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>
69 #include <sys/bootprops.h>
70 #include <sys/callb.h>
71 #include <sys/cpupart.h>
73 #include <sys/sysdc.h>
78 #include "zfs_comutil.h"
80 typedef enum zti_modes
{
81 zti_mode_fixed
, /* value is # of threads (min 1) */
82 zti_mode_online_percent
, /* value is % of online CPUs */
83 zti_mode_batch
, /* cpu-intensive; value is ignored */
84 zti_mode_null
, /* don't create a taskq */
88 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
89 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
90 #define ZTI_BATCH { zti_mode_batch, 0 }
91 #define ZTI_NULL { zti_mode_null, 0 }
93 #define ZTI_ONE ZTI_FIX(1)
95 typedef struct zio_taskq_info
{
96 enum zti_modes zti_mode
;
100 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
101 "iss", "iss_h", "int", "int_h"
105 * Define the taskq threads for the following I/O types:
106 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
108 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
109 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
110 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
111 { ZTI_FIX(8), ZTI_NULL
, ZTI_BATCH
, ZTI_NULL
},
112 { ZTI_BATCH
, ZTI_FIX(5), ZTI_FIX(16), ZTI_FIX(5) },
113 { ZTI_PCT(100), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
114 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
115 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
118 static dsl_syncfunc_t spa_sync_version
;
119 static dsl_syncfunc_t spa_sync_props
;
120 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
121 static inline int spa_load_impl(spa_t
*spa
, uint64_t, nvlist_t
*config
,
122 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
124 static void spa_vdev_resilver_done(spa_t
*spa
);
126 uint_t zio_taskq_batch_pct
= 100; /* 1 thread per cpu in pset */
127 id_t zio_taskq_psrset_bind
= PS_NONE
;
128 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
129 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
131 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
134 * This (illegal) pool name is used when temporarily importing a spa_t in order
135 * to get the vdev stats associated with the imported devices.
137 #define TRYIMPORT_NAME "$import"
140 * ==========================================================================
141 * SPA properties routines
142 * ==========================================================================
146 * Add a (source=src, propname=propval) list to an nvlist.
149 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
150 uint64_t intval
, zprop_source_t src
)
152 const char *propname
= zpool_prop_to_name(prop
);
155 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
156 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
159 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
161 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
163 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
164 nvlist_free(propval
);
168 * Get property values from the spa configuration.
171 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
173 vdev_t
*rvd
= spa
->spa_root_vdev
;
174 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
178 uint64_t cap
, version
;
179 zprop_source_t src
= ZPROP_SRC_NONE
;
180 spa_config_dirent_t
*dp
;
183 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
186 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
187 size
= metaslab_class_get_space(spa_normal_class(spa
));
188 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
189 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
190 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
191 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
195 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
196 vdev_t
*tvd
= rvd
->vdev_child
[c
];
197 space
+= tvd
->vdev_max_asize
- tvd
->vdev_asize
;
199 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
, space
,
202 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
203 (spa_mode(spa
) == FREAD
), src
);
205 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
206 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
208 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
209 ddt_get_pool_dedup_ratio(spa
), src
);
211 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
212 rvd
->vdev_state
, src
);
214 version
= spa_version(spa
);
215 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
))
216 src
= ZPROP_SRC_DEFAULT
;
218 src
= ZPROP_SRC_LOCAL
;
219 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
223 dsl_dir_t
*freedir
= pool
->dp_free_dir
;
226 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
227 * when opening pools before this version freedir will be NULL.
229 if (freedir
!= NULL
) {
230 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
231 freedir
->dd_phys
->dd_used_bytes
, src
);
233 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
238 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
240 if (spa
->spa_comment
!= NULL
) {
241 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
245 if (spa
->spa_root
!= NULL
)
246 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
249 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
250 if (dp
->scd_path
== NULL
) {
251 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
252 "none", 0, ZPROP_SRC_LOCAL
);
253 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
254 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
255 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
261 * Get zpool property values.
264 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
266 objset_t
*mos
= spa
->spa_meta_objset
;
271 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_PUSHPAGE
);
275 mutex_enter(&spa
->spa_props_lock
);
278 * Get properties from the spa config.
280 spa_prop_get_config(spa
, nvp
);
282 /* If no pool property object, no more prop to get. */
283 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
284 mutex_exit(&spa
->spa_props_lock
);
289 * Get properties from the MOS pool property object.
291 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
292 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
293 zap_cursor_advance(&zc
)) {
296 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
299 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
302 switch (za
.za_integer_length
) {
304 /* integer property */
305 if (za
.za_first_integer
!=
306 zpool_prop_default_numeric(prop
))
307 src
= ZPROP_SRC_LOCAL
;
309 if (prop
== ZPOOL_PROP_BOOTFS
) {
311 dsl_dataset_t
*ds
= NULL
;
313 dp
= spa_get_dsl(spa
);
314 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
315 if ((err
= dsl_dataset_hold_obj(dp
,
316 za
.za_first_integer
, FTAG
, &ds
))) {
317 rw_exit(&dp
->dp_config_rwlock
);
322 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1,
324 dsl_dataset_name(ds
, strval
);
325 dsl_dataset_rele(ds
, FTAG
);
326 rw_exit(&dp
->dp_config_rwlock
);
329 intval
= za
.za_first_integer
;
332 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
336 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1);
341 /* string property */
342 strval
= kmem_alloc(za
.za_num_integers
, KM_PUSHPAGE
);
343 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
344 za
.za_name
, 1, za
.za_num_integers
, strval
);
346 kmem_free(strval
, za
.za_num_integers
);
349 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
350 kmem_free(strval
, za
.za_num_integers
);
357 zap_cursor_fini(&zc
);
358 mutex_exit(&spa
->spa_props_lock
);
360 if (err
&& err
!= ENOENT
) {
370 * Validate the given pool properties nvlist and modify the list
371 * for the property values to be set.
374 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
377 int error
= 0, reset_bootfs
= 0;
379 boolean_t has_feature
= B_FALSE
;
382 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
384 char *strval
, *slash
, *check
, *fname
;
385 const char *propname
= nvpair_name(elem
);
386 zpool_prop_t prop
= zpool_name_to_prop(propname
);
390 if (!zpool_prop_feature(propname
)) {
396 * Sanitize the input.
398 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
403 if (nvpair_value_uint64(elem
, &intval
) != 0) {
413 fname
= strchr(propname
, '@') + 1;
414 if (zfeature_lookup_name(fname
, NULL
) != 0) {
419 has_feature
= B_TRUE
;
422 case ZPOOL_PROP_VERSION
:
423 error
= nvpair_value_uint64(elem
, &intval
);
425 (intval
< spa_version(spa
) ||
426 intval
> SPA_VERSION_BEFORE_FEATURES
||
431 case ZPOOL_PROP_DELEGATION
:
432 case ZPOOL_PROP_AUTOREPLACE
:
433 case ZPOOL_PROP_LISTSNAPS
:
434 case ZPOOL_PROP_AUTOEXPAND
:
435 error
= nvpair_value_uint64(elem
, &intval
);
436 if (!error
&& intval
> 1)
440 case ZPOOL_PROP_BOOTFS
:
442 * If the pool version is less than SPA_VERSION_BOOTFS,
443 * or the pool is still being created (version == 0),
444 * the bootfs property cannot be set.
446 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
452 * Make sure the vdev config is bootable
454 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
461 error
= nvpair_value_string(elem
, &strval
);
467 if (strval
== NULL
|| strval
[0] == '\0') {
468 objnum
= zpool_prop_default_numeric(
473 if ((error
= dmu_objset_hold(strval
,FTAG
,&os
)))
476 /* Must be ZPL and not gzip compressed. */
478 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
480 } else if ((error
= dsl_prop_get_integer(strval
,
481 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
482 &compress
, NULL
)) == 0 &&
483 !BOOTFS_COMPRESS_VALID(compress
)) {
486 objnum
= dmu_objset_id(os
);
488 dmu_objset_rele(os
, FTAG
);
492 case ZPOOL_PROP_FAILUREMODE
:
493 error
= nvpair_value_uint64(elem
, &intval
);
494 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
495 intval
> ZIO_FAILURE_MODE_PANIC
))
499 * This is a special case which only occurs when
500 * the pool has completely failed. This allows
501 * the user to change the in-core failmode property
502 * without syncing it out to disk (I/Os might
503 * currently be blocked). We do this by returning
504 * EIO to the caller (spa_prop_set) to trick it
505 * into thinking we encountered a property validation
508 if (!error
&& spa_suspended(spa
)) {
509 spa
->spa_failmode
= intval
;
514 case ZPOOL_PROP_CACHEFILE
:
515 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
518 if (strval
[0] == '\0')
521 if (strcmp(strval
, "none") == 0)
524 if (strval
[0] != '/') {
529 slash
= strrchr(strval
, '/');
530 ASSERT(slash
!= NULL
);
532 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
533 strcmp(slash
, "/..") == 0)
537 case ZPOOL_PROP_COMMENT
:
538 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
540 for (check
= strval
; *check
!= '\0'; check
++) {
541 if (!isprint(*check
)) {
547 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
551 case ZPOOL_PROP_DEDUPDITTO
:
552 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
555 error
= nvpair_value_uint64(elem
, &intval
);
557 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
569 if (!error
&& reset_bootfs
) {
570 error
= nvlist_remove(props
,
571 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
574 error
= nvlist_add_uint64(props
,
575 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
583 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
586 spa_config_dirent_t
*dp
;
588 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
592 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
595 if (cachefile
[0] == '\0')
596 dp
->scd_path
= spa_strdup(spa_config_path
);
597 else if (strcmp(cachefile
, "none") == 0)
600 dp
->scd_path
= spa_strdup(cachefile
);
602 list_insert_head(&spa
->spa_config_list
, dp
);
604 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
608 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
611 nvpair_t
*elem
= NULL
;
612 boolean_t need_sync
= B_FALSE
;
614 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
617 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
618 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
620 if (prop
== ZPOOL_PROP_CACHEFILE
||
621 prop
== ZPOOL_PROP_ALTROOT
||
622 prop
== ZPOOL_PROP_READONLY
)
625 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPROP_INVAL
) {
628 if (prop
== ZPOOL_PROP_VERSION
) {
629 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
631 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
632 ver
= SPA_VERSION_FEATURES
;
636 /* Save time if the version is already set. */
637 if (ver
== spa_version(spa
))
641 * In addition to the pool directory object, we might
642 * create the pool properties object, the features for
643 * read object, the features for write object, or the
644 * feature descriptions object.
646 error
= dsl_sync_task_do(spa_get_dsl(spa
), NULL
,
647 spa_sync_version
, spa
, &ver
, 6);
658 return (dsl_sync_task_do(spa_get_dsl(spa
), NULL
, spa_sync_props
,
666 * If the bootfs property value is dsobj, clear it.
669 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
671 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
672 VERIFY(zap_remove(spa
->spa_meta_objset
,
673 spa
->spa_pool_props_object
,
674 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
680 * Change the GUID for the pool. This is done so that we can later
681 * re-import a pool built from a clone of our own vdevs. We will modify
682 * the root vdev's guid, our own pool guid, and then mark all of our
683 * vdevs dirty. Note that we must make sure that all our vdevs are
684 * online when we do this, or else any vdevs that weren't present
685 * would be orphaned from our pool. We are also going to issue a
686 * sysevent to update any watchers.
689 spa_change_guid(spa_t
*spa
)
691 uint64_t oldguid
, newguid
;
694 if (!(spa_mode_global
& FWRITE
))
697 txg
= spa_vdev_enter(spa
);
699 if (spa
->spa_root_vdev
->vdev_state
!= VDEV_STATE_HEALTHY
)
700 return (spa_vdev_exit(spa
, NULL
, txg
, ENXIO
));
702 oldguid
= spa_guid(spa
);
703 newguid
= spa_generate_guid(NULL
);
704 ASSERT3U(oldguid
, !=, newguid
);
706 spa
->spa_root_vdev
->vdev_guid
= newguid
;
707 spa
->spa_root_vdev
->vdev_guid_sum
+= (newguid
- oldguid
);
709 vdev_config_dirty(spa
->spa_root_vdev
);
711 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_REGUID
);
713 return (spa_vdev_exit(spa
, NULL
, txg
, 0));
717 * ==========================================================================
718 * SPA state manipulation (open/create/destroy/import/export)
719 * ==========================================================================
723 spa_error_entry_compare(const void *a
, const void *b
)
725 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
726 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
729 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
730 sizeof (zbookmark_t
));
741 * Utility function which retrieves copies of the current logs and
742 * re-initializes them in the process.
745 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
747 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
749 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
750 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
752 avl_create(&spa
->spa_errlist_scrub
,
753 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
754 offsetof(spa_error_entry_t
, se_avl
));
755 avl_create(&spa
->spa_errlist_last
,
756 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
757 offsetof(spa_error_entry_t
, se_avl
));
761 spa_taskq_create(spa_t
*spa
, const char *name
, enum zti_modes mode
,
764 uint_t flags
= TASKQ_PREPOPULATE
;
765 boolean_t batch
= B_FALSE
;
769 return (NULL
); /* no taskq needed */
772 ASSERT3U(value
, >=, 1);
773 value
= MAX(value
, 1);
778 flags
|= TASKQ_THREADS_CPU_PCT
;
779 value
= zio_taskq_batch_pct
;
782 case zti_mode_online_percent
:
783 flags
|= TASKQ_THREADS_CPU_PCT
;
787 panic("unrecognized mode for %s taskq (%u:%u) in "
793 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
795 flags
|= TASKQ_DC_BATCH
;
797 return (taskq_create_sysdc(name
, value
, 50, INT_MAX
,
798 spa
->spa_proc
, zio_taskq_basedc
, flags
));
800 return (taskq_create_proc(name
, value
, maxclsyspri
, 50, INT_MAX
,
801 spa
->spa_proc
, flags
));
805 spa_create_zio_taskqs(spa_t
*spa
)
809 for (t
= 0; t
< ZIO_TYPES
; t
++) {
810 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
811 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
812 enum zti_modes mode
= ztip
->zti_mode
;
813 uint_t value
= ztip
->zti_value
;
816 (void) snprintf(name
, sizeof (name
),
817 "%s_%s", zio_type_name
[t
], zio_taskq_types
[q
]);
819 spa
->spa_zio_taskq
[t
][q
] =
820 spa_taskq_create(spa
, name
, mode
, value
);
825 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
827 spa_thread(void *arg
)
832 user_t
*pu
= PTOU(curproc
);
834 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
837 ASSERT(curproc
!= &p0
);
838 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
839 "zpool-%s", spa
->spa_name
);
840 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
842 /* bind this thread to the requested psrset */
843 if (zio_taskq_psrset_bind
!= PS_NONE
) {
845 mutex_enter(&cpu_lock
);
846 mutex_enter(&pidlock
);
847 mutex_enter(&curproc
->p_lock
);
849 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
850 0, NULL
, NULL
) == 0) {
851 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
854 "Couldn't bind process for zfs pool \"%s\" to "
855 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
858 mutex_exit(&curproc
->p_lock
);
859 mutex_exit(&pidlock
);
860 mutex_exit(&cpu_lock
);
864 if (zio_taskq_sysdc
) {
865 sysdc_thread_enter(curthread
, 100, 0);
868 spa
->spa_proc
= curproc
;
869 spa
->spa_did
= curthread
->t_did
;
871 spa_create_zio_taskqs(spa
);
873 mutex_enter(&spa
->spa_proc_lock
);
874 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
876 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
877 cv_broadcast(&spa
->spa_proc_cv
);
879 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
880 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
881 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
882 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
884 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
885 spa
->spa_proc_state
= SPA_PROC_GONE
;
887 cv_broadcast(&spa
->spa_proc_cv
);
888 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
890 mutex_enter(&curproc
->p_lock
);
896 * Activate an uninitialized pool.
899 spa_activate(spa_t
*spa
, int mode
)
901 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
903 spa
->spa_state
= POOL_STATE_ACTIVE
;
904 spa
->spa_mode
= mode
;
906 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
907 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
909 /* Try to create a covering process */
910 mutex_enter(&spa
->spa_proc_lock
);
911 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
912 ASSERT(spa
->spa_proc
== &p0
);
915 #ifdef HAVE_SPA_THREAD
916 /* Only create a process if we're going to be around a while. */
917 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
918 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
920 spa
->spa_proc_state
= SPA_PROC_CREATED
;
921 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
922 cv_wait(&spa
->spa_proc_cv
,
923 &spa
->spa_proc_lock
);
925 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
926 ASSERT(spa
->spa_proc
!= &p0
);
927 ASSERT(spa
->spa_did
!= 0);
931 "Couldn't create process for zfs pool \"%s\"\n",
936 #endif /* HAVE_SPA_THREAD */
937 mutex_exit(&spa
->spa_proc_lock
);
939 /* If we didn't create a process, we need to create our taskqs. */
940 if (spa
->spa_proc
== &p0
) {
941 spa_create_zio_taskqs(spa
);
944 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
945 offsetof(vdev_t
, vdev_config_dirty_node
));
946 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
947 offsetof(vdev_t
, vdev_state_dirty_node
));
949 txg_list_create(&spa
->spa_vdev_txg_list
,
950 offsetof(struct vdev
, vdev_txg_node
));
952 avl_create(&spa
->spa_errlist_scrub
,
953 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
954 offsetof(spa_error_entry_t
, se_avl
));
955 avl_create(&spa
->spa_errlist_last
,
956 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
957 offsetof(spa_error_entry_t
, se_avl
));
961 * Opposite of spa_activate().
964 spa_deactivate(spa_t
*spa
)
968 ASSERT(spa
->spa_sync_on
== B_FALSE
);
969 ASSERT(spa
->spa_dsl_pool
== NULL
);
970 ASSERT(spa
->spa_root_vdev
== NULL
);
971 ASSERT(spa
->spa_async_zio_root
== NULL
);
972 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
974 txg_list_destroy(&spa
->spa_vdev_txg_list
);
976 list_destroy(&spa
->spa_config_dirty_list
);
977 list_destroy(&spa
->spa_state_dirty_list
);
979 for (t
= 0; t
< ZIO_TYPES
; t
++) {
980 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
981 if (spa
->spa_zio_taskq
[t
][q
] != NULL
)
982 taskq_destroy(spa
->spa_zio_taskq
[t
][q
]);
983 spa
->spa_zio_taskq
[t
][q
] = NULL
;
987 metaslab_class_destroy(spa
->spa_normal_class
);
988 spa
->spa_normal_class
= NULL
;
990 metaslab_class_destroy(spa
->spa_log_class
);
991 spa
->spa_log_class
= NULL
;
994 * If this was part of an import or the open otherwise failed, we may
995 * still have errors left in the queues. Empty them just in case.
997 spa_errlog_drain(spa
);
999 avl_destroy(&spa
->spa_errlist_scrub
);
1000 avl_destroy(&spa
->spa_errlist_last
);
1002 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1004 mutex_enter(&spa
->spa_proc_lock
);
1005 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1006 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1007 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1008 cv_broadcast(&spa
->spa_proc_cv
);
1009 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1010 ASSERT(spa
->spa_proc
!= &p0
);
1011 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1013 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1014 spa
->spa_proc_state
= SPA_PROC_NONE
;
1016 ASSERT(spa
->spa_proc
== &p0
);
1017 mutex_exit(&spa
->spa_proc_lock
);
1020 * We want to make sure spa_thread() has actually exited the ZFS
1021 * module, so that the module can't be unloaded out from underneath
1024 if (spa
->spa_did
!= 0) {
1025 thread_join(spa
->spa_did
);
1031 * Verify a pool configuration, and construct the vdev tree appropriately. This
1032 * will create all the necessary vdevs in the appropriate layout, with each vdev
1033 * in the CLOSED state. This will prep the pool before open/creation/import.
1034 * All vdev validation is done by the vdev_alloc() routine.
1037 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1038 uint_t id
, int atype
)
1045 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1048 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1051 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1054 if (error
== ENOENT
)
1063 for (c
= 0; c
< children
; c
++) {
1065 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1073 ASSERT(*vdp
!= NULL
);
1079 * Opposite of spa_load().
1082 spa_unload(spa_t
*spa
)
1086 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1091 spa_async_suspend(spa
);
1096 if (spa
->spa_sync_on
) {
1097 txg_sync_stop(spa
->spa_dsl_pool
);
1098 spa
->spa_sync_on
= B_FALSE
;
1102 * Wait for any outstanding async I/O to complete.
1104 if (spa
->spa_async_zio_root
!= NULL
) {
1105 (void) zio_wait(spa
->spa_async_zio_root
);
1106 spa
->spa_async_zio_root
= NULL
;
1109 bpobj_close(&spa
->spa_deferred_bpobj
);
1112 * Close the dsl pool.
1114 if (spa
->spa_dsl_pool
) {
1115 dsl_pool_close(spa
->spa_dsl_pool
);
1116 spa
->spa_dsl_pool
= NULL
;
1117 spa
->spa_meta_objset
= NULL
;
1122 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1125 * Drop and purge level 2 cache
1127 spa_l2cache_drop(spa
);
1132 if (spa
->spa_root_vdev
)
1133 vdev_free(spa
->spa_root_vdev
);
1134 ASSERT(spa
->spa_root_vdev
== NULL
);
1136 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1137 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1138 if (spa
->spa_spares
.sav_vdevs
) {
1139 kmem_free(spa
->spa_spares
.sav_vdevs
,
1140 spa
->spa_spares
.sav_count
* sizeof (void *));
1141 spa
->spa_spares
.sav_vdevs
= NULL
;
1143 if (spa
->spa_spares
.sav_config
) {
1144 nvlist_free(spa
->spa_spares
.sav_config
);
1145 spa
->spa_spares
.sav_config
= NULL
;
1147 spa
->spa_spares
.sav_count
= 0;
1149 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1150 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1151 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1153 if (spa
->spa_l2cache
.sav_vdevs
) {
1154 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1155 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1156 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1158 if (spa
->spa_l2cache
.sav_config
) {
1159 nvlist_free(spa
->spa_l2cache
.sav_config
);
1160 spa
->spa_l2cache
.sav_config
= NULL
;
1162 spa
->spa_l2cache
.sav_count
= 0;
1164 spa
->spa_async_suspended
= 0;
1166 if (spa
->spa_comment
!= NULL
) {
1167 spa_strfree(spa
->spa_comment
);
1168 spa
->spa_comment
= NULL
;
1171 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1175 * Load (or re-load) the current list of vdevs describing the active spares for
1176 * this pool. When this is called, we have some form of basic information in
1177 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1178 * then re-generate a more complete list including status information.
1181 spa_load_spares(spa_t
*spa
)
1188 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1191 * First, close and free any existing spare vdevs.
1193 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1194 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1196 /* Undo the call to spa_activate() below */
1197 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1198 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1199 spa_spare_remove(tvd
);
1204 if (spa
->spa_spares
.sav_vdevs
)
1205 kmem_free(spa
->spa_spares
.sav_vdevs
,
1206 spa
->spa_spares
.sav_count
* sizeof (void *));
1208 if (spa
->spa_spares
.sav_config
== NULL
)
1211 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1212 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1214 spa
->spa_spares
.sav_count
= (int)nspares
;
1215 spa
->spa_spares
.sav_vdevs
= NULL
;
1221 * Construct the array of vdevs, opening them to get status in the
1222 * process. For each spare, there is potentially two different vdev_t
1223 * structures associated with it: one in the list of spares (used only
1224 * for basic validation purposes) and one in the active vdev
1225 * configuration (if it's spared in). During this phase we open and
1226 * validate each vdev on the spare list. If the vdev also exists in the
1227 * active configuration, then we also mark this vdev as an active spare.
1229 spa
->spa_spares
.sav_vdevs
= kmem_alloc(nspares
* sizeof (void *),
1231 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1232 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1233 VDEV_ALLOC_SPARE
) == 0);
1236 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1238 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1239 B_FALSE
)) != NULL
) {
1240 if (!tvd
->vdev_isspare
)
1244 * We only mark the spare active if we were successfully
1245 * able to load the vdev. Otherwise, importing a pool
1246 * with a bad active spare would result in strange
1247 * behavior, because multiple pool would think the spare
1248 * is actively in use.
1250 * There is a vulnerability here to an equally bizarre
1251 * circumstance, where a dead active spare is later
1252 * brought back to life (onlined or otherwise). Given
1253 * the rarity of this scenario, and the extra complexity
1254 * it adds, we ignore the possibility.
1256 if (!vdev_is_dead(tvd
))
1257 spa_spare_activate(tvd
);
1261 vd
->vdev_aux
= &spa
->spa_spares
;
1263 if (vdev_open(vd
) != 0)
1266 if (vdev_validate_aux(vd
) == 0)
1271 * Recompute the stashed list of spares, with status information
1274 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1275 DATA_TYPE_NVLIST_ARRAY
) == 0);
1277 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1279 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1280 spares
[i
] = vdev_config_generate(spa
,
1281 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1282 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1283 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1284 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1285 nvlist_free(spares
[i
]);
1286 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1290 * Load (or re-load) the current list of vdevs describing the active l2cache for
1291 * this pool. When this is called, we have some form of basic information in
1292 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1293 * then re-generate a more complete list including status information.
1294 * Devices which are already active have their details maintained, and are
1298 spa_load_l2cache(spa_t
*spa
)
1302 int i
, j
, oldnvdevs
;
1304 vdev_t
*vd
, **oldvdevs
, **newvdevs
= NULL
;
1305 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1307 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1309 if (sav
->sav_config
!= NULL
) {
1310 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1311 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1312 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_PUSHPAGE
);
1317 oldvdevs
= sav
->sav_vdevs
;
1318 oldnvdevs
= sav
->sav_count
;
1319 sav
->sav_vdevs
= NULL
;
1323 * Process new nvlist of vdevs.
1325 for (i
= 0; i
< nl2cache
; i
++) {
1326 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1330 for (j
= 0; j
< oldnvdevs
; j
++) {
1332 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1334 * Retain previous vdev for add/remove ops.
1342 if (newvdevs
[i
] == NULL
) {
1346 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1347 VDEV_ALLOC_L2CACHE
) == 0);
1352 * Commit this vdev as an l2cache device,
1353 * even if it fails to open.
1355 spa_l2cache_add(vd
);
1360 spa_l2cache_activate(vd
);
1362 if (vdev_open(vd
) != 0)
1365 (void) vdev_validate_aux(vd
);
1367 if (!vdev_is_dead(vd
))
1368 l2arc_add_vdev(spa
, vd
);
1373 * Purge vdevs that were dropped
1375 for (i
= 0; i
< oldnvdevs
; i
++) {
1380 ASSERT(vd
->vdev_isl2cache
);
1382 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1383 pool
!= 0ULL && l2arc_vdev_present(vd
))
1384 l2arc_remove_vdev(vd
);
1385 vdev_clear_stats(vd
);
1391 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1393 if (sav
->sav_config
== NULL
)
1396 sav
->sav_vdevs
= newvdevs
;
1397 sav
->sav_count
= (int)nl2cache
;
1400 * Recompute the stashed list of l2cache devices, with status
1401 * information this time.
1403 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1404 DATA_TYPE_NVLIST_ARRAY
) == 0);
1406 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_PUSHPAGE
);
1407 for (i
= 0; i
< sav
->sav_count
; i
++)
1408 l2cache
[i
] = vdev_config_generate(spa
,
1409 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1410 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1411 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1413 for (i
= 0; i
< sav
->sav_count
; i
++)
1414 nvlist_free(l2cache
[i
]);
1416 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1420 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1423 char *packed
= NULL
;
1428 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1432 nvsize
= *(uint64_t *)db
->db_data
;
1433 dmu_buf_rele(db
, FTAG
);
1435 packed
= kmem_alloc(nvsize
, KM_PUSHPAGE
| KM_NODEBUG
);
1436 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1439 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1440 kmem_free(packed
, nvsize
);
1446 * Checks to see if the given vdev could not be opened, in which case we post a
1447 * sysevent to notify the autoreplace code that the device has been removed.
1450 spa_check_removed(vdev_t
*vd
)
1454 for (c
= 0; c
< vd
->vdev_children
; c
++)
1455 spa_check_removed(vd
->vdev_child
[c
]);
1457 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
)) {
1458 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE
,
1459 vd
->vdev_spa
, vd
, NULL
, 0, 0);
1460 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_CHECK
);
1465 * Validate the current config against the MOS config
1468 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1470 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1474 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1476 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1477 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1479 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1482 * If we're doing a normal import, then build up any additional
1483 * diagnostic information about missing devices in this config.
1484 * We'll pass this up to the user for further processing.
1486 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1487 nvlist_t
**child
, *nv
;
1490 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1492 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
1494 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1495 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1496 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1498 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1499 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1501 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1506 VERIFY(nvlist_add_nvlist_array(nv
,
1507 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1508 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1509 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1511 for (i
= 0; i
< idx
; i
++)
1512 nvlist_free(child
[i
]);
1515 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1519 * Compare the root vdev tree with the information we have
1520 * from the MOS config (mrvd). Check each top-level vdev
1521 * with the corresponding MOS config top-level (mtvd).
1523 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1524 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1525 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1528 * Resolve any "missing" vdevs in the current configuration.
1529 * If we find that the MOS config has more accurate information
1530 * about the top-level vdev then use that vdev instead.
1532 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1533 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1535 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1539 * Device specific actions.
1541 if (mtvd
->vdev_islog
) {
1542 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1545 * XXX - once we have 'readonly' pool
1546 * support we should be able to handle
1547 * missing data devices by transitioning
1548 * the pool to readonly.
1554 * Swap the missing vdev with the data we were
1555 * able to obtain from the MOS config.
1557 vdev_remove_child(rvd
, tvd
);
1558 vdev_remove_child(mrvd
, mtvd
);
1560 vdev_add_child(rvd
, mtvd
);
1561 vdev_add_child(mrvd
, tvd
);
1563 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1565 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1568 } else if (mtvd
->vdev_islog
) {
1570 * Load the slog device's state from the MOS config
1571 * since it's possible that the label does not
1572 * contain the most up-to-date information.
1574 vdev_load_log_state(tvd
, mtvd
);
1579 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1582 * Ensure we were able to validate the config.
1584 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1588 * Check for missing log devices
1591 spa_check_logs(spa_t
*spa
)
1593 switch (spa
->spa_log_state
) {
1596 case SPA_LOG_MISSING
:
1597 /* need to recheck in case slog has been restored */
1598 case SPA_LOG_UNKNOWN
:
1599 if (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
, NULL
,
1600 DS_FIND_CHILDREN
)) {
1601 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1610 spa_passivate_log(spa_t
*spa
)
1612 vdev_t
*rvd
= spa
->spa_root_vdev
;
1613 boolean_t slog_found
= B_FALSE
;
1616 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1618 if (!spa_has_slogs(spa
))
1621 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1622 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1623 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1625 if (tvd
->vdev_islog
) {
1626 metaslab_group_passivate(mg
);
1627 slog_found
= B_TRUE
;
1631 return (slog_found
);
1635 spa_activate_log(spa_t
*spa
)
1637 vdev_t
*rvd
= spa
->spa_root_vdev
;
1640 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1642 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1643 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1644 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1646 if (tvd
->vdev_islog
)
1647 metaslab_group_activate(mg
);
1652 spa_offline_log(spa_t
*spa
)
1656 if ((error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1657 NULL
, DS_FIND_CHILDREN
)) == 0) {
1660 * We successfully offlined the log device, sync out the
1661 * current txg so that the "stubby" block can be removed
1664 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1670 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1674 for (i
= 0; i
< sav
->sav_count
; i
++)
1675 spa_check_removed(sav
->sav_vdevs
[i
]);
1679 spa_claim_notify(zio_t
*zio
)
1681 spa_t
*spa
= zio
->io_spa
;
1686 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1687 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1688 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1689 mutex_exit(&spa
->spa_props_lock
);
1692 typedef struct spa_load_error
{
1693 uint64_t sle_meta_count
;
1694 uint64_t sle_data_count
;
1698 spa_load_verify_done(zio_t
*zio
)
1700 blkptr_t
*bp
= zio
->io_bp
;
1701 spa_load_error_t
*sle
= zio
->io_private
;
1702 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1703 int error
= zio
->io_error
;
1706 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1707 type
!= DMU_OT_INTENT_LOG
)
1708 atomic_add_64(&sle
->sle_meta_count
, 1);
1710 atomic_add_64(&sle
->sle_data_count
, 1);
1712 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1717 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1718 arc_buf_t
*pbuf
, const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1722 size_t size
= BP_GET_PSIZE(bp
);
1723 void *data
= zio_data_buf_alloc(size
);
1725 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1726 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1727 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1728 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1734 spa_load_verify(spa_t
*spa
)
1737 spa_load_error_t sle
= { 0 };
1738 zpool_rewind_policy_t policy
;
1739 boolean_t verify_ok
= B_FALSE
;
1742 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1744 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1747 rio
= zio_root(spa
, NULL
, &sle
,
1748 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1750 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1751 TRAVERSE_PRE
| TRAVERSE_PREFETCH
, spa_load_verify_cb
, rio
);
1753 (void) zio_wait(rio
);
1755 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1756 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1758 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1759 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1763 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1764 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1766 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1767 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1768 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1769 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1770 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1771 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1772 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1774 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1778 if (error
!= ENXIO
&& error
!= EIO
)
1783 return (verify_ok
? 0 : EIO
);
1787 * Find a value in the pool props object.
1790 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
1792 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
1793 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
1797 * Find a value in the pool directory object.
1800 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
1802 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1803 name
, sizeof (uint64_t), 1, val
));
1807 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
1809 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
1814 * Fix up config after a partly-completed split. This is done with the
1815 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1816 * pool have that entry in their config, but only the splitting one contains
1817 * a list of all the guids of the vdevs that are being split off.
1819 * This function determines what to do with that list: either rejoin
1820 * all the disks to the pool, or complete the splitting process. To attempt
1821 * the rejoin, each disk that is offlined is marked online again, and
1822 * we do a reopen() call. If the vdev label for every disk that was
1823 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1824 * then we call vdev_split() on each disk, and complete the split.
1826 * Otherwise we leave the config alone, with all the vdevs in place in
1827 * the original pool.
1830 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
1837 boolean_t attempt_reopen
;
1839 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
1842 /* check that the config is complete */
1843 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
1844 &glist
, &gcount
) != 0)
1847 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_PUSHPAGE
);
1849 /* attempt to online all the vdevs & validate */
1850 attempt_reopen
= B_TRUE
;
1851 for (i
= 0; i
< gcount
; i
++) {
1852 if (glist
[i
] == 0) /* vdev is hole */
1855 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
1856 if (vd
[i
] == NULL
) {
1858 * Don't bother attempting to reopen the disks;
1859 * just do the split.
1861 attempt_reopen
= B_FALSE
;
1863 /* attempt to re-online it */
1864 vd
[i
]->vdev_offline
= B_FALSE
;
1868 if (attempt_reopen
) {
1869 vdev_reopen(spa
->spa_root_vdev
);
1871 /* check each device to see what state it's in */
1872 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
1873 if (vd
[i
] != NULL
&&
1874 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
1881 * If every disk has been moved to the new pool, or if we never
1882 * even attempted to look at them, then we split them off for
1885 if (!attempt_reopen
|| gcount
== extracted
) {
1886 for (i
= 0; i
< gcount
; i
++)
1889 vdev_reopen(spa
->spa_root_vdev
);
1892 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
1896 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
1897 boolean_t mosconfig
)
1899 nvlist_t
*config
= spa
->spa_config
;
1900 char *ereport
= FM_EREPORT_ZFS_POOL
;
1906 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
1909 ASSERT(spa
->spa_comment
== NULL
);
1910 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
1911 spa
->spa_comment
= spa_strdup(comment
);
1914 * Versioning wasn't explicitly added to the label until later, so if
1915 * it's not present treat it as the initial version.
1917 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
1918 &spa
->spa_ubsync
.ub_version
) != 0)
1919 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
1921 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
1922 &spa
->spa_config_txg
);
1924 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
1925 spa_guid_exists(pool_guid
, 0)) {
1928 spa
->spa_config_guid
= pool_guid
;
1930 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
1932 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
1936 nvlist_free(spa
->spa_load_info
);
1937 spa
->spa_load_info
= fnvlist_alloc();
1939 gethrestime(&spa
->spa_loaded_ts
);
1940 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
1941 mosconfig
, &ereport
);
1944 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
1946 if (error
!= EEXIST
) {
1947 spa
->spa_loaded_ts
.tv_sec
= 0;
1948 spa
->spa_loaded_ts
.tv_nsec
= 0;
1950 if (error
!= EBADF
) {
1951 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
1954 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
1961 * Load an existing storage pool, using the pool's builtin spa_config as a
1962 * source of configuration information.
1964 __attribute__((always_inline
))
1966 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
1967 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
1971 nvlist_t
*nvroot
= NULL
;
1974 uberblock_t
*ub
= &spa
->spa_uberblock
;
1975 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
1976 int orig_mode
= spa
->spa_mode
;
1979 boolean_t missing_feat_write
= B_FALSE
;
1982 * If this is an untrusted config, access the pool in read-only mode.
1983 * This prevents things like resilvering recently removed devices.
1986 spa
->spa_mode
= FREAD
;
1988 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1990 spa
->spa_load_state
= state
;
1992 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
1995 parse
= (type
== SPA_IMPORT_EXISTING
?
1996 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
1999 * Create "The Godfather" zio to hold all async IOs
2001 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
2002 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
2005 * Parse the configuration into a vdev tree. We explicitly set the
2006 * value that will be returned by spa_version() since parsing the
2007 * configuration requires knowing the version number.
2009 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2010 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2011 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2016 ASSERT(spa
->spa_root_vdev
== rvd
);
2018 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2019 ASSERT(spa_guid(spa
) == pool_guid
);
2023 * Try to open all vdevs, loading each label in the process.
2025 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2026 error
= vdev_open(rvd
);
2027 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2032 * We need to validate the vdev labels against the configuration that
2033 * we have in hand, which is dependent on the setting of mosconfig. If
2034 * mosconfig is true then we're validating the vdev labels based on
2035 * that config. Otherwise, we're validating against the cached config
2036 * (zpool.cache) that was read when we loaded the zfs module, and then
2037 * later we will recursively call spa_load() and validate against
2040 * If we're assembling a new pool that's been split off from an
2041 * existing pool, the labels haven't yet been updated so we skip
2042 * validation for now.
2044 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2045 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2046 error
= vdev_validate(rvd
, mosconfig
);
2047 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2052 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2057 * Find the best uberblock.
2059 vdev_uberblock_load(rvd
, ub
, &label
);
2062 * If we weren't able to find a single valid uberblock, return failure.
2064 if (ub
->ub_txg
== 0) {
2066 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2070 * If the pool has an unsupported version we can't open it.
2072 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2074 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2077 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2081 * If we weren't able to find what's necessary for reading the
2082 * MOS in the label, return failure.
2084 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2085 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2087 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2092 * Update our in-core representation with the definitive values
2095 nvlist_free(spa
->spa_label_features
);
2096 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2102 * Look through entries in the label nvlist's features_for_read. If
2103 * there is a feature listed there which we don't understand then we
2104 * cannot open a pool.
2106 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2107 nvlist_t
*unsup_feat
;
2110 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2113 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2115 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2116 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2117 VERIFY(nvlist_add_string(unsup_feat
,
2118 nvpair_name(nvp
), "") == 0);
2122 if (!nvlist_empty(unsup_feat
)) {
2123 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2124 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2125 nvlist_free(unsup_feat
);
2126 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2130 nvlist_free(unsup_feat
);
2134 * If the vdev guid sum doesn't match the uberblock, we have an
2135 * incomplete configuration. We first check to see if the pool
2136 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2137 * If it is, defer the vdev_guid_sum check till later so we
2138 * can handle missing vdevs.
2140 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2141 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2142 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2143 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2145 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2146 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2147 spa_try_repair(spa
, config
);
2148 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2149 nvlist_free(spa
->spa_config_splitting
);
2150 spa
->spa_config_splitting
= NULL
;
2154 * Initialize internal SPA structures.
2156 spa
->spa_state
= POOL_STATE_ACTIVE
;
2157 spa
->spa_ubsync
= spa
->spa_uberblock
;
2158 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2159 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2160 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2161 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2162 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2163 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2165 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2167 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2168 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2170 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2171 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2173 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2174 boolean_t missing_feat_read
= B_FALSE
;
2175 nvlist_t
*unsup_feat
;
2177 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2178 &spa
->spa_feat_for_read_obj
) != 0) {
2179 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2182 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2183 &spa
->spa_feat_for_write_obj
) != 0) {
2184 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2187 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2188 &spa
->spa_feat_desc_obj
) != 0) {
2189 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2192 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2195 if (!feature_is_supported(spa
->spa_meta_objset
,
2196 spa
->spa_feat_for_read_obj
, spa
->spa_feat_desc_obj
,
2198 missing_feat_read
= B_TRUE
;
2200 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2201 if (!feature_is_supported(spa
->spa_meta_objset
,
2202 spa
->spa_feat_for_write_obj
, spa
->spa_feat_desc_obj
,
2204 missing_feat_write
= B_TRUE
;
2207 if (!nvlist_empty(unsup_feat
)) {
2208 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2209 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2212 nvlist_free(unsup_feat
);
2214 if (!missing_feat_read
) {
2215 fnvlist_add_boolean(spa
->spa_load_info
,
2216 ZPOOL_CONFIG_CAN_RDONLY
);
2220 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2221 * twofold: to determine whether the pool is available for
2222 * import in read-write mode and (if it is not) whether the
2223 * pool is available for import in read-only mode. If the pool
2224 * is available for import in read-write mode, it is displayed
2225 * as available in userland; if it is not available for import
2226 * in read-only mode, it is displayed as unavailable in
2227 * userland. If the pool is available for import in read-only
2228 * mode but not read-write mode, it is displayed as unavailable
2229 * in userland with a special note that the pool is actually
2230 * available for open in read-only mode.
2232 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2233 * missing a feature for write, we must first determine whether
2234 * the pool can be opened read-only before returning to
2235 * userland in order to know whether to display the
2236 * abovementioned note.
2238 if (missing_feat_read
|| (missing_feat_write
&&
2239 spa_writeable(spa
))) {
2240 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2245 spa
->spa_is_initializing
= B_TRUE
;
2246 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2247 spa
->spa_is_initializing
= B_FALSE
;
2249 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2253 nvlist_t
*policy
= NULL
, *nvconfig
;
2255 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2256 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2258 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2259 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2261 unsigned long myhostid
= 0;
2263 VERIFY(nvlist_lookup_string(nvconfig
,
2264 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2267 myhostid
= zone_get_hostid(NULL
);
2270 * We're emulating the system's hostid in userland, so
2271 * we can't use zone_get_hostid().
2273 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2274 #endif /* _KERNEL */
2275 if (hostid
!= 0 && myhostid
!= 0 &&
2276 hostid
!= myhostid
) {
2277 nvlist_free(nvconfig
);
2278 cmn_err(CE_WARN
, "pool '%s' could not be "
2279 "loaded as it was last accessed by "
2280 "another system (host: %s hostid: 0x%lx). "
2281 "See: http://zfsonlinux.org/msg/ZFS-8000-EY",
2282 spa_name(spa
), hostname
,
2283 (unsigned long)hostid
);
2287 if (nvlist_lookup_nvlist(spa
->spa_config
,
2288 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2289 VERIFY(nvlist_add_nvlist(nvconfig
,
2290 ZPOOL_REWIND_POLICY
, policy
) == 0);
2292 spa_config_set(spa
, nvconfig
);
2294 spa_deactivate(spa
);
2295 spa_activate(spa
, orig_mode
);
2297 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2300 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2301 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2302 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2304 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2307 * Load the bit that tells us to use the new accounting function
2308 * (raid-z deflation). If we have an older pool, this will not
2311 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2312 if (error
!= 0 && error
!= ENOENT
)
2313 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2315 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2316 &spa
->spa_creation_version
);
2317 if (error
!= 0 && error
!= ENOENT
)
2318 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2321 * Load the persistent error log. If we have an older pool, this will
2324 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2325 if (error
!= 0 && error
!= ENOENT
)
2326 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2328 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2329 &spa
->spa_errlog_scrub
);
2330 if (error
!= 0 && error
!= ENOENT
)
2331 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2334 * Load the history object. If we have an older pool, this
2335 * will not be present.
2337 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2338 if (error
!= 0 && error
!= ENOENT
)
2339 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2342 * If we're assembling the pool from the split-off vdevs of
2343 * an existing pool, we don't want to attach the spares & cache
2348 * Load any hot spares for this pool.
2350 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2351 if (error
!= 0 && error
!= ENOENT
)
2352 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2353 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2354 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2355 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2356 &spa
->spa_spares
.sav_config
) != 0)
2357 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2359 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2360 spa_load_spares(spa
);
2361 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2362 } else if (error
== 0) {
2363 spa
->spa_spares
.sav_sync
= B_TRUE
;
2367 * Load any level 2 ARC devices for this pool.
2369 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2370 &spa
->spa_l2cache
.sav_object
);
2371 if (error
!= 0 && error
!= ENOENT
)
2372 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2373 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2374 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2375 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2376 &spa
->spa_l2cache
.sav_config
) != 0)
2377 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2379 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2380 spa_load_l2cache(spa
);
2381 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2382 } else if (error
== 0) {
2383 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2386 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2388 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2389 if (error
&& error
!= ENOENT
)
2390 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2393 uint64_t autoreplace
;
2395 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2396 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2397 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2398 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2399 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2400 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2401 &spa
->spa_dedup_ditto
);
2403 spa
->spa_autoreplace
= (autoreplace
!= 0);
2407 * If the 'autoreplace' property is set, then post a resource notifying
2408 * the ZFS DE that it should not issue any faults for unopenable
2409 * devices. We also iterate over the vdevs, and post a sysevent for any
2410 * unopenable vdevs so that the normal autoreplace handler can take
2413 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2414 spa_check_removed(spa
->spa_root_vdev
);
2416 * For the import case, this is done in spa_import(), because
2417 * at this point we're using the spare definitions from
2418 * the MOS config, not necessarily from the userland config.
2420 if (state
!= SPA_LOAD_IMPORT
) {
2421 spa_aux_check_removed(&spa
->spa_spares
);
2422 spa_aux_check_removed(&spa
->spa_l2cache
);
2427 * Load the vdev state for all toplevel vdevs.
2432 * Propagate the leaf DTLs we just loaded all the way up the tree.
2434 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2435 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2436 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2439 * Load the DDTs (dedup tables).
2441 error
= ddt_load(spa
);
2443 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2445 spa_update_dspace(spa
);
2448 * Validate the config, using the MOS config to fill in any
2449 * information which might be missing. If we fail to validate
2450 * the config then declare the pool unfit for use. If we're
2451 * assembling a pool from a split, the log is not transferred
2454 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2457 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2458 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2460 if (!spa_config_valid(spa
, nvconfig
)) {
2461 nvlist_free(nvconfig
);
2462 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2465 nvlist_free(nvconfig
);
2468 * Now that we've validated the config, check the state of the
2469 * root vdev. If it can't be opened, it indicates one or
2470 * more toplevel vdevs are faulted.
2472 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2475 if (spa_check_logs(spa
)) {
2476 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2477 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2481 if (missing_feat_write
) {
2482 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2485 * At this point, we know that we can open the pool in
2486 * read-only mode but not read-write mode. We now have enough
2487 * information and can return to userland.
2489 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2493 * We've successfully opened the pool, verify that we're ready
2494 * to start pushing transactions.
2496 if (state
!= SPA_LOAD_TRYIMPORT
) {
2497 if ((error
= spa_load_verify(spa
)))
2498 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2502 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2503 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2505 int need_update
= B_FALSE
;
2508 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2511 * Claim log blocks that haven't been committed yet.
2512 * This must all happen in a single txg.
2513 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2514 * invoked from zil_claim_log_block()'s i/o done callback.
2515 * Price of rollback is that we abandon the log.
2517 spa
->spa_claiming
= B_TRUE
;
2519 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
2520 spa_first_txg(spa
));
2521 (void) dmu_objset_find(spa_name(spa
),
2522 zil_claim
, tx
, DS_FIND_CHILDREN
);
2525 spa
->spa_claiming
= B_FALSE
;
2527 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2528 spa
->spa_sync_on
= B_TRUE
;
2529 txg_sync_start(spa
->spa_dsl_pool
);
2532 * Wait for all claims to sync. We sync up to the highest
2533 * claimed log block birth time so that claimed log blocks
2534 * don't appear to be from the future. spa_claim_max_txg
2535 * will have been set for us by either zil_check_log_chain()
2536 * (invoked from spa_check_logs()) or zil_claim() above.
2538 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2541 * If the config cache is stale, or we have uninitialized
2542 * metaslabs (see spa_vdev_add()), then update the config.
2544 * If this is a verbatim import, trust the current
2545 * in-core spa_config and update the disk labels.
2547 if (config_cache_txg
!= spa
->spa_config_txg
||
2548 state
== SPA_LOAD_IMPORT
||
2549 state
== SPA_LOAD_RECOVER
||
2550 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2551 need_update
= B_TRUE
;
2553 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2554 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2555 need_update
= B_TRUE
;
2558 * Update the config cache asychronously in case we're the
2559 * root pool, in which case the config cache isn't writable yet.
2562 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2565 * Check all DTLs to see if anything needs resilvering.
2567 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2568 vdev_resilver_needed(rvd
, NULL
, NULL
))
2569 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2572 * Delete any inconsistent datasets.
2574 (void) dmu_objset_find(spa_name(spa
),
2575 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2578 * Clean up any stale temporary dataset userrefs.
2580 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2587 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2589 int mode
= spa
->spa_mode
;
2592 spa_deactivate(spa
);
2594 spa
->spa_load_max_txg
--;
2596 spa_activate(spa
, mode
);
2597 spa_async_suspend(spa
);
2599 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2603 * If spa_load() fails this function will try loading prior txg's. If
2604 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2605 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2606 * function will not rewind the pool and will return the same error as
2610 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2611 uint64_t max_request
, int rewind_flags
)
2613 nvlist_t
*loadinfo
= NULL
;
2614 nvlist_t
*config
= NULL
;
2615 int load_error
, rewind_error
;
2616 uint64_t safe_rewind_txg
;
2619 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2620 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2621 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2623 spa
->spa_load_max_txg
= max_request
;
2626 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2628 if (load_error
== 0)
2631 if (spa
->spa_root_vdev
!= NULL
)
2632 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2634 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2635 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2637 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2638 nvlist_free(config
);
2639 return (load_error
);
2642 if (state
== SPA_LOAD_RECOVER
) {
2643 /* Price of rolling back is discarding txgs, including log */
2644 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2647 * If we aren't rolling back save the load info from our first
2648 * import attempt so that we can restore it after attempting
2651 loadinfo
= spa
->spa_load_info
;
2652 spa
->spa_load_info
= fnvlist_alloc();
2655 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2656 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2657 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2658 TXG_INITIAL
: safe_rewind_txg
;
2661 * Continue as long as we're finding errors, we're still within
2662 * the acceptable rewind range, and we're still finding uberblocks
2664 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2665 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2666 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2667 spa
->spa_extreme_rewind
= B_TRUE
;
2668 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2671 spa
->spa_extreme_rewind
= B_FALSE
;
2672 spa
->spa_load_max_txg
= UINT64_MAX
;
2674 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2675 spa_config_set(spa
, config
);
2677 if (state
== SPA_LOAD_RECOVER
) {
2678 ASSERT3P(loadinfo
, ==, NULL
);
2679 return (rewind_error
);
2681 /* Store the rewind info as part of the initial load info */
2682 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
2683 spa
->spa_load_info
);
2685 /* Restore the initial load info */
2686 fnvlist_free(spa
->spa_load_info
);
2687 spa
->spa_load_info
= loadinfo
;
2689 return (load_error
);
2696 * The import case is identical to an open except that the configuration is sent
2697 * down from userland, instead of grabbed from the configuration cache. For the
2698 * case of an open, the pool configuration will exist in the
2699 * POOL_STATE_UNINITIALIZED state.
2701 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2702 * the same time open the pool, without having to keep around the spa_t in some
2706 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2710 spa_load_state_t state
= SPA_LOAD_OPEN
;
2712 int locked
= B_FALSE
;
2717 * As disgusting as this is, we need to support recursive calls to this
2718 * function because dsl_dir_open() is called during spa_load(), and ends
2719 * up calling spa_open() again. The real fix is to figure out how to
2720 * avoid dsl_dir_open() calling this in the first place.
2722 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2723 mutex_enter(&spa_namespace_lock
);
2727 if ((spa
= spa_lookup(pool
)) == NULL
) {
2729 mutex_exit(&spa_namespace_lock
);
2733 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
2734 zpool_rewind_policy_t policy
;
2736 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
2738 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
2739 state
= SPA_LOAD_RECOVER
;
2741 spa_activate(spa
, spa_mode_global
);
2743 if (state
!= SPA_LOAD_RECOVER
)
2744 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
2746 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
2747 policy
.zrp_request
);
2749 if (error
== EBADF
) {
2751 * If vdev_validate() returns failure (indicated by
2752 * EBADF), it indicates that one of the vdevs indicates
2753 * that the pool has been exported or destroyed. If
2754 * this is the case, the config cache is out of sync and
2755 * we should remove the pool from the namespace.
2758 spa_deactivate(spa
);
2759 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
2762 mutex_exit(&spa_namespace_lock
);
2768 * We can't open the pool, but we still have useful
2769 * information: the state of each vdev after the
2770 * attempted vdev_open(). Return this to the user.
2772 if (config
!= NULL
&& spa
->spa_config
) {
2773 VERIFY(nvlist_dup(spa
->spa_config
, config
,
2775 VERIFY(nvlist_add_nvlist(*config
,
2776 ZPOOL_CONFIG_LOAD_INFO
,
2777 spa
->spa_load_info
) == 0);
2780 spa_deactivate(spa
);
2781 spa
->spa_last_open_failed
= error
;
2783 mutex_exit(&spa_namespace_lock
);
2789 spa_open_ref(spa
, tag
);
2792 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2795 * If we've recovered the pool, pass back any information we
2796 * gathered while doing the load.
2798 if (state
== SPA_LOAD_RECOVER
) {
2799 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
2800 spa
->spa_load_info
) == 0);
2804 spa
->spa_last_open_failed
= 0;
2805 spa
->spa_last_ubsync_txg
= 0;
2806 spa
->spa_load_txg
= 0;
2807 mutex_exit(&spa_namespace_lock
);
2816 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
2819 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
2823 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
2825 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
2829 * Lookup the given spa_t, incrementing the inject count in the process,
2830 * preventing it from being exported or destroyed.
2833 spa_inject_addref(char *name
)
2837 mutex_enter(&spa_namespace_lock
);
2838 if ((spa
= spa_lookup(name
)) == NULL
) {
2839 mutex_exit(&spa_namespace_lock
);
2842 spa
->spa_inject_ref
++;
2843 mutex_exit(&spa_namespace_lock
);
2849 spa_inject_delref(spa_t
*spa
)
2851 mutex_enter(&spa_namespace_lock
);
2852 spa
->spa_inject_ref
--;
2853 mutex_exit(&spa_namespace_lock
);
2857 * Add spares device information to the nvlist.
2860 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
2870 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2872 if (spa
->spa_spares
.sav_count
== 0)
2875 VERIFY(nvlist_lookup_nvlist(config
,
2876 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
2877 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
2878 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
2880 VERIFY(nvlist_add_nvlist_array(nvroot
,
2881 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
2882 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
2883 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
2886 * Go through and find any spares which have since been
2887 * repurposed as an active spare. If this is the case, update
2888 * their status appropriately.
2890 for (i
= 0; i
< nspares
; i
++) {
2891 VERIFY(nvlist_lookup_uint64(spares
[i
],
2892 ZPOOL_CONFIG_GUID
, &guid
) == 0);
2893 if (spa_spare_exists(guid
, &pool
, NULL
) &&
2895 VERIFY(nvlist_lookup_uint64_array(
2896 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
2897 (uint64_t **)&vs
, &vsc
) == 0);
2898 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
2899 vs
->vs_aux
= VDEV_AUX_SPARED
;
2906 * Add l2cache device information to the nvlist, including vdev stats.
2909 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
2912 uint_t i
, j
, nl2cache
;
2919 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2921 if (spa
->spa_l2cache
.sav_count
== 0)
2924 VERIFY(nvlist_lookup_nvlist(config
,
2925 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
2926 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
2927 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
2928 if (nl2cache
!= 0) {
2929 VERIFY(nvlist_add_nvlist_array(nvroot
,
2930 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
2931 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
2932 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
2935 * Update level 2 cache device stats.
2938 for (i
= 0; i
< nl2cache
; i
++) {
2939 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
2940 ZPOOL_CONFIG_GUID
, &guid
) == 0);
2943 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
2945 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
2946 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
2952 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
2953 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
2955 vdev_get_stats(vd
, vs
);
2961 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
2967 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2968 VERIFY(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2970 if (spa
->spa_feat_for_read_obj
!= 0) {
2971 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
2972 spa
->spa_feat_for_read_obj
);
2973 zap_cursor_retrieve(&zc
, &za
) == 0;
2974 zap_cursor_advance(&zc
)) {
2975 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
2976 za
.za_num_integers
== 1);
2977 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
2978 za
.za_first_integer
));
2980 zap_cursor_fini(&zc
);
2983 if (spa
->spa_feat_for_write_obj
!= 0) {
2984 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
2985 spa
->spa_feat_for_write_obj
);
2986 zap_cursor_retrieve(&zc
, &za
) == 0;
2987 zap_cursor_advance(&zc
)) {
2988 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
2989 za
.za_num_integers
== 1);
2990 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
2991 za
.za_first_integer
));
2993 zap_cursor_fini(&zc
);
2996 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
2998 nvlist_free(features
);
3002 spa_get_stats(const char *name
, nvlist_t
**config
,
3003 char *altroot
, size_t buflen
)
3009 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3013 * This still leaves a window of inconsistency where the spares
3014 * or l2cache devices could change and the config would be
3015 * self-inconsistent.
3017 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3019 if (*config
!= NULL
) {
3020 uint64_t loadtimes
[2];
3022 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3023 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3024 VERIFY(nvlist_add_uint64_array(*config
,
3025 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3027 VERIFY(nvlist_add_uint64(*config
,
3028 ZPOOL_CONFIG_ERRCOUNT
,
3029 spa_get_errlog_size(spa
)) == 0);
3031 if (spa_suspended(spa
))
3032 VERIFY(nvlist_add_uint64(*config
,
3033 ZPOOL_CONFIG_SUSPENDED
,
3034 spa
->spa_failmode
) == 0);
3036 spa_add_spares(spa
, *config
);
3037 spa_add_l2cache(spa
, *config
);
3038 spa_add_feature_stats(spa
, *config
);
3043 * We want to get the alternate root even for faulted pools, so we cheat
3044 * and call spa_lookup() directly.
3048 mutex_enter(&spa_namespace_lock
);
3049 spa
= spa_lookup(name
);
3051 spa_altroot(spa
, altroot
, buflen
);
3055 mutex_exit(&spa_namespace_lock
);
3057 spa_altroot(spa
, altroot
, buflen
);
3062 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3063 spa_close(spa
, FTAG
);
3070 * Validate that the auxiliary device array is well formed. We must have an
3071 * array of nvlists, each which describes a valid leaf vdev. If this is an
3072 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3073 * specified, as long as they are well-formed.
3076 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3077 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3078 vdev_labeltype_t label
)
3085 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3088 * It's acceptable to have no devs specified.
3090 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3097 * Make sure the pool is formatted with a version that supports this
3100 if (spa_version(spa
) < version
)
3104 * Set the pending device list so we correctly handle device in-use
3107 sav
->sav_pending
= dev
;
3108 sav
->sav_npending
= ndev
;
3110 for (i
= 0; i
< ndev
; i
++) {
3111 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3115 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3122 * The L2ARC currently only supports disk devices in
3123 * kernel context. For user-level testing, we allow it.
3126 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3127 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3135 if ((error
= vdev_open(vd
)) == 0 &&
3136 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3137 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3138 vd
->vdev_guid
) == 0);
3144 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3151 sav
->sav_pending
= NULL
;
3152 sav
->sav_npending
= 0;
3157 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3161 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3163 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3164 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3165 VDEV_LABEL_SPARE
)) != 0) {
3169 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3170 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3171 VDEV_LABEL_L2CACHE
));
3175 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3180 if (sav
->sav_config
!= NULL
) {
3186 * Generate new dev list by concatentating with the
3189 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3190 &olddevs
, &oldndevs
) == 0);
3192 newdevs
= kmem_alloc(sizeof (void *) *
3193 (ndevs
+ oldndevs
), KM_PUSHPAGE
);
3194 for (i
= 0; i
< oldndevs
; i
++)
3195 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3197 for (i
= 0; i
< ndevs
; i
++)
3198 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3201 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3202 DATA_TYPE_NVLIST_ARRAY
) == 0);
3204 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3205 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3206 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3207 nvlist_free(newdevs
[i
]);
3208 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3211 * Generate a new dev list.
3213 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3215 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3221 * Stop and drop level 2 ARC devices
3224 spa_l2cache_drop(spa_t
*spa
)
3228 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3230 for (i
= 0; i
< sav
->sav_count
; i
++) {
3233 vd
= sav
->sav_vdevs
[i
];
3236 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3237 pool
!= 0ULL && l2arc_vdev_present(vd
))
3238 l2arc_remove_vdev(vd
);
3246 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3247 const char *history_str
, nvlist_t
*zplprops
)
3250 char *altroot
= NULL
;
3255 uint64_t txg
= TXG_INITIAL
;
3256 nvlist_t
**spares
, **l2cache
;
3257 uint_t nspares
, nl2cache
;
3258 uint64_t version
, obj
;
3259 boolean_t has_features
;
3264 * If this pool already exists, return failure.
3266 mutex_enter(&spa_namespace_lock
);
3267 if (spa_lookup(pool
) != NULL
) {
3268 mutex_exit(&spa_namespace_lock
);
3273 * Allocate a new spa_t structure.
3275 (void) nvlist_lookup_string(props
,
3276 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3277 spa
= spa_add(pool
, NULL
, altroot
);
3278 spa_activate(spa
, spa_mode_global
);
3280 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3281 spa_deactivate(spa
);
3283 mutex_exit(&spa_namespace_lock
);
3287 has_features
= B_FALSE
;
3288 for (elem
= nvlist_next_nvpair(props
, NULL
);
3289 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3290 if (zpool_prop_feature(nvpair_name(elem
)))
3291 has_features
= B_TRUE
;
3294 if (has_features
|| nvlist_lookup_uint64(props
,
3295 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3296 version
= SPA_VERSION
;
3298 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3300 spa
->spa_first_txg
= txg
;
3301 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3302 spa
->spa_uberblock
.ub_version
= version
;
3303 spa
->spa_ubsync
= spa
->spa_uberblock
;
3306 * Create "The Godfather" zio to hold all async IOs
3308 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
3309 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
3312 * Create the root vdev.
3314 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3316 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3318 ASSERT(error
!= 0 || rvd
!= NULL
);
3319 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3321 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3325 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3326 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3327 VDEV_ALLOC_ADD
)) == 0) {
3328 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
3329 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3330 vdev_expand(rvd
->vdev_child
[c
], txg
);
3334 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3338 spa_deactivate(spa
);
3340 mutex_exit(&spa_namespace_lock
);
3345 * Get the list of spares, if specified.
3347 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3348 &spares
, &nspares
) == 0) {
3349 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3351 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3352 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3353 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3354 spa_load_spares(spa
);
3355 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3356 spa
->spa_spares
.sav_sync
= B_TRUE
;
3360 * Get the list of level 2 cache devices, if specified.
3362 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3363 &l2cache
, &nl2cache
) == 0) {
3364 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3365 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3366 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3367 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3368 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3369 spa_load_l2cache(spa
);
3370 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3371 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3374 spa
->spa_is_initializing
= B_TRUE
;
3375 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3376 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3377 spa
->spa_is_initializing
= B_FALSE
;
3380 * Create DDTs (dedup tables).
3384 spa_update_dspace(spa
);
3386 tx
= dmu_tx_create_assigned(dp
, txg
);
3389 * Create the pool config object.
3391 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3392 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3393 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3395 if (zap_add(spa
->spa_meta_objset
,
3396 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3397 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3398 cmn_err(CE_PANIC
, "failed to add pool config");
3401 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3402 spa_feature_create_zap_objects(spa
, tx
);
3404 if (zap_add(spa
->spa_meta_objset
,
3405 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3406 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3407 cmn_err(CE_PANIC
, "failed to add pool version");
3410 /* Newly created pools with the right version are always deflated. */
3411 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3412 spa
->spa_deflate
= TRUE
;
3413 if (zap_add(spa
->spa_meta_objset
,
3414 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3415 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3416 cmn_err(CE_PANIC
, "failed to add deflate");
3421 * Create the deferred-free bpobj. Turn off compression
3422 * because sync-to-convergence takes longer if the blocksize
3425 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3426 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3427 ZIO_COMPRESS_OFF
, tx
);
3428 if (zap_add(spa
->spa_meta_objset
,
3429 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3430 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3431 cmn_err(CE_PANIC
, "failed to add bpobj");
3433 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3434 spa
->spa_meta_objset
, obj
));
3437 * Create the pool's history object.
3439 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3440 spa_history_create_obj(spa
, tx
);
3443 * Set pool properties.
3445 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3446 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3447 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3448 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3450 if (props
!= NULL
) {
3451 spa_configfile_set(spa
, props
, B_FALSE
);
3452 spa_sync_props(spa
, props
, tx
);
3457 spa
->spa_sync_on
= B_TRUE
;
3458 txg_sync_start(spa
->spa_dsl_pool
);
3461 * We explicitly wait for the first transaction to complete so that our
3462 * bean counters are appropriately updated.
3464 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3466 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3468 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& history_str
!= NULL
)
3469 (void) spa_history_log(spa
, history_str
, LOG_CMD_POOL_CREATE
);
3470 spa_history_log_version(spa
, LOG_POOL_CREATE
);
3472 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3474 mutex_exit(&spa_namespace_lock
);
3481 * Get the root pool information from the root disk, then import the root pool
3482 * during the system boot up time.
3484 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3487 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3490 nvlist_t
*nvtop
, *nvroot
;
3493 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3497 * Add this top-level vdev to the child array.
3499 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3501 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3503 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3506 * Put this pool's top-level vdevs into a root vdev.
3508 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3509 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3510 VDEV_TYPE_ROOT
) == 0);
3511 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3512 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3513 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3517 * Replace the existing vdev_tree with the new root vdev in
3518 * this pool's configuration (remove the old, add the new).
3520 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3521 nvlist_free(nvroot
);
3526 * Walk the vdev tree and see if we can find a device with "better"
3527 * configuration. A configuration is "better" if the label on that
3528 * device has a more recent txg.
3531 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3535 for (c
= 0; c
< vd
->vdev_children
; c
++)
3536 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3538 if (vd
->vdev_ops
->vdev_op_leaf
) {
3542 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3546 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3550 * Do we have a better boot device?
3552 if (label_txg
> *txg
) {
3561 * Import a root pool.
3563 * For x86. devpath_list will consist of devid and/or physpath name of
3564 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3565 * The GRUB "findroot" command will return the vdev we should boot.
3567 * For Sparc, devpath_list consists the physpath name of the booting device
3568 * no matter the rootpool is a single device pool or a mirrored pool.
3570 * "/pci@1f,0/ide@d/disk@0,0:a"
3573 spa_import_rootpool(char *devpath
, char *devid
)
3576 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3577 nvlist_t
*config
, *nvtop
;
3583 * Read the label from the boot device and generate a configuration.
3585 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3586 #if defined(_OBP) && defined(_KERNEL)
3587 if (config
== NULL
) {
3588 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3590 get_iscsi_bootpath_phy(devpath
);
3591 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3595 if (config
== NULL
) {
3596 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
3601 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3603 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3605 mutex_enter(&spa_namespace_lock
);
3606 if ((spa
= spa_lookup(pname
)) != NULL
) {
3608 * Remove the existing root pool from the namespace so that we
3609 * can replace it with the correct config we just read in.
3614 spa
= spa_add(pname
, config
, NULL
);
3615 spa
->spa_is_root
= B_TRUE
;
3616 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3619 * Build up a vdev tree based on the boot device's label config.
3621 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3623 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3624 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3625 VDEV_ALLOC_ROOTPOOL
);
3626 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3628 mutex_exit(&spa_namespace_lock
);
3629 nvlist_free(config
);
3630 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3636 * Get the boot vdev.
3638 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3639 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3640 (u_longlong_t
)guid
);
3646 * Determine if there is a better boot device.
3649 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3651 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3652 "try booting from '%s'", avd
->vdev_path
);
3658 * If the boot device is part of a spare vdev then ensure that
3659 * we're booting off the active spare.
3661 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3662 !bvd
->vdev_isspare
) {
3663 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
3664 "try booting from '%s'",
3666 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
3672 spa_history_log_version(spa
, LOG_POOL_IMPORT
);
3674 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3676 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3677 mutex_exit(&spa_namespace_lock
);
3679 nvlist_free(config
);
3686 * Import a non-root pool into the system.
3689 spa_import(const char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
3692 char *altroot
= NULL
;
3693 spa_load_state_t state
= SPA_LOAD_IMPORT
;
3694 zpool_rewind_policy_t policy
;
3695 uint64_t mode
= spa_mode_global
;
3696 uint64_t readonly
= B_FALSE
;
3699 nvlist_t
**spares
, **l2cache
;
3700 uint_t nspares
, nl2cache
;
3703 * If a pool with this name exists, return failure.
3705 mutex_enter(&spa_namespace_lock
);
3706 if (spa_lookup(pool
) != NULL
) {
3707 mutex_exit(&spa_namespace_lock
);
3712 * Create and initialize the spa structure.
3714 (void) nvlist_lookup_string(props
,
3715 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3716 (void) nvlist_lookup_uint64(props
,
3717 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
3720 spa
= spa_add(pool
, config
, altroot
);
3721 spa
->spa_import_flags
= flags
;
3724 * Verbatim import - Take a pool and insert it into the namespace
3725 * as if it had been loaded at boot.
3727 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
3729 spa_configfile_set(spa
, props
, B_FALSE
);
3731 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3733 mutex_exit(&spa_namespace_lock
);
3734 spa_history_log_version(spa
, LOG_POOL_IMPORT
);
3739 spa_activate(spa
, mode
);
3742 * Don't start async tasks until we know everything is healthy.
3744 spa_async_suspend(spa
);
3746 zpool_get_rewind_policy(config
, &policy
);
3747 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3748 state
= SPA_LOAD_RECOVER
;
3751 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3752 * because the user-supplied config is actually the one to trust when
3755 if (state
!= SPA_LOAD_RECOVER
)
3756 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3758 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
3759 policy
.zrp_request
);
3762 * Propagate anything learned while loading the pool and pass it
3763 * back to caller (i.e. rewind info, missing devices, etc).
3765 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
3766 spa
->spa_load_info
) == 0);
3768 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3770 * Toss any existing sparelist, as it doesn't have any validity
3771 * anymore, and conflicts with spa_has_spare().
3773 if (spa
->spa_spares
.sav_config
) {
3774 nvlist_free(spa
->spa_spares
.sav_config
);
3775 spa
->spa_spares
.sav_config
= NULL
;
3776 spa_load_spares(spa
);
3778 if (spa
->spa_l2cache
.sav_config
) {
3779 nvlist_free(spa
->spa_l2cache
.sav_config
);
3780 spa
->spa_l2cache
.sav_config
= NULL
;
3781 spa_load_l2cache(spa
);
3784 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3787 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3790 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3791 VDEV_ALLOC_L2CACHE
);
3792 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3795 spa_configfile_set(spa
, props
, B_FALSE
);
3797 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
3798 (error
= spa_prop_set(spa
, props
)))) {
3800 spa_deactivate(spa
);
3802 mutex_exit(&spa_namespace_lock
);
3806 spa_async_resume(spa
);
3809 * Override any spares and level 2 cache devices as specified by
3810 * the user, as these may have correct device names/devids, etc.
3812 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3813 &spares
, &nspares
) == 0) {
3814 if (spa
->spa_spares
.sav_config
)
3815 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
3816 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3818 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
3819 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3820 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3821 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3822 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3823 spa_load_spares(spa
);
3824 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3825 spa
->spa_spares
.sav_sync
= B_TRUE
;
3827 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3828 &l2cache
, &nl2cache
) == 0) {
3829 if (spa
->spa_l2cache
.sav_config
)
3830 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
3831 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3833 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3834 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
3835 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3836 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3837 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3838 spa_load_l2cache(spa
);
3839 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3840 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3844 * Check for any removed devices.
3846 if (spa
->spa_autoreplace
) {
3847 spa_aux_check_removed(&spa
->spa_spares
);
3848 spa_aux_check_removed(&spa
->spa_l2cache
);
3851 if (spa_writeable(spa
)) {
3853 * Update the config cache to include the newly-imported pool.
3855 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
3859 * It's possible that the pool was expanded while it was exported.
3860 * We kick off an async task to handle this for us.
3862 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
3864 mutex_exit(&spa_namespace_lock
);
3865 spa_history_log_version(spa
, LOG_POOL_IMPORT
);
3871 spa_tryimport(nvlist_t
*tryconfig
)
3873 nvlist_t
*config
= NULL
;
3879 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
3882 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
3886 * Create and initialize the spa structure.
3888 mutex_enter(&spa_namespace_lock
);
3889 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
3890 spa_activate(spa
, FREAD
);
3893 * Pass off the heavy lifting to spa_load().
3894 * Pass TRUE for mosconfig because the user-supplied config
3895 * is actually the one to trust when doing an import.
3897 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
3900 * If 'tryconfig' was at least parsable, return the current config.
3902 if (spa
->spa_root_vdev
!= NULL
) {
3903 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3904 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3906 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
3908 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
3909 spa
->spa_uberblock
.ub_timestamp
) == 0);
3910 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
3911 spa
->spa_load_info
) == 0);
3914 * If the bootfs property exists on this pool then we
3915 * copy it out so that external consumers can tell which
3916 * pools are bootable.
3918 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
3919 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
3922 * We have to play games with the name since the
3923 * pool was opened as TRYIMPORT_NAME.
3925 if (dsl_dsobj_to_dsname(spa_name(spa
),
3926 spa
->spa_bootfs
, tmpname
) == 0) {
3928 char *dsname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
3930 cp
= strchr(tmpname
, '/');
3932 (void) strlcpy(dsname
, tmpname
,
3935 (void) snprintf(dsname
, MAXPATHLEN
,
3936 "%s/%s", poolname
, ++cp
);
3938 VERIFY(nvlist_add_string(config
,
3939 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
3940 kmem_free(dsname
, MAXPATHLEN
);
3942 kmem_free(tmpname
, MAXPATHLEN
);
3946 * Add the list of hot spares and level 2 cache devices.
3948 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3949 spa_add_spares(spa
, config
);
3950 spa_add_l2cache(spa
, config
);
3951 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3955 spa_deactivate(spa
);
3957 mutex_exit(&spa_namespace_lock
);
3963 * Pool export/destroy
3965 * The act of destroying or exporting a pool is very simple. We make sure there
3966 * is no more pending I/O and any references to the pool are gone. Then, we
3967 * update the pool state and sync all the labels to disk, removing the
3968 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3969 * we don't sync the labels or remove the configuration cache.
3972 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
3973 boolean_t force
, boolean_t hardforce
)
3980 if (!(spa_mode_global
& FWRITE
))
3983 mutex_enter(&spa_namespace_lock
);
3984 if ((spa
= spa_lookup(pool
)) == NULL
) {
3985 mutex_exit(&spa_namespace_lock
);
3990 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3991 * reacquire the namespace lock, and see if we can export.
3993 spa_open_ref(spa
, FTAG
);
3994 mutex_exit(&spa_namespace_lock
);
3995 spa_async_suspend(spa
);
3996 mutex_enter(&spa_namespace_lock
);
3997 spa_close(spa
, FTAG
);
4000 * The pool will be in core if it's openable,
4001 * in which case we can modify its state.
4003 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
&& spa
->spa_sync_on
) {
4005 * Objsets may be open only because they're dirty, so we
4006 * have to force it to sync before checking spa_refcnt.
4008 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4011 * A pool cannot be exported or destroyed if there are active
4012 * references. If we are resetting a pool, allow references by
4013 * fault injection handlers.
4015 if (!spa_refcount_zero(spa
) ||
4016 (spa
->spa_inject_ref
!= 0 &&
4017 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4018 spa_async_resume(spa
);
4019 mutex_exit(&spa_namespace_lock
);
4024 * A pool cannot be exported if it has an active shared spare.
4025 * This is to prevent other pools stealing the active spare
4026 * from an exported pool. At user's own will, such pool can
4027 * be forcedly exported.
4029 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4030 spa_has_active_shared_spare(spa
)) {
4031 spa_async_resume(spa
);
4032 mutex_exit(&spa_namespace_lock
);
4037 * We want this to be reflected on every label,
4038 * so mark them all dirty. spa_unload() will do the
4039 * final sync that pushes these changes out.
4041 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4042 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4043 spa
->spa_state
= new_state
;
4044 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4046 vdev_config_dirty(spa
->spa_root_vdev
);
4047 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4051 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_DESTROY
);
4053 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4055 spa_deactivate(spa
);
4058 if (oldconfig
&& spa
->spa_config
)
4059 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4061 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4063 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4066 mutex_exit(&spa_namespace_lock
);
4072 * Destroy a storage pool.
4075 spa_destroy(char *pool
)
4077 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4082 * Export a storage pool.
4085 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4086 boolean_t hardforce
)
4088 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4093 * Similar to spa_export(), this unloads the spa_t without actually removing it
4094 * from the namespace in any way.
4097 spa_reset(char *pool
)
4099 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4104 * ==========================================================================
4105 * Device manipulation
4106 * ==========================================================================
4110 * Add a device to a storage pool.
4113 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4117 vdev_t
*rvd
= spa
->spa_root_vdev
;
4119 nvlist_t
**spares
, **l2cache
;
4120 uint_t nspares
, nl2cache
;
4123 ASSERT(spa_writeable(spa
));
4125 txg
= spa_vdev_enter(spa
);
4127 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4128 VDEV_ALLOC_ADD
)) != 0)
4129 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4131 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4133 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4137 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4141 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4142 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4144 if (vd
->vdev_children
!= 0 &&
4145 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4146 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4149 * We must validate the spares and l2cache devices after checking the
4150 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4152 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4153 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4156 * Transfer each new top-level vdev from vd to rvd.
4158 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4161 * Set the vdev id to the first hole, if one exists.
4163 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4164 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4165 vdev_free(rvd
->vdev_child
[id
]);
4169 tvd
= vd
->vdev_child
[c
];
4170 vdev_remove_child(vd
, tvd
);
4172 vdev_add_child(rvd
, tvd
);
4173 vdev_config_dirty(tvd
);
4177 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4178 ZPOOL_CONFIG_SPARES
);
4179 spa_load_spares(spa
);
4180 spa
->spa_spares
.sav_sync
= B_TRUE
;
4183 if (nl2cache
!= 0) {
4184 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4185 ZPOOL_CONFIG_L2CACHE
);
4186 spa_load_l2cache(spa
);
4187 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4191 * We have to be careful when adding new vdevs to an existing pool.
4192 * If other threads start allocating from these vdevs before we
4193 * sync the config cache, and we lose power, then upon reboot we may
4194 * fail to open the pool because there are DVAs that the config cache
4195 * can't translate. Therefore, we first add the vdevs without
4196 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4197 * and then let spa_config_update() initialize the new metaslabs.
4199 * spa_load() checks for added-but-not-initialized vdevs, so that
4200 * if we lose power at any point in this sequence, the remaining
4201 * steps will be completed the next time we load the pool.
4203 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4205 mutex_enter(&spa_namespace_lock
);
4206 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4207 mutex_exit(&spa_namespace_lock
);
4213 * Attach a device to a mirror. The arguments are the path to any device
4214 * in the mirror, and the nvroot for the new device. If the path specifies
4215 * a device that is not mirrored, we automatically insert the mirror vdev.
4217 * If 'replacing' is specified, the new device is intended to replace the
4218 * existing device; in this case the two devices are made into their own
4219 * mirror using the 'replacing' vdev, which is functionally identical to
4220 * the mirror vdev (it actually reuses all the same ops) but has a few
4221 * extra rules: you can't attach to it after it's been created, and upon
4222 * completion of resilvering, the first disk (the one being replaced)
4223 * is automatically detached.
4226 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4228 uint64_t txg
, dtl_max_txg
;
4229 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
;)
4230 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4232 char *oldvdpath
, *newvdpath
;
4236 ASSERT(spa_writeable(spa
));
4238 txg
= spa_vdev_enter(spa
);
4240 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4243 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4245 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4246 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4248 pvd
= oldvd
->vdev_parent
;
4250 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4251 VDEV_ALLOC_ATTACH
)) != 0)
4252 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4254 if (newrootvd
->vdev_children
!= 1)
4255 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4257 newvd
= newrootvd
->vdev_child
[0];
4259 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4260 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4262 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4263 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4266 * Spares can't replace logs
4268 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4269 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4273 * For attach, the only allowable parent is a mirror or the root
4276 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4277 pvd
->vdev_ops
!= &vdev_root_ops
)
4278 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4280 pvops
= &vdev_mirror_ops
;
4283 * Active hot spares can only be replaced by inactive hot
4286 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4287 oldvd
->vdev_isspare
&&
4288 !spa_has_spare(spa
, newvd
->vdev_guid
))
4289 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4292 * If the source is a hot spare, and the parent isn't already a
4293 * spare, then we want to create a new hot spare. Otherwise, we
4294 * want to create a replacing vdev. The user is not allowed to
4295 * attach to a spared vdev child unless the 'isspare' state is
4296 * the same (spare replaces spare, non-spare replaces
4299 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4300 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4301 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4302 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4303 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4304 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4307 if (newvd
->vdev_isspare
)
4308 pvops
= &vdev_spare_ops
;
4310 pvops
= &vdev_replacing_ops
;
4314 * Make sure the new device is big enough.
4316 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4317 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4320 * The new device cannot have a higher alignment requirement
4321 * than the top-level vdev.
4323 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4324 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4327 * If this is an in-place replacement, update oldvd's path and devid
4328 * to make it distinguishable from newvd, and unopenable from now on.
4330 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4331 spa_strfree(oldvd
->vdev_path
);
4332 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4334 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4335 newvd
->vdev_path
, "old");
4336 if (oldvd
->vdev_devid
!= NULL
) {
4337 spa_strfree(oldvd
->vdev_devid
);
4338 oldvd
->vdev_devid
= NULL
;
4342 /* mark the device being resilvered */
4343 newvd
->vdev_resilvering
= B_TRUE
;
4346 * If the parent is not a mirror, or if we're replacing, insert the new
4347 * mirror/replacing/spare vdev above oldvd.
4349 if (pvd
->vdev_ops
!= pvops
)
4350 pvd
= vdev_add_parent(oldvd
, pvops
);
4352 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4353 ASSERT(pvd
->vdev_ops
== pvops
);
4354 ASSERT(oldvd
->vdev_parent
== pvd
);
4357 * Extract the new device from its root and add it to pvd.
4359 vdev_remove_child(newrootvd
, newvd
);
4360 newvd
->vdev_id
= pvd
->vdev_children
;
4361 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4362 vdev_add_child(pvd
, newvd
);
4364 tvd
= newvd
->vdev_top
;
4365 ASSERT(pvd
->vdev_top
== tvd
);
4366 ASSERT(tvd
->vdev_parent
== rvd
);
4368 vdev_config_dirty(tvd
);
4371 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4372 * for any dmu_sync-ed blocks. It will propagate upward when
4373 * spa_vdev_exit() calls vdev_dtl_reassess().
4375 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4377 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4378 dtl_max_txg
- TXG_INITIAL
);
4380 if (newvd
->vdev_isspare
) {
4381 spa_spare_activate(newvd
);
4382 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_DEVICE_SPARE
);
4385 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4386 newvdpath
= spa_strdup(newvd
->vdev_path
);
4387 newvd_isspare
= newvd
->vdev_isspare
;
4390 * Mark newvd's DTL dirty in this txg.
4392 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4395 * Restart the resilver
4397 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4402 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4404 spa_history_log_internal(LOG_POOL_VDEV_ATTACH
, spa
, NULL
,
4405 "%s vdev=%s %s vdev=%s",
4406 replacing
&& newvd_isspare
? "spare in" :
4407 replacing
? "replace" : "attach", newvdpath
,
4408 replacing
? "for" : "to", oldvdpath
);
4410 spa_strfree(oldvdpath
);
4411 spa_strfree(newvdpath
);
4413 if (spa
->spa_bootfs
)
4414 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH
);
4420 * Detach a device from a mirror or replacing vdev.
4421 * If 'replace_done' is specified, only detach if the parent
4422 * is a replacing vdev.
4425 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4429 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
;)
4430 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4431 boolean_t unspare
= B_FALSE
;
4432 uint64_t unspare_guid
= 0;
4436 ASSERT(spa_writeable(spa
));
4438 txg
= spa_vdev_enter(spa
);
4440 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4443 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4445 if (!vd
->vdev_ops
->vdev_op_leaf
)
4446 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4448 pvd
= vd
->vdev_parent
;
4451 * If the parent/child relationship is not as expected, don't do it.
4452 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4453 * vdev that's replacing B with C. The user's intent in replacing
4454 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4455 * the replace by detaching C, the expected behavior is to end up
4456 * M(A,B). But suppose that right after deciding to detach C,
4457 * the replacement of B completes. We would have M(A,C), and then
4458 * ask to detach C, which would leave us with just A -- not what
4459 * the user wanted. To prevent this, we make sure that the
4460 * parent/child relationship hasn't changed -- in this example,
4461 * that C's parent is still the replacing vdev R.
4463 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4464 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4467 * Only 'replacing' or 'spare' vdevs can be replaced.
4469 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4470 pvd
->vdev_ops
!= &vdev_spare_ops
)
4471 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4473 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4474 spa_version(spa
) >= SPA_VERSION_SPARES
);
4477 * Only mirror, replacing, and spare vdevs support detach.
4479 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4480 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4481 pvd
->vdev_ops
!= &vdev_spare_ops
)
4482 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4485 * If this device has the only valid copy of some data,
4486 * we cannot safely detach it.
4488 if (vdev_dtl_required(vd
))
4489 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4491 ASSERT(pvd
->vdev_children
>= 2);
4494 * If we are detaching the second disk from a replacing vdev, then
4495 * check to see if we changed the original vdev's path to have "/old"
4496 * at the end in spa_vdev_attach(). If so, undo that change now.
4498 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4499 vd
->vdev_path
!= NULL
) {
4500 size_t len
= strlen(vd
->vdev_path
);
4502 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4503 cvd
= pvd
->vdev_child
[c
];
4505 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4508 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4509 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4510 spa_strfree(cvd
->vdev_path
);
4511 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4518 * If we are detaching the original disk from a spare, then it implies
4519 * that the spare should become a real disk, and be removed from the
4520 * active spare list for the pool.
4522 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4524 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4528 * Erase the disk labels so the disk can be used for other things.
4529 * This must be done after all other error cases are handled,
4530 * but before we disembowel vd (so we can still do I/O to it).
4531 * But if we can't do it, don't treat the error as fatal --
4532 * it may be that the unwritability of the disk is the reason
4533 * it's being detached!
4535 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4538 * Remove vd from its parent and compact the parent's children.
4540 vdev_remove_child(pvd
, vd
);
4541 vdev_compact_children(pvd
);
4544 * Remember one of the remaining children so we can get tvd below.
4546 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4549 * If we need to remove the remaining child from the list of hot spares,
4550 * do it now, marking the vdev as no longer a spare in the process.
4551 * We must do this before vdev_remove_parent(), because that can
4552 * change the GUID if it creates a new toplevel GUID. For a similar
4553 * reason, we must remove the spare now, in the same txg as the detach;
4554 * otherwise someone could attach a new sibling, change the GUID, and
4555 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4558 ASSERT(cvd
->vdev_isspare
);
4559 spa_spare_remove(cvd
);
4560 unspare_guid
= cvd
->vdev_guid
;
4561 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4562 cvd
->vdev_unspare
= B_TRUE
;
4566 * If the parent mirror/replacing vdev only has one child,
4567 * the parent is no longer needed. Remove it from the tree.
4569 if (pvd
->vdev_children
== 1) {
4570 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4571 cvd
->vdev_unspare
= B_FALSE
;
4572 vdev_remove_parent(cvd
);
4573 cvd
->vdev_resilvering
= B_FALSE
;
4578 * We don't set tvd until now because the parent we just removed
4579 * may have been the previous top-level vdev.
4581 tvd
= cvd
->vdev_top
;
4582 ASSERT(tvd
->vdev_parent
== rvd
);
4585 * Reevaluate the parent vdev state.
4587 vdev_propagate_state(cvd
);
4590 * If the 'autoexpand' property is set on the pool then automatically
4591 * try to expand the size of the pool. For example if the device we
4592 * just detached was smaller than the others, it may be possible to
4593 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4594 * first so that we can obtain the updated sizes of the leaf vdevs.
4596 if (spa
->spa_autoexpand
) {
4598 vdev_expand(tvd
, txg
);
4601 vdev_config_dirty(tvd
);
4604 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4605 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4606 * But first make sure we're not on any *other* txg's DTL list, to
4607 * prevent vd from being accessed after it's freed.
4609 vdpath
= spa_strdup(vd
->vdev_path
);
4610 for (t
= 0; t
< TXG_SIZE
; t
++)
4611 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4612 vd
->vdev_detached
= B_TRUE
;
4613 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4615 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_REMOVE
);
4617 /* hang on to the spa before we release the lock */
4618 spa_open_ref(spa
, FTAG
);
4620 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4622 spa_history_log_internal(LOG_POOL_VDEV_DETACH
, spa
, NULL
,
4624 spa_strfree(vdpath
);
4627 * If this was the removal of the original device in a hot spare vdev,
4628 * then we want to go through and remove the device from the hot spare
4629 * list of every other pool.
4632 spa_t
*altspa
= NULL
;
4634 mutex_enter(&spa_namespace_lock
);
4635 while ((altspa
= spa_next(altspa
)) != NULL
) {
4636 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4640 spa_open_ref(altspa
, FTAG
);
4641 mutex_exit(&spa_namespace_lock
);
4642 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4643 mutex_enter(&spa_namespace_lock
);
4644 spa_close(altspa
, FTAG
);
4646 mutex_exit(&spa_namespace_lock
);
4648 /* search the rest of the vdevs for spares to remove */
4649 spa_vdev_resilver_done(spa
);
4652 /* all done with the spa; OK to release */
4653 mutex_enter(&spa_namespace_lock
);
4654 spa_close(spa
, FTAG
);
4655 mutex_exit(&spa_namespace_lock
);
4661 * Split a set of devices from their mirrors, and create a new pool from them.
4664 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
4665 nvlist_t
*props
, boolean_t exp
)
4668 uint64_t txg
, *glist
;
4670 uint_t c
, children
, lastlog
;
4671 nvlist_t
**child
, *nvl
, *tmp
;
4673 char *altroot
= NULL
;
4674 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
4675 boolean_t activate_slog
;
4677 ASSERT(spa_writeable(spa
));
4679 txg
= spa_vdev_enter(spa
);
4681 /* clear the log and flush everything up to now */
4682 activate_slog
= spa_passivate_log(spa
);
4683 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4684 error
= spa_offline_log(spa
);
4685 txg
= spa_vdev_config_enter(spa
);
4688 spa_activate_log(spa
);
4691 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4693 /* check new spa name before going any further */
4694 if (spa_lookup(newname
) != NULL
)
4695 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
4698 * scan through all the children to ensure they're all mirrors
4700 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
4701 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
4703 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4705 /* first, check to ensure we've got the right child count */
4706 rvd
= spa
->spa_root_vdev
;
4708 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
4709 vdev_t
*vd
= rvd
->vdev_child
[c
];
4711 /* don't count the holes & logs as children */
4712 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
4720 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
4721 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4723 /* next, ensure no spare or cache devices are part of the split */
4724 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
4725 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
4726 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4728 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_PUSHPAGE
);
4729 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_PUSHPAGE
);
4731 /* then, loop over each vdev and validate it */
4732 for (c
= 0; c
< children
; c
++) {
4733 uint64_t is_hole
= 0;
4735 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
4739 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
4740 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
4748 /* which disk is going to be split? */
4749 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
4755 /* look it up in the spa */
4756 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
4757 if (vml
[c
] == NULL
) {
4762 /* make sure there's nothing stopping the split */
4763 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
4764 vml
[c
]->vdev_islog
||
4765 vml
[c
]->vdev_ishole
||
4766 vml
[c
]->vdev_isspare
||
4767 vml
[c
]->vdev_isl2cache
||
4768 !vdev_writeable(vml
[c
]) ||
4769 vml
[c
]->vdev_children
!= 0 ||
4770 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
4771 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
4776 if (vdev_dtl_required(vml
[c
])) {
4781 /* we need certain info from the top level */
4782 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
4783 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
4784 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
4785 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
4786 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
4787 vml
[c
]->vdev_top
->vdev_asize
) == 0);
4788 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
4789 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
4793 kmem_free(vml
, children
* sizeof (vdev_t
*));
4794 kmem_free(glist
, children
* sizeof (uint64_t));
4795 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4798 /* stop writers from using the disks */
4799 for (c
= 0; c
< children
; c
++) {
4801 vml
[c
]->vdev_offline
= B_TRUE
;
4803 vdev_reopen(spa
->spa_root_vdev
);
4806 * Temporarily record the splitting vdevs in the spa config. This
4807 * will disappear once the config is regenerated.
4809 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
4810 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
4811 glist
, children
) == 0);
4812 kmem_free(glist
, children
* sizeof (uint64_t));
4814 mutex_enter(&spa
->spa_props_lock
);
4815 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
4817 mutex_exit(&spa
->spa_props_lock
);
4818 spa
->spa_config_splitting
= nvl
;
4819 vdev_config_dirty(spa
->spa_root_vdev
);
4821 /* configure and create the new pool */
4822 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
4823 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4824 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
4825 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
4826 spa_version(spa
)) == 0);
4827 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
4828 spa
->spa_config_txg
) == 0);
4829 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
4830 spa_generate_guid(NULL
)) == 0);
4831 (void) nvlist_lookup_string(props
,
4832 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4834 /* add the new pool to the namespace */
4835 newspa
= spa_add(newname
, config
, altroot
);
4836 newspa
->spa_config_txg
= spa
->spa_config_txg
;
4837 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
4839 /* release the spa config lock, retaining the namespace lock */
4840 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4842 if (zio_injection_enabled
)
4843 zio_handle_panic_injection(spa
, FTAG
, 1);
4845 spa_activate(newspa
, spa_mode_global
);
4846 spa_async_suspend(newspa
);
4848 /* create the new pool from the disks of the original pool */
4849 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
4853 /* if that worked, generate a real config for the new pool */
4854 if (newspa
->spa_root_vdev
!= NULL
) {
4855 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
4856 NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
4857 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
4858 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
4859 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
4864 if (props
!= NULL
) {
4865 spa_configfile_set(newspa
, props
, B_FALSE
);
4866 error
= spa_prop_set(newspa
, props
);
4871 /* flush everything */
4872 txg
= spa_vdev_config_enter(newspa
);
4873 vdev_config_dirty(newspa
->spa_root_vdev
);
4874 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
4876 if (zio_injection_enabled
)
4877 zio_handle_panic_injection(spa
, FTAG
, 2);
4879 spa_async_resume(newspa
);
4881 /* finally, update the original pool's config */
4882 txg
= spa_vdev_config_enter(spa
);
4883 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
4884 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4887 for (c
= 0; c
< children
; c
++) {
4888 if (vml
[c
] != NULL
) {
4891 spa_history_log_internal(LOG_POOL_VDEV_DETACH
,
4897 vdev_config_dirty(spa
->spa_root_vdev
);
4898 spa
->spa_config_splitting
= NULL
;
4902 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
4904 if (zio_injection_enabled
)
4905 zio_handle_panic_injection(spa
, FTAG
, 3);
4907 /* split is complete; log a history record */
4908 spa_history_log_internal(LOG_POOL_SPLIT
, newspa
, NULL
,
4909 "split new pool %s from pool %s", newname
, spa_name(spa
));
4911 kmem_free(vml
, children
* sizeof (vdev_t
*));
4913 /* if we're not going to mount the filesystems in userland, export */
4915 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
4922 spa_deactivate(newspa
);
4925 txg
= spa_vdev_config_enter(spa
);
4927 /* re-online all offlined disks */
4928 for (c
= 0; c
< children
; c
++) {
4930 vml
[c
]->vdev_offline
= B_FALSE
;
4932 vdev_reopen(spa
->spa_root_vdev
);
4934 nvlist_free(spa
->spa_config_splitting
);
4935 spa
->spa_config_splitting
= NULL
;
4936 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
4938 kmem_free(vml
, children
* sizeof (vdev_t
*));
4943 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
4947 for (i
= 0; i
< count
; i
++) {
4950 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
4953 if (guid
== target_guid
)
4961 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
4962 nvlist_t
*dev_to_remove
)
4964 nvlist_t
**newdev
= NULL
;
4968 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_PUSHPAGE
);
4970 for (i
= 0, j
= 0; i
< count
; i
++) {
4971 if (dev
[i
] == dev_to_remove
)
4973 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_PUSHPAGE
) == 0);
4976 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4977 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
4979 for (i
= 0; i
< count
- 1; i
++)
4980 nvlist_free(newdev
[i
]);
4983 kmem_free(newdev
, (count
- 1) * sizeof (void *));
4987 * Evacuate the device.
4990 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
4995 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4996 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
4997 ASSERT(vd
== vd
->vdev_top
);
5000 * Evacuate the device. We don't hold the config lock as writer
5001 * since we need to do I/O but we do keep the
5002 * spa_namespace_lock held. Once this completes the device
5003 * should no longer have any blocks allocated on it.
5005 if (vd
->vdev_islog
) {
5006 if (vd
->vdev_stat
.vs_alloc
!= 0)
5007 error
= spa_offline_log(spa
);
5016 * The evacuation succeeded. Remove any remaining MOS metadata
5017 * associated with this vdev, and wait for these changes to sync.
5019 ASSERT3U(vd
->vdev_stat
.vs_alloc
, ==, 0);
5020 txg
= spa_vdev_config_enter(spa
);
5021 vd
->vdev_removing
= B_TRUE
;
5022 vdev_dirty(vd
, 0, NULL
, txg
);
5023 vdev_config_dirty(vd
);
5024 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5030 * Complete the removal by cleaning up the namespace.
5033 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5035 vdev_t
*rvd
= spa
->spa_root_vdev
;
5036 uint64_t id
= vd
->vdev_id
;
5037 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5039 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5040 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5041 ASSERT(vd
== vd
->vdev_top
);
5044 * Only remove any devices which are empty.
5046 if (vd
->vdev_stat
.vs_alloc
!= 0)
5049 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5051 if (list_link_active(&vd
->vdev_state_dirty_node
))
5052 vdev_state_clean(vd
);
5053 if (list_link_active(&vd
->vdev_config_dirty_node
))
5054 vdev_config_clean(vd
);
5059 vdev_compact_children(rvd
);
5061 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5062 vdev_add_child(rvd
, vd
);
5064 vdev_config_dirty(rvd
);
5067 * Reassess the health of our root vdev.
5073 * Remove a device from the pool -
5075 * Removing a device from the vdev namespace requires several steps
5076 * and can take a significant amount of time. As a result we use
5077 * the spa_vdev_config_[enter/exit] functions which allow us to
5078 * grab and release the spa_config_lock while still holding the namespace
5079 * lock. During each step the configuration is synced out.
5083 * Remove a device from the pool. Currently, this supports removing only hot
5084 * spares, slogs, and level 2 ARC devices.
5087 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5090 metaslab_group_t
*mg
;
5091 nvlist_t
**spares
, **l2cache
, *nv
;
5093 uint_t nspares
, nl2cache
;
5095 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5097 ASSERT(spa_writeable(spa
));
5100 txg
= spa_vdev_enter(spa
);
5102 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5104 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5105 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5106 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5107 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5109 * Only remove the hot spare if it's not currently in use
5112 if (vd
== NULL
|| unspare
) {
5113 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5114 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5115 spa_load_spares(spa
);
5116 spa
->spa_spares
.sav_sync
= B_TRUE
;
5120 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5121 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5122 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5123 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5125 * Cache devices can always be removed.
5127 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5128 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5129 spa_load_l2cache(spa
);
5130 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5131 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5133 ASSERT(vd
== vd
->vdev_top
);
5136 * XXX - Once we have bp-rewrite this should
5137 * become the common case.
5143 * Stop allocating from this vdev.
5145 metaslab_group_passivate(mg
);
5148 * Wait for the youngest allocations and frees to sync,
5149 * and then wait for the deferral of those frees to finish.
5151 spa_vdev_config_exit(spa
, NULL
,
5152 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5155 * Attempt to evacuate the vdev.
5157 error
= spa_vdev_remove_evacuate(spa
, vd
);
5159 txg
= spa_vdev_config_enter(spa
);
5162 * If we couldn't evacuate the vdev, unwind.
5165 metaslab_group_activate(mg
);
5166 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5170 * Clean up the vdev namespace.
5172 spa_vdev_remove_from_namespace(spa
, vd
);
5174 } else if (vd
!= NULL
) {
5176 * Normal vdevs cannot be removed (yet).
5181 * There is no vdev of any kind with the specified guid.
5187 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5193 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5194 * current spared, so we can detach it.
5197 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5199 vdev_t
*newvd
, *oldvd
;
5202 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5203 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5209 * Check for a completed replacement. We always consider the first
5210 * vdev in the list to be the oldest vdev, and the last one to be
5211 * the newest (see spa_vdev_attach() for how that works). In
5212 * the case where the newest vdev is faulted, we will not automatically
5213 * remove it after a resilver completes. This is OK as it will require
5214 * user intervention to determine which disk the admin wishes to keep.
5216 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5217 ASSERT(vd
->vdev_children
> 1);
5219 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5220 oldvd
= vd
->vdev_child
[0];
5222 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5223 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5224 !vdev_dtl_required(oldvd
))
5229 * Check for a completed resilver with the 'unspare' flag set.
5231 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5232 vdev_t
*first
= vd
->vdev_child
[0];
5233 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5235 if (last
->vdev_unspare
) {
5238 } else if (first
->vdev_unspare
) {
5245 if (oldvd
!= NULL
&&
5246 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5247 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5248 !vdev_dtl_required(oldvd
))
5252 * If there are more than two spares attached to a disk,
5253 * and those spares are not required, then we want to
5254 * attempt to free them up now so that they can be used
5255 * by other pools. Once we're back down to a single
5256 * disk+spare, we stop removing them.
5258 if (vd
->vdev_children
> 2) {
5259 newvd
= vd
->vdev_child
[1];
5261 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5262 vdev_dtl_empty(last
, DTL_MISSING
) &&
5263 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5264 !vdev_dtl_required(newvd
))
5273 spa_vdev_resilver_done(spa_t
*spa
)
5275 vdev_t
*vd
, *pvd
, *ppvd
;
5276 uint64_t guid
, sguid
, pguid
, ppguid
;
5278 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5280 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5281 pvd
= vd
->vdev_parent
;
5282 ppvd
= pvd
->vdev_parent
;
5283 guid
= vd
->vdev_guid
;
5284 pguid
= pvd
->vdev_guid
;
5285 ppguid
= ppvd
->vdev_guid
;
5288 * If we have just finished replacing a hot spared device, then
5289 * we need to detach the parent's first child (the original hot
5292 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5293 ppvd
->vdev_children
== 2) {
5294 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5295 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5297 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5298 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5300 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5302 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5305 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5309 * Update the stored path or FRU for this vdev.
5312 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5316 boolean_t sync
= B_FALSE
;
5318 ASSERT(spa_writeable(spa
));
5320 spa_vdev_state_enter(spa
, SCL_ALL
);
5322 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5323 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5325 if (!vd
->vdev_ops
->vdev_op_leaf
)
5326 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5329 if (strcmp(value
, vd
->vdev_path
) != 0) {
5330 spa_strfree(vd
->vdev_path
);
5331 vd
->vdev_path
= spa_strdup(value
);
5335 if (vd
->vdev_fru
== NULL
) {
5336 vd
->vdev_fru
= spa_strdup(value
);
5338 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5339 spa_strfree(vd
->vdev_fru
);
5340 vd
->vdev_fru
= spa_strdup(value
);
5345 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5349 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5351 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5355 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5357 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5361 * ==========================================================================
5363 * ==========================================================================
5367 spa_scan_stop(spa_t
*spa
)
5369 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5370 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5372 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5376 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5378 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5380 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5384 * If a resilver was requested, but there is no DTL on a
5385 * writeable leaf device, we have nothing to do.
5387 if (func
== POOL_SCAN_RESILVER
&&
5388 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5389 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5393 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5397 * ==========================================================================
5398 * SPA async task processing
5399 * ==========================================================================
5403 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5407 if (vd
->vdev_remove_wanted
) {
5408 vd
->vdev_remove_wanted
= B_FALSE
;
5409 vd
->vdev_delayed_close
= B_FALSE
;
5410 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5413 * We want to clear the stats, but we don't want to do a full
5414 * vdev_clear() as that will cause us to throw away
5415 * degraded/faulted state as well as attempt to reopen the
5416 * device, all of which is a waste.
5418 vd
->vdev_stat
.vs_read_errors
= 0;
5419 vd
->vdev_stat
.vs_write_errors
= 0;
5420 vd
->vdev_stat
.vs_checksum_errors
= 0;
5422 vdev_state_dirty(vd
->vdev_top
);
5425 for (c
= 0; c
< vd
->vdev_children
; c
++)
5426 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5430 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5434 if (vd
->vdev_probe_wanted
) {
5435 vd
->vdev_probe_wanted
= B_FALSE
;
5436 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5439 for (c
= 0; c
< vd
->vdev_children
; c
++)
5440 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5444 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5448 if (!spa
->spa_autoexpand
)
5451 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5452 vdev_t
*cvd
= vd
->vdev_child
[c
];
5453 spa_async_autoexpand(spa
, cvd
);
5456 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5459 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND
);
5463 spa_async_thread(spa_t
*spa
)
5467 ASSERT(spa
->spa_sync_on
);
5469 mutex_enter(&spa
->spa_async_lock
);
5470 tasks
= spa
->spa_async_tasks
;
5471 spa
->spa_async_tasks
= 0;
5472 mutex_exit(&spa
->spa_async_lock
);
5475 * See if the config needs to be updated.
5477 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5478 uint64_t old_space
, new_space
;
5480 mutex_enter(&spa_namespace_lock
);
5481 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5482 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5483 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5484 mutex_exit(&spa_namespace_lock
);
5487 * If the pool grew as a result of the config update,
5488 * then log an internal history event.
5490 if (new_space
!= old_space
) {
5491 spa_history_log_internal(LOG_POOL_VDEV_ONLINE
,
5493 "pool '%s' size: %llu(+%llu)",
5494 spa_name(spa
), new_space
, new_space
- old_space
);
5499 * See if any devices need to be marked REMOVED.
5501 if (tasks
& SPA_ASYNC_REMOVE
) {
5502 spa_vdev_state_enter(spa
, SCL_NONE
);
5503 spa_async_remove(spa
, spa
->spa_root_vdev
);
5504 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5505 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5506 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5507 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5508 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5511 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5512 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5513 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5514 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5518 * See if any devices need to be probed.
5520 if (tasks
& SPA_ASYNC_PROBE
) {
5521 spa_vdev_state_enter(spa
, SCL_NONE
);
5522 spa_async_probe(spa
, spa
->spa_root_vdev
);
5523 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5527 * If any devices are done replacing, detach them.
5529 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5530 spa_vdev_resilver_done(spa
);
5533 * Kick off a resilver.
5535 if (tasks
& SPA_ASYNC_RESILVER
)
5536 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5539 * Let the world know that we're done.
5541 mutex_enter(&spa
->spa_async_lock
);
5542 spa
->spa_async_thread
= NULL
;
5543 cv_broadcast(&spa
->spa_async_cv
);
5544 mutex_exit(&spa
->spa_async_lock
);
5549 spa_async_suspend(spa_t
*spa
)
5551 mutex_enter(&spa
->spa_async_lock
);
5552 spa
->spa_async_suspended
++;
5553 while (spa
->spa_async_thread
!= NULL
)
5554 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5555 mutex_exit(&spa
->spa_async_lock
);
5559 spa_async_resume(spa_t
*spa
)
5561 mutex_enter(&spa
->spa_async_lock
);
5562 ASSERT(spa
->spa_async_suspended
!= 0);
5563 spa
->spa_async_suspended
--;
5564 mutex_exit(&spa
->spa_async_lock
);
5568 spa_async_dispatch(spa_t
*spa
)
5570 mutex_enter(&spa
->spa_async_lock
);
5571 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5572 spa
->spa_async_thread
== NULL
&&
5573 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5574 spa
->spa_async_thread
= thread_create(NULL
, 0,
5575 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5576 mutex_exit(&spa
->spa_async_lock
);
5580 spa_async_request(spa_t
*spa
, int task
)
5582 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5583 mutex_enter(&spa
->spa_async_lock
);
5584 spa
->spa_async_tasks
|= task
;
5585 mutex_exit(&spa
->spa_async_lock
);
5589 * ==========================================================================
5590 * SPA syncing routines
5591 * ==========================================================================
5595 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5598 bpobj_enqueue(bpo
, bp
, tx
);
5603 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5607 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5613 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5615 char *packed
= NULL
;
5620 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5623 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5624 * information. This avoids the dbuf_will_dirty() path and
5625 * saves us a pre-read to get data we don't actually care about.
5627 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
5628 packed
= vmem_alloc(bufsize
, KM_PUSHPAGE
);
5630 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
5632 bzero(packed
+ nvsize
, bufsize
- nvsize
);
5634 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
5636 vmem_free(packed
, bufsize
);
5638 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
5639 dmu_buf_will_dirty(db
, tx
);
5640 *(uint64_t *)db
->db_data
= nvsize
;
5641 dmu_buf_rele(db
, FTAG
);
5645 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
5646 const char *config
, const char *entry
)
5656 * Update the MOS nvlist describing the list of available devices.
5657 * spa_validate_aux() will have already made sure this nvlist is
5658 * valid and the vdevs are labeled appropriately.
5660 if (sav
->sav_object
== 0) {
5661 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
5662 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
5663 sizeof (uint64_t), tx
);
5664 VERIFY(zap_update(spa
->spa_meta_objset
,
5665 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
5666 &sav
->sav_object
, tx
) == 0);
5669 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_PUSHPAGE
) == 0);
5670 if (sav
->sav_count
== 0) {
5671 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
5673 list
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_PUSHPAGE
);
5674 for (i
= 0; i
< sav
->sav_count
; i
++)
5675 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
5676 B_FALSE
, VDEV_CONFIG_L2CACHE
);
5677 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
5678 sav
->sav_count
) == 0);
5679 for (i
= 0; i
< sav
->sav_count
; i
++)
5680 nvlist_free(list
[i
]);
5681 kmem_free(list
, sav
->sav_count
* sizeof (void *));
5684 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
5685 nvlist_free(nvroot
);
5687 sav
->sav_sync
= B_FALSE
;
5691 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
5695 if (list_is_empty(&spa
->spa_config_dirty_list
))
5698 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5700 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
5701 dmu_tx_get_txg(tx
), B_FALSE
);
5703 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5705 if (spa
->spa_config_syncing
)
5706 nvlist_free(spa
->spa_config_syncing
);
5707 spa
->spa_config_syncing
= config
;
5709 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
5713 spa_sync_version(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
5716 uint64_t version
= *(uint64_t *)arg2
;
5719 * Setting the version is special cased when first creating the pool.
5721 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
5723 ASSERT(version
<= SPA_VERSION
);
5724 ASSERT(version
>= spa_version(spa
));
5726 spa
->spa_uberblock
.ub_version
= version
;
5727 vdev_config_dirty(spa
->spa_root_vdev
);
5731 * Set zpool properties.
5734 spa_sync_props(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
5737 objset_t
*mos
= spa
->spa_meta_objset
;
5738 nvlist_t
*nvp
= arg2
;
5739 nvpair_t
*elem
= NULL
;
5741 mutex_enter(&spa
->spa_props_lock
);
5743 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
5745 char *strval
, *fname
;
5747 const char *propname
;
5748 zprop_type_t proptype
;
5749 zfeature_info_t
*feature
;
5751 prop
= zpool_name_to_prop(nvpair_name(elem
));
5752 switch ((int)prop
) {
5755 * We checked this earlier in spa_prop_validate().
5757 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
5759 fname
= strchr(nvpair_name(elem
), '@') + 1;
5760 VERIFY3U(0, ==, zfeature_lookup_name(fname
, &feature
));
5762 spa_feature_enable(spa
, feature
, tx
);
5765 case ZPOOL_PROP_VERSION
:
5766 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
5768 * The version is synced seperatly before other
5769 * properties and should be correct by now.
5771 ASSERT3U(spa_version(spa
), >=, intval
);
5774 case ZPOOL_PROP_ALTROOT
:
5776 * 'altroot' is a non-persistent property. It should
5777 * have been set temporarily at creation or import time.
5779 ASSERT(spa
->spa_root
!= NULL
);
5782 case ZPOOL_PROP_READONLY
:
5783 case ZPOOL_PROP_CACHEFILE
:
5785 * 'readonly' and 'cachefile' are also non-persisitent
5789 case ZPOOL_PROP_COMMENT
:
5790 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
5791 if (spa
->spa_comment
!= NULL
)
5792 spa_strfree(spa
->spa_comment
);
5793 spa
->spa_comment
= spa_strdup(strval
);
5795 * We need to dirty the configuration on all the vdevs
5796 * so that their labels get updated. It's unnecessary
5797 * to do this for pool creation since the vdev's
5798 * configuratoin has already been dirtied.
5800 if (tx
->tx_txg
!= TXG_INITIAL
)
5801 vdev_config_dirty(spa
->spa_root_vdev
);
5805 * Set pool property values in the poolprops mos object.
5807 if (spa
->spa_pool_props_object
== 0) {
5808 spa
->spa_pool_props_object
=
5809 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
5810 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
5814 /* normalize the property name */
5815 propname
= zpool_prop_to_name(prop
);
5816 proptype
= zpool_prop_get_type(prop
);
5818 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
5819 ASSERT(proptype
== PROP_TYPE_STRING
);
5820 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
5821 VERIFY(zap_update(mos
,
5822 spa
->spa_pool_props_object
, propname
,
5823 1, strlen(strval
) + 1, strval
, tx
) == 0);
5825 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
5826 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
5828 if (proptype
== PROP_TYPE_INDEX
) {
5830 VERIFY(zpool_prop_index_to_string(
5831 prop
, intval
, &unused
) == 0);
5833 VERIFY(zap_update(mos
,
5834 spa
->spa_pool_props_object
, propname
,
5835 8, 1, &intval
, tx
) == 0);
5837 ASSERT(0); /* not allowed */
5841 case ZPOOL_PROP_DELEGATION
:
5842 spa
->spa_delegation
= intval
;
5844 case ZPOOL_PROP_BOOTFS
:
5845 spa
->spa_bootfs
= intval
;
5847 case ZPOOL_PROP_FAILUREMODE
:
5848 spa
->spa_failmode
= intval
;
5850 case ZPOOL_PROP_AUTOEXPAND
:
5851 spa
->spa_autoexpand
= intval
;
5852 if (tx
->tx_txg
!= TXG_INITIAL
)
5853 spa_async_request(spa
,
5854 SPA_ASYNC_AUTOEXPAND
);
5856 case ZPOOL_PROP_DEDUPDITTO
:
5857 spa
->spa_dedup_ditto
= intval
;
5864 /* log internal history if this is not a zpool create */
5865 if (spa_version(spa
) >= SPA_VERSION_ZPOOL_HISTORY
&&
5866 tx
->tx_txg
!= TXG_INITIAL
) {
5867 spa_history_log_internal(LOG_POOL_PROPSET
,
5868 spa
, tx
, "%s %lld %s",
5869 nvpair_name(elem
), intval
, spa_name(spa
));
5873 mutex_exit(&spa
->spa_props_lock
);
5877 * Perform one-time upgrade on-disk changes. spa_version() does not
5878 * reflect the new version this txg, so there must be no changes this
5879 * txg to anything that the upgrade code depends on after it executes.
5880 * Therefore this must be called after dsl_pool_sync() does the sync
5884 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
5886 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
5888 ASSERT(spa
->spa_sync_pass
== 1);
5890 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
5891 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
5892 dsl_pool_create_origin(dp
, tx
);
5894 /* Keeping the origin open increases spa_minref */
5895 spa
->spa_minref
+= 3;
5898 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
5899 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
5900 dsl_pool_upgrade_clones(dp
, tx
);
5903 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
5904 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
5905 dsl_pool_upgrade_dir_clones(dp
, tx
);
5907 /* Keeping the freedir open increases spa_minref */
5908 spa
->spa_minref
+= 3;
5911 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
5912 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
5913 spa_feature_create_zap_objects(spa
, tx
);
5918 * Sync the specified transaction group. New blocks may be dirtied as
5919 * part of the process, so we iterate until it converges.
5922 spa_sync(spa_t
*spa
, uint64_t txg
)
5924 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
5925 objset_t
*mos
= spa
->spa_meta_objset
;
5926 bpobj_t
*defer_bpo
= &spa
->spa_deferred_bpobj
;
5927 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
5928 vdev_t
*rvd
= spa
->spa_root_vdev
;
5934 VERIFY(spa_writeable(spa
));
5937 * Lock out configuration changes.
5939 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5941 spa
->spa_syncing_txg
= txg
;
5942 spa
->spa_sync_pass
= 0;
5945 * If there are any pending vdev state changes, convert them
5946 * into config changes that go out with this transaction group.
5948 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5949 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
5951 * We need the write lock here because, for aux vdevs,
5952 * calling vdev_config_dirty() modifies sav_config.
5953 * This is ugly and will become unnecessary when we
5954 * eliminate the aux vdev wart by integrating all vdevs
5955 * into the root vdev tree.
5957 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
5958 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
5959 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
5960 vdev_state_clean(vd
);
5961 vdev_config_dirty(vd
);
5963 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
5964 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
5966 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5968 tx
= dmu_tx_create_assigned(dp
, txg
);
5971 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5972 * set spa_deflate if we have no raid-z vdevs.
5974 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
5975 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
5978 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
5979 vd
= rvd
->vdev_child
[i
];
5980 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
5983 if (i
== rvd
->vdev_children
) {
5984 spa
->spa_deflate
= TRUE
;
5985 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
5986 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
5987 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
5992 * If anything has changed in this txg, or if someone is waiting
5993 * for this txg to sync (eg, spa_vdev_remove()), push the
5994 * deferred frees from the previous txg. If not, leave them
5995 * alone so that we don't generate work on an otherwise idle
5998 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
5999 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
6000 !txg_list_empty(&dp
->dp_sync_tasks
, txg
) ||
6001 ((dsl_scan_active(dp
->dp_scan
) ||
6002 txg_sync_waiting(dp
)) && !spa_shutting_down(spa
))) {
6003 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6004 VERIFY3U(bpobj_iterate(defer_bpo
,
6005 spa_free_sync_cb
, zio
, tx
), ==, 0);
6006 VERIFY3U(zio_wait(zio
), ==, 0);
6010 * Iterate to convergence.
6013 int pass
= ++spa
->spa_sync_pass
;
6015 spa_sync_config_object(spa
, tx
);
6016 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6017 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6018 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6019 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6020 spa_errlog_sync(spa
, txg
);
6021 dsl_pool_sync(dp
, txg
);
6023 if (pass
<= SYNC_PASS_DEFERRED_FREE
) {
6024 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6025 bplist_iterate(free_bpl
, spa_free_sync_cb
,
6027 VERIFY(zio_wait(zio
) == 0);
6029 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6034 dsl_scan_sync(dp
, tx
);
6036 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6040 spa_sync_upgrades(spa
, tx
);
6042 } while (dmu_objset_is_dirty(mos
, txg
));
6045 * Rewrite the vdev configuration (which includes the uberblock)
6046 * to commit the transaction group.
6048 * If there are no dirty vdevs, we sync the uberblock to a few
6049 * random top-level vdevs that are known to be visible in the
6050 * config cache (see spa_vdev_add() for a complete description).
6051 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6055 * We hold SCL_STATE to prevent vdev open/close/etc.
6056 * while we're attempting to write the vdev labels.
6058 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6060 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6061 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6063 int children
= rvd
->vdev_children
;
6064 int c0
= spa_get_random(children
);
6066 for (c
= 0; c
< children
; c
++) {
6067 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6068 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6070 svd
[svdcount
++] = vd
;
6071 if (svdcount
== SPA_DVAS_PER_BP
)
6074 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
6076 error
= vdev_config_sync(svd
, svdcount
, txg
,
6079 error
= vdev_config_sync(rvd
->vdev_child
,
6080 rvd
->vdev_children
, txg
, B_FALSE
);
6082 error
= vdev_config_sync(rvd
->vdev_child
,
6083 rvd
->vdev_children
, txg
, B_TRUE
);
6086 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6090 zio_suspend(spa
, NULL
);
6091 zio_resume_wait(spa
);
6096 * Clear the dirty config list.
6098 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6099 vdev_config_clean(vd
);
6102 * Now that the new config has synced transactionally,
6103 * let it become visible to the config cache.
6105 if (spa
->spa_config_syncing
!= NULL
) {
6106 spa_config_set(spa
, spa
->spa_config_syncing
);
6107 spa
->spa_config_txg
= txg
;
6108 spa
->spa_config_syncing
= NULL
;
6111 spa
->spa_ubsync
= spa
->spa_uberblock
;
6113 dsl_pool_sync_done(dp
, txg
);
6116 * Update usable space statistics.
6118 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6119 vdev_sync_done(vd
, txg
);
6121 spa_update_dspace(spa
);
6124 * It had better be the case that we didn't dirty anything
6125 * since vdev_config_sync().
6127 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6128 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6129 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6131 spa
->spa_sync_pass
= 0;
6133 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6135 spa_handle_ignored_writes(spa
);
6138 * If any async tasks have been requested, kick them off.
6140 spa_async_dispatch(spa
);
6144 * Sync all pools. We don't want to hold the namespace lock across these
6145 * operations, so we take a reference on the spa_t and drop the lock during the
6149 spa_sync_allpools(void)
6152 mutex_enter(&spa_namespace_lock
);
6153 while ((spa
= spa_next(spa
)) != NULL
) {
6154 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6155 !spa_writeable(spa
) || spa_suspended(spa
))
6157 spa_open_ref(spa
, FTAG
);
6158 mutex_exit(&spa_namespace_lock
);
6159 txg_wait_synced(spa_get_dsl(spa
), 0);
6160 mutex_enter(&spa_namespace_lock
);
6161 spa_close(spa
, FTAG
);
6163 mutex_exit(&spa_namespace_lock
);
6167 * ==========================================================================
6168 * Miscellaneous routines
6169 * ==========================================================================
6173 * Remove all pools in the system.
6181 * Remove all cached state. All pools should be closed now,
6182 * so every spa in the AVL tree should be unreferenced.
6184 mutex_enter(&spa_namespace_lock
);
6185 while ((spa
= spa_next(NULL
)) != NULL
) {
6187 * Stop async tasks. The async thread may need to detach
6188 * a device that's been replaced, which requires grabbing
6189 * spa_namespace_lock, so we must drop it here.
6191 spa_open_ref(spa
, FTAG
);
6192 mutex_exit(&spa_namespace_lock
);
6193 spa_async_suspend(spa
);
6194 mutex_enter(&spa_namespace_lock
);
6195 spa_close(spa
, FTAG
);
6197 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6199 spa_deactivate(spa
);
6203 mutex_exit(&spa_namespace_lock
);
6207 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6212 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6216 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6217 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6218 if (vd
->vdev_guid
== guid
)
6222 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6223 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6224 if (vd
->vdev_guid
== guid
)
6233 spa_upgrade(spa_t
*spa
, uint64_t version
)
6235 ASSERT(spa_writeable(spa
));
6237 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6240 * This should only be called for a non-faulted pool, and since a
6241 * future version would result in an unopenable pool, this shouldn't be
6244 ASSERT(spa
->spa_uberblock
.ub_version
<= SPA_VERSION
);
6245 ASSERT(version
>= spa
->spa_uberblock
.ub_version
);
6247 spa
->spa_uberblock
.ub_version
= version
;
6248 vdev_config_dirty(spa
->spa_root_vdev
);
6250 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6252 txg_wait_synced(spa_get_dsl(spa
), 0);
6256 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6260 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6262 for (i
= 0; i
< sav
->sav_count
; i
++)
6263 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6266 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6267 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6268 &spareguid
) == 0 && spareguid
== guid
)
6276 * Check if a pool has an active shared spare device.
6277 * Note: reference count of an active spare is 2, as a spare and as a replace
6280 spa_has_active_shared_spare(spa_t
*spa
)
6284 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6286 for (i
= 0; i
< sav
->sav_count
; i
++) {
6287 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6288 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6297 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6298 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6299 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6300 * or zdb as real changes.
6303 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6306 zfs_ereport_post(name
, spa
, vd
, NULL
, 0, 0);
6310 #if defined(_KERNEL) && defined(HAVE_SPL)
6311 /* state manipulation functions */
6312 EXPORT_SYMBOL(spa_open
);
6313 EXPORT_SYMBOL(spa_open_rewind
);
6314 EXPORT_SYMBOL(spa_get_stats
);
6315 EXPORT_SYMBOL(spa_create
);
6316 EXPORT_SYMBOL(spa_import_rootpool
);
6317 EXPORT_SYMBOL(spa_import
);
6318 EXPORT_SYMBOL(spa_tryimport
);
6319 EXPORT_SYMBOL(spa_destroy
);
6320 EXPORT_SYMBOL(spa_export
);
6321 EXPORT_SYMBOL(spa_reset
);
6322 EXPORT_SYMBOL(spa_async_request
);
6323 EXPORT_SYMBOL(spa_async_suspend
);
6324 EXPORT_SYMBOL(spa_async_resume
);
6325 EXPORT_SYMBOL(spa_inject_addref
);
6326 EXPORT_SYMBOL(spa_inject_delref
);
6327 EXPORT_SYMBOL(spa_scan_stat_init
);
6328 EXPORT_SYMBOL(spa_scan_get_stats
);
6330 /* device maniion */
6331 EXPORT_SYMBOL(spa_vdev_add
);
6332 EXPORT_SYMBOL(spa_vdev_attach
);
6333 EXPORT_SYMBOL(spa_vdev_detach
);
6334 EXPORT_SYMBOL(spa_vdev_remove
);
6335 EXPORT_SYMBOL(spa_vdev_setpath
);
6336 EXPORT_SYMBOL(spa_vdev_setfru
);
6337 EXPORT_SYMBOL(spa_vdev_split_mirror
);
6339 /* spare statech is global across all pools) */
6340 EXPORT_SYMBOL(spa_spare_add
);
6341 EXPORT_SYMBOL(spa_spare_remove
);
6342 EXPORT_SYMBOL(spa_spare_exists
);
6343 EXPORT_SYMBOL(spa_spare_activate
);
6345 /* L2ARC statech is global across all pools) */
6346 EXPORT_SYMBOL(spa_l2cache_add
);
6347 EXPORT_SYMBOL(spa_l2cache_remove
);
6348 EXPORT_SYMBOL(spa_l2cache_exists
);
6349 EXPORT_SYMBOL(spa_l2cache_activate
);
6350 EXPORT_SYMBOL(spa_l2cache_drop
);
6353 EXPORT_SYMBOL(spa_scan
);
6354 EXPORT_SYMBOL(spa_scan_stop
);
6357 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
6358 EXPORT_SYMBOL(spa_sync_allpools
);
6361 EXPORT_SYMBOL(spa_prop_set
);
6362 EXPORT_SYMBOL(spa_prop_get
);
6363 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
6365 /* asynchronous event notification */
6366 EXPORT_SYMBOL(spa_event_notify
);