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3838 (u_longlong_t
)guid
);
3839 error
= SET_ERROR(ENOENT
);
3844 * Determine if there is a better boot device.
3847 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3849 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3850 "try booting from '%s'", avd
->vdev_path
);
3851 error
= SET_ERROR(EINVAL
);
3856 * If the boot device is part of a spare vdev then ensure that
3857 * we're booting off the active spare.
3859 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3860 !bvd
->vdev_isspare
) {
3861 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
3862 "try booting from '%s'",
3864 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
3865 error
= SET_ERROR(EINVAL
);
3871 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3873 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3874 mutex_exit(&spa_namespace_lock
);
3876 nvlist_free(config
);
3883 * Import a non-root pool into the system.
3886 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
3889 char *altroot
= NULL
;
3890 spa_load_state_t state
= SPA_LOAD_IMPORT
;
3891 zpool_rewind_policy_t policy
;
3892 uint64_t mode
= spa_mode_global
;
3893 uint64_t readonly
= B_FALSE
;
3896 nvlist_t
**spares
, **l2cache
;
3897 uint_t nspares
, nl2cache
;
3900 * If a pool with this name exists, return failure.
3902 mutex_enter(&spa_namespace_lock
);
3903 if (spa_lookup(pool
) != NULL
) {
3904 mutex_exit(&spa_namespace_lock
);
3905 return (SET_ERROR(EEXIST
));
3909 * Create and initialize the spa structure.
3911 (void) nvlist_lookup_string(props
,
3912 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3913 (void) nvlist_lookup_uint64(props
,
3914 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
3917 spa
= spa_add(pool
, config
, altroot
);
3918 spa
->spa_import_flags
= flags
;
3921 * Verbatim import - Take a pool and insert it into the namespace
3922 * as if it had been loaded at boot.
3924 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
3926 spa_configfile_set(spa
, props
, B_FALSE
);
3928 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3930 mutex_exit(&spa_namespace_lock
);
3934 spa_activate(spa
, mode
);
3937 * Don't start async tasks until we know everything is healthy.
3939 spa_async_suspend(spa
);
3941 zpool_get_rewind_policy(config
, &policy
);
3942 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3943 state
= SPA_LOAD_RECOVER
;
3946 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3947 * because the user-supplied config is actually the one to trust when
3950 if (state
!= SPA_LOAD_RECOVER
)
3951 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3953 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
3954 policy
.zrp_request
);
3957 * Propagate anything learned while loading the pool and pass it
3958 * back to caller (i.e. rewind info, missing devices, etc).
3960 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
3961 spa
->spa_load_info
) == 0);
3963 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3965 * Toss any existing sparelist, as it doesn't have any validity
3966 * anymore, and conflicts with spa_has_spare().
3968 if (spa
->spa_spares
.sav_config
) {
3969 nvlist_free(spa
->spa_spares
.sav_config
);
3970 spa
->spa_spares
.sav_config
= NULL
;
3971 spa_load_spares(spa
);
3973 if (spa
->spa_l2cache
.sav_config
) {
3974 nvlist_free(spa
->spa_l2cache
.sav_config
);
3975 spa
->spa_l2cache
.sav_config
= NULL
;
3976 spa_load_l2cache(spa
);
3979 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3982 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3985 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3986 VDEV_ALLOC_L2CACHE
);
3987 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3990 spa_configfile_set(spa
, props
, B_FALSE
);
3992 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
3993 (error
= spa_prop_set(spa
, props
)))) {
3995 spa_deactivate(spa
);
3997 mutex_exit(&spa_namespace_lock
);
4001 spa_async_resume(spa
);
4004 * Override any spares and level 2 cache devices as specified by
4005 * the user, as these may have correct device names/devids, etc.
4007 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4008 &spares
, &nspares
) == 0) {
4009 if (spa
->spa_spares
.sav_config
)
4010 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4011 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4013 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4014 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
4015 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4016 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4017 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4018 spa_load_spares(spa
);
4019 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4020 spa
->spa_spares
.sav_sync
= B_TRUE
;
4022 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4023 &l2cache
, &nl2cache
) == 0) {
4024 if (spa
->spa_l2cache
.sav_config
)
4025 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4026 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4028 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4029 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
4030 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4031 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4032 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4033 spa_load_l2cache(spa
);
4034 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4035 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4039 * Check for any removed devices.
4041 if (spa
->spa_autoreplace
) {
4042 spa_aux_check_removed(&spa
->spa_spares
);
4043 spa_aux_check_removed(&spa
->spa_l2cache
);
4046 if (spa_writeable(spa
)) {
4048 * Update the config cache to include the newly-imported pool.
4050 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4054 * It's possible that the pool was expanded while it was exported.
4055 * We kick off an async task to handle this for us.
4057 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4059 mutex_exit(&spa_namespace_lock
);
4060 spa_history_log_version(spa
, "import");
4063 zvol_create_minors(pool
);
4070 spa_tryimport(nvlist_t
*tryconfig
)
4072 nvlist_t
*config
= NULL
;
4078 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4081 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4085 * Create and initialize the spa structure.
4087 mutex_enter(&spa_namespace_lock
);
4088 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4089 spa_activate(spa
, FREAD
);
4092 * Pass off the heavy lifting to spa_load().
4093 * Pass TRUE for mosconfig because the user-supplied config
4094 * is actually the one to trust when doing an import.
4096 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4099 * If 'tryconfig' was at least parsable, return the current config.
4101 if (spa
->spa_root_vdev
!= NULL
) {
4102 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4103 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4105 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4107 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4108 spa
->spa_uberblock
.ub_timestamp
) == 0);
4109 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4110 spa
->spa_load_info
) == 0);
4111 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4112 spa
->spa_errata
) == 0);
4115 * If the bootfs property exists on this pool then we
4116 * copy it out so that external consumers can tell which
4117 * pools are bootable.
4119 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4120 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
4123 * We have to play games with the name since the
4124 * pool was opened as TRYIMPORT_NAME.
4126 if (dsl_dsobj_to_dsname(spa_name(spa
),
4127 spa
->spa_bootfs
, tmpname
) == 0) {
4131 dsname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
4133 cp
= strchr(tmpname
, '/');
4135 (void) strlcpy(dsname
, tmpname
,
4138 (void) snprintf(dsname
, MAXPATHLEN
,
4139 "%s/%s", poolname
, ++cp
);
4141 VERIFY(nvlist_add_string(config
,
4142 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4143 kmem_free(dsname
, MAXPATHLEN
);
4145 kmem_free(tmpname
, MAXPATHLEN
);
4149 * Add the list of hot spares and level 2 cache devices.
4151 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4152 spa_add_spares(spa
, config
);
4153 spa_add_l2cache(spa
, config
);
4154 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4158 spa_deactivate(spa
);
4160 mutex_exit(&spa_namespace_lock
);
4166 * Pool export/destroy
4168 * The act of destroying or exporting a pool is very simple. We make sure there
4169 * is no more pending I/O and any references to the pool are gone. Then, we
4170 * update the pool state and sync all the labels to disk, removing the
4171 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4172 * we don't sync the labels or remove the configuration cache.
4175 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4176 boolean_t force
, boolean_t hardforce
)
4183 if (!(spa_mode_global
& FWRITE
))
4184 return (SET_ERROR(EROFS
));
4186 mutex_enter(&spa_namespace_lock
);
4187 if ((spa
= spa_lookup(pool
)) == NULL
) {
4188 mutex_exit(&spa_namespace_lock
);
4189 return (SET_ERROR(ENOENT
));
4193 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4194 * reacquire the namespace lock, and see if we can export.
4196 spa_open_ref(spa
, FTAG
);
4197 mutex_exit(&spa_namespace_lock
);
4198 spa_async_suspend(spa
);
4199 mutex_enter(&spa_namespace_lock
);
4200 spa_close(spa
, FTAG
);
4203 * The pool will be in core if it's openable,
4204 * in which case we can modify its state.
4206 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
&& spa
->spa_sync_on
) {
4208 * Objsets may be open only because they're dirty, so we
4209 * have to force it to sync before checking spa_refcnt.
4211 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4214 * A pool cannot be exported or destroyed if there are active
4215 * references. If we are resetting a pool, allow references by
4216 * fault injection handlers.
4218 if (!spa_refcount_zero(spa
) ||
4219 (spa
->spa_inject_ref
!= 0 &&
4220 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4221 spa_async_resume(spa
);
4222 mutex_exit(&spa_namespace_lock
);
4223 return (SET_ERROR(EBUSY
));
4227 * A pool cannot be exported if it has an active shared spare.
4228 * This is to prevent other pools stealing the active spare
4229 * from an exported pool. At user's own will, such pool can
4230 * be forcedly exported.
4232 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4233 spa_has_active_shared_spare(spa
)) {
4234 spa_async_resume(spa
);
4235 mutex_exit(&spa_namespace_lock
);
4236 return (SET_ERROR(EXDEV
));
4240 * We want this to be reflected on every label,
4241 * so mark them all dirty. spa_unload() will do the
4242 * final sync that pushes these changes out.
4244 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4245 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4246 spa
->spa_state
= new_state
;
4247 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4249 vdev_config_dirty(spa
->spa_root_vdev
);
4250 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4254 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_DESTROY
);
4256 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4258 spa_deactivate(spa
);
4261 if (oldconfig
&& spa
->spa_config
)
4262 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4264 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4266 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4269 mutex_exit(&spa_namespace_lock
);
4275 * Destroy a storage pool.
4278 spa_destroy(char *pool
)
4280 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4285 * Export a storage pool.
4288 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4289 boolean_t hardforce
)
4291 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4296 * Similar to spa_export(), this unloads the spa_t without actually removing it
4297 * from the namespace in any way.
4300 spa_reset(char *pool
)
4302 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4307 * ==========================================================================
4308 * Device manipulation
4309 * ==========================================================================
4313 * Add a device to a storage pool.
4316 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4320 vdev_t
*rvd
= spa
->spa_root_vdev
;
4322 nvlist_t
**spares
, **l2cache
;
4323 uint_t nspares
, nl2cache
;
4326 ASSERT(spa_writeable(spa
));
4328 txg
= spa_vdev_enter(spa
);
4330 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4331 VDEV_ALLOC_ADD
)) != 0)
4332 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4334 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4336 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4340 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4344 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4345 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4347 if (vd
->vdev_children
!= 0 &&
4348 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4349 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4352 * We must validate the spares and l2cache devices after checking the
4353 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4355 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4356 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4359 * Transfer each new top-level vdev from vd to rvd.
4361 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4364 * Set the vdev id to the first hole, if one exists.
4366 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4367 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4368 vdev_free(rvd
->vdev_child
[id
]);
4372 tvd
= vd
->vdev_child
[c
];
4373 vdev_remove_child(vd
, tvd
);
4375 vdev_add_child(rvd
, tvd
);
4376 vdev_config_dirty(tvd
);
4380 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4381 ZPOOL_CONFIG_SPARES
);
4382 spa_load_spares(spa
);
4383 spa
->spa_spares
.sav_sync
= B_TRUE
;
4386 if (nl2cache
!= 0) {
4387 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4388 ZPOOL_CONFIG_L2CACHE
);
4389 spa_load_l2cache(spa
);
4390 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4394 * We have to be careful when adding new vdevs to an existing pool.
4395 * If other threads start allocating from these vdevs before we
4396 * sync the config cache, and we lose power, then upon reboot we may
4397 * fail to open the pool because there are DVAs that the config cache
4398 * can't translate. Therefore, we first add the vdevs without
4399 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4400 * and then let spa_config_update() initialize the new metaslabs.
4402 * spa_load() checks for added-but-not-initialized vdevs, so that
4403 * if we lose power at any point in this sequence, the remaining
4404 * steps will be completed the next time we load the pool.
4406 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4408 mutex_enter(&spa_namespace_lock
);
4409 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4410 mutex_exit(&spa_namespace_lock
);
4416 * Attach a device to a mirror. The arguments are the path to any device
4417 * in the mirror, and the nvroot for the new device. If the path specifies
4418 * a device that is not mirrored, we automatically insert the mirror vdev.
4420 * If 'replacing' is specified, the new device is intended to replace the
4421 * existing device; in this case the two devices are made into their own
4422 * mirror using the 'replacing' vdev, which is functionally identical to
4423 * the mirror vdev (it actually reuses all the same ops) but has a few
4424 * extra rules: you can't attach to it after it's been created, and upon
4425 * completion of resilvering, the first disk (the one being replaced)
4426 * is automatically detached.
4429 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4431 uint64_t txg
, dtl_max_txg
;
4432 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4434 char *oldvdpath
, *newvdpath
;
4437 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4439 ASSERT(spa_writeable(spa
));
4441 txg
= spa_vdev_enter(spa
);
4443 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4446 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4448 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4449 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4451 pvd
= oldvd
->vdev_parent
;
4453 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4454 VDEV_ALLOC_ATTACH
)) != 0)
4455 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4457 if (newrootvd
->vdev_children
!= 1)
4458 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4460 newvd
= newrootvd
->vdev_child
[0];
4462 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4463 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4465 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4466 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4469 * Spares can't replace logs
4471 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4472 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4476 * For attach, the only allowable parent is a mirror or the root
4479 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4480 pvd
->vdev_ops
!= &vdev_root_ops
)
4481 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4483 pvops
= &vdev_mirror_ops
;
4486 * Active hot spares can only be replaced by inactive hot
4489 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4490 oldvd
->vdev_isspare
&&
4491 !spa_has_spare(spa
, newvd
->vdev_guid
))
4492 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4495 * If the source is a hot spare, and the parent isn't already a
4496 * spare, then we want to create a new hot spare. Otherwise, we
4497 * want to create a replacing vdev. The user is not allowed to
4498 * attach to a spared vdev child unless the 'isspare' state is
4499 * the same (spare replaces spare, non-spare replaces
4502 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4503 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4504 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4505 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4506 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4507 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4510 if (newvd
->vdev_isspare
)
4511 pvops
= &vdev_spare_ops
;
4513 pvops
= &vdev_replacing_ops
;
4517 * Make sure the new device is big enough.
4519 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4520 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4523 * The new device cannot have a higher alignment requirement
4524 * than the top-level vdev.
4526 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4527 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4530 * If this is an in-place replacement, update oldvd's path and devid
4531 * to make it distinguishable from newvd, and unopenable from now on.
4533 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4534 spa_strfree(oldvd
->vdev_path
);
4535 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4537 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4538 newvd
->vdev_path
, "old");
4539 if (oldvd
->vdev_devid
!= NULL
) {
4540 spa_strfree(oldvd
->vdev_devid
);
4541 oldvd
->vdev_devid
= NULL
;
4545 /* mark the device being resilvered */
4546 newvd
->vdev_resilver_txg
= txg
;
4549 * If the parent is not a mirror, or if we're replacing, insert the new
4550 * mirror/replacing/spare vdev above oldvd.
4552 if (pvd
->vdev_ops
!= pvops
)
4553 pvd
= vdev_add_parent(oldvd
, pvops
);
4555 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4556 ASSERT(pvd
->vdev_ops
== pvops
);
4557 ASSERT(oldvd
->vdev_parent
== pvd
);
4560 * Extract the new device from its root and add it to pvd.
4562 vdev_remove_child(newrootvd
, newvd
);
4563 newvd
->vdev_id
= pvd
->vdev_children
;
4564 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4565 vdev_add_child(pvd
, newvd
);
4567 tvd
= newvd
->vdev_top
;
4568 ASSERT(pvd
->vdev_top
== tvd
);
4569 ASSERT(tvd
->vdev_parent
== rvd
);
4571 vdev_config_dirty(tvd
);
4574 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4575 * for any dmu_sync-ed blocks. It will propagate upward when
4576 * spa_vdev_exit() calls vdev_dtl_reassess().
4578 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4580 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4581 dtl_max_txg
- TXG_INITIAL
);
4583 if (newvd
->vdev_isspare
) {
4584 spa_spare_activate(newvd
);
4585 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_DEVICE_SPARE
);
4588 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4589 newvdpath
= spa_strdup(newvd
->vdev_path
);
4590 newvd_isspare
= newvd
->vdev_isspare
;
4593 * Mark newvd's DTL dirty in this txg.
4595 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4598 * Schedule the resilver to restart in the future. We do this to
4599 * ensure that dmu_sync-ed blocks have been stitched into the
4600 * respective datasets.
4602 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4607 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4609 spa_history_log_internal(spa
, "vdev attach", NULL
,
4610 "%s vdev=%s %s vdev=%s",
4611 replacing
&& newvd_isspare
? "spare in" :
4612 replacing
? "replace" : "attach", newvdpath
,
4613 replacing
? "for" : "to", oldvdpath
);
4615 spa_strfree(oldvdpath
);
4616 spa_strfree(newvdpath
);
4618 if (spa
->spa_bootfs
)
4619 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH
);
4625 * Detach a device from a mirror or replacing vdev.
4627 * If 'replace_done' is specified, only detach if the parent
4628 * is a replacing vdev.
4631 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4635 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4636 boolean_t unspare
= B_FALSE
;
4637 uint64_t unspare_guid
= 0;
4640 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4641 ASSERT(spa_writeable(spa
));
4643 txg
= spa_vdev_enter(spa
);
4645 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4648 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4650 if (!vd
->vdev_ops
->vdev_op_leaf
)
4651 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4653 pvd
= vd
->vdev_parent
;
4656 * If the parent/child relationship is not as expected, don't do it.
4657 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4658 * vdev that's replacing B with C. The user's intent in replacing
4659 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4660 * the replace by detaching C, the expected behavior is to end up
4661 * M(A,B). But suppose that right after deciding to detach C,
4662 * the replacement of B completes. We would have M(A,C), and then
4663 * ask to detach C, which would leave us with just A -- not what
4664 * the user wanted. To prevent this, we make sure that the
4665 * parent/child relationship hasn't changed -- in this example,
4666 * that C's parent is still the replacing vdev R.
4668 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4669 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4672 * Only 'replacing' or 'spare' vdevs can be replaced.
4674 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4675 pvd
->vdev_ops
!= &vdev_spare_ops
)
4676 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4678 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4679 spa_version(spa
) >= SPA_VERSION_SPARES
);
4682 * Only mirror, replacing, and spare vdevs support detach.
4684 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4685 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4686 pvd
->vdev_ops
!= &vdev_spare_ops
)
4687 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4690 * If this device has the only valid copy of some data,
4691 * we cannot safely detach it.
4693 if (vdev_dtl_required(vd
))
4694 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4696 ASSERT(pvd
->vdev_children
>= 2);
4699 * If we are detaching the second disk from a replacing vdev, then
4700 * check to see if we changed the original vdev's path to have "/old"
4701 * at the end in spa_vdev_attach(). If so, undo that change now.
4703 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4704 vd
->vdev_path
!= NULL
) {
4705 size_t len
= strlen(vd
->vdev_path
);
4707 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4708 cvd
= pvd
->vdev_child
[c
];
4710 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4713 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4714 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4715 spa_strfree(cvd
->vdev_path
);
4716 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4723 * If we are detaching the original disk from a spare, then it implies
4724 * that the spare should become a real disk, and be removed from the
4725 * active spare list for the pool.
4727 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4729 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4733 * Erase the disk labels so the disk can be used for other things.
4734 * This must be done after all other error cases are handled,
4735 * but before we disembowel vd (so we can still do I/O to it).
4736 * But if we can't do it, don't treat the error as fatal --
4737 * it may be that the unwritability of the disk is the reason
4738 * it's being detached!
4740 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4743 * Remove vd from its parent and compact the parent's children.
4745 vdev_remove_child(pvd
, vd
);
4746 vdev_compact_children(pvd
);
4749 * Remember one of the remaining children so we can get tvd below.
4751 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4754 * If we need to remove the remaining child from the list of hot spares,
4755 * do it now, marking the vdev as no longer a spare in the process.
4756 * We must do this before vdev_remove_parent(), because that can
4757 * change the GUID if it creates a new toplevel GUID. For a similar
4758 * reason, we must remove the spare now, in the same txg as the detach;
4759 * otherwise someone could attach a new sibling, change the GUID, and
4760 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4763 ASSERT(cvd
->vdev_isspare
);
4764 spa_spare_remove(cvd
);
4765 unspare_guid
= cvd
->vdev_guid
;
4766 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4767 cvd
->vdev_unspare
= B_TRUE
;
4771 * If the parent mirror/replacing vdev only has one child,
4772 * the parent is no longer needed. Remove it from the tree.
4774 if (pvd
->vdev_children
== 1) {
4775 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4776 cvd
->vdev_unspare
= B_FALSE
;
4777 vdev_remove_parent(cvd
);
4782 * We don't set tvd until now because the parent we just removed
4783 * may have been the previous top-level vdev.
4785 tvd
= cvd
->vdev_top
;
4786 ASSERT(tvd
->vdev_parent
== rvd
);
4789 * Reevaluate the parent vdev state.
4791 vdev_propagate_state(cvd
);
4794 * If the 'autoexpand' property is set on the pool then automatically
4795 * try to expand the size of the pool. For example if the device we
4796 * just detached was smaller than the others, it may be possible to
4797 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4798 * first so that we can obtain the updated sizes of the leaf vdevs.
4800 if (spa
->spa_autoexpand
) {
4802 vdev_expand(tvd
, txg
);
4805 vdev_config_dirty(tvd
);
4808 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4809 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4810 * But first make sure we're not on any *other* txg's DTL list, to
4811 * prevent vd from being accessed after it's freed.
4813 vdpath
= spa_strdup(vd
->vdev_path
);
4814 for (t
= 0; t
< TXG_SIZE
; t
++)
4815 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4816 vd
->vdev_detached
= B_TRUE
;
4817 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4819 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_REMOVE
);
4821 /* hang on to the spa before we release the lock */
4822 spa_open_ref(spa
, FTAG
);
4824 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4826 spa_history_log_internal(spa
, "detach", NULL
,
4828 spa_strfree(vdpath
);
4831 * If this was the removal of the original device in a hot spare vdev,
4832 * then we want to go through and remove the device from the hot spare
4833 * list of every other pool.
4836 spa_t
*altspa
= NULL
;
4838 mutex_enter(&spa_namespace_lock
);
4839 while ((altspa
= spa_next(altspa
)) != NULL
) {
4840 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4844 spa_open_ref(altspa
, FTAG
);
4845 mutex_exit(&spa_namespace_lock
);
4846 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4847 mutex_enter(&spa_namespace_lock
);
4848 spa_close(altspa
, FTAG
);
4850 mutex_exit(&spa_namespace_lock
);
4852 /* search the rest of the vdevs for spares to remove */
4853 spa_vdev_resilver_done(spa
);
4856 /* all done with the spa; OK to release */
4857 mutex_enter(&spa_namespace_lock
);
4858 spa_close(spa
, FTAG
);
4859 mutex_exit(&spa_namespace_lock
);
4865 * Split a set of devices from their mirrors, and create a new pool from them.
4868 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
4869 nvlist_t
*props
, boolean_t exp
)
4872 uint64_t txg
, *glist
;
4874 uint_t c
, children
, lastlog
;
4875 nvlist_t
**child
, *nvl
, *tmp
;
4877 char *altroot
= NULL
;
4878 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
4879 boolean_t activate_slog
;
4881 ASSERT(spa_writeable(spa
));
4883 txg
= spa_vdev_enter(spa
);
4885 /* clear the log and flush everything up to now */
4886 activate_slog
= spa_passivate_log(spa
);
4887 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4888 error
= spa_offline_log(spa
);
4889 txg
= spa_vdev_config_enter(spa
);
4892 spa_activate_log(spa
);
4895 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4897 /* check new spa name before going any further */
4898 if (spa_lookup(newname
) != NULL
)
4899 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
4902 * scan through all the children to ensure they're all mirrors
4904 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
4905 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
4907 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4909 /* first, check to ensure we've got the right child count */
4910 rvd
= spa
->spa_root_vdev
;
4912 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
4913 vdev_t
*vd
= rvd
->vdev_child
[c
];
4915 /* don't count the holes & logs as children */
4916 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
4924 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
4925 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4927 /* next, ensure no spare or cache devices are part of the split */
4928 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
4929 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
4930 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4932 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_PUSHPAGE
);
4933 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_PUSHPAGE
);
4935 /* then, loop over each vdev and validate it */
4936 for (c
= 0; c
< children
; c
++) {
4937 uint64_t is_hole
= 0;
4939 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
4943 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
4944 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
4947 error
= SET_ERROR(EINVAL
);
4952 /* which disk is going to be split? */
4953 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
4955 error
= SET_ERROR(EINVAL
);
4959 /* look it up in the spa */
4960 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
4961 if (vml
[c
] == NULL
) {
4962 error
= SET_ERROR(ENODEV
);
4966 /* make sure there's nothing stopping the split */
4967 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
4968 vml
[c
]->vdev_islog
||
4969 vml
[c
]->vdev_ishole
||
4970 vml
[c
]->vdev_isspare
||
4971 vml
[c
]->vdev_isl2cache
||
4972 !vdev_writeable(vml
[c
]) ||
4973 vml
[c
]->vdev_children
!= 0 ||
4974 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
4975 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
4976 error
= SET_ERROR(EINVAL
);
4980 if (vdev_dtl_required(vml
[c
])) {
4981 error
= SET_ERROR(EBUSY
);
4985 /* we need certain info from the top level */
4986 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
4987 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
4988 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
4989 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
4990 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
4991 vml
[c
]->vdev_top
->vdev_asize
) == 0);
4992 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
4993 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
4997 kmem_free(vml
, children
* sizeof (vdev_t
*));
4998 kmem_free(glist
, children
* sizeof (uint64_t));
4999 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5002 /* stop writers from using the disks */
5003 for (c
= 0; c
< children
; c
++) {
5005 vml
[c
]->vdev_offline
= B_TRUE
;
5007 vdev_reopen(spa
->spa_root_vdev
);
5010 * Temporarily record the splitting vdevs in the spa config. This
5011 * will disappear once the config is regenerated.
5013 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
5014 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5015 glist
, children
) == 0);
5016 kmem_free(glist
, children
* sizeof (uint64_t));
5018 mutex_enter(&spa
->spa_props_lock
);
5019 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5021 mutex_exit(&spa
->spa_props_lock
);
5022 spa
->spa_config_splitting
= nvl
;
5023 vdev_config_dirty(spa
->spa_root_vdev
);
5025 /* configure and create the new pool */
5026 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5027 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5028 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5029 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5030 spa_version(spa
)) == 0);
5031 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5032 spa
->spa_config_txg
) == 0);
5033 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5034 spa_generate_guid(NULL
)) == 0);
5035 (void) nvlist_lookup_string(props
,
5036 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5038 /* add the new pool to the namespace */
5039 newspa
= spa_add(newname
, config
, altroot
);
5040 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5041 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5043 /* release the spa config lock, retaining the namespace lock */
5044 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5046 if (zio_injection_enabled
)
5047 zio_handle_panic_injection(spa
, FTAG
, 1);
5049 spa_activate(newspa
, spa_mode_global
);
5050 spa_async_suspend(newspa
);
5052 /* create the new pool from the disks of the original pool */
5053 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5057 /* if that worked, generate a real config for the new pool */
5058 if (newspa
->spa_root_vdev
!= NULL
) {
5059 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5060 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
5061 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5062 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5063 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5068 if (props
!= NULL
) {
5069 spa_configfile_set(newspa
, props
, B_FALSE
);
5070 error
= spa_prop_set(newspa
, props
);
5075 /* flush everything */
5076 txg
= spa_vdev_config_enter(newspa
);
5077 vdev_config_dirty(newspa
->spa_root_vdev
);
5078 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5080 if (zio_injection_enabled
)
5081 zio_handle_panic_injection(spa
, FTAG
, 2);
5083 spa_async_resume(newspa
);
5085 /* finally, update the original pool's config */
5086 txg
= spa_vdev_config_enter(spa
);
5087 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5088 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5091 for (c
= 0; c
< children
; c
++) {
5092 if (vml
[c
] != NULL
) {
5095 spa_history_log_internal(spa
, "detach", tx
,
5096 "vdev=%s", vml
[c
]->vdev_path
);
5100 vdev_config_dirty(spa
->spa_root_vdev
);
5101 spa
->spa_config_splitting
= NULL
;
5105 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5107 if (zio_injection_enabled
)
5108 zio_handle_panic_injection(spa
, FTAG
, 3);
5110 /* split is complete; log a history record */
5111 spa_history_log_internal(newspa
, "split", NULL
,
5112 "from pool %s", spa_name(spa
));
5114 kmem_free(vml
, children
* sizeof (vdev_t
*));
5116 /* if we're not going to mount the filesystems in userland, export */
5118 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5125 spa_deactivate(newspa
);
5128 txg
= spa_vdev_config_enter(spa
);
5130 /* re-online all offlined disks */
5131 for (c
= 0; c
< children
; c
++) {
5133 vml
[c
]->vdev_offline
= B_FALSE
;
5135 vdev_reopen(spa
->spa_root_vdev
);
5137 nvlist_free(spa
->spa_config_splitting
);
5138 spa
->spa_config_splitting
= NULL
;
5139 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5141 kmem_free(vml
, children
* sizeof (vdev_t
*));
5146 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5150 for (i
= 0; i
< count
; i
++) {
5153 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5156 if (guid
== target_guid
)
5164 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5165 nvlist_t
*dev_to_remove
)
5167 nvlist_t
**newdev
= NULL
;
5171 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_PUSHPAGE
);
5173 for (i
= 0, j
= 0; i
< count
; i
++) {
5174 if (dev
[i
] == dev_to_remove
)
5176 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_PUSHPAGE
) == 0);
5179 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5180 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5182 for (i
= 0; i
< count
- 1; i
++)
5183 nvlist_free(newdev
[i
]);
5186 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5190 * Evacuate the device.
5193 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5198 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5199 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5200 ASSERT(vd
== vd
->vdev_top
);
5203 * Evacuate the device. We don't hold the config lock as writer
5204 * since we need to do I/O but we do keep the
5205 * spa_namespace_lock held. Once this completes the device
5206 * should no longer have any blocks allocated on it.
5208 if (vd
->vdev_islog
) {
5209 if (vd
->vdev_stat
.vs_alloc
!= 0)
5210 error
= spa_offline_log(spa
);
5212 error
= SET_ERROR(ENOTSUP
);
5219 * The evacuation succeeded. Remove any remaining MOS metadata
5220 * associated with this vdev, and wait for these changes to sync.
5222 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5223 txg
= spa_vdev_config_enter(spa
);
5224 vd
->vdev_removing
= B_TRUE
;
5225 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5226 vdev_config_dirty(vd
);
5227 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5233 * Complete the removal by cleaning up the namespace.
5236 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5238 vdev_t
*rvd
= spa
->spa_root_vdev
;
5239 uint64_t id
= vd
->vdev_id
;
5240 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5242 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5243 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5244 ASSERT(vd
== vd
->vdev_top
);
5247 * Only remove any devices which are empty.
5249 if (vd
->vdev_stat
.vs_alloc
!= 0)
5252 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5254 if (list_link_active(&vd
->vdev_state_dirty_node
))
5255 vdev_state_clean(vd
);
5256 if (list_link_active(&vd
->vdev_config_dirty_node
))
5257 vdev_config_clean(vd
);
5262 vdev_compact_children(rvd
);
5264 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5265 vdev_add_child(rvd
, vd
);
5267 vdev_config_dirty(rvd
);
5270 * Reassess the health of our root vdev.
5276 * Remove a device from the pool -
5278 * Removing a device from the vdev namespace requires several steps
5279 * and can take a significant amount of time. As a result we use
5280 * the spa_vdev_config_[enter/exit] functions which allow us to
5281 * grab and release the spa_config_lock while still holding the namespace
5282 * lock. During each step the configuration is synced out.
5284 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5288 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5291 metaslab_group_t
*mg
;
5292 nvlist_t
**spares
, **l2cache
, *nv
;
5294 uint_t nspares
, nl2cache
;
5296 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5298 ASSERT(spa_writeable(spa
));
5301 txg
= spa_vdev_enter(spa
);
5303 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5305 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5306 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5307 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5308 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5310 * Only remove the hot spare if it's not currently in use
5313 if (vd
== NULL
|| unspare
) {
5314 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5315 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5316 spa_load_spares(spa
);
5317 spa
->spa_spares
.sav_sync
= B_TRUE
;
5319 error
= SET_ERROR(EBUSY
);
5321 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5322 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5323 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5324 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5326 * Cache devices can always be removed.
5328 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5329 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5330 spa_load_l2cache(spa
);
5331 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5332 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5334 ASSERT(vd
== vd
->vdev_top
);
5337 * XXX - Once we have bp-rewrite this should
5338 * become the common case.
5344 * Stop allocating from this vdev.
5346 metaslab_group_passivate(mg
);
5349 * Wait for the youngest allocations and frees to sync,
5350 * and then wait for the deferral of those frees to finish.
5352 spa_vdev_config_exit(spa
, NULL
,
5353 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5356 * Attempt to evacuate the vdev.
5358 error
= spa_vdev_remove_evacuate(spa
, vd
);
5360 txg
= spa_vdev_config_enter(spa
);
5363 * If we couldn't evacuate the vdev, unwind.
5366 metaslab_group_activate(mg
);
5367 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5371 * Clean up the vdev namespace.
5373 spa_vdev_remove_from_namespace(spa
, vd
);
5375 } else if (vd
!= NULL
) {
5377 * Normal vdevs cannot be removed (yet).
5379 error
= SET_ERROR(ENOTSUP
);
5382 * There is no vdev of any kind with the specified guid.
5384 error
= SET_ERROR(ENOENT
);
5388 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5394 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5395 * currently spared, so we can detach it.
5398 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5400 vdev_t
*newvd
, *oldvd
;
5403 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5404 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5410 * Check for a completed replacement. We always consider the first
5411 * vdev in the list to be the oldest vdev, and the last one to be
5412 * the newest (see spa_vdev_attach() for how that works). In
5413 * the case where the newest vdev is faulted, we will not automatically
5414 * remove it after a resilver completes. This is OK as it will require
5415 * user intervention to determine which disk the admin wishes to keep.
5417 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5418 ASSERT(vd
->vdev_children
> 1);
5420 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5421 oldvd
= vd
->vdev_child
[0];
5423 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5424 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5425 !vdev_dtl_required(oldvd
))
5430 * Check for a completed resilver with the 'unspare' flag set.
5432 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5433 vdev_t
*first
= vd
->vdev_child
[0];
5434 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5436 if (last
->vdev_unspare
) {
5439 } else if (first
->vdev_unspare
) {
5446 if (oldvd
!= NULL
&&
5447 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5448 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5449 !vdev_dtl_required(oldvd
))
5453 * If there are more than two spares attached to a disk,
5454 * and those spares are not required, then we want to
5455 * attempt to free them up now so that they can be used
5456 * by other pools. Once we're back down to a single
5457 * disk+spare, we stop removing them.
5459 if (vd
->vdev_children
> 2) {
5460 newvd
= vd
->vdev_child
[1];
5462 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5463 vdev_dtl_empty(last
, DTL_MISSING
) &&
5464 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5465 !vdev_dtl_required(newvd
))
5474 spa_vdev_resilver_done(spa_t
*spa
)
5476 vdev_t
*vd
, *pvd
, *ppvd
;
5477 uint64_t guid
, sguid
, pguid
, ppguid
;
5479 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5481 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5482 pvd
= vd
->vdev_parent
;
5483 ppvd
= pvd
->vdev_parent
;
5484 guid
= vd
->vdev_guid
;
5485 pguid
= pvd
->vdev_guid
;
5486 ppguid
= ppvd
->vdev_guid
;
5489 * If we have just finished replacing a hot spared device, then
5490 * we need to detach the parent's first child (the original hot
5493 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5494 ppvd
->vdev_children
== 2) {
5495 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5496 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5498 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5500 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5501 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5503 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5505 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5508 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5512 * Update the stored path or FRU for this vdev.
5515 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5519 boolean_t sync
= B_FALSE
;
5521 ASSERT(spa_writeable(spa
));
5523 spa_vdev_state_enter(spa
, SCL_ALL
);
5525 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5526 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5528 if (!vd
->vdev_ops
->vdev_op_leaf
)
5529 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5532 if (strcmp(value
, vd
->vdev_path
) != 0) {
5533 spa_strfree(vd
->vdev_path
);
5534 vd
->vdev_path
= spa_strdup(value
);
5538 if (vd
->vdev_fru
== NULL
) {
5539 vd
->vdev_fru
= spa_strdup(value
);
5541 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5542 spa_strfree(vd
->vdev_fru
);
5543 vd
->vdev_fru
= spa_strdup(value
);
5548 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5552 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5554 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5558 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5560 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5564 * ==========================================================================
5566 * ==========================================================================
5570 spa_scan_stop(spa_t
*spa
)
5572 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5573 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5574 return (SET_ERROR(EBUSY
));
5575 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5579 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5581 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5583 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5584 return (SET_ERROR(ENOTSUP
));
5587 * If a resilver was requested, but there is no DTL on a
5588 * writeable leaf device, we have nothing to do.
5590 if (func
== POOL_SCAN_RESILVER
&&
5591 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5592 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5596 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5600 * ==========================================================================
5601 * SPA async task processing
5602 * ==========================================================================
5606 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5610 if (vd
->vdev_remove_wanted
) {
5611 vd
->vdev_remove_wanted
= B_FALSE
;
5612 vd
->vdev_delayed_close
= B_FALSE
;
5613 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5616 * We want to clear the stats, but we don't want to do a full
5617 * vdev_clear() as that will cause us to throw away
5618 * degraded/faulted state as well as attempt to reopen the
5619 * device, all of which is a waste.
5621 vd
->vdev_stat
.vs_read_errors
= 0;
5622 vd
->vdev_stat
.vs_write_errors
= 0;
5623 vd
->vdev_stat
.vs_checksum_errors
= 0;
5625 vdev_state_dirty(vd
->vdev_top
);
5628 for (c
= 0; c
< vd
->vdev_children
; c
++)
5629 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5633 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5637 if (vd
->vdev_probe_wanted
) {
5638 vd
->vdev_probe_wanted
= B_FALSE
;
5639 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5642 for (c
= 0; c
< vd
->vdev_children
; c
++)
5643 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5647 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5651 if (!spa
->spa_autoexpand
)
5654 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5655 vdev_t
*cvd
= vd
->vdev_child
[c
];
5656 spa_async_autoexpand(spa
, cvd
);
5659 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5662 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND
);
5666 spa_async_thread(spa_t
*spa
)
5670 ASSERT(spa
->spa_sync_on
);
5672 mutex_enter(&spa
->spa_async_lock
);
5673 tasks
= spa
->spa_async_tasks
;
5674 spa
->spa_async_tasks
= 0;
5675 mutex_exit(&spa
->spa_async_lock
);
5678 * See if the config needs to be updated.
5680 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5681 uint64_t old_space
, new_space
;
5683 mutex_enter(&spa_namespace_lock
);
5684 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5685 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5686 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5687 mutex_exit(&spa_namespace_lock
);
5690 * If the pool grew as a result of the config update,
5691 * then log an internal history event.
5693 if (new_space
!= old_space
) {
5694 spa_history_log_internal(spa
, "vdev online", NULL
,
5695 "pool '%s' size: %llu(+%llu)",
5696 spa_name(spa
), new_space
, new_space
- old_space
);
5701 * See if any devices need to be marked REMOVED.
5703 if (tasks
& SPA_ASYNC_REMOVE
) {
5704 spa_vdev_state_enter(spa
, SCL_NONE
);
5705 spa_async_remove(spa
, spa
->spa_root_vdev
);
5706 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5707 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5708 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5709 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5710 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5713 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5714 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5715 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5716 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5720 * See if any devices need to be probed.
5722 if (tasks
& SPA_ASYNC_PROBE
) {
5723 spa_vdev_state_enter(spa
, SCL_NONE
);
5724 spa_async_probe(spa
, spa
->spa_root_vdev
);
5725 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5729 * If any devices are done replacing, detach them.
5731 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5732 spa_vdev_resilver_done(spa
);
5735 * Kick off a resilver.
5737 if (tasks
& SPA_ASYNC_RESILVER
)
5738 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5741 * Let the world know that we're done.
5743 mutex_enter(&spa
->spa_async_lock
);
5744 spa
->spa_async_thread
= NULL
;
5745 cv_broadcast(&spa
->spa_async_cv
);
5746 mutex_exit(&spa
->spa_async_lock
);
5751 spa_async_suspend(spa_t
*spa
)
5753 mutex_enter(&spa
->spa_async_lock
);
5754 spa
->spa_async_suspended
++;
5755 while (spa
->spa_async_thread
!= NULL
)
5756 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5757 mutex_exit(&spa
->spa_async_lock
);
5761 spa_async_resume(spa_t
*spa
)
5763 mutex_enter(&spa
->spa_async_lock
);
5764 ASSERT(spa
->spa_async_suspended
!= 0);
5765 spa
->spa_async_suspended
--;
5766 mutex_exit(&spa
->spa_async_lock
);
5770 spa_async_dispatch(spa_t
*spa
)
5772 mutex_enter(&spa
->spa_async_lock
);
5773 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5774 spa
->spa_async_thread
== NULL
&&
5775 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5776 spa
->spa_async_thread
= thread_create(NULL
, 0,
5777 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5778 mutex_exit(&spa
->spa_async_lock
);
5782 spa_async_request(spa_t
*spa
, int task
)
5784 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5785 mutex_enter(&spa
->spa_async_lock
);
5786 spa
->spa_async_tasks
|= task
;
5787 mutex_exit(&spa
->spa_async_lock
);
5791 * ==========================================================================
5792 * SPA syncing routines
5793 * ==========================================================================
5797 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5800 bpobj_enqueue(bpo
, bp
, tx
);
5805 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5809 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5815 * Note: this simple function is not inlined to make it easier to dtrace the
5816 * amount of time spent syncing frees.
5819 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
5821 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5822 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
5823 VERIFY(zio_wait(zio
) == 0);
5827 * Note: this simple function is not inlined to make it easier to dtrace the
5828 * amount of time spent syncing deferred frees.
5831 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
5833 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5834 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
5835 spa_free_sync_cb
, zio
, tx
), ==, 0);
5836 VERIFY0(zio_wait(zio
));
5840 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5842 char *packed
= NULL
;
5847 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5850 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5851 * information. This avoids the dmu_buf_will_dirty() path and
5852 * saves us a pre-read to get data we don't actually care about.
5854 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
5855 packed
= vmem_alloc(bufsize
, KM_PUSHPAGE
);
5857 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
5859 bzero(packed
+ nvsize
, bufsize
- nvsize
);
5861 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
5863 vmem_free(packed
, bufsize
);
5865 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
5866 dmu_buf_will_dirty(db
, tx
);
5867 *(uint64_t *)db
->db_data
= nvsize
;
5868 dmu_buf_rele(db
, FTAG
);
5872 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
5873 const char *config
, const char *entry
)
5883 * Update the MOS nvlist describing the list of available devices.
5884 * spa_validate_aux() will have already made sure this nvlist is
5885 * valid and the vdevs are labeled appropriately.
5887 if (sav
->sav_object
== 0) {
5888 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
5889 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
5890 sizeof (uint64_t), tx
);
5891 VERIFY(zap_update(spa
->spa_meta_objset
,
5892 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
5893 &sav
->sav_object
, tx
) == 0);
5896 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
5897 if (sav
->sav_count
== 0) {
5898 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
5900 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_PUSHPAGE
);
5901 for (i
= 0; i
< sav
->sav_count
; i
++)
5902 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
5903 B_FALSE
, VDEV_CONFIG_L2CACHE
);
5904 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
5905 sav
->sav_count
) == 0);
5906 for (i
= 0; i
< sav
->sav_count
; i
++)
5907 nvlist_free(list
[i
]);
5908 kmem_free(list
, sav
->sav_count
* sizeof (void *));
5911 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
5912 nvlist_free(nvroot
);
5914 sav
->sav_sync
= B_FALSE
;
5918 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
5922 if (list_is_empty(&spa
->spa_config_dirty_list
))
5925 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5927 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
5928 dmu_tx_get_txg(tx
), B_FALSE
);
5931 * If we're upgrading the spa version then make sure that
5932 * the config object gets updated with the correct version.
5934 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
5935 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5936 spa
->spa_uberblock
.ub_version
);
5938 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5940 if (spa
->spa_config_syncing
)
5941 nvlist_free(spa
->spa_config_syncing
);
5942 spa
->spa_config_syncing
= config
;
5944 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
5948 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
5950 uint64_t *versionp
= arg
;
5951 uint64_t version
= *versionp
;
5952 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5955 * Setting the version is special cased when first creating the pool.
5957 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
5959 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
5960 ASSERT(version
>= spa_version(spa
));
5962 spa
->spa_uberblock
.ub_version
= version
;
5963 vdev_config_dirty(spa
->spa_root_vdev
);
5964 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
5968 * Set zpool properties.
5971 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
5973 nvlist_t
*nvp
= arg
;
5974 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5975 objset_t
*mos
= spa
->spa_meta_objset
;
5976 nvpair_t
*elem
= NULL
;
5978 mutex_enter(&spa
->spa_props_lock
);
5980 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
5982 char *strval
, *fname
;
5984 const char *propname
;
5985 zprop_type_t proptype
;
5988 prop
= zpool_name_to_prop(nvpair_name(elem
));
5989 switch ((int)prop
) {
5992 * We checked this earlier in spa_prop_validate().
5994 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
5996 fname
= strchr(nvpair_name(elem
), '@') + 1;
5997 VERIFY0(zfeature_lookup_name(fname
, &fid
));
5999 spa_feature_enable(spa
, fid
, tx
);
6000 spa_history_log_internal(spa
, "set", tx
,
6001 "%s=enabled", nvpair_name(elem
));
6004 case ZPOOL_PROP_VERSION
:
6005 intval
= fnvpair_value_uint64(elem
);
6007 * The version is synced seperatly before other
6008 * properties and should be correct by now.
6010 ASSERT3U(spa_version(spa
), >=, intval
);
6013 case ZPOOL_PROP_ALTROOT
:
6015 * 'altroot' is a non-persistent property. It should
6016 * have been set temporarily at creation or import time.
6018 ASSERT(spa
->spa_root
!= NULL
);
6021 case ZPOOL_PROP_READONLY
:
6022 case ZPOOL_PROP_CACHEFILE
:
6024 * 'readonly' and 'cachefile' are also non-persisitent
6028 case ZPOOL_PROP_COMMENT
:
6029 strval
= fnvpair_value_string(elem
);
6030 if (spa
->spa_comment
!= NULL
)
6031 spa_strfree(spa
->spa_comment
);
6032 spa
->spa_comment
= spa_strdup(strval
);
6034 * We need to dirty the configuration on all the vdevs
6035 * so that their labels get updated. It's unnecessary
6036 * to do this for pool creation since the vdev's
6037 * configuratoin has already been dirtied.
6039 if (tx
->tx_txg
!= TXG_INITIAL
)
6040 vdev_config_dirty(spa
->spa_root_vdev
);
6041 spa_history_log_internal(spa
, "set", tx
,
6042 "%s=%s", nvpair_name(elem
), strval
);
6046 * Set pool property values in the poolprops mos object.
6048 if (spa
->spa_pool_props_object
== 0) {
6049 spa
->spa_pool_props_object
=
6050 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6051 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6055 /* normalize the property name */
6056 propname
= zpool_prop_to_name(prop
);
6057 proptype
= zpool_prop_get_type(prop
);
6059 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6060 ASSERT(proptype
== PROP_TYPE_STRING
);
6061 strval
= fnvpair_value_string(elem
);
6062 VERIFY0(zap_update(mos
,
6063 spa
->spa_pool_props_object
, propname
,
6064 1, strlen(strval
) + 1, strval
, tx
));
6065 spa_history_log_internal(spa
, "set", tx
,
6066 "%s=%s", nvpair_name(elem
), strval
);
6067 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6068 intval
= fnvpair_value_uint64(elem
);
6070 if (proptype
== PROP_TYPE_INDEX
) {
6072 VERIFY0(zpool_prop_index_to_string(
6073 prop
, intval
, &unused
));
6075 VERIFY0(zap_update(mos
,
6076 spa
->spa_pool_props_object
, propname
,
6077 8, 1, &intval
, tx
));
6078 spa_history_log_internal(spa
, "set", tx
,
6079 "%s=%lld", nvpair_name(elem
), intval
);
6081 ASSERT(0); /* not allowed */
6085 case ZPOOL_PROP_DELEGATION
:
6086 spa
->spa_delegation
= intval
;
6088 case ZPOOL_PROP_BOOTFS
:
6089 spa
->spa_bootfs
= intval
;
6091 case ZPOOL_PROP_FAILUREMODE
:
6092 spa
->spa_failmode
= intval
;
6094 case ZPOOL_PROP_AUTOEXPAND
:
6095 spa
->spa_autoexpand
= intval
;
6096 if (tx
->tx_txg
!= TXG_INITIAL
)
6097 spa_async_request(spa
,
6098 SPA_ASYNC_AUTOEXPAND
);
6100 case ZPOOL_PROP_DEDUPDITTO
:
6101 spa
->spa_dedup_ditto
= intval
;
6110 mutex_exit(&spa
->spa_props_lock
);
6114 * Perform one-time upgrade on-disk changes. spa_version() does not
6115 * reflect the new version this txg, so there must be no changes this
6116 * txg to anything that the upgrade code depends on after it executes.
6117 * Therefore this must be called after dsl_pool_sync() does the sync
6121 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6123 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6125 ASSERT(spa
->spa_sync_pass
== 1);
6127 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6129 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6130 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6131 dsl_pool_create_origin(dp
, tx
);
6133 /* Keeping the origin open increases spa_minref */
6134 spa
->spa_minref
+= 3;
6137 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6138 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6139 dsl_pool_upgrade_clones(dp
, tx
);
6142 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6143 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6144 dsl_pool_upgrade_dir_clones(dp
, tx
);
6146 /* Keeping the freedir open increases spa_minref */
6147 spa
->spa_minref
+= 3;
6150 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6151 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6152 spa_feature_create_zap_objects(spa
, tx
);
6154 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6158 * Sync the specified transaction group. New blocks may be dirtied as
6159 * part of the process, so we iterate until it converges.
6162 spa_sync(spa_t
*spa
, uint64_t txg
)
6164 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6165 objset_t
*mos
= spa
->spa_meta_objset
;
6166 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6167 vdev_t
*rvd
= spa
->spa_root_vdev
;
6173 VERIFY(spa_writeable(spa
));
6176 * Lock out configuration changes.
6178 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6180 spa
->spa_syncing_txg
= txg
;
6181 spa
->spa_sync_pass
= 0;
6184 * If there are any pending vdev state changes, convert them
6185 * into config changes that go out with this transaction group.
6187 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6188 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6190 * We need the write lock here because, for aux vdevs,
6191 * calling vdev_config_dirty() modifies sav_config.
6192 * This is ugly and will become unnecessary when we
6193 * eliminate the aux vdev wart by integrating all vdevs
6194 * into the root vdev tree.
6196 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6197 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6198 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6199 vdev_state_clean(vd
);
6200 vdev_config_dirty(vd
);
6202 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6203 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6205 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6207 tx
= dmu_tx_create_assigned(dp
, txg
);
6209 spa
->spa_sync_starttime
= gethrtime();
6210 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6211 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
6212 spa_deadman
, spa
, TQ_PUSHPAGE
, ddi_get_lbolt() +
6213 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6216 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6217 * set spa_deflate if we have no raid-z vdevs.
6219 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6220 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6223 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6224 vd
= rvd
->vdev_child
[i
];
6225 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6228 if (i
== rvd
->vdev_children
) {
6229 spa
->spa_deflate
= TRUE
;
6230 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6231 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6232 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6237 * If anything has changed in this txg, or if someone is waiting
6238 * for this txg to sync (eg, spa_vdev_remove()), push the
6239 * deferred frees from the previous txg. If not, leave them
6240 * alone so that we don't generate work on an otherwise idle
6243 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
6244 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
6245 !txg_list_empty(&dp
->dp_sync_tasks
, txg
) ||
6246 ((dsl_scan_active(dp
->dp_scan
) ||
6247 txg_sync_waiting(dp
)) && !spa_shutting_down(spa
))) {
6248 spa_sync_deferred_frees(spa
, tx
);
6252 * Iterate to convergence.
6255 int pass
= ++spa
->spa_sync_pass
;
6257 spa_sync_config_object(spa
, tx
);
6258 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6259 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6260 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6261 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6262 spa_errlog_sync(spa
, txg
);
6263 dsl_pool_sync(dp
, txg
);
6265 if (pass
< zfs_sync_pass_deferred_free
) {
6266 spa_sync_frees(spa
, free_bpl
, tx
);
6268 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6269 &spa
->spa_deferred_bpobj
, tx
);
6273 dsl_scan_sync(dp
, tx
);
6275 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6279 spa_sync_upgrades(spa
, tx
);
6281 } while (dmu_objset_is_dirty(mos
, txg
));
6284 * Rewrite the vdev configuration (which includes the uberblock)
6285 * to commit the transaction group.
6287 * If there are no dirty vdevs, we sync the uberblock to a few
6288 * random top-level vdevs that are known to be visible in the
6289 * config cache (see spa_vdev_add() for a complete description).
6290 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6294 * We hold SCL_STATE to prevent vdev open/close/etc.
6295 * while we're attempting to write the vdev labels.
6297 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6299 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6300 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6302 int children
= rvd
->vdev_children
;
6303 int c0
= spa_get_random(children
);
6305 for (c
= 0; c
< children
; c
++) {
6306 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6307 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6309 svd
[svdcount
++] = vd
;
6310 if (svdcount
== SPA_DVAS_PER_BP
)
6313 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
6315 error
= vdev_config_sync(svd
, svdcount
, txg
,
6318 error
= vdev_config_sync(rvd
->vdev_child
,
6319 rvd
->vdev_children
, txg
, B_FALSE
);
6321 error
= vdev_config_sync(rvd
->vdev_child
,
6322 rvd
->vdev_children
, txg
, B_TRUE
);
6326 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6328 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6332 zio_suspend(spa
, NULL
);
6333 zio_resume_wait(spa
);
6337 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6338 spa
->spa_deadman_tqid
= 0;
6341 * Clear the dirty config list.
6343 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6344 vdev_config_clean(vd
);
6347 * Now that the new config has synced transactionally,
6348 * let it become visible to the config cache.
6350 if (spa
->spa_config_syncing
!= NULL
) {
6351 spa_config_set(spa
, spa
->spa_config_syncing
);
6352 spa
->spa_config_txg
= txg
;
6353 spa
->spa_config_syncing
= NULL
;
6356 spa
->spa_ubsync
= spa
->spa_uberblock
;
6358 dsl_pool_sync_done(dp
, txg
);
6361 * Update usable space statistics.
6363 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6364 vdev_sync_done(vd
, txg
);
6366 spa_update_dspace(spa
);
6369 * It had better be the case that we didn't dirty anything
6370 * since vdev_config_sync().
6372 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6373 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6374 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6376 spa
->spa_sync_pass
= 0;
6378 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6380 spa_handle_ignored_writes(spa
);
6383 * If any async tasks have been requested, kick them off.
6385 spa_async_dispatch(spa
);
6389 * Sync all pools. We don't want to hold the namespace lock across these
6390 * operations, so we take a reference on the spa_t and drop the lock during the
6394 spa_sync_allpools(void)
6397 mutex_enter(&spa_namespace_lock
);
6398 while ((spa
= spa_next(spa
)) != NULL
) {
6399 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6400 !spa_writeable(spa
) || spa_suspended(spa
))
6402 spa_open_ref(spa
, FTAG
);
6403 mutex_exit(&spa_namespace_lock
);
6404 txg_wait_synced(spa_get_dsl(spa
), 0);
6405 mutex_enter(&spa_namespace_lock
);
6406 spa_close(spa
, FTAG
);
6408 mutex_exit(&spa_namespace_lock
);
6412 * ==========================================================================
6413 * Miscellaneous routines
6414 * ==========================================================================
6418 * Remove all pools in the system.
6426 * Remove all cached state. All pools should be closed now,
6427 * so every spa in the AVL tree should be unreferenced.
6429 mutex_enter(&spa_namespace_lock
);
6430 while ((spa
= spa_next(NULL
)) != NULL
) {
6432 * Stop async tasks. The async thread may need to detach
6433 * a device that's been replaced, which requires grabbing
6434 * spa_namespace_lock, so we must drop it here.
6436 spa_open_ref(spa
, FTAG
);
6437 mutex_exit(&spa_namespace_lock
);
6438 spa_async_suspend(spa
);
6439 mutex_enter(&spa_namespace_lock
);
6440 spa_close(spa
, FTAG
);
6442 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6444 spa_deactivate(spa
);
6448 mutex_exit(&spa_namespace_lock
);
6452 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6457 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6461 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6462 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6463 if (vd
->vdev_guid
== guid
)
6467 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6468 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6469 if (vd
->vdev_guid
== guid
)
6478 spa_upgrade(spa_t
*spa
, uint64_t version
)
6480 ASSERT(spa_writeable(spa
));
6482 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6485 * This should only be called for a non-faulted pool, and since a
6486 * future version would result in an unopenable pool, this shouldn't be
6489 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
6490 ASSERT(version
>= spa
->spa_uberblock
.ub_version
);
6492 spa
->spa_uberblock
.ub_version
= version
;
6493 vdev_config_dirty(spa
->spa_root_vdev
);
6495 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6497 txg_wait_synced(spa_get_dsl(spa
), 0);
6501 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6505 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6507 for (i
= 0; i
< sav
->sav_count
; i
++)
6508 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6511 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6512 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6513 &spareguid
) == 0 && spareguid
== guid
)
6521 * Check if a pool has an active shared spare device.
6522 * Note: reference count of an active spare is 2, as a spare and as a replace
6525 spa_has_active_shared_spare(spa_t
*spa
)
6529 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6531 for (i
= 0; i
< sav
->sav_count
; i
++) {
6532 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6533 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6542 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6543 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6544 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6545 * or zdb as real changes.
6548 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6551 zfs_ereport_post(name
, spa
, vd
, NULL
, 0, 0);
6555 #if defined(_KERNEL) && defined(HAVE_SPL)
6556 /* state manipulation functions */
6557 EXPORT_SYMBOL(spa_open
);
6558 EXPORT_SYMBOL(spa_open_rewind
);
6559 EXPORT_SYMBOL(spa_get_stats
);
6560 EXPORT_SYMBOL(spa_create
);
6561 EXPORT_SYMBOL(spa_import_rootpool
);
6562 EXPORT_SYMBOL(spa_import
);
6563 EXPORT_SYMBOL(spa_tryimport
);
6564 EXPORT_SYMBOL(spa_destroy
);
6565 EXPORT_SYMBOL(spa_export
);
6566 EXPORT_SYMBOL(spa_reset
);
6567 EXPORT_SYMBOL(spa_async_request
);
6568 EXPORT_SYMBOL(spa_async_suspend
);
6569 EXPORT_SYMBOL(spa_async_resume
);
6570 EXPORT_SYMBOL(spa_inject_addref
);
6571 EXPORT_SYMBOL(spa_inject_delref
);
6572 EXPORT_SYMBOL(spa_scan_stat_init
);
6573 EXPORT_SYMBOL(spa_scan_get_stats
);
6575 /* device maniion */
6576 EXPORT_SYMBOL(spa_vdev_add
);
6577 EXPORT_SYMBOL(spa_vdev_attach
);
6578 EXPORT_SYMBOL(spa_vdev_detach
);
6579 EXPORT_SYMBOL(spa_vdev_remove
);
6580 EXPORT_SYMBOL(spa_vdev_setpath
);
6581 EXPORT_SYMBOL(spa_vdev_setfru
);
6582 EXPORT_SYMBOL(spa_vdev_split_mirror
);
6584 /* spare statech is global across all pools) */
6585 EXPORT_SYMBOL(spa_spare_add
);
6586 EXPORT_SYMBOL(spa_spare_remove
);
6587 EXPORT_SYMBOL(spa_spare_exists
);
6588 EXPORT_SYMBOL(spa_spare_activate
);
6590 /* L2ARC statech is global across all pools) */
6591 EXPORT_SYMBOL(spa_l2cache_add
);
6592 EXPORT_SYMBOL(spa_l2cache_remove
);
6593 EXPORT_SYMBOL(spa_l2cache_exists
);
6594 EXPORT_SYMBOL(spa_l2cache_activate
);
6595 EXPORT_SYMBOL(spa_l2cache_drop
);
6598 EXPORT_SYMBOL(spa_scan
);
6599 EXPORT_SYMBOL(spa_scan_stop
);
6602 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
6603 EXPORT_SYMBOL(spa_sync_allpools
);
6606 EXPORT_SYMBOL(spa_prop_set
);
6607 EXPORT_SYMBOL(spa_prop_get
);
6608 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
6610 /* asynchronous event notification */
6611 EXPORT_SYMBOL(spa_event_notify
);