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 2009 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 #include <sys/zfs_context.h>
34 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
37 #include <sys/zio_checksum.h>
39 #include <sys/dmu_tx.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/metaslab.h>
45 #include <sys/metaslab_impl.h>
46 #include <sys/uberblock_impl.h>
49 #include <sys/dmu_traverse.h>
50 #include <sys/dmu_objset.h>
51 #include <sys/unique.h>
52 #include <sys/dsl_pool.h>
53 #include <sys/dsl_dataset.h>
54 #include <sys/dsl_dir.h>
55 #include <sys/dsl_prop.h>
56 #include <sys/dsl_synctask.h>
57 #include <sys/fs/zfs.h>
59 #include <sys/callb.h>
60 #include <sys/systeminfo.h>
61 #include <sys/sunddi.h>
62 #include <sys/spa_boot.h>
63 #include <sys/zfs_ioctl.h>
67 #include <sys/bootprops.h>
71 #include "zfs_comutil.h"
74 zti_mode_fixed
, /* value is # of threads (min 1) */
75 zti_mode_online_percent
, /* value is % of online CPUs */
76 zti_mode_tune
, /* fill from zio_taskq_tune_* */
80 #define ZTI_THREAD_FIX(n) { zti_mode_fixed, (n) }
81 #define ZTI_THREAD_PCT(n) { zti_mode_online_percent, (n) }
82 #define ZTI_THREAD_TUNE { zti_mode_tune, 0 }
84 #define ZTI_THREAD_ONE ZTI_THREAD_FIX(1)
86 typedef struct zio_taskq_info
{
89 enum zti_modes zti_mode
;
91 } zti_nthreads
[ZIO_TASKQ_TYPES
];
94 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
98 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
] = {
100 { "spa_zio_null", { ZTI_THREAD_ONE
, ZTI_THREAD_ONE
} },
101 { "spa_zio_read", { ZTI_THREAD_FIX(8), ZTI_THREAD_TUNE
} },
102 { "spa_zio_write", { ZTI_THREAD_TUNE
, ZTI_THREAD_FIX(8) } },
103 { "spa_zio_free", { ZTI_THREAD_ONE
, ZTI_THREAD_ONE
} },
104 { "spa_zio_claim", { ZTI_THREAD_ONE
, ZTI_THREAD_ONE
} },
105 { "spa_zio_ioctl", { ZTI_THREAD_ONE
, ZTI_THREAD_ONE
} },
108 enum zti_modes zio_taskq_tune_mode
= zti_mode_online_percent
;
109 uint_t zio_taskq_tune_value
= 80; /* #threads = 80% of # online CPUs */
111 static void spa_sync_props(void *arg1
, void *arg2
, cred_t
*cr
, dmu_tx_t
*tx
);
112 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
115 * ==========================================================================
116 * SPA properties routines
117 * ==========================================================================
121 * Add a (source=src, propname=propval) list to an nvlist.
124 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
125 uint64_t intval
, zprop_source_t src
)
127 const char *propname
= zpool_prop_to_name(prop
);
130 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
131 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
134 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
136 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
138 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
139 nvlist_free(propval
);
143 * Get property values from the spa configuration.
146 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
150 uint64_t cap
, version
;
151 zprop_source_t src
= ZPROP_SRC_NONE
;
152 spa_config_dirent_t
*dp
;
154 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
156 if (spa
->spa_root_vdev
!= NULL
) {
157 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
158 size
= metaslab_class_get_space(spa_normal_class(spa
));
159 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
160 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
161 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
162 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
165 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
166 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
168 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
169 ddt_get_pool_dedup_ratio(spa
), src
);
171 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
172 spa
->spa_root_vdev
->vdev_state
, src
);
174 version
= spa_version(spa
);
175 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
))
176 src
= ZPROP_SRC_DEFAULT
;
178 src
= ZPROP_SRC_LOCAL
;
179 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
182 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
184 if (spa
->spa_root
!= NULL
)
185 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
188 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
189 if (dp
->scd_path
== NULL
) {
190 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
191 "none", 0, ZPROP_SRC_LOCAL
);
192 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
193 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
194 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
200 * Get zpool property values.
203 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
205 objset_t
*mos
= spa
->spa_meta_objset
;
210 VERIFY(nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
212 mutex_enter(&spa
->spa_props_lock
);
215 * Get properties from the spa config.
217 spa_prop_get_config(spa
, nvp
);
219 /* If no pool property object, no more prop to get. */
220 if (spa
->spa_pool_props_object
== 0) {
221 mutex_exit(&spa
->spa_props_lock
);
226 * Get properties from the MOS pool property object.
228 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
229 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
230 zap_cursor_advance(&zc
)) {
233 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
236 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
239 switch (za
.za_integer_length
) {
241 /* integer property */
242 if (za
.za_first_integer
!=
243 zpool_prop_default_numeric(prop
))
244 src
= ZPROP_SRC_LOCAL
;
246 if (prop
== ZPOOL_PROP_BOOTFS
) {
248 dsl_dataset_t
*ds
= NULL
;
250 dp
= spa_get_dsl(spa
);
251 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
252 if (err
= dsl_dataset_hold_obj(dp
,
253 za
.za_first_integer
, FTAG
, &ds
)) {
254 rw_exit(&dp
->dp_config_rwlock
);
259 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1,
261 dsl_dataset_name(ds
, strval
);
262 dsl_dataset_rele(ds
, FTAG
);
263 rw_exit(&dp
->dp_config_rwlock
);
266 intval
= za
.za_first_integer
;
269 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
273 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1);
278 /* string property */
279 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
280 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
281 za
.za_name
, 1, za
.za_num_integers
, strval
);
283 kmem_free(strval
, za
.za_num_integers
);
286 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
287 kmem_free(strval
, za
.za_num_integers
);
294 zap_cursor_fini(&zc
);
295 mutex_exit(&spa
->spa_props_lock
);
297 if (err
&& err
!= ENOENT
) {
307 * Validate the given pool properties nvlist and modify the list
308 * for the property values to be set.
311 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
314 int error
= 0, reset_bootfs
= 0;
318 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
320 char *propname
, *strval
;
325 propname
= nvpair_name(elem
);
327 if ((prop
= zpool_name_to_prop(propname
)) == ZPROP_INVAL
)
331 case ZPOOL_PROP_VERSION
:
332 error
= nvpair_value_uint64(elem
, &intval
);
334 (intval
< spa_version(spa
) || intval
> SPA_VERSION
))
338 case ZPOOL_PROP_DELEGATION
:
339 case ZPOOL_PROP_AUTOREPLACE
:
340 case ZPOOL_PROP_LISTSNAPS
:
341 case ZPOOL_PROP_AUTOEXPAND
:
342 error
= nvpair_value_uint64(elem
, &intval
);
343 if (!error
&& intval
> 1)
347 case ZPOOL_PROP_BOOTFS
:
349 * If the pool version is less than SPA_VERSION_BOOTFS,
350 * or the pool is still being created (version == 0),
351 * the bootfs property cannot be set.
353 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
359 * Make sure the vdev config is bootable
361 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
368 error
= nvpair_value_string(elem
, &strval
);
373 if (strval
== NULL
|| strval
[0] == '\0') {
374 objnum
= zpool_prop_default_numeric(
379 if (error
= dmu_objset_hold(strval
, FTAG
, &os
))
382 /* Must be ZPL and not gzip compressed. */
384 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
386 } else if ((error
= dsl_prop_get_integer(strval
,
387 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
388 &compress
, NULL
)) == 0 &&
389 !BOOTFS_COMPRESS_VALID(compress
)) {
392 objnum
= dmu_objset_id(os
);
394 dmu_objset_rele(os
, FTAG
);
398 case ZPOOL_PROP_FAILUREMODE
:
399 error
= nvpair_value_uint64(elem
, &intval
);
400 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
401 intval
> ZIO_FAILURE_MODE_PANIC
))
405 * This is a special case which only occurs when
406 * the pool has completely failed. This allows
407 * the user to change the in-core failmode property
408 * without syncing it out to disk (I/Os might
409 * currently be blocked). We do this by returning
410 * EIO to the caller (spa_prop_set) to trick it
411 * into thinking we encountered a property validation
414 if (!error
&& spa_suspended(spa
)) {
415 spa
->spa_failmode
= intval
;
420 case ZPOOL_PROP_CACHEFILE
:
421 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
424 if (strval
[0] == '\0')
427 if (strcmp(strval
, "none") == 0)
430 if (strval
[0] != '/') {
435 slash
= strrchr(strval
, '/');
436 ASSERT(slash
!= NULL
);
438 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
439 strcmp(slash
, "/..") == 0)
443 case ZPOOL_PROP_DEDUPDITTO
:
444 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
447 error
= nvpair_value_uint64(elem
, &intval
);
449 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
458 if (!error
&& reset_bootfs
) {
459 error
= nvlist_remove(props
,
460 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
463 error
= nvlist_add_uint64(props
,
464 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
472 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
475 spa_config_dirent_t
*dp
;
477 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
481 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
484 if (cachefile
[0] == '\0')
485 dp
->scd_path
= spa_strdup(spa_config_path
);
486 else if (strcmp(cachefile
, "none") == 0)
489 dp
->scd_path
= spa_strdup(cachefile
);
491 list_insert_head(&spa
->spa_config_list
, dp
);
493 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
497 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
501 boolean_t need_sync
= B_FALSE
;
504 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
508 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
509 if ((prop
= zpool_name_to_prop(
510 nvpair_name(elem
))) == ZPROP_INVAL
)
513 if (prop
== ZPOOL_PROP_CACHEFILE
|| prop
== ZPOOL_PROP_ALTROOT
)
521 return (dsl_sync_task_do(spa_get_dsl(spa
), NULL
, spa_sync_props
,
528 * If the bootfs property value is dsobj, clear it.
531 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
533 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
534 VERIFY(zap_remove(spa
->spa_meta_objset
,
535 spa
->spa_pool_props_object
,
536 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
542 * ==========================================================================
543 * SPA state manipulation (open/create/destroy/import/export)
544 * ==========================================================================
548 spa_error_entry_compare(const void *a
, const void *b
)
550 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
551 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
554 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
555 sizeof (zbookmark_t
));
566 * Utility function which retrieves copies of the current logs and
567 * re-initializes them in the process.
570 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
572 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
574 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
575 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
577 avl_create(&spa
->spa_errlist_scrub
,
578 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
579 offsetof(spa_error_entry_t
, se_avl
));
580 avl_create(&spa
->spa_errlist_last
,
581 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
582 offsetof(spa_error_entry_t
, se_avl
));
586 * Activate an uninitialized pool.
589 spa_activate(spa_t
*spa
, int mode
)
591 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
593 spa
->spa_state
= POOL_STATE_ACTIVE
;
594 spa
->spa_mode
= mode
;
596 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
597 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
599 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
600 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
];
601 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
602 enum zti_modes mode
= ztip
->zti_nthreads
[q
].zti_mode
;
603 uint_t value
= ztip
->zti_nthreads
[q
].zti_value
;
606 (void) snprintf(name
, sizeof (name
),
607 "%s_%s", ztip
->zti_name
, zio_taskq_types
[q
]);
609 if (mode
== zti_mode_tune
) {
610 mode
= zio_taskq_tune_mode
;
611 value
= zio_taskq_tune_value
;
612 if (mode
== zti_mode_tune
)
613 mode
= zti_mode_online_percent
;
618 ASSERT3U(value
, >=, 1);
619 value
= MAX(value
, 1);
621 spa
->spa_zio_taskq
[t
][q
] = taskq_create(name
,
622 value
, maxclsyspri
, 50, INT_MAX
,
626 case zti_mode_online_percent
:
627 spa
->spa_zio_taskq
[t
][q
] = taskq_create(name
,
628 value
, maxclsyspri
, 50, INT_MAX
,
629 TASKQ_PREPOPULATE
| TASKQ_THREADS_CPU_PCT
);
634 panic("unrecognized mode for "
635 "zio_taskqs[%u]->zti_nthreads[%u] (%u:%u) "
643 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
644 offsetof(vdev_t
, vdev_config_dirty_node
));
645 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
646 offsetof(vdev_t
, vdev_state_dirty_node
));
648 txg_list_create(&spa
->spa_vdev_txg_list
,
649 offsetof(struct vdev
, vdev_txg_node
));
651 avl_create(&spa
->spa_errlist_scrub
,
652 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
653 offsetof(spa_error_entry_t
, se_avl
));
654 avl_create(&spa
->spa_errlist_last
,
655 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
656 offsetof(spa_error_entry_t
, se_avl
));
660 * Opposite of spa_activate().
663 spa_deactivate(spa_t
*spa
)
665 ASSERT(spa
->spa_sync_on
== B_FALSE
);
666 ASSERT(spa
->spa_dsl_pool
== NULL
);
667 ASSERT(spa
->spa_root_vdev
== NULL
);
668 ASSERT(spa
->spa_async_zio_root
== NULL
);
669 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
671 txg_list_destroy(&spa
->spa_vdev_txg_list
);
673 list_destroy(&spa
->spa_config_dirty_list
);
674 list_destroy(&spa
->spa_state_dirty_list
);
676 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
677 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
678 taskq_destroy(spa
->spa_zio_taskq
[t
][q
]);
679 spa
->spa_zio_taskq
[t
][q
] = NULL
;
683 metaslab_class_destroy(spa
->spa_normal_class
);
684 spa
->spa_normal_class
= NULL
;
686 metaslab_class_destroy(spa
->spa_log_class
);
687 spa
->spa_log_class
= NULL
;
690 * If this was part of an import or the open otherwise failed, we may
691 * still have errors left in the queues. Empty them just in case.
693 spa_errlog_drain(spa
);
695 avl_destroy(&spa
->spa_errlist_scrub
);
696 avl_destroy(&spa
->spa_errlist_last
);
698 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
702 * Verify a pool configuration, and construct the vdev tree appropriately. This
703 * will create all the necessary vdevs in the appropriate layout, with each vdev
704 * in the CLOSED state. This will prep the pool before open/creation/import.
705 * All vdev validation is done by the vdev_alloc() routine.
708 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
709 uint_t id
, int atype
)
715 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
718 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
721 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
733 for (int c
= 0; c
< children
; c
++) {
735 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
743 ASSERT(*vdp
!= NULL
);
749 * Opposite of spa_load().
752 spa_unload(spa_t
*spa
)
756 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
761 spa_async_suspend(spa
);
766 if (spa
->spa_sync_on
) {
767 txg_sync_stop(spa
->spa_dsl_pool
);
768 spa
->spa_sync_on
= B_FALSE
;
772 * Wait for any outstanding async I/O to complete.
774 if (spa
->spa_async_zio_root
!= NULL
) {
775 (void) zio_wait(spa
->spa_async_zio_root
);
776 spa
->spa_async_zio_root
= NULL
;
780 * Close the dsl pool.
782 if (spa
->spa_dsl_pool
) {
783 dsl_pool_close(spa
->spa_dsl_pool
);
784 spa
->spa_dsl_pool
= NULL
;
789 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
792 * Drop and purge level 2 cache
794 spa_l2cache_drop(spa
);
799 if (spa
->spa_root_vdev
)
800 vdev_free(spa
->spa_root_vdev
);
801 ASSERT(spa
->spa_root_vdev
== NULL
);
803 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
804 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
805 if (spa
->spa_spares
.sav_vdevs
) {
806 kmem_free(spa
->spa_spares
.sav_vdevs
,
807 spa
->spa_spares
.sav_count
* sizeof (void *));
808 spa
->spa_spares
.sav_vdevs
= NULL
;
810 if (spa
->spa_spares
.sav_config
) {
811 nvlist_free(spa
->spa_spares
.sav_config
);
812 spa
->spa_spares
.sav_config
= NULL
;
814 spa
->spa_spares
.sav_count
= 0;
816 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
817 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
818 if (spa
->spa_l2cache
.sav_vdevs
) {
819 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
820 spa
->spa_l2cache
.sav_count
* sizeof (void *));
821 spa
->spa_l2cache
.sav_vdevs
= NULL
;
823 if (spa
->spa_l2cache
.sav_config
) {
824 nvlist_free(spa
->spa_l2cache
.sav_config
);
825 spa
->spa_l2cache
.sav_config
= NULL
;
827 spa
->spa_l2cache
.sav_count
= 0;
829 spa
->spa_async_suspended
= 0;
831 spa_config_exit(spa
, SCL_ALL
, FTAG
);
835 * Load (or re-load) the current list of vdevs describing the active spares for
836 * this pool. When this is called, we have some form of basic information in
837 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
838 * then re-generate a more complete list including status information.
841 spa_load_spares(spa_t
*spa
)
848 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
851 * First, close and free any existing spare vdevs.
853 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
854 vd
= spa
->spa_spares
.sav_vdevs
[i
];
856 /* Undo the call to spa_activate() below */
857 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
858 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
859 spa_spare_remove(tvd
);
864 if (spa
->spa_spares
.sav_vdevs
)
865 kmem_free(spa
->spa_spares
.sav_vdevs
,
866 spa
->spa_spares
.sav_count
* sizeof (void *));
868 if (spa
->spa_spares
.sav_config
== NULL
)
871 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
872 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
874 spa
->spa_spares
.sav_count
= (int)nspares
;
875 spa
->spa_spares
.sav_vdevs
= NULL
;
881 * Construct the array of vdevs, opening them to get status in the
882 * process. For each spare, there is potentially two different vdev_t
883 * structures associated with it: one in the list of spares (used only
884 * for basic validation purposes) and one in the active vdev
885 * configuration (if it's spared in). During this phase we open and
886 * validate each vdev on the spare list. If the vdev also exists in the
887 * active configuration, then we also mark this vdev as an active spare.
889 spa
->spa_spares
.sav_vdevs
= kmem_alloc(nspares
* sizeof (void *),
891 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
892 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
893 VDEV_ALLOC_SPARE
) == 0);
896 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
898 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
900 if (!tvd
->vdev_isspare
)
904 * We only mark the spare active if we were successfully
905 * able to load the vdev. Otherwise, importing a pool
906 * with a bad active spare would result in strange
907 * behavior, because multiple pool would think the spare
908 * is actively in use.
910 * There is a vulnerability here to an equally bizarre
911 * circumstance, where a dead active spare is later
912 * brought back to life (onlined or otherwise). Given
913 * the rarity of this scenario, and the extra complexity
914 * it adds, we ignore the possibility.
916 if (!vdev_is_dead(tvd
))
917 spa_spare_activate(tvd
);
921 vd
->vdev_aux
= &spa
->spa_spares
;
923 if (vdev_open(vd
) != 0)
926 if (vdev_validate_aux(vd
) == 0)
931 * Recompute the stashed list of spares, with status information
934 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
935 DATA_TYPE_NVLIST_ARRAY
) == 0);
937 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
939 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
940 spares
[i
] = vdev_config_generate(spa
,
941 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, B_TRUE
, B_FALSE
);
942 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
943 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
944 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
945 nvlist_free(spares
[i
]);
946 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
950 * Load (or re-load) the current list of vdevs describing the active l2cache for
951 * this pool. When this is called, we have some form of basic information in
952 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
953 * then re-generate a more complete list including status information.
954 * Devices which are already active have their details maintained, and are
958 spa_load_l2cache(spa_t
*spa
)
964 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
965 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
967 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
969 if (sav
->sav_config
!= NULL
) {
970 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
971 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
972 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
977 oldvdevs
= sav
->sav_vdevs
;
978 oldnvdevs
= sav
->sav_count
;
979 sav
->sav_vdevs
= NULL
;
983 * Process new nvlist of vdevs.
985 for (i
= 0; i
< nl2cache
; i
++) {
986 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
990 for (j
= 0; j
< oldnvdevs
; j
++) {
992 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
994 * Retain previous vdev for add/remove ops.
1002 if (newvdevs
[i
] == NULL
) {
1006 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1007 VDEV_ALLOC_L2CACHE
) == 0);
1012 * Commit this vdev as an l2cache device,
1013 * even if it fails to open.
1015 spa_l2cache_add(vd
);
1020 spa_l2cache_activate(vd
);
1022 if (vdev_open(vd
) != 0)
1025 (void) vdev_validate_aux(vd
);
1027 if (!vdev_is_dead(vd
))
1028 l2arc_add_vdev(spa
, vd
);
1033 * Purge vdevs that were dropped
1035 for (i
= 0; i
< oldnvdevs
; i
++) {
1040 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1041 pool
!= 0ULL && l2arc_vdev_present(vd
))
1042 l2arc_remove_vdev(vd
);
1043 (void) vdev_close(vd
);
1044 spa_l2cache_remove(vd
);
1049 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1051 if (sav
->sav_config
== NULL
)
1054 sav
->sav_vdevs
= newvdevs
;
1055 sav
->sav_count
= (int)nl2cache
;
1058 * Recompute the stashed list of l2cache devices, with status
1059 * information this time.
1061 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1062 DATA_TYPE_NVLIST_ARRAY
) == 0);
1064 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1065 for (i
= 0; i
< sav
->sav_count
; i
++)
1066 l2cache
[i
] = vdev_config_generate(spa
,
1067 sav
->sav_vdevs
[i
], B_TRUE
, B_FALSE
, B_TRUE
);
1068 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1069 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1071 for (i
= 0; i
< sav
->sav_count
; i
++)
1072 nvlist_free(l2cache
[i
]);
1074 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1078 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1081 char *packed
= NULL
;
1086 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
1087 nvsize
= *(uint64_t *)db
->db_data
;
1088 dmu_buf_rele(db
, FTAG
);
1090 packed
= kmem_alloc(nvsize
, KM_SLEEP
);
1091 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1094 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1095 kmem_free(packed
, nvsize
);
1101 * Checks to see if the given vdev could not be opened, in which case we post a
1102 * sysevent to notify the autoreplace code that the device has been removed.
1105 spa_check_removed(vdev_t
*vd
)
1107 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1108 spa_check_removed(vd
->vdev_child
[c
]);
1110 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
)) {
1111 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1112 spa_event_notify(vd
->vdev_spa
, vd
, ESC_ZFS_VDEV_CHECK
);
1117 * Load the slog device state from the config object since it's possible
1118 * that the label does not contain the most up-to-date information.
1121 spa_load_log_state(spa_t
*spa
, nvlist_t
*nv
)
1123 vdev_t
*ovd
, *rvd
= spa
->spa_root_vdev
;
1126 * Load the original root vdev tree from the passed config.
1128 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1129 VERIFY(spa_config_parse(spa
, &ovd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1131 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1132 vdev_t
*cvd
= rvd
->vdev_child
[c
];
1133 if (cvd
->vdev_islog
)
1134 vdev_load_log_state(cvd
, ovd
->vdev_child
[c
]);
1137 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1141 * Check for missing log devices
1144 spa_check_logs(spa_t
*spa
)
1146 switch (spa
->spa_log_state
) {
1147 case SPA_LOG_MISSING
:
1148 /* need to recheck in case slog has been restored */
1149 case SPA_LOG_UNKNOWN
:
1150 if (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
, NULL
,
1151 DS_FIND_CHILDREN
)) {
1152 spa
->spa_log_state
= SPA_LOG_MISSING
;
1161 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1163 for (int i
= 0; i
< sav
->sav_count
; i
++)
1164 spa_check_removed(sav
->sav_vdevs
[i
]);
1168 spa_claim_notify(zio_t
*zio
)
1170 spa_t
*spa
= zio
->io_spa
;
1175 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1176 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1177 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1178 mutex_exit(&spa
->spa_props_lock
);
1181 typedef struct spa_load_error
{
1182 uint64_t sle_metadata_count
;
1183 uint64_t sle_data_count
;
1187 spa_load_verify_done(zio_t
*zio
)
1189 blkptr_t
*bp
= zio
->io_bp
;
1190 spa_load_error_t
*sle
= zio
->io_private
;
1191 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1192 int error
= zio
->io_error
;
1195 if ((BP_GET_LEVEL(bp
) != 0 || dmu_ot
[type
].ot_metadata
) &&
1196 type
!= DMU_OT_INTENT_LOG
)
1197 atomic_add_64(&sle
->sle_metadata_count
, 1);
1199 atomic_add_64(&sle
->sle_data_count
, 1);
1201 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1206 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1207 const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1211 size_t size
= BP_GET_PSIZE(bp
);
1212 void *data
= zio_data_buf_alloc(size
);
1214 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1215 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1216 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1217 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1223 spa_load_verify(spa_t
*spa
)
1226 spa_load_error_t sle
= { 0 };
1227 zpool_rewind_policy_t policy
;
1228 boolean_t verify_ok
= B_FALSE
;
1231 rio
= zio_root(spa
, NULL
, &sle
,
1232 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1234 error
= traverse_pool(spa
, spa_load_verify_cb
, rio
,
1235 spa
->spa_verify_min_txg
);
1237 (void) zio_wait(rio
);
1239 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1241 spa
->spa_load_meta_errors
= sle
.sle_metadata_count
;
1242 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1244 if (!error
&& sle
.sle_metadata_count
<= policy
.zrp_maxmeta
&&
1245 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1247 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1248 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1252 if (error
!= ENXIO
&& error
!= EIO
)
1257 return (verify_ok
? 0 : EIO
);
1261 * Load an existing storage pool, using the pool's builtin spa_config as a
1262 * source of configuration information.
1265 spa_load(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
1268 nvlist_t
*nvconfig
, *nvroot
= NULL
;
1270 uberblock_t
*ub
= &spa
->spa_uberblock
;
1271 uint64_t config_cache_txg
= spa
->spa_config_txg
;
1274 uint64_t autoreplace
= 0;
1275 int orig_mode
= spa
->spa_mode
;
1276 char *ereport
= FM_EREPORT_ZFS_POOL
;
1277 nvlist_t
*config
= spa
->spa_config
;
1280 * If this is an untrusted config, access the pool in read-only mode.
1281 * This prevents things like resilvering recently removed devices.
1284 spa
->spa_mode
= FREAD
;
1286 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1288 spa
->spa_load_state
= state
;
1290 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) ||
1291 nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
1297 * Versioning wasn't explicitly added to the label until later, so if
1298 * it's not present treat it as the initial version.
1300 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
, &version
) != 0)
1301 version
= SPA_VERSION_INITIAL
;
1303 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
1304 &spa
->spa_config_txg
);
1306 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
1307 spa_guid_exists(pool_guid
, 0)) {
1312 spa
->spa_load_guid
= pool_guid
;
1315 * Create "The Godfather" zio to hold all async IOs
1317 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
1318 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
1321 * Parse the configuration into a vdev tree. We explicitly set the
1322 * value that will be returned by spa_version() since parsing the
1323 * configuration requires knowing the version number.
1325 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1326 spa
->spa_ubsync
.ub_version
= version
;
1327 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_LOAD
);
1328 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1333 ASSERT(spa
->spa_root_vdev
== rvd
);
1334 ASSERT(spa_guid(spa
) == pool_guid
);
1337 * Try to open all vdevs, loading each label in the process.
1339 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1340 error
= vdev_open(rvd
);
1341 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1346 * We need to validate the vdev labels against the configuration that
1347 * we have in hand, which is dependent on the setting of mosconfig. If
1348 * mosconfig is true then we're validating the vdev labels based on
1349 * that config. Otherwise, we're validating against the cached config
1350 * (zpool.cache) that was read when we loaded the zfs module, and then
1351 * later we will recursively call spa_load() and validate against
1354 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1355 error
= vdev_validate(rvd
);
1356 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1360 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
1366 * Find the best uberblock.
1368 vdev_uberblock_load(NULL
, rvd
, ub
);
1371 * If we weren't able to find a single valid uberblock, return failure.
1373 if (ub
->ub_txg
== 0) {
1374 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1375 VDEV_AUX_CORRUPT_DATA
);
1381 * If the pool is newer than the code, we can't open it.
1383 if (ub
->ub_version
> SPA_VERSION
) {
1384 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1385 VDEV_AUX_VERSION_NEWER
);
1391 * If the vdev guid sum doesn't match the uberblock, we have an
1392 * incomplete configuration.
1394 if (rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
&& mosconfig
) {
1395 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1396 VDEV_AUX_BAD_GUID_SUM
);
1402 * Initialize internal SPA structures.
1404 spa
->spa_state
= POOL_STATE_ACTIVE
;
1405 spa
->spa_ubsync
= spa
->spa_uberblock
;
1406 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
1407 TXG_INITIAL
: spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
;
1408 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
1409 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
1410 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
1412 error
= dsl_pool_open(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
1414 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1415 VDEV_AUX_CORRUPT_DATA
);
1419 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
1421 if (zap_lookup(spa
->spa_meta_objset
,
1422 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
1423 sizeof (uint64_t), 1, &spa
->spa_config_object
) != 0) {
1424 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1425 VDEV_AUX_CORRUPT_DATA
);
1430 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0) {
1431 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1432 VDEV_AUX_CORRUPT_DATA
);
1440 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
1441 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
1443 unsigned long myhostid
= 0;
1445 VERIFY(nvlist_lookup_string(nvconfig
,
1446 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
1449 myhostid
= zone_get_hostid(NULL
);
1452 * We're emulating the system's hostid in userland, so
1453 * we can't use zone_get_hostid().
1455 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
1456 #endif /* _KERNEL */
1457 if (hostid
!= 0 && myhostid
!= 0 &&
1458 hostid
!= myhostid
) {
1459 cmn_err(CE_WARN
, "pool '%s' could not be "
1460 "loaded as it was last accessed by "
1461 "another system (host: %s hostid: 0x%lx). "
1462 "See: http://www.sun.com/msg/ZFS-8000-EY",
1463 spa_name(spa
), hostname
,
1464 (unsigned long)hostid
);
1470 spa_config_set(spa
, nvconfig
);
1472 spa_deactivate(spa
);
1473 spa_activate(spa
, orig_mode
);
1475 return (spa_load(spa
, state
, B_TRUE
));
1478 if (zap_lookup(spa
->spa_meta_objset
,
1479 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPLIST
,
1480 sizeof (uint64_t), 1, &spa
->spa_deferred_bplist_obj
) != 0) {
1481 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1482 VDEV_AUX_CORRUPT_DATA
);
1488 * Load the bit that tells us to use the new accounting function
1489 * (raid-z deflation). If we have an older pool, this will not
1492 error
= zap_lookup(spa
->spa_meta_objset
,
1493 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
1494 sizeof (uint64_t), 1, &spa
->spa_deflate
);
1495 if (error
!= 0 && error
!= ENOENT
) {
1496 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1497 VDEV_AUX_CORRUPT_DATA
);
1503 * Load the persistent error log. If we have an older pool, this will
1506 error
= zap_lookup(spa
->spa_meta_objset
,
1507 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ERRLOG_LAST
,
1508 sizeof (uint64_t), 1, &spa
->spa_errlog_last
);
1509 if (error
!= 0 && error
!= ENOENT
) {
1510 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1511 VDEV_AUX_CORRUPT_DATA
);
1516 error
= zap_lookup(spa
->spa_meta_objset
,
1517 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ERRLOG_SCRUB
,
1518 sizeof (uint64_t), 1, &spa
->spa_errlog_scrub
);
1519 if (error
!= 0 && error
!= ENOENT
) {
1520 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1521 VDEV_AUX_CORRUPT_DATA
);
1527 * Load the history object. If we have an older pool, this
1528 * will not be present.
1530 error
= zap_lookup(spa
->spa_meta_objset
,
1531 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_HISTORY
,
1532 sizeof (uint64_t), 1, &spa
->spa_history
);
1533 if (error
!= 0 && error
!= ENOENT
) {
1534 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1535 VDEV_AUX_CORRUPT_DATA
);
1541 * Load any hot spares for this pool.
1543 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1544 DMU_POOL_SPARES
, sizeof (uint64_t), 1, &spa
->spa_spares
.sav_object
);
1545 if (error
!= 0 && error
!= ENOENT
) {
1546 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1547 VDEV_AUX_CORRUPT_DATA
);
1552 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
1553 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
1554 &spa
->spa_spares
.sav_config
) != 0) {
1555 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1556 VDEV_AUX_CORRUPT_DATA
);
1561 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1562 spa_load_spares(spa
);
1563 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1567 * Load any level 2 ARC devices for this pool.
1569 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1570 DMU_POOL_L2CACHE
, sizeof (uint64_t), 1,
1571 &spa
->spa_l2cache
.sav_object
);
1572 if (error
!= 0 && error
!= ENOENT
) {
1573 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1574 VDEV_AUX_CORRUPT_DATA
);
1579 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
1580 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
1581 &spa
->spa_l2cache
.sav_config
) != 0) {
1582 vdev_set_state(rvd
, B_TRUE
,
1583 VDEV_STATE_CANT_OPEN
,
1584 VDEV_AUX_CORRUPT_DATA
);
1589 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1590 spa_load_l2cache(spa
);
1591 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1594 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
1596 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1597 DMU_POOL_PROPS
, sizeof (uint64_t), 1, &spa
->spa_pool_props_object
);
1599 if (error
&& error
!= ENOENT
) {
1600 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1601 VDEV_AUX_CORRUPT_DATA
);
1607 (void) zap_lookup(spa
->spa_meta_objset
,
1608 spa
->spa_pool_props_object
,
1609 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
),
1610 sizeof (uint64_t), 1, &spa
->spa_bootfs
);
1611 (void) zap_lookup(spa
->spa_meta_objset
,
1612 spa
->spa_pool_props_object
,
1613 zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE
),
1614 sizeof (uint64_t), 1, &autoreplace
);
1615 spa
->spa_autoreplace
= (autoreplace
!= 0);
1616 (void) zap_lookup(spa
->spa_meta_objset
,
1617 spa
->spa_pool_props_object
,
1618 zpool_prop_to_name(ZPOOL_PROP_DELEGATION
),
1619 sizeof (uint64_t), 1, &spa
->spa_delegation
);
1620 (void) zap_lookup(spa
->spa_meta_objset
,
1621 spa
->spa_pool_props_object
,
1622 zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE
),
1623 sizeof (uint64_t), 1, &spa
->spa_failmode
);
1624 (void) zap_lookup(spa
->spa_meta_objset
,
1625 spa
->spa_pool_props_object
,
1626 zpool_prop_to_name(ZPOOL_PROP_AUTOEXPAND
),
1627 sizeof (uint64_t), 1, &spa
->spa_autoexpand
);
1628 (void) zap_lookup(spa
->spa_meta_objset
,
1629 spa
->spa_pool_props_object
,
1630 zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO
),
1631 sizeof (uint64_t), 1, &spa
->spa_dedup_ditto
);
1635 * If the 'autoreplace' property is set, then post a resource notifying
1636 * the ZFS DE that it should not issue any faults for unopenable
1637 * devices. We also iterate over the vdevs, and post a sysevent for any
1638 * unopenable vdevs so that the normal autoreplace handler can take
1641 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
1642 spa_check_removed(spa
->spa_root_vdev
);
1644 * For the import case, this is done in spa_import(), because
1645 * at this point we're using the spare definitions from
1646 * the MOS config, not necessarily from the userland config.
1648 if (state
!= SPA_LOAD_IMPORT
) {
1649 spa_aux_check_removed(&spa
->spa_spares
);
1650 spa_aux_check_removed(&spa
->spa_l2cache
);
1655 * Load the vdev state for all toplevel vdevs.
1660 * Propagate the leaf DTLs we just loaded all the way up the tree.
1662 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1663 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
1664 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1667 * Check the state of the root vdev. If it can't be opened, it
1668 * indicates one or more toplevel vdevs are faulted.
1670 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
1676 * Load the DDTs (dedup tables).
1678 error
= ddt_load(spa
);
1680 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1681 VDEV_AUX_CORRUPT_DATA
);
1686 spa_update_dspace(spa
);
1688 if (state
!= SPA_LOAD_TRYIMPORT
) {
1689 error
= spa_load_verify(spa
);
1691 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1692 VDEV_AUX_CORRUPT_DATA
);
1698 * Load the intent log state and check log integrity.
1700 VERIFY(nvlist_lookup_nvlist(nvconfig
, ZPOOL_CONFIG_VDEV_TREE
,
1702 spa_load_log_state(spa
, nvroot
);
1703 nvlist_free(nvconfig
);
1705 if (spa_check_logs(spa
)) {
1706 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1709 ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
1713 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
1714 spa
->spa_load_max_txg
== UINT64_MAX
)) {
1716 int need_update
= B_FALSE
;
1718 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
1721 * Claim log blocks that haven't been committed yet.
1722 * This must all happen in a single txg.
1723 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
1724 * invoked from zil_claim_log_block()'s i/o done callback.
1725 * Price of rollback is that we abandon the log.
1727 spa
->spa_claiming
= B_TRUE
;
1729 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
1730 spa_first_txg(spa
));
1731 (void) dmu_objset_find(spa_name(spa
),
1732 zil_claim
, tx
, DS_FIND_CHILDREN
);
1735 spa
->spa_claiming
= B_FALSE
;
1737 spa
->spa_log_state
= SPA_LOG_GOOD
;
1738 spa
->spa_sync_on
= B_TRUE
;
1739 txg_sync_start(spa
->spa_dsl_pool
);
1742 * Wait for all claims to sync. We sync up to the highest
1743 * claimed log block birth time so that claimed log blocks
1744 * don't appear to be from the future. spa_claim_max_txg
1745 * will have been set for us by either zil_check_log_chain()
1746 * (invoked from spa_check_logs()) or zil_claim() above.
1748 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
1751 * If the config cache is stale, or we have uninitialized
1752 * metaslabs (see spa_vdev_add()), then update the config.
1754 * If spa_load_verbatim is true, trust the current
1755 * in-core spa_config and update the disk labels.
1757 if (config_cache_txg
!= spa
->spa_config_txg
||
1758 state
== SPA_LOAD_IMPORT
|| spa
->spa_load_verbatim
||
1759 state
== SPA_LOAD_RECOVER
)
1760 need_update
= B_TRUE
;
1762 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
1763 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
1764 need_update
= B_TRUE
;
1767 * Update the config cache asychronously in case we're the
1768 * root pool, in which case the config cache isn't writable yet.
1771 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
1774 * Check all DTLs to see if anything needs resilvering.
1776 if (vdev_resilver_needed(rvd
, NULL
, NULL
))
1777 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1780 * Delete any inconsistent datasets.
1782 (void) dmu_objset_find(spa_name(spa
),
1783 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
1786 * Clean up any stale temporary dataset userrefs.
1788 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
1794 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
1795 if (error
&& error
!= EBADF
)
1796 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
1797 spa
->spa_load_state
= SPA_LOAD_NONE
;
1804 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
1807 spa_deactivate(spa
);
1809 spa
->spa_load_max_txg
--;
1811 spa_activate(spa
, spa_mode_global
);
1812 spa_async_suspend(spa
);
1814 return (spa_load(spa
, state
, mosconfig
));
1818 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
1819 uint64_t max_request
, boolean_t extreme
)
1821 nvlist_t
*config
= NULL
;
1822 int load_error
, rewind_error
;
1823 uint64_t safe_rollback_txg
;
1826 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
)
1827 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
1829 spa
->spa_load_max_txg
= max_request
;
1831 load_error
= rewind_error
= spa_load(spa
, state
, mosconfig
);
1832 if (load_error
== 0)
1835 if (spa
->spa_root_vdev
!= NULL
)
1836 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
1838 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
1839 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1841 /* specific txg requested */
1842 if (spa
->spa_load_max_txg
!= UINT64_MAX
&& !extreme
) {
1843 nvlist_free(config
);
1844 return (load_error
);
1847 /* Price of rolling back is discarding txgs, including log */
1848 if (state
== SPA_LOAD_RECOVER
)
1849 spa
->spa_log_state
= SPA_LOG_CLEAR
;
1851 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1852 safe_rollback_txg
= spa
->spa_uberblock
.ub_txg
- TXG_DEFER_SIZE
;
1854 min_txg
= extreme
? TXG_INITIAL
: safe_rollback_txg
;
1855 while (rewind_error
&& (spa
->spa_uberblock
.ub_txg
>= min_txg
)) {
1856 if (spa
->spa_load_max_txg
< safe_rollback_txg
)
1857 spa
->spa_extreme_rewind
= B_TRUE
;
1858 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
1862 spa_rewind_data_to_nvlist(spa
, config
);
1864 spa
->spa_extreme_rewind
= B_FALSE
;
1865 spa
->spa_load_max_txg
= UINT64_MAX
;
1867 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
1868 spa_config_set(spa
, config
);
1870 return (state
== SPA_LOAD_RECOVER
? rewind_error
: load_error
);
1876 * The import case is identical to an open except that the configuration is sent
1877 * down from userland, instead of grabbed from the configuration cache. For the
1878 * case of an open, the pool configuration will exist in the
1879 * POOL_STATE_UNINITIALIZED state.
1881 * The stats information (gen/count/ustats) is used to gather vdev statistics at
1882 * the same time open the pool, without having to keep around the spa_t in some
1886 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
1892 zpool_rewind_policy_t policy
;
1893 spa_load_state_t state
= SPA_LOAD_OPEN
;
1895 int locked
= B_FALSE
;
1899 zpool_get_rewind_policy(nvpolicy
, &policy
);
1900 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
1901 state
= SPA_LOAD_RECOVER
;
1902 norewind
= (policy
.zrp_request
== ZPOOL_NO_REWIND
);
1903 extreme
= ((policy
.zrp_request
& ZPOOL_EXTREME_REWIND
) != 0);
1906 * As disgusting as this is, we need to support recursive calls to this
1907 * function because dsl_dir_open() is called during spa_load(), and ends
1908 * up calling spa_open() again. The real fix is to figure out how to
1909 * avoid dsl_dir_open() calling this in the first place.
1911 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
1912 mutex_enter(&spa_namespace_lock
);
1916 if ((spa
= spa_lookup(pool
)) == NULL
) {
1918 mutex_exit(&spa_namespace_lock
);
1922 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
1924 spa_activate(spa
, spa_mode_global
);
1926 if (spa
->spa_last_open_failed
&& norewind
) {
1927 if (config
!= NULL
&& spa
->spa_config
)
1928 VERIFY(nvlist_dup(spa
->spa_config
,
1929 config
, KM_SLEEP
) == 0);
1930 spa_deactivate(spa
);
1932 mutex_exit(&spa_namespace_lock
);
1933 return (spa
->spa_last_open_failed
);
1936 if (state
!= SPA_LOAD_RECOVER
)
1937 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
1939 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
1942 if (error
== EBADF
) {
1944 * If vdev_validate() returns failure (indicated by
1945 * EBADF), it indicates that one of the vdevs indicates
1946 * that the pool has been exported or destroyed. If
1947 * this is the case, the config cache is out of sync and
1948 * we should remove the pool from the namespace.
1951 spa_deactivate(spa
);
1952 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
1955 mutex_exit(&spa_namespace_lock
);
1961 * We can't open the pool, but we still have useful
1962 * information: the state of each vdev after the
1963 * attempted vdev_open(). Return this to the user.
1965 if (config
!= NULL
&& spa
->spa_config
)
1966 VERIFY(nvlist_dup(spa
->spa_config
, config
,
1969 spa_deactivate(spa
);
1970 spa
->spa_last_open_failed
= error
;
1972 mutex_exit(&spa_namespace_lock
);
1979 spa_open_ref(spa
, tag
);
1981 spa
->spa_last_open_failed
= 0;
1984 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
1986 spa
->spa_last_ubsync_txg
= 0;
1987 spa
->spa_load_txg
= 0;
1990 mutex_exit(&spa_namespace_lock
);
1998 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
2001 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
2005 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
2007 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
2011 * Lookup the given spa_t, incrementing the inject count in the process,
2012 * preventing it from being exported or destroyed.
2015 spa_inject_addref(char *name
)
2019 mutex_enter(&spa_namespace_lock
);
2020 if ((spa
= spa_lookup(name
)) == NULL
) {
2021 mutex_exit(&spa_namespace_lock
);
2024 spa
->spa_inject_ref
++;
2025 mutex_exit(&spa_namespace_lock
);
2031 spa_inject_delref(spa_t
*spa
)
2033 mutex_enter(&spa_namespace_lock
);
2034 spa
->spa_inject_ref
--;
2035 mutex_exit(&spa_namespace_lock
);
2039 * Add spares device information to the nvlist.
2042 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
2052 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2054 if (spa
->spa_spares
.sav_count
== 0)
2057 VERIFY(nvlist_lookup_nvlist(config
,
2058 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
2059 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
2060 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
2062 VERIFY(nvlist_add_nvlist_array(nvroot
,
2063 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
2064 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
2065 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
2068 * Go through and find any spares which have since been
2069 * repurposed as an active spare. If this is the case, update
2070 * their status appropriately.
2072 for (i
= 0; i
< nspares
; i
++) {
2073 VERIFY(nvlist_lookup_uint64(spares
[i
],
2074 ZPOOL_CONFIG_GUID
, &guid
) == 0);
2075 if (spa_spare_exists(guid
, &pool
, NULL
) &&
2077 VERIFY(nvlist_lookup_uint64_array(
2078 spares
[i
], ZPOOL_CONFIG_STATS
,
2079 (uint64_t **)&vs
, &vsc
) == 0);
2080 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
2081 vs
->vs_aux
= VDEV_AUX_SPARED
;
2088 * Add l2cache device information to the nvlist, including vdev stats.
2091 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
2094 uint_t i
, j
, nl2cache
;
2101 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2103 if (spa
->spa_l2cache
.sav_count
== 0)
2106 VERIFY(nvlist_lookup_nvlist(config
,
2107 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
2108 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
2109 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
2110 if (nl2cache
!= 0) {
2111 VERIFY(nvlist_add_nvlist_array(nvroot
,
2112 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
2113 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
2114 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
2117 * Update level 2 cache device stats.
2120 for (i
= 0; i
< nl2cache
; i
++) {
2121 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
2122 ZPOOL_CONFIG_GUID
, &guid
) == 0);
2125 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
2127 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
2128 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
2134 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
2135 ZPOOL_CONFIG_STATS
, (uint64_t **)&vs
, &vsc
) == 0);
2136 vdev_get_stats(vd
, vs
);
2142 spa_get_stats(const char *name
, nvlist_t
**config
, char *altroot
, size_t buflen
)
2148 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
2152 * This still leaves a window of inconsistency where the spares
2153 * or l2cache devices could change and the config would be
2154 * self-inconsistent.
2156 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
2158 if (*config
!= NULL
) {
2159 VERIFY(nvlist_add_uint64(*config
,
2160 ZPOOL_CONFIG_ERRCOUNT
,
2161 spa_get_errlog_size(spa
)) == 0);
2163 if (spa_suspended(spa
))
2164 VERIFY(nvlist_add_uint64(*config
,
2165 ZPOOL_CONFIG_SUSPENDED
,
2166 spa
->spa_failmode
) == 0);
2168 spa_add_spares(spa
, *config
);
2169 spa_add_l2cache(spa
, *config
);
2174 * We want to get the alternate root even for faulted pools, so we cheat
2175 * and call spa_lookup() directly.
2179 mutex_enter(&spa_namespace_lock
);
2180 spa
= spa_lookup(name
);
2182 spa_altroot(spa
, altroot
, buflen
);
2186 mutex_exit(&spa_namespace_lock
);
2188 spa_altroot(spa
, altroot
, buflen
);
2193 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
2194 spa_close(spa
, FTAG
);
2201 * Validate that the auxiliary device array is well formed. We must have an
2202 * array of nvlists, each which describes a valid leaf vdev. If this is an
2203 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2204 * specified, as long as they are well-formed.
2207 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
2208 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
2209 vdev_labeltype_t label
)
2216 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
2219 * It's acceptable to have no devs specified.
2221 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
2228 * Make sure the pool is formatted with a version that supports this
2231 if (spa_version(spa
) < version
)
2235 * Set the pending device list so we correctly handle device in-use
2238 sav
->sav_pending
= dev
;
2239 sav
->sav_npending
= ndev
;
2241 for (i
= 0; i
< ndev
; i
++) {
2242 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
2246 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2253 * The L2ARC currently only supports disk devices in
2254 * kernel context. For user-level testing, we allow it.
2257 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
2258 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
2265 if ((error
= vdev_open(vd
)) == 0 &&
2266 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
2267 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
2268 vd
->vdev_guid
) == 0);
2274 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
2281 sav
->sav_pending
= NULL
;
2282 sav
->sav_npending
= 0;
2287 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
2291 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
2293 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
2294 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
2295 VDEV_LABEL_SPARE
)) != 0) {
2299 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
2300 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
2301 VDEV_LABEL_L2CACHE
));
2305 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
2310 if (sav
->sav_config
!= NULL
) {
2316 * Generate new dev list by concatentating with the
2319 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
2320 &olddevs
, &oldndevs
) == 0);
2322 newdevs
= kmem_alloc(sizeof (void *) *
2323 (ndevs
+ oldndevs
), KM_SLEEP
);
2324 for (i
= 0; i
< oldndevs
; i
++)
2325 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
2327 for (i
= 0; i
< ndevs
; i
++)
2328 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
2331 VERIFY(nvlist_remove(sav
->sav_config
, config
,
2332 DATA_TYPE_NVLIST_ARRAY
) == 0);
2334 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
2335 config
, newdevs
, ndevs
+ oldndevs
) == 0);
2336 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
2337 nvlist_free(newdevs
[i
]);
2338 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
2341 * Generate a new dev list.
2343 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
2345 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
2351 * Stop and drop level 2 ARC devices
2354 spa_l2cache_drop(spa_t
*spa
)
2358 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
2360 for (i
= 0; i
< sav
->sav_count
; i
++) {
2363 vd
= sav
->sav_vdevs
[i
];
2366 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
2367 pool
!= 0ULL && l2arc_vdev_present(vd
))
2368 l2arc_remove_vdev(vd
);
2369 if (vd
->vdev_isl2cache
)
2370 spa_l2cache_remove(vd
);
2371 vdev_clear_stats(vd
);
2372 (void) vdev_close(vd
);
2380 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
2381 const char *history_str
, nvlist_t
*zplprops
)
2384 char *altroot
= NULL
;
2389 uint64_t txg
= TXG_INITIAL
;
2390 nvlist_t
**spares
, **l2cache
;
2391 uint_t nspares
, nl2cache
;
2395 * If this pool already exists, return failure.
2397 mutex_enter(&spa_namespace_lock
);
2398 if (spa_lookup(pool
) != NULL
) {
2399 mutex_exit(&spa_namespace_lock
);
2404 * Allocate a new spa_t structure.
2406 (void) nvlist_lookup_string(props
,
2407 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
2408 spa
= spa_add(pool
, NULL
, altroot
);
2409 spa_activate(spa
, spa_mode_global
);
2411 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
2412 spa_deactivate(spa
);
2414 mutex_exit(&spa_namespace_lock
);
2418 if (nvlist_lookup_uint64(props
, zpool_prop_to_name(ZPOOL_PROP_VERSION
),
2420 version
= SPA_VERSION
;
2421 ASSERT(version
<= SPA_VERSION
);
2423 spa
->spa_first_txg
= txg
;
2424 spa
->spa_uberblock
.ub_txg
= txg
- 1;
2425 spa
->spa_uberblock
.ub_version
= version
;
2426 spa
->spa_ubsync
= spa
->spa_uberblock
;
2429 * Create "The Godfather" zio to hold all async IOs
2431 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
2432 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
2435 * Create the root vdev.
2437 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2439 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
2441 ASSERT(error
!= 0 || rvd
!= NULL
);
2442 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
2444 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
2448 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
2449 (error
= spa_validate_aux(spa
, nvroot
, txg
,
2450 VDEV_ALLOC_ADD
)) == 0) {
2451 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2452 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
2453 vdev_expand(rvd
->vdev_child
[c
], txg
);
2457 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2461 spa_deactivate(spa
);
2463 mutex_exit(&spa_namespace_lock
);
2468 * Get the list of spares, if specified.
2470 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
2471 &spares
, &nspares
) == 0) {
2472 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
2474 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
2475 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
2476 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2477 spa_load_spares(spa
);
2478 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2479 spa
->spa_spares
.sav_sync
= B_TRUE
;
2483 * Get the list of level 2 cache devices, if specified.
2485 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
2486 &l2cache
, &nl2cache
) == 0) {
2487 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
2488 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2489 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
2490 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
2491 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2492 spa_load_l2cache(spa
);
2493 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2494 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2497 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
2498 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
2501 * Create DDTs (dedup tables).
2505 spa_update_dspace(spa
);
2507 tx
= dmu_tx_create_assigned(dp
, txg
);
2510 * Create the pool config object.
2512 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
2513 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
2514 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
2516 if (zap_add(spa
->spa_meta_objset
,
2517 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
2518 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
2519 cmn_err(CE_PANIC
, "failed to add pool config");
2522 /* Newly created pools with the right version are always deflated. */
2523 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
2524 spa
->spa_deflate
= TRUE
;
2525 if (zap_add(spa
->spa_meta_objset
,
2526 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
2527 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
2528 cmn_err(CE_PANIC
, "failed to add deflate");
2533 * Create the deferred-free bplist object. Turn off compression
2534 * because sync-to-convergence takes longer if the blocksize
2537 spa
->spa_deferred_bplist_obj
= bplist_create(spa
->spa_meta_objset
,
2539 dmu_object_set_compress(spa
->spa_meta_objset
,
2540 spa
->spa_deferred_bplist_obj
, ZIO_COMPRESS_OFF
, tx
);
2542 if (zap_add(spa
->spa_meta_objset
,
2543 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPLIST
,
2544 sizeof (uint64_t), 1, &spa
->spa_deferred_bplist_obj
, tx
) != 0) {
2545 cmn_err(CE_PANIC
, "failed to add bplist");
2549 * Create the pool's history object.
2551 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
2552 spa_history_create_obj(spa
, tx
);
2555 * Set pool properties.
2557 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
2558 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2559 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
2560 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
2562 if (props
!= NULL
) {
2563 spa_configfile_set(spa
, props
, B_FALSE
);
2564 spa_sync_props(spa
, props
, CRED(), tx
);
2569 spa
->spa_sync_on
= B_TRUE
;
2570 txg_sync_start(spa
->spa_dsl_pool
);
2573 * We explicitly wait for the first transaction to complete so that our
2574 * bean counters are appropriately updated.
2576 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
2578 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
2580 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& history_str
!= NULL
)
2581 (void) spa_history_log(spa
, history_str
, LOG_CMD_POOL_CREATE
);
2582 spa_history_log_version(spa
, LOG_POOL_CREATE
);
2584 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2586 mutex_exit(&spa_namespace_lock
);
2593 * Get the root pool information from the root disk, then import the root pool
2594 * during the system boot up time.
2596 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
2599 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
2602 nvlist_t
*nvtop
, *nvroot
;
2605 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
2609 * Add this top-level vdev to the child array.
2611 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
2613 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
2615 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
2618 * Put this pool's top-level vdevs into a root vdev.
2620 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2621 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
2622 VDEV_TYPE_ROOT
) == 0);
2623 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
2624 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
2625 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
2629 * Replace the existing vdev_tree with the new root vdev in
2630 * this pool's configuration (remove the old, add the new).
2632 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
2633 nvlist_free(nvroot
);
2638 * Walk the vdev tree and see if we can find a device with "better"
2639 * configuration. A configuration is "better" if the label on that
2640 * device has a more recent txg.
2643 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
2645 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2646 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
2648 if (vd
->vdev_ops
->vdev_op_leaf
) {
2652 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
2656 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
2660 * Do we have a better boot device?
2662 if (label_txg
> *txg
) {
2671 * Import a root pool.
2673 * For x86. devpath_list will consist of devid and/or physpath name of
2674 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2675 * The GRUB "findroot" command will return the vdev we should boot.
2677 * For Sparc, devpath_list consists the physpath name of the booting device
2678 * no matter the rootpool is a single device pool or a mirrored pool.
2680 * "/pci@1f,0/ide@d/disk@0,0:a"
2683 spa_import_rootpool(char *devpath
, char *devid
)
2686 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
2687 nvlist_t
*config
, *nvtop
;
2693 * Read the label from the boot device and generate a configuration.
2695 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
2696 #if defined(_OBP) && defined(_KERNEL)
2697 if (config
== NULL
) {
2698 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
2700 get_iscsi_bootpath_phy(devpath
);
2701 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
2705 if (config
== NULL
) {
2706 cmn_err(CE_NOTE
, "Can not read the pool label from '%s'",
2711 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
2713 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
2715 mutex_enter(&spa_namespace_lock
);
2716 if ((spa
= spa_lookup(pname
)) != NULL
) {
2718 * Remove the existing root pool from the namespace so that we
2719 * can replace it with the correct config we just read in.
2724 spa
= spa_add(pname
, config
, NULL
);
2725 spa
->spa_is_root
= B_TRUE
;
2726 spa
->spa_load_verbatim
= B_TRUE
;
2729 * Build up a vdev tree based on the boot device's label config.
2731 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
2733 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2734 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
2735 VDEV_ALLOC_ROOTPOOL
);
2736 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2738 mutex_exit(&spa_namespace_lock
);
2739 nvlist_free(config
);
2740 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
2746 * Get the boot vdev.
2748 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
2749 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
2750 (u_longlong_t
)guid
);
2756 * Determine if there is a better boot device.
2759 spa_alt_rootvdev(rvd
, &avd
, &txg
);
2761 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
2762 "try booting from '%s'", avd
->vdev_path
);
2768 * If the boot device is part of a spare vdev then ensure that
2769 * we're booting off the active spare.
2771 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2772 !bvd
->vdev_isspare
) {
2773 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
2774 "try booting from '%s'",
2775 bvd
->vdev_parent
->vdev_child
[1]->vdev_path
);
2781 spa_history_log_version(spa
, LOG_POOL_IMPORT
);
2783 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2785 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2786 mutex_exit(&spa_namespace_lock
);
2788 nvlist_free(config
);
2795 * Take a pool and insert it into the namespace as if it had been loaded at
2799 spa_import_verbatim(const char *pool
, nvlist_t
*config
, nvlist_t
*props
)
2802 zpool_rewind_policy_t policy
;
2803 char *altroot
= NULL
;
2805 mutex_enter(&spa_namespace_lock
);
2806 if (spa_lookup(pool
) != NULL
) {
2807 mutex_exit(&spa_namespace_lock
);
2811 (void) nvlist_lookup_string(props
,
2812 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
2813 spa
= spa_add(pool
, config
, altroot
);
2815 zpool_get_rewind_policy(config
, &policy
);
2816 spa
->spa_load_max_txg
= policy
.zrp_txg
;
2818 spa
->spa_load_verbatim
= B_TRUE
;
2821 spa_configfile_set(spa
, props
, B_FALSE
);
2823 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
2825 mutex_exit(&spa_namespace_lock
);
2826 spa_history_log_version(spa
, LOG_POOL_IMPORT
);
2832 * Import a non-root pool into the system.
2835 spa_import(const char *pool
, nvlist_t
*config
, nvlist_t
*props
)
2838 char *altroot
= NULL
;
2839 spa_load_state_t state
= SPA_LOAD_IMPORT
;
2840 zpool_rewind_policy_t policy
;
2843 nvlist_t
**spares
, **l2cache
;
2844 uint_t nspares
, nl2cache
;
2847 * If a pool with this name exists, return failure.
2849 mutex_enter(&spa_namespace_lock
);
2850 if ((spa
= spa_lookup(pool
)) != NULL
) {
2851 mutex_exit(&spa_namespace_lock
);
2855 zpool_get_rewind_policy(config
, &policy
);
2856 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
2857 state
= SPA_LOAD_RECOVER
;
2860 * Create and initialize the spa structure.
2862 (void) nvlist_lookup_string(props
,
2863 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
2864 spa
= spa_add(pool
, config
, altroot
);
2865 spa_activate(spa
, spa_mode_global
);
2868 * Don't start async tasks until we know everything is healthy.
2870 spa_async_suspend(spa
);
2873 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
2874 * because the user-supplied config is actually the one to trust when
2877 if (state
!= SPA_LOAD_RECOVER
)
2878 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
2879 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
2880 ((policy
.zrp_request
& ZPOOL_EXTREME_REWIND
) != 0));
2883 * Propagate anything learned about failing or best txgs
2886 spa_rewind_data_to_nvlist(spa
, config
);
2888 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2890 * Toss any existing sparelist, as it doesn't have any validity
2891 * anymore, and conflicts with spa_has_spare().
2893 if (spa
->spa_spares
.sav_config
) {
2894 nvlist_free(spa
->spa_spares
.sav_config
);
2895 spa
->spa_spares
.sav_config
= NULL
;
2896 spa_load_spares(spa
);
2898 if (spa
->spa_l2cache
.sav_config
) {
2899 nvlist_free(spa
->spa_l2cache
.sav_config
);
2900 spa
->spa_l2cache
.sav_config
= NULL
;
2901 spa_load_l2cache(spa
);
2904 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
2907 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
2910 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
2911 VDEV_ALLOC_L2CACHE
);
2912 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2915 spa_configfile_set(spa
, props
, B_FALSE
);
2917 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
2918 (error
= spa_prop_set(spa
, props
)))) {
2920 spa_deactivate(spa
);
2922 mutex_exit(&spa_namespace_lock
);
2926 spa_async_resume(spa
);
2929 * Override any spares and level 2 cache devices as specified by
2930 * the user, as these may have correct device names/devids, etc.
2932 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
2933 &spares
, &nspares
) == 0) {
2934 if (spa
->spa_spares
.sav_config
)
2935 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
2936 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
2938 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
2939 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2940 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
2941 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
2942 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2943 spa_load_spares(spa
);
2944 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2945 spa
->spa_spares
.sav_sync
= B_TRUE
;
2947 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
2948 &l2cache
, &nl2cache
) == 0) {
2949 if (spa
->spa_l2cache
.sav_config
)
2950 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
2951 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
2953 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
2954 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2955 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
2956 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
2957 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2958 spa_load_l2cache(spa
);
2959 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2960 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2964 * Check for any removed devices.
2966 if (spa
->spa_autoreplace
) {
2967 spa_aux_check_removed(&spa
->spa_spares
);
2968 spa_aux_check_removed(&spa
->spa_l2cache
);
2971 if (spa_writeable(spa
)) {
2973 * Update the config cache to include the newly-imported pool.
2975 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
2979 * It's possible that the pool was expanded while it was exported.
2980 * We kick off an async task to handle this for us.
2982 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
2984 mutex_exit(&spa_namespace_lock
);
2985 spa_history_log_version(spa
, LOG_POOL_IMPORT
);
2992 * This (illegal) pool name is used when temporarily importing a spa_t in order
2993 * to get the vdev stats associated with the imported devices.
2995 #define TRYIMPORT_NAME "$import"
2998 spa_tryimport(nvlist_t
*tryconfig
)
3000 nvlist_t
*config
= NULL
;
3006 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
3009 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
3013 * Create and initialize the spa structure.
3015 mutex_enter(&spa_namespace_lock
);
3016 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
3017 spa_activate(spa
, FREAD
);
3020 * Pass off the heavy lifting to spa_load().
3021 * Pass TRUE for mosconfig because the user-supplied config
3022 * is actually the one to trust when doing an import.
3024 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, B_TRUE
);
3027 * If 'tryconfig' was at least parsable, return the current config.
3029 if (spa
->spa_root_vdev
!= NULL
) {
3030 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3031 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3033 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
3035 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
3036 spa
->spa_uberblock
.ub_timestamp
) == 0);
3039 * If the bootfs property exists on this pool then we
3040 * copy it out so that external consumers can tell which
3041 * pools are bootable.
3043 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
3044 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
3047 * We have to play games with the name since the
3048 * pool was opened as TRYIMPORT_NAME.
3050 if (dsl_dsobj_to_dsname(spa_name(spa
),
3051 spa
->spa_bootfs
, tmpname
) == 0) {
3053 char *dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
3055 cp
= strchr(tmpname
, '/');
3057 (void) strlcpy(dsname
, tmpname
,
3060 (void) snprintf(dsname
, MAXPATHLEN
,
3061 "%s/%s", poolname
, ++cp
);
3063 VERIFY(nvlist_add_string(config
,
3064 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
3065 kmem_free(dsname
, MAXPATHLEN
);
3067 kmem_free(tmpname
, MAXPATHLEN
);
3071 * Add the list of hot spares and level 2 cache devices.
3073 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3074 spa_add_spares(spa
, config
);
3075 spa_add_l2cache(spa
, config
);
3076 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3080 spa_deactivate(spa
);
3082 mutex_exit(&spa_namespace_lock
);
3088 * Pool export/destroy
3090 * The act of destroying or exporting a pool is very simple. We make sure there
3091 * is no more pending I/O and any references to the pool are gone. Then, we
3092 * update the pool state and sync all the labels to disk, removing the
3093 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3094 * we don't sync the labels or remove the configuration cache.
3097 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
3098 boolean_t force
, boolean_t hardforce
)
3105 if (!(spa_mode_global
& FWRITE
))
3108 mutex_enter(&spa_namespace_lock
);
3109 if ((spa
= spa_lookup(pool
)) == NULL
) {
3110 mutex_exit(&spa_namespace_lock
);
3115 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3116 * reacquire the namespace lock, and see if we can export.
3118 spa_open_ref(spa
, FTAG
);
3119 mutex_exit(&spa_namespace_lock
);
3120 spa_async_suspend(spa
);
3121 mutex_enter(&spa_namespace_lock
);
3122 spa_close(spa
, FTAG
);
3125 * The pool will be in core if it's openable,
3126 * in which case we can modify its state.
3128 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
&& spa
->spa_sync_on
) {
3130 * Objsets may be open only because they're dirty, so we
3131 * have to force it to sync before checking spa_refcnt.
3133 txg_wait_synced(spa
->spa_dsl_pool
, 0);
3136 * A pool cannot be exported or destroyed if there are active
3137 * references. If we are resetting a pool, allow references by
3138 * fault injection handlers.
3140 if (!spa_refcount_zero(spa
) ||
3141 (spa
->spa_inject_ref
!= 0 &&
3142 new_state
!= POOL_STATE_UNINITIALIZED
)) {
3143 spa_async_resume(spa
);
3144 mutex_exit(&spa_namespace_lock
);
3149 * A pool cannot be exported if it has an active shared spare.
3150 * This is to prevent other pools stealing the active spare
3151 * from an exported pool. At user's own will, such pool can
3152 * be forcedly exported.
3154 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
3155 spa_has_active_shared_spare(spa
)) {
3156 spa_async_resume(spa
);
3157 mutex_exit(&spa_namespace_lock
);
3162 * We want this to be reflected on every label,
3163 * so mark them all dirty. spa_unload() will do the
3164 * final sync that pushes these changes out.
3166 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
3167 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3168 spa
->spa_state
= new_state
;
3169 spa
->spa_final_txg
= spa_last_synced_txg(spa
) + 1;
3170 vdev_config_dirty(spa
->spa_root_vdev
);
3171 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3175 spa_event_notify(spa
, NULL
, ESC_ZFS_POOL_DESTROY
);
3177 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
3179 spa_deactivate(spa
);
3182 if (oldconfig
&& spa
->spa_config
)
3183 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
3185 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
3187 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3190 mutex_exit(&spa_namespace_lock
);
3196 * Destroy a storage pool.
3199 spa_destroy(char *pool
)
3201 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
3206 * Export a storage pool.
3209 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
3210 boolean_t hardforce
)
3212 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
3217 * Similar to spa_export(), this unloads the spa_t without actually removing it
3218 * from the namespace in any way.
3221 spa_reset(char *pool
)
3223 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
3228 * ==========================================================================
3229 * Device manipulation
3230 * ==========================================================================
3234 * Add a device to a storage pool.
3237 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
3241 vdev_t
*rvd
= spa
->spa_root_vdev
;
3243 nvlist_t
**spares
, **l2cache
;
3244 uint_t nspares
, nl2cache
;
3246 txg
= spa_vdev_enter(spa
);
3248 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
3249 VDEV_ALLOC_ADD
)) != 0)
3250 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
3252 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
3254 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
3258 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
3262 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
3263 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
3265 if (vd
->vdev_children
!= 0 &&
3266 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
3267 return (spa_vdev_exit(spa
, vd
, txg
, error
));
3270 * We must validate the spares and l2cache devices after checking the
3271 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3273 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
3274 return (spa_vdev_exit(spa
, vd
, txg
, error
));
3277 * Transfer each new top-level vdev from vd to rvd.
3279 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3282 * Set the vdev id to the first hole, if one exists.
3284 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
3285 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
3286 vdev_free(rvd
->vdev_child
[id
]);
3290 tvd
= vd
->vdev_child
[c
];
3291 vdev_remove_child(vd
, tvd
);
3293 vdev_add_child(rvd
, tvd
);
3294 vdev_config_dirty(tvd
);
3298 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
3299 ZPOOL_CONFIG_SPARES
);
3300 spa_load_spares(spa
);
3301 spa
->spa_spares
.sav_sync
= B_TRUE
;
3304 if (nl2cache
!= 0) {
3305 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
3306 ZPOOL_CONFIG_L2CACHE
);
3307 spa_load_l2cache(spa
);
3308 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3312 * We have to be careful when adding new vdevs to an existing pool.
3313 * If other threads start allocating from these vdevs before we
3314 * sync the config cache, and we lose power, then upon reboot we may
3315 * fail to open the pool because there are DVAs that the config cache
3316 * can't translate. Therefore, we first add the vdevs without
3317 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3318 * and then let spa_config_update() initialize the new metaslabs.
3320 * spa_load() checks for added-but-not-initialized vdevs, so that
3321 * if we lose power at any point in this sequence, the remaining
3322 * steps will be completed the next time we load the pool.
3324 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
3326 mutex_enter(&spa_namespace_lock
);
3327 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
3328 mutex_exit(&spa_namespace_lock
);
3334 * Attach a device to a mirror. The arguments are the path to any device
3335 * in the mirror, and the nvroot for the new device. If the path specifies
3336 * a device that is not mirrored, we automatically insert the mirror vdev.
3338 * If 'replacing' is specified, the new device is intended to replace the
3339 * existing device; in this case the two devices are made into their own
3340 * mirror using the 'replacing' vdev, which is functionally identical to
3341 * the mirror vdev (it actually reuses all the same ops) but has a few
3342 * extra rules: you can't attach to it after it's been created, and upon
3343 * completion of resilvering, the first disk (the one being replaced)
3344 * is automatically detached.
3347 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
3349 uint64_t txg
, open_txg
;
3350 vdev_t
*rvd
= spa
->spa_root_vdev
;
3351 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
3353 char *oldvdpath
, *newvdpath
;
3357 txg
= spa_vdev_enter(spa
);
3359 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
3362 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
3364 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
3365 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3367 pvd
= oldvd
->vdev_parent
;
3369 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
3370 VDEV_ALLOC_ADD
)) != 0)
3371 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
3373 if (newrootvd
->vdev_children
!= 1)
3374 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
3376 newvd
= newrootvd
->vdev_child
[0];
3378 if (!newvd
->vdev_ops
->vdev_op_leaf
)
3379 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
3381 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
3382 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
3385 * Spares can't replace logs
3387 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
3388 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3392 * For attach, the only allowable parent is a mirror or the root
3395 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
3396 pvd
->vdev_ops
!= &vdev_root_ops
)
3397 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3399 pvops
= &vdev_mirror_ops
;
3402 * Active hot spares can only be replaced by inactive hot
3405 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
3406 pvd
->vdev_child
[1] == oldvd
&&
3407 !spa_has_spare(spa
, newvd
->vdev_guid
))
3408 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3411 * If the source is a hot spare, and the parent isn't already a
3412 * spare, then we want to create a new hot spare. Otherwise, we
3413 * want to create a replacing vdev. The user is not allowed to
3414 * attach to a spared vdev child unless the 'isspare' state is
3415 * the same (spare replaces spare, non-spare replaces
3418 if (pvd
->vdev_ops
== &vdev_replacing_ops
)
3419 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3420 else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
3421 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
)
3422 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3423 else if (pvd
->vdev_ops
!= &vdev_spare_ops
&&
3424 newvd
->vdev_isspare
)
3425 pvops
= &vdev_spare_ops
;
3427 pvops
= &vdev_replacing_ops
;
3431 * Make sure the new device is big enough.
3433 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
3434 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
3437 * The new device cannot have a higher alignment requirement
3438 * than the top-level vdev.
3440 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
3441 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
3444 * If this is an in-place replacement, update oldvd's path and devid
3445 * to make it distinguishable from newvd, and unopenable from now on.
3447 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
3448 spa_strfree(oldvd
->vdev_path
);
3449 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
3451 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
3452 newvd
->vdev_path
, "old");
3453 if (oldvd
->vdev_devid
!= NULL
) {
3454 spa_strfree(oldvd
->vdev_devid
);
3455 oldvd
->vdev_devid
= NULL
;
3460 * If the parent is not a mirror, or if we're replacing, insert the new
3461 * mirror/replacing/spare vdev above oldvd.
3463 if (pvd
->vdev_ops
!= pvops
)
3464 pvd
= vdev_add_parent(oldvd
, pvops
);
3466 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
3467 ASSERT(pvd
->vdev_ops
== pvops
);
3468 ASSERT(oldvd
->vdev_parent
== pvd
);
3471 * Extract the new device from its root and add it to pvd.
3473 vdev_remove_child(newrootvd
, newvd
);
3474 newvd
->vdev_id
= pvd
->vdev_children
;
3475 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
3476 vdev_add_child(pvd
, newvd
);
3478 tvd
= newvd
->vdev_top
;
3479 ASSERT(pvd
->vdev_top
== tvd
);
3480 ASSERT(tvd
->vdev_parent
== rvd
);
3482 vdev_config_dirty(tvd
);
3485 * Set newvd's DTL to [TXG_INITIAL, open_txg]. It will propagate
3486 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3488 open_txg
= txg
+ TXG_CONCURRENT_STATES
- 1;
3490 vdev_dtl_dirty(newvd
, DTL_MISSING
,
3491 TXG_INITIAL
, open_txg
- TXG_INITIAL
+ 1);
3493 if (newvd
->vdev_isspare
) {
3494 spa_spare_activate(newvd
);
3495 spa_event_notify(spa
, newvd
, ESC_ZFS_VDEV_SPARE
);
3498 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
3499 newvdpath
= spa_strdup(newvd
->vdev_path
);
3500 newvd_isspare
= newvd
->vdev_isspare
;
3503 * Mark newvd's DTL dirty in this txg.
3505 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
3507 (void) spa_vdev_exit(spa
, newrootvd
, open_txg
, 0);
3509 spa_history_internal_log(LOG_POOL_VDEV_ATTACH
, spa
, NULL
,
3510 CRED(), "%s vdev=%s %s vdev=%s",
3511 replacing
&& newvd_isspare
? "spare in" :
3512 replacing
? "replace" : "attach", newvdpath
,
3513 replacing
? "for" : "to", oldvdpath
);
3515 spa_strfree(oldvdpath
);
3516 spa_strfree(newvdpath
);
3519 * Kick off a resilver to update newvd.
3521 VERIFY3U(spa_scrub(spa
, POOL_SCRUB_RESILVER
), ==, 0);
3527 * Detach a device from a mirror or replacing vdev.
3528 * If 'replace_done' is specified, only detach if the parent
3529 * is a replacing vdev.
3532 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
3536 vdev_t
*rvd
= spa
->spa_root_vdev
;
3537 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
3538 boolean_t unspare
= B_FALSE
;
3539 uint64_t unspare_guid
;
3542 txg
= spa_vdev_enter(spa
);
3544 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
3547 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
3549 if (!vd
->vdev_ops
->vdev_op_leaf
)
3550 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3552 pvd
= vd
->vdev_parent
;
3555 * If the parent/child relationship is not as expected, don't do it.
3556 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3557 * vdev that's replacing B with C. The user's intent in replacing
3558 * is to go from M(A,B) to M(A,C). If the user decides to cancel
3559 * the replace by detaching C, the expected behavior is to end up
3560 * M(A,B). But suppose that right after deciding to detach C,
3561 * the replacement of B completes. We would have M(A,C), and then
3562 * ask to detach C, which would leave us with just A -- not what
3563 * the user wanted. To prevent this, we make sure that the
3564 * parent/child relationship hasn't changed -- in this example,
3565 * that C's parent is still the replacing vdev R.
3567 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
3568 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
3571 * If replace_done is specified, only remove this device if it's
3572 * the first child of a replacing vdev. For the 'spare' vdev, either
3573 * disk can be removed.
3576 if (pvd
->vdev_ops
== &vdev_replacing_ops
) {
3577 if (vd
->vdev_id
!= 0)
3578 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3579 } else if (pvd
->vdev_ops
!= &vdev_spare_ops
) {
3580 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3584 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
3585 spa_version(spa
) >= SPA_VERSION_SPARES
);
3588 * Only mirror, replacing, and spare vdevs support detach.
3590 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
3591 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
3592 pvd
->vdev_ops
!= &vdev_spare_ops
)
3593 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3596 * If this device has the only valid copy of some data,
3597 * we cannot safely detach it.
3599 if (vdev_dtl_required(vd
))
3600 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
3602 ASSERT(pvd
->vdev_children
>= 2);
3605 * If we are detaching the second disk from a replacing vdev, then
3606 * check to see if we changed the original vdev's path to have "/old"
3607 * at the end in spa_vdev_attach(). If so, undo that change now.
3609 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
== 1 &&
3610 pvd
->vdev_child
[0]->vdev_path
!= NULL
&&
3611 pvd
->vdev_child
[1]->vdev_path
!= NULL
) {
3612 ASSERT(pvd
->vdev_child
[1] == vd
);
3613 cvd
= pvd
->vdev_child
[0];
3614 len
= strlen(vd
->vdev_path
);
3615 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
3616 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
3617 spa_strfree(cvd
->vdev_path
);
3618 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
3623 * If we are detaching the original disk from a spare, then it implies
3624 * that the spare should become a real disk, and be removed from the
3625 * active spare list for the pool.
3627 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
3628 vd
->vdev_id
== 0 && pvd
->vdev_child
[1]->vdev_isspare
)
3632 * Erase the disk labels so the disk can be used for other things.
3633 * This must be done after all other error cases are handled,
3634 * but before we disembowel vd (so we can still do I/O to it).
3635 * But if we can't do it, don't treat the error as fatal --
3636 * it may be that the unwritability of the disk is the reason
3637 * it's being detached!
3639 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
3642 * Remove vd from its parent and compact the parent's children.
3644 vdev_remove_child(pvd
, vd
);
3645 vdev_compact_children(pvd
);
3648 * Remember one of the remaining children so we can get tvd below.
3650 cvd
= pvd
->vdev_child
[0];
3653 * If we need to remove the remaining child from the list of hot spares,
3654 * do it now, marking the vdev as no longer a spare in the process.
3655 * We must do this before vdev_remove_parent(), because that can
3656 * change the GUID if it creates a new toplevel GUID. For a similar
3657 * reason, we must remove the spare now, in the same txg as the detach;
3658 * otherwise someone could attach a new sibling, change the GUID, and
3659 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3662 ASSERT(cvd
->vdev_isspare
);
3663 spa_spare_remove(cvd
);
3664 unspare_guid
= cvd
->vdev_guid
;
3665 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
3669 * If the parent mirror/replacing vdev only has one child,
3670 * the parent is no longer needed. Remove it from the tree.
3672 if (pvd
->vdev_children
== 1)
3673 vdev_remove_parent(cvd
);
3676 * We don't set tvd until now because the parent we just removed
3677 * may have been the previous top-level vdev.
3679 tvd
= cvd
->vdev_top
;
3680 ASSERT(tvd
->vdev_parent
== rvd
);
3683 * Reevaluate the parent vdev state.
3685 vdev_propagate_state(cvd
);
3688 * If the 'autoexpand' property is set on the pool then automatically
3689 * try to expand the size of the pool. For example if the device we
3690 * just detached was smaller than the others, it may be possible to
3691 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3692 * first so that we can obtain the updated sizes of the leaf vdevs.
3694 if (spa
->spa_autoexpand
) {
3696 vdev_expand(tvd
, txg
);
3699 vdev_config_dirty(tvd
);
3702 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
3703 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3704 * But first make sure we're not on any *other* txg's DTL list, to
3705 * prevent vd from being accessed after it's freed.
3707 for (int t
= 0; t
< TXG_SIZE
; t
++)
3708 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
3709 vd
->vdev_detached
= B_TRUE
;
3710 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
3712 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_REMOVE
);
3714 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
3717 * If this was the removal of the original device in a hot spare vdev,
3718 * then we want to go through and remove the device from the hot spare
3719 * list of every other pool.
3724 mutex_enter(&spa_namespace_lock
);
3725 while ((spa
= spa_next(spa
)) != NULL
) {
3726 if (spa
->spa_state
!= POOL_STATE_ACTIVE
)
3730 spa_open_ref(spa
, FTAG
);
3731 mutex_exit(&spa_namespace_lock
);
3732 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
3733 mutex_enter(&spa_namespace_lock
);
3734 spa_close(spa
, FTAG
);
3736 mutex_exit(&spa_namespace_lock
);
3743 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
3745 for (int i
= 0; i
< count
; i
++) {
3748 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
3751 if (guid
== target_guid
)
3759 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
3760 nvlist_t
*dev_to_remove
)
3762 nvlist_t
**newdev
= NULL
;
3765 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
3767 for (int i
= 0, j
= 0; i
< count
; i
++) {
3768 if (dev
[i
] == dev_to_remove
)
3770 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
3773 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3774 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
3776 for (int i
= 0; i
< count
- 1; i
++)
3777 nvlist_free(newdev
[i
]);
3780 kmem_free(newdev
, (count
- 1) * sizeof (void *));
3784 * Removing a device from the vdev namespace requires several steps
3785 * and can take a significant amount of time. As a result we use
3786 * the spa_vdev_config_[enter/exit] functions which allow us to
3787 * grab and release the spa_config_lock while still holding the namespace
3788 * lock. During each step the configuration is synced out.
3792 * Initial phase of device removal - stop future allocations from this device.
3795 spa_vdev_remove_start(spa_t
*spa
, vdev_t
*vd
)
3797 metaslab_group_t
*mg
= vd
->vdev_mg
;
3799 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
3800 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3801 ASSERT(vd
== vd
->vdev_top
);
3804 * Remove our vdev from the allocatable vdevs
3807 metaslab_class_remove(mg
->mg_class
, mg
);
3811 * Evacuate the device.
3814 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
3819 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
3820 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
3821 ASSERT(vd
== vd
->vdev_top
);
3824 * Evacuate the device. We don't hold the config lock as writer
3825 * since we need to do I/O but we do keep the
3826 * spa_namespace_lock held. Once this completes the device
3827 * should no longer have any blocks allocated on it.
3829 if (vd
->vdev_islog
) {
3831 * Evacuate the device.
3833 if (error
= dmu_objset_find(spa_name(spa
),
3834 zil_vdev_offline
, NULL
, DS_FIND_CHILDREN
)) {
3837 txg
= spa_vdev_config_enter(spa
);
3838 metaslab_class_add(spa
->spa_log_class
,
3840 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
3842 txg_wait_synced(spa_get_dsl(spa
), 0);
3846 * Remove any remaining MOS metadata associated with the device.
3848 txg
= spa_vdev_config_enter(spa
);
3849 vd
->vdev_removing
= B_TRUE
;
3850 vdev_dirty(vd
, 0, NULL
, txg
);
3851 vdev_config_dirty(vd
);
3852 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
3858 * Complete the removal by cleaning up the namespace.
3861 spa_vdev_remove_done(spa_t
*spa
, vdev_t
*vd
)
3863 vdev_t
*rvd
= spa
->spa_root_vdev
;
3864 metaslab_group_t
*mg
= vd
->vdev_mg
;
3865 uint64_t id
= vd
->vdev_id
;
3866 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
3868 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
3869 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3870 ASSERT(vd
== vd
->vdev_top
);
3872 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
3874 if (list_link_active(&vd
->vdev_state_dirty_node
))
3875 vdev_state_clean(vd
);
3876 if (list_link_active(&vd
->vdev_config_dirty_node
))
3877 vdev_config_clean(vd
);
3882 * It's possible that another thread is trying todo a spa_vdev_add()
3883 * at the same time we're trying remove it. As a result the
3884 * added vdev may not have initialized its metaslabs yet.
3887 metaslab_group_destroy(mg
);
3890 vdev_compact_children(rvd
);
3892 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
3893 vdev_add_child(rvd
, vd
);
3895 vdev_config_dirty(rvd
);
3898 * Reassess the health of our root vdev.
3904 * Remove a device from the pool. Currently, this supports removing only hot
3905 * spares, slogs, and level 2 ARC devices.
3908 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
3911 nvlist_t
**spares
, **l2cache
, *nv
;
3913 uint_t nspares
, nl2cache
;
3915 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
3918 txg
= spa_vdev_enter(spa
);
3920 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
3922 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
3923 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3924 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
3925 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
3927 * Only remove the hot spare if it's not currently in use
3930 if (vd
== NULL
|| unspare
) {
3931 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
3932 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
3933 spa_load_spares(spa
);
3934 spa
->spa_spares
.sav_sync
= B_TRUE
;
3938 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
3939 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3940 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
3941 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
3943 * Cache devices can always be removed.
3945 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
3946 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
3947 spa_load_l2cache(spa
);
3948 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3949 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
3951 ASSERT(vd
== vd
->vdev_top
);
3954 * XXX - Once we have bp-rewrite this should
3955 * become the common case.
3959 * 1. Stop allocations
3960 * 2. Evacuate the device (i.e. kill off stubby and
3961 * metadata) and wait for it to complete (i.e. sync).
3962 * 3. Cleanup the vdev namespace.
3964 spa_vdev_remove_start(spa
, vd
);
3967 * Wait for the youngest allocations and frees to sync,
3968 * and then wait for the deferral of those frees to finish.
3970 spa_vdev_config_exit(spa
, NULL
,
3971 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
3973 if ((error
= spa_vdev_remove_evacuate(spa
, vd
)) != 0)
3975 txg
= spa_vdev_config_enter(spa
);
3977 spa_vdev_remove_done(spa
, vd
);
3979 } else if (vd
!= NULL
) {
3981 * Normal vdevs cannot be removed (yet).
3986 * There is no vdev of any kind with the specified guid.
3992 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
3998 * Find any device that's done replacing, or a vdev marked 'unspare' that's
3999 * current spared, so we can detach it.
4002 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
4004 vdev_t
*newvd
, *oldvd
;
4006 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4007 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
4013 * Check for a completed replacement.
4015 if (vd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_children
== 2) {
4016 oldvd
= vd
->vdev_child
[0];
4017 newvd
= vd
->vdev_child
[1];
4019 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
4020 !vdev_dtl_required(oldvd
))
4025 * Check for a completed resilver with the 'unspare' flag set.
4027 if (vd
->vdev_ops
== &vdev_spare_ops
&& vd
->vdev_children
== 2) {
4028 newvd
= vd
->vdev_child
[0];
4029 oldvd
= vd
->vdev_child
[1];
4031 if (newvd
->vdev_unspare
&&
4032 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
4033 !vdev_dtl_required(oldvd
)) {
4034 newvd
->vdev_unspare
= 0;
4043 spa_vdev_resilver_done(spa_t
*spa
)
4045 vdev_t
*vd
, *pvd
, *ppvd
;
4046 uint64_t guid
, sguid
, pguid
, ppguid
;
4048 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4050 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
4051 pvd
= vd
->vdev_parent
;
4052 ppvd
= pvd
->vdev_parent
;
4053 guid
= vd
->vdev_guid
;
4054 pguid
= pvd
->vdev_guid
;
4055 ppguid
= ppvd
->vdev_guid
;
4058 * If we have just finished replacing a hot spared device, then
4059 * we need to detach the parent's first child (the original hot
4062 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0) {
4063 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
4064 ASSERT(ppvd
->vdev_children
== 2);
4065 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
4067 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4068 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
4070 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
4072 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4075 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4079 * Update the stored path or FRU for this vdev. Dirty the vdev configuration,
4080 * relying on spa_vdev_enter/exit() to synchronize the labels and cache.
4083 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
4089 txg
= spa_vdev_enter(spa
);
4091 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4092 return (spa_vdev_exit(spa
, NULL
, txg
, ENOENT
));
4094 if (!vd
->vdev_ops
->vdev_op_leaf
)
4095 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4098 spa_strfree(vd
->vdev_path
);
4099 vd
->vdev_path
= spa_strdup(value
);
4101 if (vd
->vdev_fru
!= NULL
)
4102 spa_strfree(vd
->vdev_fru
);
4103 vd
->vdev_fru
= spa_strdup(value
);
4106 vdev_config_dirty(vd
->vdev_top
);
4108 return (spa_vdev_exit(spa
, NULL
, txg
, 0));
4112 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
4114 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
4118 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
4120 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
4124 * ==========================================================================
4126 * ==========================================================================
4130 spa_scrub(spa_t
*spa
, pool_scrub_type_t type
)
4132 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
4134 if ((uint_t
)type
>= POOL_SCRUB_TYPES
)
4138 * If a resilver was requested, but there is no DTL on a
4139 * writeable leaf device, we have nothing to do.
4141 if (type
== POOL_SCRUB_RESILVER
&&
4142 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
4143 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
4147 if (type
== POOL_SCRUB_EVERYTHING
&&
4148 spa
->spa_dsl_pool
->dp_scrub_func
!= SCRUB_FUNC_NONE
&&
4149 spa
->spa_dsl_pool
->dp_scrub_isresilver
)
4152 if (type
== POOL_SCRUB_EVERYTHING
|| type
== POOL_SCRUB_RESILVER
) {
4153 return (dsl_pool_scrub_clean(spa
->spa_dsl_pool
));
4154 } else if (type
== POOL_SCRUB_NONE
) {
4155 return (dsl_pool_scrub_cancel(spa
->spa_dsl_pool
));
4162 * ==========================================================================
4163 * SPA async task processing
4164 * ==========================================================================
4168 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
4170 if (vd
->vdev_remove_wanted
) {
4171 vd
->vdev_remove_wanted
= 0;
4172 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
4175 * We want to clear the stats, but we don't want to do a full
4176 * vdev_clear() as that will cause us to throw away
4177 * degraded/faulted state as well as attempt to reopen the
4178 * device, all of which is a waste.
4180 vd
->vdev_stat
.vs_read_errors
= 0;
4181 vd
->vdev_stat
.vs_write_errors
= 0;
4182 vd
->vdev_stat
.vs_checksum_errors
= 0;
4184 vdev_state_dirty(vd
->vdev_top
);
4187 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4188 spa_async_remove(spa
, vd
->vdev_child
[c
]);
4192 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
4194 if (vd
->vdev_probe_wanted
) {
4195 vd
->vdev_probe_wanted
= 0;
4196 vdev_reopen(vd
); /* vdev_open() does the actual probe */
4199 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4200 spa_async_probe(spa
, vd
->vdev_child
[c
]);
4204 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
4210 if (!spa
->spa_autoexpand
)
4213 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4214 vdev_t
*cvd
= vd
->vdev_child
[c
];
4215 spa_async_autoexpand(spa
, cvd
);
4218 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
4221 physpath
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
4222 (void) snprintf(physpath
, MAXPATHLEN
, "/devices%s", vd
->vdev_physpath
);
4224 VERIFY(nvlist_alloc(&attr
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4225 VERIFY(nvlist_add_string(attr
, DEV_PHYS_PATH
, physpath
) == 0);
4227 (void) ddi_log_sysevent(zfs_dip
, SUNW_VENDOR
, EC_DEV_STATUS
,
4228 ESC_DEV_DLE
, attr
, &eid
, DDI_SLEEP
);
4231 kmem_free(physpath
, MAXPATHLEN
);
4235 spa_async_thread(spa_t
*spa
)
4239 ASSERT(spa
->spa_sync_on
);
4241 mutex_enter(&spa
->spa_async_lock
);
4242 tasks
= spa
->spa_async_tasks
;
4243 spa
->spa_async_tasks
= 0;
4244 mutex_exit(&spa
->spa_async_lock
);
4247 * See if the config needs to be updated.
4249 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
4250 uint64_t old_space
, new_space
;
4252 mutex_enter(&spa_namespace_lock
);
4253 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
4254 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4255 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
4256 mutex_exit(&spa_namespace_lock
);
4259 * If the pool grew as a result of the config update,
4260 * then log an internal history event.
4262 if (new_space
!= old_space
) {
4263 spa_history_internal_log(LOG_POOL_VDEV_ONLINE
,
4265 "pool '%s' size: %llu(+%llu)",
4266 spa_name(spa
), new_space
, new_space
- old_space
);
4271 * See if any devices need to be marked REMOVED.
4273 if (tasks
& SPA_ASYNC_REMOVE
) {
4274 spa_vdev_state_enter(spa
, SCL_NONE
);
4275 spa_async_remove(spa
, spa
->spa_root_vdev
);
4276 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
4277 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
4278 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
4279 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
4280 (void) spa_vdev_state_exit(spa
, NULL
, 0);
4283 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
4284 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4285 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
4286 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4290 * See if any devices need to be probed.
4292 if (tasks
& SPA_ASYNC_PROBE
) {
4293 spa_vdev_state_enter(spa
, SCL_NONE
);
4294 spa_async_probe(spa
, spa
->spa_root_vdev
);
4295 (void) spa_vdev_state_exit(spa
, NULL
, 0);
4299 * If any devices are done replacing, detach them.
4301 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
4302 spa_vdev_resilver_done(spa
);
4305 * Kick off a resilver.
4307 if (tasks
& SPA_ASYNC_RESILVER
)
4308 VERIFY(spa_scrub(spa
, POOL_SCRUB_RESILVER
) == 0);
4311 * Let the world know that we're done.
4313 mutex_enter(&spa
->spa_async_lock
);
4314 spa
->spa_async_thread
= NULL
;
4315 cv_broadcast(&spa
->spa_async_cv
);
4316 mutex_exit(&spa
->spa_async_lock
);
4321 spa_async_suspend(spa_t
*spa
)
4323 mutex_enter(&spa
->spa_async_lock
);
4324 spa
->spa_async_suspended
++;
4325 while (spa
->spa_async_thread
!= NULL
)
4326 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
4327 mutex_exit(&spa
->spa_async_lock
);
4331 spa_async_resume(spa_t
*spa
)
4333 mutex_enter(&spa
->spa_async_lock
);
4334 ASSERT(spa
->spa_async_suspended
!= 0);
4335 spa
->spa_async_suspended
--;
4336 mutex_exit(&spa
->spa_async_lock
);
4340 spa_async_dispatch(spa_t
*spa
)
4342 mutex_enter(&spa
->spa_async_lock
);
4343 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
4344 spa
->spa_async_thread
== NULL
&&
4345 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
4346 spa
->spa_async_thread
= thread_create(NULL
, 0,
4347 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
4348 mutex_exit(&spa
->spa_async_lock
);
4352 spa_async_request(spa_t
*spa
, int task
)
4354 mutex_enter(&spa
->spa_async_lock
);
4355 spa
->spa_async_tasks
|= task
;
4356 mutex_exit(&spa
->spa_async_lock
);
4360 * ==========================================================================
4361 * SPA syncing routines
4362 * ==========================================================================
4365 spa_sync_deferred_bplist(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
, uint64_t txg
)
4371 while (bplist_iterate(bpl
, &itor
, &blk
) == 0) {
4372 ASSERT(blk
.blk_birth
< txg
);
4373 zio_free(spa
, txg
, &blk
);
4376 bplist_vacate(bpl
, tx
);
4379 * Pre-dirty the first block so we sync to convergence faster.
4380 * (Usually only the first block is needed.)
4382 dmu_write(bpl
->bpl_mos
, spa
->spa_deferred_bplist_obj
, 0, 1, &c
, tx
);
4386 spa_sync_free(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
4390 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
4395 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
4397 char *packed
= NULL
;
4402 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
4405 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4406 * information. This avoids the dbuf_will_dirty() path and
4407 * saves us a pre-read to get data we don't actually care about.
4409 bufsize
= P2ROUNDUP(nvsize
, SPA_CONFIG_BLOCKSIZE
);
4410 packed
= kmem_alloc(bufsize
, KM_SLEEP
);
4412 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
4414 bzero(packed
+ nvsize
, bufsize
- nvsize
);
4416 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
4418 kmem_free(packed
, bufsize
);
4420 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
4421 dmu_buf_will_dirty(db
, tx
);
4422 *(uint64_t *)db
->db_data
= nvsize
;
4423 dmu_buf_rele(db
, FTAG
);
4427 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
4428 const char *config
, const char *entry
)
4438 * Update the MOS nvlist describing the list of available devices.
4439 * spa_validate_aux() will have already made sure this nvlist is
4440 * valid and the vdevs are labeled appropriately.
4442 if (sav
->sav_object
== 0) {
4443 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4444 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
4445 sizeof (uint64_t), tx
);
4446 VERIFY(zap_update(spa
->spa_meta_objset
,
4447 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
4448 &sav
->sav_object
, tx
) == 0);
4451 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4452 if (sav
->sav_count
== 0) {
4453 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
4455 list
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
4456 for (i
= 0; i
< sav
->sav_count
; i
++)
4457 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
4458 B_FALSE
, B_FALSE
, B_TRUE
);
4459 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
4460 sav
->sav_count
) == 0);
4461 for (i
= 0; i
< sav
->sav_count
; i
++)
4462 nvlist_free(list
[i
]);
4463 kmem_free(list
, sav
->sav_count
* sizeof (void *));
4466 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
4467 nvlist_free(nvroot
);
4469 sav
->sav_sync
= B_FALSE
;
4473 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
4477 if (list_is_empty(&spa
->spa_config_dirty_list
))
4480 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4482 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
4483 dmu_tx_get_txg(tx
), B_FALSE
);
4485 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4487 if (spa
->spa_config_syncing
)
4488 nvlist_free(spa
->spa_config_syncing
);
4489 spa
->spa_config_syncing
= config
;
4491 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
4495 * Set zpool properties.
4498 spa_sync_props(void *arg1
, void *arg2
, cred_t
*cr
, dmu_tx_t
*tx
)
4501 objset_t
*mos
= spa
->spa_meta_objset
;
4502 nvlist_t
*nvp
= arg2
;
4507 const char *propname
;
4508 zprop_type_t proptype
;
4510 mutex_enter(&spa
->spa_props_lock
);
4513 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
4514 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
4515 case ZPOOL_PROP_VERSION
:
4517 * Only set version for non-zpool-creation cases
4518 * (set/import). spa_create() needs special care
4519 * for version setting.
4521 if (tx
->tx_txg
!= TXG_INITIAL
) {
4522 VERIFY(nvpair_value_uint64(elem
,
4524 ASSERT(intval
<= SPA_VERSION
);
4525 ASSERT(intval
>= spa_version(spa
));
4526 spa
->spa_uberblock
.ub_version
= intval
;
4527 vdev_config_dirty(spa
->spa_root_vdev
);
4531 case ZPOOL_PROP_ALTROOT
:
4533 * 'altroot' is a non-persistent property. It should
4534 * have been set temporarily at creation or import time.
4536 ASSERT(spa
->spa_root
!= NULL
);
4539 case ZPOOL_PROP_CACHEFILE
:
4541 * 'cachefile' is also a non-persisitent property.
4546 * Set pool property values in the poolprops mos object.
4548 if (spa
->spa_pool_props_object
== 0) {
4549 VERIFY((spa
->spa_pool_props_object
=
4550 zap_create(mos
, DMU_OT_POOL_PROPS
,
4551 DMU_OT_NONE
, 0, tx
)) > 0);
4553 VERIFY(zap_update(mos
,
4554 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
4555 8, 1, &spa
->spa_pool_props_object
, tx
)
4559 /* normalize the property name */
4560 propname
= zpool_prop_to_name(prop
);
4561 proptype
= zpool_prop_get_type(prop
);
4563 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
4564 ASSERT(proptype
== PROP_TYPE_STRING
);
4565 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
4566 VERIFY(zap_update(mos
,
4567 spa
->spa_pool_props_object
, propname
,
4568 1, strlen(strval
) + 1, strval
, tx
) == 0);
4570 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
4571 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
4573 if (proptype
== PROP_TYPE_INDEX
) {
4575 VERIFY(zpool_prop_index_to_string(
4576 prop
, intval
, &unused
) == 0);
4578 VERIFY(zap_update(mos
,
4579 spa
->spa_pool_props_object
, propname
,
4580 8, 1, &intval
, tx
) == 0);
4582 ASSERT(0); /* not allowed */
4586 case ZPOOL_PROP_DELEGATION
:
4587 spa
->spa_delegation
= intval
;
4589 case ZPOOL_PROP_BOOTFS
:
4590 spa
->spa_bootfs
= intval
;
4592 case ZPOOL_PROP_FAILUREMODE
:
4593 spa
->spa_failmode
= intval
;
4595 case ZPOOL_PROP_AUTOEXPAND
:
4596 spa
->spa_autoexpand
= intval
;
4597 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4599 case ZPOOL_PROP_DEDUPDITTO
:
4600 spa
->spa_dedup_ditto
= intval
;
4607 /* log internal history if this is not a zpool create */
4608 if (spa_version(spa
) >= SPA_VERSION_ZPOOL_HISTORY
&&
4609 tx
->tx_txg
!= TXG_INITIAL
) {
4610 spa_history_internal_log(LOG_POOL_PROPSET
,
4611 spa
, tx
, cr
, "%s %lld %s",
4612 nvpair_name(elem
), intval
, spa_name(spa
));
4616 mutex_exit(&spa
->spa_props_lock
);
4620 * Sync the specified transaction group. New blocks may be dirtied as
4621 * part of the process, so we iterate until it converges.
4624 spa_sync(spa_t
*spa
, uint64_t txg
)
4626 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
4627 objset_t
*mos
= spa
->spa_meta_objset
;
4628 bplist_t
*defer_bpl
= &spa
->spa_deferred_bplist
;
4629 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
4630 vdev_t
*rvd
= spa
->spa_root_vdev
;
4636 * Lock out configuration changes.
4638 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4640 spa
->spa_syncing_txg
= txg
;
4641 spa
->spa_sync_pass
= 0;
4644 * If there are any pending vdev state changes, convert them
4645 * into config changes that go out with this transaction group.
4647 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4648 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
4650 * We need the write lock here because, for aux vdevs,
4651 * calling vdev_config_dirty() modifies sav_config.
4652 * This is ugly and will become unnecessary when we
4653 * eliminate the aux vdev wart by integrating all vdevs
4654 * into the root vdev tree.
4656 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
4657 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
4658 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
4659 vdev_state_clean(vd
);
4660 vdev_config_dirty(vd
);
4662 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
4663 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
4665 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4667 VERIFY(0 == bplist_open(defer_bpl
, mos
, spa
->spa_deferred_bplist_obj
));
4669 tx
= dmu_tx_create_assigned(dp
, txg
);
4672 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
4673 * set spa_deflate if we have no raid-z vdevs.
4675 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
4676 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4679 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
4680 vd
= rvd
->vdev_child
[i
];
4681 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
4684 if (i
== rvd
->vdev_children
) {
4685 spa
->spa_deflate
= TRUE
;
4686 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
4687 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4688 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
4692 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
4693 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
4694 dsl_pool_create_origin(dp
, tx
);
4696 /* Keeping the origin open increases spa_minref */
4697 spa
->spa_minref
+= 3;
4700 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
4701 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
4702 dsl_pool_upgrade_clones(dp
, tx
);
4706 * If anything has changed in this txg, push the deferred frees
4707 * from the previous txg. If not, leave them alone so that we
4708 * don't generate work on an otherwise idle system.
4710 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
4711 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
4712 !txg_list_empty(&dp
->dp_sync_tasks
, txg
))
4713 spa_sync_deferred_bplist(spa
, defer_bpl
, tx
, txg
);
4716 * Iterate to convergence.
4719 int pass
= ++spa
->spa_sync_pass
;
4721 spa_sync_config_object(spa
, tx
);
4722 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
4723 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
4724 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
4725 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
4726 spa_errlog_sync(spa
, txg
);
4727 dsl_pool_sync(dp
, txg
);
4729 if (pass
<= SYNC_PASS_DEFERRED_FREE
) {
4730 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
4731 bplist_sync(free_bpl
, spa_sync_free
, zio
, tx
);
4732 VERIFY(zio_wait(zio
) == 0);
4734 bplist_sync(free_bpl
, bplist_enqueue_cb
, defer_bpl
, tx
);
4739 while (vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
4742 } while (dmu_objset_is_dirty(mos
, txg
));
4744 ASSERT(free_bpl
->bpl_queue
== NULL
);
4746 bplist_close(defer_bpl
);
4749 * Rewrite the vdev configuration (which includes the uberblock)
4750 * to commit the transaction group.
4752 * If there are no dirty vdevs, we sync the uberblock to a few
4753 * random top-level vdevs that are known to be visible in the
4754 * config cache (see spa_vdev_add() for a complete description).
4755 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4759 * We hold SCL_STATE to prevent vdev open/close/etc.
4760 * while we're attempting to write the vdev labels.
4762 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4764 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
4765 vdev_t
*svd
[SPA_DVAS_PER_BP
];
4767 int children
= rvd
->vdev_children
;
4768 int c0
= spa_get_random(children
);
4770 for (int c
= 0; c
< children
; c
++) {
4771 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
4772 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
4774 svd
[svdcount
++] = vd
;
4775 if (svdcount
== SPA_DVAS_PER_BP
)
4778 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
4780 error
= vdev_config_sync(svd
, svdcount
, txg
,
4783 error
= vdev_config_sync(rvd
->vdev_child
,
4784 rvd
->vdev_children
, txg
, B_FALSE
);
4786 error
= vdev_config_sync(rvd
->vdev_child
,
4787 rvd
->vdev_children
, txg
, B_TRUE
);
4790 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4794 zio_suspend(spa
, NULL
);
4795 zio_resume_wait(spa
);
4800 * Clear the dirty config list.
4802 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
4803 vdev_config_clean(vd
);
4806 * Now that the new config has synced transactionally,
4807 * let it become visible to the config cache.
4809 if (spa
->spa_config_syncing
!= NULL
) {
4810 spa_config_set(spa
, spa
->spa_config_syncing
);
4811 spa
->spa_config_txg
= txg
;
4812 spa
->spa_config_syncing
= NULL
;
4815 spa
->spa_ubsync
= spa
->spa_uberblock
;
4817 dsl_pool_sync_done(dp
, txg
);
4820 * Update usable space statistics.
4822 while (vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
4823 vdev_sync_done(vd
, txg
);
4825 spa_update_dspace(spa
);
4828 * It had better be the case that we didn't dirty anything
4829 * since vdev_config_sync().
4831 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
4832 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
4833 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
4834 ASSERT(defer_bpl
->bpl_queue
== NULL
);
4835 ASSERT(free_bpl
->bpl_queue
== NULL
);
4837 spa
->spa_sync_pass
= 0;
4839 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4841 spa_handle_ignored_writes(spa
);
4844 * If any async tasks have been requested, kick them off.
4846 spa_async_dispatch(spa
);
4850 * Sync all pools. We don't want to hold the namespace lock across these
4851 * operations, so we take a reference on the spa_t and drop the lock during the
4855 spa_sync_allpools(void)
4858 mutex_enter(&spa_namespace_lock
);
4859 while ((spa
= spa_next(spa
)) != NULL
) {
4860 if (spa_state(spa
) != POOL_STATE_ACTIVE
|| spa_suspended(spa
))
4862 spa_open_ref(spa
, FTAG
);
4863 mutex_exit(&spa_namespace_lock
);
4864 txg_wait_synced(spa_get_dsl(spa
), 0);
4865 mutex_enter(&spa_namespace_lock
);
4866 spa_close(spa
, FTAG
);
4868 mutex_exit(&spa_namespace_lock
);
4872 * ==========================================================================
4873 * Miscellaneous routines
4874 * ==========================================================================
4878 * Remove all pools in the system.
4886 * Remove all cached state. All pools should be closed now,
4887 * so every spa in the AVL tree should be unreferenced.
4889 mutex_enter(&spa_namespace_lock
);
4890 while ((spa
= spa_next(NULL
)) != NULL
) {
4892 * Stop async tasks. The async thread may need to detach
4893 * a device that's been replaced, which requires grabbing
4894 * spa_namespace_lock, so we must drop it here.
4896 spa_open_ref(spa
, FTAG
);
4897 mutex_exit(&spa_namespace_lock
);
4898 spa_async_suspend(spa
);
4899 mutex_enter(&spa_namespace_lock
);
4900 spa_close(spa
, FTAG
);
4902 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4904 spa_deactivate(spa
);
4908 mutex_exit(&spa_namespace_lock
);
4912 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
4917 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
4921 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
4922 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
4923 if (vd
->vdev_guid
== guid
)
4927 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
4928 vd
= spa
->spa_spares
.sav_vdevs
[i
];
4929 if (vd
->vdev_guid
== guid
)
4938 spa_upgrade(spa_t
*spa
, uint64_t version
)
4940 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4943 * This should only be called for a non-faulted pool, and since a
4944 * future version would result in an unopenable pool, this shouldn't be
4947 ASSERT(spa
->spa_uberblock
.ub_version
<= SPA_VERSION
);
4948 ASSERT(version
>= spa
->spa_uberblock
.ub_version
);
4950 spa
->spa_uberblock
.ub_version
= version
;
4951 vdev_config_dirty(spa
->spa_root_vdev
);
4953 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4955 txg_wait_synced(spa_get_dsl(spa
), 0);
4959 spa_has_spare(spa_t
*spa
, uint64_t guid
)
4963 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
4965 for (i
= 0; i
< sav
->sav_count
; i
++)
4966 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
4969 for (i
= 0; i
< sav
->sav_npending
; i
++) {
4970 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
4971 &spareguid
) == 0 && spareguid
== guid
)
4979 * Check if a pool has an active shared spare device.
4980 * Note: reference count of an active spare is 2, as a spare and as a replace
4983 spa_has_active_shared_spare(spa_t
*spa
)
4987 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
4989 for (i
= 0; i
< sav
->sav_count
; i
++) {
4990 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
4991 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
5000 * Post a sysevent corresponding to the given event. The 'name' must be one of
5001 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
5002 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5003 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5004 * or zdb as real changes.
5007 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
5011 sysevent_attr_list_t
*attr
= NULL
;
5012 sysevent_value_t value
;
5015 ev
= sysevent_alloc(EC_ZFS
, (char *)name
, SUNW_KERN_PUB
"zfs",
5018 value
.value_type
= SE_DATA_TYPE_STRING
;
5019 value
.value
.sv_string
= spa_name(spa
);
5020 if (sysevent_add_attr(&attr
, ZFS_EV_POOL_NAME
, &value
, SE_SLEEP
) != 0)
5023 value
.value_type
= SE_DATA_TYPE_UINT64
;
5024 value
.value
.sv_uint64
= spa_guid(spa
);
5025 if (sysevent_add_attr(&attr
, ZFS_EV_POOL_GUID
, &value
, SE_SLEEP
) != 0)
5029 value
.value_type
= SE_DATA_TYPE_UINT64
;
5030 value
.value
.sv_uint64
= vd
->vdev_guid
;
5031 if (sysevent_add_attr(&attr
, ZFS_EV_VDEV_GUID
, &value
,
5035 if (vd
->vdev_path
) {
5036 value
.value_type
= SE_DATA_TYPE_STRING
;
5037 value
.value
.sv_string
= vd
->vdev_path
;
5038 if (sysevent_add_attr(&attr
, ZFS_EV_VDEV_PATH
,
5039 &value
, SE_SLEEP
) != 0)
5044 if (sysevent_attach_attributes(ev
, attr
) != 0)
5048 (void) log_sysevent(ev
, SE_SLEEP
, &eid
);
5052 sysevent_free_attr(attr
);