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
);
1263 * Close the dsl pool.
1265 if (spa
->spa_dsl_pool
) {
1266 dsl_pool_close(spa
->spa_dsl_pool
);
1267 spa
->spa_dsl_pool
= NULL
;
1268 spa
->spa_meta_objset
= NULL
;
1273 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1276 * Drop and purge level 2 cache
1278 spa_l2cache_drop(spa
);
1283 if (spa
->spa_root_vdev
)
1284 vdev_free(spa
->spa_root_vdev
);
1285 ASSERT(spa
->spa_root_vdev
== NULL
);
1287 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1288 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1289 if (spa
->spa_spares
.sav_vdevs
) {
1290 kmem_free(spa
->spa_spares
.sav_vdevs
,
1291 spa
->spa_spares
.sav_count
* sizeof (void *));
1292 spa
->spa_spares
.sav_vdevs
= NULL
;
1294 if (spa
->spa_spares
.sav_config
) {
1295 nvlist_free(spa
->spa_spares
.sav_config
);
1296 spa
->spa_spares
.sav_config
= NULL
;
1298 spa
->spa_spares
.sav_count
= 0;
1300 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1301 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1302 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1304 if (spa
->spa_l2cache
.sav_vdevs
) {
1305 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1306 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1307 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1309 if (spa
->spa_l2cache
.sav_config
) {
1310 nvlist_free(spa
->spa_l2cache
.sav_config
);
1311 spa
->spa_l2cache
.sav_config
= NULL
;
1313 spa
->spa_l2cache
.sav_count
= 0;
1315 spa
->spa_async_suspended
= 0;
1317 if (spa
->spa_comment
!= NULL
) {
1318 spa_strfree(spa
->spa_comment
);
1319 spa
->spa_comment
= NULL
;
1322 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1326 * Load (or re-load) the current list of vdevs describing the active spares for
1327 * this pool. When this is called, we have some form of basic information in
1328 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1329 * then re-generate a more complete list including status information.
1332 spa_load_spares(spa_t
*spa
)
1339 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1342 * First, close and free any existing spare vdevs.
1344 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1345 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1347 /* Undo the call to spa_activate() below */
1348 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1349 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1350 spa_spare_remove(tvd
);
1355 if (spa
->spa_spares
.sav_vdevs
)
1356 kmem_free(spa
->spa_spares
.sav_vdevs
,
1357 spa
->spa_spares
.sav_count
* sizeof (void *));
1359 if (spa
->spa_spares
.sav_config
== NULL
)
1362 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1363 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1365 spa
->spa_spares
.sav_count
= (int)nspares
;
1366 spa
->spa_spares
.sav_vdevs
= NULL
;
1372 * Construct the array of vdevs, opening them to get status in the
1373 * process. For each spare, there is potentially two different vdev_t
1374 * structures associated with it: one in the list of spares (used only
1375 * for basic validation purposes) and one in the active vdev
1376 * configuration (if it's spared in). During this phase we open and
1377 * validate each vdev on the spare list. If the vdev also exists in the
1378 * active configuration, then we also mark this vdev as an active spare.
1380 spa
->spa_spares
.sav_vdevs
= kmem_alloc(nspares
* sizeof (void *),
1382 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1383 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1384 VDEV_ALLOC_SPARE
) == 0);
1387 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1389 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1390 B_FALSE
)) != NULL
) {
1391 if (!tvd
->vdev_isspare
)
1395 * We only mark the spare active if we were successfully
1396 * able to load the vdev. Otherwise, importing a pool
1397 * with a bad active spare would result in strange
1398 * behavior, because multiple pool would think the spare
1399 * is actively in use.
1401 * There is a vulnerability here to an equally bizarre
1402 * circumstance, where a dead active spare is later
1403 * brought back to life (onlined or otherwise). Given
1404 * the rarity of this scenario, and the extra complexity
1405 * it adds, we ignore the possibility.
1407 if (!vdev_is_dead(tvd
))
1408 spa_spare_activate(tvd
);
1412 vd
->vdev_aux
= &spa
->spa_spares
;
1414 if (vdev_open(vd
) != 0)
1417 if (vdev_validate_aux(vd
) == 0)
1422 * Recompute the stashed list of spares, with status information
1425 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1426 DATA_TYPE_NVLIST_ARRAY
) == 0);
1428 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1430 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1431 spares
[i
] = vdev_config_generate(spa
,
1432 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1433 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1434 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1435 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1436 nvlist_free(spares
[i
]);
1437 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1441 * Load (or re-load) the current list of vdevs describing the active l2cache for
1442 * this pool. When this is called, we have some form of basic information in
1443 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1444 * then re-generate a more complete list including status information.
1445 * Devices which are already active have their details maintained, and are
1449 spa_load_l2cache(spa_t
*spa
)
1453 int i
, j
, oldnvdevs
;
1455 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1456 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1458 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1460 if (sav
->sav_config
!= NULL
) {
1461 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1462 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1463 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_PUSHPAGE
);
1469 oldvdevs
= sav
->sav_vdevs
;
1470 oldnvdevs
= sav
->sav_count
;
1471 sav
->sav_vdevs
= NULL
;
1475 * Process new nvlist of vdevs.
1477 for (i
= 0; i
< nl2cache
; i
++) {
1478 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1482 for (j
= 0; j
< oldnvdevs
; j
++) {
1484 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1486 * Retain previous vdev for add/remove ops.
1494 if (newvdevs
[i
] == NULL
) {
1498 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1499 VDEV_ALLOC_L2CACHE
) == 0);
1504 * Commit this vdev as an l2cache device,
1505 * even if it fails to open.
1507 spa_l2cache_add(vd
);
1512 spa_l2cache_activate(vd
);
1514 if (vdev_open(vd
) != 0)
1517 (void) vdev_validate_aux(vd
);
1519 if (!vdev_is_dead(vd
))
1520 l2arc_add_vdev(spa
, vd
);
1525 * Purge vdevs that were dropped
1527 for (i
= 0; i
< oldnvdevs
; i
++) {
1532 ASSERT(vd
->vdev_isl2cache
);
1534 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1535 pool
!= 0ULL && l2arc_vdev_present(vd
))
1536 l2arc_remove_vdev(vd
);
1537 vdev_clear_stats(vd
);
1543 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1545 if (sav
->sav_config
== NULL
)
1548 sav
->sav_vdevs
= newvdevs
;
1549 sav
->sav_count
= (int)nl2cache
;
1552 * Recompute the stashed list of l2cache devices, with status
1553 * information this time.
1555 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1556 DATA_TYPE_NVLIST_ARRAY
) == 0);
1558 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_PUSHPAGE
);
1559 for (i
= 0; i
< sav
->sav_count
; i
++)
1560 l2cache
[i
] = vdev_config_generate(spa
,
1561 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1562 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1563 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1565 for (i
= 0; i
< sav
->sav_count
; i
++)
1566 nvlist_free(l2cache
[i
]);
1568 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1572 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1575 char *packed
= NULL
;
1580 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1584 nvsize
= *(uint64_t *)db
->db_data
;
1585 dmu_buf_rele(db
, FTAG
);
1587 packed
= kmem_alloc(nvsize
, KM_PUSHPAGE
| KM_NODEBUG
);
1588 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1591 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1592 kmem_free(packed
, nvsize
);
1598 * Checks to see if the given vdev could not be opened, in which case we post a
1599 * sysevent to notify the autoreplace code that the device has been removed.
1602 spa_check_removed(vdev_t
*vd
)
1606 for (c
= 0; c
< vd
->vdev_children
; c
++)
1607 spa_check_removed(vd
->vdev_child
[c
]);
1609 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1611 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE
,
1612 vd
->vdev_spa
, vd
, NULL
, 0, 0);
1613 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_CHECK
);
1618 * Validate the current config against the MOS config
1621 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1623 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1627 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1629 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1630 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1632 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1635 * If we're doing a normal import, then build up any additional
1636 * diagnostic information about missing devices in this config.
1637 * We'll pass this up to the user for further processing.
1639 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1640 nvlist_t
**child
, *nv
;
1643 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1645 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
1647 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1648 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1649 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1651 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1652 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1654 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1659 VERIFY(nvlist_add_nvlist_array(nv
,
1660 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1661 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1662 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1664 for (i
= 0; i
< idx
; i
++)
1665 nvlist_free(child
[i
]);
1668 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1672 * Compare the root vdev tree with the information we have
1673 * from the MOS config (mrvd). Check each top-level vdev
1674 * with the corresponding MOS config top-level (mtvd).
1676 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1677 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1678 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1681 * Resolve any "missing" vdevs in the current configuration.
1682 * If we find that the MOS config has more accurate information
1683 * about the top-level vdev then use that vdev instead.
1685 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1686 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1688 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1692 * Device specific actions.
1694 if (mtvd
->vdev_islog
) {
1695 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1698 * XXX - once we have 'readonly' pool
1699 * support we should be able to handle
1700 * missing data devices by transitioning
1701 * the pool to readonly.
1707 * Swap the missing vdev with the data we were
1708 * able to obtain from the MOS config.
1710 vdev_remove_child(rvd
, tvd
);
1711 vdev_remove_child(mrvd
, mtvd
);
1713 vdev_add_child(rvd
, mtvd
);
1714 vdev_add_child(mrvd
, tvd
);
1716 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1718 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1721 } else if (mtvd
->vdev_islog
) {
1723 * Load the slog device's state from the MOS config
1724 * since it's possible that the label does not
1725 * contain the most up-to-date information.
1727 vdev_load_log_state(tvd
, mtvd
);
1732 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1735 * Ensure we were able to validate the config.
1737 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1741 * Check for missing log devices
1744 spa_check_logs(spa_t
*spa
)
1746 boolean_t rv
= B_FALSE
;
1748 switch (spa
->spa_log_state
) {
1751 case SPA_LOG_MISSING
:
1752 /* need to recheck in case slog has been restored */
1753 case SPA_LOG_UNKNOWN
:
1754 rv
= (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
,
1755 NULL
, DS_FIND_CHILDREN
) != 0);
1757 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1764 spa_passivate_log(spa_t
*spa
)
1766 vdev_t
*rvd
= spa
->spa_root_vdev
;
1767 boolean_t slog_found
= B_FALSE
;
1770 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1772 if (!spa_has_slogs(spa
))
1775 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1776 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1777 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1779 if (tvd
->vdev_islog
) {
1780 metaslab_group_passivate(mg
);
1781 slog_found
= B_TRUE
;
1785 return (slog_found
);
1789 spa_activate_log(spa_t
*spa
)
1791 vdev_t
*rvd
= spa
->spa_root_vdev
;
1794 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1796 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1797 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1798 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1800 if (tvd
->vdev_islog
)
1801 metaslab_group_activate(mg
);
1806 spa_offline_log(spa_t
*spa
)
1810 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1811 NULL
, DS_FIND_CHILDREN
);
1814 * We successfully offlined the log device, sync out the
1815 * current txg so that the "stubby" block can be removed
1818 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1824 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1828 for (i
= 0; i
< sav
->sav_count
; i
++)
1829 spa_check_removed(sav
->sav_vdevs
[i
]);
1833 spa_claim_notify(zio_t
*zio
)
1835 spa_t
*spa
= zio
->io_spa
;
1840 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1841 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1842 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1843 mutex_exit(&spa
->spa_props_lock
);
1846 typedef struct spa_load_error
{
1847 uint64_t sle_meta_count
;
1848 uint64_t sle_data_count
;
1852 spa_load_verify_done(zio_t
*zio
)
1854 blkptr_t
*bp
= zio
->io_bp
;
1855 spa_load_error_t
*sle
= zio
->io_private
;
1856 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1857 int error
= zio
->io_error
;
1860 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1861 type
!= DMU_OT_INTENT_LOG
)
1862 atomic_add_64(&sle
->sle_meta_count
, 1);
1864 atomic_add_64(&sle
->sle_data_count
, 1);
1866 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1871 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1872 const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1876 size_t size
= BP_GET_PSIZE(bp
);
1877 void *data
= zio_data_buf_alloc(size
);
1879 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1880 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1881 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1882 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1888 spa_load_verify(spa_t
*spa
)
1891 spa_load_error_t sle
= { 0 };
1892 zpool_rewind_policy_t policy
;
1893 boolean_t verify_ok
= B_FALSE
;
1896 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1898 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1901 rio
= zio_root(spa
, NULL
, &sle
,
1902 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1904 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1905 TRAVERSE_PRE
| TRAVERSE_PREFETCH
, spa_load_verify_cb
, rio
);
1907 (void) zio_wait(rio
);
1909 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1910 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1912 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1913 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1917 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1918 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1920 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1921 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1922 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1923 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1924 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1925 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1926 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1928 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1932 if (error
!= ENXIO
&& error
!= EIO
)
1933 error
= SET_ERROR(EIO
);
1937 return (verify_ok
? 0 : EIO
);
1941 * Find a value in the pool props object.
1944 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
1946 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
1947 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
1951 * Find a value in the pool directory object.
1954 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
1956 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1957 name
, sizeof (uint64_t), 1, val
));
1961 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
1963 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
1968 * Fix up config after a partly-completed split. This is done with the
1969 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1970 * pool have that entry in their config, but only the splitting one contains
1971 * a list of all the guids of the vdevs that are being split off.
1973 * This function determines what to do with that list: either rejoin
1974 * all the disks to the pool, or complete the splitting process. To attempt
1975 * the rejoin, each disk that is offlined is marked online again, and
1976 * we do a reopen() call. If the vdev label for every disk that was
1977 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1978 * then we call vdev_split() on each disk, and complete the split.
1980 * Otherwise we leave the config alone, with all the vdevs in place in
1981 * the original pool.
1984 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
1991 boolean_t attempt_reopen
;
1993 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
1996 /* check that the config is complete */
1997 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
1998 &glist
, &gcount
) != 0)
2001 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_PUSHPAGE
);
2003 /* attempt to online all the vdevs & validate */
2004 attempt_reopen
= B_TRUE
;
2005 for (i
= 0; i
< gcount
; i
++) {
2006 if (glist
[i
] == 0) /* vdev is hole */
2009 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2010 if (vd
[i
] == NULL
) {
2012 * Don't bother attempting to reopen the disks;
2013 * just do the split.
2015 attempt_reopen
= B_FALSE
;
2017 /* attempt to re-online it */
2018 vd
[i
]->vdev_offline
= B_FALSE
;
2022 if (attempt_reopen
) {
2023 vdev_reopen(spa
->spa_root_vdev
);
2025 /* check each device to see what state it's in */
2026 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2027 if (vd
[i
] != NULL
&&
2028 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2035 * If every disk has been moved to the new pool, or if we never
2036 * even attempted to look at them, then we split them off for
2039 if (!attempt_reopen
|| gcount
== extracted
) {
2040 for (i
= 0; i
< gcount
; i
++)
2043 vdev_reopen(spa
->spa_root_vdev
);
2046 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2050 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2051 boolean_t mosconfig
)
2053 nvlist_t
*config
= spa
->spa_config
;
2054 char *ereport
= FM_EREPORT_ZFS_POOL
;
2060 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2061 return (SET_ERROR(EINVAL
));
2063 ASSERT(spa
->spa_comment
== NULL
);
2064 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2065 spa
->spa_comment
= spa_strdup(comment
);
2068 * Versioning wasn't explicitly added to the label until later, so if
2069 * it's not present treat it as the initial version.
2071 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2072 &spa
->spa_ubsync
.ub_version
) != 0)
2073 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2075 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2076 &spa
->spa_config_txg
);
2078 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2079 spa_guid_exists(pool_guid
, 0)) {
2080 error
= SET_ERROR(EEXIST
);
2082 spa
->spa_config_guid
= pool_guid
;
2084 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2086 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2090 nvlist_free(spa
->spa_load_info
);
2091 spa
->spa_load_info
= fnvlist_alloc();
2093 gethrestime(&spa
->spa_loaded_ts
);
2094 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2095 mosconfig
, &ereport
);
2098 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2100 if (error
!= EEXIST
) {
2101 spa
->spa_loaded_ts
.tv_sec
= 0;
2102 spa
->spa_loaded_ts
.tv_nsec
= 0;
2104 if (error
!= EBADF
) {
2105 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2108 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2115 * Load an existing storage pool, using the pool's builtin spa_config as a
2116 * source of configuration information.
2118 __attribute__((always_inline
))
2120 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2121 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2125 nvlist_t
*nvroot
= NULL
;
2128 uberblock_t
*ub
= &spa
->spa_uberblock
;
2129 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2130 int orig_mode
= spa
->spa_mode
;
2133 boolean_t missing_feat_write
= B_FALSE
;
2136 * If this is an untrusted config, access the pool in read-only mode.
2137 * This prevents things like resilvering recently removed devices.
2140 spa
->spa_mode
= FREAD
;
2142 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2144 spa
->spa_load_state
= state
;
2146 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2147 return (SET_ERROR(EINVAL
));
2149 parse
= (type
== SPA_IMPORT_EXISTING
?
2150 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2153 * Create "The Godfather" zio to hold all async IOs
2155 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
2156 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
2159 * Parse the configuration into a vdev tree. We explicitly set the
2160 * value that will be returned by spa_version() since parsing the
2161 * configuration requires knowing the version number.
2163 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2164 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2165 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2170 ASSERT(spa
->spa_root_vdev
== rvd
);
2172 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2173 ASSERT(spa_guid(spa
) == pool_guid
);
2177 * Try to open all vdevs, loading each label in the process.
2179 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2180 error
= vdev_open(rvd
);
2181 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2186 * We need to validate the vdev labels against the configuration that
2187 * we have in hand, which is dependent on the setting of mosconfig. If
2188 * mosconfig is true then we're validating the vdev labels based on
2189 * that config. Otherwise, we're validating against the cached config
2190 * (zpool.cache) that was read when we loaded the zfs module, and then
2191 * later we will recursively call spa_load() and validate against
2194 * If we're assembling a new pool that's been split off from an
2195 * existing pool, the labels haven't yet been updated so we skip
2196 * validation for now.
2198 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2199 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2200 error
= vdev_validate(rvd
, mosconfig
);
2201 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2206 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2207 return (SET_ERROR(ENXIO
));
2211 * Find the best uberblock.
2213 vdev_uberblock_load(rvd
, ub
, &label
);
2216 * If we weren't able to find a single valid uberblock, return failure.
2218 if (ub
->ub_txg
== 0) {
2220 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2224 * If the pool has an unsupported version we can't open it.
2226 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2228 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2231 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2235 * If we weren't able to find what's necessary for reading the
2236 * MOS in the label, return failure.
2238 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2239 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2241 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2246 * Update our in-core representation with the definitive values
2249 nvlist_free(spa
->spa_label_features
);
2250 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2256 * Look through entries in the label nvlist's features_for_read. If
2257 * there is a feature listed there which we don't understand then we
2258 * cannot open a pool.
2260 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2261 nvlist_t
*unsup_feat
;
2264 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2267 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2269 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2270 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2271 VERIFY(nvlist_add_string(unsup_feat
,
2272 nvpair_name(nvp
), "") == 0);
2276 if (!nvlist_empty(unsup_feat
)) {
2277 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2278 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2279 nvlist_free(unsup_feat
);
2280 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2284 nvlist_free(unsup_feat
);
2288 * If the vdev guid sum doesn't match the uberblock, we have an
2289 * incomplete configuration. We first check to see if the pool
2290 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2291 * If it is, defer the vdev_guid_sum check till later so we
2292 * can handle missing vdevs.
2294 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2295 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2296 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2297 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2299 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2300 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2301 spa_try_repair(spa
, config
);
2302 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2303 nvlist_free(spa
->spa_config_splitting
);
2304 spa
->spa_config_splitting
= NULL
;
2308 * Initialize internal SPA structures.
2310 spa
->spa_state
= POOL_STATE_ACTIVE
;
2311 spa
->spa_ubsync
= spa
->spa_uberblock
;
2312 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2313 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2314 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2315 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2316 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2317 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2319 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2321 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2322 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2324 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2325 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2327 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2328 boolean_t missing_feat_read
= B_FALSE
;
2329 nvlist_t
*unsup_feat
, *enabled_feat
;
2331 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2332 &spa
->spa_feat_for_read_obj
) != 0) {
2333 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2336 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2337 &spa
->spa_feat_for_write_obj
) != 0) {
2338 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2341 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2342 &spa
->spa_feat_desc_obj
) != 0) {
2343 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2346 enabled_feat
= fnvlist_alloc();
2347 unsup_feat
= fnvlist_alloc();
2349 if (!feature_is_supported(spa
->spa_meta_objset
,
2350 spa
->spa_feat_for_read_obj
, spa
->spa_feat_desc_obj
,
2351 unsup_feat
, enabled_feat
))
2352 missing_feat_read
= B_TRUE
;
2354 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2355 if (!feature_is_supported(spa
->spa_meta_objset
,
2356 spa
->spa_feat_for_write_obj
, spa
->spa_feat_desc_obj
,
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 spa
->spa_is_initializing
= B_TRUE
;
2405 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2406 spa
->spa_is_initializing
= B_FALSE
;
2408 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2412 nvlist_t
*policy
= NULL
, *nvconfig
;
2414 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2415 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2417 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2418 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2420 unsigned long myhostid
= 0;
2422 VERIFY(nvlist_lookup_string(nvconfig
,
2423 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2426 myhostid
= zone_get_hostid(NULL
);
2429 * We're emulating the system's hostid in userland, so
2430 * we can't use zone_get_hostid().
2432 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2433 #endif /* _KERNEL */
2434 if (hostid
!= 0 && myhostid
!= 0 &&
2435 hostid
!= myhostid
) {
2436 nvlist_free(nvconfig
);
2437 cmn_err(CE_WARN
, "pool '%s' could not be "
2438 "loaded as it was last accessed by another "
2439 "system (host: %s hostid: 0x%lx). See: "
2440 "http://zfsonlinux.org/msg/ZFS-8000-EY",
2441 spa_name(spa
), hostname
,
2442 (unsigned long)hostid
);
2443 return (SET_ERROR(EBADF
));
2446 if (nvlist_lookup_nvlist(spa
->spa_config
,
2447 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2448 VERIFY(nvlist_add_nvlist(nvconfig
,
2449 ZPOOL_REWIND_POLICY
, policy
) == 0);
2451 spa_config_set(spa
, nvconfig
);
2453 spa_deactivate(spa
);
2454 spa_activate(spa
, orig_mode
);
2456 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2459 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2460 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2461 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2463 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2466 * Load the bit that tells us to use the new accounting function
2467 * (raid-z deflation). If we have an older pool, this will not
2470 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2471 if (error
!= 0 && error
!= ENOENT
)
2472 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2474 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2475 &spa
->spa_creation_version
);
2476 if (error
!= 0 && error
!= ENOENT
)
2477 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2480 * Load the persistent error log. If we have an older pool, this will
2483 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2484 if (error
!= 0 && error
!= ENOENT
)
2485 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2487 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2488 &spa
->spa_errlog_scrub
);
2489 if (error
!= 0 && error
!= ENOENT
)
2490 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2493 * Load the history object. If we have an older pool, this
2494 * will not be present.
2496 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2497 if (error
!= 0 && error
!= ENOENT
)
2498 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2501 * If we're assembling the pool from the split-off vdevs of
2502 * an existing pool, we don't want to attach the spares & cache
2507 * Load any hot spares for this pool.
2509 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2510 if (error
!= 0 && error
!= ENOENT
)
2511 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2512 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2513 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2514 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2515 &spa
->spa_spares
.sav_config
) != 0)
2516 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2518 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2519 spa_load_spares(spa
);
2520 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2521 } else if (error
== 0) {
2522 spa
->spa_spares
.sav_sync
= B_TRUE
;
2526 * Load any level 2 ARC devices for this pool.
2528 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2529 &spa
->spa_l2cache
.sav_object
);
2530 if (error
!= 0 && error
!= ENOENT
)
2531 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2532 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2533 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2534 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2535 &spa
->spa_l2cache
.sav_config
) != 0)
2536 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2538 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2539 spa_load_l2cache(spa
);
2540 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2541 } else if (error
== 0) {
2542 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2545 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2547 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2548 if (error
&& error
!= ENOENT
)
2549 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2552 uint64_t autoreplace
;
2554 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2555 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2556 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2557 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2558 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2559 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2560 &spa
->spa_dedup_ditto
);
2562 spa
->spa_autoreplace
= (autoreplace
!= 0);
2566 * If the 'autoreplace' property is set, then post a resource notifying
2567 * the ZFS DE that it should not issue any faults for unopenable
2568 * devices. We also iterate over the vdevs, and post a sysevent for any
2569 * unopenable vdevs so that the normal autoreplace handler can take
2572 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2573 spa_check_removed(spa
->spa_root_vdev
);
2575 * For the import case, this is done in spa_import(), because
2576 * at this point we're using the spare definitions from
2577 * the MOS config, not necessarily from the userland config.
2579 if (state
!= SPA_LOAD_IMPORT
) {
2580 spa_aux_check_removed(&spa
->spa_spares
);
2581 spa_aux_check_removed(&spa
->spa_l2cache
);
2586 * Load the vdev state for all toplevel vdevs.
2591 * Propagate the leaf DTLs we just loaded all the way up the tree.
2593 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2594 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2595 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2598 * Load the DDTs (dedup tables).
2600 error
= ddt_load(spa
);
2602 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2604 spa_update_dspace(spa
);
2607 * Validate the config, using the MOS config to fill in any
2608 * information which might be missing. If we fail to validate
2609 * the config then declare the pool unfit for use. If we're
2610 * assembling a pool from a split, the log is not transferred
2613 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2616 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2617 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2619 if (!spa_config_valid(spa
, nvconfig
)) {
2620 nvlist_free(nvconfig
);
2621 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2624 nvlist_free(nvconfig
);
2627 * Now that we've validated the config, check the state of the
2628 * root vdev. If it can't be opened, it indicates one or
2629 * more toplevel vdevs are faulted.
2631 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2632 return (SET_ERROR(ENXIO
));
2634 if (spa_check_logs(spa
)) {
2635 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2636 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2640 if (missing_feat_write
) {
2641 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2644 * At this point, we know that we can open the pool in
2645 * read-only mode but not read-write mode. We now have enough
2646 * information and can return to userland.
2648 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2652 * We've successfully opened the pool, verify that we're ready
2653 * to start pushing transactions.
2655 if (state
!= SPA_LOAD_TRYIMPORT
) {
2656 if ((error
= spa_load_verify(spa
)))
2657 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2661 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2662 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2664 int need_update
= B_FALSE
;
2667 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2670 * Claim log blocks that haven't been committed yet.
2671 * This must all happen in a single txg.
2672 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2673 * invoked from zil_claim_log_block()'s i/o done callback.
2674 * Price of rollback is that we abandon the log.
2676 spa
->spa_claiming
= B_TRUE
;
2678 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
2679 spa_first_txg(spa
));
2680 (void) dmu_objset_find(spa_name(spa
),
2681 zil_claim
, tx
, DS_FIND_CHILDREN
);
2684 spa
->spa_claiming
= B_FALSE
;
2686 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2687 spa
->spa_sync_on
= B_TRUE
;
2688 txg_sync_start(spa
->spa_dsl_pool
);
2691 * Wait for all claims to sync. We sync up to the highest
2692 * claimed log block birth time so that claimed log blocks
2693 * don't appear to be from the future. spa_claim_max_txg
2694 * will have been set for us by either zil_check_log_chain()
2695 * (invoked from spa_check_logs()) or zil_claim() above.
2697 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2700 * If the config cache is stale, or we have uninitialized
2701 * metaslabs (see spa_vdev_add()), then update the config.
2703 * If this is a verbatim import, trust the current
2704 * in-core spa_config and update the disk labels.
2706 if (config_cache_txg
!= spa
->spa_config_txg
||
2707 state
== SPA_LOAD_IMPORT
||
2708 state
== SPA_LOAD_RECOVER
||
2709 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2710 need_update
= B_TRUE
;
2712 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2713 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2714 need_update
= B_TRUE
;
2717 * Update the config cache asychronously in case we're the
2718 * root pool, in which case the config cache isn't writable yet.
2721 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2724 * Check all DTLs to see if anything needs resilvering.
2726 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2727 vdev_resilver_needed(rvd
, NULL
, NULL
))
2728 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2731 * Log the fact that we booted up (so that we can detect if
2732 * we rebooted in the middle of an operation).
2734 spa_history_log_version(spa
, "open");
2737 * Delete any inconsistent datasets.
2739 (void) dmu_objset_find(spa_name(spa
),
2740 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2743 * Clean up any stale temporary dataset userrefs.
2745 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2752 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2754 int mode
= spa
->spa_mode
;
2757 spa_deactivate(spa
);
2759 spa
->spa_load_max_txg
--;
2761 spa_activate(spa
, mode
);
2762 spa_async_suspend(spa
);
2764 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2768 * If spa_load() fails this function will try loading prior txg's. If
2769 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2770 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2771 * function will not rewind the pool and will return the same error as
2775 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2776 uint64_t max_request
, int rewind_flags
)
2778 nvlist_t
*loadinfo
= NULL
;
2779 nvlist_t
*config
= NULL
;
2780 int load_error
, rewind_error
;
2781 uint64_t safe_rewind_txg
;
2784 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2785 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2786 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2788 spa
->spa_load_max_txg
= max_request
;
2791 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2793 if (load_error
== 0)
2796 if (spa
->spa_root_vdev
!= NULL
)
2797 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2799 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2800 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2802 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2803 nvlist_free(config
);
2804 return (load_error
);
2807 if (state
== SPA_LOAD_RECOVER
) {
2808 /* Price of rolling back is discarding txgs, including log */
2809 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2812 * If we aren't rolling back save the load info from our first
2813 * import attempt so that we can restore it after attempting
2816 loadinfo
= spa
->spa_load_info
;
2817 spa
->spa_load_info
= fnvlist_alloc();
2820 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2821 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2822 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2823 TXG_INITIAL
: safe_rewind_txg
;
2826 * Continue as long as we're finding errors, we're still within
2827 * the acceptable rewind range, and we're still finding uberblocks
2829 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2830 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2831 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2832 spa
->spa_extreme_rewind
= B_TRUE
;
2833 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2836 spa
->spa_extreme_rewind
= B_FALSE
;
2837 spa
->spa_load_max_txg
= UINT64_MAX
;
2839 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2840 spa_config_set(spa
, config
);
2842 if (state
== SPA_LOAD_RECOVER
) {
2843 ASSERT3P(loadinfo
, ==, NULL
);
2844 return (rewind_error
);
2846 /* Store the rewind info as part of the initial load info */
2847 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
2848 spa
->spa_load_info
);
2850 /* Restore the initial load info */
2851 fnvlist_free(spa
->spa_load_info
);
2852 spa
->spa_load_info
= loadinfo
;
2854 return (load_error
);
2861 * The import case is identical to an open except that the configuration is sent
2862 * down from userland, instead of grabbed from the configuration cache. For the
2863 * case of an open, the pool configuration will exist in the
2864 * POOL_STATE_UNINITIALIZED state.
2866 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2867 * the same time open the pool, without having to keep around the spa_t in some
2871 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2875 spa_load_state_t state
= SPA_LOAD_OPEN
;
2877 int locked
= B_FALSE
;
2878 int firstopen
= B_FALSE
;
2883 * As disgusting as this is, we need to support recursive calls to this
2884 * function because dsl_dir_open() is called during spa_load(), and ends
2885 * up calling spa_open() again. The real fix is to figure out how to
2886 * avoid dsl_dir_open() calling this in the first place.
2888 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2889 mutex_enter(&spa_namespace_lock
);
2893 if ((spa
= spa_lookup(pool
)) == NULL
) {
2895 mutex_exit(&spa_namespace_lock
);
2896 return (SET_ERROR(ENOENT
));
2899 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
2900 zpool_rewind_policy_t policy
;
2904 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
2906 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
2907 state
= SPA_LOAD_RECOVER
;
2909 spa_activate(spa
, spa_mode_global
);
2911 if (state
!= SPA_LOAD_RECOVER
)
2912 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
2914 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
2915 policy
.zrp_request
);
2917 if (error
== EBADF
) {
2919 * If vdev_validate() returns failure (indicated by
2920 * EBADF), it indicates that one of the vdevs indicates
2921 * that the pool has been exported or destroyed. If
2922 * this is the case, the config cache is out of sync and
2923 * we should remove the pool from the namespace.
2926 spa_deactivate(spa
);
2927 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
2930 mutex_exit(&spa_namespace_lock
);
2931 return (SET_ERROR(ENOENT
));
2936 * We can't open the pool, but we still have useful
2937 * information: the state of each vdev after the
2938 * attempted vdev_open(). Return this to the user.
2940 if (config
!= NULL
&& spa
->spa_config
) {
2941 VERIFY(nvlist_dup(spa
->spa_config
, config
,
2943 VERIFY(nvlist_add_nvlist(*config
,
2944 ZPOOL_CONFIG_LOAD_INFO
,
2945 spa
->spa_load_info
) == 0);
2948 spa_deactivate(spa
);
2949 spa
->spa_last_open_failed
= error
;
2951 mutex_exit(&spa_namespace_lock
);
2957 spa_open_ref(spa
, tag
);
2960 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2963 * If we've recovered the pool, pass back any information we
2964 * gathered while doing the load.
2966 if (state
== SPA_LOAD_RECOVER
) {
2967 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
2968 spa
->spa_load_info
) == 0);
2972 spa
->spa_last_open_failed
= 0;
2973 spa
->spa_last_ubsync_txg
= 0;
2974 spa
->spa_load_txg
= 0;
2975 mutex_exit(&spa_namespace_lock
);
2980 zvol_create_minors(spa
->spa_name
);
2989 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
2992 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
2996 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
2998 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3002 * Lookup the given spa_t, incrementing the inject count in the process,
3003 * preventing it from being exported or destroyed.
3006 spa_inject_addref(char *name
)
3010 mutex_enter(&spa_namespace_lock
);
3011 if ((spa
= spa_lookup(name
)) == NULL
) {
3012 mutex_exit(&spa_namespace_lock
);
3015 spa
->spa_inject_ref
++;
3016 mutex_exit(&spa_namespace_lock
);
3022 spa_inject_delref(spa_t
*spa
)
3024 mutex_enter(&spa_namespace_lock
);
3025 spa
->spa_inject_ref
--;
3026 mutex_exit(&spa_namespace_lock
);
3030 * Add spares device information to the nvlist.
3033 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3043 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3045 if (spa
->spa_spares
.sav_count
== 0)
3048 VERIFY(nvlist_lookup_nvlist(config
,
3049 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3050 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3051 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3053 VERIFY(nvlist_add_nvlist_array(nvroot
,
3054 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3055 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3056 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3059 * Go through and find any spares which have since been
3060 * repurposed as an active spare. If this is the case, update
3061 * their status appropriately.
3063 for (i
= 0; i
< nspares
; i
++) {
3064 VERIFY(nvlist_lookup_uint64(spares
[i
],
3065 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3066 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3068 VERIFY(nvlist_lookup_uint64_array(
3069 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3070 (uint64_t **)&vs
, &vsc
) == 0);
3071 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3072 vs
->vs_aux
= VDEV_AUX_SPARED
;
3079 * Add l2cache device information to the nvlist, including vdev stats.
3082 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3085 uint_t i
, j
, nl2cache
;
3092 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3094 if (spa
->spa_l2cache
.sav_count
== 0)
3097 VERIFY(nvlist_lookup_nvlist(config
,
3098 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3099 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3100 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3101 if (nl2cache
!= 0) {
3102 VERIFY(nvlist_add_nvlist_array(nvroot
,
3103 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3104 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3105 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3108 * Update level 2 cache device stats.
3111 for (i
= 0; i
< nl2cache
; i
++) {
3112 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3113 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3116 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3118 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3119 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3125 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3126 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3128 vdev_get_stats(vd
, vs
);
3134 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3140 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3141 VERIFY(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3143 if (spa
->spa_feat_for_read_obj
!= 0) {
3144 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3145 spa
->spa_feat_for_read_obj
);
3146 zap_cursor_retrieve(&zc
, &za
) == 0;
3147 zap_cursor_advance(&zc
)) {
3148 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3149 za
.za_num_integers
== 1);
3150 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
3151 za
.za_first_integer
));
3153 zap_cursor_fini(&zc
);
3156 if (spa
->spa_feat_for_write_obj
!= 0) {
3157 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3158 spa
->spa_feat_for_write_obj
);
3159 zap_cursor_retrieve(&zc
, &za
) == 0;
3160 zap_cursor_advance(&zc
)) {
3161 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3162 za
.za_num_integers
== 1);
3163 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
3164 za
.za_first_integer
));
3166 zap_cursor_fini(&zc
);
3169 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3171 nvlist_free(features
);
3175 spa_get_stats(const char *name
, nvlist_t
**config
,
3176 char *altroot
, size_t buflen
)
3182 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3186 * This still leaves a window of inconsistency where the spares
3187 * or l2cache devices could change and the config would be
3188 * self-inconsistent.
3190 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3192 if (*config
!= NULL
) {
3193 uint64_t loadtimes
[2];
3195 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3196 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3197 VERIFY(nvlist_add_uint64_array(*config
,
3198 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3200 VERIFY(nvlist_add_uint64(*config
,
3201 ZPOOL_CONFIG_ERRCOUNT
,
3202 spa_get_errlog_size(spa
)) == 0);
3204 if (spa_suspended(spa
))
3205 VERIFY(nvlist_add_uint64(*config
,
3206 ZPOOL_CONFIG_SUSPENDED
,
3207 spa
->spa_failmode
) == 0);
3209 spa_add_spares(spa
, *config
);
3210 spa_add_l2cache(spa
, *config
);
3211 spa_add_feature_stats(spa
, *config
);
3216 * We want to get the alternate root even for faulted pools, so we cheat
3217 * and call spa_lookup() directly.
3221 mutex_enter(&spa_namespace_lock
);
3222 spa
= spa_lookup(name
);
3224 spa_altroot(spa
, altroot
, buflen
);
3228 mutex_exit(&spa_namespace_lock
);
3230 spa_altroot(spa
, altroot
, buflen
);
3235 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3236 spa_close(spa
, FTAG
);
3243 * Validate that the auxiliary device array is well formed. We must have an
3244 * array of nvlists, each which describes a valid leaf vdev. If this is an
3245 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3246 * specified, as long as they are well-formed.
3249 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3250 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3251 vdev_labeltype_t label
)
3258 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3261 * It's acceptable to have no devs specified.
3263 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3267 return (SET_ERROR(EINVAL
));
3270 * Make sure the pool is formatted with a version that supports this
3273 if (spa_version(spa
) < version
)
3274 return (SET_ERROR(ENOTSUP
));
3277 * Set the pending device list so we correctly handle device in-use
3280 sav
->sav_pending
= dev
;
3281 sav
->sav_npending
= ndev
;
3283 for (i
= 0; i
< ndev
; i
++) {
3284 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3288 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3290 error
= SET_ERROR(EINVAL
);
3295 * The L2ARC currently only supports disk devices in
3296 * kernel context. For user-level testing, we allow it.
3299 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3300 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3301 error
= SET_ERROR(ENOTBLK
);
3308 if ((error
= vdev_open(vd
)) == 0 &&
3309 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3310 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3311 vd
->vdev_guid
) == 0);
3317 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3324 sav
->sav_pending
= NULL
;
3325 sav
->sav_npending
= 0;
3330 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3334 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3336 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3337 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3338 VDEV_LABEL_SPARE
)) != 0) {
3342 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3343 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3344 VDEV_LABEL_L2CACHE
));
3348 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3353 if (sav
->sav_config
!= NULL
) {
3359 * Generate new dev list by concatentating with the
3362 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3363 &olddevs
, &oldndevs
) == 0);
3365 newdevs
= kmem_alloc(sizeof (void *) *
3366 (ndevs
+ oldndevs
), KM_PUSHPAGE
);
3367 for (i
= 0; i
< oldndevs
; i
++)
3368 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3370 for (i
= 0; i
< ndevs
; i
++)
3371 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3374 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3375 DATA_TYPE_NVLIST_ARRAY
) == 0);
3377 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3378 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3379 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3380 nvlist_free(newdevs
[i
]);
3381 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3384 * Generate a new dev list.
3386 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3388 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3394 * Stop and drop level 2 ARC devices
3397 spa_l2cache_drop(spa_t
*spa
)
3401 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3403 for (i
= 0; i
< sav
->sav_count
; i
++) {
3406 vd
= sav
->sav_vdevs
[i
];
3409 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3410 pool
!= 0ULL && l2arc_vdev_present(vd
))
3411 l2arc_remove_vdev(vd
);
3419 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3423 char *altroot
= NULL
;
3428 uint64_t txg
= TXG_INITIAL
;
3429 nvlist_t
**spares
, **l2cache
;
3430 uint_t nspares
, nl2cache
;
3431 uint64_t version
, obj
;
3432 boolean_t has_features
;
3437 * If this pool already exists, return failure.
3439 mutex_enter(&spa_namespace_lock
);
3440 if (spa_lookup(pool
) != NULL
) {
3441 mutex_exit(&spa_namespace_lock
);
3442 return (SET_ERROR(EEXIST
));
3446 * Allocate a new spa_t structure.
3448 (void) nvlist_lookup_string(props
,
3449 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3450 spa
= spa_add(pool
, NULL
, altroot
);
3451 spa_activate(spa
, spa_mode_global
);
3453 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3454 spa_deactivate(spa
);
3456 mutex_exit(&spa_namespace_lock
);
3460 has_features
= B_FALSE
;
3461 for (elem
= nvlist_next_nvpair(props
, NULL
);
3462 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3463 if (zpool_prop_feature(nvpair_name(elem
)))
3464 has_features
= B_TRUE
;
3467 if (has_features
|| nvlist_lookup_uint64(props
,
3468 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3469 version
= SPA_VERSION
;
3471 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3473 spa
->spa_first_txg
= txg
;
3474 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3475 spa
->spa_uberblock
.ub_version
= version
;
3476 spa
->spa_ubsync
= spa
->spa_uberblock
;
3479 * Create "The Godfather" zio to hold all async IOs
3481 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
3482 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
3485 * Create the root vdev.
3487 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3489 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3491 ASSERT(error
!= 0 || rvd
!= NULL
);
3492 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3494 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3495 error
= SET_ERROR(EINVAL
);
3498 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3499 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3500 VDEV_ALLOC_ADD
)) == 0) {
3501 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
3502 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3503 vdev_expand(rvd
->vdev_child
[c
], txg
);
3507 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3511 spa_deactivate(spa
);
3513 mutex_exit(&spa_namespace_lock
);
3518 * Get the list of spares, if specified.
3520 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3521 &spares
, &nspares
) == 0) {
3522 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3524 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3525 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3526 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3527 spa_load_spares(spa
);
3528 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3529 spa
->spa_spares
.sav_sync
= B_TRUE
;
3533 * Get the list of level 2 cache devices, if specified.
3535 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3536 &l2cache
, &nl2cache
) == 0) {
3537 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3538 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3539 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3540 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3541 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3542 spa_load_l2cache(spa
);
3543 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3544 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3547 spa
->spa_is_initializing
= B_TRUE
;
3548 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3549 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3550 spa
->spa_is_initializing
= B_FALSE
;
3553 * Create DDTs (dedup tables).
3557 spa_update_dspace(spa
);
3559 tx
= dmu_tx_create_assigned(dp
, txg
);
3562 * Create the pool config object.
3564 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3565 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3566 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3568 if (zap_add(spa
->spa_meta_objset
,
3569 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3570 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3571 cmn_err(CE_PANIC
, "failed to add pool config");
3574 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3575 spa_feature_create_zap_objects(spa
, tx
);
3577 if (zap_add(spa
->spa_meta_objset
,
3578 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3579 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3580 cmn_err(CE_PANIC
, "failed to add pool version");
3583 /* Newly created pools with the right version are always deflated. */
3584 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3585 spa
->spa_deflate
= TRUE
;
3586 if (zap_add(spa
->spa_meta_objset
,
3587 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3588 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3589 cmn_err(CE_PANIC
, "failed to add deflate");
3594 * Create the deferred-free bpobj. Turn off compression
3595 * because sync-to-convergence takes longer if the blocksize
3598 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3599 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3600 ZIO_COMPRESS_OFF
, tx
);
3601 if (zap_add(spa
->spa_meta_objset
,
3602 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3603 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3604 cmn_err(CE_PANIC
, "failed to add bpobj");
3606 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3607 spa
->spa_meta_objset
, obj
));
3610 * Create the pool's history object.
3612 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3613 spa_history_create_obj(spa
, tx
);
3616 * Set pool properties.
3618 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3619 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3620 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3621 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3623 if (props
!= NULL
) {
3624 spa_configfile_set(spa
, props
, B_FALSE
);
3625 spa_sync_props(props
, tx
);
3630 spa
->spa_sync_on
= B_TRUE
;
3631 txg_sync_start(spa
->spa_dsl_pool
);
3634 * We explicitly wait for the first transaction to complete so that our
3635 * bean counters are appropriately updated.
3637 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3639 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3641 spa_history_log_version(spa
, "create");
3643 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3645 mutex_exit(&spa_namespace_lock
);
3652 * Get the root pool information from the root disk, then import the root pool
3653 * during the system boot up time.
3655 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3658 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3661 nvlist_t
*nvtop
, *nvroot
;
3664 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3668 * Add this top-level vdev to the child array.
3670 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3672 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3674 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3677 * Put this pool's top-level vdevs into a root vdev.
3679 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3680 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3681 VDEV_TYPE_ROOT
) == 0);
3682 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3683 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3684 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3688 * Replace the existing vdev_tree with the new root vdev in
3689 * this pool's configuration (remove the old, add the new).
3691 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3692 nvlist_free(nvroot
);
3697 * Walk the vdev tree and see if we can find a device with "better"
3698 * configuration. A configuration is "better" if the label on that
3699 * device has a more recent txg.
3702 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3706 for (c
= 0; c
< vd
->vdev_children
; c
++)
3707 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3709 if (vd
->vdev_ops
->vdev_op_leaf
) {
3713 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3717 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3721 * Do we have a better boot device?
3723 if (label_txg
> *txg
) {
3732 * Import a root pool.
3734 * For x86. devpath_list will consist of devid and/or physpath name of
3735 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3736 * The GRUB "findroot" command will return the vdev we should boot.
3738 * For Sparc, devpath_list consists the physpath name of the booting device
3739 * no matter the rootpool is a single device pool or a mirrored pool.
3741 * "/pci@1f,0/ide@d/disk@0,0:a"
3744 spa_import_rootpool(char *devpath
, char *devid
)
3747 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3748 nvlist_t
*config
, *nvtop
;
3754 * Read the label from the boot device and generate a configuration.
3756 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3757 #if defined(_OBP) && defined(_KERNEL)
3758 if (config
== NULL
) {
3759 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3761 get_iscsi_bootpath_phy(devpath
);
3762 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3766 if (config
== NULL
) {
3767 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
3769 return (SET_ERROR(EIO
));
3772 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3774 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3776 mutex_enter(&spa_namespace_lock
);
3777 if ((spa
= spa_lookup(pname
)) != NULL
) {
3779 * Remove the existing root pool from the namespace so that we
3780 * can replace it with the correct config we just read in.
3785 spa
= spa_add(pname
, config
, NULL
);
3786 spa
->spa_is_root
= B_TRUE
;
3787 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3790 * Build up a vdev tree based on the boot device's label config.
3792 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3794 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3795 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3796 VDEV_ALLOC_ROOTPOOL
);
3797 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3799 mutex_exit(&spa_namespace_lock
);
3800 nvlist_free(config
);
3801 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3807 * Get the boot vdev.
3809 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3810 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3811 (u_longlong_t
)guid
);
3812 error
= SET_ERROR(ENOENT
);
3817 * Determine if there is a better boot device.
3820 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3822 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3823 "try booting from '%s'", avd
->vdev_path
);
3824 error
= SET_ERROR(EINVAL
);
3829 * If the boot device is part of a spare vdev then ensure that
3830 * we're booting off the active spare.
3832 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3833 !bvd
->vdev_isspare
) {
3834 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
3835 "try booting from '%s'",
3837 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
3838 error
= SET_ERROR(EINVAL
);
3844 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3846 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3847 mutex_exit(&spa_namespace_lock
);
3849 nvlist_free(config
);
3856 * Import a non-root pool into the system.
3859 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
3862 char *altroot
= NULL
;
3863 spa_load_state_t state
= SPA_LOAD_IMPORT
;
3864 zpool_rewind_policy_t policy
;
3865 uint64_t mode
= spa_mode_global
;
3866 uint64_t readonly
= B_FALSE
;
3869 nvlist_t
**spares
, **l2cache
;
3870 uint_t nspares
, nl2cache
;
3873 * If a pool with this name exists, return failure.
3875 mutex_enter(&spa_namespace_lock
);
3876 if (spa_lookup(pool
) != NULL
) {
3877 mutex_exit(&spa_namespace_lock
);
3878 return (SET_ERROR(EEXIST
));
3882 * Create and initialize the spa structure.
3884 (void) nvlist_lookup_string(props
,
3885 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3886 (void) nvlist_lookup_uint64(props
,
3887 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
3890 spa
= spa_add(pool
, config
, altroot
);
3891 spa
->spa_import_flags
= flags
;
3894 * Verbatim import - Take a pool and insert it into the namespace
3895 * as if it had been loaded at boot.
3897 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
3899 spa_configfile_set(spa
, props
, B_FALSE
);
3901 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3903 mutex_exit(&spa_namespace_lock
);
3904 spa_history_log_version(spa
, "import");
3909 spa_activate(spa
, mode
);
3912 * Don't start async tasks until we know everything is healthy.
3914 spa_async_suspend(spa
);
3916 zpool_get_rewind_policy(config
, &policy
);
3917 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3918 state
= SPA_LOAD_RECOVER
;
3921 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3922 * because the user-supplied config is actually the one to trust when
3925 if (state
!= SPA_LOAD_RECOVER
)
3926 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3928 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
3929 policy
.zrp_request
);
3932 * Propagate anything learned while loading the pool and pass it
3933 * back to caller (i.e. rewind info, missing devices, etc).
3935 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
3936 spa
->spa_load_info
) == 0);
3938 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3940 * Toss any existing sparelist, as it doesn't have any validity
3941 * anymore, and conflicts with spa_has_spare().
3943 if (spa
->spa_spares
.sav_config
) {
3944 nvlist_free(spa
->spa_spares
.sav_config
);
3945 spa
->spa_spares
.sav_config
= NULL
;
3946 spa_load_spares(spa
);
3948 if (spa
->spa_l2cache
.sav_config
) {
3949 nvlist_free(spa
->spa_l2cache
.sav_config
);
3950 spa
->spa_l2cache
.sav_config
= NULL
;
3951 spa_load_l2cache(spa
);
3954 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3957 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3960 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3961 VDEV_ALLOC_L2CACHE
);
3962 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3965 spa_configfile_set(spa
, props
, B_FALSE
);
3967 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
3968 (error
= spa_prop_set(spa
, props
)))) {
3970 spa_deactivate(spa
);
3972 mutex_exit(&spa_namespace_lock
);
3976 spa_async_resume(spa
);
3979 * Override any spares and level 2 cache devices as specified by
3980 * the user, as these may have correct device names/devids, etc.
3982 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3983 &spares
, &nspares
) == 0) {
3984 if (spa
->spa_spares
.sav_config
)
3985 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
3986 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3988 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
3989 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3990 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3991 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3992 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3993 spa_load_spares(spa
);
3994 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3995 spa
->spa_spares
.sav_sync
= B_TRUE
;
3997 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3998 &l2cache
, &nl2cache
) == 0) {
3999 if (spa
->spa_l2cache
.sav_config
)
4000 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4001 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4003 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4004 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
4005 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4006 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4007 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4008 spa_load_l2cache(spa
);
4009 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4010 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4014 * Check for any removed devices.
4016 if (spa
->spa_autoreplace
) {
4017 spa_aux_check_removed(&spa
->spa_spares
);
4018 spa_aux_check_removed(&spa
->spa_l2cache
);
4021 if (spa_writeable(spa
)) {
4023 * Update the config cache to include the newly-imported pool.
4025 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4029 * It's possible that the pool was expanded while it was exported.
4030 * We kick off an async task to handle this for us.
4032 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4034 mutex_exit(&spa_namespace_lock
);
4035 spa_history_log_version(spa
, "import");
4038 zvol_create_minors(pool
);
4045 spa_tryimport(nvlist_t
*tryconfig
)
4047 nvlist_t
*config
= NULL
;
4053 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4056 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4060 * Create and initialize the spa structure.
4062 mutex_enter(&spa_namespace_lock
);
4063 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4064 spa_activate(spa
, FREAD
);
4067 * Pass off the heavy lifting to spa_load().
4068 * Pass TRUE for mosconfig because the user-supplied config
4069 * is actually the one to trust when doing an import.
4071 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4074 * If 'tryconfig' was at least parsable, return the current config.
4076 if (spa
->spa_root_vdev
!= NULL
) {
4077 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4078 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4080 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4082 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4083 spa
->spa_uberblock
.ub_timestamp
) == 0);
4084 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4085 spa
->spa_load_info
) == 0);
4088 * If the bootfs property exists on this pool then we
4089 * copy it out so that external consumers can tell which
4090 * pools are bootable.
4092 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4093 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
4096 * We have to play games with the name since the
4097 * pool was opened as TRYIMPORT_NAME.
4099 if (dsl_dsobj_to_dsname(spa_name(spa
),
4100 spa
->spa_bootfs
, tmpname
) == 0) {
4104 dsname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
4106 cp
= strchr(tmpname
, '/');
4108 (void) strlcpy(dsname
, tmpname
,
4111 (void) snprintf(dsname
, MAXPATHLEN
,
4112 "%s/%s", poolname
, ++cp
);
4114 VERIFY(nvlist_add_string(config
,
4115 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4116 kmem_free(dsname
, MAXPATHLEN
);
4118 kmem_free(tmpname
, MAXPATHLEN
);
4122 * Add the list of hot spares and level 2 cache devices.
4124 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4125 spa_add_spares(spa
, config
);
4126 spa_add_l2cache(spa
, config
);
4127 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4131 spa_deactivate(spa
);
4133 mutex_exit(&spa_namespace_lock
);
4139 * Pool export/destroy
4141 * The act of destroying or exporting a pool is very simple. We make sure there
4142 * is no more pending I/O and any references to the pool are gone. Then, we
4143 * update the pool state and sync all the labels to disk, removing the
4144 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4145 * we don't sync the labels or remove the configuration cache.
4148 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4149 boolean_t force
, boolean_t hardforce
)
4156 if (!(spa_mode_global
& FWRITE
))
4157 return (SET_ERROR(EROFS
));
4159 mutex_enter(&spa_namespace_lock
);
4160 if ((spa
= spa_lookup(pool
)) == NULL
) {
4161 mutex_exit(&spa_namespace_lock
);
4162 return (SET_ERROR(ENOENT
));
4166 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4167 * reacquire the namespace lock, and see if we can export.
4169 spa_open_ref(spa
, FTAG
);
4170 mutex_exit(&spa_namespace_lock
);
4171 spa_async_suspend(spa
);
4172 mutex_enter(&spa_namespace_lock
);
4173 spa_close(spa
, FTAG
);
4176 * The pool will be in core if it's openable,
4177 * in which case we can modify its state.
4179 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
&& spa
->spa_sync_on
) {
4181 * Objsets may be open only because they're dirty, so we
4182 * have to force it to sync before checking spa_refcnt.
4184 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4187 * A pool cannot be exported or destroyed if there are active
4188 * references. If we are resetting a pool, allow references by
4189 * fault injection handlers.
4191 if (!spa_refcount_zero(spa
) ||
4192 (spa
->spa_inject_ref
!= 0 &&
4193 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4194 spa_async_resume(spa
);
4195 mutex_exit(&spa_namespace_lock
);
4196 return (SET_ERROR(EBUSY
));
4200 * A pool cannot be exported if it has an active shared spare.
4201 * This is to prevent other pools stealing the active spare
4202 * from an exported pool. At user's own will, such pool can
4203 * be forcedly exported.
4205 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4206 spa_has_active_shared_spare(spa
)) {
4207 spa_async_resume(spa
);
4208 mutex_exit(&spa_namespace_lock
);
4209 return (SET_ERROR(EXDEV
));
4213 * We want this to be reflected on every label,
4214 * so mark them all dirty. spa_unload() will do the
4215 * final sync that pushes these changes out.
4217 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4218 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4219 spa
->spa_state
= new_state
;
4220 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4222 vdev_config_dirty(spa
->spa_root_vdev
);
4223 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4227 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_DESTROY
);
4229 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4231 spa_deactivate(spa
);
4234 if (oldconfig
&& spa
->spa_config
)
4235 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4237 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4239 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4242 mutex_exit(&spa_namespace_lock
);
4248 * Destroy a storage pool.
4251 spa_destroy(char *pool
)
4253 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4258 * Export a storage pool.
4261 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4262 boolean_t hardforce
)
4264 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4269 * Similar to spa_export(), this unloads the spa_t without actually removing it
4270 * from the namespace in any way.
4273 spa_reset(char *pool
)
4275 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4280 * ==========================================================================
4281 * Device manipulation
4282 * ==========================================================================
4286 * Add a device to a storage pool.
4289 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4293 vdev_t
*rvd
= spa
->spa_root_vdev
;
4295 nvlist_t
**spares
, **l2cache
;
4296 uint_t nspares
, nl2cache
;
4299 ASSERT(spa_writeable(spa
));
4301 txg
= spa_vdev_enter(spa
);
4303 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4304 VDEV_ALLOC_ADD
)) != 0)
4305 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4307 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4309 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4313 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4317 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4318 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4320 if (vd
->vdev_children
!= 0 &&
4321 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4322 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4325 * We must validate the spares and l2cache devices after checking the
4326 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4328 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4329 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4332 * Transfer each new top-level vdev from vd to rvd.
4334 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4337 * Set the vdev id to the first hole, if one exists.
4339 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4340 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4341 vdev_free(rvd
->vdev_child
[id
]);
4345 tvd
= vd
->vdev_child
[c
];
4346 vdev_remove_child(vd
, tvd
);
4348 vdev_add_child(rvd
, tvd
);
4349 vdev_config_dirty(tvd
);
4353 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4354 ZPOOL_CONFIG_SPARES
);
4355 spa_load_spares(spa
);
4356 spa
->spa_spares
.sav_sync
= B_TRUE
;
4359 if (nl2cache
!= 0) {
4360 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4361 ZPOOL_CONFIG_L2CACHE
);
4362 spa_load_l2cache(spa
);
4363 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4367 * We have to be careful when adding new vdevs to an existing pool.
4368 * If other threads start allocating from these vdevs before we
4369 * sync the config cache, and we lose power, then upon reboot we may
4370 * fail to open the pool because there are DVAs that the config cache
4371 * can't translate. Therefore, we first add the vdevs without
4372 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4373 * and then let spa_config_update() initialize the new metaslabs.
4375 * spa_load() checks for added-but-not-initialized vdevs, so that
4376 * if we lose power at any point in this sequence, the remaining
4377 * steps will be completed the next time we load the pool.
4379 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4381 mutex_enter(&spa_namespace_lock
);
4382 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4383 mutex_exit(&spa_namespace_lock
);
4389 * Attach a device to a mirror. The arguments are the path to any device
4390 * in the mirror, and the nvroot for the new device. If the path specifies
4391 * a device that is not mirrored, we automatically insert the mirror vdev.
4393 * If 'replacing' is specified, the new device is intended to replace the
4394 * existing device; in this case the two devices are made into their own
4395 * mirror using the 'replacing' vdev, which is functionally identical to
4396 * the mirror vdev (it actually reuses all the same ops) but has a few
4397 * extra rules: you can't attach to it after it's been created, and upon
4398 * completion of resilvering, the first disk (the one being replaced)
4399 * is automatically detached.
4402 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4404 uint64_t txg
, dtl_max_txg
;
4405 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4407 char *oldvdpath
, *newvdpath
;
4410 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4412 ASSERT(spa_writeable(spa
));
4414 txg
= spa_vdev_enter(spa
);
4416 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4419 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4421 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4422 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4424 pvd
= oldvd
->vdev_parent
;
4426 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4427 VDEV_ALLOC_ATTACH
)) != 0)
4428 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4430 if (newrootvd
->vdev_children
!= 1)
4431 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4433 newvd
= newrootvd
->vdev_child
[0];
4435 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4436 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4438 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4439 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4442 * Spares can't replace logs
4444 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4445 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4449 * For attach, the only allowable parent is a mirror or the root
4452 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4453 pvd
->vdev_ops
!= &vdev_root_ops
)
4454 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4456 pvops
= &vdev_mirror_ops
;
4459 * Active hot spares can only be replaced by inactive hot
4462 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4463 oldvd
->vdev_isspare
&&
4464 !spa_has_spare(spa
, newvd
->vdev_guid
))
4465 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4468 * If the source is a hot spare, and the parent isn't already a
4469 * spare, then we want to create a new hot spare. Otherwise, we
4470 * want to create a replacing vdev. The user is not allowed to
4471 * attach to a spared vdev child unless the 'isspare' state is
4472 * the same (spare replaces spare, non-spare replaces
4475 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4476 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4477 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4478 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4479 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4480 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4483 if (newvd
->vdev_isspare
)
4484 pvops
= &vdev_spare_ops
;
4486 pvops
= &vdev_replacing_ops
;
4490 * Make sure the new device is big enough.
4492 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4493 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4496 * The new device cannot have a higher alignment requirement
4497 * than the top-level vdev.
4499 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4500 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4503 * If this is an in-place replacement, update oldvd's path and devid
4504 * to make it distinguishable from newvd, and unopenable from now on.
4506 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4507 spa_strfree(oldvd
->vdev_path
);
4508 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4510 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4511 newvd
->vdev_path
, "old");
4512 if (oldvd
->vdev_devid
!= NULL
) {
4513 spa_strfree(oldvd
->vdev_devid
);
4514 oldvd
->vdev_devid
= NULL
;
4518 /* mark the device being resilvered */
4519 newvd
->vdev_resilver_txg
= txg
;
4522 * If the parent is not a mirror, or if we're replacing, insert the new
4523 * mirror/replacing/spare vdev above oldvd.
4525 if (pvd
->vdev_ops
!= pvops
)
4526 pvd
= vdev_add_parent(oldvd
, pvops
);
4528 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4529 ASSERT(pvd
->vdev_ops
== pvops
);
4530 ASSERT(oldvd
->vdev_parent
== pvd
);
4533 * Extract the new device from its root and add it to pvd.
4535 vdev_remove_child(newrootvd
, newvd
);
4536 newvd
->vdev_id
= pvd
->vdev_children
;
4537 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4538 vdev_add_child(pvd
, newvd
);
4540 tvd
= newvd
->vdev_top
;
4541 ASSERT(pvd
->vdev_top
== tvd
);
4542 ASSERT(tvd
->vdev_parent
== rvd
);
4544 vdev_config_dirty(tvd
);
4547 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4548 * for any dmu_sync-ed blocks. It will propagate upward when
4549 * spa_vdev_exit() calls vdev_dtl_reassess().
4551 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4553 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4554 dtl_max_txg
- TXG_INITIAL
);
4556 if (newvd
->vdev_isspare
) {
4557 spa_spare_activate(newvd
);
4558 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_DEVICE_SPARE
);
4561 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4562 newvdpath
= spa_strdup(newvd
->vdev_path
);
4563 newvd_isspare
= newvd
->vdev_isspare
;
4566 * Mark newvd's DTL dirty in this txg.
4568 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4571 * Restart the resilver
4573 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4578 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4580 spa_history_log_internal(spa
, "vdev attach", NULL
,
4581 "%s vdev=%s %s vdev=%s",
4582 replacing
&& newvd_isspare
? "spare in" :
4583 replacing
? "replace" : "attach", newvdpath
,
4584 replacing
? "for" : "to", oldvdpath
);
4586 spa_strfree(oldvdpath
);
4587 spa_strfree(newvdpath
);
4589 if (spa
->spa_bootfs
)
4590 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH
);
4596 * Detach a device from a mirror or replacing vdev.
4598 * If 'replace_done' is specified, only detach if the parent
4599 * is a replacing vdev.
4602 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4606 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4607 boolean_t unspare
= B_FALSE
;
4608 uint64_t unspare_guid
= 0;
4611 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4612 ASSERT(spa_writeable(spa
));
4614 txg
= spa_vdev_enter(spa
);
4616 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4619 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4621 if (!vd
->vdev_ops
->vdev_op_leaf
)
4622 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4624 pvd
= vd
->vdev_parent
;
4627 * If the parent/child relationship is not as expected, don't do it.
4628 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4629 * vdev that's replacing B with C. The user's intent in replacing
4630 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4631 * the replace by detaching C, the expected behavior is to end up
4632 * M(A,B). But suppose that right after deciding to detach C,
4633 * the replacement of B completes. We would have M(A,C), and then
4634 * ask to detach C, which would leave us with just A -- not what
4635 * the user wanted. To prevent this, we make sure that the
4636 * parent/child relationship hasn't changed -- in this example,
4637 * that C's parent is still the replacing vdev R.
4639 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4640 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4643 * Only 'replacing' or 'spare' vdevs can be replaced.
4645 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4646 pvd
->vdev_ops
!= &vdev_spare_ops
)
4647 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4649 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4650 spa_version(spa
) >= SPA_VERSION_SPARES
);
4653 * Only mirror, replacing, and spare vdevs support detach.
4655 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4656 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4657 pvd
->vdev_ops
!= &vdev_spare_ops
)
4658 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4661 * If this device has the only valid copy of some data,
4662 * we cannot safely detach it.
4664 if (vdev_dtl_required(vd
))
4665 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4667 ASSERT(pvd
->vdev_children
>= 2);
4670 * If we are detaching the second disk from a replacing vdev, then
4671 * check to see if we changed the original vdev's path to have "/old"
4672 * at the end in spa_vdev_attach(). If so, undo that change now.
4674 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4675 vd
->vdev_path
!= NULL
) {
4676 size_t len
= strlen(vd
->vdev_path
);
4678 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4679 cvd
= pvd
->vdev_child
[c
];
4681 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4684 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4685 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4686 spa_strfree(cvd
->vdev_path
);
4687 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4694 * If we are detaching the original disk from a spare, then it implies
4695 * that the spare should become a real disk, and be removed from the
4696 * active spare list for the pool.
4698 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4700 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4704 * Erase the disk labels so the disk can be used for other things.
4705 * This must be done after all other error cases are handled,
4706 * but before we disembowel vd (so we can still do I/O to it).
4707 * But if we can't do it, don't treat the error as fatal --
4708 * it may be that the unwritability of the disk is the reason
4709 * it's being detached!
4711 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4714 * Remove vd from its parent and compact the parent's children.
4716 vdev_remove_child(pvd
, vd
);
4717 vdev_compact_children(pvd
);
4720 * Remember one of the remaining children so we can get tvd below.
4722 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4725 * If we need to remove the remaining child from the list of hot spares,
4726 * do it now, marking the vdev as no longer a spare in the process.
4727 * We must do this before vdev_remove_parent(), because that can
4728 * change the GUID if it creates a new toplevel GUID. For a similar
4729 * reason, we must remove the spare now, in the same txg as the detach;
4730 * otherwise someone could attach a new sibling, change the GUID, and
4731 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4734 ASSERT(cvd
->vdev_isspare
);
4735 spa_spare_remove(cvd
);
4736 unspare_guid
= cvd
->vdev_guid
;
4737 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4738 cvd
->vdev_unspare
= B_TRUE
;
4742 * If the parent mirror/replacing vdev only has one child,
4743 * the parent is no longer needed. Remove it from the tree.
4745 if (pvd
->vdev_children
== 1) {
4746 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4747 cvd
->vdev_unspare
= B_FALSE
;
4748 vdev_remove_parent(cvd
);
4753 * We don't set tvd until now because the parent we just removed
4754 * may have been the previous top-level vdev.
4756 tvd
= cvd
->vdev_top
;
4757 ASSERT(tvd
->vdev_parent
== rvd
);
4760 * Reevaluate the parent vdev state.
4762 vdev_propagate_state(cvd
);
4765 * If the 'autoexpand' property is set on the pool then automatically
4766 * try to expand the size of the pool. For example if the device we
4767 * just detached was smaller than the others, it may be possible to
4768 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4769 * first so that we can obtain the updated sizes of the leaf vdevs.
4771 if (spa
->spa_autoexpand
) {
4773 vdev_expand(tvd
, txg
);
4776 vdev_config_dirty(tvd
);
4779 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4780 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4781 * But first make sure we're not on any *other* txg's DTL list, to
4782 * prevent vd from being accessed after it's freed.
4784 vdpath
= spa_strdup(vd
->vdev_path
);
4785 for (t
= 0; t
< TXG_SIZE
; t
++)
4786 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4787 vd
->vdev_detached
= B_TRUE
;
4788 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4790 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_REMOVE
);
4792 /* hang on to the spa before we release the lock */
4793 spa_open_ref(spa
, FTAG
);
4795 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4797 spa_history_log_internal(spa
, "detach", NULL
,
4799 spa_strfree(vdpath
);
4802 * If this was the removal of the original device in a hot spare vdev,
4803 * then we want to go through and remove the device from the hot spare
4804 * list of every other pool.
4807 spa_t
*altspa
= NULL
;
4809 mutex_enter(&spa_namespace_lock
);
4810 while ((altspa
= spa_next(altspa
)) != NULL
) {
4811 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4815 spa_open_ref(altspa
, FTAG
);
4816 mutex_exit(&spa_namespace_lock
);
4817 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4818 mutex_enter(&spa_namespace_lock
);
4819 spa_close(altspa
, FTAG
);
4821 mutex_exit(&spa_namespace_lock
);
4823 /* search the rest of the vdevs for spares to remove */
4824 spa_vdev_resilver_done(spa
);
4827 /* all done with the spa; OK to release */
4828 mutex_enter(&spa_namespace_lock
);
4829 spa_close(spa
, FTAG
);
4830 mutex_exit(&spa_namespace_lock
);
4836 * Split a set of devices from their mirrors, and create a new pool from them.
4839 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
4840 nvlist_t
*props
, boolean_t exp
)
4843 uint64_t txg
, *glist
;
4845 uint_t c
, children
, lastlog
;
4846 nvlist_t
**child
, *nvl
, *tmp
;
4848 char *altroot
= NULL
;
4849 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
4850 boolean_t activate_slog
;
4852 ASSERT(spa_writeable(spa
));
4854 txg
= spa_vdev_enter(spa
);
4856 /* clear the log and flush everything up to now */
4857 activate_slog
= spa_passivate_log(spa
);
4858 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4859 error
= spa_offline_log(spa
);
4860 txg
= spa_vdev_config_enter(spa
);
4863 spa_activate_log(spa
);
4866 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4868 /* check new spa name before going any further */
4869 if (spa_lookup(newname
) != NULL
)
4870 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
4873 * scan through all the children to ensure they're all mirrors
4875 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
4876 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
4878 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4880 /* first, check to ensure we've got the right child count */
4881 rvd
= spa
->spa_root_vdev
;
4883 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
4884 vdev_t
*vd
= rvd
->vdev_child
[c
];
4886 /* don't count the holes & logs as children */
4887 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
4895 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
4896 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4898 /* next, ensure no spare or cache devices are part of the split */
4899 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
4900 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
4901 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4903 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_PUSHPAGE
);
4904 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_PUSHPAGE
);
4906 /* then, loop over each vdev and validate it */
4907 for (c
= 0; c
< children
; c
++) {
4908 uint64_t is_hole
= 0;
4910 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
4914 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
4915 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
4918 error
= SET_ERROR(EINVAL
);
4923 /* which disk is going to be split? */
4924 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
4926 error
= SET_ERROR(EINVAL
);
4930 /* look it up in the spa */
4931 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
4932 if (vml
[c
] == NULL
) {
4933 error
= SET_ERROR(ENODEV
);
4937 /* make sure there's nothing stopping the split */
4938 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
4939 vml
[c
]->vdev_islog
||
4940 vml
[c
]->vdev_ishole
||
4941 vml
[c
]->vdev_isspare
||
4942 vml
[c
]->vdev_isl2cache
||
4943 !vdev_writeable(vml
[c
]) ||
4944 vml
[c
]->vdev_children
!= 0 ||
4945 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
4946 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
4947 error
= SET_ERROR(EINVAL
);
4951 if (vdev_dtl_required(vml
[c
])) {
4952 error
= SET_ERROR(EBUSY
);
4956 /* we need certain info from the top level */
4957 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
4958 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
4959 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
4960 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
4961 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
4962 vml
[c
]->vdev_top
->vdev_asize
) == 0);
4963 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
4964 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
4968 kmem_free(vml
, children
* sizeof (vdev_t
*));
4969 kmem_free(glist
, children
* sizeof (uint64_t));
4970 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4973 /* stop writers from using the disks */
4974 for (c
= 0; c
< children
; c
++) {
4976 vml
[c
]->vdev_offline
= B_TRUE
;
4978 vdev_reopen(spa
->spa_root_vdev
);
4981 * Temporarily record the splitting vdevs in the spa config. This
4982 * will disappear once the config is regenerated.
4984 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
4985 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
4986 glist
, children
) == 0);
4987 kmem_free(glist
, children
* sizeof (uint64_t));
4989 mutex_enter(&spa
->spa_props_lock
);
4990 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
4992 mutex_exit(&spa
->spa_props_lock
);
4993 spa
->spa_config_splitting
= nvl
;
4994 vdev_config_dirty(spa
->spa_root_vdev
);
4996 /* configure and create the new pool */
4997 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
4998 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4999 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5000 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5001 spa_version(spa
)) == 0);
5002 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5003 spa
->spa_config_txg
) == 0);
5004 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5005 spa_generate_guid(NULL
)) == 0);
5006 (void) nvlist_lookup_string(props
,
5007 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5009 /* add the new pool to the namespace */
5010 newspa
= spa_add(newname
, config
, altroot
);
5011 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5012 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5014 /* release the spa config lock, retaining the namespace lock */
5015 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5017 if (zio_injection_enabled
)
5018 zio_handle_panic_injection(spa
, FTAG
, 1);
5020 spa_activate(newspa
, spa_mode_global
);
5021 spa_async_suspend(newspa
);
5023 /* create the new pool from the disks of the original pool */
5024 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5028 /* if that worked, generate a real config for the new pool */
5029 if (newspa
->spa_root_vdev
!= NULL
) {
5030 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5031 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
5032 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5033 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5034 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5039 if (props
!= NULL
) {
5040 spa_configfile_set(newspa
, props
, B_FALSE
);
5041 error
= spa_prop_set(newspa
, props
);
5046 /* flush everything */
5047 txg
= spa_vdev_config_enter(newspa
);
5048 vdev_config_dirty(newspa
->spa_root_vdev
);
5049 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5051 if (zio_injection_enabled
)
5052 zio_handle_panic_injection(spa
, FTAG
, 2);
5054 spa_async_resume(newspa
);
5056 /* finally, update the original pool's config */
5057 txg
= spa_vdev_config_enter(spa
);
5058 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5059 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5062 for (c
= 0; c
< children
; c
++) {
5063 if (vml
[c
] != NULL
) {
5066 spa_history_log_internal(spa
, "detach", tx
,
5067 "vdev=%s", vml
[c
]->vdev_path
);
5071 vdev_config_dirty(spa
->spa_root_vdev
);
5072 spa
->spa_config_splitting
= NULL
;
5076 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5078 if (zio_injection_enabled
)
5079 zio_handle_panic_injection(spa
, FTAG
, 3);
5081 /* split is complete; log a history record */
5082 spa_history_log_internal(newspa
, "split", NULL
,
5083 "from pool %s", spa_name(spa
));
5085 kmem_free(vml
, children
* sizeof (vdev_t
*));
5087 /* if we're not going to mount the filesystems in userland, export */
5089 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5096 spa_deactivate(newspa
);
5099 txg
= spa_vdev_config_enter(spa
);
5101 /* re-online all offlined disks */
5102 for (c
= 0; c
< children
; c
++) {
5104 vml
[c
]->vdev_offline
= B_FALSE
;
5106 vdev_reopen(spa
->spa_root_vdev
);
5108 nvlist_free(spa
->spa_config_splitting
);
5109 spa
->spa_config_splitting
= NULL
;
5110 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5112 kmem_free(vml
, children
* sizeof (vdev_t
*));
5117 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5121 for (i
= 0; i
< count
; i
++) {
5124 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5127 if (guid
== target_guid
)
5135 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5136 nvlist_t
*dev_to_remove
)
5138 nvlist_t
**newdev
= NULL
;
5142 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_PUSHPAGE
);
5144 for (i
= 0, j
= 0; i
< count
; i
++) {
5145 if (dev
[i
] == dev_to_remove
)
5147 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_PUSHPAGE
) == 0);
5150 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5151 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5153 for (i
= 0; i
< count
- 1; i
++)
5154 nvlist_free(newdev
[i
]);
5157 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5161 * Evacuate the device.
5164 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5169 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5170 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5171 ASSERT(vd
== vd
->vdev_top
);
5174 * Evacuate the device. We don't hold the config lock as writer
5175 * since we need to do I/O but we do keep the
5176 * spa_namespace_lock held. Once this completes the device
5177 * should no longer have any blocks allocated on it.
5179 if (vd
->vdev_islog
) {
5180 if (vd
->vdev_stat
.vs_alloc
!= 0)
5181 error
= spa_offline_log(spa
);
5183 error
= SET_ERROR(ENOTSUP
);
5190 * The evacuation succeeded. Remove any remaining MOS metadata
5191 * associated with this vdev, and wait for these changes to sync.
5193 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5194 txg
= spa_vdev_config_enter(spa
);
5195 vd
->vdev_removing
= B_TRUE
;
5196 vdev_dirty(vd
, 0, NULL
, txg
);
5197 vdev_config_dirty(vd
);
5198 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5204 * Complete the removal by cleaning up the namespace.
5207 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5209 vdev_t
*rvd
= spa
->spa_root_vdev
;
5210 uint64_t id
= vd
->vdev_id
;
5211 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5213 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5214 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5215 ASSERT(vd
== vd
->vdev_top
);
5218 * Only remove any devices which are empty.
5220 if (vd
->vdev_stat
.vs_alloc
!= 0)
5223 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5225 if (list_link_active(&vd
->vdev_state_dirty_node
))
5226 vdev_state_clean(vd
);
5227 if (list_link_active(&vd
->vdev_config_dirty_node
))
5228 vdev_config_clean(vd
);
5233 vdev_compact_children(rvd
);
5235 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5236 vdev_add_child(rvd
, vd
);
5238 vdev_config_dirty(rvd
);
5241 * Reassess the health of our root vdev.
5247 * Remove a device from the pool -
5249 * Removing a device from the vdev namespace requires several steps
5250 * and can take a significant amount of time. As a result we use
5251 * the spa_vdev_config_[enter/exit] functions which allow us to
5252 * grab and release the spa_config_lock while still holding the namespace
5253 * lock. During each step the configuration is synced out.
5255 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5259 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5262 metaslab_group_t
*mg
;
5263 nvlist_t
**spares
, **l2cache
, *nv
;
5265 uint_t nspares
, nl2cache
;
5267 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5269 ASSERT(spa_writeable(spa
));
5272 txg
= spa_vdev_enter(spa
);
5274 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5276 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5277 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5278 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5279 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5281 * Only remove the hot spare if it's not currently in use
5284 if (vd
== NULL
|| unspare
) {
5285 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5286 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5287 spa_load_spares(spa
);
5288 spa
->spa_spares
.sav_sync
= B_TRUE
;
5290 error
= SET_ERROR(EBUSY
);
5292 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5293 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5294 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5295 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5297 * Cache devices can always be removed.
5299 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5300 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5301 spa_load_l2cache(spa
);
5302 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5303 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5305 ASSERT(vd
== vd
->vdev_top
);
5308 * XXX - Once we have bp-rewrite this should
5309 * become the common case.
5315 * Stop allocating from this vdev.
5317 metaslab_group_passivate(mg
);
5320 * Wait for the youngest allocations and frees to sync,
5321 * and then wait for the deferral of those frees to finish.
5323 spa_vdev_config_exit(spa
, NULL
,
5324 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5327 * Attempt to evacuate the vdev.
5329 error
= spa_vdev_remove_evacuate(spa
, vd
);
5331 txg
= spa_vdev_config_enter(spa
);
5334 * If we couldn't evacuate the vdev, unwind.
5337 metaslab_group_activate(mg
);
5338 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5342 * Clean up the vdev namespace.
5344 spa_vdev_remove_from_namespace(spa
, vd
);
5346 } else if (vd
!= NULL
) {
5348 * Normal vdevs cannot be removed (yet).
5350 error
= SET_ERROR(ENOTSUP
);
5353 * There is no vdev of any kind with the specified guid.
5355 error
= SET_ERROR(ENOENT
);
5359 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5365 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5366 * currently spared, so we can detach it.
5369 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5371 vdev_t
*newvd
, *oldvd
;
5374 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5375 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5381 * Check for a completed replacement. We always consider the first
5382 * vdev in the list to be the oldest vdev, and the last one to be
5383 * the newest (see spa_vdev_attach() for how that works). In
5384 * the case where the newest vdev is faulted, we will not automatically
5385 * remove it after a resilver completes. This is OK as it will require
5386 * user intervention to determine which disk the admin wishes to keep.
5388 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5389 ASSERT(vd
->vdev_children
> 1);
5391 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5392 oldvd
= vd
->vdev_child
[0];
5394 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5395 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5396 !vdev_dtl_required(oldvd
))
5401 * Check for a completed resilver with the 'unspare' flag set.
5403 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5404 vdev_t
*first
= vd
->vdev_child
[0];
5405 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5407 if (last
->vdev_unspare
) {
5410 } else if (first
->vdev_unspare
) {
5417 if (oldvd
!= NULL
&&
5418 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5419 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5420 !vdev_dtl_required(oldvd
))
5424 * If there are more than two spares attached to a disk,
5425 * and those spares are not required, then we want to
5426 * attempt to free them up now so that they can be used
5427 * by other pools. Once we're back down to a single
5428 * disk+spare, we stop removing them.
5430 if (vd
->vdev_children
> 2) {
5431 newvd
= vd
->vdev_child
[1];
5433 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5434 vdev_dtl_empty(last
, DTL_MISSING
) &&
5435 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5436 !vdev_dtl_required(newvd
))
5445 spa_vdev_resilver_done(spa_t
*spa
)
5447 vdev_t
*vd
, *pvd
, *ppvd
;
5448 uint64_t guid
, sguid
, pguid
, ppguid
;
5450 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5452 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5453 pvd
= vd
->vdev_parent
;
5454 ppvd
= pvd
->vdev_parent
;
5455 guid
= vd
->vdev_guid
;
5456 pguid
= pvd
->vdev_guid
;
5457 ppguid
= ppvd
->vdev_guid
;
5460 * If we have just finished replacing a hot spared device, then
5461 * we need to detach the parent's first child (the original hot
5464 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5465 ppvd
->vdev_children
== 2) {
5466 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5467 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5469 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5471 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5472 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5474 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5476 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5479 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5483 * Update the stored path or FRU for this vdev.
5486 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5490 boolean_t sync
= B_FALSE
;
5492 ASSERT(spa_writeable(spa
));
5494 spa_vdev_state_enter(spa
, SCL_ALL
);
5496 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5497 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5499 if (!vd
->vdev_ops
->vdev_op_leaf
)
5500 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5503 if (strcmp(value
, vd
->vdev_path
) != 0) {
5504 spa_strfree(vd
->vdev_path
);
5505 vd
->vdev_path
= spa_strdup(value
);
5509 if (vd
->vdev_fru
== NULL
) {
5510 vd
->vdev_fru
= spa_strdup(value
);
5512 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5513 spa_strfree(vd
->vdev_fru
);
5514 vd
->vdev_fru
= spa_strdup(value
);
5519 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5523 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5525 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5529 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5531 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5535 * ==========================================================================
5537 * ==========================================================================
5541 spa_scan_stop(spa_t
*spa
)
5543 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5544 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5545 return (SET_ERROR(EBUSY
));
5546 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5550 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5552 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5554 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5555 return (SET_ERROR(ENOTSUP
));
5558 * If a resilver was requested, but there is no DTL on a
5559 * writeable leaf device, we have nothing to do.
5561 if (func
== POOL_SCAN_RESILVER
&&
5562 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5563 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5567 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5571 * ==========================================================================
5572 * SPA async task processing
5573 * ==========================================================================
5577 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5581 if (vd
->vdev_remove_wanted
) {
5582 vd
->vdev_remove_wanted
= B_FALSE
;
5583 vd
->vdev_delayed_close
= B_FALSE
;
5584 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5587 * We want to clear the stats, but we don't want to do a full
5588 * vdev_clear() as that will cause us to throw away
5589 * degraded/faulted state as well as attempt to reopen the
5590 * device, all of which is a waste.
5592 vd
->vdev_stat
.vs_read_errors
= 0;
5593 vd
->vdev_stat
.vs_write_errors
= 0;
5594 vd
->vdev_stat
.vs_checksum_errors
= 0;
5596 vdev_state_dirty(vd
->vdev_top
);
5599 for (c
= 0; c
< vd
->vdev_children
; c
++)
5600 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5604 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5608 if (vd
->vdev_probe_wanted
) {
5609 vd
->vdev_probe_wanted
= B_FALSE
;
5610 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5613 for (c
= 0; c
< vd
->vdev_children
; c
++)
5614 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5618 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5622 if (!spa
->spa_autoexpand
)
5625 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5626 vdev_t
*cvd
= vd
->vdev_child
[c
];
5627 spa_async_autoexpand(spa
, cvd
);
5630 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5633 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND
);
5637 spa_async_thread(spa_t
*spa
)
5641 ASSERT(spa
->spa_sync_on
);
5643 mutex_enter(&spa
->spa_async_lock
);
5644 tasks
= spa
->spa_async_tasks
;
5645 spa
->spa_async_tasks
= 0;
5646 mutex_exit(&spa
->spa_async_lock
);
5649 * See if the config needs to be updated.
5651 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5652 uint64_t old_space
, new_space
;
5654 mutex_enter(&spa_namespace_lock
);
5655 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5656 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5657 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5658 mutex_exit(&spa_namespace_lock
);
5661 * If the pool grew as a result of the config update,
5662 * then log an internal history event.
5664 if (new_space
!= old_space
) {
5665 spa_history_log_internal(spa
, "vdev online", NULL
,
5666 "pool '%s' size: %llu(+%llu)",
5667 spa_name(spa
), new_space
, new_space
- old_space
);
5672 * See if any devices need to be marked REMOVED.
5674 if (tasks
& SPA_ASYNC_REMOVE
) {
5675 spa_vdev_state_enter(spa
, SCL_NONE
);
5676 spa_async_remove(spa
, spa
->spa_root_vdev
);
5677 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5678 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5679 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5680 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5681 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5684 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5685 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5686 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5687 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5691 * See if any devices need to be probed.
5693 if (tasks
& SPA_ASYNC_PROBE
) {
5694 spa_vdev_state_enter(spa
, SCL_NONE
);
5695 spa_async_probe(spa
, spa
->spa_root_vdev
);
5696 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5700 * If any devices are done replacing, detach them.
5702 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5703 spa_vdev_resilver_done(spa
);
5706 * Kick off a resilver.
5708 if (tasks
& SPA_ASYNC_RESILVER
)
5709 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5712 * Let the world know that we're done.
5714 mutex_enter(&spa
->spa_async_lock
);
5715 spa
->spa_async_thread
= NULL
;
5716 cv_broadcast(&spa
->spa_async_cv
);
5717 mutex_exit(&spa
->spa_async_lock
);
5722 spa_async_suspend(spa_t
*spa
)
5724 mutex_enter(&spa
->spa_async_lock
);
5725 spa
->spa_async_suspended
++;
5726 while (spa
->spa_async_thread
!= NULL
)
5727 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5728 mutex_exit(&spa
->spa_async_lock
);
5732 spa_async_resume(spa_t
*spa
)
5734 mutex_enter(&spa
->spa_async_lock
);
5735 ASSERT(spa
->spa_async_suspended
!= 0);
5736 spa
->spa_async_suspended
--;
5737 mutex_exit(&spa
->spa_async_lock
);
5741 spa_async_dispatch(spa_t
*spa
)
5743 mutex_enter(&spa
->spa_async_lock
);
5744 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5745 spa
->spa_async_thread
== NULL
&&
5746 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5747 spa
->spa_async_thread
= thread_create(NULL
, 0,
5748 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5749 mutex_exit(&spa
->spa_async_lock
);
5753 spa_async_request(spa_t
*spa
, int task
)
5755 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5756 mutex_enter(&spa
->spa_async_lock
);
5757 spa
->spa_async_tasks
|= task
;
5758 mutex_exit(&spa
->spa_async_lock
);
5762 * ==========================================================================
5763 * SPA syncing routines
5764 * ==========================================================================
5768 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5771 bpobj_enqueue(bpo
, bp
, tx
);
5776 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5780 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5786 * Note: this simple function is not inlined to make it easier to dtrace the
5787 * amount of time spent syncing frees.
5790 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
5792 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5793 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
5794 VERIFY(zio_wait(zio
) == 0);
5798 * Note: this simple function is not inlined to make it easier to dtrace the
5799 * amount of time spent syncing deferred frees.
5802 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
5804 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5805 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
5806 spa_free_sync_cb
, zio
, tx
), ==, 0);
5807 VERIFY0(zio_wait(zio
));
5811 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5813 char *packed
= NULL
;
5818 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5821 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5822 * information. This avoids the dbuf_will_dirty() path and
5823 * saves us a pre-read to get data we don't actually care about.
5825 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
5826 packed
= vmem_alloc(bufsize
, KM_PUSHPAGE
);
5828 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
5830 bzero(packed
+ nvsize
, bufsize
- nvsize
);
5832 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
5834 vmem_free(packed
, bufsize
);
5836 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
5837 dmu_buf_will_dirty(db
, tx
);
5838 *(uint64_t *)db
->db_data
= nvsize
;
5839 dmu_buf_rele(db
, FTAG
);
5843 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
5844 const char *config
, const char *entry
)
5854 * Update the MOS nvlist describing the list of available devices.
5855 * spa_validate_aux() will have already made sure this nvlist is
5856 * valid and the vdevs are labeled appropriately.
5858 if (sav
->sav_object
== 0) {
5859 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
5860 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
5861 sizeof (uint64_t), tx
);
5862 VERIFY(zap_update(spa
->spa_meta_objset
,
5863 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
5864 &sav
->sav_object
, tx
) == 0);
5867 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
5868 if (sav
->sav_count
== 0) {
5869 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
5871 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_PUSHPAGE
);
5872 for (i
= 0; i
< sav
->sav_count
; i
++)
5873 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
5874 B_FALSE
, VDEV_CONFIG_L2CACHE
);
5875 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
5876 sav
->sav_count
) == 0);
5877 for (i
= 0; i
< sav
->sav_count
; i
++)
5878 nvlist_free(list
[i
]);
5879 kmem_free(list
, sav
->sav_count
* sizeof (void *));
5882 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
5883 nvlist_free(nvroot
);
5885 sav
->sav_sync
= B_FALSE
;
5889 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
5893 if (list_is_empty(&spa
->spa_config_dirty_list
))
5896 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5898 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
5899 dmu_tx_get_txg(tx
), B_FALSE
);
5902 * If we're upgrading the spa version then make sure that
5903 * the config object gets updated with the correct version.
5905 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
5906 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5907 spa
->spa_uberblock
.ub_version
);
5909 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5911 if (spa
->spa_config_syncing
)
5912 nvlist_free(spa
->spa_config_syncing
);
5913 spa
->spa_config_syncing
= config
;
5915 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
5919 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
5921 uint64_t *versionp
= arg
;
5922 uint64_t version
= *versionp
;
5923 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5926 * Setting the version is special cased when first creating the pool.
5928 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
5930 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
5931 ASSERT(version
>= spa_version(spa
));
5933 spa
->spa_uberblock
.ub_version
= version
;
5934 vdev_config_dirty(spa
->spa_root_vdev
);
5935 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
5939 * Set zpool properties.
5942 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
5944 nvlist_t
*nvp
= arg
;
5945 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5946 objset_t
*mos
= spa
->spa_meta_objset
;
5947 nvpair_t
*elem
= NULL
;
5949 mutex_enter(&spa
->spa_props_lock
);
5951 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
5953 char *strval
, *fname
;
5955 const char *propname
;
5956 zprop_type_t proptype
;
5957 zfeature_info_t
*feature
;
5959 prop
= zpool_name_to_prop(nvpair_name(elem
));
5960 switch ((int)prop
) {
5963 * We checked this earlier in spa_prop_validate().
5965 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
5967 fname
= strchr(nvpair_name(elem
), '@') + 1;
5968 VERIFY3U(0, ==, zfeature_lookup_name(fname
, &feature
));
5970 spa_feature_enable(spa
, feature
, tx
);
5971 spa_history_log_internal(spa
, "set", tx
,
5972 "%s=enabled", nvpair_name(elem
));
5975 case ZPOOL_PROP_VERSION
:
5976 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
5978 * The version is synced seperatly before other
5979 * properties and should be correct by now.
5981 ASSERT3U(spa_version(spa
), >=, intval
);
5984 case ZPOOL_PROP_ALTROOT
:
5986 * 'altroot' is a non-persistent property. It should
5987 * have been set temporarily at creation or import time.
5989 ASSERT(spa
->spa_root
!= NULL
);
5992 case ZPOOL_PROP_READONLY
:
5993 case ZPOOL_PROP_CACHEFILE
:
5995 * 'readonly' and 'cachefile' are also non-persisitent
5999 case ZPOOL_PROP_COMMENT
:
6000 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
6001 if (spa
->spa_comment
!= NULL
)
6002 spa_strfree(spa
->spa_comment
);
6003 spa
->spa_comment
= spa_strdup(strval
);
6005 * We need to dirty the configuration on all the vdevs
6006 * so that their labels get updated. It's unnecessary
6007 * to do this for pool creation since the vdev's
6008 * configuratoin has already been dirtied.
6010 if (tx
->tx_txg
!= TXG_INITIAL
)
6011 vdev_config_dirty(spa
->spa_root_vdev
);
6012 spa_history_log_internal(spa
, "set", tx
,
6013 "%s=%s", nvpair_name(elem
), strval
);
6017 * Set pool property values in the poolprops mos object.
6019 if (spa
->spa_pool_props_object
== 0) {
6020 spa
->spa_pool_props_object
=
6021 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6022 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6026 /* normalize the property name */
6027 propname
= zpool_prop_to_name(prop
);
6028 proptype
= zpool_prop_get_type(prop
);
6030 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6031 ASSERT(proptype
== PROP_TYPE_STRING
);
6032 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
6033 VERIFY(zap_update(mos
,
6034 spa
->spa_pool_props_object
, propname
,
6035 1, strlen(strval
) + 1, strval
, tx
) == 0);
6036 spa_history_log_internal(spa
, "set", tx
,
6037 "%s=%s", nvpair_name(elem
), strval
);
6038 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6039 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
6041 if (proptype
== PROP_TYPE_INDEX
) {
6043 VERIFY(zpool_prop_index_to_string(
6044 prop
, intval
, &unused
) == 0);
6046 VERIFY(zap_update(mos
,
6047 spa
->spa_pool_props_object
, propname
,
6048 8, 1, &intval
, tx
) == 0);
6049 spa_history_log_internal(spa
, "set", tx
,
6050 "%s=%lld", nvpair_name(elem
), intval
);
6052 ASSERT(0); /* not allowed */
6056 case ZPOOL_PROP_DELEGATION
:
6057 spa
->spa_delegation
= intval
;
6059 case ZPOOL_PROP_BOOTFS
:
6060 spa
->spa_bootfs
= intval
;
6062 case ZPOOL_PROP_FAILUREMODE
:
6063 spa
->spa_failmode
= intval
;
6065 case ZPOOL_PROP_AUTOEXPAND
:
6066 spa
->spa_autoexpand
= intval
;
6067 if (tx
->tx_txg
!= TXG_INITIAL
)
6068 spa_async_request(spa
,
6069 SPA_ASYNC_AUTOEXPAND
);
6071 case ZPOOL_PROP_DEDUPDITTO
:
6072 spa
->spa_dedup_ditto
= intval
;
6081 mutex_exit(&spa
->spa_props_lock
);
6085 * Perform one-time upgrade on-disk changes. spa_version() does not
6086 * reflect the new version this txg, so there must be no changes this
6087 * txg to anything that the upgrade code depends on after it executes.
6088 * Therefore this must be called after dsl_pool_sync() does the sync
6092 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6094 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6096 ASSERT(spa
->spa_sync_pass
== 1);
6098 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6100 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6101 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6102 dsl_pool_create_origin(dp
, tx
);
6104 /* Keeping the origin open increases spa_minref */
6105 spa
->spa_minref
+= 3;
6108 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6109 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6110 dsl_pool_upgrade_clones(dp
, tx
);
6113 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6114 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6115 dsl_pool_upgrade_dir_clones(dp
, tx
);
6117 /* Keeping the freedir open increases spa_minref */
6118 spa
->spa_minref
+= 3;
6121 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6122 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6123 spa_feature_create_zap_objects(spa
, tx
);
6125 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6129 * Sync the specified transaction group. New blocks may be dirtied as
6130 * part of the process, so we iterate until it converges.
6133 spa_sync(spa_t
*spa
, uint64_t txg
)
6135 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6136 objset_t
*mos
= spa
->spa_meta_objset
;
6137 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6138 vdev_t
*rvd
= spa
->spa_root_vdev
;
6144 VERIFY(spa_writeable(spa
));
6147 * Lock out configuration changes.
6149 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6151 spa
->spa_syncing_txg
= txg
;
6152 spa
->spa_sync_pass
= 0;
6155 * If there are any pending vdev state changes, convert them
6156 * into config changes that go out with this transaction group.
6158 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6159 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6161 * We need the write lock here because, for aux vdevs,
6162 * calling vdev_config_dirty() modifies sav_config.
6163 * This is ugly and will become unnecessary when we
6164 * eliminate the aux vdev wart by integrating all vdevs
6165 * into the root vdev tree.
6167 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6168 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6169 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6170 vdev_state_clean(vd
);
6171 vdev_config_dirty(vd
);
6173 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6174 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6176 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6178 tx
= dmu_tx_create_assigned(dp
, txg
);
6180 spa
->spa_sync_starttime
= gethrtime();
6181 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6182 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
6183 spa_deadman
, spa
, TQ_PUSHPAGE
, ddi_get_lbolt() +
6184 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6187 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6188 * set spa_deflate if we have no raid-z vdevs.
6190 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6191 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6194 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6195 vd
= rvd
->vdev_child
[i
];
6196 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6199 if (i
== rvd
->vdev_children
) {
6200 spa
->spa_deflate
= TRUE
;
6201 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6202 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6203 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6208 * If anything has changed in this txg, or if someone is waiting
6209 * for this txg to sync (eg, spa_vdev_remove()), push the
6210 * deferred frees from the previous txg. If not, leave them
6211 * alone so that we don't generate work on an otherwise idle
6214 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
6215 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
6216 !txg_list_empty(&dp
->dp_sync_tasks
, txg
) ||
6217 ((dsl_scan_active(dp
->dp_scan
) ||
6218 txg_sync_waiting(dp
)) && !spa_shutting_down(spa
))) {
6219 spa_sync_deferred_frees(spa
, tx
);
6223 * Iterate to convergence.
6226 int pass
= ++spa
->spa_sync_pass
;
6228 spa_sync_config_object(spa
, tx
);
6229 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6230 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6231 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6232 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6233 spa_errlog_sync(spa
, txg
);
6234 dsl_pool_sync(dp
, txg
);
6236 if (pass
< zfs_sync_pass_deferred_free
) {
6237 spa_sync_frees(spa
, free_bpl
, tx
);
6239 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6240 &spa
->spa_deferred_bpobj
, tx
);
6244 dsl_scan_sync(dp
, tx
);
6246 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6250 spa_sync_upgrades(spa
, tx
);
6252 } while (dmu_objset_is_dirty(mos
, txg
));
6255 * Rewrite the vdev configuration (which includes the uberblock)
6256 * to commit the transaction group.
6258 * If there are no dirty vdevs, we sync the uberblock to a few
6259 * random top-level vdevs that are known to be visible in the
6260 * config cache (see spa_vdev_add() for a complete description).
6261 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6265 * We hold SCL_STATE to prevent vdev open/close/etc.
6266 * while we're attempting to write the vdev labels.
6268 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6270 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6271 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6273 int children
= rvd
->vdev_children
;
6274 int c0
= spa_get_random(children
);
6276 for (c
= 0; c
< children
; c
++) {
6277 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6278 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6280 svd
[svdcount
++] = vd
;
6281 if (svdcount
== SPA_DVAS_PER_BP
)
6284 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
6286 error
= vdev_config_sync(svd
, svdcount
, txg
,
6289 error
= vdev_config_sync(rvd
->vdev_child
,
6290 rvd
->vdev_children
, txg
, B_FALSE
);
6292 error
= vdev_config_sync(rvd
->vdev_child
,
6293 rvd
->vdev_children
, txg
, B_TRUE
);
6297 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6299 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6303 zio_suspend(spa
, NULL
);
6304 zio_resume_wait(spa
);
6308 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6309 spa
->spa_deadman_tqid
= 0;
6312 * Clear the dirty config list.
6314 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6315 vdev_config_clean(vd
);
6318 * Now that the new config has synced transactionally,
6319 * let it become visible to the config cache.
6321 if (spa
->spa_config_syncing
!= NULL
) {
6322 spa_config_set(spa
, spa
->spa_config_syncing
);
6323 spa
->spa_config_txg
= txg
;
6324 spa
->spa_config_syncing
= NULL
;
6327 spa
->spa_ubsync
= spa
->spa_uberblock
;
6329 dsl_pool_sync_done(dp
, txg
);
6332 * Update usable space statistics.
6334 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6335 vdev_sync_done(vd
, txg
);
6337 spa_update_dspace(spa
);
6340 * It had better be the case that we didn't dirty anything
6341 * since vdev_config_sync().
6343 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6344 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6345 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6347 spa
->spa_sync_pass
= 0;
6349 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6351 spa_handle_ignored_writes(spa
);
6354 * If any async tasks have been requested, kick them off.
6356 spa_async_dispatch(spa
);
6360 * Sync all pools. We don't want to hold the namespace lock across these
6361 * operations, so we take a reference on the spa_t and drop the lock during the
6365 spa_sync_allpools(void)
6368 mutex_enter(&spa_namespace_lock
);
6369 while ((spa
= spa_next(spa
)) != NULL
) {
6370 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6371 !spa_writeable(spa
) || spa_suspended(spa
))
6373 spa_open_ref(spa
, FTAG
);
6374 mutex_exit(&spa_namespace_lock
);
6375 txg_wait_synced(spa_get_dsl(spa
), 0);
6376 mutex_enter(&spa_namespace_lock
);
6377 spa_close(spa
, FTAG
);
6379 mutex_exit(&spa_namespace_lock
);
6383 * ==========================================================================
6384 * Miscellaneous routines
6385 * ==========================================================================
6389 * Remove all pools in the system.
6397 * Remove all cached state. All pools should be closed now,
6398 * so every spa in the AVL tree should be unreferenced.
6400 mutex_enter(&spa_namespace_lock
);
6401 while ((spa
= spa_next(NULL
)) != NULL
) {
6403 * Stop async tasks. The async thread may need to detach
6404 * a device that's been replaced, which requires grabbing
6405 * spa_namespace_lock, so we must drop it here.
6407 spa_open_ref(spa
, FTAG
);
6408 mutex_exit(&spa_namespace_lock
);
6409 spa_async_suspend(spa
);
6410 mutex_enter(&spa_namespace_lock
);
6411 spa_close(spa
, FTAG
);
6413 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6415 spa_deactivate(spa
);
6419 mutex_exit(&spa_namespace_lock
);
6423 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6428 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6432 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6433 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6434 if (vd
->vdev_guid
== guid
)
6438 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6439 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6440 if (vd
->vdev_guid
== guid
)
6449 spa_upgrade(spa_t
*spa
, uint64_t version
)
6451 ASSERT(spa_writeable(spa
));
6453 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6456 * This should only be called for a non-faulted pool, and since a
6457 * future version would result in an unopenable pool, this shouldn't be
6460 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
6461 ASSERT(version
>= spa
->spa_uberblock
.ub_version
);
6463 spa
->spa_uberblock
.ub_version
= version
;
6464 vdev_config_dirty(spa
->spa_root_vdev
);
6466 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6468 txg_wait_synced(spa_get_dsl(spa
), 0);
6472 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6476 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6478 for (i
= 0; i
< sav
->sav_count
; i
++)
6479 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6482 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6483 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6484 &spareguid
) == 0 && spareguid
== guid
)
6492 * Check if a pool has an active shared spare device.
6493 * Note: reference count of an active spare is 2, as a spare and as a replace
6496 spa_has_active_shared_spare(spa_t
*spa
)
6500 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6502 for (i
= 0; i
< sav
->sav_count
; i
++) {
6503 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6504 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6513 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6514 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6515 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6516 * or zdb as real changes.
6519 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6522 zfs_ereport_post(name
, spa
, vd
, NULL
, 0, 0);
6526 #if defined(_KERNEL) && defined(HAVE_SPL)
6527 /* state manipulation functions */
6528 EXPORT_SYMBOL(spa_open
);
6529 EXPORT_SYMBOL(spa_open_rewind
);
6530 EXPORT_SYMBOL(spa_get_stats
);
6531 EXPORT_SYMBOL(spa_create
);
6532 EXPORT_SYMBOL(spa_import_rootpool
);
6533 EXPORT_SYMBOL(spa_import
);
6534 EXPORT_SYMBOL(spa_tryimport
);
6535 EXPORT_SYMBOL(spa_destroy
);
6536 EXPORT_SYMBOL(spa_export
);
6537 EXPORT_SYMBOL(spa_reset
);
6538 EXPORT_SYMBOL(spa_async_request
);
6539 EXPORT_SYMBOL(spa_async_suspend
);
6540 EXPORT_SYMBOL(spa_async_resume
);
6541 EXPORT_SYMBOL(spa_inject_addref
);
6542 EXPORT_SYMBOL(spa_inject_delref
);
6543 EXPORT_SYMBOL(spa_scan_stat_init
);
6544 EXPORT_SYMBOL(spa_scan_get_stats
);
6546 /* device maniion */
6547 EXPORT_SYMBOL(spa_vdev_add
);
6548 EXPORT_SYMBOL(spa_vdev_attach
);
6549 EXPORT_SYMBOL(spa_vdev_detach
);
6550 EXPORT_SYMBOL(spa_vdev_remove
);
6551 EXPORT_SYMBOL(spa_vdev_setpath
);
6552 EXPORT_SYMBOL(spa_vdev_setfru
);
6553 EXPORT_SYMBOL(spa_vdev_split_mirror
);
6555 /* spare statech is global across all pools) */
6556 EXPORT_SYMBOL(spa_spare_add
);
6557 EXPORT_SYMBOL(spa_spare_remove
);
6558 EXPORT_SYMBOL(spa_spare_exists
);
6559 EXPORT_SYMBOL(spa_spare_activate
);
6561 /* L2ARC statech is global across all pools) */
6562 EXPORT_SYMBOL(spa_l2cache_add
);
6563 EXPORT_SYMBOL(spa_l2cache_remove
);
6564 EXPORT_SYMBOL(spa_l2cache_exists
);
6565 EXPORT_SYMBOL(spa_l2cache_activate
);
6566 EXPORT_SYMBOL(spa_l2cache_drop
);
6569 EXPORT_SYMBOL(spa_scan
);
6570 EXPORT_SYMBOL(spa_scan_stop
);
6573 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
6574 EXPORT_SYMBOL(spa_sync_allpools
);
6577 EXPORT_SYMBOL(spa_prop_set
);
6578 EXPORT_SYMBOL(spa_prop_get
);
6579 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
6581 /* asynchronous event notification */
6582 EXPORT_SYMBOL(spa_event_notify
);