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/metaslab.h>
46 #include <sys/metaslab_impl.h>
47 #include <sys/uberblock_impl.h>
50 #include <sys/dmu_traverse.h>
51 #include <sys/dmu_objset.h>
52 #include <sys/unique.h>
53 #include <sys/dsl_pool.h>
54 #include <sys/dsl_dataset.h>
55 #include <sys/dsl_dir.h>
56 #include <sys/dsl_prop.h>
57 #include <sys/dsl_synctask.h>
58 #include <sys/fs/zfs.h>
60 #include <sys/callb.h>
61 #include <sys/systeminfo.h>
62 #include <sys/spa_boot.h>
63 #include <sys/zfs_ioctl.h>
64 #include <sys/dsl_scan.h>
65 #include <sys/zfeature.h>
68 #include <sys/bootprops.h>
69 #include <sys/callb.h>
70 #include <sys/cpupart.h>
72 #include <sys/sysdc.h>
77 #include "zfs_comutil.h"
79 typedef enum zti_modes
{
80 zti_mode_fixed
, /* value is # of threads (min 1) */
81 zti_mode_online_percent
, /* value is % of online CPUs */
82 zti_mode_batch
, /* cpu-intensive; value is ignored */
83 zti_mode_null
, /* don't create a taskq */
87 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
88 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
89 #define ZTI_BATCH { zti_mode_batch, 0 }
90 #define ZTI_NULL { zti_mode_null, 0 }
92 #define ZTI_ONE ZTI_FIX(1)
94 typedef struct zio_taskq_info
{
95 enum zti_modes zti_mode
;
99 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
100 "issue", "issue_high", "intr", "intr_high"
104 * Define the taskq threads for the following I/O types:
105 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
107 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
108 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
109 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
110 { ZTI_FIX(8), ZTI_NULL
, ZTI_BATCH
, ZTI_NULL
},
111 { ZTI_BATCH
, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
112 { ZTI_FIX(100), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
113 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
114 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
117 static dsl_syncfunc_t spa_sync_version
;
118 static dsl_syncfunc_t spa_sync_props
;
119 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
120 static int spa_load_impl(spa_t
*spa
, uint64_t, nvlist_t
*config
,
121 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
123 static void spa_vdev_resilver_done(spa_t
*spa
);
125 uint_t zio_taskq_batch_pct
= 100; /* 1 thread per cpu in pset */
126 id_t zio_taskq_psrset_bind
= PS_NONE
;
127 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
128 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
130 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
133 * This (illegal) pool name is used when temporarily importing a spa_t in order
134 * to get the vdev stats associated with the imported devices.
136 #define TRYIMPORT_NAME "$import"
139 * ==========================================================================
140 * SPA properties routines
141 * ==========================================================================
145 * Add a (source=src, propname=propval) list to an nvlist.
148 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
149 uint64_t intval
, zprop_source_t src
)
151 const char *propname
= zpool_prop_to_name(prop
);
154 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
155 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
158 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
160 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
162 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
163 nvlist_free(propval
);
167 * Get property values from the spa configuration.
170 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
172 vdev_t
*rvd
= spa
->spa_root_vdev
;
173 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
177 uint64_t cap
, version
;
178 zprop_source_t src
= ZPROP_SRC_NONE
;
179 spa_config_dirent_t
*dp
;
181 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
184 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
185 size
= metaslab_class_get_space(spa_normal_class(spa
));
186 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
187 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
188 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
189 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
193 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
194 vdev_t
*tvd
= rvd
->vdev_child
[c
];
195 space
+= tvd
->vdev_max_asize
- tvd
->vdev_asize
;
197 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
, space
,
200 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
201 (spa_mode(spa
) == FREAD
), src
);
203 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
204 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
206 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
207 ddt_get_pool_dedup_ratio(spa
), src
);
209 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
210 rvd
->vdev_state
, src
);
212 version
= spa_version(spa
);
213 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
))
214 src
= ZPROP_SRC_DEFAULT
;
216 src
= ZPROP_SRC_LOCAL
;
217 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
221 dsl_dir_t
*freedir
= pool
->dp_free_dir
;
224 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
225 * when opening pools before this version freedir will be NULL.
227 if (freedir
!= NULL
) {
228 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
229 freedir
->dd_phys
->dd_used_bytes
, src
);
231 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
236 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
238 if (spa
->spa_comment
!= NULL
) {
239 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
243 if (spa
->spa_root
!= NULL
)
244 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
247 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
248 if (dp
->scd_path
== NULL
) {
249 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
250 "none", 0, ZPROP_SRC_LOCAL
);
251 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
252 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
253 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
259 * Get zpool property values.
262 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
264 objset_t
*mos
= spa
->spa_meta_objset
;
269 VERIFY(nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
271 mutex_enter(&spa
->spa_props_lock
);
274 * Get properties from the spa config.
276 spa_prop_get_config(spa
, nvp
);
278 /* If no pool property object, no more prop to get. */
279 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
280 mutex_exit(&spa
->spa_props_lock
);
285 * Get properties from the MOS pool property object.
287 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
288 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
289 zap_cursor_advance(&zc
)) {
292 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
295 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
298 switch (za
.za_integer_length
) {
300 /* integer property */
301 if (za
.za_first_integer
!=
302 zpool_prop_default_numeric(prop
))
303 src
= ZPROP_SRC_LOCAL
;
305 if (prop
== ZPOOL_PROP_BOOTFS
) {
307 dsl_dataset_t
*ds
= NULL
;
309 dp
= spa_get_dsl(spa
);
310 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
311 if (err
= dsl_dataset_hold_obj(dp
,
312 za
.za_first_integer
, FTAG
, &ds
)) {
313 rw_exit(&dp
->dp_config_rwlock
);
318 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1,
320 dsl_dataset_name(ds
, strval
);
321 dsl_dataset_rele(ds
, FTAG
);
322 rw_exit(&dp
->dp_config_rwlock
);
325 intval
= za
.za_first_integer
;
328 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
332 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1);
337 /* string property */
338 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
339 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
340 za
.za_name
, 1, za
.za_num_integers
, strval
);
342 kmem_free(strval
, za
.za_num_integers
);
345 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
346 kmem_free(strval
, za
.za_num_integers
);
353 zap_cursor_fini(&zc
);
354 mutex_exit(&spa
->spa_props_lock
);
356 if (err
&& err
!= ENOENT
) {
366 * Validate the given pool properties nvlist and modify the list
367 * for the property values to be set.
370 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
373 int error
= 0, reset_bootfs
= 0;
375 boolean_t has_feature
= B_FALSE
;
378 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
380 char *strval
, *slash
, *check
, *fname
;
381 const char *propname
= nvpair_name(elem
);
382 zpool_prop_t prop
= zpool_name_to_prop(propname
);
386 if (!zpool_prop_feature(propname
)) {
392 * Sanitize the input.
394 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
399 if (nvpair_value_uint64(elem
, &intval
) != 0) {
409 fname
= strchr(propname
, '@') + 1;
410 if (zfeature_lookup_name(fname
, NULL
) != 0) {
415 has_feature
= B_TRUE
;
418 case ZPOOL_PROP_VERSION
:
419 error
= nvpair_value_uint64(elem
, &intval
);
421 (intval
< spa_version(spa
) ||
422 intval
> SPA_VERSION_BEFORE_FEATURES
||
427 case ZPOOL_PROP_DELEGATION
:
428 case ZPOOL_PROP_AUTOREPLACE
:
429 case ZPOOL_PROP_LISTSNAPS
:
430 case ZPOOL_PROP_AUTOEXPAND
:
431 error
= nvpair_value_uint64(elem
, &intval
);
432 if (!error
&& intval
> 1)
436 case ZPOOL_PROP_BOOTFS
:
438 * If the pool version is less than SPA_VERSION_BOOTFS,
439 * or the pool is still being created (version == 0),
440 * the bootfs property cannot be set.
442 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
448 * Make sure the vdev config is bootable
450 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
457 error
= nvpair_value_string(elem
, &strval
);
463 if (strval
== NULL
|| strval
[0] == '\0') {
464 objnum
= zpool_prop_default_numeric(
469 if (error
= dmu_objset_hold(strval
, FTAG
, &os
))
472 /* Must be ZPL and not gzip compressed. */
474 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
476 } else if ((error
= dsl_prop_get_integer(strval
,
477 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
478 &compress
, NULL
)) == 0 &&
479 !BOOTFS_COMPRESS_VALID(compress
)) {
482 objnum
= dmu_objset_id(os
);
484 dmu_objset_rele(os
, FTAG
);
488 case ZPOOL_PROP_FAILUREMODE
:
489 error
= nvpair_value_uint64(elem
, &intval
);
490 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
491 intval
> ZIO_FAILURE_MODE_PANIC
))
495 * This is a special case which only occurs when
496 * the pool has completely failed. This allows
497 * the user to change the in-core failmode property
498 * without syncing it out to disk (I/Os might
499 * currently be blocked). We do this by returning
500 * EIO to the caller (spa_prop_set) to trick it
501 * into thinking we encountered a property validation
504 if (!error
&& spa_suspended(spa
)) {
505 spa
->spa_failmode
= intval
;
510 case ZPOOL_PROP_CACHEFILE
:
511 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
514 if (strval
[0] == '\0')
517 if (strcmp(strval
, "none") == 0)
520 if (strval
[0] != '/') {
525 slash
= strrchr(strval
, '/');
526 ASSERT(slash
!= NULL
);
528 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
529 strcmp(slash
, "/..") == 0)
533 case ZPOOL_PROP_COMMENT
:
534 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
536 for (check
= strval
; *check
!= '\0'; check
++) {
538 * The kernel doesn't have an easy isprint()
539 * check. For this kernel check, we merely
540 * check ASCII apart from DEL. Fix this if
541 * there is an easy-to-use kernel isprint().
543 if (*check
>= 0x7f) {
549 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
553 case ZPOOL_PROP_DEDUPDITTO
:
554 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
557 error
= nvpair_value_uint64(elem
, &intval
);
559 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
568 if (!error
&& reset_bootfs
) {
569 error
= nvlist_remove(props
,
570 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
573 error
= nvlist_add_uint64(props
,
574 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
582 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
585 spa_config_dirent_t
*dp
;
587 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
591 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
594 if (cachefile
[0] == '\0')
595 dp
->scd_path
= spa_strdup(spa_config_path
);
596 else if (strcmp(cachefile
, "none") == 0)
599 dp
->scd_path
= spa_strdup(cachefile
);
601 list_insert_head(&spa
->spa_config_list
, dp
);
603 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
607 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
610 nvpair_t
*elem
= NULL
;
611 boolean_t need_sync
= B_FALSE
;
613 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
616 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
617 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
619 if (prop
== ZPOOL_PROP_CACHEFILE
||
620 prop
== ZPOOL_PROP_ALTROOT
||
621 prop
== ZPOOL_PROP_READONLY
)
624 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPROP_INVAL
) {
627 if (prop
== ZPOOL_PROP_VERSION
) {
628 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
630 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
631 ver
= SPA_VERSION_FEATURES
;
635 /* Save time if the version is already set. */
636 if (ver
== spa_version(spa
))
640 * In addition to the pool directory object, we might
641 * create the pool properties object, the features for
642 * read object, the features for write object, or the
643 * feature descriptions object.
645 error
= dsl_sync_task_do(spa_get_dsl(spa
), NULL
,
646 spa_sync_version
, spa
, &ver
, 6);
657 return (dsl_sync_task_do(spa_get_dsl(spa
), NULL
, spa_sync_props
,
665 * If the bootfs property value is dsobj, clear it.
668 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
670 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
671 VERIFY(zap_remove(spa
->spa_meta_objset
,
672 spa
->spa_pool_props_object
,
673 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
679 * Change the GUID for the pool. This is done so that we can later
680 * re-import a pool built from a clone of our own vdevs. We will modify
681 * the root vdev's guid, our own pool guid, and then mark all of our
682 * vdevs dirty. Note that we must make sure that all our vdevs are
683 * online when we do this, or else any vdevs that weren't present
684 * would be orphaned from our pool. We are also going to issue a
685 * sysevent to update any watchers.
688 spa_change_guid(spa_t
*spa
)
690 uint64_t oldguid
, newguid
;
693 if (!(spa_mode_global
& FWRITE
))
696 txg
= spa_vdev_enter(spa
);
698 if (spa
->spa_root_vdev
->vdev_state
!= VDEV_STATE_HEALTHY
)
699 return (spa_vdev_exit(spa
, NULL
, txg
, ENXIO
));
701 oldguid
= spa_guid(spa
);
702 newguid
= spa_generate_guid(NULL
);
703 ASSERT3U(oldguid
, !=, newguid
);
705 spa
->spa_root_vdev
->vdev_guid
= newguid
;
706 spa
->spa_root_vdev
->vdev_guid_sum
+= (newguid
- oldguid
);
708 vdev_config_dirty(spa
->spa_root_vdev
);
710 spa_event_notify(spa
, NULL
, ESC_ZFS_POOL_REGUID
);
712 return (spa_vdev_exit(spa
, NULL
, txg
, 0));
716 * ==========================================================================
717 * SPA state manipulation (open/create/destroy/import/export)
718 * ==========================================================================
722 spa_error_entry_compare(const void *a
, const void *b
)
724 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
725 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
728 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
729 sizeof (zbookmark_t
));
740 * Utility function which retrieves copies of the current logs and
741 * re-initializes them in the process.
744 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
746 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
748 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
749 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
751 avl_create(&spa
->spa_errlist_scrub
,
752 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
753 offsetof(spa_error_entry_t
, se_avl
));
754 avl_create(&spa
->spa_errlist_last
,
755 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
756 offsetof(spa_error_entry_t
, se_avl
));
760 spa_taskq_create(spa_t
*spa
, const char *name
, enum zti_modes mode
,
764 boolean_t batch
= B_FALSE
;
768 return (NULL
); /* no taskq needed */
771 ASSERT3U(value
, >=, 1);
772 value
= MAX(value
, 1);
777 flags
|= TASKQ_THREADS_CPU_PCT
;
778 value
= zio_taskq_batch_pct
;
781 case zti_mode_online_percent
:
782 flags
|= TASKQ_THREADS_CPU_PCT
;
786 panic("unrecognized mode for %s taskq (%u:%u) in "
792 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
794 flags
|= TASKQ_DC_BATCH
;
796 return (taskq_create_sysdc(name
, value
, 50, INT_MAX
,
797 spa
->spa_proc
, zio_taskq_basedc
, flags
));
799 return (taskq_create_proc(name
, value
, maxclsyspri
, 50, INT_MAX
,
800 spa
->spa_proc
, flags
));
804 spa_create_zio_taskqs(spa_t
*spa
)
806 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
807 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
808 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
809 enum zti_modes mode
= ztip
->zti_mode
;
810 uint_t value
= ztip
->zti_value
;
813 (void) snprintf(name
, sizeof (name
),
814 "%s_%s", zio_type_name
[t
], zio_taskq_types
[q
]);
816 spa
->spa_zio_taskq
[t
][q
] =
817 spa_taskq_create(spa
, name
, mode
, value
);
824 spa_thread(void *arg
)
829 user_t
*pu
= PTOU(curproc
);
831 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
834 ASSERT(curproc
!= &p0
);
835 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
836 "zpool-%s", spa
->spa_name
);
837 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
839 /* bind this thread to the requested psrset */
840 if (zio_taskq_psrset_bind
!= PS_NONE
) {
842 mutex_enter(&cpu_lock
);
843 mutex_enter(&pidlock
);
844 mutex_enter(&curproc
->p_lock
);
846 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
847 0, NULL
, NULL
) == 0) {
848 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
851 "Couldn't bind process for zfs pool \"%s\" to "
852 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
855 mutex_exit(&curproc
->p_lock
);
856 mutex_exit(&pidlock
);
857 mutex_exit(&cpu_lock
);
861 if (zio_taskq_sysdc
) {
862 sysdc_thread_enter(curthread
, 100, 0);
865 spa
->spa_proc
= curproc
;
866 spa
->spa_did
= curthread
->t_did
;
868 spa_create_zio_taskqs(spa
);
870 mutex_enter(&spa
->spa_proc_lock
);
871 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
873 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
874 cv_broadcast(&spa
->spa_proc_cv
);
876 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
877 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
878 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
879 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
881 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
882 spa
->spa_proc_state
= SPA_PROC_GONE
;
884 cv_broadcast(&spa
->spa_proc_cv
);
885 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
887 mutex_enter(&curproc
->p_lock
);
893 * Activate an uninitialized pool.
896 spa_activate(spa_t
*spa
, int mode
)
898 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
900 spa
->spa_state
= POOL_STATE_ACTIVE
;
901 spa
->spa_mode
= mode
;
903 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
904 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
906 /* Try to create a covering process */
907 mutex_enter(&spa
->spa_proc_lock
);
908 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
909 ASSERT(spa
->spa_proc
== &p0
);
912 /* Only create a process if we're going to be around a while. */
913 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
914 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
916 spa
->spa_proc_state
= SPA_PROC_CREATED
;
917 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
918 cv_wait(&spa
->spa_proc_cv
,
919 &spa
->spa_proc_lock
);
921 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
922 ASSERT(spa
->spa_proc
!= &p0
);
923 ASSERT(spa
->spa_did
!= 0);
927 "Couldn't create process for zfs pool \"%s\"\n",
932 mutex_exit(&spa
->spa_proc_lock
);
934 /* If we didn't create a process, we need to create our taskqs. */
935 if (spa
->spa_proc
== &p0
) {
936 spa_create_zio_taskqs(spa
);
939 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
940 offsetof(vdev_t
, vdev_config_dirty_node
));
941 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
942 offsetof(vdev_t
, vdev_state_dirty_node
));
944 txg_list_create(&spa
->spa_vdev_txg_list
,
945 offsetof(struct vdev
, vdev_txg_node
));
947 avl_create(&spa
->spa_errlist_scrub
,
948 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
949 offsetof(spa_error_entry_t
, se_avl
));
950 avl_create(&spa
->spa_errlist_last
,
951 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
952 offsetof(spa_error_entry_t
, se_avl
));
956 * Opposite of spa_activate().
959 spa_deactivate(spa_t
*spa
)
961 ASSERT(spa
->spa_sync_on
== B_FALSE
);
962 ASSERT(spa
->spa_dsl_pool
== NULL
);
963 ASSERT(spa
->spa_root_vdev
== NULL
);
964 ASSERT(spa
->spa_async_zio_root
== NULL
);
965 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
967 txg_list_destroy(&spa
->spa_vdev_txg_list
);
969 list_destroy(&spa
->spa_config_dirty_list
);
970 list_destroy(&spa
->spa_state_dirty_list
);
972 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
973 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
974 if (spa
->spa_zio_taskq
[t
][q
] != NULL
)
975 taskq_destroy(spa
->spa_zio_taskq
[t
][q
]);
976 spa
->spa_zio_taskq
[t
][q
] = NULL
;
980 metaslab_class_destroy(spa
->spa_normal_class
);
981 spa
->spa_normal_class
= NULL
;
983 metaslab_class_destroy(spa
->spa_log_class
);
984 spa
->spa_log_class
= NULL
;
987 * If this was part of an import or the open otherwise failed, we may
988 * still have errors left in the queues. Empty them just in case.
990 spa_errlog_drain(spa
);
992 avl_destroy(&spa
->spa_errlist_scrub
);
993 avl_destroy(&spa
->spa_errlist_last
);
995 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
997 mutex_enter(&spa
->spa_proc_lock
);
998 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
999 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1000 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1001 cv_broadcast(&spa
->spa_proc_cv
);
1002 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1003 ASSERT(spa
->spa_proc
!= &p0
);
1004 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1006 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1007 spa
->spa_proc_state
= SPA_PROC_NONE
;
1009 ASSERT(spa
->spa_proc
== &p0
);
1010 mutex_exit(&spa
->spa_proc_lock
);
1013 * We want to make sure spa_thread() has actually exited the ZFS
1014 * module, so that the module can't be unloaded out from underneath
1017 if (spa
->spa_did
!= 0) {
1018 thread_join(spa
->spa_did
);
1024 * Verify a pool configuration, and construct the vdev tree appropriately. This
1025 * will create all the necessary vdevs in the appropriate layout, with each vdev
1026 * in the CLOSED state. This will prep the pool before open/creation/import.
1027 * All vdev validation is done by the vdev_alloc() routine.
1030 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1031 uint_t id
, int atype
)
1037 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1040 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1043 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1046 if (error
== ENOENT
)
1055 for (int c
= 0; c
< children
; c
++) {
1057 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1065 ASSERT(*vdp
!= NULL
);
1071 * Opposite of spa_load().
1074 spa_unload(spa_t
*spa
)
1078 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1083 spa_async_suspend(spa
);
1088 if (spa
->spa_sync_on
) {
1089 txg_sync_stop(spa
->spa_dsl_pool
);
1090 spa
->spa_sync_on
= B_FALSE
;
1094 * Wait for any outstanding async I/O to complete.
1096 if (spa
->spa_async_zio_root
!= NULL
) {
1097 (void) zio_wait(spa
->spa_async_zio_root
);
1098 spa
->spa_async_zio_root
= NULL
;
1101 bpobj_close(&spa
->spa_deferred_bpobj
);
1104 * Close the dsl pool.
1106 if (spa
->spa_dsl_pool
) {
1107 dsl_pool_close(spa
->spa_dsl_pool
);
1108 spa
->spa_dsl_pool
= NULL
;
1109 spa
->spa_meta_objset
= NULL
;
1114 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1117 * Drop and purge level 2 cache
1119 spa_l2cache_drop(spa
);
1124 if (spa
->spa_root_vdev
)
1125 vdev_free(spa
->spa_root_vdev
);
1126 ASSERT(spa
->spa_root_vdev
== NULL
);
1128 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1129 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1130 if (spa
->spa_spares
.sav_vdevs
) {
1131 kmem_free(spa
->spa_spares
.sav_vdevs
,
1132 spa
->spa_spares
.sav_count
* sizeof (void *));
1133 spa
->spa_spares
.sav_vdevs
= NULL
;
1135 if (spa
->spa_spares
.sav_config
) {
1136 nvlist_free(spa
->spa_spares
.sav_config
);
1137 spa
->spa_spares
.sav_config
= NULL
;
1139 spa
->spa_spares
.sav_count
= 0;
1141 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1142 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1143 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1145 if (spa
->spa_l2cache
.sav_vdevs
) {
1146 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1147 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1148 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1150 if (spa
->spa_l2cache
.sav_config
) {
1151 nvlist_free(spa
->spa_l2cache
.sav_config
);
1152 spa
->spa_l2cache
.sav_config
= NULL
;
1154 spa
->spa_l2cache
.sav_count
= 0;
1156 spa
->spa_async_suspended
= 0;
1158 if (spa
->spa_comment
!= NULL
) {
1159 spa_strfree(spa
->spa_comment
);
1160 spa
->spa_comment
= NULL
;
1163 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1167 * Load (or re-load) the current list of vdevs describing the active spares for
1168 * this pool. When this is called, we have some form of basic information in
1169 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1170 * then re-generate a more complete list including status information.
1173 spa_load_spares(spa_t
*spa
)
1180 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1183 * First, close and free any existing spare vdevs.
1185 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1186 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1188 /* Undo the call to spa_activate() below */
1189 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1190 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1191 spa_spare_remove(tvd
);
1196 if (spa
->spa_spares
.sav_vdevs
)
1197 kmem_free(spa
->spa_spares
.sav_vdevs
,
1198 spa
->spa_spares
.sav_count
* sizeof (void *));
1200 if (spa
->spa_spares
.sav_config
== NULL
)
1203 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1204 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1206 spa
->spa_spares
.sav_count
= (int)nspares
;
1207 spa
->spa_spares
.sav_vdevs
= NULL
;
1213 * Construct the array of vdevs, opening them to get status in the
1214 * process. For each spare, there is potentially two different vdev_t
1215 * structures associated with it: one in the list of spares (used only
1216 * for basic validation purposes) and one in the active vdev
1217 * configuration (if it's spared in). During this phase we open and
1218 * validate each vdev on the spare list. If the vdev also exists in the
1219 * active configuration, then we also mark this vdev as an active spare.
1221 spa
->spa_spares
.sav_vdevs
= kmem_alloc(nspares
* sizeof (void *),
1223 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1224 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1225 VDEV_ALLOC_SPARE
) == 0);
1228 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1230 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1231 B_FALSE
)) != NULL
) {
1232 if (!tvd
->vdev_isspare
)
1236 * We only mark the spare active if we were successfully
1237 * able to load the vdev. Otherwise, importing a pool
1238 * with a bad active spare would result in strange
1239 * behavior, because multiple pool would think the spare
1240 * is actively in use.
1242 * There is a vulnerability here to an equally bizarre
1243 * circumstance, where a dead active spare is later
1244 * brought back to life (onlined or otherwise). Given
1245 * the rarity of this scenario, and the extra complexity
1246 * it adds, we ignore the possibility.
1248 if (!vdev_is_dead(tvd
))
1249 spa_spare_activate(tvd
);
1253 vd
->vdev_aux
= &spa
->spa_spares
;
1255 if (vdev_open(vd
) != 0)
1258 if (vdev_validate_aux(vd
) == 0)
1263 * Recompute the stashed list of spares, with status information
1266 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1267 DATA_TYPE_NVLIST_ARRAY
) == 0);
1269 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1271 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1272 spares
[i
] = vdev_config_generate(spa
,
1273 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1274 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1275 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1276 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1277 nvlist_free(spares
[i
]);
1278 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1282 * Load (or re-load) the current list of vdevs describing the active l2cache for
1283 * this pool. When this is called, we have some form of basic information in
1284 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1285 * then re-generate a more complete list including status information.
1286 * Devices which are already active have their details maintained, and are
1290 spa_load_l2cache(spa_t
*spa
)
1294 int i
, j
, oldnvdevs
;
1296 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1297 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1299 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1301 if (sav
->sav_config
!= NULL
) {
1302 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1303 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1304 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1309 oldvdevs
= sav
->sav_vdevs
;
1310 oldnvdevs
= sav
->sav_count
;
1311 sav
->sav_vdevs
= NULL
;
1315 * Process new nvlist of vdevs.
1317 for (i
= 0; i
< nl2cache
; i
++) {
1318 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1322 for (j
= 0; j
< oldnvdevs
; j
++) {
1324 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1326 * Retain previous vdev for add/remove ops.
1334 if (newvdevs
[i
] == NULL
) {
1338 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1339 VDEV_ALLOC_L2CACHE
) == 0);
1344 * Commit this vdev as an l2cache device,
1345 * even if it fails to open.
1347 spa_l2cache_add(vd
);
1352 spa_l2cache_activate(vd
);
1354 if (vdev_open(vd
) != 0)
1357 (void) vdev_validate_aux(vd
);
1359 if (!vdev_is_dead(vd
))
1360 l2arc_add_vdev(spa
, vd
);
1365 * Purge vdevs that were dropped
1367 for (i
= 0; i
< oldnvdevs
; i
++) {
1372 ASSERT(vd
->vdev_isl2cache
);
1374 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1375 pool
!= 0ULL && l2arc_vdev_present(vd
))
1376 l2arc_remove_vdev(vd
);
1377 vdev_clear_stats(vd
);
1383 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1385 if (sav
->sav_config
== NULL
)
1388 sav
->sav_vdevs
= newvdevs
;
1389 sav
->sav_count
= (int)nl2cache
;
1392 * Recompute the stashed list of l2cache devices, with status
1393 * information this time.
1395 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1396 DATA_TYPE_NVLIST_ARRAY
) == 0);
1398 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1399 for (i
= 0; i
< sav
->sav_count
; i
++)
1400 l2cache
[i
] = vdev_config_generate(spa
,
1401 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1402 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1403 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1405 for (i
= 0; i
< sav
->sav_count
; i
++)
1406 nvlist_free(l2cache
[i
]);
1408 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1412 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1415 char *packed
= NULL
;
1420 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
1421 nvsize
= *(uint64_t *)db
->db_data
;
1422 dmu_buf_rele(db
, FTAG
);
1424 packed
= kmem_alloc(nvsize
, KM_SLEEP
);
1425 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1428 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1429 kmem_free(packed
, nvsize
);
1435 * Checks to see if the given vdev could not be opened, in which case we post a
1436 * sysevent to notify the autoreplace code that the device has been removed.
1439 spa_check_removed(vdev_t
*vd
)
1441 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1442 spa_check_removed(vd
->vdev_child
[c
]);
1444 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
)) {
1445 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1446 spa_event_notify(vd
->vdev_spa
, vd
, ESC_ZFS_VDEV_CHECK
);
1451 * Validate the current config against the MOS config
1454 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1456 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1459 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1461 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1462 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1464 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1467 * If we're doing a normal import, then build up any additional
1468 * diagnostic information about missing devices in this config.
1469 * We'll pass this up to the user for further processing.
1471 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1472 nvlist_t
**child
, *nv
;
1475 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1477 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1479 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1480 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1481 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1483 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1484 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1486 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1491 VERIFY(nvlist_add_nvlist_array(nv
,
1492 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1493 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1494 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1496 for (int i
= 0; i
< idx
; i
++)
1497 nvlist_free(child
[i
]);
1500 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1504 * Compare the root vdev tree with the information we have
1505 * from the MOS config (mrvd). Check each top-level vdev
1506 * with the corresponding MOS config top-level (mtvd).
1508 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1509 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1510 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1513 * Resolve any "missing" vdevs in the current configuration.
1514 * If we find that the MOS config has more accurate information
1515 * about the top-level vdev then use that vdev instead.
1517 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1518 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1520 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1524 * Device specific actions.
1526 if (mtvd
->vdev_islog
) {
1527 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1530 * XXX - once we have 'readonly' pool
1531 * support we should be able to handle
1532 * missing data devices by transitioning
1533 * the pool to readonly.
1539 * Swap the missing vdev with the data we were
1540 * able to obtain from the MOS config.
1542 vdev_remove_child(rvd
, tvd
);
1543 vdev_remove_child(mrvd
, mtvd
);
1545 vdev_add_child(rvd
, mtvd
);
1546 vdev_add_child(mrvd
, tvd
);
1548 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1550 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1553 } else if (mtvd
->vdev_islog
) {
1555 * Load the slog device's state from the MOS config
1556 * since it's possible that the label does not
1557 * contain the most up-to-date information.
1559 vdev_load_log_state(tvd
, mtvd
);
1564 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1567 * Ensure we were able to validate the config.
1569 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1573 * Check for missing log devices
1576 spa_check_logs(spa_t
*spa
)
1578 switch (spa
->spa_log_state
) {
1579 case SPA_LOG_MISSING
:
1580 /* need to recheck in case slog has been restored */
1581 case SPA_LOG_UNKNOWN
:
1582 if (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
, NULL
,
1583 DS_FIND_CHILDREN
)) {
1584 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1593 spa_passivate_log(spa_t
*spa
)
1595 vdev_t
*rvd
= spa
->spa_root_vdev
;
1596 boolean_t slog_found
= B_FALSE
;
1598 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1600 if (!spa_has_slogs(spa
))
1603 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1604 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1605 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1607 if (tvd
->vdev_islog
) {
1608 metaslab_group_passivate(mg
);
1609 slog_found
= B_TRUE
;
1613 return (slog_found
);
1617 spa_activate_log(spa_t
*spa
)
1619 vdev_t
*rvd
= spa
->spa_root_vdev
;
1621 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1623 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1624 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1625 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1627 if (tvd
->vdev_islog
)
1628 metaslab_group_activate(mg
);
1633 spa_offline_log(spa_t
*spa
)
1637 if ((error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1638 NULL
, DS_FIND_CHILDREN
)) == 0) {
1641 * We successfully offlined the log device, sync out the
1642 * current txg so that the "stubby" block can be removed
1645 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1651 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1653 for (int i
= 0; i
< sav
->sav_count
; i
++)
1654 spa_check_removed(sav
->sav_vdevs
[i
]);
1658 spa_claim_notify(zio_t
*zio
)
1660 spa_t
*spa
= zio
->io_spa
;
1665 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1666 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1667 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1668 mutex_exit(&spa
->spa_props_lock
);
1671 typedef struct spa_load_error
{
1672 uint64_t sle_meta_count
;
1673 uint64_t sle_data_count
;
1677 spa_load_verify_done(zio_t
*zio
)
1679 blkptr_t
*bp
= zio
->io_bp
;
1680 spa_load_error_t
*sle
= zio
->io_private
;
1681 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1682 int error
= zio
->io_error
;
1685 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1686 type
!= DMU_OT_INTENT_LOG
)
1687 atomic_add_64(&sle
->sle_meta_count
, 1);
1689 atomic_add_64(&sle
->sle_data_count
, 1);
1691 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1696 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1697 arc_buf_t
*pbuf
, const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1701 size_t size
= BP_GET_PSIZE(bp
);
1702 void *data
= zio_data_buf_alloc(size
);
1704 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1705 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1706 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1707 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1713 spa_load_verify(spa_t
*spa
)
1716 spa_load_error_t sle
= { 0 };
1717 zpool_rewind_policy_t policy
;
1718 boolean_t verify_ok
= B_FALSE
;
1721 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1723 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1726 rio
= zio_root(spa
, NULL
, &sle
,
1727 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1729 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1730 TRAVERSE_PRE
| TRAVERSE_PREFETCH
, spa_load_verify_cb
, rio
);
1732 (void) zio_wait(rio
);
1734 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1735 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1737 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1738 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1742 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1743 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1745 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1746 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1747 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1748 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1749 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1750 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1751 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1753 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1757 if (error
!= ENXIO
&& error
!= EIO
)
1762 return (verify_ok
? 0 : EIO
);
1766 * Find a value in the pool props object.
1769 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
1771 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
1772 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
1776 * Find a value in the pool directory object.
1779 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
1781 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1782 name
, sizeof (uint64_t), 1, val
));
1786 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
1788 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
1793 * Fix up config after a partly-completed split. This is done with the
1794 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1795 * pool have that entry in their config, but only the splitting one contains
1796 * a list of all the guids of the vdevs that are being split off.
1798 * This function determines what to do with that list: either rejoin
1799 * all the disks to the pool, or complete the splitting process. To attempt
1800 * the rejoin, each disk that is offlined is marked online again, and
1801 * we do a reopen() call. If the vdev label for every disk that was
1802 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1803 * then we call vdev_split() on each disk, and complete the split.
1805 * Otherwise we leave the config alone, with all the vdevs in place in
1806 * the original pool.
1809 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
1816 boolean_t attempt_reopen
;
1818 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
1821 /* check that the config is complete */
1822 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
1823 &glist
, &gcount
) != 0)
1826 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
1828 /* attempt to online all the vdevs & validate */
1829 attempt_reopen
= B_TRUE
;
1830 for (i
= 0; i
< gcount
; i
++) {
1831 if (glist
[i
] == 0) /* vdev is hole */
1834 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
1835 if (vd
[i
] == NULL
) {
1837 * Don't bother attempting to reopen the disks;
1838 * just do the split.
1840 attempt_reopen
= B_FALSE
;
1842 /* attempt to re-online it */
1843 vd
[i
]->vdev_offline
= B_FALSE
;
1847 if (attempt_reopen
) {
1848 vdev_reopen(spa
->spa_root_vdev
);
1850 /* check each device to see what state it's in */
1851 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
1852 if (vd
[i
] != NULL
&&
1853 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
1860 * If every disk has been moved to the new pool, or if we never
1861 * even attempted to look at them, then we split them off for
1864 if (!attempt_reopen
|| gcount
== extracted
) {
1865 for (i
= 0; i
< gcount
; i
++)
1868 vdev_reopen(spa
->spa_root_vdev
);
1871 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
1875 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
1876 boolean_t mosconfig
)
1878 nvlist_t
*config
= spa
->spa_config
;
1879 char *ereport
= FM_EREPORT_ZFS_POOL
;
1885 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
1888 ASSERT(spa
->spa_comment
== NULL
);
1889 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
1890 spa
->spa_comment
= spa_strdup(comment
);
1893 * Versioning wasn't explicitly added to the label until later, so if
1894 * it's not present treat it as the initial version.
1896 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
1897 &spa
->spa_ubsync
.ub_version
) != 0)
1898 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
1900 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
1901 &spa
->spa_config_txg
);
1903 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
1904 spa_guid_exists(pool_guid
, 0)) {
1907 spa
->spa_config_guid
= pool_guid
;
1909 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
1911 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
1915 nvlist_free(spa
->spa_load_info
);
1916 spa
->spa_load_info
= fnvlist_alloc();
1918 gethrestime(&spa
->spa_loaded_ts
);
1919 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
1920 mosconfig
, &ereport
);
1923 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
1925 if (error
!= EEXIST
) {
1926 spa
->spa_loaded_ts
.tv_sec
= 0;
1927 spa
->spa_loaded_ts
.tv_nsec
= 0;
1929 if (error
!= EBADF
) {
1930 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
1933 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
1940 * Load an existing storage pool, using the pool's builtin spa_config as a
1941 * source of configuration information.
1944 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
1945 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
1949 nvlist_t
*nvroot
= NULL
;
1952 uberblock_t
*ub
= &spa
->spa_uberblock
;
1953 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
1954 int orig_mode
= spa
->spa_mode
;
1957 boolean_t missing_feat_write
= B_FALSE
;
1960 * If this is an untrusted config, access the pool in read-only mode.
1961 * This prevents things like resilvering recently removed devices.
1964 spa
->spa_mode
= FREAD
;
1966 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1968 spa
->spa_load_state
= state
;
1970 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
1973 parse
= (type
== SPA_IMPORT_EXISTING
?
1974 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
1977 * Create "The Godfather" zio to hold all async IOs
1979 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
1980 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
1983 * Parse the configuration into a vdev tree. We explicitly set the
1984 * value that will be returned by spa_version() since parsing the
1985 * configuration requires knowing the version number.
1987 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1988 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
1989 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1994 ASSERT(spa
->spa_root_vdev
== rvd
);
1996 if (type
!= SPA_IMPORT_ASSEMBLE
) {
1997 ASSERT(spa_guid(spa
) == pool_guid
);
2001 * Try to open all vdevs, loading each label in the process.
2003 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2004 error
= vdev_open(rvd
);
2005 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2010 * We need to validate the vdev labels against the configuration that
2011 * we have in hand, which is dependent on the setting of mosconfig. If
2012 * mosconfig is true then we're validating the vdev labels based on
2013 * that config. Otherwise, we're validating against the cached config
2014 * (zpool.cache) that was read when we loaded the zfs module, and then
2015 * later we will recursively call spa_load() and validate against
2018 * If we're assembling a new pool that's been split off from an
2019 * existing pool, the labels haven't yet been updated so we skip
2020 * validation for now.
2022 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2023 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2024 error
= vdev_validate(rvd
, mosconfig
);
2025 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2030 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2035 * Find the best uberblock.
2037 vdev_uberblock_load(rvd
, ub
, &label
);
2040 * If we weren't able to find a single valid uberblock, return failure.
2042 if (ub
->ub_txg
== 0) {
2044 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2048 * If the pool has an unsupported version we can't open it.
2050 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2052 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2055 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2059 * If we weren't able to find what's necessary for reading the
2060 * MOS in the label, return failure.
2062 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2063 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2065 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2070 * Update our in-core representation with the definitive values
2073 nvlist_free(spa
->spa_label_features
);
2074 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2080 * Look through entries in the label nvlist's features_for_read. If
2081 * there is a feature listed there which we don't understand then we
2082 * cannot open a pool.
2084 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2085 nvlist_t
*unsup_feat
;
2087 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2090 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
2092 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2093 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2094 VERIFY(nvlist_add_string(unsup_feat
,
2095 nvpair_name(nvp
), "") == 0);
2099 if (!nvlist_empty(unsup_feat
)) {
2100 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2101 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2102 nvlist_free(unsup_feat
);
2103 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2107 nvlist_free(unsup_feat
);
2111 * If the vdev guid sum doesn't match the uberblock, we have an
2112 * incomplete configuration. We first check to see if the pool
2113 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2114 * If it is, defer the vdev_guid_sum check till later so we
2115 * can handle missing vdevs.
2117 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2118 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2119 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2120 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2122 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2123 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2124 spa_try_repair(spa
, config
);
2125 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2126 nvlist_free(spa
->spa_config_splitting
);
2127 spa
->spa_config_splitting
= NULL
;
2131 * Initialize internal SPA structures.
2133 spa
->spa_state
= POOL_STATE_ACTIVE
;
2134 spa
->spa_ubsync
= spa
->spa_uberblock
;
2135 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2136 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2137 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2138 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2139 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2140 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2142 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2144 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2145 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2147 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2148 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2150 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2151 boolean_t missing_feat_read
= B_FALSE
;
2152 nvlist_t
*unsup_feat
;
2154 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2155 &spa
->spa_feat_for_read_obj
) != 0) {
2156 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2159 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2160 &spa
->spa_feat_for_write_obj
) != 0) {
2161 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2164 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2165 &spa
->spa_feat_desc_obj
) != 0) {
2166 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2169 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2172 if (!feature_is_supported(spa
->spa_meta_objset
,
2173 spa
->spa_feat_for_read_obj
, spa
->spa_feat_desc_obj
,
2175 missing_feat_read
= B_TRUE
;
2177 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2178 if (!feature_is_supported(spa
->spa_meta_objset
,
2179 spa
->spa_feat_for_write_obj
, spa
->spa_feat_desc_obj
,
2181 missing_feat_write
= B_TRUE
;
2184 if (!nvlist_empty(unsup_feat
)) {
2185 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2186 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2189 nvlist_free(unsup_feat
);
2191 if (!missing_feat_read
) {
2192 fnvlist_add_boolean(spa
->spa_load_info
,
2193 ZPOOL_CONFIG_CAN_RDONLY
);
2197 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2198 * twofold: to determine whether the pool is available for
2199 * import in read-write mode and (if it is not) whether the
2200 * pool is available for import in read-only mode. If the pool
2201 * is available for import in read-write mode, it is displayed
2202 * as available in userland; if it is not available for import
2203 * in read-only mode, it is displayed as unavailable in
2204 * userland. If the pool is available for import in read-only
2205 * mode but not read-write mode, it is displayed as unavailable
2206 * in userland with a special note that the pool is actually
2207 * available for open in read-only mode.
2209 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2210 * missing a feature for write, we must first determine whether
2211 * the pool can be opened read-only before returning to
2212 * userland in order to know whether to display the
2213 * abovementioned note.
2215 if (missing_feat_read
|| (missing_feat_write
&&
2216 spa_writeable(spa
))) {
2217 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2222 spa
->spa_is_initializing
= B_TRUE
;
2223 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2224 spa
->spa_is_initializing
= B_FALSE
;
2226 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2230 nvlist_t
*policy
= NULL
, *nvconfig
;
2232 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2233 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2235 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2236 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2238 unsigned long myhostid
= 0;
2240 VERIFY(nvlist_lookup_string(nvconfig
,
2241 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2244 myhostid
= zone_get_hostid(NULL
);
2247 * We're emulating the system's hostid in userland, so
2248 * we can't use zone_get_hostid().
2250 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2251 #endif /* _KERNEL */
2252 if (hostid
!= 0 && myhostid
!= 0 &&
2253 hostid
!= myhostid
) {
2254 nvlist_free(nvconfig
);
2255 cmn_err(CE_WARN
, "pool '%s' could not be "
2256 "loaded as it was last accessed by "
2257 "another system (host: %s hostid: 0x%lx). "
2258 "See: http://illumos.org/msg/ZFS-8000-EY",
2259 spa_name(spa
), hostname
,
2260 (unsigned long)hostid
);
2264 if (nvlist_lookup_nvlist(spa
->spa_config
,
2265 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2266 VERIFY(nvlist_add_nvlist(nvconfig
,
2267 ZPOOL_REWIND_POLICY
, policy
) == 0);
2269 spa_config_set(spa
, nvconfig
);
2271 spa_deactivate(spa
);
2272 spa_activate(spa
, orig_mode
);
2274 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2277 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2278 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2279 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2281 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2284 * Load the bit that tells us to use the new accounting function
2285 * (raid-z deflation). If we have an older pool, this will not
2288 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2289 if (error
!= 0 && error
!= ENOENT
)
2290 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2292 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2293 &spa
->spa_creation_version
);
2294 if (error
!= 0 && error
!= ENOENT
)
2295 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2298 * Load the persistent error log. If we have an older pool, this will
2301 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2302 if (error
!= 0 && error
!= ENOENT
)
2303 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2305 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2306 &spa
->spa_errlog_scrub
);
2307 if (error
!= 0 && error
!= ENOENT
)
2308 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2311 * Load the history object. If we have an older pool, this
2312 * will not be present.
2314 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2315 if (error
!= 0 && error
!= ENOENT
)
2316 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2319 * If we're assembling the pool from the split-off vdevs of
2320 * an existing pool, we don't want to attach the spares & cache
2325 * Load any hot spares for this pool.
2327 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2328 if (error
!= 0 && error
!= ENOENT
)
2329 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2330 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2331 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2332 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2333 &spa
->spa_spares
.sav_config
) != 0)
2334 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2336 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2337 spa_load_spares(spa
);
2338 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2339 } else if (error
== 0) {
2340 spa
->spa_spares
.sav_sync
= B_TRUE
;
2344 * Load any level 2 ARC devices for this pool.
2346 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2347 &spa
->spa_l2cache
.sav_object
);
2348 if (error
!= 0 && error
!= ENOENT
)
2349 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2350 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2351 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2352 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2353 &spa
->spa_l2cache
.sav_config
) != 0)
2354 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2356 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2357 spa_load_l2cache(spa
);
2358 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2359 } else if (error
== 0) {
2360 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2363 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2365 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2366 if (error
&& error
!= ENOENT
)
2367 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2370 uint64_t autoreplace
;
2372 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2373 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2374 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2375 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2376 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2377 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2378 &spa
->spa_dedup_ditto
);
2380 spa
->spa_autoreplace
= (autoreplace
!= 0);
2384 * If the 'autoreplace' property is set, then post a resource notifying
2385 * the ZFS DE that it should not issue any faults for unopenable
2386 * devices. We also iterate over the vdevs, and post a sysevent for any
2387 * unopenable vdevs so that the normal autoreplace handler can take
2390 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2391 spa_check_removed(spa
->spa_root_vdev
);
2393 * For the import case, this is done in spa_import(), because
2394 * at this point we're using the spare definitions from
2395 * the MOS config, not necessarily from the userland config.
2397 if (state
!= SPA_LOAD_IMPORT
) {
2398 spa_aux_check_removed(&spa
->spa_spares
);
2399 spa_aux_check_removed(&spa
->spa_l2cache
);
2404 * Load the vdev state for all toplevel vdevs.
2409 * Propagate the leaf DTLs we just loaded all the way up the tree.
2411 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2412 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2413 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2416 * Load the DDTs (dedup tables).
2418 error
= ddt_load(spa
);
2420 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2422 spa_update_dspace(spa
);
2425 * Validate the config, using the MOS config to fill in any
2426 * information which might be missing. If we fail to validate
2427 * the config then declare the pool unfit for use. If we're
2428 * assembling a pool from a split, the log is not transferred
2431 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2434 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2435 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2437 if (!spa_config_valid(spa
, nvconfig
)) {
2438 nvlist_free(nvconfig
);
2439 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2442 nvlist_free(nvconfig
);
2445 * Now that we've validated the config, check the state of the
2446 * root vdev. If it can't be opened, it indicates one or
2447 * more toplevel vdevs are faulted.
2449 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2452 if (spa_check_logs(spa
)) {
2453 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2454 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2458 if (missing_feat_write
) {
2459 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2462 * At this point, we know that we can open the pool in
2463 * read-only mode but not read-write mode. We now have enough
2464 * information and can return to userland.
2466 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2470 * We've successfully opened the pool, verify that we're ready
2471 * to start pushing transactions.
2473 if (state
!= SPA_LOAD_TRYIMPORT
) {
2474 if (error
= spa_load_verify(spa
))
2475 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2479 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2480 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2482 int need_update
= B_FALSE
;
2484 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2487 * Claim log blocks that haven't been committed yet.
2488 * This must all happen in a single txg.
2489 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2490 * invoked from zil_claim_log_block()'s i/o done callback.
2491 * Price of rollback is that we abandon the log.
2493 spa
->spa_claiming
= B_TRUE
;
2495 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
2496 spa_first_txg(spa
));
2497 (void) dmu_objset_find(spa_name(spa
),
2498 zil_claim
, tx
, DS_FIND_CHILDREN
);
2501 spa
->spa_claiming
= B_FALSE
;
2503 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2504 spa
->spa_sync_on
= B_TRUE
;
2505 txg_sync_start(spa
->spa_dsl_pool
);
2508 * Wait for all claims to sync. We sync up to the highest
2509 * claimed log block birth time so that claimed log blocks
2510 * don't appear to be from the future. spa_claim_max_txg
2511 * will have been set for us by either zil_check_log_chain()
2512 * (invoked from spa_check_logs()) or zil_claim() above.
2514 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2517 * If the config cache is stale, or we have uninitialized
2518 * metaslabs (see spa_vdev_add()), then update the config.
2520 * If this is a verbatim import, trust the current
2521 * in-core spa_config and update the disk labels.
2523 if (config_cache_txg
!= spa
->spa_config_txg
||
2524 state
== SPA_LOAD_IMPORT
||
2525 state
== SPA_LOAD_RECOVER
||
2526 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2527 need_update
= B_TRUE
;
2529 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
2530 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2531 need_update
= B_TRUE
;
2534 * Update the config cache asychronously in case we're the
2535 * root pool, in which case the config cache isn't writable yet.
2538 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2541 * Check all DTLs to see if anything needs resilvering.
2543 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2544 vdev_resilver_needed(rvd
, NULL
, NULL
))
2545 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2548 * Log the fact that we booted up (so that we can detect if
2549 * we rebooted in the middle of an operation).
2551 spa_history_log_version(spa
, "open");
2554 * Delete any inconsistent datasets.
2556 (void) dmu_objset_find(spa_name(spa
),
2557 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2560 * Clean up any stale temporary dataset userrefs.
2562 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2569 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2571 int mode
= spa
->spa_mode
;
2574 spa_deactivate(spa
);
2576 spa
->spa_load_max_txg
--;
2578 spa_activate(spa
, mode
);
2579 spa_async_suspend(spa
);
2581 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2585 * If spa_load() fails this function will try loading prior txg's. If
2586 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2587 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2588 * function will not rewind the pool and will return the same error as
2592 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2593 uint64_t max_request
, int rewind_flags
)
2595 nvlist_t
*loadinfo
= NULL
;
2596 nvlist_t
*config
= NULL
;
2597 int load_error
, rewind_error
;
2598 uint64_t safe_rewind_txg
;
2601 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2602 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2603 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2605 spa
->spa_load_max_txg
= max_request
;
2608 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2610 if (load_error
== 0)
2613 if (spa
->spa_root_vdev
!= NULL
)
2614 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2616 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2617 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2619 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2620 nvlist_free(config
);
2621 return (load_error
);
2624 if (state
== SPA_LOAD_RECOVER
) {
2625 /* Price of rolling back is discarding txgs, including log */
2626 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2629 * If we aren't rolling back save the load info from our first
2630 * import attempt so that we can restore it after attempting
2633 loadinfo
= spa
->spa_load_info
;
2634 spa
->spa_load_info
= fnvlist_alloc();
2637 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2638 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2639 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2640 TXG_INITIAL
: safe_rewind_txg
;
2643 * Continue as long as we're finding errors, we're still within
2644 * the acceptable rewind range, and we're still finding uberblocks
2646 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2647 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2648 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2649 spa
->spa_extreme_rewind
= B_TRUE
;
2650 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2653 spa
->spa_extreme_rewind
= B_FALSE
;
2654 spa
->spa_load_max_txg
= UINT64_MAX
;
2656 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2657 spa_config_set(spa
, config
);
2659 if (state
== SPA_LOAD_RECOVER
) {
2660 ASSERT3P(loadinfo
, ==, NULL
);
2661 return (rewind_error
);
2663 /* Store the rewind info as part of the initial load info */
2664 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
2665 spa
->spa_load_info
);
2667 /* Restore the initial load info */
2668 fnvlist_free(spa
->spa_load_info
);
2669 spa
->spa_load_info
= loadinfo
;
2671 return (load_error
);
2678 * The import case is identical to an open except that the configuration is sent
2679 * down from userland, instead of grabbed from the configuration cache. For the
2680 * case of an open, the pool configuration will exist in the
2681 * POOL_STATE_UNINITIALIZED state.
2683 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2684 * the same time open the pool, without having to keep around the spa_t in some
2688 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2692 spa_load_state_t state
= SPA_LOAD_OPEN
;
2694 int locked
= B_FALSE
;
2699 * As disgusting as this is, we need to support recursive calls to this
2700 * function because dsl_dir_open() is called during spa_load(), and ends
2701 * up calling spa_open() again. The real fix is to figure out how to
2702 * avoid dsl_dir_open() calling this in the first place.
2704 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2705 mutex_enter(&spa_namespace_lock
);
2709 if ((spa
= spa_lookup(pool
)) == NULL
) {
2711 mutex_exit(&spa_namespace_lock
);
2715 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
2716 zpool_rewind_policy_t policy
;
2718 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
2720 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
2721 state
= SPA_LOAD_RECOVER
;
2723 spa_activate(spa
, spa_mode_global
);
2725 if (state
!= SPA_LOAD_RECOVER
)
2726 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
2728 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
2729 policy
.zrp_request
);
2731 if (error
== EBADF
) {
2733 * If vdev_validate() returns failure (indicated by
2734 * EBADF), it indicates that one of the vdevs indicates
2735 * that the pool has been exported or destroyed. If
2736 * this is the case, the config cache is out of sync and
2737 * we should remove the pool from the namespace.
2740 spa_deactivate(spa
);
2741 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
2744 mutex_exit(&spa_namespace_lock
);
2750 * We can't open the pool, but we still have useful
2751 * information: the state of each vdev after the
2752 * attempted vdev_open(). Return this to the user.
2754 if (config
!= NULL
&& spa
->spa_config
) {
2755 VERIFY(nvlist_dup(spa
->spa_config
, config
,
2757 VERIFY(nvlist_add_nvlist(*config
,
2758 ZPOOL_CONFIG_LOAD_INFO
,
2759 spa
->spa_load_info
) == 0);
2762 spa_deactivate(spa
);
2763 spa
->spa_last_open_failed
= error
;
2765 mutex_exit(&spa_namespace_lock
);
2771 spa_open_ref(spa
, tag
);
2774 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2777 * If we've recovered the pool, pass back any information we
2778 * gathered while doing the load.
2780 if (state
== SPA_LOAD_RECOVER
) {
2781 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
2782 spa
->spa_load_info
) == 0);
2786 spa
->spa_last_open_failed
= 0;
2787 spa
->spa_last_ubsync_txg
= 0;
2788 spa
->spa_load_txg
= 0;
2789 mutex_exit(&spa_namespace_lock
);
2798 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
2801 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
2805 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
2807 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
2811 * Lookup the given spa_t, incrementing the inject count in the process,
2812 * preventing it from being exported or destroyed.
2815 spa_inject_addref(char *name
)
2819 mutex_enter(&spa_namespace_lock
);
2820 if ((spa
= spa_lookup(name
)) == NULL
) {
2821 mutex_exit(&spa_namespace_lock
);
2824 spa
->spa_inject_ref
++;
2825 mutex_exit(&spa_namespace_lock
);
2831 spa_inject_delref(spa_t
*spa
)
2833 mutex_enter(&spa_namespace_lock
);
2834 spa
->spa_inject_ref
--;
2835 mutex_exit(&spa_namespace_lock
);
2839 * Add spares device information to the nvlist.
2842 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
2852 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2854 if (spa
->spa_spares
.sav_count
== 0)
2857 VERIFY(nvlist_lookup_nvlist(config
,
2858 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
2859 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
2860 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
2862 VERIFY(nvlist_add_nvlist_array(nvroot
,
2863 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
2864 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
2865 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
2868 * Go through and find any spares which have since been
2869 * repurposed as an active spare. If this is the case, update
2870 * their status appropriately.
2872 for (i
= 0; i
< nspares
; i
++) {
2873 VERIFY(nvlist_lookup_uint64(spares
[i
],
2874 ZPOOL_CONFIG_GUID
, &guid
) == 0);
2875 if (spa_spare_exists(guid
, &pool
, NULL
) &&
2877 VERIFY(nvlist_lookup_uint64_array(
2878 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
2879 (uint64_t **)&vs
, &vsc
) == 0);
2880 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
2881 vs
->vs_aux
= VDEV_AUX_SPARED
;
2888 * Add l2cache device information to the nvlist, including vdev stats.
2891 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
2894 uint_t i
, j
, nl2cache
;
2901 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2903 if (spa
->spa_l2cache
.sav_count
== 0)
2906 VERIFY(nvlist_lookup_nvlist(config
,
2907 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
2908 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
2909 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
2910 if (nl2cache
!= 0) {
2911 VERIFY(nvlist_add_nvlist_array(nvroot
,
2912 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
2913 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
2914 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
2917 * Update level 2 cache device stats.
2920 for (i
= 0; i
< nl2cache
; i
++) {
2921 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
2922 ZPOOL_CONFIG_GUID
, &guid
) == 0);
2925 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
2927 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
2928 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
2934 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
2935 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
2937 vdev_get_stats(vd
, vs
);
2943 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
2949 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2950 VERIFY(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2952 if (spa
->spa_feat_for_read_obj
!= 0) {
2953 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
2954 spa
->spa_feat_for_read_obj
);
2955 zap_cursor_retrieve(&zc
, &za
) == 0;
2956 zap_cursor_advance(&zc
)) {
2957 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
2958 za
.za_num_integers
== 1);
2959 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
2960 za
.za_first_integer
));
2962 zap_cursor_fini(&zc
);
2965 if (spa
->spa_feat_for_write_obj
!= 0) {
2966 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
2967 spa
->spa_feat_for_write_obj
);
2968 zap_cursor_retrieve(&zc
, &za
) == 0;
2969 zap_cursor_advance(&zc
)) {
2970 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
2971 za
.za_num_integers
== 1);
2972 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
2973 za
.za_first_integer
));
2975 zap_cursor_fini(&zc
);
2978 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
2980 nvlist_free(features
);
2984 spa_get_stats(const char *name
, nvlist_t
**config
,
2985 char *altroot
, size_t buflen
)
2991 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
2995 * This still leaves a window of inconsistency where the spares
2996 * or l2cache devices could change and the config would be
2997 * self-inconsistent.
2999 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3001 if (*config
!= NULL
) {
3002 uint64_t loadtimes
[2];
3004 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3005 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3006 VERIFY(nvlist_add_uint64_array(*config
,
3007 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3009 VERIFY(nvlist_add_uint64(*config
,
3010 ZPOOL_CONFIG_ERRCOUNT
,
3011 spa_get_errlog_size(spa
)) == 0);
3013 if (spa_suspended(spa
))
3014 VERIFY(nvlist_add_uint64(*config
,
3015 ZPOOL_CONFIG_SUSPENDED
,
3016 spa
->spa_failmode
) == 0);
3018 spa_add_spares(spa
, *config
);
3019 spa_add_l2cache(spa
, *config
);
3020 spa_add_feature_stats(spa
, *config
);
3025 * We want to get the alternate root even for faulted pools, so we cheat
3026 * and call spa_lookup() directly.
3030 mutex_enter(&spa_namespace_lock
);
3031 spa
= spa_lookup(name
);
3033 spa_altroot(spa
, altroot
, buflen
);
3037 mutex_exit(&spa_namespace_lock
);
3039 spa_altroot(spa
, altroot
, buflen
);
3044 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3045 spa_close(spa
, FTAG
);
3052 * Validate that the auxiliary device array is well formed. We must have an
3053 * array of nvlists, each which describes a valid leaf vdev. If this is an
3054 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3055 * specified, as long as they are well-formed.
3058 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3059 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3060 vdev_labeltype_t label
)
3067 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3070 * It's acceptable to have no devs specified.
3072 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3079 * Make sure the pool is formatted with a version that supports this
3082 if (spa_version(spa
) < version
)
3086 * Set the pending device list so we correctly handle device in-use
3089 sav
->sav_pending
= dev
;
3090 sav
->sav_npending
= ndev
;
3092 for (i
= 0; i
< ndev
; i
++) {
3093 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3097 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3104 * The L2ARC currently only supports disk devices in
3105 * kernel context. For user-level testing, we allow it.
3108 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3109 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3117 if ((error
= vdev_open(vd
)) == 0 &&
3118 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3119 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3120 vd
->vdev_guid
) == 0);
3126 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3133 sav
->sav_pending
= NULL
;
3134 sav
->sav_npending
= 0;
3139 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3143 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3145 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3146 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3147 VDEV_LABEL_SPARE
)) != 0) {
3151 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3152 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3153 VDEV_LABEL_L2CACHE
));
3157 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3162 if (sav
->sav_config
!= NULL
) {
3168 * Generate new dev list by concatentating with the
3171 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3172 &olddevs
, &oldndevs
) == 0);
3174 newdevs
= kmem_alloc(sizeof (void *) *
3175 (ndevs
+ oldndevs
), KM_SLEEP
);
3176 for (i
= 0; i
< oldndevs
; i
++)
3177 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3179 for (i
= 0; i
< ndevs
; i
++)
3180 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3183 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3184 DATA_TYPE_NVLIST_ARRAY
) == 0);
3186 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3187 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3188 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3189 nvlist_free(newdevs
[i
]);
3190 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3193 * Generate a new dev list.
3195 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3197 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3203 * Stop and drop level 2 ARC devices
3206 spa_l2cache_drop(spa_t
*spa
)
3210 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3212 for (i
= 0; i
< sav
->sav_count
; i
++) {
3215 vd
= sav
->sav_vdevs
[i
];
3218 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3219 pool
!= 0ULL && l2arc_vdev_present(vd
))
3220 l2arc_remove_vdev(vd
);
3228 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3232 char *altroot
= NULL
;
3237 uint64_t txg
= TXG_INITIAL
;
3238 nvlist_t
**spares
, **l2cache
;
3239 uint_t nspares
, nl2cache
;
3240 uint64_t version
, obj
;
3241 boolean_t has_features
;
3244 * If this pool already exists, return failure.
3246 mutex_enter(&spa_namespace_lock
);
3247 if (spa_lookup(pool
) != NULL
) {
3248 mutex_exit(&spa_namespace_lock
);
3253 * Allocate a new spa_t structure.
3255 (void) nvlist_lookup_string(props
,
3256 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3257 spa
= spa_add(pool
, NULL
, altroot
);
3258 spa_activate(spa
, spa_mode_global
);
3260 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3261 spa_deactivate(spa
);
3263 mutex_exit(&spa_namespace_lock
);
3267 has_features
= B_FALSE
;
3268 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
3269 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3270 if (zpool_prop_feature(nvpair_name(elem
)))
3271 has_features
= B_TRUE
;
3274 if (has_features
|| nvlist_lookup_uint64(props
,
3275 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3276 version
= SPA_VERSION
;
3278 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3280 spa
->spa_first_txg
= txg
;
3281 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3282 spa
->spa_uberblock
.ub_version
= version
;
3283 spa
->spa_ubsync
= spa
->spa_uberblock
;
3286 * Create "The Godfather" zio to hold all async IOs
3288 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
3289 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
3292 * Create the root vdev.
3294 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3296 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3298 ASSERT(error
!= 0 || rvd
!= NULL
);
3299 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3301 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3305 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3306 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3307 VDEV_ALLOC_ADD
)) == 0) {
3308 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
3309 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3310 vdev_expand(rvd
->vdev_child
[c
], txg
);
3314 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3318 spa_deactivate(spa
);
3320 mutex_exit(&spa_namespace_lock
);
3325 * Get the list of spares, if specified.
3327 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3328 &spares
, &nspares
) == 0) {
3329 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3331 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3332 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3333 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3334 spa_load_spares(spa
);
3335 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3336 spa
->spa_spares
.sav_sync
= B_TRUE
;
3340 * Get the list of level 2 cache devices, if specified.
3342 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3343 &l2cache
, &nl2cache
) == 0) {
3344 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3345 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3346 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3347 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3348 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3349 spa_load_l2cache(spa
);
3350 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3351 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3354 spa
->spa_is_initializing
= B_TRUE
;
3355 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3356 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3357 spa
->spa_is_initializing
= B_FALSE
;
3360 * Create DDTs (dedup tables).
3364 spa_update_dspace(spa
);
3366 tx
= dmu_tx_create_assigned(dp
, txg
);
3369 * Create the pool config object.
3371 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3372 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3373 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3375 if (zap_add(spa
->spa_meta_objset
,
3376 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3377 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3378 cmn_err(CE_PANIC
, "failed to add pool config");
3381 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3382 spa_feature_create_zap_objects(spa
, tx
);
3384 if (zap_add(spa
->spa_meta_objset
,
3385 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3386 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3387 cmn_err(CE_PANIC
, "failed to add pool version");
3390 /* Newly created pools with the right version are always deflated. */
3391 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3392 spa
->spa_deflate
= TRUE
;
3393 if (zap_add(spa
->spa_meta_objset
,
3394 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3395 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3396 cmn_err(CE_PANIC
, "failed to add deflate");
3401 * Create the deferred-free bpobj. Turn off compression
3402 * because sync-to-convergence takes longer if the blocksize
3405 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3406 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3407 ZIO_COMPRESS_OFF
, tx
);
3408 if (zap_add(spa
->spa_meta_objset
,
3409 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3410 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3411 cmn_err(CE_PANIC
, "failed to add bpobj");
3413 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3414 spa
->spa_meta_objset
, obj
));
3417 * Create the pool's history object.
3419 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3420 spa_history_create_obj(spa
, tx
);
3423 * Set pool properties.
3425 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3426 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3427 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3428 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3430 if (props
!= NULL
) {
3431 spa_configfile_set(spa
, props
, B_FALSE
);
3432 spa_sync_props(spa
, props
, tx
);
3437 spa
->spa_sync_on
= B_TRUE
;
3438 txg_sync_start(spa
->spa_dsl_pool
);
3441 * We explicitly wait for the first transaction to complete so that our
3442 * bean counters are appropriately updated.
3444 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3446 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3448 spa_history_log_version(spa
, "create");
3450 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3452 mutex_exit(&spa_namespace_lock
);
3459 * Get the root pool information from the root disk, then import the root pool
3460 * during the system boot up time.
3462 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3465 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3468 nvlist_t
*nvtop
, *nvroot
;
3471 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3475 * Add this top-level vdev to the child array.
3477 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3479 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3481 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3484 * Put this pool's top-level vdevs into a root vdev.
3486 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3487 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3488 VDEV_TYPE_ROOT
) == 0);
3489 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3490 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3491 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3495 * Replace the existing vdev_tree with the new root vdev in
3496 * this pool's configuration (remove the old, add the new).
3498 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3499 nvlist_free(nvroot
);
3504 * Walk the vdev tree and see if we can find a device with "better"
3505 * configuration. A configuration is "better" if the label on that
3506 * device has a more recent txg.
3509 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3511 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3512 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3514 if (vd
->vdev_ops
->vdev_op_leaf
) {
3518 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3522 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3526 * Do we have a better boot device?
3528 if (label_txg
> *txg
) {
3537 * Import a root pool.
3539 * For x86. devpath_list will consist of devid and/or physpath name of
3540 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3541 * The GRUB "findroot" command will return the vdev we should boot.
3543 * For Sparc, devpath_list consists the physpath name of the booting device
3544 * no matter the rootpool is a single device pool or a mirrored pool.
3546 * "/pci@1f,0/ide@d/disk@0,0:a"
3549 spa_import_rootpool(char *devpath
, char *devid
)
3552 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3553 nvlist_t
*config
, *nvtop
;
3559 * Read the label from the boot device and generate a configuration.
3561 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3562 #if defined(_OBP) && defined(_KERNEL)
3563 if (config
== NULL
) {
3564 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3566 get_iscsi_bootpath_phy(devpath
);
3567 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3571 if (config
== NULL
) {
3572 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
3577 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3579 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3581 mutex_enter(&spa_namespace_lock
);
3582 if ((spa
= spa_lookup(pname
)) != NULL
) {
3584 * Remove the existing root pool from the namespace so that we
3585 * can replace it with the correct config we just read in.
3590 spa
= spa_add(pname
, config
, NULL
);
3591 spa
->spa_is_root
= B_TRUE
;
3592 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3595 * Build up a vdev tree based on the boot device's label config.
3597 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3599 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3600 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3601 VDEV_ALLOC_ROOTPOOL
);
3602 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3604 mutex_exit(&spa_namespace_lock
);
3605 nvlist_free(config
);
3606 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3612 * Get the boot vdev.
3614 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3615 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3616 (u_longlong_t
)guid
);
3622 * Determine if there is a better boot device.
3625 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3627 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3628 "try booting from '%s'", avd
->vdev_path
);
3634 * If the boot device is part of a spare vdev then ensure that
3635 * we're booting off the active spare.
3637 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3638 !bvd
->vdev_isspare
) {
3639 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
3640 "try booting from '%s'",
3642 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
3649 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3651 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3652 mutex_exit(&spa_namespace_lock
);
3654 nvlist_free(config
);
3661 * Import a non-root pool into the system.
3664 spa_import(const char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
3667 char *altroot
= NULL
;
3668 spa_load_state_t state
= SPA_LOAD_IMPORT
;
3669 zpool_rewind_policy_t policy
;
3670 uint64_t mode
= spa_mode_global
;
3671 uint64_t readonly
= B_FALSE
;
3674 nvlist_t
**spares
, **l2cache
;
3675 uint_t nspares
, nl2cache
;
3678 * If a pool with this name exists, return failure.
3680 mutex_enter(&spa_namespace_lock
);
3681 if (spa_lookup(pool
) != NULL
) {
3682 mutex_exit(&spa_namespace_lock
);
3687 * Create and initialize the spa structure.
3689 (void) nvlist_lookup_string(props
,
3690 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3691 (void) nvlist_lookup_uint64(props
,
3692 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
3695 spa
= spa_add(pool
, config
, altroot
);
3696 spa
->spa_import_flags
= flags
;
3699 * Verbatim import - Take a pool and insert it into the namespace
3700 * as if it had been loaded at boot.
3702 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
3704 spa_configfile_set(spa
, props
, B_FALSE
);
3706 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3708 mutex_exit(&spa_namespace_lock
);
3709 spa_history_log_version(spa
, "import");
3714 spa_activate(spa
, mode
);
3717 * Don't start async tasks until we know everything is healthy.
3719 spa_async_suspend(spa
);
3721 zpool_get_rewind_policy(config
, &policy
);
3722 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3723 state
= SPA_LOAD_RECOVER
;
3726 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3727 * because the user-supplied config is actually the one to trust when
3730 if (state
!= SPA_LOAD_RECOVER
)
3731 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3733 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
3734 policy
.zrp_request
);
3737 * Propagate anything learned while loading the pool and pass it
3738 * back to caller (i.e. rewind info, missing devices, etc).
3740 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
3741 spa
->spa_load_info
) == 0);
3743 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3745 * Toss any existing sparelist, as it doesn't have any validity
3746 * anymore, and conflicts with spa_has_spare().
3748 if (spa
->spa_spares
.sav_config
) {
3749 nvlist_free(spa
->spa_spares
.sav_config
);
3750 spa
->spa_spares
.sav_config
= NULL
;
3751 spa_load_spares(spa
);
3753 if (spa
->spa_l2cache
.sav_config
) {
3754 nvlist_free(spa
->spa_l2cache
.sav_config
);
3755 spa
->spa_l2cache
.sav_config
= NULL
;
3756 spa_load_l2cache(spa
);
3759 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3762 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3765 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3766 VDEV_ALLOC_L2CACHE
);
3767 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3770 spa_configfile_set(spa
, props
, B_FALSE
);
3772 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
3773 (error
= spa_prop_set(spa
, props
)))) {
3775 spa_deactivate(spa
);
3777 mutex_exit(&spa_namespace_lock
);
3781 spa_async_resume(spa
);
3784 * Override any spares and level 2 cache devices as specified by
3785 * the user, as these may have correct device names/devids, etc.
3787 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3788 &spares
, &nspares
) == 0) {
3789 if (spa
->spa_spares
.sav_config
)
3790 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
3791 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3793 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
3794 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3795 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3796 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3797 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3798 spa_load_spares(spa
);
3799 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3800 spa
->spa_spares
.sav_sync
= B_TRUE
;
3802 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3803 &l2cache
, &nl2cache
) == 0) {
3804 if (spa
->spa_l2cache
.sav_config
)
3805 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
3806 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3808 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3809 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3810 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3811 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3812 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3813 spa_load_l2cache(spa
);
3814 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3815 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3819 * Check for any removed devices.
3821 if (spa
->spa_autoreplace
) {
3822 spa_aux_check_removed(&spa
->spa_spares
);
3823 spa_aux_check_removed(&spa
->spa_l2cache
);
3826 if (spa_writeable(spa
)) {
3828 * Update the config cache to include the newly-imported pool.
3830 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
3834 * It's possible that the pool was expanded while it was exported.
3835 * We kick off an async task to handle this for us.
3837 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
3839 mutex_exit(&spa_namespace_lock
);
3840 spa_history_log_version(spa
, "import");
3846 spa_tryimport(nvlist_t
*tryconfig
)
3848 nvlist_t
*config
= NULL
;
3854 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
3857 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
3861 * Create and initialize the spa structure.
3863 mutex_enter(&spa_namespace_lock
);
3864 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
3865 spa_activate(spa
, FREAD
);
3868 * Pass off the heavy lifting to spa_load().
3869 * Pass TRUE for mosconfig because the user-supplied config
3870 * is actually the one to trust when doing an import.
3872 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
3875 * If 'tryconfig' was at least parsable, return the current config.
3877 if (spa
->spa_root_vdev
!= NULL
) {
3878 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3879 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3881 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
3883 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
3884 spa
->spa_uberblock
.ub_timestamp
) == 0);
3885 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
3886 spa
->spa_load_info
) == 0);
3889 * If the bootfs property exists on this pool then we
3890 * copy it out so that external consumers can tell which
3891 * pools are bootable.
3893 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
3894 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
3897 * We have to play games with the name since the
3898 * pool was opened as TRYIMPORT_NAME.
3900 if (dsl_dsobj_to_dsname(spa_name(spa
),
3901 spa
->spa_bootfs
, tmpname
) == 0) {
3903 char *dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
3905 cp
= strchr(tmpname
, '/');
3907 (void) strlcpy(dsname
, tmpname
,
3910 (void) snprintf(dsname
, MAXPATHLEN
,
3911 "%s/%s", poolname
, ++cp
);
3913 VERIFY(nvlist_add_string(config
,
3914 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
3915 kmem_free(dsname
, MAXPATHLEN
);
3917 kmem_free(tmpname
, MAXPATHLEN
);
3921 * Add the list of hot spares and level 2 cache devices.
3923 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3924 spa_add_spares(spa
, config
);
3925 spa_add_l2cache(spa
, config
);
3926 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3930 spa_deactivate(spa
);
3932 mutex_exit(&spa_namespace_lock
);
3938 * Pool export/destroy
3940 * The act of destroying or exporting a pool is very simple. We make sure there
3941 * is no more pending I/O and any references to the pool are gone. Then, we
3942 * update the pool state and sync all the labels to disk, removing the
3943 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3944 * we don't sync the labels or remove the configuration cache.
3947 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
3948 boolean_t force
, boolean_t hardforce
)
3955 if (!(spa_mode_global
& FWRITE
))
3958 mutex_enter(&spa_namespace_lock
);
3959 if ((spa
= spa_lookup(pool
)) == NULL
) {
3960 mutex_exit(&spa_namespace_lock
);
3965 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3966 * reacquire the namespace lock, and see if we can export.
3968 spa_open_ref(spa
, FTAG
);
3969 mutex_exit(&spa_namespace_lock
);
3970 spa_async_suspend(spa
);
3971 mutex_enter(&spa_namespace_lock
);
3972 spa_close(spa
, FTAG
);
3975 * The pool will be in core if it's openable,
3976 * in which case we can modify its state.
3978 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
&& spa
->spa_sync_on
) {
3980 * Objsets may be open only because they're dirty, so we
3981 * have to force it to sync before checking spa_refcnt.
3983 txg_wait_synced(spa
->spa_dsl_pool
, 0);
3986 * A pool cannot be exported or destroyed if there are active
3987 * references. If we are resetting a pool, allow references by
3988 * fault injection handlers.
3990 if (!spa_refcount_zero(spa
) ||
3991 (spa
->spa_inject_ref
!= 0 &&
3992 new_state
!= POOL_STATE_UNINITIALIZED
)) {
3993 spa_async_resume(spa
);
3994 mutex_exit(&spa_namespace_lock
);
3999 * A pool cannot be exported if it has an active shared spare.
4000 * This is to prevent other pools stealing the active spare
4001 * from an exported pool. At user's own will, such pool can
4002 * be forcedly exported.
4004 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4005 spa_has_active_shared_spare(spa
)) {
4006 spa_async_resume(spa
);
4007 mutex_exit(&spa_namespace_lock
);
4012 * We want this to be reflected on every label,
4013 * so mark them all dirty. spa_unload() will do the
4014 * final sync that pushes these changes out.
4016 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4017 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4018 spa
->spa_state
= new_state
;
4019 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4021 vdev_config_dirty(spa
->spa_root_vdev
);
4022 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4026 spa_event_notify(spa
, NULL
, ESC_ZFS_POOL_DESTROY
);
4028 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4030 spa_deactivate(spa
);
4033 if (oldconfig
&& spa
->spa_config
)
4034 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4036 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4038 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4041 mutex_exit(&spa_namespace_lock
);
4047 * Destroy a storage pool.
4050 spa_destroy(char *pool
)
4052 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4057 * Export a storage pool.
4060 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4061 boolean_t hardforce
)
4063 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4068 * Similar to spa_export(), this unloads the spa_t without actually removing it
4069 * from the namespace in any way.
4072 spa_reset(char *pool
)
4074 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4079 * ==========================================================================
4080 * Device manipulation
4081 * ==========================================================================
4085 * Add a device to a storage pool.
4088 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4092 vdev_t
*rvd
= spa
->spa_root_vdev
;
4094 nvlist_t
**spares
, **l2cache
;
4095 uint_t nspares
, nl2cache
;
4097 ASSERT(spa_writeable(spa
));
4099 txg
= spa_vdev_enter(spa
);
4101 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4102 VDEV_ALLOC_ADD
)) != 0)
4103 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4105 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4107 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4111 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4115 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4116 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4118 if (vd
->vdev_children
!= 0 &&
4119 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4120 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4123 * We must validate the spares and l2cache devices after checking the
4124 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4126 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4127 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4130 * Transfer each new top-level vdev from vd to rvd.
4132 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4135 * Set the vdev id to the first hole, if one exists.
4137 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4138 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4139 vdev_free(rvd
->vdev_child
[id
]);
4143 tvd
= vd
->vdev_child
[c
];
4144 vdev_remove_child(vd
, tvd
);
4146 vdev_add_child(rvd
, tvd
);
4147 vdev_config_dirty(tvd
);
4151 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4152 ZPOOL_CONFIG_SPARES
);
4153 spa_load_spares(spa
);
4154 spa
->spa_spares
.sav_sync
= B_TRUE
;
4157 if (nl2cache
!= 0) {
4158 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4159 ZPOOL_CONFIG_L2CACHE
);
4160 spa_load_l2cache(spa
);
4161 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4165 * We have to be careful when adding new vdevs to an existing pool.
4166 * If other threads start allocating from these vdevs before we
4167 * sync the config cache, and we lose power, then upon reboot we may
4168 * fail to open the pool because there are DVAs that the config cache
4169 * can't translate. Therefore, we first add the vdevs without
4170 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4171 * and then let spa_config_update() initialize the new metaslabs.
4173 * spa_load() checks for added-but-not-initialized vdevs, so that
4174 * if we lose power at any point in this sequence, the remaining
4175 * steps will be completed the next time we load the pool.
4177 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4179 mutex_enter(&spa_namespace_lock
);
4180 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4181 mutex_exit(&spa_namespace_lock
);
4187 * Attach a device to a mirror. The arguments are the path to any device
4188 * in the mirror, and the nvroot for the new device. If the path specifies
4189 * a device that is not mirrored, we automatically insert the mirror vdev.
4191 * If 'replacing' is specified, the new device is intended to replace the
4192 * existing device; in this case the two devices are made into their own
4193 * mirror using the 'replacing' vdev, which is functionally identical to
4194 * the mirror vdev (it actually reuses all the same ops) but has a few
4195 * extra rules: you can't attach to it after it's been created, and upon
4196 * completion of resilvering, the first disk (the one being replaced)
4197 * is automatically detached.
4200 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4202 uint64_t txg
, dtl_max_txg
;
4203 vdev_t
*rvd
= spa
->spa_root_vdev
;
4204 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4206 char *oldvdpath
, *newvdpath
;
4210 ASSERT(spa_writeable(spa
));
4212 txg
= spa_vdev_enter(spa
);
4214 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4217 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4219 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4220 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4222 pvd
= oldvd
->vdev_parent
;
4224 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4225 VDEV_ALLOC_ATTACH
)) != 0)
4226 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4228 if (newrootvd
->vdev_children
!= 1)
4229 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4231 newvd
= newrootvd
->vdev_child
[0];
4233 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4234 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4236 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4237 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4240 * Spares can't replace logs
4242 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4243 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4247 * For attach, the only allowable parent is a mirror or the root
4250 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4251 pvd
->vdev_ops
!= &vdev_root_ops
)
4252 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4254 pvops
= &vdev_mirror_ops
;
4257 * Active hot spares can only be replaced by inactive hot
4260 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4261 oldvd
->vdev_isspare
&&
4262 !spa_has_spare(spa
, newvd
->vdev_guid
))
4263 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4266 * If the source is a hot spare, and the parent isn't already a
4267 * spare, then we want to create a new hot spare. Otherwise, we
4268 * want to create a replacing vdev. The user is not allowed to
4269 * attach to a spared vdev child unless the 'isspare' state is
4270 * the same (spare replaces spare, non-spare replaces
4273 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4274 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4275 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4276 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4277 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4278 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4281 if (newvd
->vdev_isspare
)
4282 pvops
= &vdev_spare_ops
;
4284 pvops
= &vdev_replacing_ops
;
4288 * Make sure the new device is big enough.
4290 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4291 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4294 * The new device cannot have a higher alignment requirement
4295 * than the top-level vdev.
4297 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4298 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4301 * If this is an in-place replacement, update oldvd's path and devid
4302 * to make it distinguishable from newvd, and unopenable from now on.
4304 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4305 spa_strfree(oldvd
->vdev_path
);
4306 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4308 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4309 newvd
->vdev_path
, "old");
4310 if (oldvd
->vdev_devid
!= NULL
) {
4311 spa_strfree(oldvd
->vdev_devid
);
4312 oldvd
->vdev_devid
= NULL
;
4316 /* mark the device being resilvered */
4317 newvd
->vdev_resilvering
= B_TRUE
;
4320 * If the parent is not a mirror, or if we're replacing, insert the new
4321 * mirror/replacing/spare vdev above oldvd.
4323 if (pvd
->vdev_ops
!= pvops
)
4324 pvd
= vdev_add_parent(oldvd
, pvops
);
4326 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4327 ASSERT(pvd
->vdev_ops
== pvops
);
4328 ASSERT(oldvd
->vdev_parent
== pvd
);
4331 * Extract the new device from its root and add it to pvd.
4333 vdev_remove_child(newrootvd
, newvd
);
4334 newvd
->vdev_id
= pvd
->vdev_children
;
4335 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4336 vdev_add_child(pvd
, newvd
);
4338 tvd
= newvd
->vdev_top
;
4339 ASSERT(pvd
->vdev_top
== tvd
);
4340 ASSERT(tvd
->vdev_parent
== rvd
);
4342 vdev_config_dirty(tvd
);
4345 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4346 * for any dmu_sync-ed blocks. It will propagate upward when
4347 * spa_vdev_exit() calls vdev_dtl_reassess().
4349 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4351 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4352 dtl_max_txg
- TXG_INITIAL
);
4354 if (newvd
->vdev_isspare
) {
4355 spa_spare_activate(newvd
);
4356 spa_event_notify(spa
, newvd
, ESC_ZFS_VDEV_SPARE
);
4359 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4360 newvdpath
= spa_strdup(newvd
->vdev_path
);
4361 newvd_isspare
= newvd
->vdev_isspare
;
4364 * Mark newvd's DTL dirty in this txg.
4366 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4369 * Restart the resilver
4371 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4376 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4378 spa_history_log_internal(spa
, "vdev attach", NULL
,
4379 "%s vdev=%s %s vdev=%s",
4380 replacing
&& newvd_isspare
? "spare in" :
4381 replacing
? "replace" : "attach", newvdpath
,
4382 replacing
? "for" : "to", oldvdpath
);
4384 spa_strfree(oldvdpath
);
4385 spa_strfree(newvdpath
);
4387 if (spa
->spa_bootfs
)
4388 spa_event_notify(spa
, newvd
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
4394 * Detach a device from a mirror or replacing vdev.
4395 * If 'replace_done' is specified, only detach if the parent
4396 * is a replacing vdev.
4399 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4403 vdev_t
*rvd
= spa
->spa_root_vdev
;
4404 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4405 boolean_t unspare
= B_FALSE
;
4406 uint64_t unspare_guid
;
4409 ASSERT(spa_writeable(spa
));
4411 txg
= spa_vdev_enter(spa
);
4413 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4416 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4418 if (!vd
->vdev_ops
->vdev_op_leaf
)
4419 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4421 pvd
= vd
->vdev_parent
;
4424 * If the parent/child relationship is not as expected, don't do it.
4425 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4426 * vdev that's replacing B with C. The user's intent in replacing
4427 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4428 * the replace by detaching C, the expected behavior is to end up
4429 * M(A,B). But suppose that right after deciding to detach C,
4430 * the replacement of B completes. We would have M(A,C), and then
4431 * ask to detach C, which would leave us with just A -- not what
4432 * the user wanted. To prevent this, we make sure that the
4433 * parent/child relationship hasn't changed -- in this example,
4434 * that C's parent is still the replacing vdev R.
4436 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4437 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4440 * Only 'replacing' or 'spare' vdevs can be replaced.
4442 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4443 pvd
->vdev_ops
!= &vdev_spare_ops
)
4444 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4446 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4447 spa_version(spa
) >= SPA_VERSION_SPARES
);
4450 * Only mirror, replacing, and spare vdevs support detach.
4452 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4453 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4454 pvd
->vdev_ops
!= &vdev_spare_ops
)
4455 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4458 * If this device has the only valid copy of some data,
4459 * we cannot safely detach it.
4461 if (vdev_dtl_required(vd
))
4462 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4464 ASSERT(pvd
->vdev_children
>= 2);
4467 * If we are detaching the second disk from a replacing vdev, then
4468 * check to see if we changed the original vdev's path to have "/old"
4469 * at the end in spa_vdev_attach(). If so, undo that change now.
4471 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4472 vd
->vdev_path
!= NULL
) {
4473 size_t len
= strlen(vd
->vdev_path
);
4475 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
4476 cvd
= pvd
->vdev_child
[c
];
4478 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4481 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4482 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4483 spa_strfree(cvd
->vdev_path
);
4484 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4491 * If we are detaching the original disk from a spare, then it implies
4492 * that the spare should become a real disk, and be removed from the
4493 * active spare list for the pool.
4495 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4497 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4501 * Erase the disk labels so the disk can be used for other things.
4502 * This must be done after all other error cases are handled,
4503 * but before we disembowel vd (so we can still do I/O to it).
4504 * But if we can't do it, don't treat the error as fatal --
4505 * it may be that the unwritability of the disk is the reason
4506 * it's being detached!
4508 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4511 * Remove vd from its parent and compact the parent's children.
4513 vdev_remove_child(pvd
, vd
);
4514 vdev_compact_children(pvd
);
4517 * Remember one of the remaining children so we can get tvd below.
4519 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4522 * If we need to remove the remaining child from the list of hot spares,
4523 * do it now, marking the vdev as no longer a spare in the process.
4524 * We must do this before vdev_remove_parent(), because that can
4525 * change the GUID if it creates a new toplevel GUID. For a similar
4526 * reason, we must remove the spare now, in the same txg as the detach;
4527 * otherwise someone could attach a new sibling, change the GUID, and
4528 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4531 ASSERT(cvd
->vdev_isspare
);
4532 spa_spare_remove(cvd
);
4533 unspare_guid
= cvd
->vdev_guid
;
4534 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4535 cvd
->vdev_unspare
= B_TRUE
;
4539 * If the parent mirror/replacing vdev only has one child,
4540 * the parent is no longer needed. Remove it from the tree.
4542 if (pvd
->vdev_children
== 1) {
4543 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4544 cvd
->vdev_unspare
= B_FALSE
;
4545 vdev_remove_parent(cvd
);
4546 cvd
->vdev_resilvering
= B_FALSE
;
4551 * We don't set tvd until now because the parent we just removed
4552 * may have been the previous top-level vdev.
4554 tvd
= cvd
->vdev_top
;
4555 ASSERT(tvd
->vdev_parent
== rvd
);
4558 * Reevaluate the parent vdev state.
4560 vdev_propagate_state(cvd
);
4563 * If the 'autoexpand' property is set on the pool then automatically
4564 * try to expand the size of the pool. For example if the device we
4565 * just detached was smaller than the others, it may be possible to
4566 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4567 * first so that we can obtain the updated sizes of the leaf vdevs.
4569 if (spa
->spa_autoexpand
) {
4571 vdev_expand(tvd
, txg
);
4574 vdev_config_dirty(tvd
);
4577 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4578 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4579 * But first make sure we're not on any *other* txg's DTL list, to
4580 * prevent vd from being accessed after it's freed.
4582 vdpath
= spa_strdup(vd
->vdev_path
);
4583 for (int t
= 0; t
< TXG_SIZE
; t
++)
4584 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4585 vd
->vdev_detached
= B_TRUE
;
4586 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4588 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_REMOVE
);
4590 /* hang on to the spa before we release the lock */
4591 spa_open_ref(spa
, FTAG
);
4593 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4595 spa_history_log_internal(spa
, "detach", NULL
,
4597 spa_strfree(vdpath
);
4600 * If this was the removal of the original device in a hot spare vdev,
4601 * then we want to go through and remove the device from the hot spare
4602 * list of every other pool.
4605 spa_t
*altspa
= NULL
;
4607 mutex_enter(&spa_namespace_lock
);
4608 while ((altspa
= spa_next(altspa
)) != NULL
) {
4609 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4613 spa_open_ref(altspa
, FTAG
);
4614 mutex_exit(&spa_namespace_lock
);
4615 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4616 mutex_enter(&spa_namespace_lock
);
4617 spa_close(altspa
, FTAG
);
4619 mutex_exit(&spa_namespace_lock
);
4621 /* search the rest of the vdevs for spares to remove */
4622 spa_vdev_resilver_done(spa
);
4625 /* all done with the spa; OK to release */
4626 mutex_enter(&spa_namespace_lock
);
4627 spa_close(spa
, FTAG
);
4628 mutex_exit(&spa_namespace_lock
);
4634 * Split a set of devices from their mirrors, and create a new pool from them.
4637 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
4638 nvlist_t
*props
, boolean_t exp
)
4641 uint64_t txg
, *glist
;
4643 uint_t c
, children
, lastlog
;
4644 nvlist_t
**child
, *nvl
, *tmp
;
4646 char *altroot
= NULL
;
4647 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
4648 boolean_t activate_slog
;
4650 ASSERT(spa_writeable(spa
));
4652 txg
= spa_vdev_enter(spa
);
4654 /* clear the log and flush everything up to now */
4655 activate_slog
= spa_passivate_log(spa
);
4656 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4657 error
= spa_offline_log(spa
);
4658 txg
= spa_vdev_config_enter(spa
);
4661 spa_activate_log(spa
);
4664 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4666 /* check new spa name before going any further */
4667 if (spa_lookup(newname
) != NULL
)
4668 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
4671 * scan through all the children to ensure they're all mirrors
4673 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
4674 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
4676 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4678 /* first, check to ensure we've got the right child count */
4679 rvd
= spa
->spa_root_vdev
;
4681 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
4682 vdev_t
*vd
= rvd
->vdev_child
[c
];
4684 /* don't count the holes & logs as children */
4685 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
4693 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
4694 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4696 /* next, ensure no spare or cache devices are part of the split */
4697 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
4698 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
4699 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4701 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
4702 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
4704 /* then, loop over each vdev and validate it */
4705 for (c
= 0; c
< children
; c
++) {
4706 uint64_t is_hole
= 0;
4708 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
4712 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
4713 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
4721 /* which disk is going to be split? */
4722 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
4728 /* look it up in the spa */
4729 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
4730 if (vml
[c
] == NULL
) {
4735 /* make sure there's nothing stopping the split */
4736 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
4737 vml
[c
]->vdev_islog
||
4738 vml
[c
]->vdev_ishole
||
4739 vml
[c
]->vdev_isspare
||
4740 vml
[c
]->vdev_isl2cache
||
4741 !vdev_writeable(vml
[c
]) ||
4742 vml
[c
]->vdev_children
!= 0 ||
4743 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
4744 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
4749 if (vdev_dtl_required(vml
[c
])) {
4754 /* we need certain info from the top level */
4755 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
4756 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
4757 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
4758 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
4759 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
4760 vml
[c
]->vdev_top
->vdev_asize
) == 0);
4761 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
4762 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
4766 kmem_free(vml
, children
* sizeof (vdev_t
*));
4767 kmem_free(glist
, children
* sizeof (uint64_t));
4768 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4771 /* stop writers from using the disks */
4772 for (c
= 0; c
< children
; c
++) {
4774 vml
[c
]->vdev_offline
= B_TRUE
;
4776 vdev_reopen(spa
->spa_root_vdev
);
4779 * Temporarily record the splitting vdevs in the spa config. This
4780 * will disappear once the config is regenerated.
4782 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4783 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
4784 glist
, children
) == 0);
4785 kmem_free(glist
, children
* sizeof (uint64_t));
4787 mutex_enter(&spa
->spa_props_lock
);
4788 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
4790 mutex_exit(&spa
->spa_props_lock
);
4791 spa
->spa_config_splitting
= nvl
;
4792 vdev_config_dirty(spa
->spa_root_vdev
);
4794 /* configure and create the new pool */
4795 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
4796 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4797 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
4798 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
4799 spa_version(spa
)) == 0);
4800 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
4801 spa
->spa_config_txg
) == 0);
4802 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
4803 spa_generate_guid(NULL
)) == 0);
4804 (void) nvlist_lookup_string(props
,
4805 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4807 /* add the new pool to the namespace */
4808 newspa
= spa_add(newname
, config
, altroot
);
4809 newspa
->spa_config_txg
= spa
->spa_config_txg
;
4810 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
4812 /* release the spa config lock, retaining the namespace lock */
4813 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4815 if (zio_injection_enabled
)
4816 zio_handle_panic_injection(spa
, FTAG
, 1);
4818 spa_activate(newspa
, spa_mode_global
);
4819 spa_async_suspend(newspa
);
4821 /* create the new pool from the disks of the original pool */
4822 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
4826 /* if that worked, generate a real config for the new pool */
4827 if (newspa
->spa_root_vdev
!= NULL
) {
4828 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
4829 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4830 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
4831 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
4832 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
4837 if (props
!= NULL
) {
4838 spa_configfile_set(newspa
, props
, B_FALSE
);
4839 error
= spa_prop_set(newspa
, props
);
4844 /* flush everything */
4845 txg
= spa_vdev_config_enter(newspa
);
4846 vdev_config_dirty(newspa
->spa_root_vdev
);
4847 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
4849 if (zio_injection_enabled
)
4850 zio_handle_panic_injection(spa
, FTAG
, 2);
4852 spa_async_resume(newspa
);
4854 /* finally, update the original pool's config */
4855 txg
= spa_vdev_config_enter(spa
);
4856 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
4857 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4860 for (c
= 0; c
< children
; c
++) {
4861 if (vml
[c
] != NULL
) {
4864 spa_history_log_internal(spa
, "detach", tx
,
4865 "vdev=%s", vml
[c
]->vdev_path
);
4869 vdev_config_dirty(spa
->spa_root_vdev
);
4870 spa
->spa_config_splitting
= NULL
;
4874 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
4876 if (zio_injection_enabled
)
4877 zio_handle_panic_injection(spa
, FTAG
, 3);
4879 /* split is complete; log a history record */
4880 spa_history_log_internal(newspa
, "split", NULL
,
4881 "from pool %s", spa_name(spa
));
4883 kmem_free(vml
, children
* sizeof (vdev_t
*));
4885 /* if we're not going to mount the filesystems in userland, export */
4887 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
4894 spa_deactivate(newspa
);
4897 txg
= spa_vdev_config_enter(spa
);
4899 /* re-online all offlined disks */
4900 for (c
= 0; c
< children
; c
++) {
4902 vml
[c
]->vdev_offline
= B_FALSE
;
4904 vdev_reopen(spa
->spa_root_vdev
);
4906 nvlist_free(spa
->spa_config_splitting
);
4907 spa
->spa_config_splitting
= NULL
;
4908 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
4910 kmem_free(vml
, children
* sizeof (vdev_t
*));
4915 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
4917 for (int i
= 0; i
< count
; i
++) {
4920 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
4923 if (guid
== target_guid
)
4931 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
4932 nvlist_t
*dev_to_remove
)
4934 nvlist_t
**newdev
= NULL
;
4937 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
4939 for (int i
= 0, j
= 0; i
< count
; i
++) {
4940 if (dev
[i
] == dev_to_remove
)
4942 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
4945 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4946 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
4948 for (int i
= 0; i
< count
- 1; i
++)
4949 nvlist_free(newdev
[i
]);
4952 kmem_free(newdev
, (count
- 1) * sizeof (void *));
4956 * Evacuate the device.
4959 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
4964 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4965 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
4966 ASSERT(vd
== vd
->vdev_top
);
4969 * Evacuate the device. We don't hold the config lock as writer
4970 * since we need to do I/O but we do keep the
4971 * spa_namespace_lock held. Once this completes the device
4972 * should no longer have any blocks allocated on it.
4974 if (vd
->vdev_islog
) {
4975 if (vd
->vdev_stat
.vs_alloc
!= 0)
4976 error
= spa_offline_log(spa
);
4985 * The evacuation succeeded. Remove any remaining MOS metadata
4986 * associated with this vdev, and wait for these changes to sync.
4988 ASSERT3U(vd
->vdev_stat
.vs_alloc
, ==, 0);
4989 txg
= spa_vdev_config_enter(spa
);
4990 vd
->vdev_removing
= B_TRUE
;
4991 vdev_dirty(vd
, 0, NULL
, txg
);
4992 vdev_config_dirty(vd
);
4993 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4999 * Complete the removal by cleaning up the namespace.
5002 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5004 vdev_t
*rvd
= spa
->spa_root_vdev
;
5005 uint64_t id
= vd
->vdev_id
;
5006 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5008 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5009 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5010 ASSERT(vd
== vd
->vdev_top
);
5013 * Only remove any devices which are empty.
5015 if (vd
->vdev_stat
.vs_alloc
!= 0)
5018 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5020 if (list_link_active(&vd
->vdev_state_dirty_node
))
5021 vdev_state_clean(vd
);
5022 if (list_link_active(&vd
->vdev_config_dirty_node
))
5023 vdev_config_clean(vd
);
5028 vdev_compact_children(rvd
);
5030 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5031 vdev_add_child(rvd
, vd
);
5033 vdev_config_dirty(rvd
);
5036 * Reassess the health of our root vdev.
5042 * Remove a device from the pool -
5044 * Removing a device from the vdev namespace requires several steps
5045 * and can take a significant amount of time. As a result we use
5046 * the spa_vdev_config_[enter/exit] functions which allow us to
5047 * grab and release the spa_config_lock while still holding the namespace
5048 * lock. During each step the configuration is synced out.
5052 * Remove a device from the pool. Currently, this supports removing only hot
5053 * spares, slogs, and level 2 ARC devices.
5056 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5059 metaslab_group_t
*mg
;
5060 nvlist_t
**spares
, **l2cache
, *nv
;
5062 uint_t nspares
, nl2cache
;
5064 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5066 ASSERT(spa_writeable(spa
));
5069 txg
= spa_vdev_enter(spa
);
5071 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5073 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5074 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5075 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5076 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5078 * Only remove the hot spare if it's not currently in use
5081 if (vd
== NULL
|| unspare
) {
5082 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5083 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5084 spa_load_spares(spa
);
5085 spa
->spa_spares
.sav_sync
= B_TRUE
;
5089 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5090 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5091 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5092 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5094 * Cache devices can always be removed.
5096 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5097 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5098 spa_load_l2cache(spa
);
5099 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5100 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5102 ASSERT(vd
== vd
->vdev_top
);
5105 * XXX - Once we have bp-rewrite this should
5106 * become the common case.
5112 * Stop allocating from this vdev.
5114 metaslab_group_passivate(mg
);
5117 * Wait for the youngest allocations and frees to sync,
5118 * and then wait for the deferral of those frees to finish.
5120 spa_vdev_config_exit(spa
, NULL
,
5121 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5124 * Attempt to evacuate the vdev.
5126 error
= spa_vdev_remove_evacuate(spa
, vd
);
5128 txg
= spa_vdev_config_enter(spa
);
5131 * If we couldn't evacuate the vdev, unwind.
5134 metaslab_group_activate(mg
);
5135 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5139 * Clean up the vdev namespace.
5141 spa_vdev_remove_from_namespace(spa
, vd
);
5143 } else if (vd
!= NULL
) {
5145 * Normal vdevs cannot be removed (yet).
5150 * There is no vdev of any kind with the specified guid.
5156 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5162 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5163 * current spared, so we can detach it.
5166 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5168 vdev_t
*newvd
, *oldvd
;
5170 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5171 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5177 * Check for a completed replacement. We always consider the first
5178 * vdev in the list to be the oldest vdev, and the last one to be
5179 * the newest (see spa_vdev_attach() for how that works). In
5180 * the case where the newest vdev is faulted, we will not automatically
5181 * remove it after a resilver completes. This is OK as it will require
5182 * user intervention to determine which disk the admin wishes to keep.
5184 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5185 ASSERT(vd
->vdev_children
> 1);
5187 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5188 oldvd
= vd
->vdev_child
[0];
5190 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5191 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5192 !vdev_dtl_required(oldvd
))
5197 * Check for a completed resilver with the 'unspare' flag set.
5199 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5200 vdev_t
*first
= vd
->vdev_child
[0];
5201 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5203 if (last
->vdev_unspare
) {
5206 } else if (first
->vdev_unspare
) {
5213 if (oldvd
!= NULL
&&
5214 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5215 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5216 !vdev_dtl_required(oldvd
))
5220 * If there are more than two spares attached to a disk,
5221 * and those spares are not required, then we want to
5222 * attempt to free them up now so that they can be used
5223 * by other pools. Once we're back down to a single
5224 * disk+spare, we stop removing them.
5226 if (vd
->vdev_children
> 2) {
5227 newvd
= vd
->vdev_child
[1];
5229 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5230 vdev_dtl_empty(last
, DTL_MISSING
) &&
5231 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5232 !vdev_dtl_required(newvd
))
5241 spa_vdev_resilver_done(spa_t
*spa
)
5243 vdev_t
*vd
, *pvd
, *ppvd
;
5244 uint64_t guid
, sguid
, pguid
, ppguid
;
5246 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5248 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5249 pvd
= vd
->vdev_parent
;
5250 ppvd
= pvd
->vdev_parent
;
5251 guid
= vd
->vdev_guid
;
5252 pguid
= pvd
->vdev_guid
;
5253 ppguid
= ppvd
->vdev_guid
;
5256 * If we have just finished replacing a hot spared device, then
5257 * we need to detach the parent's first child (the original hot
5260 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5261 ppvd
->vdev_children
== 2) {
5262 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5263 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5265 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5266 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5268 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5270 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5273 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5277 * Update the stored path or FRU for this vdev.
5280 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5284 boolean_t sync
= B_FALSE
;
5286 ASSERT(spa_writeable(spa
));
5288 spa_vdev_state_enter(spa
, SCL_ALL
);
5290 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5291 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5293 if (!vd
->vdev_ops
->vdev_op_leaf
)
5294 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5297 if (strcmp(value
, vd
->vdev_path
) != 0) {
5298 spa_strfree(vd
->vdev_path
);
5299 vd
->vdev_path
= spa_strdup(value
);
5303 if (vd
->vdev_fru
== NULL
) {
5304 vd
->vdev_fru
= spa_strdup(value
);
5306 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5307 spa_strfree(vd
->vdev_fru
);
5308 vd
->vdev_fru
= spa_strdup(value
);
5313 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5317 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5319 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5323 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5325 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5329 * ==========================================================================
5331 * ==========================================================================
5335 spa_scan_stop(spa_t
*spa
)
5337 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5338 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5340 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5344 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5346 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5348 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5352 * If a resilver was requested, but there is no DTL on a
5353 * writeable leaf device, we have nothing to do.
5355 if (func
== POOL_SCAN_RESILVER
&&
5356 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5357 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5361 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5365 * ==========================================================================
5366 * SPA async task processing
5367 * ==========================================================================
5371 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5373 if (vd
->vdev_remove_wanted
) {
5374 vd
->vdev_remove_wanted
= B_FALSE
;
5375 vd
->vdev_delayed_close
= B_FALSE
;
5376 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5379 * We want to clear the stats, but we don't want to do a full
5380 * vdev_clear() as that will cause us to throw away
5381 * degraded/faulted state as well as attempt to reopen the
5382 * device, all of which is a waste.
5384 vd
->vdev_stat
.vs_read_errors
= 0;
5385 vd
->vdev_stat
.vs_write_errors
= 0;
5386 vd
->vdev_stat
.vs_checksum_errors
= 0;
5388 vdev_state_dirty(vd
->vdev_top
);
5391 for (int c
= 0; c
< vd
->vdev_children
; c
++)
5392 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5396 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5398 if (vd
->vdev_probe_wanted
) {
5399 vd
->vdev_probe_wanted
= B_FALSE
;
5400 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5403 for (int c
= 0; c
< vd
->vdev_children
; c
++)
5404 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5408 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5414 if (!spa
->spa_autoexpand
)
5417 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5418 vdev_t
*cvd
= vd
->vdev_child
[c
];
5419 spa_async_autoexpand(spa
, cvd
);
5422 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5425 physpath
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
5426 (void) snprintf(physpath
, MAXPATHLEN
, "/devices%s", vd
->vdev_physpath
);
5428 VERIFY(nvlist_alloc(&attr
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5429 VERIFY(nvlist_add_string(attr
, DEV_PHYS_PATH
, physpath
) == 0);
5431 (void) ddi_log_sysevent(zfs_dip
, SUNW_VENDOR
, EC_DEV_STATUS
,
5432 ESC_DEV_DLE
, attr
, &eid
, DDI_SLEEP
);
5435 kmem_free(physpath
, MAXPATHLEN
);
5439 spa_async_thread(spa_t
*spa
)
5443 ASSERT(spa
->spa_sync_on
);
5445 mutex_enter(&spa
->spa_async_lock
);
5446 tasks
= spa
->spa_async_tasks
;
5447 spa
->spa_async_tasks
= 0;
5448 mutex_exit(&spa
->spa_async_lock
);
5451 * See if the config needs to be updated.
5453 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5454 uint64_t old_space
, new_space
;
5456 mutex_enter(&spa_namespace_lock
);
5457 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5458 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5459 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5460 mutex_exit(&spa_namespace_lock
);
5463 * If the pool grew as a result of the config update,
5464 * then log an internal history event.
5466 if (new_space
!= old_space
) {
5467 spa_history_log_internal(spa
, "vdev online", NULL
,
5468 "pool '%s' size: %llu(+%llu)",
5469 spa_name(spa
), new_space
, new_space
- old_space
);
5474 * See if any devices need to be marked REMOVED.
5476 if (tasks
& SPA_ASYNC_REMOVE
) {
5477 spa_vdev_state_enter(spa
, SCL_NONE
);
5478 spa_async_remove(spa
, spa
->spa_root_vdev
);
5479 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5480 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5481 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5482 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5483 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5486 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5487 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5488 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5489 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5493 * See if any devices need to be probed.
5495 if (tasks
& SPA_ASYNC_PROBE
) {
5496 spa_vdev_state_enter(spa
, SCL_NONE
);
5497 spa_async_probe(spa
, spa
->spa_root_vdev
);
5498 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5502 * If any devices are done replacing, detach them.
5504 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5505 spa_vdev_resilver_done(spa
);
5508 * Kick off a resilver.
5510 if (tasks
& SPA_ASYNC_RESILVER
)
5511 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5514 * Let the world know that we're done.
5516 mutex_enter(&spa
->spa_async_lock
);
5517 spa
->spa_async_thread
= NULL
;
5518 cv_broadcast(&spa
->spa_async_cv
);
5519 mutex_exit(&spa
->spa_async_lock
);
5524 spa_async_suspend(spa_t
*spa
)
5526 mutex_enter(&spa
->spa_async_lock
);
5527 spa
->spa_async_suspended
++;
5528 while (spa
->spa_async_thread
!= NULL
)
5529 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5530 mutex_exit(&spa
->spa_async_lock
);
5534 spa_async_resume(spa_t
*spa
)
5536 mutex_enter(&spa
->spa_async_lock
);
5537 ASSERT(spa
->spa_async_suspended
!= 0);
5538 spa
->spa_async_suspended
--;
5539 mutex_exit(&spa
->spa_async_lock
);
5543 spa_async_dispatch(spa_t
*spa
)
5545 mutex_enter(&spa
->spa_async_lock
);
5546 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5547 spa
->spa_async_thread
== NULL
&&
5548 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5549 spa
->spa_async_thread
= thread_create(NULL
, 0,
5550 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5551 mutex_exit(&spa
->spa_async_lock
);
5555 spa_async_request(spa_t
*spa
, int task
)
5557 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5558 mutex_enter(&spa
->spa_async_lock
);
5559 spa
->spa_async_tasks
|= task
;
5560 mutex_exit(&spa
->spa_async_lock
);
5564 * ==========================================================================
5565 * SPA syncing routines
5566 * ==========================================================================
5570 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5573 bpobj_enqueue(bpo
, bp
, tx
);
5578 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5582 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5588 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5590 char *packed
= NULL
;
5595 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5598 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5599 * information. This avoids the dbuf_will_dirty() path and
5600 * saves us a pre-read to get data we don't actually care about.
5602 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
5603 packed
= kmem_alloc(bufsize
, KM_SLEEP
);
5605 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
5607 bzero(packed
+ nvsize
, bufsize
- nvsize
);
5609 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
5611 kmem_free(packed
, bufsize
);
5613 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
5614 dmu_buf_will_dirty(db
, tx
);
5615 *(uint64_t *)db
->db_data
= nvsize
;
5616 dmu_buf_rele(db
, FTAG
);
5620 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
5621 const char *config
, const char *entry
)
5631 * Update the MOS nvlist describing the list of available devices.
5632 * spa_validate_aux() will have already made sure this nvlist is
5633 * valid and the vdevs are labeled appropriately.
5635 if (sav
->sav_object
== 0) {
5636 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
5637 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
5638 sizeof (uint64_t), tx
);
5639 VERIFY(zap_update(spa
->spa_meta_objset
,
5640 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
5641 &sav
->sav_object
, tx
) == 0);
5644 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5645 if (sav
->sav_count
== 0) {
5646 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
5648 list
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
5649 for (i
= 0; i
< sav
->sav_count
; i
++)
5650 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
5651 B_FALSE
, VDEV_CONFIG_L2CACHE
);
5652 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
5653 sav
->sav_count
) == 0);
5654 for (i
= 0; i
< sav
->sav_count
; i
++)
5655 nvlist_free(list
[i
]);
5656 kmem_free(list
, sav
->sav_count
* sizeof (void *));
5659 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
5660 nvlist_free(nvroot
);
5662 sav
->sav_sync
= B_FALSE
;
5666 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
5670 if (list_is_empty(&spa
->spa_config_dirty_list
))
5673 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5675 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
5676 dmu_tx_get_txg(tx
), B_FALSE
);
5678 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5680 if (spa
->spa_config_syncing
)
5681 nvlist_free(spa
->spa_config_syncing
);
5682 spa
->spa_config_syncing
= config
;
5684 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
5688 spa_sync_version(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
5691 uint64_t version
= *(uint64_t *)arg2
;
5694 * Setting the version is special cased when first creating the pool.
5696 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
5698 ASSERT(version
<= SPA_VERSION
);
5699 ASSERT(version
>= spa_version(spa
));
5701 spa
->spa_uberblock
.ub_version
= version
;
5702 vdev_config_dirty(spa
->spa_root_vdev
);
5703 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
5707 * Set zpool properties.
5710 spa_sync_props(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
5713 objset_t
*mos
= spa
->spa_meta_objset
;
5714 nvlist_t
*nvp
= arg2
;
5715 nvpair_t
*elem
= NULL
;
5717 mutex_enter(&spa
->spa_props_lock
);
5719 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
5721 char *strval
, *fname
;
5723 const char *propname
;
5724 zprop_type_t proptype
;
5725 zfeature_info_t
*feature
;
5727 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
5730 * We checked this earlier in spa_prop_validate().
5732 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
5734 fname
= strchr(nvpair_name(elem
), '@') + 1;
5735 VERIFY3U(0, ==, zfeature_lookup_name(fname
, &feature
));
5737 spa_feature_enable(spa
, feature
, tx
);
5738 spa_history_log_internal(spa
, "set", tx
,
5739 "%s=enabled", nvpair_name(elem
));
5742 case ZPOOL_PROP_VERSION
:
5743 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
5745 * The version is synced seperatly before other
5746 * properties and should be correct by now.
5748 ASSERT3U(spa_version(spa
), >=, intval
);
5751 case ZPOOL_PROP_ALTROOT
:
5753 * 'altroot' is a non-persistent property. It should
5754 * have been set temporarily at creation or import time.
5756 ASSERT(spa
->spa_root
!= NULL
);
5759 case ZPOOL_PROP_READONLY
:
5760 case ZPOOL_PROP_CACHEFILE
:
5762 * 'readonly' and 'cachefile' are also non-persisitent
5766 case ZPOOL_PROP_COMMENT
:
5767 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
5768 if (spa
->spa_comment
!= NULL
)
5769 spa_strfree(spa
->spa_comment
);
5770 spa
->spa_comment
= spa_strdup(strval
);
5772 * We need to dirty the configuration on all the vdevs
5773 * so that their labels get updated. It's unnecessary
5774 * to do this for pool creation since the vdev's
5775 * configuratoin has already been dirtied.
5777 if (tx
->tx_txg
!= TXG_INITIAL
)
5778 vdev_config_dirty(spa
->spa_root_vdev
);
5779 spa_history_log_internal(spa
, "set", tx
,
5780 "%s=%s", nvpair_name(elem
), strval
);
5784 * Set pool property values in the poolprops mos object.
5786 if (spa
->spa_pool_props_object
== 0) {
5787 spa
->spa_pool_props_object
=
5788 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
5789 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
5793 /* normalize the property name */
5794 propname
= zpool_prop_to_name(prop
);
5795 proptype
= zpool_prop_get_type(prop
);
5797 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
5798 ASSERT(proptype
== PROP_TYPE_STRING
);
5799 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
5800 VERIFY(zap_update(mos
,
5801 spa
->spa_pool_props_object
, propname
,
5802 1, strlen(strval
) + 1, strval
, tx
) == 0);
5803 spa_history_log_internal(spa
, "set", tx
,
5804 "%s=%s", nvpair_name(elem
), strval
);
5805 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
5806 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
5808 if (proptype
== PROP_TYPE_INDEX
) {
5810 VERIFY(zpool_prop_index_to_string(
5811 prop
, intval
, &unused
) == 0);
5813 VERIFY(zap_update(mos
,
5814 spa
->spa_pool_props_object
, propname
,
5815 8, 1, &intval
, tx
) == 0);
5816 spa_history_log_internal(spa
, "set", tx
,
5817 "%s=%lld", nvpair_name(elem
), intval
);
5819 ASSERT(0); /* not allowed */
5823 case ZPOOL_PROP_DELEGATION
:
5824 spa
->spa_delegation
= intval
;
5826 case ZPOOL_PROP_BOOTFS
:
5827 spa
->spa_bootfs
= intval
;
5829 case ZPOOL_PROP_FAILUREMODE
:
5830 spa
->spa_failmode
= intval
;
5832 case ZPOOL_PROP_AUTOEXPAND
:
5833 spa
->spa_autoexpand
= intval
;
5834 if (tx
->tx_txg
!= TXG_INITIAL
)
5835 spa_async_request(spa
,
5836 SPA_ASYNC_AUTOEXPAND
);
5838 case ZPOOL_PROP_DEDUPDITTO
:
5839 spa
->spa_dedup_ditto
= intval
;
5848 mutex_exit(&spa
->spa_props_lock
);
5852 * Perform one-time upgrade on-disk changes. spa_version() does not
5853 * reflect the new version this txg, so there must be no changes this
5854 * txg to anything that the upgrade code depends on after it executes.
5855 * Therefore this must be called after dsl_pool_sync() does the sync
5859 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
5861 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
5863 ASSERT(spa
->spa_sync_pass
== 1);
5865 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
5866 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
5867 dsl_pool_create_origin(dp
, tx
);
5869 /* Keeping the origin open increases spa_minref */
5870 spa
->spa_minref
+= 3;
5873 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
5874 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
5875 dsl_pool_upgrade_clones(dp
, tx
);
5878 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
5879 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
5880 dsl_pool_upgrade_dir_clones(dp
, tx
);
5882 /* Keeping the freedir open increases spa_minref */
5883 spa
->spa_minref
+= 3;
5886 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
5887 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
5888 spa_feature_create_zap_objects(spa
, tx
);
5893 * Sync the specified transaction group. New blocks may be dirtied as
5894 * part of the process, so we iterate until it converges.
5897 spa_sync(spa_t
*spa
, uint64_t txg
)
5899 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
5900 objset_t
*mos
= spa
->spa_meta_objset
;
5901 bpobj_t
*defer_bpo
= &spa
->spa_deferred_bpobj
;
5902 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
5903 vdev_t
*rvd
= spa
->spa_root_vdev
;
5908 VERIFY(spa_writeable(spa
));
5911 * Lock out configuration changes.
5913 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5915 spa
->spa_syncing_txg
= txg
;
5916 spa
->spa_sync_pass
= 0;
5919 * If there are any pending vdev state changes, convert them
5920 * into config changes that go out with this transaction group.
5922 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5923 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
5925 * We need the write lock here because, for aux vdevs,
5926 * calling vdev_config_dirty() modifies sav_config.
5927 * This is ugly and will become unnecessary when we
5928 * eliminate the aux vdev wart by integrating all vdevs
5929 * into the root vdev tree.
5931 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
5932 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
5933 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
5934 vdev_state_clean(vd
);
5935 vdev_config_dirty(vd
);
5937 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
5938 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
5940 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5942 tx
= dmu_tx_create_assigned(dp
, txg
);
5945 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5946 * set spa_deflate if we have no raid-z vdevs.
5948 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
5949 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
5952 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
5953 vd
= rvd
->vdev_child
[i
];
5954 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
5957 if (i
== rvd
->vdev_children
) {
5958 spa
->spa_deflate
= TRUE
;
5959 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
5960 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
5961 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
5966 * If anything has changed in this txg, or if someone is waiting
5967 * for this txg to sync (eg, spa_vdev_remove()), push the
5968 * deferred frees from the previous txg. If not, leave them
5969 * alone so that we don't generate work on an otherwise idle
5972 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
5973 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
5974 !txg_list_empty(&dp
->dp_sync_tasks
, txg
) ||
5975 ((dsl_scan_active(dp
->dp_scan
) ||
5976 txg_sync_waiting(dp
)) && !spa_shutting_down(spa
))) {
5977 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5978 VERIFY3U(bpobj_iterate(defer_bpo
,
5979 spa_free_sync_cb
, zio
, tx
), ==, 0);
5980 VERIFY3U(zio_wait(zio
), ==, 0);
5984 * Iterate to convergence.
5987 int pass
= ++spa
->spa_sync_pass
;
5989 spa_sync_config_object(spa
, tx
);
5990 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
5991 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
5992 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
5993 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
5994 spa_errlog_sync(spa
, txg
);
5995 dsl_pool_sync(dp
, txg
);
5997 if (pass
<= SYNC_PASS_DEFERRED_FREE
) {
5998 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5999 bplist_iterate(free_bpl
, spa_free_sync_cb
,
6001 VERIFY(zio_wait(zio
) == 0);
6003 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6008 dsl_scan_sync(dp
, tx
);
6010 while (vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
6014 spa_sync_upgrades(spa
, tx
);
6016 } while (dmu_objset_is_dirty(mos
, txg
));
6019 * Rewrite the vdev configuration (which includes the uberblock)
6020 * to commit the transaction group.
6022 * If there are no dirty vdevs, we sync the uberblock to a few
6023 * random top-level vdevs that are known to be visible in the
6024 * config cache (see spa_vdev_add() for a complete description).
6025 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6029 * We hold SCL_STATE to prevent vdev open/close/etc.
6030 * while we're attempting to write the vdev labels.
6032 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6034 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6035 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6037 int children
= rvd
->vdev_children
;
6038 int c0
= spa_get_random(children
);
6040 for (int c
= 0; c
< children
; c
++) {
6041 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6042 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6044 svd
[svdcount
++] = vd
;
6045 if (svdcount
== SPA_DVAS_PER_BP
)
6048 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
6050 error
= vdev_config_sync(svd
, svdcount
, txg
,
6053 error
= vdev_config_sync(rvd
->vdev_child
,
6054 rvd
->vdev_children
, txg
, B_FALSE
);
6056 error
= vdev_config_sync(rvd
->vdev_child
,
6057 rvd
->vdev_children
, txg
, B_TRUE
);
6060 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6064 zio_suspend(spa
, NULL
);
6065 zio_resume_wait(spa
);
6070 * Clear the dirty config list.
6072 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6073 vdev_config_clean(vd
);
6076 * Now that the new config has synced transactionally,
6077 * let it become visible to the config cache.
6079 if (spa
->spa_config_syncing
!= NULL
) {
6080 spa_config_set(spa
, spa
->spa_config_syncing
);
6081 spa
->spa_config_txg
= txg
;
6082 spa
->spa_config_syncing
= NULL
;
6085 spa
->spa_ubsync
= spa
->spa_uberblock
;
6087 dsl_pool_sync_done(dp
, txg
);
6090 * Update usable space statistics.
6092 while (vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
6093 vdev_sync_done(vd
, txg
);
6095 spa_update_dspace(spa
);
6098 * It had better be the case that we didn't dirty anything
6099 * since vdev_config_sync().
6101 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6102 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6103 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6105 spa
->spa_sync_pass
= 0;
6107 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6109 spa_handle_ignored_writes(spa
);
6112 * If any async tasks have been requested, kick them off.
6114 spa_async_dispatch(spa
);
6118 * Sync all pools. We don't want to hold the namespace lock across these
6119 * operations, so we take a reference on the spa_t and drop the lock during the
6123 spa_sync_allpools(void)
6126 mutex_enter(&spa_namespace_lock
);
6127 while ((spa
= spa_next(spa
)) != NULL
) {
6128 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6129 !spa_writeable(spa
) || spa_suspended(spa
))
6131 spa_open_ref(spa
, FTAG
);
6132 mutex_exit(&spa_namespace_lock
);
6133 txg_wait_synced(spa_get_dsl(spa
), 0);
6134 mutex_enter(&spa_namespace_lock
);
6135 spa_close(spa
, FTAG
);
6137 mutex_exit(&spa_namespace_lock
);
6141 * ==========================================================================
6142 * Miscellaneous routines
6143 * ==========================================================================
6147 * Remove all pools in the system.
6155 * Remove all cached state. All pools should be closed now,
6156 * so every spa in the AVL tree should be unreferenced.
6158 mutex_enter(&spa_namespace_lock
);
6159 while ((spa
= spa_next(NULL
)) != NULL
) {
6161 * Stop async tasks. The async thread may need to detach
6162 * a device that's been replaced, which requires grabbing
6163 * spa_namespace_lock, so we must drop it here.
6165 spa_open_ref(spa
, FTAG
);
6166 mutex_exit(&spa_namespace_lock
);
6167 spa_async_suspend(spa
);
6168 mutex_enter(&spa_namespace_lock
);
6169 spa_close(spa
, FTAG
);
6171 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6173 spa_deactivate(spa
);
6177 mutex_exit(&spa_namespace_lock
);
6181 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6186 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6190 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6191 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6192 if (vd
->vdev_guid
== guid
)
6196 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6197 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6198 if (vd
->vdev_guid
== guid
)
6207 spa_upgrade(spa_t
*spa
, uint64_t version
)
6209 ASSERT(spa_writeable(spa
));
6211 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6214 * This should only be called for a non-faulted pool, and since a
6215 * future version would result in an unopenable pool, this shouldn't be
6218 ASSERT(spa
->spa_uberblock
.ub_version
<= SPA_VERSION
);
6219 ASSERT(version
>= spa
->spa_uberblock
.ub_version
);
6221 spa
->spa_uberblock
.ub_version
= version
;
6222 vdev_config_dirty(spa
->spa_root_vdev
);
6224 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6226 txg_wait_synced(spa_get_dsl(spa
), 0);
6230 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6234 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6236 for (i
= 0; i
< sav
->sav_count
; i
++)
6237 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6240 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6241 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6242 &spareguid
) == 0 && spareguid
== guid
)
6250 * Check if a pool has an active shared spare device.
6251 * Note: reference count of an active spare is 2, as a spare and as a replace
6254 spa_has_active_shared_spare(spa_t
*spa
)
6258 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6260 for (i
= 0; i
< sav
->sav_count
; i
++) {
6261 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6262 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6271 * Post a sysevent corresponding to the given event. The 'name' must be one of
6272 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6273 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6274 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6275 * or zdb as real changes.
6278 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6282 sysevent_attr_list_t
*attr
= NULL
;
6283 sysevent_value_t value
;
6286 ev
= sysevent_alloc(EC_ZFS
, (char *)name
, SUNW_KERN_PUB
"zfs",
6289 value
.value_type
= SE_DATA_TYPE_STRING
;
6290 value
.value
.sv_string
= spa_name(spa
);
6291 if (sysevent_add_attr(&attr
, ZFS_EV_POOL_NAME
, &value
, SE_SLEEP
) != 0)
6294 value
.value_type
= SE_DATA_TYPE_UINT64
;
6295 value
.value
.sv_uint64
= spa_guid(spa
);
6296 if (sysevent_add_attr(&attr
, ZFS_EV_POOL_GUID
, &value
, SE_SLEEP
) != 0)
6300 value
.value_type
= SE_DATA_TYPE_UINT64
;
6301 value
.value
.sv_uint64
= vd
->vdev_guid
;
6302 if (sysevent_add_attr(&attr
, ZFS_EV_VDEV_GUID
, &value
,
6306 if (vd
->vdev_path
) {
6307 value
.value_type
= SE_DATA_TYPE_STRING
;
6308 value
.value
.sv_string
= vd
->vdev_path
;
6309 if (sysevent_add_attr(&attr
, ZFS_EV_VDEV_PATH
,
6310 &value
, SE_SLEEP
) != 0)
6315 if (sysevent_attach_attributes(ev
, attr
) != 0)
6319 (void) log_sysevent(ev
, SE_SLEEP
, &eid
);
6323 sysevent_free_attr(attr
);