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 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
34 #include <sys/zfs_context.h>
35 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
38 #include <sys/zio_checksum.h>
40 #include <sys/dmu_tx.h>
44 #include <sys/vdev_impl.h>
45 #include <sys/vdev_disk.h>
46 #include <sys/metaslab.h>
47 #include <sys/metaslab_impl.h>
48 #include <sys/uberblock_impl.h>
51 #include <sys/dmu_traverse.h>
52 #include <sys/dmu_objset.h>
53 #include <sys/unique.h>
54 #include <sys/dsl_pool.h>
55 #include <sys/dsl_dataset.h>
56 #include <sys/dsl_dir.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dsl_synctask.h>
59 #include <sys/fs/zfs.h>
61 #include <sys/callb.h>
62 #include <sys/systeminfo.h>
63 #include <sys/spa_boot.h>
64 #include <sys/zfs_ioctl.h>
65 #include <sys/dsl_scan.h>
66 #include <sys/zfeature.h>
67 #include <sys/dsl_destroy.h>
71 #include <sys/bootprops.h>
72 #include <sys/callb.h>
73 #include <sys/cpupart.h>
75 #include <sys/sysdc.h>
80 #include "zfs_comutil.h"
82 typedef enum zti_modes
{
83 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
84 ZTI_MODE_ONLINE_PERCENT
, /* value is % of online CPUs */
85 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
86 ZTI_MODE_NULL
, /* don't create a taskq */
90 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
91 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
92 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
93 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
95 #define ZTI_N(n) ZTI_P(n, 1)
96 #define ZTI_ONE ZTI_N(1)
98 typedef struct zio_taskq_info
{
104 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
105 "iss", "iss_h", "int", "int_h"
109 * This table defines the taskq settings for each ZFS I/O type. When
110 * initializing a pool, we use this table to create an appropriately sized
111 * taskq. Some operations are low volume and therefore have a small, static
112 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
113 * macros. Other operations process a large amount of data; the ZTI_BATCH
114 * macro causes us to create a taskq oriented for throughput. Some operations
115 * are so high frequency and short-lived that the taskq itself can become a a
116 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
117 * additional degree of parallelism specified by the number of threads per-
118 * taskq and the number of taskqs; when dispatching an event in this case, the
119 * particular taskq is chosen at random.
121 * The different taskq priorities are to handle the different contexts (issue
122 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
123 * need to be handled with minimum delay.
125 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
126 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
127 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
128 { ZTI_N(8), ZTI_NULL
, ZTI_BATCH
, ZTI_NULL
}, /* READ */
129 { ZTI_BATCH
, ZTI_N(5), ZTI_N(16), ZTI_N(5) }, /* WRITE */
130 { ZTI_P(4, 8), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
131 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
132 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
135 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
136 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
137 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
138 static inline int spa_load_impl(spa_t
*spa
, uint64_t, nvlist_t
*config
,
139 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
141 static void spa_vdev_resilver_done(spa_t
*spa
);
143 uint_t zio_taskq_batch_pct
= 100; /* 1 thread per cpu in pset */
144 id_t zio_taskq_psrset_bind
= PS_NONE
;
145 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
146 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
148 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
151 * This (illegal) pool name is used when temporarily importing a spa_t in order
152 * to get the vdev stats associated with the imported devices.
154 #define TRYIMPORT_NAME "$import"
157 * ==========================================================================
158 * SPA properties routines
159 * ==========================================================================
163 * Add a (source=src, propname=propval) list to an nvlist.
166 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
167 uint64_t intval
, zprop_source_t src
)
169 const char *propname
= zpool_prop_to_name(prop
);
172 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
173 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
176 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
178 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
180 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
181 nvlist_free(propval
);
185 * Get property values from the spa configuration.
188 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
190 vdev_t
*rvd
= spa
->spa_root_vdev
;
191 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
195 uint64_t cap
, version
;
196 zprop_source_t src
= ZPROP_SRC_NONE
;
197 spa_config_dirent_t
*dp
;
200 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
203 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
204 size
= metaslab_class_get_space(spa_normal_class(spa
));
205 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
206 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
207 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
208 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
212 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
213 vdev_t
*tvd
= rvd
->vdev_child
[c
];
214 space
+= tvd
->vdev_max_asize
- tvd
->vdev_asize
;
216 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
, space
,
219 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
220 (spa_mode(spa
) == FREAD
), src
);
222 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
223 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
225 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
226 ddt_get_pool_dedup_ratio(spa
), src
);
228 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
229 rvd
->vdev_state
, src
);
231 version
= spa_version(spa
);
232 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
))
233 src
= ZPROP_SRC_DEFAULT
;
235 src
= ZPROP_SRC_LOCAL
;
236 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
240 dsl_dir_t
*freedir
= pool
->dp_free_dir
;
243 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
244 * when opening pools before this version freedir will be NULL.
246 if (freedir
!= NULL
) {
247 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
248 freedir
->dd_phys
->dd_used_bytes
, src
);
250 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
255 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
257 if (spa
->spa_comment
!= NULL
) {
258 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
262 if (spa
->spa_root
!= NULL
)
263 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
266 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
267 if (dp
->scd_path
== NULL
) {
268 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
269 "none", 0, ZPROP_SRC_LOCAL
);
270 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
271 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
272 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
278 * Get zpool property values.
281 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
283 objset_t
*mos
= spa
->spa_meta_objset
;
288 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_PUSHPAGE
);
292 mutex_enter(&spa
->spa_props_lock
);
295 * Get properties from the spa config.
297 spa_prop_get_config(spa
, nvp
);
299 /* If no pool property object, no more prop to get. */
300 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
301 mutex_exit(&spa
->spa_props_lock
);
306 * Get properties from the MOS pool property object.
308 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
309 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
310 zap_cursor_advance(&zc
)) {
313 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
316 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
319 switch (za
.za_integer_length
) {
321 /* integer property */
322 if (za
.za_first_integer
!=
323 zpool_prop_default_numeric(prop
))
324 src
= ZPROP_SRC_LOCAL
;
326 if (prop
== ZPOOL_PROP_BOOTFS
) {
328 dsl_dataset_t
*ds
= NULL
;
330 dp
= spa_get_dsl(spa
);
331 dsl_pool_config_enter(dp
, FTAG
);
332 if ((err
= dsl_dataset_hold_obj(dp
,
333 za
.za_first_integer
, FTAG
, &ds
))) {
334 dsl_pool_config_exit(dp
, FTAG
);
339 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1,
341 dsl_dataset_name(ds
, strval
);
342 dsl_dataset_rele(ds
, FTAG
);
343 dsl_pool_config_exit(dp
, FTAG
);
346 intval
= za
.za_first_integer
;
349 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
353 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1);
358 /* string property */
359 strval
= kmem_alloc(za
.za_num_integers
, KM_PUSHPAGE
);
360 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
361 za
.za_name
, 1, za
.za_num_integers
, strval
);
363 kmem_free(strval
, za
.za_num_integers
);
366 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
367 kmem_free(strval
, za
.za_num_integers
);
374 zap_cursor_fini(&zc
);
375 mutex_exit(&spa
->spa_props_lock
);
377 if (err
&& err
!= ENOENT
) {
387 * Validate the given pool properties nvlist and modify the list
388 * for the property values to be set.
391 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
394 int error
= 0, reset_bootfs
= 0;
396 boolean_t has_feature
= B_FALSE
;
399 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
401 char *strval
, *slash
, *check
, *fname
;
402 const char *propname
= nvpair_name(elem
);
403 zpool_prop_t prop
= zpool_name_to_prop(propname
);
407 if (!zpool_prop_feature(propname
)) {
413 * Sanitize the input.
415 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
420 if (nvpair_value_uint64(elem
, &intval
) != 0) {
430 fname
= strchr(propname
, '@') + 1;
431 if (zfeature_lookup_name(fname
, NULL
) != 0) {
436 has_feature
= B_TRUE
;
439 case ZPOOL_PROP_VERSION
:
440 error
= nvpair_value_uint64(elem
, &intval
);
442 (intval
< spa_version(spa
) ||
443 intval
> SPA_VERSION_BEFORE_FEATURES
||
448 case ZPOOL_PROP_DELEGATION
:
449 case ZPOOL_PROP_AUTOREPLACE
:
450 case ZPOOL_PROP_LISTSNAPS
:
451 case ZPOOL_PROP_AUTOEXPAND
:
452 error
= nvpair_value_uint64(elem
, &intval
);
453 if (!error
&& intval
> 1)
457 case ZPOOL_PROP_BOOTFS
:
459 * If the pool version is less than SPA_VERSION_BOOTFS,
460 * or the pool is still being created (version == 0),
461 * the bootfs property cannot be set.
463 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
469 * Make sure the vdev config is bootable
471 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
478 error
= nvpair_value_string(elem
, &strval
);
484 if (strval
== NULL
|| strval
[0] == '\0') {
485 objnum
= zpool_prop_default_numeric(
490 if ((error
= dmu_objset_hold(strval
,FTAG
,&os
)))
493 /* Must be ZPL and not gzip compressed. */
495 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
498 dsl_prop_get_int_ds(dmu_objset_ds(os
),
499 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
501 !BOOTFS_COMPRESS_VALID(compress
)) {
504 objnum
= dmu_objset_id(os
);
506 dmu_objset_rele(os
, FTAG
);
510 case ZPOOL_PROP_FAILUREMODE
:
511 error
= nvpair_value_uint64(elem
, &intval
);
512 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
513 intval
> ZIO_FAILURE_MODE_PANIC
))
517 * This is a special case which only occurs when
518 * the pool has completely failed. This allows
519 * the user to change the in-core failmode property
520 * without syncing it out to disk (I/Os might
521 * currently be blocked). We do this by returning
522 * EIO to the caller (spa_prop_set) to trick it
523 * into thinking we encountered a property validation
526 if (!error
&& spa_suspended(spa
)) {
527 spa
->spa_failmode
= intval
;
532 case ZPOOL_PROP_CACHEFILE
:
533 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
536 if (strval
[0] == '\0')
539 if (strcmp(strval
, "none") == 0)
542 if (strval
[0] != '/') {
547 slash
= strrchr(strval
, '/');
548 ASSERT(slash
!= NULL
);
550 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
551 strcmp(slash
, "/..") == 0)
555 case ZPOOL_PROP_COMMENT
:
556 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
558 for (check
= strval
; *check
!= '\0'; check
++) {
559 if (!isprint(*check
)) {
565 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
569 case ZPOOL_PROP_DEDUPDITTO
:
570 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
573 error
= nvpair_value_uint64(elem
, &intval
);
575 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
587 if (!error
&& reset_bootfs
) {
588 error
= nvlist_remove(props
,
589 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
592 error
= nvlist_add_uint64(props
,
593 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
601 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
604 spa_config_dirent_t
*dp
;
606 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
610 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
613 if (cachefile
[0] == '\0')
614 dp
->scd_path
= spa_strdup(spa_config_path
);
615 else if (strcmp(cachefile
, "none") == 0)
618 dp
->scd_path
= spa_strdup(cachefile
);
620 list_insert_head(&spa
->spa_config_list
, dp
);
622 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
626 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
629 nvpair_t
*elem
= NULL
;
630 boolean_t need_sync
= B_FALSE
;
632 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
635 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
636 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
638 if (prop
== ZPOOL_PROP_CACHEFILE
||
639 prop
== ZPOOL_PROP_ALTROOT
||
640 prop
== ZPOOL_PROP_READONLY
)
643 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPROP_INVAL
) {
646 if (prop
== ZPOOL_PROP_VERSION
) {
647 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
649 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
650 ver
= SPA_VERSION_FEATURES
;
654 /* Save time if the version is already set. */
655 if (ver
== spa_version(spa
))
659 * In addition to the pool directory object, we might
660 * create the pool properties object, the features for
661 * read object, the features for write object, or the
662 * feature descriptions object.
664 error
= dsl_sync_task(spa
->spa_name
, NULL
,
665 spa_sync_version
, &ver
, 6);
676 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
684 * If the bootfs property value is dsobj, clear it.
687 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
689 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
690 VERIFY(zap_remove(spa
->spa_meta_objset
,
691 spa
->spa_pool_props_object
,
692 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
699 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
701 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
702 vdev_t
*rvd
= spa
->spa_root_vdev
;
704 ASSERTV(uint64_t *newguid
= arg
);
706 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
707 vdev_state
= rvd
->vdev_state
;
708 spa_config_exit(spa
, SCL_STATE
, FTAG
);
710 if (vdev_state
!= VDEV_STATE_HEALTHY
)
713 ASSERT3U(spa_guid(spa
), !=, *newguid
);
719 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
721 uint64_t *newguid
= arg
;
722 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
724 vdev_t
*rvd
= spa
->spa_root_vdev
;
726 oldguid
= spa_guid(spa
);
728 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
729 rvd
->vdev_guid
= *newguid
;
730 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
731 vdev_config_dirty(rvd
);
732 spa_config_exit(spa
, SCL_STATE
, FTAG
);
734 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
739 * Change the GUID for the pool. This is done so that we can later
740 * re-import a pool built from a clone of our own vdevs. We will modify
741 * the root vdev's guid, our own pool guid, and then mark all of our
742 * vdevs dirty. Note that we must make sure that all our vdevs are
743 * online when we do this, or else any vdevs that weren't present
744 * would be orphaned from our pool. We are also going to issue a
745 * sysevent to update any watchers.
748 spa_change_guid(spa_t
*spa
)
753 mutex_enter(&spa_namespace_lock
);
754 guid
= spa_generate_guid(NULL
);
756 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
757 spa_change_guid_sync
, &guid
, 5);
760 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
761 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_REGUID
);
764 mutex_exit(&spa_namespace_lock
);
770 * ==========================================================================
771 * SPA state manipulation (open/create/destroy/import/export)
772 * ==========================================================================
776 spa_error_entry_compare(const void *a
, const void *b
)
778 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
779 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
782 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
783 sizeof (zbookmark_t
));
794 * Utility function which retrieves copies of the current logs and
795 * re-initializes them in the process.
798 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
800 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
802 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
803 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
805 avl_create(&spa
->spa_errlist_scrub
,
806 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
807 offsetof(spa_error_entry_t
, se_avl
));
808 avl_create(&spa
->spa_errlist_last
,
809 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
810 offsetof(spa_error_entry_t
, se_avl
));
814 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
816 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
817 enum zti_modes mode
= ztip
->zti_mode
;
818 uint_t value
= ztip
->zti_value
;
819 uint_t count
= ztip
->zti_count
;
820 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
823 boolean_t batch
= B_FALSE
;
825 if (mode
== ZTI_MODE_NULL
) {
827 tqs
->stqs_taskq
= NULL
;
831 ASSERT3U(count
, >, 0);
833 tqs
->stqs_count
= count
;
834 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
836 for (i
= 0; i
< count
; i
++) {
841 ASSERT3U(value
, >=, 1);
842 value
= MAX(value
, 1);
847 flags
|= TASKQ_THREADS_CPU_PCT
;
848 value
= zio_taskq_batch_pct
;
851 case ZTI_MODE_ONLINE_PERCENT
:
852 flags
|= TASKQ_THREADS_CPU_PCT
;
856 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
858 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
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 tq
= taskq_create_proc(name
, value
, maxclsyspri
, 50,
878 INT_MAX
, spa
->spa_proc
, flags
);
881 tqs
->stqs_taskq
[i
] = tq
;
886 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
888 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
891 if (tqs
->stqs_taskq
== NULL
) {
892 ASSERT3U(tqs
->stqs_count
, ==, 0);
896 for (i
= 0; i
< tqs
->stqs_count
; i
++) {
897 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
898 taskq_destroy(tqs
->stqs_taskq
[i
]);
901 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
902 tqs
->stqs_taskq
= NULL
;
906 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
907 * Note that a type may have multiple discrete taskqs to avoid lock contention
908 * on the taskq itself. In that case we choose which taskq at random by using
909 * the low bits of gethrtime().
912 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
913 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
915 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
918 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
919 ASSERT3U(tqs
->stqs_count
, !=, 0);
921 if (tqs
->stqs_count
== 1) {
922 tq
= tqs
->stqs_taskq
[0];
924 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
927 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
931 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
934 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
935 task_func_t
*func
, void *arg
, uint_t flags
)
937 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
941 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
942 ASSERT3U(tqs
->stqs_count
, !=, 0);
944 if (tqs
->stqs_count
== 1) {
945 tq
= tqs
->stqs_taskq
[0];
947 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
950 id
= taskq_dispatch(tq
, func
, arg
, flags
);
952 taskq_wait_id(tq
, id
);
956 spa_create_zio_taskqs(spa_t
*spa
)
960 for (t
= 0; t
< ZIO_TYPES
; t
++) {
961 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
962 spa_taskqs_init(spa
, t
, q
);
967 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
969 spa_thread(void *arg
)
974 user_t
*pu
= PTOU(curproc
);
976 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
979 ASSERT(curproc
!= &p0
);
980 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
981 "zpool-%s", spa
->spa_name
);
982 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
984 /* bind this thread to the requested psrset */
985 if (zio_taskq_psrset_bind
!= PS_NONE
) {
987 mutex_enter(&cpu_lock
);
988 mutex_enter(&pidlock
);
989 mutex_enter(&curproc
->p_lock
);
991 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
992 0, NULL
, NULL
) == 0) {
993 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
996 "Couldn't bind process for zfs pool \"%s\" to "
997 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1000 mutex_exit(&curproc
->p_lock
);
1001 mutex_exit(&pidlock
);
1002 mutex_exit(&cpu_lock
);
1006 if (zio_taskq_sysdc
) {
1007 sysdc_thread_enter(curthread
, 100, 0);
1010 spa
->spa_proc
= curproc
;
1011 spa
->spa_did
= curthread
->t_did
;
1013 spa_create_zio_taskqs(spa
);
1015 mutex_enter(&spa
->spa_proc_lock
);
1016 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1018 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1019 cv_broadcast(&spa
->spa_proc_cv
);
1021 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1022 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1023 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1024 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1026 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1027 spa
->spa_proc_state
= SPA_PROC_GONE
;
1028 spa
->spa_proc
= &p0
;
1029 cv_broadcast(&spa
->spa_proc_cv
);
1030 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1032 mutex_enter(&curproc
->p_lock
);
1038 * Activate an uninitialized pool.
1041 spa_activate(spa_t
*spa
, int mode
)
1043 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1045 spa
->spa_state
= POOL_STATE_ACTIVE
;
1046 spa
->spa_mode
= mode
;
1048 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1049 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1051 /* Try to create a covering process */
1052 mutex_enter(&spa
->spa_proc_lock
);
1053 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1054 ASSERT(spa
->spa_proc
== &p0
);
1057 #ifdef HAVE_SPA_THREAD
1058 /* Only create a process if we're going to be around a while. */
1059 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1060 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1062 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1063 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1064 cv_wait(&spa
->spa_proc_cv
,
1065 &spa
->spa_proc_lock
);
1067 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1068 ASSERT(spa
->spa_proc
!= &p0
);
1069 ASSERT(spa
->spa_did
!= 0);
1073 "Couldn't create process for zfs pool \"%s\"\n",
1078 #endif /* HAVE_SPA_THREAD */
1079 mutex_exit(&spa
->spa_proc_lock
);
1081 /* If we didn't create a process, we need to create our taskqs. */
1082 if (spa
->spa_proc
== &p0
) {
1083 spa_create_zio_taskqs(spa
);
1086 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1087 offsetof(vdev_t
, vdev_config_dirty_node
));
1088 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1089 offsetof(vdev_t
, vdev_state_dirty_node
));
1091 txg_list_create(&spa
->spa_vdev_txg_list
,
1092 offsetof(struct vdev
, vdev_txg_node
));
1094 avl_create(&spa
->spa_errlist_scrub
,
1095 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1096 offsetof(spa_error_entry_t
, se_avl
));
1097 avl_create(&spa
->spa_errlist_last
,
1098 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1099 offsetof(spa_error_entry_t
, se_avl
));
1103 * Opposite of spa_activate().
1106 spa_deactivate(spa_t
*spa
)
1110 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1111 ASSERT(spa
->spa_dsl_pool
== NULL
);
1112 ASSERT(spa
->spa_root_vdev
== NULL
);
1113 ASSERT(spa
->spa_async_zio_root
== NULL
);
1114 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1116 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1118 list_destroy(&spa
->spa_config_dirty_list
);
1119 list_destroy(&spa
->spa_state_dirty_list
);
1121 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
1123 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1124 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1125 spa_taskqs_fini(spa
, t
, q
);
1129 metaslab_class_destroy(spa
->spa_normal_class
);
1130 spa
->spa_normal_class
= NULL
;
1132 metaslab_class_destroy(spa
->spa_log_class
);
1133 spa
->spa_log_class
= NULL
;
1136 * If this was part of an import or the open otherwise failed, we may
1137 * still have errors left in the queues. Empty them just in case.
1139 spa_errlog_drain(spa
);
1141 avl_destroy(&spa
->spa_errlist_scrub
);
1142 avl_destroy(&spa
->spa_errlist_last
);
1144 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1146 mutex_enter(&spa
->spa_proc_lock
);
1147 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1148 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1149 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1150 cv_broadcast(&spa
->spa_proc_cv
);
1151 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1152 ASSERT(spa
->spa_proc
!= &p0
);
1153 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1155 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1156 spa
->spa_proc_state
= SPA_PROC_NONE
;
1158 ASSERT(spa
->spa_proc
== &p0
);
1159 mutex_exit(&spa
->spa_proc_lock
);
1162 * We want to make sure spa_thread() has actually exited the ZFS
1163 * module, so that the module can't be unloaded out from underneath
1166 if (spa
->spa_did
!= 0) {
1167 thread_join(spa
->spa_did
);
1173 * Verify a pool configuration, and construct the vdev tree appropriately. This
1174 * will create all the necessary vdevs in the appropriate layout, with each vdev
1175 * in the CLOSED state. This will prep the pool before open/creation/import.
1176 * All vdev validation is done by the vdev_alloc() routine.
1179 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1180 uint_t id
, int atype
)
1187 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1190 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1193 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1196 if (error
== ENOENT
)
1205 for (c
= 0; c
< children
; c
++) {
1207 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1215 ASSERT(*vdp
!= NULL
);
1221 * Opposite of spa_load().
1224 spa_unload(spa_t
*spa
)
1228 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1233 spa_async_suspend(spa
);
1238 if (spa
->spa_sync_on
) {
1239 txg_sync_stop(spa
->spa_dsl_pool
);
1240 spa
->spa_sync_on
= B_FALSE
;
1244 * Wait for any outstanding async I/O to complete.
1246 if (spa
->spa_async_zio_root
!= NULL
) {
1247 (void) zio_wait(spa
->spa_async_zio_root
);
1248 spa
->spa_async_zio_root
= NULL
;
1251 bpobj_close(&spa
->spa_deferred_bpobj
);
1254 * Close the dsl pool.
1256 if (spa
->spa_dsl_pool
) {
1257 dsl_pool_close(spa
->spa_dsl_pool
);
1258 spa
->spa_dsl_pool
= NULL
;
1259 spa
->spa_meta_objset
= NULL
;
1264 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1267 * Drop and purge level 2 cache
1269 spa_l2cache_drop(spa
);
1274 if (spa
->spa_root_vdev
)
1275 vdev_free(spa
->spa_root_vdev
);
1276 ASSERT(spa
->spa_root_vdev
== NULL
);
1278 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1279 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1280 if (spa
->spa_spares
.sav_vdevs
) {
1281 kmem_free(spa
->spa_spares
.sav_vdevs
,
1282 spa
->spa_spares
.sav_count
* sizeof (void *));
1283 spa
->spa_spares
.sav_vdevs
= NULL
;
1285 if (spa
->spa_spares
.sav_config
) {
1286 nvlist_free(spa
->spa_spares
.sav_config
);
1287 spa
->spa_spares
.sav_config
= NULL
;
1289 spa
->spa_spares
.sav_count
= 0;
1291 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1292 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1293 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1295 if (spa
->spa_l2cache
.sav_vdevs
) {
1296 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1297 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1298 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1300 if (spa
->spa_l2cache
.sav_config
) {
1301 nvlist_free(spa
->spa_l2cache
.sav_config
);
1302 spa
->spa_l2cache
.sav_config
= NULL
;
1304 spa
->spa_l2cache
.sav_count
= 0;
1306 spa
->spa_async_suspended
= 0;
1308 if (spa
->spa_comment
!= NULL
) {
1309 spa_strfree(spa
->spa_comment
);
1310 spa
->spa_comment
= NULL
;
1313 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1317 * Load (or re-load) the current list of vdevs describing the active spares for
1318 * this pool. When this is called, we have some form of basic information in
1319 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1320 * then re-generate a more complete list including status information.
1323 spa_load_spares(spa_t
*spa
)
1330 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1333 * First, close and free any existing spare vdevs.
1335 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1336 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1338 /* Undo the call to spa_activate() below */
1339 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1340 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1341 spa_spare_remove(tvd
);
1346 if (spa
->spa_spares
.sav_vdevs
)
1347 kmem_free(spa
->spa_spares
.sav_vdevs
,
1348 spa
->spa_spares
.sav_count
* sizeof (void *));
1350 if (spa
->spa_spares
.sav_config
== NULL
)
1353 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1354 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1356 spa
->spa_spares
.sav_count
= (int)nspares
;
1357 spa
->spa_spares
.sav_vdevs
= NULL
;
1363 * Construct the array of vdevs, opening them to get status in the
1364 * process. For each spare, there is potentially two different vdev_t
1365 * structures associated with it: one in the list of spares (used only
1366 * for basic validation purposes) and one in the active vdev
1367 * configuration (if it's spared in). During this phase we open and
1368 * validate each vdev on the spare list. If the vdev also exists in the
1369 * active configuration, then we also mark this vdev as an active spare.
1371 spa
->spa_spares
.sav_vdevs
= kmem_alloc(nspares
* sizeof (void *),
1373 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1374 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1375 VDEV_ALLOC_SPARE
) == 0);
1378 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1380 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1381 B_FALSE
)) != NULL
) {
1382 if (!tvd
->vdev_isspare
)
1386 * We only mark the spare active if we were successfully
1387 * able to load the vdev. Otherwise, importing a pool
1388 * with a bad active spare would result in strange
1389 * behavior, because multiple pool would think the spare
1390 * is actively in use.
1392 * There is a vulnerability here to an equally bizarre
1393 * circumstance, where a dead active spare is later
1394 * brought back to life (onlined or otherwise). Given
1395 * the rarity of this scenario, and the extra complexity
1396 * it adds, we ignore the possibility.
1398 if (!vdev_is_dead(tvd
))
1399 spa_spare_activate(tvd
);
1403 vd
->vdev_aux
= &spa
->spa_spares
;
1405 if (vdev_open(vd
) != 0)
1408 if (vdev_validate_aux(vd
) == 0)
1413 * Recompute the stashed list of spares, with status information
1416 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1417 DATA_TYPE_NVLIST_ARRAY
) == 0);
1419 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1421 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1422 spares
[i
] = vdev_config_generate(spa
,
1423 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1424 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1425 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1426 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1427 nvlist_free(spares
[i
]);
1428 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1432 * Load (or re-load) the current list of vdevs describing the active l2cache for
1433 * this pool. When this is called, we have some form of basic information in
1434 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1435 * then re-generate a more complete list including status information.
1436 * Devices which are already active have their details maintained, and are
1440 spa_load_l2cache(spa_t
*spa
)
1444 int i
, j
, oldnvdevs
;
1446 vdev_t
*vd
, **oldvdevs
, **newvdevs
= NULL
;
1447 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1449 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1451 if (sav
->sav_config
!= NULL
) {
1452 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1453 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1454 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_PUSHPAGE
);
1459 oldvdevs
= sav
->sav_vdevs
;
1460 oldnvdevs
= sav
->sav_count
;
1461 sav
->sav_vdevs
= NULL
;
1465 * Process new nvlist of vdevs.
1467 for (i
= 0; i
< nl2cache
; i
++) {
1468 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1472 for (j
= 0; j
< oldnvdevs
; j
++) {
1474 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1476 * Retain previous vdev for add/remove ops.
1484 if (newvdevs
[i
] == NULL
) {
1488 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1489 VDEV_ALLOC_L2CACHE
) == 0);
1494 * Commit this vdev as an l2cache device,
1495 * even if it fails to open.
1497 spa_l2cache_add(vd
);
1502 spa_l2cache_activate(vd
);
1504 if (vdev_open(vd
) != 0)
1507 (void) vdev_validate_aux(vd
);
1509 if (!vdev_is_dead(vd
))
1510 l2arc_add_vdev(spa
, vd
);
1515 * Purge vdevs that were dropped
1517 for (i
= 0; i
< oldnvdevs
; i
++) {
1522 ASSERT(vd
->vdev_isl2cache
);
1524 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1525 pool
!= 0ULL && l2arc_vdev_present(vd
))
1526 l2arc_remove_vdev(vd
);
1527 vdev_clear_stats(vd
);
1533 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1535 if (sav
->sav_config
== NULL
)
1538 sav
->sav_vdevs
= newvdevs
;
1539 sav
->sav_count
= (int)nl2cache
;
1542 * Recompute the stashed list of l2cache devices, with status
1543 * information this time.
1545 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1546 DATA_TYPE_NVLIST_ARRAY
) == 0);
1548 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_PUSHPAGE
);
1549 for (i
= 0; i
< sav
->sav_count
; i
++)
1550 l2cache
[i
] = vdev_config_generate(spa
,
1551 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1552 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1553 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1555 for (i
= 0; i
< sav
->sav_count
; i
++)
1556 nvlist_free(l2cache
[i
]);
1558 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1562 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1565 char *packed
= NULL
;
1570 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1574 nvsize
= *(uint64_t *)db
->db_data
;
1575 dmu_buf_rele(db
, FTAG
);
1577 packed
= kmem_alloc(nvsize
, KM_PUSHPAGE
| KM_NODEBUG
);
1578 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1581 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1582 kmem_free(packed
, nvsize
);
1588 * Checks to see if the given vdev could not be opened, in which case we post a
1589 * sysevent to notify the autoreplace code that the device has been removed.
1592 spa_check_removed(vdev_t
*vd
)
1596 for (c
= 0; c
< vd
->vdev_children
; c
++)
1597 spa_check_removed(vd
->vdev_child
[c
]);
1599 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
)) {
1600 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE
,
1601 vd
->vdev_spa
, vd
, NULL
, 0, 0);
1602 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_CHECK
);
1607 * Validate the current config against the MOS config
1610 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1612 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1616 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1618 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1619 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1621 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1624 * If we're doing a normal import, then build up any additional
1625 * diagnostic information about missing devices in this config.
1626 * We'll pass this up to the user for further processing.
1628 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1629 nvlist_t
**child
, *nv
;
1632 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1634 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
1636 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1637 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1638 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1640 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1641 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1643 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1648 VERIFY(nvlist_add_nvlist_array(nv
,
1649 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1650 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1651 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1653 for (i
= 0; i
< idx
; i
++)
1654 nvlist_free(child
[i
]);
1657 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1661 * Compare the root vdev tree with the information we have
1662 * from the MOS config (mrvd). Check each top-level vdev
1663 * with the corresponding MOS config top-level (mtvd).
1665 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1666 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1667 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1670 * Resolve any "missing" vdevs in the current configuration.
1671 * If we find that the MOS config has more accurate information
1672 * about the top-level vdev then use that vdev instead.
1674 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1675 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1677 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1681 * Device specific actions.
1683 if (mtvd
->vdev_islog
) {
1684 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1687 * XXX - once we have 'readonly' pool
1688 * support we should be able to handle
1689 * missing data devices by transitioning
1690 * the pool to readonly.
1696 * Swap the missing vdev with the data we were
1697 * able to obtain from the MOS config.
1699 vdev_remove_child(rvd
, tvd
);
1700 vdev_remove_child(mrvd
, mtvd
);
1702 vdev_add_child(rvd
, mtvd
);
1703 vdev_add_child(mrvd
, tvd
);
1705 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1707 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1710 } else if (mtvd
->vdev_islog
) {
1712 * Load the slog device's state from the MOS config
1713 * since it's possible that the label does not
1714 * contain the most up-to-date information.
1716 vdev_load_log_state(tvd
, mtvd
);
1721 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1724 * Ensure we were able to validate the config.
1726 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1730 * Check for missing log devices
1733 spa_check_logs(spa_t
*spa
)
1735 boolean_t rv
= B_FALSE
;
1737 switch (spa
->spa_log_state
) {
1740 case SPA_LOG_MISSING
:
1741 /* need to recheck in case slog has been restored */
1742 case SPA_LOG_UNKNOWN
:
1743 rv
= (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
,
1744 NULL
, DS_FIND_CHILDREN
) != 0);
1746 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1753 spa_passivate_log(spa_t
*spa
)
1755 vdev_t
*rvd
= spa
->spa_root_vdev
;
1756 boolean_t slog_found
= B_FALSE
;
1759 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1761 if (!spa_has_slogs(spa
))
1764 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1765 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1766 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1768 if (tvd
->vdev_islog
) {
1769 metaslab_group_passivate(mg
);
1770 slog_found
= B_TRUE
;
1774 return (slog_found
);
1778 spa_activate_log(spa_t
*spa
)
1780 vdev_t
*rvd
= spa
->spa_root_vdev
;
1783 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1785 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1786 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1787 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1789 if (tvd
->vdev_islog
)
1790 metaslab_group_activate(mg
);
1795 spa_offline_log(spa_t
*spa
)
1799 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1800 NULL
, DS_FIND_CHILDREN
);
1803 * We successfully offlined the log device, sync out the
1804 * current txg so that the "stubby" block can be removed
1807 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1813 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1817 for (i
= 0; i
< sav
->sav_count
; i
++)
1818 spa_check_removed(sav
->sav_vdevs
[i
]);
1822 spa_claim_notify(zio_t
*zio
)
1824 spa_t
*spa
= zio
->io_spa
;
1829 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1830 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1831 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1832 mutex_exit(&spa
->spa_props_lock
);
1835 typedef struct spa_load_error
{
1836 uint64_t sle_meta_count
;
1837 uint64_t sle_data_count
;
1841 spa_load_verify_done(zio_t
*zio
)
1843 blkptr_t
*bp
= zio
->io_bp
;
1844 spa_load_error_t
*sle
= zio
->io_private
;
1845 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1846 int error
= zio
->io_error
;
1849 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1850 type
!= DMU_OT_INTENT_LOG
)
1851 atomic_add_64(&sle
->sle_meta_count
, 1);
1853 atomic_add_64(&sle
->sle_data_count
, 1);
1855 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1860 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1861 const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1865 size_t size
= BP_GET_PSIZE(bp
);
1866 void *data
= zio_data_buf_alloc(size
);
1868 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1869 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1870 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1871 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1877 spa_load_verify(spa_t
*spa
)
1880 spa_load_error_t sle
= { 0 };
1881 zpool_rewind_policy_t policy
;
1882 boolean_t verify_ok
= B_FALSE
;
1885 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1887 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1890 rio
= zio_root(spa
, NULL
, &sle
,
1891 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1893 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1894 TRAVERSE_PRE
| TRAVERSE_PREFETCH
, spa_load_verify_cb
, rio
);
1896 (void) zio_wait(rio
);
1898 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1899 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1901 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1902 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1906 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1907 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1909 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1910 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1911 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1912 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1913 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1914 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1915 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1917 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1921 if (error
!= ENXIO
&& error
!= EIO
)
1926 return (verify_ok
? 0 : EIO
);
1930 * Find a value in the pool props object.
1933 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
1935 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
1936 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
1940 * Find a value in the pool directory object.
1943 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
1945 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1946 name
, sizeof (uint64_t), 1, val
));
1950 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
1952 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
1957 * Fix up config after a partly-completed split. This is done with the
1958 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1959 * pool have that entry in their config, but only the splitting one contains
1960 * a list of all the guids of the vdevs that are being split off.
1962 * This function determines what to do with that list: either rejoin
1963 * all the disks to the pool, or complete the splitting process. To attempt
1964 * the rejoin, each disk that is offlined is marked online again, and
1965 * we do a reopen() call. If the vdev label for every disk that was
1966 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1967 * then we call vdev_split() on each disk, and complete the split.
1969 * Otherwise we leave the config alone, with all the vdevs in place in
1970 * the original pool.
1973 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
1980 boolean_t attempt_reopen
;
1982 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
1985 /* check that the config is complete */
1986 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
1987 &glist
, &gcount
) != 0)
1990 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_PUSHPAGE
);
1992 /* attempt to online all the vdevs & validate */
1993 attempt_reopen
= B_TRUE
;
1994 for (i
= 0; i
< gcount
; i
++) {
1995 if (glist
[i
] == 0) /* vdev is hole */
1998 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
1999 if (vd
[i
] == NULL
) {
2001 * Don't bother attempting to reopen the disks;
2002 * just do the split.
2004 attempt_reopen
= B_FALSE
;
2006 /* attempt to re-online it */
2007 vd
[i
]->vdev_offline
= B_FALSE
;
2011 if (attempt_reopen
) {
2012 vdev_reopen(spa
->spa_root_vdev
);
2014 /* check each device to see what state it's in */
2015 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2016 if (vd
[i
] != NULL
&&
2017 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2024 * If every disk has been moved to the new pool, or if we never
2025 * even attempted to look at them, then we split them off for
2028 if (!attempt_reopen
|| gcount
== extracted
) {
2029 for (i
= 0; i
< gcount
; i
++)
2032 vdev_reopen(spa
->spa_root_vdev
);
2035 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2039 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2040 boolean_t mosconfig
)
2042 nvlist_t
*config
= spa
->spa_config
;
2043 char *ereport
= FM_EREPORT_ZFS_POOL
;
2049 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2052 ASSERT(spa
->spa_comment
== NULL
);
2053 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2054 spa
->spa_comment
= spa_strdup(comment
);
2057 * Versioning wasn't explicitly added to the label until later, so if
2058 * it's not present treat it as the initial version.
2060 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2061 &spa
->spa_ubsync
.ub_version
) != 0)
2062 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2064 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2065 &spa
->spa_config_txg
);
2067 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2068 spa_guid_exists(pool_guid
, 0)) {
2071 spa
->spa_config_guid
= pool_guid
;
2073 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2075 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2079 nvlist_free(spa
->spa_load_info
);
2080 spa
->spa_load_info
= fnvlist_alloc();
2082 gethrestime(&spa
->spa_loaded_ts
);
2083 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2084 mosconfig
, &ereport
);
2087 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2089 if (error
!= EEXIST
) {
2090 spa
->spa_loaded_ts
.tv_sec
= 0;
2091 spa
->spa_loaded_ts
.tv_nsec
= 0;
2093 if (error
!= EBADF
) {
2094 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2097 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2104 * Load an existing storage pool, using the pool's builtin spa_config as a
2105 * source of configuration information.
2107 __attribute__((always_inline
))
2109 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2110 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2114 nvlist_t
*nvroot
= NULL
;
2117 uberblock_t
*ub
= &spa
->spa_uberblock
;
2118 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2119 int orig_mode
= spa
->spa_mode
;
2122 boolean_t missing_feat_write
= B_FALSE
;
2125 * If this is an untrusted config, access the pool in read-only mode.
2126 * This prevents things like resilvering recently removed devices.
2129 spa
->spa_mode
= FREAD
;
2131 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2133 spa
->spa_load_state
= state
;
2135 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2138 parse
= (type
== SPA_IMPORT_EXISTING
?
2139 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2142 * Create "The Godfather" zio to hold all async IOs
2144 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
2145 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
2148 * Parse the configuration into a vdev tree. We explicitly set the
2149 * value that will be returned by spa_version() since parsing the
2150 * configuration requires knowing the version number.
2152 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2153 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2154 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2159 ASSERT(spa
->spa_root_vdev
== rvd
);
2161 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2162 ASSERT(spa_guid(spa
) == pool_guid
);
2166 * Try to open all vdevs, loading each label in the process.
2168 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2169 error
= vdev_open(rvd
);
2170 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2175 * We need to validate the vdev labels against the configuration that
2176 * we have in hand, which is dependent on the setting of mosconfig. If
2177 * mosconfig is true then we're validating the vdev labels based on
2178 * that config. Otherwise, we're validating against the cached config
2179 * (zpool.cache) that was read when we loaded the zfs module, and then
2180 * later we will recursively call spa_load() and validate against
2183 * If we're assembling a new pool that's been split off from an
2184 * existing pool, the labels haven't yet been updated so we skip
2185 * validation for now.
2187 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2188 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2189 error
= vdev_validate(rvd
, mosconfig
);
2190 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2195 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2200 * Find the best uberblock.
2202 vdev_uberblock_load(rvd
, ub
, &label
);
2205 * If we weren't able to find a single valid uberblock, return failure.
2207 if (ub
->ub_txg
== 0) {
2209 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2213 * If the pool has an unsupported version we can't open it.
2215 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2217 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2220 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2224 * If we weren't able to find what's necessary for reading the
2225 * MOS in the label, return failure.
2227 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2228 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2230 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2235 * Update our in-core representation with the definitive values
2238 nvlist_free(spa
->spa_label_features
);
2239 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2245 * Look through entries in the label nvlist's features_for_read. If
2246 * there is a feature listed there which we don't understand then we
2247 * cannot open a pool.
2249 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2250 nvlist_t
*unsup_feat
;
2253 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2256 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2258 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2259 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2260 VERIFY(nvlist_add_string(unsup_feat
,
2261 nvpair_name(nvp
), "") == 0);
2265 if (!nvlist_empty(unsup_feat
)) {
2266 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2267 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2268 nvlist_free(unsup_feat
);
2269 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2273 nvlist_free(unsup_feat
);
2277 * If the vdev guid sum doesn't match the uberblock, we have an
2278 * incomplete configuration. We first check to see if the pool
2279 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2280 * If it is, defer the vdev_guid_sum check till later so we
2281 * can handle missing vdevs.
2283 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2284 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2285 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2286 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2288 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2289 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2290 spa_try_repair(spa
, config
);
2291 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2292 nvlist_free(spa
->spa_config_splitting
);
2293 spa
->spa_config_splitting
= NULL
;
2297 * Initialize internal SPA structures.
2299 spa
->spa_state
= POOL_STATE_ACTIVE
;
2300 spa
->spa_ubsync
= spa
->spa_uberblock
;
2301 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2302 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2303 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2304 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2305 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2306 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2308 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2310 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2311 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2313 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2314 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2316 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2317 boolean_t missing_feat_read
= B_FALSE
;
2318 nvlist_t
*unsup_feat
, *enabled_feat
;
2320 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2321 &spa
->spa_feat_for_read_obj
) != 0) {
2322 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2325 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2326 &spa
->spa_feat_for_write_obj
) != 0) {
2327 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2330 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2331 &spa
->spa_feat_desc_obj
) != 0) {
2332 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2335 enabled_feat
= fnvlist_alloc();
2336 unsup_feat
= fnvlist_alloc();
2338 if (!feature_is_supported(spa
->spa_meta_objset
,
2339 spa
->spa_feat_for_read_obj
, spa
->spa_feat_desc_obj
,
2340 unsup_feat
, enabled_feat
))
2341 missing_feat_read
= B_TRUE
;
2343 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2344 if (!feature_is_supported(spa
->spa_meta_objset
,
2345 spa
->spa_feat_for_write_obj
, spa
->spa_feat_desc_obj
,
2346 unsup_feat
, enabled_feat
)) {
2347 missing_feat_write
= B_TRUE
;
2351 fnvlist_add_nvlist(spa
->spa_load_info
,
2352 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2354 if (!nvlist_empty(unsup_feat
)) {
2355 fnvlist_add_nvlist(spa
->spa_load_info
,
2356 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2359 fnvlist_free(enabled_feat
);
2360 fnvlist_free(unsup_feat
);
2362 if (!missing_feat_read
) {
2363 fnvlist_add_boolean(spa
->spa_load_info
,
2364 ZPOOL_CONFIG_CAN_RDONLY
);
2368 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2369 * twofold: to determine whether the pool is available for
2370 * import in read-write mode and (if it is not) whether the
2371 * pool is available for import in read-only mode. If the pool
2372 * is available for import in read-write mode, it is displayed
2373 * as available in userland; if it is not available for import
2374 * in read-only mode, it is displayed as unavailable in
2375 * userland. If the pool is available for import in read-only
2376 * mode but not read-write mode, it is displayed as unavailable
2377 * in userland with a special note that the pool is actually
2378 * available for open in read-only mode.
2380 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2381 * missing a feature for write, we must first determine whether
2382 * the pool can be opened read-only before returning to
2383 * userland in order to know whether to display the
2384 * abovementioned note.
2386 if (missing_feat_read
|| (missing_feat_write
&&
2387 spa_writeable(spa
))) {
2388 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2393 spa
->spa_is_initializing
= B_TRUE
;
2394 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2395 spa
->spa_is_initializing
= B_FALSE
;
2397 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2401 nvlist_t
*policy
= NULL
, *nvconfig
;
2403 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2404 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2406 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2407 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2409 unsigned long myhostid
= 0;
2411 VERIFY(nvlist_lookup_string(nvconfig
,
2412 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2415 myhostid
= zone_get_hostid(NULL
);
2418 * We're emulating the system's hostid in userland, so
2419 * we can't use zone_get_hostid().
2421 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2422 #endif /* _KERNEL */
2423 if (hostid
!= 0 && myhostid
!= 0 &&
2424 hostid
!= myhostid
) {
2425 nvlist_free(nvconfig
);
2426 cmn_err(CE_WARN
, "pool '%s' could not be "
2427 "loaded as it was last accessed by "
2428 "another system (host: %s hostid: 0x%lx). "
2429 "See: http://zfsonlinux.org/msg/ZFS-8000-EY",
2430 spa_name(spa
), hostname
,
2431 (unsigned long)hostid
);
2435 if (nvlist_lookup_nvlist(spa
->spa_config
,
2436 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2437 VERIFY(nvlist_add_nvlist(nvconfig
,
2438 ZPOOL_REWIND_POLICY
, policy
) == 0);
2440 spa_config_set(spa
, nvconfig
);
2442 spa_deactivate(spa
);
2443 spa_activate(spa
, orig_mode
);
2445 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2448 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2449 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2450 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2452 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2455 * Load the bit that tells us to use the new accounting function
2456 * (raid-z deflation). If we have an older pool, this will not
2459 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2460 if (error
!= 0 && error
!= ENOENT
)
2461 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2463 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2464 &spa
->spa_creation_version
);
2465 if (error
!= 0 && error
!= ENOENT
)
2466 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2469 * Load the persistent error log. If we have an older pool, this will
2472 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2473 if (error
!= 0 && error
!= ENOENT
)
2474 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2476 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2477 &spa
->spa_errlog_scrub
);
2478 if (error
!= 0 && error
!= ENOENT
)
2479 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2482 * Load the history object. If we have an older pool, this
2483 * will not be present.
2485 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2486 if (error
!= 0 && error
!= ENOENT
)
2487 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2490 * If we're assembling the pool from the split-off vdevs of
2491 * an existing pool, we don't want to attach the spares & cache
2496 * Load any hot spares for this pool.
2498 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2499 if (error
!= 0 && error
!= ENOENT
)
2500 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2501 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2502 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2503 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2504 &spa
->spa_spares
.sav_config
) != 0)
2505 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2507 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2508 spa_load_spares(spa
);
2509 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2510 } else if (error
== 0) {
2511 spa
->spa_spares
.sav_sync
= B_TRUE
;
2515 * Load any level 2 ARC devices for this pool.
2517 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2518 &spa
->spa_l2cache
.sav_object
);
2519 if (error
!= 0 && error
!= ENOENT
)
2520 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2521 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2522 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2523 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2524 &spa
->spa_l2cache
.sav_config
) != 0)
2525 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2527 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2528 spa_load_l2cache(spa
);
2529 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2530 } else if (error
== 0) {
2531 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2534 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2536 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2537 if (error
&& error
!= ENOENT
)
2538 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2541 uint64_t autoreplace
;
2543 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2544 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2545 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2546 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2547 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2548 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2549 &spa
->spa_dedup_ditto
);
2551 spa
->spa_autoreplace
= (autoreplace
!= 0);
2555 * If the 'autoreplace' property is set, then post a resource notifying
2556 * the ZFS DE that it should not issue any faults for unopenable
2557 * devices. We also iterate over the vdevs, and post a sysevent for any
2558 * unopenable vdevs so that the normal autoreplace handler can take
2561 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2562 spa_check_removed(spa
->spa_root_vdev
);
2564 * For the import case, this is done in spa_import(), because
2565 * at this point we're using the spare definitions from
2566 * the MOS config, not necessarily from the userland config.
2568 if (state
!= SPA_LOAD_IMPORT
) {
2569 spa_aux_check_removed(&spa
->spa_spares
);
2570 spa_aux_check_removed(&spa
->spa_l2cache
);
2575 * Load the vdev state for all toplevel vdevs.
2580 * Propagate the leaf DTLs we just loaded all the way up the tree.
2582 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2583 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2584 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2587 * Load the DDTs (dedup tables).
2589 error
= ddt_load(spa
);
2591 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2593 spa_update_dspace(spa
);
2596 * Validate the config, using the MOS config to fill in any
2597 * information which might be missing. If we fail to validate
2598 * the config then declare the pool unfit for use. If we're
2599 * assembling a pool from a split, the log is not transferred
2602 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2605 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2606 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2608 if (!spa_config_valid(spa
, nvconfig
)) {
2609 nvlist_free(nvconfig
);
2610 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2613 nvlist_free(nvconfig
);
2616 * Now that we've validated the config, check the state of the
2617 * root vdev. If it can't be opened, it indicates one or
2618 * more toplevel vdevs are faulted.
2620 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2623 if (spa_check_logs(spa
)) {
2624 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2625 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2629 if (missing_feat_write
) {
2630 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2633 * At this point, we know that we can open the pool in
2634 * read-only mode but not read-write mode. We now have enough
2635 * information and can return to userland.
2637 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2641 * We've successfully opened the pool, verify that we're ready
2642 * to start pushing transactions.
2644 if (state
!= SPA_LOAD_TRYIMPORT
) {
2645 if ((error
= spa_load_verify(spa
)))
2646 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2650 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2651 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2653 int need_update
= B_FALSE
;
2656 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2659 * Claim log blocks that haven't been committed yet.
2660 * This must all happen in a single txg.
2661 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2662 * invoked from zil_claim_log_block()'s i/o done callback.
2663 * Price of rollback is that we abandon the log.
2665 spa
->spa_claiming
= B_TRUE
;
2667 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
2668 spa_first_txg(spa
));
2669 (void) dmu_objset_find(spa_name(spa
),
2670 zil_claim
, tx
, DS_FIND_CHILDREN
);
2673 spa
->spa_claiming
= B_FALSE
;
2675 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2676 spa
->spa_sync_on
= B_TRUE
;
2677 txg_sync_start(spa
->spa_dsl_pool
);
2680 * Wait for all claims to sync. We sync up to the highest
2681 * claimed log block birth time so that claimed log blocks
2682 * don't appear to be from the future. spa_claim_max_txg
2683 * will have been set for us by either zil_check_log_chain()
2684 * (invoked from spa_check_logs()) or zil_claim() above.
2686 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2689 * If the config cache is stale, or we have uninitialized
2690 * metaslabs (see spa_vdev_add()), then update the config.
2692 * If this is a verbatim import, trust the current
2693 * in-core spa_config and update the disk labels.
2695 if (config_cache_txg
!= spa
->spa_config_txg
||
2696 state
== SPA_LOAD_IMPORT
||
2697 state
== SPA_LOAD_RECOVER
||
2698 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2699 need_update
= B_TRUE
;
2701 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2702 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2703 need_update
= B_TRUE
;
2706 * Update the config cache asychronously in case we're the
2707 * root pool, in which case the config cache isn't writable yet.
2710 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2713 * Check all DTLs to see if anything needs resilvering.
2715 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2716 vdev_resilver_needed(rvd
, NULL
, NULL
))
2717 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2720 * Log the fact that we booted up (so that we can detect if
2721 * we rebooted in the middle of an operation).
2723 spa_history_log_version(spa
, "open");
2726 * Delete any inconsistent datasets.
2728 (void) dmu_objset_find(spa_name(spa
),
2729 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2732 * Clean up any stale temporary dataset userrefs.
2734 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2741 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2743 int mode
= spa
->spa_mode
;
2746 spa_deactivate(spa
);
2748 spa
->spa_load_max_txg
--;
2750 spa_activate(spa
, mode
);
2751 spa_async_suspend(spa
);
2753 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2757 * If spa_load() fails this function will try loading prior txg's. If
2758 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2759 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2760 * function will not rewind the pool and will return the same error as
2764 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2765 uint64_t max_request
, int rewind_flags
)
2767 nvlist_t
*loadinfo
= NULL
;
2768 nvlist_t
*config
= NULL
;
2769 int load_error
, rewind_error
;
2770 uint64_t safe_rewind_txg
;
2773 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2774 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2775 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2777 spa
->spa_load_max_txg
= max_request
;
2780 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2782 if (load_error
== 0)
2785 if (spa
->spa_root_vdev
!= NULL
)
2786 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2788 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2789 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2791 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2792 nvlist_free(config
);
2793 return (load_error
);
2796 if (state
== SPA_LOAD_RECOVER
) {
2797 /* Price of rolling back is discarding txgs, including log */
2798 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2801 * If we aren't rolling back save the load info from our first
2802 * import attempt so that we can restore it after attempting
2805 loadinfo
= spa
->spa_load_info
;
2806 spa
->spa_load_info
= fnvlist_alloc();
2809 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2810 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2811 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2812 TXG_INITIAL
: safe_rewind_txg
;
2815 * Continue as long as we're finding errors, we're still within
2816 * the acceptable rewind range, and we're still finding uberblocks
2818 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2819 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2820 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2821 spa
->spa_extreme_rewind
= B_TRUE
;
2822 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2825 spa
->spa_extreme_rewind
= B_FALSE
;
2826 spa
->spa_load_max_txg
= UINT64_MAX
;
2828 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2829 spa_config_set(spa
, config
);
2831 if (state
== SPA_LOAD_RECOVER
) {
2832 ASSERT3P(loadinfo
, ==, NULL
);
2833 return (rewind_error
);
2835 /* Store the rewind info as part of the initial load info */
2836 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
2837 spa
->spa_load_info
);
2839 /* Restore the initial load info */
2840 fnvlist_free(spa
->spa_load_info
);
2841 spa
->spa_load_info
= loadinfo
;
2843 return (load_error
);
2850 * The import case is identical to an open except that the configuration is sent
2851 * down from userland, instead of grabbed from the configuration cache. For the
2852 * case of an open, the pool configuration will exist in the
2853 * POOL_STATE_UNINITIALIZED state.
2855 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2856 * the same time open the pool, without having to keep around the spa_t in some
2860 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2864 spa_load_state_t state
= SPA_LOAD_OPEN
;
2866 int locked
= B_FALSE
;
2867 int firstopen
= B_FALSE
;
2872 * As disgusting as this is, we need to support recursive calls to this
2873 * function because dsl_dir_open() is called during spa_load(), and ends
2874 * up calling spa_open() again. The real fix is to figure out how to
2875 * avoid dsl_dir_open() calling this in the first place.
2877 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2878 mutex_enter(&spa_namespace_lock
);
2882 if ((spa
= spa_lookup(pool
)) == NULL
) {
2884 mutex_exit(&spa_namespace_lock
);
2888 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
2889 zpool_rewind_policy_t policy
;
2893 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
2895 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
2896 state
= SPA_LOAD_RECOVER
;
2898 spa_activate(spa
, spa_mode_global
);
2900 if (state
!= SPA_LOAD_RECOVER
)
2901 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
2903 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
2904 policy
.zrp_request
);
2906 if (error
== EBADF
) {
2908 * If vdev_validate() returns failure (indicated by
2909 * EBADF), it indicates that one of the vdevs indicates
2910 * that the pool has been exported or destroyed. If
2911 * this is the case, the config cache is out of sync and
2912 * we should remove the pool from the namespace.
2915 spa_deactivate(spa
);
2916 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
2919 mutex_exit(&spa_namespace_lock
);
2925 * We can't open the pool, but we still have useful
2926 * information: the state of each vdev after the
2927 * attempted vdev_open(). Return this to the user.
2929 if (config
!= NULL
&& spa
->spa_config
) {
2930 VERIFY(nvlist_dup(spa
->spa_config
, config
,
2932 VERIFY(nvlist_add_nvlist(*config
,
2933 ZPOOL_CONFIG_LOAD_INFO
,
2934 spa
->spa_load_info
) == 0);
2937 spa_deactivate(spa
);
2938 spa
->spa_last_open_failed
= error
;
2940 mutex_exit(&spa_namespace_lock
);
2946 spa_open_ref(spa
, tag
);
2949 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2952 * If we've recovered the pool, pass back any information we
2953 * gathered while doing the load.
2955 if (state
== SPA_LOAD_RECOVER
) {
2956 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
2957 spa
->spa_load_info
) == 0);
2961 spa
->spa_last_open_failed
= 0;
2962 spa
->spa_last_ubsync_txg
= 0;
2963 spa
->spa_load_txg
= 0;
2964 mutex_exit(&spa_namespace_lock
);
2969 zvol_create_minors(spa
->spa_name
);
2978 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
2981 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
2985 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
2987 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
2991 * Lookup the given spa_t, incrementing the inject count in the process,
2992 * preventing it from being exported or destroyed.
2995 spa_inject_addref(char *name
)
2999 mutex_enter(&spa_namespace_lock
);
3000 if ((spa
= spa_lookup(name
)) == NULL
) {
3001 mutex_exit(&spa_namespace_lock
);
3004 spa
->spa_inject_ref
++;
3005 mutex_exit(&spa_namespace_lock
);
3011 spa_inject_delref(spa_t
*spa
)
3013 mutex_enter(&spa_namespace_lock
);
3014 spa
->spa_inject_ref
--;
3015 mutex_exit(&spa_namespace_lock
);
3019 * Add spares device information to the nvlist.
3022 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3032 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3034 if (spa
->spa_spares
.sav_count
== 0)
3037 VERIFY(nvlist_lookup_nvlist(config
,
3038 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3039 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3040 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3042 VERIFY(nvlist_add_nvlist_array(nvroot
,
3043 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3044 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3045 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3048 * Go through and find any spares which have since been
3049 * repurposed as an active spare. If this is the case, update
3050 * their status appropriately.
3052 for (i
= 0; i
< nspares
; i
++) {
3053 VERIFY(nvlist_lookup_uint64(spares
[i
],
3054 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3055 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3057 VERIFY(nvlist_lookup_uint64_array(
3058 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3059 (uint64_t **)&vs
, &vsc
) == 0);
3060 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3061 vs
->vs_aux
= VDEV_AUX_SPARED
;
3068 * Add l2cache device information to the nvlist, including vdev stats.
3071 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3074 uint_t i
, j
, nl2cache
;
3081 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3083 if (spa
->spa_l2cache
.sav_count
== 0)
3086 VERIFY(nvlist_lookup_nvlist(config
,
3087 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3088 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3089 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3090 if (nl2cache
!= 0) {
3091 VERIFY(nvlist_add_nvlist_array(nvroot
,
3092 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3093 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3094 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3097 * Update level 2 cache device stats.
3100 for (i
= 0; i
< nl2cache
; i
++) {
3101 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3102 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3105 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3107 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3108 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3114 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3115 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3117 vdev_get_stats(vd
, vs
);
3123 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3129 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3130 VERIFY(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3132 if (spa
->spa_feat_for_read_obj
!= 0) {
3133 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3134 spa
->spa_feat_for_read_obj
);
3135 zap_cursor_retrieve(&zc
, &za
) == 0;
3136 zap_cursor_advance(&zc
)) {
3137 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3138 za
.za_num_integers
== 1);
3139 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
3140 za
.za_first_integer
));
3142 zap_cursor_fini(&zc
);
3145 if (spa
->spa_feat_for_write_obj
!= 0) {
3146 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3147 spa
->spa_feat_for_write_obj
);
3148 zap_cursor_retrieve(&zc
, &za
) == 0;
3149 zap_cursor_advance(&zc
)) {
3150 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3151 za
.za_num_integers
== 1);
3152 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
3153 za
.za_first_integer
));
3155 zap_cursor_fini(&zc
);
3158 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3160 nvlist_free(features
);
3164 spa_get_stats(const char *name
, nvlist_t
**config
,
3165 char *altroot
, size_t buflen
)
3171 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3175 * This still leaves a window of inconsistency where the spares
3176 * or l2cache devices could change and the config would be
3177 * self-inconsistent.
3179 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3181 if (*config
!= NULL
) {
3182 uint64_t loadtimes
[2];
3184 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3185 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3186 VERIFY(nvlist_add_uint64_array(*config
,
3187 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3189 VERIFY(nvlist_add_uint64(*config
,
3190 ZPOOL_CONFIG_ERRCOUNT
,
3191 spa_get_errlog_size(spa
)) == 0);
3193 if (spa_suspended(spa
))
3194 VERIFY(nvlist_add_uint64(*config
,
3195 ZPOOL_CONFIG_SUSPENDED
,
3196 spa
->spa_failmode
) == 0);
3198 spa_add_spares(spa
, *config
);
3199 spa_add_l2cache(spa
, *config
);
3200 spa_add_feature_stats(spa
, *config
);
3205 * We want to get the alternate root even for faulted pools, so we cheat
3206 * and call spa_lookup() directly.
3210 mutex_enter(&spa_namespace_lock
);
3211 spa
= spa_lookup(name
);
3213 spa_altroot(spa
, altroot
, buflen
);
3217 mutex_exit(&spa_namespace_lock
);
3219 spa_altroot(spa
, altroot
, buflen
);
3224 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3225 spa_close(spa
, FTAG
);
3232 * Validate that the auxiliary device array is well formed. We must have an
3233 * array of nvlists, each which describes a valid leaf vdev. If this is an
3234 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3235 * specified, as long as they are well-formed.
3238 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3239 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3240 vdev_labeltype_t label
)
3247 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3250 * It's acceptable to have no devs specified.
3252 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3259 * Make sure the pool is formatted with a version that supports this
3262 if (spa_version(spa
) < version
)
3266 * Set the pending device list so we correctly handle device in-use
3269 sav
->sav_pending
= dev
;
3270 sav
->sav_npending
= ndev
;
3272 for (i
= 0; i
< ndev
; i
++) {
3273 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3277 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3284 * The L2ARC currently only supports disk devices in
3285 * kernel context. For user-level testing, we allow it.
3288 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3289 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3297 if ((error
= vdev_open(vd
)) == 0 &&
3298 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3299 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3300 vd
->vdev_guid
) == 0);
3306 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3313 sav
->sav_pending
= NULL
;
3314 sav
->sav_npending
= 0;
3319 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3323 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3325 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3326 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3327 VDEV_LABEL_SPARE
)) != 0) {
3331 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3332 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3333 VDEV_LABEL_L2CACHE
));
3337 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3342 if (sav
->sav_config
!= NULL
) {
3348 * Generate new dev list by concatentating with the
3351 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3352 &olddevs
, &oldndevs
) == 0);
3354 newdevs
= kmem_alloc(sizeof (void *) *
3355 (ndevs
+ oldndevs
), KM_PUSHPAGE
);
3356 for (i
= 0; i
< oldndevs
; i
++)
3357 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3359 for (i
= 0; i
< ndevs
; i
++)
3360 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3363 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3364 DATA_TYPE_NVLIST_ARRAY
) == 0);
3366 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3367 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3368 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3369 nvlist_free(newdevs
[i
]);
3370 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3373 * Generate a new dev list.
3375 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3377 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3383 * Stop and drop level 2 ARC devices
3386 spa_l2cache_drop(spa_t
*spa
)
3390 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3392 for (i
= 0; i
< sav
->sav_count
; i
++) {
3395 vd
= sav
->sav_vdevs
[i
];
3398 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3399 pool
!= 0ULL && l2arc_vdev_present(vd
))
3400 l2arc_remove_vdev(vd
);
3408 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3412 char *altroot
= NULL
;
3417 uint64_t txg
= TXG_INITIAL
;
3418 nvlist_t
**spares
, **l2cache
;
3419 uint_t nspares
, nl2cache
;
3420 uint64_t version
, obj
;
3421 boolean_t has_features
;
3426 * If this pool already exists, return failure.
3428 mutex_enter(&spa_namespace_lock
);
3429 if (spa_lookup(pool
) != NULL
) {
3430 mutex_exit(&spa_namespace_lock
);
3435 * Allocate a new spa_t structure.
3437 (void) nvlist_lookup_string(props
,
3438 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3439 spa
= spa_add(pool
, NULL
, altroot
);
3440 spa_activate(spa
, spa_mode_global
);
3442 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3443 spa_deactivate(spa
);
3445 mutex_exit(&spa_namespace_lock
);
3449 has_features
= B_FALSE
;
3450 for (elem
= nvlist_next_nvpair(props
, NULL
);
3451 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3452 if (zpool_prop_feature(nvpair_name(elem
)))
3453 has_features
= B_TRUE
;
3456 if (has_features
|| nvlist_lookup_uint64(props
,
3457 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3458 version
= SPA_VERSION
;
3460 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3462 spa
->spa_first_txg
= txg
;
3463 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3464 spa
->spa_uberblock
.ub_version
= version
;
3465 spa
->spa_ubsync
= spa
->spa_uberblock
;
3468 * Create "The Godfather" zio to hold all async IOs
3470 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
3471 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
3474 * Create the root vdev.
3476 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3478 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3480 ASSERT(error
!= 0 || rvd
!= NULL
);
3481 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3483 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3487 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3488 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3489 VDEV_ALLOC_ADD
)) == 0) {
3490 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
3491 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3492 vdev_expand(rvd
->vdev_child
[c
], txg
);
3496 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3500 spa_deactivate(spa
);
3502 mutex_exit(&spa_namespace_lock
);
3507 * Get the list of spares, if specified.
3509 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3510 &spares
, &nspares
) == 0) {
3511 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3513 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3514 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3515 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3516 spa_load_spares(spa
);
3517 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3518 spa
->spa_spares
.sav_sync
= B_TRUE
;
3522 * Get the list of level 2 cache devices, if specified.
3524 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3525 &l2cache
, &nl2cache
) == 0) {
3526 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3527 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3528 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3529 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3530 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3531 spa_load_l2cache(spa
);
3532 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3533 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3536 spa
->spa_is_initializing
= B_TRUE
;
3537 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3538 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3539 spa
->spa_is_initializing
= B_FALSE
;
3542 * Create DDTs (dedup tables).
3546 spa_update_dspace(spa
);
3548 tx
= dmu_tx_create_assigned(dp
, txg
);
3551 * Create the pool config object.
3553 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3554 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3555 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3557 if (zap_add(spa
->spa_meta_objset
,
3558 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3559 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3560 cmn_err(CE_PANIC
, "failed to add pool config");
3563 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3564 spa_feature_create_zap_objects(spa
, tx
);
3566 if (zap_add(spa
->spa_meta_objset
,
3567 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3568 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3569 cmn_err(CE_PANIC
, "failed to add pool version");
3572 /* Newly created pools with the right version are always deflated. */
3573 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3574 spa
->spa_deflate
= TRUE
;
3575 if (zap_add(spa
->spa_meta_objset
,
3576 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3577 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3578 cmn_err(CE_PANIC
, "failed to add deflate");
3583 * Create the deferred-free bpobj. Turn off compression
3584 * because sync-to-convergence takes longer if the blocksize
3587 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3588 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3589 ZIO_COMPRESS_OFF
, tx
);
3590 if (zap_add(spa
->spa_meta_objset
,
3591 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3592 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3593 cmn_err(CE_PANIC
, "failed to add bpobj");
3595 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3596 spa
->spa_meta_objset
, obj
));
3599 * Create the pool's history object.
3601 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3602 spa_history_create_obj(spa
, tx
);
3605 * Set pool properties.
3607 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3608 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3609 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3610 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3612 if (props
!= NULL
) {
3613 spa_configfile_set(spa
, props
, B_FALSE
);
3614 spa_sync_props(props
, tx
);
3619 spa
->spa_sync_on
= B_TRUE
;
3620 txg_sync_start(spa
->spa_dsl_pool
);
3623 * We explicitly wait for the first transaction to complete so that our
3624 * bean counters are appropriately updated.
3626 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3628 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3630 spa_history_log_version(spa
, "create");
3632 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3634 mutex_exit(&spa_namespace_lock
);
3641 * Get the root pool information from the root disk, then import the root pool
3642 * during the system boot up time.
3644 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3647 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3650 nvlist_t
*nvtop
, *nvroot
;
3653 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3657 * Add this top-level vdev to the child array.
3659 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3661 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3663 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3666 * Put this pool's top-level vdevs into a root vdev.
3668 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3669 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3670 VDEV_TYPE_ROOT
) == 0);
3671 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3672 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3673 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3677 * Replace the existing vdev_tree with the new root vdev in
3678 * this pool's configuration (remove the old, add the new).
3680 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3681 nvlist_free(nvroot
);
3686 * Walk the vdev tree and see if we can find a device with "better"
3687 * configuration. A configuration is "better" if the label on that
3688 * device has a more recent txg.
3691 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3695 for (c
= 0; c
< vd
->vdev_children
; c
++)
3696 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3698 if (vd
->vdev_ops
->vdev_op_leaf
) {
3702 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3706 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3710 * Do we have a better boot device?
3712 if (label_txg
> *txg
) {
3721 * Import a root pool.
3723 * For x86. devpath_list will consist of devid and/or physpath name of
3724 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3725 * The GRUB "findroot" command will return the vdev we should boot.
3727 * For Sparc, devpath_list consists the physpath name of the booting device
3728 * no matter the rootpool is a single device pool or a mirrored pool.
3730 * "/pci@1f,0/ide@d/disk@0,0:a"
3733 spa_import_rootpool(char *devpath
, char *devid
)
3736 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3737 nvlist_t
*config
, *nvtop
;
3743 * Read the label from the boot device and generate a configuration.
3745 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3746 #if defined(_OBP) && defined(_KERNEL)
3747 if (config
== NULL
) {
3748 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3750 get_iscsi_bootpath_phy(devpath
);
3751 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3755 if (config
== NULL
) {
3756 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
3761 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3763 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3765 mutex_enter(&spa_namespace_lock
);
3766 if ((spa
= spa_lookup(pname
)) != NULL
) {
3768 * Remove the existing root pool from the namespace so that we
3769 * can replace it with the correct config we just read in.
3774 spa
= spa_add(pname
, config
, NULL
);
3775 spa
->spa_is_root
= B_TRUE
;
3776 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3779 * Build up a vdev tree based on the boot device's label config.
3781 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3783 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3784 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3785 VDEV_ALLOC_ROOTPOOL
);
3786 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3788 mutex_exit(&spa_namespace_lock
);
3789 nvlist_free(config
);
3790 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3796 * Get the boot vdev.
3798 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3799 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3800 (u_longlong_t
)guid
);
3806 * Determine if there is a better boot device.
3809 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3811 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3812 "try booting from '%s'", avd
->vdev_path
);
3818 * If the boot device is part of a spare vdev then ensure that
3819 * we're booting off the active spare.
3821 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3822 !bvd
->vdev_isspare
) {
3823 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
3824 "try booting from '%s'",
3826 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
3833 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3835 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3836 mutex_exit(&spa_namespace_lock
);
3838 nvlist_free(config
);
3845 * Import a non-root pool into the system.
3848 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
3851 char *altroot
= NULL
;
3852 spa_load_state_t state
= SPA_LOAD_IMPORT
;
3853 zpool_rewind_policy_t policy
;
3854 uint64_t mode
= spa_mode_global
;
3855 uint64_t readonly
= B_FALSE
;
3858 nvlist_t
**spares
, **l2cache
;
3859 uint_t nspares
, nl2cache
;
3862 * If a pool with this name exists, return failure.
3864 mutex_enter(&spa_namespace_lock
);
3865 if (spa_lookup(pool
) != NULL
) {
3866 mutex_exit(&spa_namespace_lock
);
3871 * Create and initialize the spa structure.
3873 (void) nvlist_lookup_string(props
,
3874 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3875 (void) nvlist_lookup_uint64(props
,
3876 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
3879 spa
= spa_add(pool
, config
, altroot
);
3880 spa
->spa_import_flags
= flags
;
3883 * Verbatim import - Take a pool and insert it into the namespace
3884 * as if it had been loaded at boot.
3886 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
3888 spa_configfile_set(spa
, props
, B_FALSE
);
3890 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3892 mutex_exit(&spa_namespace_lock
);
3893 spa_history_log_version(spa
, "import");
3898 spa_activate(spa
, mode
);
3901 * Don't start async tasks until we know everything is healthy.
3903 spa_async_suspend(spa
);
3905 zpool_get_rewind_policy(config
, &policy
);
3906 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3907 state
= SPA_LOAD_RECOVER
;
3910 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3911 * because the user-supplied config is actually the one to trust when
3914 if (state
!= SPA_LOAD_RECOVER
)
3915 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3917 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
3918 policy
.zrp_request
);
3921 * Propagate anything learned while loading the pool and pass it
3922 * back to caller (i.e. rewind info, missing devices, etc).
3924 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
3925 spa
->spa_load_info
) == 0);
3927 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3929 * Toss any existing sparelist, as it doesn't have any validity
3930 * anymore, and conflicts with spa_has_spare().
3932 if (spa
->spa_spares
.sav_config
) {
3933 nvlist_free(spa
->spa_spares
.sav_config
);
3934 spa
->spa_spares
.sav_config
= NULL
;
3935 spa_load_spares(spa
);
3937 if (spa
->spa_l2cache
.sav_config
) {
3938 nvlist_free(spa
->spa_l2cache
.sav_config
);
3939 spa
->spa_l2cache
.sav_config
= NULL
;
3940 spa_load_l2cache(spa
);
3943 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3946 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3949 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3950 VDEV_ALLOC_L2CACHE
);
3951 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3954 spa_configfile_set(spa
, props
, B_FALSE
);
3956 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
3957 (error
= spa_prop_set(spa
, props
)))) {
3959 spa_deactivate(spa
);
3961 mutex_exit(&spa_namespace_lock
);
3965 spa_async_resume(spa
);
3968 * Override any spares and level 2 cache devices as specified by
3969 * the user, as these may have correct device names/devids, etc.
3971 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3972 &spares
, &nspares
) == 0) {
3973 if (spa
->spa_spares
.sav_config
)
3974 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
3975 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3977 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
3978 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3979 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3980 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3981 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3982 spa_load_spares(spa
);
3983 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3984 spa
->spa_spares
.sav_sync
= B_TRUE
;
3986 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3987 &l2cache
, &nl2cache
) == 0) {
3988 if (spa
->spa_l2cache
.sav_config
)
3989 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
3990 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3992 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3993 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3994 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3995 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3996 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3997 spa_load_l2cache(spa
);
3998 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3999 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4003 * Check for any removed devices.
4005 if (spa
->spa_autoreplace
) {
4006 spa_aux_check_removed(&spa
->spa_spares
);
4007 spa_aux_check_removed(&spa
->spa_l2cache
);
4010 if (spa_writeable(spa
)) {
4012 * Update the config cache to include the newly-imported pool.
4014 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4018 * It's possible that the pool was expanded while it was exported.
4019 * We kick off an async task to handle this for us.
4021 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4023 mutex_exit(&spa_namespace_lock
);
4024 spa_history_log_version(spa
, "import");
4027 zvol_create_minors(pool
);
4034 spa_tryimport(nvlist_t
*tryconfig
)
4036 nvlist_t
*config
= NULL
;
4042 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4045 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4049 * Create and initialize the spa structure.
4051 mutex_enter(&spa_namespace_lock
);
4052 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4053 spa_activate(spa
, FREAD
);
4056 * Pass off the heavy lifting to spa_load().
4057 * Pass TRUE for mosconfig because the user-supplied config
4058 * is actually the one to trust when doing an import.
4060 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4063 * If 'tryconfig' was at least parsable, return the current config.
4065 if (spa
->spa_root_vdev
!= NULL
) {
4066 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4067 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4069 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4071 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4072 spa
->spa_uberblock
.ub_timestamp
) == 0);
4073 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4074 spa
->spa_load_info
) == 0);
4077 * If the bootfs property exists on this pool then we
4078 * copy it out so that external consumers can tell which
4079 * pools are bootable.
4081 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4082 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
4085 * We have to play games with the name since the
4086 * pool was opened as TRYIMPORT_NAME.
4088 if (dsl_dsobj_to_dsname(spa_name(spa
),
4089 spa
->spa_bootfs
, tmpname
) == 0) {
4091 char *dsname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
4093 cp
= strchr(tmpname
, '/');
4095 (void) strlcpy(dsname
, tmpname
,
4098 (void) snprintf(dsname
, MAXPATHLEN
,
4099 "%s/%s", poolname
, ++cp
);
4101 VERIFY(nvlist_add_string(config
,
4102 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4103 kmem_free(dsname
, MAXPATHLEN
);
4105 kmem_free(tmpname
, MAXPATHLEN
);
4109 * Add the list of hot spares and level 2 cache devices.
4111 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4112 spa_add_spares(spa
, config
);
4113 spa_add_l2cache(spa
, config
);
4114 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4118 spa_deactivate(spa
);
4120 mutex_exit(&spa_namespace_lock
);
4126 * Pool export/destroy
4128 * The act of destroying or exporting a pool is very simple. We make sure there
4129 * is no more pending I/O and any references to the pool are gone. Then, we
4130 * update the pool state and sync all the labels to disk, removing the
4131 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4132 * we don't sync the labels or remove the configuration cache.
4135 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4136 boolean_t force
, boolean_t hardforce
)
4143 if (!(spa_mode_global
& FWRITE
))
4146 mutex_enter(&spa_namespace_lock
);
4147 if ((spa
= spa_lookup(pool
)) == NULL
) {
4148 mutex_exit(&spa_namespace_lock
);
4153 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4154 * reacquire the namespace lock, and see if we can export.
4156 spa_open_ref(spa
, FTAG
);
4157 mutex_exit(&spa_namespace_lock
);
4158 spa_async_suspend(spa
);
4159 mutex_enter(&spa_namespace_lock
);
4160 spa_close(spa
, FTAG
);
4163 * The pool will be in core if it's openable,
4164 * in which case we can modify its state.
4166 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
&& spa
->spa_sync_on
) {
4168 * Objsets may be open only because they're dirty, so we
4169 * have to force it to sync before checking spa_refcnt.
4171 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4174 * A pool cannot be exported or destroyed if there are active
4175 * references. If we are resetting a pool, allow references by
4176 * fault injection handlers.
4178 if (!spa_refcount_zero(spa
) ||
4179 (spa
->spa_inject_ref
!= 0 &&
4180 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4181 spa_async_resume(spa
);
4182 mutex_exit(&spa_namespace_lock
);
4187 * A pool cannot be exported if it has an active shared spare.
4188 * This is to prevent other pools stealing the active spare
4189 * from an exported pool. At user's own will, such pool can
4190 * be forcedly exported.
4192 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4193 spa_has_active_shared_spare(spa
)) {
4194 spa_async_resume(spa
);
4195 mutex_exit(&spa_namespace_lock
);
4200 * We want this to be reflected on every label,
4201 * so mark them all dirty. spa_unload() will do the
4202 * final sync that pushes these changes out.
4204 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4205 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4206 spa
->spa_state
= new_state
;
4207 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4209 vdev_config_dirty(spa
->spa_root_vdev
);
4210 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4214 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_DESTROY
);
4216 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4218 spa_deactivate(spa
);
4221 if (oldconfig
&& spa
->spa_config
)
4222 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4224 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4226 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4229 mutex_exit(&spa_namespace_lock
);
4235 * Destroy a storage pool.
4238 spa_destroy(char *pool
)
4240 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4245 * Export a storage pool.
4248 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4249 boolean_t hardforce
)
4251 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4256 * Similar to spa_export(), this unloads the spa_t without actually removing it
4257 * from the namespace in any way.
4260 spa_reset(char *pool
)
4262 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4267 * ==========================================================================
4268 * Device manipulation
4269 * ==========================================================================
4273 * Add a device to a storage pool.
4276 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4280 vdev_t
*rvd
= spa
->spa_root_vdev
;
4282 nvlist_t
**spares
, **l2cache
;
4283 uint_t nspares
, nl2cache
;
4286 ASSERT(spa_writeable(spa
));
4288 txg
= spa_vdev_enter(spa
);
4290 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4291 VDEV_ALLOC_ADD
)) != 0)
4292 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4294 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4296 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4300 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4304 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4305 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4307 if (vd
->vdev_children
!= 0 &&
4308 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4309 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4312 * We must validate the spares and l2cache devices after checking the
4313 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4315 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4316 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4319 * Transfer each new top-level vdev from vd to rvd.
4321 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4324 * Set the vdev id to the first hole, if one exists.
4326 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4327 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4328 vdev_free(rvd
->vdev_child
[id
]);
4332 tvd
= vd
->vdev_child
[c
];
4333 vdev_remove_child(vd
, tvd
);
4335 vdev_add_child(rvd
, tvd
);
4336 vdev_config_dirty(tvd
);
4340 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4341 ZPOOL_CONFIG_SPARES
);
4342 spa_load_spares(spa
);
4343 spa
->spa_spares
.sav_sync
= B_TRUE
;
4346 if (nl2cache
!= 0) {
4347 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4348 ZPOOL_CONFIG_L2CACHE
);
4349 spa_load_l2cache(spa
);
4350 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4354 * We have to be careful when adding new vdevs to an existing pool.
4355 * If other threads start allocating from these vdevs before we
4356 * sync the config cache, and we lose power, then upon reboot we may
4357 * fail to open the pool because there are DVAs that the config cache
4358 * can't translate. Therefore, we first add the vdevs without
4359 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4360 * and then let spa_config_update() initialize the new metaslabs.
4362 * spa_load() checks for added-but-not-initialized vdevs, so that
4363 * if we lose power at any point in this sequence, the remaining
4364 * steps will be completed the next time we load the pool.
4366 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4368 mutex_enter(&spa_namespace_lock
);
4369 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4370 mutex_exit(&spa_namespace_lock
);
4376 * Attach a device to a mirror. The arguments are the path to any device
4377 * in the mirror, and the nvroot for the new device. If the path specifies
4378 * a device that is not mirrored, we automatically insert the mirror vdev.
4380 * If 'replacing' is specified, the new device is intended to replace the
4381 * existing device; in this case the two devices are made into their own
4382 * mirror using the 'replacing' vdev, which is functionally identical to
4383 * the mirror vdev (it actually reuses all the same ops) but has a few
4384 * extra rules: you can't attach to it after it's been created, and upon
4385 * completion of resilvering, the first disk (the one being replaced)
4386 * is automatically detached.
4389 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4391 uint64_t txg
, dtl_max_txg
;
4392 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
;)
4393 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4395 char *oldvdpath
, *newvdpath
;
4399 ASSERT(spa_writeable(spa
));
4401 txg
= spa_vdev_enter(spa
);
4403 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4406 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4408 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4409 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4411 pvd
= oldvd
->vdev_parent
;
4413 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4414 VDEV_ALLOC_ATTACH
)) != 0)
4415 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4417 if (newrootvd
->vdev_children
!= 1)
4418 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4420 newvd
= newrootvd
->vdev_child
[0];
4422 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4423 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4425 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4426 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4429 * Spares can't replace logs
4431 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4432 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4436 * For attach, the only allowable parent is a mirror or the root
4439 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4440 pvd
->vdev_ops
!= &vdev_root_ops
)
4441 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4443 pvops
= &vdev_mirror_ops
;
4446 * Active hot spares can only be replaced by inactive hot
4449 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4450 oldvd
->vdev_isspare
&&
4451 !spa_has_spare(spa
, newvd
->vdev_guid
))
4452 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4455 * If the source is a hot spare, and the parent isn't already a
4456 * spare, then we want to create a new hot spare. Otherwise, we
4457 * want to create a replacing vdev. The user is not allowed to
4458 * attach to a spared vdev child unless the 'isspare' state is
4459 * the same (spare replaces spare, non-spare replaces
4462 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4463 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4464 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4465 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4466 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4467 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4470 if (newvd
->vdev_isspare
)
4471 pvops
= &vdev_spare_ops
;
4473 pvops
= &vdev_replacing_ops
;
4477 * Make sure the new device is big enough.
4479 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4480 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4483 * The new device cannot have a higher alignment requirement
4484 * than the top-level vdev.
4486 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4487 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4490 * If this is an in-place replacement, update oldvd's path and devid
4491 * to make it distinguishable from newvd, and unopenable from now on.
4493 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4494 spa_strfree(oldvd
->vdev_path
);
4495 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4497 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4498 newvd
->vdev_path
, "old");
4499 if (oldvd
->vdev_devid
!= NULL
) {
4500 spa_strfree(oldvd
->vdev_devid
);
4501 oldvd
->vdev_devid
= NULL
;
4505 /* mark the device being resilvered */
4506 newvd
->vdev_resilvering
= B_TRUE
;
4509 * If the parent is not a mirror, or if we're replacing, insert the new
4510 * mirror/replacing/spare vdev above oldvd.
4512 if (pvd
->vdev_ops
!= pvops
)
4513 pvd
= vdev_add_parent(oldvd
, pvops
);
4515 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4516 ASSERT(pvd
->vdev_ops
== pvops
);
4517 ASSERT(oldvd
->vdev_parent
== pvd
);
4520 * Extract the new device from its root and add it to pvd.
4522 vdev_remove_child(newrootvd
, newvd
);
4523 newvd
->vdev_id
= pvd
->vdev_children
;
4524 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4525 vdev_add_child(pvd
, newvd
);
4527 tvd
= newvd
->vdev_top
;
4528 ASSERT(pvd
->vdev_top
== tvd
);
4529 ASSERT(tvd
->vdev_parent
== rvd
);
4531 vdev_config_dirty(tvd
);
4534 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4535 * for any dmu_sync-ed blocks. It will propagate upward when
4536 * spa_vdev_exit() calls vdev_dtl_reassess().
4538 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4540 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4541 dtl_max_txg
- TXG_INITIAL
);
4543 if (newvd
->vdev_isspare
) {
4544 spa_spare_activate(newvd
);
4545 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_DEVICE_SPARE
);
4548 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4549 newvdpath
= spa_strdup(newvd
->vdev_path
);
4550 newvd_isspare
= newvd
->vdev_isspare
;
4553 * Mark newvd's DTL dirty in this txg.
4555 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4558 * Restart the resilver
4560 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4565 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4567 spa_history_log_internal(spa
, "vdev attach", NULL
,
4568 "%s vdev=%s %s vdev=%s",
4569 replacing
&& newvd_isspare
? "spare in" :
4570 replacing
? "replace" : "attach", newvdpath
,
4571 replacing
? "for" : "to", oldvdpath
);
4573 spa_strfree(oldvdpath
);
4574 spa_strfree(newvdpath
);
4576 if (spa
->spa_bootfs
)
4577 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH
);
4583 * Detach a device from a mirror or replacing vdev.
4584 * If 'replace_done' is specified, only detach if the parent
4585 * is a replacing vdev.
4588 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4592 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
;)
4593 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4594 boolean_t unspare
= B_FALSE
;
4595 uint64_t unspare_guid
= 0;
4599 ASSERT(spa_writeable(spa
));
4601 txg
= spa_vdev_enter(spa
);
4603 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4606 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4608 if (!vd
->vdev_ops
->vdev_op_leaf
)
4609 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4611 pvd
= vd
->vdev_parent
;
4614 * If the parent/child relationship is not as expected, don't do it.
4615 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4616 * vdev that's replacing B with C. The user's intent in replacing
4617 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4618 * the replace by detaching C, the expected behavior is to end up
4619 * M(A,B). But suppose that right after deciding to detach C,
4620 * the replacement of B completes. We would have M(A,C), and then
4621 * ask to detach C, which would leave us with just A -- not what
4622 * the user wanted. To prevent this, we make sure that the
4623 * parent/child relationship hasn't changed -- in this example,
4624 * that C's parent is still the replacing vdev R.
4626 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4627 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4630 * Only 'replacing' or 'spare' vdevs can be replaced.
4632 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4633 pvd
->vdev_ops
!= &vdev_spare_ops
)
4634 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4636 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4637 spa_version(spa
) >= SPA_VERSION_SPARES
);
4640 * Only mirror, replacing, and spare vdevs support detach.
4642 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4643 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4644 pvd
->vdev_ops
!= &vdev_spare_ops
)
4645 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4648 * If this device has the only valid copy of some data,
4649 * we cannot safely detach it.
4651 if (vdev_dtl_required(vd
))
4652 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4654 ASSERT(pvd
->vdev_children
>= 2);
4657 * If we are detaching the second disk from a replacing vdev, then
4658 * check to see if we changed the original vdev's path to have "/old"
4659 * at the end in spa_vdev_attach(). If so, undo that change now.
4661 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4662 vd
->vdev_path
!= NULL
) {
4663 size_t len
= strlen(vd
->vdev_path
);
4665 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4666 cvd
= pvd
->vdev_child
[c
];
4668 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4671 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4672 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4673 spa_strfree(cvd
->vdev_path
);
4674 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4681 * If we are detaching the original disk from a spare, then it implies
4682 * that the spare should become a real disk, and be removed from the
4683 * active spare list for the pool.
4685 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4687 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4691 * Erase the disk labels so the disk can be used for other things.
4692 * This must be done after all other error cases are handled,
4693 * but before we disembowel vd (so we can still do I/O to it).
4694 * But if we can't do it, don't treat the error as fatal --
4695 * it may be that the unwritability of the disk is the reason
4696 * it's being detached!
4698 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4701 * Remove vd from its parent and compact the parent's children.
4703 vdev_remove_child(pvd
, vd
);
4704 vdev_compact_children(pvd
);
4707 * Remember one of the remaining children so we can get tvd below.
4709 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4712 * If we need to remove the remaining child from the list of hot spares,
4713 * do it now, marking the vdev as no longer a spare in the process.
4714 * We must do this before vdev_remove_parent(), because that can
4715 * change the GUID if it creates a new toplevel GUID. For a similar
4716 * reason, we must remove the spare now, in the same txg as the detach;
4717 * otherwise someone could attach a new sibling, change the GUID, and
4718 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4721 ASSERT(cvd
->vdev_isspare
);
4722 spa_spare_remove(cvd
);
4723 unspare_guid
= cvd
->vdev_guid
;
4724 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4725 cvd
->vdev_unspare
= B_TRUE
;
4729 * If the parent mirror/replacing vdev only has one child,
4730 * the parent is no longer needed. Remove it from the tree.
4732 if (pvd
->vdev_children
== 1) {
4733 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4734 cvd
->vdev_unspare
= B_FALSE
;
4735 vdev_remove_parent(cvd
);
4736 cvd
->vdev_resilvering
= B_FALSE
;
4741 * We don't set tvd until now because the parent we just removed
4742 * may have been the previous top-level vdev.
4744 tvd
= cvd
->vdev_top
;
4745 ASSERT(tvd
->vdev_parent
== rvd
);
4748 * Reevaluate the parent vdev state.
4750 vdev_propagate_state(cvd
);
4753 * If the 'autoexpand' property is set on the pool then automatically
4754 * try to expand the size of the pool. For example if the device we
4755 * just detached was smaller than the others, it may be possible to
4756 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4757 * first so that we can obtain the updated sizes of the leaf vdevs.
4759 if (spa
->spa_autoexpand
) {
4761 vdev_expand(tvd
, txg
);
4764 vdev_config_dirty(tvd
);
4767 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4768 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4769 * But first make sure we're not on any *other* txg's DTL list, to
4770 * prevent vd from being accessed after it's freed.
4772 vdpath
= spa_strdup(vd
->vdev_path
);
4773 for (t
= 0; t
< TXG_SIZE
; t
++)
4774 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4775 vd
->vdev_detached
= B_TRUE
;
4776 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4778 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_REMOVE
);
4780 /* hang on to the spa before we release the lock */
4781 spa_open_ref(spa
, FTAG
);
4783 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4785 spa_history_log_internal(spa
, "detach", NULL
,
4787 spa_strfree(vdpath
);
4790 * If this was the removal of the original device in a hot spare vdev,
4791 * then we want to go through and remove the device from the hot spare
4792 * list of every other pool.
4795 spa_t
*altspa
= NULL
;
4797 mutex_enter(&spa_namespace_lock
);
4798 while ((altspa
= spa_next(altspa
)) != NULL
) {
4799 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4803 spa_open_ref(altspa
, FTAG
);
4804 mutex_exit(&spa_namespace_lock
);
4805 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4806 mutex_enter(&spa_namespace_lock
);
4807 spa_close(altspa
, FTAG
);
4809 mutex_exit(&spa_namespace_lock
);
4811 /* search the rest of the vdevs for spares to remove */
4812 spa_vdev_resilver_done(spa
);
4815 /* all done with the spa; OK to release */
4816 mutex_enter(&spa_namespace_lock
);
4817 spa_close(spa
, FTAG
);
4818 mutex_exit(&spa_namespace_lock
);
4824 * Split a set of devices from their mirrors, and create a new pool from them.
4827 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
4828 nvlist_t
*props
, boolean_t exp
)
4831 uint64_t txg
, *glist
;
4833 uint_t c
, children
, lastlog
;
4834 nvlist_t
**child
, *nvl
, *tmp
;
4836 char *altroot
= NULL
;
4837 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
4838 boolean_t activate_slog
;
4840 ASSERT(spa_writeable(spa
));
4842 txg
= spa_vdev_enter(spa
);
4844 /* clear the log and flush everything up to now */
4845 activate_slog
= spa_passivate_log(spa
);
4846 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4847 error
= spa_offline_log(spa
);
4848 txg
= spa_vdev_config_enter(spa
);
4851 spa_activate_log(spa
);
4854 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4856 /* check new spa name before going any further */
4857 if (spa_lookup(newname
) != NULL
)
4858 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
4861 * scan through all the children to ensure they're all mirrors
4863 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
4864 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
4866 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4868 /* first, check to ensure we've got the right child count */
4869 rvd
= spa
->spa_root_vdev
;
4871 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
4872 vdev_t
*vd
= rvd
->vdev_child
[c
];
4874 /* don't count the holes & logs as children */
4875 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
4883 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
4884 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4886 /* next, ensure no spare or cache devices are part of the split */
4887 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
4888 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
4889 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4891 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_PUSHPAGE
);
4892 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_PUSHPAGE
);
4894 /* then, loop over each vdev and validate it */
4895 for (c
= 0; c
< children
; c
++) {
4896 uint64_t is_hole
= 0;
4898 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
4902 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
4903 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
4911 /* which disk is going to be split? */
4912 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
4918 /* look it up in the spa */
4919 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
4920 if (vml
[c
] == NULL
) {
4925 /* make sure there's nothing stopping the split */
4926 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
4927 vml
[c
]->vdev_islog
||
4928 vml
[c
]->vdev_ishole
||
4929 vml
[c
]->vdev_isspare
||
4930 vml
[c
]->vdev_isl2cache
||
4931 !vdev_writeable(vml
[c
]) ||
4932 vml
[c
]->vdev_children
!= 0 ||
4933 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
4934 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
4939 if (vdev_dtl_required(vml
[c
])) {
4944 /* we need certain info from the top level */
4945 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
4946 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
4947 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
4948 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
4949 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
4950 vml
[c
]->vdev_top
->vdev_asize
) == 0);
4951 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
4952 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
4956 kmem_free(vml
, children
* sizeof (vdev_t
*));
4957 kmem_free(glist
, children
* sizeof (uint64_t));
4958 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4961 /* stop writers from using the disks */
4962 for (c
= 0; c
< children
; c
++) {
4964 vml
[c
]->vdev_offline
= B_TRUE
;
4966 vdev_reopen(spa
->spa_root_vdev
);
4969 * Temporarily record the splitting vdevs in the spa config. This
4970 * will disappear once the config is regenerated.
4972 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
4973 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
4974 glist
, children
) == 0);
4975 kmem_free(glist
, children
* sizeof (uint64_t));
4977 mutex_enter(&spa
->spa_props_lock
);
4978 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
4980 mutex_exit(&spa
->spa_props_lock
);
4981 spa
->spa_config_splitting
= nvl
;
4982 vdev_config_dirty(spa
->spa_root_vdev
);
4984 /* configure and create the new pool */
4985 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
4986 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4987 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
4988 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
4989 spa_version(spa
)) == 0);
4990 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
4991 spa
->spa_config_txg
) == 0);
4992 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
4993 spa_generate_guid(NULL
)) == 0);
4994 (void) nvlist_lookup_string(props
,
4995 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4997 /* add the new pool to the namespace */
4998 newspa
= spa_add(newname
, config
, altroot
);
4999 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5000 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5002 /* release the spa config lock, retaining the namespace lock */
5003 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5005 if (zio_injection_enabled
)
5006 zio_handle_panic_injection(spa
, FTAG
, 1);
5008 spa_activate(newspa
, spa_mode_global
);
5009 spa_async_suspend(newspa
);
5011 /* create the new pool from the disks of the original pool */
5012 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5016 /* if that worked, generate a real config for the new pool */
5017 if (newspa
->spa_root_vdev
!= NULL
) {
5018 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5019 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
5020 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5021 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5022 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5027 if (props
!= NULL
) {
5028 spa_configfile_set(newspa
, props
, B_FALSE
);
5029 error
= spa_prop_set(newspa
, props
);
5034 /* flush everything */
5035 txg
= spa_vdev_config_enter(newspa
);
5036 vdev_config_dirty(newspa
->spa_root_vdev
);
5037 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5039 if (zio_injection_enabled
)
5040 zio_handle_panic_injection(spa
, FTAG
, 2);
5042 spa_async_resume(newspa
);
5044 /* finally, update the original pool's config */
5045 txg
= spa_vdev_config_enter(spa
);
5046 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5047 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5050 for (c
= 0; c
< children
; c
++) {
5051 if (vml
[c
] != NULL
) {
5054 spa_history_log_internal(spa
, "detach", tx
,
5055 "vdev=%s", vml
[c
]->vdev_path
);
5059 vdev_config_dirty(spa
->spa_root_vdev
);
5060 spa
->spa_config_splitting
= NULL
;
5064 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5066 if (zio_injection_enabled
)
5067 zio_handle_panic_injection(spa
, FTAG
, 3);
5069 /* split is complete; log a history record */
5070 spa_history_log_internal(newspa
, "split", NULL
,
5071 "from pool %s", spa_name(spa
));
5073 kmem_free(vml
, children
* sizeof (vdev_t
*));
5075 /* if we're not going to mount the filesystems in userland, export */
5077 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5084 spa_deactivate(newspa
);
5087 txg
= spa_vdev_config_enter(spa
);
5089 /* re-online all offlined disks */
5090 for (c
= 0; c
< children
; c
++) {
5092 vml
[c
]->vdev_offline
= B_FALSE
;
5094 vdev_reopen(spa
->spa_root_vdev
);
5096 nvlist_free(spa
->spa_config_splitting
);
5097 spa
->spa_config_splitting
= NULL
;
5098 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5100 kmem_free(vml
, children
* sizeof (vdev_t
*));
5105 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5109 for (i
= 0; i
< count
; i
++) {
5112 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5115 if (guid
== target_guid
)
5123 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5124 nvlist_t
*dev_to_remove
)
5126 nvlist_t
**newdev
= NULL
;
5130 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_PUSHPAGE
);
5132 for (i
= 0, j
= 0; i
< count
; i
++) {
5133 if (dev
[i
] == dev_to_remove
)
5135 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_PUSHPAGE
) == 0);
5138 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5139 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5141 for (i
= 0; i
< count
- 1; i
++)
5142 nvlist_free(newdev
[i
]);
5145 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5149 * Evacuate the device.
5152 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5157 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5158 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5159 ASSERT(vd
== vd
->vdev_top
);
5162 * Evacuate the device. We don't hold the config lock as writer
5163 * since we need to do I/O but we do keep the
5164 * spa_namespace_lock held. Once this completes the device
5165 * should no longer have any blocks allocated on it.
5167 if (vd
->vdev_islog
) {
5168 if (vd
->vdev_stat
.vs_alloc
!= 0)
5169 error
= spa_offline_log(spa
);
5178 * The evacuation succeeded. Remove any remaining MOS metadata
5179 * associated with this vdev, and wait for these changes to sync.
5181 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5182 txg
= spa_vdev_config_enter(spa
);
5183 vd
->vdev_removing
= B_TRUE
;
5184 vdev_dirty(vd
, 0, NULL
, txg
);
5185 vdev_config_dirty(vd
);
5186 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5192 * Complete the removal by cleaning up the namespace.
5195 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5197 vdev_t
*rvd
= spa
->spa_root_vdev
;
5198 uint64_t id
= vd
->vdev_id
;
5199 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5201 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5202 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5203 ASSERT(vd
== vd
->vdev_top
);
5206 * Only remove any devices which are empty.
5208 if (vd
->vdev_stat
.vs_alloc
!= 0)
5211 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5213 if (list_link_active(&vd
->vdev_state_dirty_node
))
5214 vdev_state_clean(vd
);
5215 if (list_link_active(&vd
->vdev_config_dirty_node
))
5216 vdev_config_clean(vd
);
5221 vdev_compact_children(rvd
);
5223 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5224 vdev_add_child(rvd
, vd
);
5226 vdev_config_dirty(rvd
);
5229 * Reassess the health of our root vdev.
5235 * Remove a device from the pool -
5237 * Removing a device from the vdev namespace requires several steps
5238 * and can take a significant amount of time. As a result we use
5239 * the spa_vdev_config_[enter/exit] functions which allow us to
5240 * grab and release the spa_config_lock while still holding the namespace
5241 * lock. During each step the configuration is synced out.
5245 * Remove a device from the pool. Currently, this supports removing only hot
5246 * spares, slogs, and level 2 ARC devices.
5249 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5252 metaslab_group_t
*mg
;
5253 nvlist_t
**spares
, **l2cache
, *nv
;
5255 uint_t nspares
, nl2cache
;
5257 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5259 ASSERT(spa_writeable(spa
));
5262 txg
= spa_vdev_enter(spa
);
5264 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5266 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5267 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5268 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5269 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5271 * Only remove the hot spare if it's not currently in use
5274 if (vd
== NULL
|| unspare
) {
5275 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5276 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5277 spa_load_spares(spa
);
5278 spa
->spa_spares
.sav_sync
= B_TRUE
;
5282 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5283 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5284 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5285 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5287 * Cache devices can always be removed.
5289 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5290 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5291 spa_load_l2cache(spa
);
5292 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5293 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5295 ASSERT(vd
== vd
->vdev_top
);
5298 * XXX - Once we have bp-rewrite this should
5299 * become the common case.
5305 * Stop allocating from this vdev.
5307 metaslab_group_passivate(mg
);
5310 * Wait for the youngest allocations and frees to sync,
5311 * and then wait for the deferral of those frees to finish.
5313 spa_vdev_config_exit(spa
, NULL
,
5314 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5317 * Attempt to evacuate the vdev.
5319 error
= spa_vdev_remove_evacuate(spa
, vd
);
5321 txg
= spa_vdev_config_enter(spa
);
5324 * If we couldn't evacuate the vdev, unwind.
5327 metaslab_group_activate(mg
);
5328 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5332 * Clean up the vdev namespace.
5334 spa_vdev_remove_from_namespace(spa
, vd
);
5336 } else if (vd
!= NULL
) {
5338 * Normal vdevs cannot be removed (yet).
5343 * There is no vdev of any kind with the specified guid.
5349 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5355 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5356 * current spared, so we can detach it.
5359 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5361 vdev_t
*newvd
, *oldvd
;
5364 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5365 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5371 * Check for a completed replacement. We always consider the first
5372 * vdev in the list to be the oldest vdev, and the last one to be
5373 * the newest (see spa_vdev_attach() for how that works). In
5374 * the case where the newest vdev is faulted, we will not automatically
5375 * remove it after a resilver completes. This is OK as it will require
5376 * user intervention to determine which disk the admin wishes to keep.
5378 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5379 ASSERT(vd
->vdev_children
> 1);
5381 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5382 oldvd
= vd
->vdev_child
[0];
5384 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5385 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5386 !vdev_dtl_required(oldvd
))
5391 * Check for a completed resilver with the 'unspare' flag set.
5393 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5394 vdev_t
*first
= vd
->vdev_child
[0];
5395 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5397 if (last
->vdev_unspare
) {
5400 } else if (first
->vdev_unspare
) {
5407 if (oldvd
!= NULL
&&
5408 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5409 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5410 !vdev_dtl_required(oldvd
))
5414 * If there are more than two spares attached to a disk,
5415 * and those spares are not required, then we want to
5416 * attempt to free them up now so that they can be used
5417 * by other pools. Once we're back down to a single
5418 * disk+spare, we stop removing them.
5420 if (vd
->vdev_children
> 2) {
5421 newvd
= vd
->vdev_child
[1];
5423 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5424 vdev_dtl_empty(last
, DTL_MISSING
) &&
5425 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5426 !vdev_dtl_required(newvd
))
5435 spa_vdev_resilver_done(spa_t
*spa
)
5437 vdev_t
*vd
, *pvd
, *ppvd
;
5438 uint64_t guid
, sguid
, pguid
, ppguid
;
5440 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5442 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5443 pvd
= vd
->vdev_parent
;
5444 ppvd
= pvd
->vdev_parent
;
5445 guid
= vd
->vdev_guid
;
5446 pguid
= pvd
->vdev_guid
;
5447 ppguid
= ppvd
->vdev_guid
;
5450 * If we have just finished replacing a hot spared device, then
5451 * we need to detach the parent's first child (the original hot
5454 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5455 ppvd
->vdev_children
== 2) {
5456 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5457 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5459 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5460 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5462 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5464 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5467 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5471 * Update the stored path or FRU for this vdev.
5474 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5478 boolean_t sync
= B_FALSE
;
5480 ASSERT(spa_writeable(spa
));
5482 spa_vdev_state_enter(spa
, SCL_ALL
);
5484 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5485 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5487 if (!vd
->vdev_ops
->vdev_op_leaf
)
5488 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5491 if (strcmp(value
, vd
->vdev_path
) != 0) {
5492 spa_strfree(vd
->vdev_path
);
5493 vd
->vdev_path
= spa_strdup(value
);
5497 if (vd
->vdev_fru
== NULL
) {
5498 vd
->vdev_fru
= spa_strdup(value
);
5500 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5501 spa_strfree(vd
->vdev_fru
);
5502 vd
->vdev_fru
= spa_strdup(value
);
5507 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5511 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5513 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5517 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5519 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5523 * ==========================================================================
5525 * ==========================================================================
5529 spa_scan_stop(spa_t
*spa
)
5531 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5532 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5534 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5538 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5540 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5542 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5546 * If a resilver was requested, but there is no DTL on a
5547 * writeable leaf device, we have nothing to do.
5549 if (func
== POOL_SCAN_RESILVER
&&
5550 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5551 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5555 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5559 * ==========================================================================
5560 * SPA async task processing
5561 * ==========================================================================
5565 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5569 if (vd
->vdev_remove_wanted
) {
5570 vd
->vdev_remove_wanted
= B_FALSE
;
5571 vd
->vdev_delayed_close
= B_FALSE
;
5572 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5575 * We want to clear the stats, but we don't want to do a full
5576 * vdev_clear() as that will cause us to throw away
5577 * degraded/faulted state as well as attempt to reopen the
5578 * device, all of which is a waste.
5580 vd
->vdev_stat
.vs_read_errors
= 0;
5581 vd
->vdev_stat
.vs_write_errors
= 0;
5582 vd
->vdev_stat
.vs_checksum_errors
= 0;
5584 vdev_state_dirty(vd
->vdev_top
);
5587 for (c
= 0; c
< vd
->vdev_children
; c
++)
5588 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5592 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5596 if (vd
->vdev_probe_wanted
) {
5597 vd
->vdev_probe_wanted
= B_FALSE
;
5598 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5601 for (c
= 0; c
< vd
->vdev_children
; c
++)
5602 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5606 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5610 if (!spa
->spa_autoexpand
)
5613 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5614 vdev_t
*cvd
= vd
->vdev_child
[c
];
5615 spa_async_autoexpand(spa
, cvd
);
5618 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5621 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND
);
5625 spa_async_thread(spa_t
*spa
)
5629 ASSERT(spa
->spa_sync_on
);
5631 mutex_enter(&spa
->spa_async_lock
);
5632 tasks
= spa
->spa_async_tasks
;
5633 spa
->spa_async_tasks
= 0;
5634 mutex_exit(&spa
->spa_async_lock
);
5637 * See if the config needs to be updated.
5639 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5640 uint64_t old_space
, new_space
;
5642 mutex_enter(&spa_namespace_lock
);
5643 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5644 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5645 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5646 mutex_exit(&spa_namespace_lock
);
5649 * If the pool grew as a result of the config update,
5650 * then log an internal history event.
5652 if (new_space
!= old_space
) {
5653 spa_history_log_internal(spa
, "vdev online", NULL
,
5654 "pool '%s' size: %llu(+%llu)",
5655 spa_name(spa
), new_space
, new_space
- old_space
);
5660 * See if any devices need to be marked REMOVED.
5662 if (tasks
& SPA_ASYNC_REMOVE
) {
5663 spa_vdev_state_enter(spa
, SCL_NONE
);
5664 spa_async_remove(spa
, spa
->spa_root_vdev
);
5665 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5666 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5667 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5668 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5669 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5672 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5673 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5674 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5675 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5679 * See if any devices need to be probed.
5681 if (tasks
& SPA_ASYNC_PROBE
) {
5682 spa_vdev_state_enter(spa
, SCL_NONE
);
5683 spa_async_probe(spa
, spa
->spa_root_vdev
);
5684 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5688 * If any devices are done replacing, detach them.
5690 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5691 spa_vdev_resilver_done(spa
);
5694 * Kick off a resilver.
5696 if (tasks
& SPA_ASYNC_RESILVER
)
5697 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5700 * Let the world know that we're done.
5702 mutex_enter(&spa
->spa_async_lock
);
5703 spa
->spa_async_thread
= NULL
;
5704 cv_broadcast(&spa
->spa_async_cv
);
5705 mutex_exit(&spa
->spa_async_lock
);
5710 spa_async_suspend(spa_t
*spa
)
5712 mutex_enter(&spa
->spa_async_lock
);
5713 spa
->spa_async_suspended
++;
5714 while (spa
->spa_async_thread
!= NULL
)
5715 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5716 mutex_exit(&spa
->spa_async_lock
);
5720 spa_async_resume(spa_t
*spa
)
5722 mutex_enter(&spa
->spa_async_lock
);
5723 ASSERT(spa
->spa_async_suspended
!= 0);
5724 spa
->spa_async_suspended
--;
5725 mutex_exit(&spa
->spa_async_lock
);
5729 spa_async_dispatch(spa_t
*spa
)
5731 mutex_enter(&spa
->spa_async_lock
);
5732 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5733 spa
->spa_async_thread
== NULL
&&
5734 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5735 spa
->spa_async_thread
= thread_create(NULL
, 0,
5736 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5737 mutex_exit(&spa
->spa_async_lock
);
5741 spa_async_request(spa_t
*spa
, int task
)
5743 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5744 mutex_enter(&spa
->spa_async_lock
);
5745 spa
->spa_async_tasks
|= task
;
5746 mutex_exit(&spa
->spa_async_lock
);
5750 * ==========================================================================
5751 * SPA syncing routines
5752 * ==========================================================================
5756 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5759 bpobj_enqueue(bpo
, bp
, tx
);
5764 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5768 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5774 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5776 char *packed
= NULL
;
5781 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5784 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5785 * information. This avoids the dbuf_will_dirty() path and
5786 * saves us a pre-read to get data we don't actually care about.
5788 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
5789 packed
= vmem_alloc(bufsize
, KM_PUSHPAGE
);
5791 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
5793 bzero(packed
+ nvsize
, bufsize
- nvsize
);
5795 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
5797 vmem_free(packed
, bufsize
);
5799 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
5800 dmu_buf_will_dirty(db
, tx
);
5801 *(uint64_t *)db
->db_data
= nvsize
;
5802 dmu_buf_rele(db
, FTAG
);
5806 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
5807 const char *config
, const char *entry
)
5817 * Update the MOS nvlist describing the list of available devices.
5818 * spa_validate_aux() will have already made sure this nvlist is
5819 * valid and the vdevs are labeled appropriately.
5821 if (sav
->sav_object
== 0) {
5822 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
5823 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
5824 sizeof (uint64_t), tx
);
5825 VERIFY(zap_update(spa
->spa_meta_objset
,
5826 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
5827 &sav
->sav_object
, tx
) == 0);
5830 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
5831 if (sav
->sav_count
== 0) {
5832 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
5834 list
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_PUSHPAGE
);
5835 for (i
= 0; i
< sav
->sav_count
; i
++)
5836 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
5837 B_FALSE
, VDEV_CONFIG_L2CACHE
);
5838 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
5839 sav
->sav_count
) == 0);
5840 for (i
= 0; i
< sav
->sav_count
; i
++)
5841 nvlist_free(list
[i
]);
5842 kmem_free(list
, sav
->sav_count
* sizeof (void *));
5845 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
5846 nvlist_free(nvroot
);
5848 sav
->sav_sync
= B_FALSE
;
5852 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
5856 if (list_is_empty(&spa
->spa_config_dirty_list
))
5859 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5861 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
5862 dmu_tx_get_txg(tx
), B_FALSE
);
5865 * If we're upgrading the spa version then make sure that
5866 * the config object gets updated with the correct version.
5868 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
5869 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5870 spa
->spa_uberblock
.ub_version
);
5872 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5874 if (spa
->spa_config_syncing
)
5875 nvlist_free(spa
->spa_config_syncing
);
5876 spa
->spa_config_syncing
= config
;
5878 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
5882 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
5884 uint64_t *versionp
= arg
;
5885 uint64_t version
= *versionp
;
5886 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5889 * Setting the version is special cased when first creating the pool.
5891 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
5893 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
5894 ASSERT(version
>= spa_version(spa
));
5896 spa
->spa_uberblock
.ub_version
= version
;
5897 vdev_config_dirty(spa
->spa_root_vdev
);
5898 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
5902 * Set zpool properties.
5905 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
5907 nvlist_t
*nvp
= arg
;
5908 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5909 objset_t
*mos
= spa
->spa_meta_objset
;
5910 nvpair_t
*elem
= NULL
;
5912 mutex_enter(&spa
->spa_props_lock
);
5914 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
5916 char *strval
, *fname
;
5918 const char *propname
;
5919 zprop_type_t proptype
;
5920 zfeature_info_t
*feature
;
5922 prop
= zpool_name_to_prop(nvpair_name(elem
));
5923 switch ((int)prop
) {
5926 * We checked this earlier in spa_prop_validate().
5928 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
5930 fname
= strchr(nvpair_name(elem
), '@') + 1;
5931 VERIFY3U(0, ==, zfeature_lookup_name(fname
, &feature
));
5933 spa_feature_enable(spa
, feature
, tx
);
5934 spa_history_log_internal(spa
, "set", tx
,
5935 "%s=enabled", nvpair_name(elem
));
5938 case ZPOOL_PROP_VERSION
:
5939 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
5941 * The version is synced seperatly before other
5942 * properties and should be correct by now.
5944 ASSERT3U(spa_version(spa
), >=, intval
);
5947 case ZPOOL_PROP_ALTROOT
:
5949 * 'altroot' is a non-persistent property. It should
5950 * have been set temporarily at creation or import time.
5952 ASSERT(spa
->spa_root
!= NULL
);
5955 case ZPOOL_PROP_READONLY
:
5956 case ZPOOL_PROP_CACHEFILE
:
5958 * 'readonly' and 'cachefile' are also non-persisitent
5962 case ZPOOL_PROP_COMMENT
:
5963 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
5964 if (spa
->spa_comment
!= NULL
)
5965 spa_strfree(spa
->spa_comment
);
5966 spa
->spa_comment
= spa_strdup(strval
);
5968 * We need to dirty the configuration on all the vdevs
5969 * so that their labels get updated. It's unnecessary
5970 * to do this for pool creation since the vdev's
5971 * configuratoin has already been dirtied.
5973 if (tx
->tx_txg
!= TXG_INITIAL
)
5974 vdev_config_dirty(spa
->spa_root_vdev
);
5975 spa_history_log_internal(spa
, "set", tx
,
5976 "%s=%s", nvpair_name(elem
), strval
);
5980 * Set pool property values in the poolprops mos object.
5982 if (spa
->spa_pool_props_object
== 0) {
5983 spa
->spa_pool_props_object
=
5984 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
5985 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
5989 /* normalize the property name */
5990 propname
= zpool_prop_to_name(prop
);
5991 proptype
= zpool_prop_get_type(prop
);
5993 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
5994 ASSERT(proptype
== PROP_TYPE_STRING
);
5995 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
5996 VERIFY(zap_update(mos
,
5997 spa
->spa_pool_props_object
, propname
,
5998 1, strlen(strval
) + 1, strval
, tx
) == 0);
5999 spa_history_log_internal(spa
, "set", tx
,
6000 "%s=%s", nvpair_name(elem
), strval
);
6001 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6002 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
6004 if (proptype
== PROP_TYPE_INDEX
) {
6006 VERIFY(zpool_prop_index_to_string(
6007 prop
, intval
, &unused
) == 0);
6009 VERIFY(zap_update(mos
,
6010 spa
->spa_pool_props_object
, propname
,
6011 8, 1, &intval
, tx
) == 0);
6012 spa_history_log_internal(spa
, "set", tx
,
6013 "%s=%lld", nvpair_name(elem
), intval
);
6015 ASSERT(0); /* not allowed */
6019 case ZPOOL_PROP_DELEGATION
:
6020 spa
->spa_delegation
= intval
;
6022 case ZPOOL_PROP_BOOTFS
:
6023 spa
->spa_bootfs
= intval
;
6025 case ZPOOL_PROP_FAILUREMODE
:
6026 spa
->spa_failmode
= intval
;
6028 case ZPOOL_PROP_AUTOEXPAND
:
6029 spa
->spa_autoexpand
= intval
;
6030 if (tx
->tx_txg
!= TXG_INITIAL
)
6031 spa_async_request(spa
,
6032 SPA_ASYNC_AUTOEXPAND
);
6034 case ZPOOL_PROP_DEDUPDITTO
:
6035 spa
->spa_dedup_ditto
= intval
;
6044 mutex_exit(&spa
->spa_props_lock
);
6048 * Perform one-time upgrade on-disk changes. spa_version() does not
6049 * reflect the new version this txg, so there must be no changes this
6050 * txg to anything that the upgrade code depends on after it executes.
6051 * Therefore this must be called after dsl_pool_sync() does the sync
6055 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6057 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6059 ASSERT(spa
->spa_sync_pass
== 1);
6061 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6063 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6064 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6065 dsl_pool_create_origin(dp
, tx
);
6067 /* Keeping the origin open increases spa_minref */
6068 spa
->spa_minref
+= 3;
6071 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6072 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6073 dsl_pool_upgrade_clones(dp
, tx
);
6076 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6077 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6078 dsl_pool_upgrade_dir_clones(dp
, tx
);
6080 /* Keeping the freedir open increases spa_minref */
6081 spa
->spa_minref
+= 3;
6084 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6085 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6086 spa_feature_create_zap_objects(spa
, tx
);
6088 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6092 * Sync the specified transaction group. New blocks may be dirtied as
6093 * part of the process, so we iterate until it converges.
6096 spa_sync(spa_t
*spa
, uint64_t txg
)
6098 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6099 objset_t
*mos
= spa
->spa_meta_objset
;
6100 bpobj_t
*defer_bpo
= &spa
->spa_deferred_bpobj
;
6101 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6102 vdev_t
*rvd
= spa
->spa_root_vdev
;
6108 VERIFY(spa_writeable(spa
));
6111 * Lock out configuration changes.
6113 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6115 spa
->spa_syncing_txg
= txg
;
6116 spa
->spa_sync_pass
= 0;
6119 * If there are any pending vdev state changes, convert them
6120 * into config changes that go out with this transaction group.
6122 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6123 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6125 * We need the write lock here because, for aux vdevs,
6126 * calling vdev_config_dirty() modifies sav_config.
6127 * This is ugly and will become unnecessary when we
6128 * eliminate the aux vdev wart by integrating all vdevs
6129 * into the root vdev tree.
6131 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6132 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6133 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6134 vdev_state_clean(vd
);
6135 vdev_config_dirty(vd
);
6137 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6138 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6140 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6142 tx
= dmu_tx_create_assigned(dp
, txg
);
6144 spa
->spa_sync_starttime
= gethrtime();
6145 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6146 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
6147 spa_deadman
, spa
, TQ_PUSHPAGE
, ddi_get_lbolt() +
6148 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6151 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6152 * set spa_deflate if we have no raid-z vdevs.
6154 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6155 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6158 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6159 vd
= rvd
->vdev_child
[i
];
6160 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6163 if (i
== rvd
->vdev_children
) {
6164 spa
->spa_deflate
= TRUE
;
6165 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6166 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6167 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6172 * If anything has changed in this txg, or if someone is waiting
6173 * for this txg to sync (eg, spa_vdev_remove()), push the
6174 * deferred frees from the previous txg. If not, leave them
6175 * alone so that we don't generate work on an otherwise idle
6178 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
6179 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
6180 !txg_list_empty(&dp
->dp_sync_tasks
, txg
) ||
6181 ((dsl_scan_active(dp
->dp_scan
) ||
6182 txg_sync_waiting(dp
)) && !spa_shutting_down(spa
))) {
6183 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6184 VERIFY3U(bpobj_iterate(defer_bpo
,
6185 spa_free_sync_cb
, zio
, tx
), ==, 0);
6186 VERIFY0(zio_wait(zio
));
6190 * Iterate to convergence.
6193 int pass
= ++spa
->spa_sync_pass
;
6195 spa_sync_config_object(spa
, tx
);
6196 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6197 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6198 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6199 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6200 spa_errlog_sync(spa
, txg
);
6201 dsl_pool_sync(dp
, txg
);
6203 if (pass
< zfs_sync_pass_deferred_free
) {
6204 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6205 bplist_iterate(free_bpl
, spa_free_sync_cb
,
6207 VERIFY(zio_wait(zio
) == 0);
6209 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6214 dsl_scan_sync(dp
, tx
);
6216 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6220 spa_sync_upgrades(spa
, tx
);
6222 } while (dmu_objset_is_dirty(mos
, txg
));
6225 * Rewrite the vdev configuration (which includes the uberblock)
6226 * to commit the transaction group.
6228 * If there are no dirty vdevs, we sync the uberblock to a few
6229 * random top-level vdevs that are known to be visible in the
6230 * config cache (see spa_vdev_add() for a complete description).
6231 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6235 * We hold SCL_STATE to prevent vdev open/close/etc.
6236 * while we're attempting to write the vdev labels.
6238 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6240 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6241 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6243 int children
= rvd
->vdev_children
;
6244 int c0
= spa_get_random(children
);
6246 for (c
= 0; c
< children
; c
++) {
6247 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6248 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6250 svd
[svdcount
++] = vd
;
6251 if (svdcount
== SPA_DVAS_PER_BP
)
6254 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
6256 error
= vdev_config_sync(svd
, svdcount
, txg
,
6259 error
= vdev_config_sync(rvd
->vdev_child
,
6260 rvd
->vdev_children
, txg
, B_FALSE
);
6262 error
= vdev_config_sync(rvd
->vdev_child
,
6263 rvd
->vdev_children
, txg
, B_TRUE
);
6267 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6269 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6273 zio_suspend(spa
, NULL
);
6274 zio_resume_wait(spa
);
6278 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6279 spa
->spa_deadman_tqid
= 0;
6282 * Clear the dirty config list.
6284 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6285 vdev_config_clean(vd
);
6288 * Now that the new config has synced transactionally,
6289 * let it become visible to the config cache.
6291 if (spa
->spa_config_syncing
!= NULL
) {
6292 spa_config_set(spa
, spa
->spa_config_syncing
);
6293 spa
->spa_config_txg
= txg
;
6294 spa
->spa_config_syncing
= NULL
;
6297 spa
->spa_ubsync
= spa
->spa_uberblock
;
6299 dsl_pool_sync_done(dp
, txg
);
6302 * Update usable space statistics.
6304 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6305 vdev_sync_done(vd
, txg
);
6307 spa_update_dspace(spa
);
6310 * It had better be the case that we didn't dirty anything
6311 * since vdev_config_sync().
6313 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6314 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6315 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6317 spa
->spa_sync_pass
= 0;
6319 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6321 spa_handle_ignored_writes(spa
);
6324 * If any async tasks have been requested, kick them off.
6326 spa_async_dispatch(spa
);
6330 * Sync all pools. We don't want to hold the namespace lock across these
6331 * operations, so we take a reference on the spa_t and drop the lock during the
6335 spa_sync_allpools(void)
6338 mutex_enter(&spa_namespace_lock
);
6339 while ((spa
= spa_next(spa
)) != NULL
) {
6340 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6341 !spa_writeable(spa
) || spa_suspended(spa
))
6343 spa_open_ref(spa
, FTAG
);
6344 mutex_exit(&spa_namespace_lock
);
6345 txg_wait_synced(spa_get_dsl(spa
), 0);
6346 mutex_enter(&spa_namespace_lock
);
6347 spa_close(spa
, FTAG
);
6349 mutex_exit(&spa_namespace_lock
);
6353 * ==========================================================================
6354 * Miscellaneous routines
6355 * ==========================================================================
6359 * Remove all pools in the system.
6367 * Remove all cached state. All pools should be closed now,
6368 * so every spa in the AVL tree should be unreferenced.
6370 mutex_enter(&spa_namespace_lock
);
6371 while ((spa
= spa_next(NULL
)) != NULL
) {
6373 * Stop async tasks. The async thread may need to detach
6374 * a device that's been replaced, which requires grabbing
6375 * spa_namespace_lock, so we must drop it here.
6377 spa_open_ref(spa
, FTAG
);
6378 mutex_exit(&spa_namespace_lock
);
6379 spa_async_suspend(spa
);
6380 mutex_enter(&spa_namespace_lock
);
6381 spa_close(spa
, FTAG
);
6383 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6385 spa_deactivate(spa
);
6389 mutex_exit(&spa_namespace_lock
);
6393 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6398 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6402 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6403 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6404 if (vd
->vdev_guid
== guid
)
6408 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6409 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6410 if (vd
->vdev_guid
== guid
)
6419 spa_upgrade(spa_t
*spa
, uint64_t version
)
6421 ASSERT(spa_writeable(spa
));
6423 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6426 * This should only be called for a non-faulted pool, and since a
6427 * future version would result in an unopenable pool, this shouldn't be
6430 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
6431 ASSERT(version
>= spa
->spa_uberblock
.ub_version
);
6433 spa
->spa_uberblock
.ub_version
= version
;
6434 vdev_config_dirty(spa
->spa_root_vdev
);
6436 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6438 txg_wait_synced(spa_get_dsl(spa
), 0);
6442 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6446 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6448 for (i
= 0; i
< sav
->sav_count
; i
++)
6449 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6452 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6453 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6454 &spareguid
) == 0 && spareguid
== guid
)
6462 * Check if a pool has an active shared spare device.
6463 * Note: reference count of an active spare is 2, as a spare and as a replace
6466 spa_has_active_shared_spare(spa_t
*spa
)
6470 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6472 for (i
= 0; i
< sav
->sav_count
; i
++) {
6473 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6474 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6483 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6484 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6485 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6486 * or zdb as real changes.
6489 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6492 zfs_ereport_post(name
, spa
, vd
, NULL
, 0, 0);
6496 #if defined(_KERNEL) && defined(HAVE_SPL)
6497 /* state manipulation functions */
6498 EXPORT_SYMBOL(spa_open
);
6499 EXPORT_SYMBOL(spa_open_rewind
);
6500 EXPORT_SYMBOL(spa_get_stats
);
6501 EXPORT_SYMBOL(spa_create
);
6502 EXPORT_SYMBOL(spa_import_rootpool
);
6503 EXPORT_SYMBOL(spa_import
);
6504 EXPORT_SYMBOL(spa_tryimport
);
6505 EXPORT_SYMBOL(spa_destroy
);
6506 EXPORT_SYMBOL(spa_export
);
6507 EXPORT_SYMBOL(spa_reset
);
6508 EXPORT_SYMBOL(spa_async_request
);
6509 EXPORT_SYMBOL(spa_async_suspend
);
6510 EXPORT_SYMBOL(spa_async_resume
);
6511 EXPORT_SYMBOL(spa_inject_addref
);
6512 EXPORT_SYMBOL(spa_inject_delref
);
6513 EXPORT_SYMBOL(spa_scan_stat_init
);
6514 EXPORT_SYMBOL(spa_scan_get_stats
);
6516 /* device maniion */
6517 EXPORT_SYMBOL(spa_vdev_add
);
6518 EXPORT_SYMBOL(spa_vdev_attach
);
6519 EXPORT_SYMBOL(spa_vdev_detach
);
6520 EXPORT_SYMBOL(spa_vdev_remove
);
6521 EXPORT_SYMBOL(spa_vdev_setpath
);
6522 EXPORT_SYMBOL(spa_vdev_setfru
);
6523 EXPORT_SYMBOL(spa_vdev_split_mirror
);
6525 /* spare statech is global across all pools) */
6526 EXPORT_SYMBOL(spa_spare_add
);
6527 EXPORT_SYMBOL(spa_spare_remove
);
6528 EXPORT_SYMBOL(spa_spare_exists
);
6529 EXPORT_SYMBOL(spa_spare_activate
);
6531 /* L2ARC statech is global across all pools) */
6532 EXPORT_SYMBOL(spa_l2cache_add
);
6533 EXPORT_SYMBOL(spa_l2cache_remove
);
6534 EXPORT_SYMBOL(spa_l2cache_exists
);
6535 EXPORT_SYMBOL(spa_l2cache_activate
);
6536 EXPORT_SYMBOL(spa_l2cache_drop
);
6539 EXPORT_SYMBOL(spa_scan
);
6540 EXPORT_SYMBOL(spa_scan_stop
);
6543 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
6544 EXPORT_SYMBOL(spa_sync_allpools
);
6547 EXPORT_SYMBOL(spa_prop_set
);
6548 EXPORT_SYMBOL(spa_prop_get
);
6549 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
6551 /* asynchronous event notification */
6552 EXPORT_SYMBOL(spa_event_notify
);