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 2008 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>
38 #include <sys/zio_compress.h>
40 #include <sys/dmu_tx.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/metaslab.h>
45 #include <sys/uberblock_impl.h>
48 #include <sys/dmu_traverse.h>
49 #include <sys/dmu_objset.h>
50 #include <sys/unique.h>
51 #include <sys/dsl_pool.h>
52 #include <sys/dsl_dataset.h>
53 #include <sys/dsl_dir.h>
54 #include <sys/dsl_prop.h>
55 #include <sys/dsl_synctask.h>
56 #include <sys/fs/zfs.h>
58 #include <sys/callb.h>
59 #include <sys/systeminfo.h>
60 #include <sys/sunddi.h>
61 #include <sys/spa_boot.h>
64 #include "zfs_comutil.h"
66 int zio_taskq_threads
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
68 { 1, 1 }, /* ZIO_TYPE_NULL */
69 { 1, 8 }, /* ZIO_TYPE_READ */
70 { 8, 1 }, /* ZIO_TYPE_WRITE */
71 { 1, 1 }, /* ZIO_TYPE_FREE */
72 { 1, 1 }, /* ZIO_TYPE_CLAIM */
73 { 1, 1 }, /* ZIO_TYPE_IOCTL */
76 static void spa_sync_props(void *arg1
, void *arg2
, cred_t
*cr
, dmu_tx_t
*tx
);
77 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
80 * ==========================================================================
81 * SPA properties routines
82 * ==========================================================================
86 * Add a (source=src, propname=propval) list to an nvlist.
89 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
90 uint64_t intval
, zprop_source_t src
)
92 const char *propname
= zpool_prop_to_name(prop
);
95 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
96 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
99 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
101 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
103 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
104 nvlist_free(propval
);
108 * Get property values from the spa configuration.
111 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
113 uint64_t size
= spa_get_space(spa
);
114 uint64_t used
= spa_get_alloc(spa
);
115 uint64_t cap
, version
;
116 zprop_source_t src
= ZPROP_SRC_NONE
;
117 spa_config_dirent_t
*dp
;
119 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
122 * readonly properties
124 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
125 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
126 spa_prop_add_list(*nvp
, ZPOOL_PROP_USED
, NULL
, used
, src
);
127 spa_prop_add_list(*nvp
, ZPOOL_PROP_AVAILABLE
, NULL
, size
- used
, src
);
129 cap
= (size
== 0) ? 0 : (used
* 100 / size
);
130 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
132 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
133 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
134 spa
->spa_root_vdev
->vdev_state
, src
);
137 * settable properties that are not stored in the pool property object.
139 version
= spa_version(spa
);
140 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
))
141 src
= ZPROP_SRC_DEFAULT
;
143 src
= ZPROP_SRC_LOCAL
;
144 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
146 if (spa
->spa_root
!= NULL
)
147 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
150 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
151 if (dp
->scd_path
== NULL
) {
152 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
153 "none", 0, ZPROP_SRC_LOCAL
);
154 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
155 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
156 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
162 * Get zpool property values.
165 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
169 objset_t
*mos
= spa
->spa_meta_objset
;
172 VERIFY(nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
174 mutex_enter(&spa
->spa_props_lock
);
177 * Get properties from the spa config.
179 spa_prop_get_config(spa
, nvp
);
181 /* If no pool property object, no more prop to get. */
182 if (spa
->spa_pool_props_object
== 0) {
183 mutex_exit(&spa
->spa_props_lock
);
188 * Get properties from the MOS pool property object.
190 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
191 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
192 zap_cursor_advance(&zc
)) {
195 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
198 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
201 switch (za
.za_integer_length
) {
203 /* integer property */
204 if (za
.za_first_integer
!=
205 zpool_prop_default_numeric(prop
))
206 src
= ZPROP_SRC_LOCAL
;
208 if (prop
== ZPOOL_PROP_BOOTFS
) {
210 dsl_dataset_t
*ds
= NULL
;
212 dp
= spa_get_dsl(spa
);
213 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
214 if (err
= dsl_dataset_hold_obj(dp
,
215 za
.za_first_integer
, FTAG
, &ds
)) {
216 rw_exit(&dp
->dp_config_rwlock
);
221 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1,
223 dsl_dataset_name(ds
, strval
);
224 dsl_dataset_rele(ds
, FTAG
);
225 rw_exit(&dp
->dp_config_rwlock
);
228 intval
= za
.za_first_integer
;
231 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
235 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1);
240 /* string property */
241 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
242 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
243 za
.za_name
, 1, za
.za_num_integers
, strval
);
245 kmem_free(strval
, za
.za_num_integers
);
248 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
249 kmem_free(strval
, za
.za_num_integers
);
256 zap_cursor_fini(&zc
);
257 mutex_exit(&spa
->spa_props_lock
);
259 if (err
&& err
!= ENOENT
) {
269 * Validate the given pool properties nvlist and modify the list
270 * for the property values to be set.
273 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
276 int error
= 0, reset_bootfs
= 0;
280 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
282 char *propname
, *strval
;
287 propname
= nvpair_name(elem
);
289 if ((prop
= zpool_name_to_prop(propname
)) == ZPROP_INVAL
)
293 case ZPOOL_PROP_VERSION
:
294 error
= nvpair_value_uint64(elem
, &intval
);
296 (intval
< spa_version(spa
) || intval
> SPA_VERSION
))
300 case ZPOOL_PROP_DELEGATION
:
301 case ZPOOL_PROP_AUTOREPLACE
:
302 case ZPOOL_PROP_LISTSNAPS
:
303 error
= nvpair_value_uint64(elem
, &intval
);
304 if (!error
&& intval
> 1)
308 case ZPOOL_PROP_BOOTFS
:
309 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
315 * Make sure the vdev config is bootable
317 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
324 error
= nvpair_value_string(elem
, &strval
);
329 if (strval
== NULL
|| strval
[0] == '\0') {
330 objnum
= zpool_prop_default_numeric(
335 if (error
= dmu_objset_open(strval
, DMU_OST_ZFS
,
336 DS_MODE_USER
| DS_MODE_READONLY
, &os
))
339 /* We don't support gzip bootable datasets */
340 if ((error
= dsl_prop_get_integer(strval
,
341 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
342 &compress
, NULL
)) == 0 &&
343 !BOOTFS_COMPRESS_VALID(compress
)) {
346 objnum
= dmu_objset_id(os
);
348 dmu_objset_close(os
);
352 case ZPOOL_PROP_FAILUREMODE
:
353 error
= nvpair_value_uint64(elem
, &intval
);
354 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
355 intval
> ZIO_FAILURE_MODE_PANIC
))
359 * This is a special case which only occurs when
360 * the pool has completely failed. This allows
361 * the user to change the in-core failmode property
362 * without syncing it out to disk (I/Os might
363 * currently be blocked). We do this by returning
364 * EIO to the caller (spa_prop_set) to trick it
365 * into thinking we encountered a property validation
368 if (!error
&& spa_suspended(spa
)) {
369 spa
->spa_failmode
= intval
;
374 case ZPOOL_PROP_CACHEFILE
:
375 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
378 if (strval
[0] == '\0')
381 if (strcmp(strval
, "none") == 0)
384 if (strval
[0] != '/') {
389 slash
= strrchr(strval
, '/');
390 ASSERT(slash
!= NULL
);
392 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
393 strcmp(slash
, "/..") == 0)
402 if (!error
&& reset_bootfs
) {
403 error
= nvlist_remove(props
,
404 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
407 error
= nvlist_add_uint64(props
,
408 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
416 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
420 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
423 return (dsl_sync_task_do(spa_get_dsl(spa
), NULL
, spa_sync_props
,
428 * If the bootfs property value is dsobj, clear it.
431 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
433 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
434 VERIFY(zap_remove(spa
->spa_meta_objset
,
435 spa
->spa_pool_props_object
,
436 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
442 * ==========================================================================
443 * SPA state manipulation (open/create/destroy/import/export)
444 * ==========================================================================
448 spa_error_entry_compare(const void *a
, const void *b
)
450 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
451 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
454 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
455 sizeof (zbookmark_t
));
466 * Utility function which retrieves copies of the current logs and
467 * re-initializes them in the process.
470 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
472 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
474 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
475 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
477 avl_create(&spa
->spa_errlist_scrub
,
478 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
479 offsetof(spa_error_entry_t
, se_avl
));
480 avl_create(&spa
->spa_errlist_last
,
481 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
482 offsetof(spa_error_entry_t
, se_avl
));
486 * Activate an uninitialized pool.
489 spa_activate(spa_t
*spa
, int mode
)
491 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
493 spa
->spa_state
= POOL_STATE_ACTIVE
;
494 spa
->spa_mode
= mode
;
496 spa
->spa_normal_class
= metaslab_class_create();
497 spa
->spa_log_class
= metaslab_class_create();
499 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
500 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
501 spa
->spa_zio_taskq
[t
][q
] = taskq_create("spa_zio",
502 zio_taskq_threads
[t
][q
], maxclsyspri
, 50,
503 INT_MAX
, TASKQ_PREPOPULATE
);
507 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
508 offsetof(vdev_t
, vdev_config_dirty_node
));
509 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
510 offsetof(vdev_t
, vdev_state_dirty_node
));
512 txg_list_create(&spa
->spa_vdev_txg_list
,
513 offsetof(struct vdev
, vdev_txg_node
));
515 avl_create(&spa
->spa_errlist_scrub
,
516 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
517 offsetof(spa_error_entry_t
, se_avl
));
518 avl_create(&spa
->spa_errlist_last
,
519 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
520 offsetof(spa_error_entry_t
, se_avl
));
524 * Opposite of spa_activate().
527 spa_deactivate(spa_t
*spa
)
529 ASSERT(spa
->spa_sync_on
== B_FALSE
);
530 ASSERT(spa
->spa_dsl_pool
== NULL
);
531 ASSERT(spa
->spa_root_vdev
== NULL
);
533 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
535 txg_list_destroy(&spa
->spa_vdev_txg_list
);
537 list_destroy(&spa
->spa_config_dirty_list
);
538 list_destroy(&spa
->spa_state_dirty_list
);
540 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
541 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
542 taskq_destroy(spa
->spa_zio_taskq
[t
][q
]);
543 spa
->spa_zio_taskq
[t
][q
] = NULL
;
547 metaslab_class_destroy(spa
->spa_normal_class
);
548 spa
->spa_normal_class
= NULL
;
550 metaslab_class_destroy(spa
->spa_log_class
);
551 spa
->spa_log_class
= NULL
;
554 * If this was part of an import or the open otherwise failed, we may
555 * still have errors left in the queues. Empty them just in case.
557 spa_errlog_drain(spa
);
559 avl_destroy(&spa
->spa_errlist_scrub
);
560 avl_destroy(&spa
->spa_errlist_last
);
562 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
566 * Verify a pool configuration, and construct the vdev tree appropriately. This
567 * will create all the necessary vdevs in the appropriate layout, with each vdev
568 * in the CLOSED state. This will prep the pool before open/creation/import.
569 * All vdev validation is done by the vdev_alloc() routine.
572 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
573 uint_t id
, int atype
)
579 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
582 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
585 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
597 for (c
= 0; c
< children
; c
++) {
599 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
607 ASSERT(*vdp
!= NULL
);
613 * Opposite of spa_load().
616 spa_unload(spa_t
*spa
)
620 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
625 spa_async_suspend(spa
);
630 if (spa
->spa_sync_on
) {
631 txg_sync_stop(spa
->spa_dsl_pool
);
632 spa
->spa_sync_on
= B_FALSE
;
636 * Wait for any outstanding async I/O to complete.
638 mutex_enter(&spa
->spa_async_root_lock
);
639 while (spa
->spa_async_root_count
!= 0)
640 cv_wait(&spa
->spa_async_root_cv
, &spa
->spa_async_root_lock
);
641 mutex_exit(&spa
->spa_async_root_lock
);
644 * Close the dsl pool.
646 if (spa
->spa_dsl_pool
) {
647 dsl_pool_close(spa
->spa_dsl_pool
);
648 spa
->spa_dsl_pool
= NULL
;
651 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
654 * Drop and purge level 2 cache
656 spa_l2cache_drop(spa
);
661 if (spa
->spa_root_vdev
)
662 vdev_free(spa
->spa_root_vdev
);
663 ASSERT(spa
->spa_root_vdev
== NULL
);
665 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
666 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
667 if (spa
->spa_spares
.sav_vdevs
) {
668 kmem_free(spa
->spa_spares
.sav_vdevs
,
669 spa
->spa_spares
.sav_count
* sizeof (void *));
670 spa
->spa_spares
.sav_vdevs
= NULL
;
672 if (spa
->spa_spares
.sav_config
) {
673 nvlist_free(spa
->spa_spares
.sav_config
);
674 spa
->spa_spares
.sav_config
= NULL
;
676 spa
->spa_spares
.sav_count
= 0;
678 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
679 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
680 if (spa
->spa_l2cache
.sav_vdevs
) {
681 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
682 spa
->spa_l2cache
.sav_count
* sizeof (void *));
683 spa
->spa_l2cache
.sav_vdevs
= NULL
;
685 if (spa
->spa_l2cache
.sav_config
) {
686 nvlist_free(spa
->spa_l2cache
.sav_config
);
687 spa
->spa_l2cache
.sav_config
= NULL
;
689 spa
->spa_l2cache
.sav_count
= 0;
691 spa
->spa_async_suspended
= 0;
693 spa_config_exit(spa
, SCL_ALL
, FTAG
);
697 * Load (or re-load) the current list of vdevs describing the active spares for
698 * this pool. When this is called, we have some form of basic information in
699 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
700 * then re-generate a more complete list including status information.
703 spa_load_spares(spa_t
*spa
)
710 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
713 * First, close and free any existing spare vdevs.
715 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
716 vd
= spa
->spa_spares
.sav_vdevs
[i
];
718 /* Undo the call to spa_activate() below */
719 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
720 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
721 spa_spare_remove(tvd
);
726 if (spa
->spa_spares
.sav_vdevs
)
727 kmem_free(spa
->spa_spares
.sav_vdevs
,
728 spa
->spa_spares
.sav_count
* sizeof (void *));
730 if (spa
->spa_spares
.sav_config
== NULL
)
733 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
734 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
736 spa
->spa_spares
.sav_count
= (int)nspares
;
737 spa
->spa_spares
.sav_vdevs
= NULL
;
743 * Construct the array of vdevs, opening them to get status in the
744 * process. For each spare, there is potentially two different vdev_t
745 * structures associated with it: one in the list of spares (used only
746 * for basic validation purposes) and one in the active vdev
747 * configuration (if it's spared in). During this phase we open and
748 * validate each vdev on the spare list. If the vdev also exists in the
749 * active configuration, then we also mark this vdev as an active spare.
751 spa
->spa_spares
.sav_vdevs
= kmem_alloc(nspares
* sizeof (void *),
753 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
754 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
755 VDEV_ALLOC_SPARE
) == 0);
758 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
760 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
762 if (!tvd
->vdev_isspare
)
766 * We only mark the spare active if we were successfully
767 * able to load the vdev. Otherwise, importing a pool
768 * with a bad active spare would result in strange
769 * behavior, because multiple pool would think the spare
770 * is actively in use.
772 * There is a vulnerability here to an equally bizarre
773 * circumstance, where a dead active spare is later
774 * brought back to life (onlined or otherwise). Given
775 * the rarity of this scenario, and the extra complexity
776 * it adds, we ignore the possibility.
778 if (!vdev_is_dead(tvd
))
779 spa_spare_activate(tvd
);
784 if (vdev_open(vd
) != 0)
787 if (vdev_validate_aux(vd
) == 0)
792 * Recompute the stashed list of spares, with status information
795 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
796 DATA_TYPE_NVLIST_ARRAY
) == 0);
798 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
800 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
801 spares
[i
] = vdev_config_generate(spa
,
802 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, B_TRUE
, B_FALSE
);
803 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
804 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
805 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
806 nvlist_free(spares
[i
]);
807 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
811 * Load (or re-load) the current list of vdevs describing the active l2cache for
812 * this pool. When this is called, we have some form of basic information in
813 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
814 * then re-generate a more complete list including status information.
815 * Devices which are already active have their details maintained, and are
819 spa_load_l2cache(spa_t
*spa
)
825 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
826 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
828 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
830 if (sav
->sav_config
!= NULL
) {
831 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
832 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
833 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
838 oldvdevs
= sav
->sav_vdevs
;
839 oldnvdevs
= sav
->sav_count
;
840 sav
->sav_vdevs
= NULL
;
844 * Process new nvlist of vdevs.
846 for (i
= 0; i
< nl2cache
; i
++) {
847 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
851 for (j
= 0; j
< oldnvdevs
; j
++) {
853 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
855 * Retain previous vdev for add/remove ops.
863 if (newvdevs
[i
] == NULL
) {
867 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
868 VDEV_ALLOC_L2CACHE
) == 0);
873 * Commit this vdev as an l2cache device,
874 * even if it fails to open.
881 spa_l2cache_activate(vd
);
883 if (vdev_open(vd
) != 0)
886 (void) vdev_validate_aux(vd
);
888 if (!vdev_is_dead(vd
)) {
889 size
= vdev_get_rsize(vd
);
890 l2arc_add_vdev(spa
, vd
,
891 VDEV_LABEL_START_SIZE
,
892 size
- VDEV_LABEL_START_SIZE
);
898 * Purge vdevs that were dropped
900 for (i
= 0; i
< oldnvdevs
; i
++) {
905 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
906 pool
!= 0ULL && l2arc_vdev_present(vd
))
907 l2arc_remove_vdev(vd
);
908 (void) vdev_close(vd
);
909 spa_l2cache_remove(vd
);
914 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
916 if (sav
->sav_config
== NULL
)
919 sav
->sav_vdevs
= newvdevs
;
920 sav
->sav_count
= (int)nl2cache
;
923 * Recompute the stashed list of l2cache devices, with status
924 * information this time.
926 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
927 DATA_TYPE_NVLIST_ARRAY
) == 0);
929 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
930 for (i
= 0; i
< sav
->sav_count
; i
++)
931 l2cache
[i
] = vdev_config_generate(spa
,
932 sav
->sav_vdevs
[i
], B_TRUE
, B_FALSE
, B_TRUE
);
933 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
934 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
936 for (i
= 0; i
< sav
->sav_count
; i
++)
937 nvlist_free(l2cache
[i
]);
939 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
943 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
951 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
952 nvsize
= *(uint64_t *)db
->db_data
;
953 dmu_buf_rele(db
, FTAG
);
955 packed
= kmem_alloc(nvsize
, KM_SLEEP
);
956 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
);
958 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
959 kmem_free(packed
, nvsize
);
965 * Checks to see if the given vdev could not be opened, in which case we post a
966 * sysevent to notify the autoreplace code that the device has been removed.
969 spa_check_removed(vdev_t
*vd
)
973 for (c
= 0; c
< vd
->vdev_children
; c
++)
974 spa_check_removed(vd
->vdev_child
[c
]);
976 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
)) {
977 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
978 spa_event_notify(vd
->vdev_spa
, vd
, ESC_ZFS_VDEV_CHECK
);
983 * Check for missing log devices
986 spa_check_logs(spa_t
*spa
)
988 switch (spa
->spa_log_state
) {
989 case SPA_LOG_MISSING
:
990 /* need to recheck in case slog has been restored */
991 case SPA_LOG_UNKNOWN
:
992 if (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
, NULL
,
994 spa
->spa_log_state
= SPA_LOG_MISSING
;
1000 (void) dmu_objset_find(spa
->spa_name
, zil_clear_log_chain
, NULL
,
1004 spa
->spa_log_state
= SPA_LOG_GOOD
;
1009 * Load an existing storage pool, using the pool's builtin spa_config as a
1010 * source of configuration information.
1013 spa_load(spa_t
*spa
, nvlist_t
*config
, spa_load_state_t state
, int mosconfig
)
1016 nvlist_t
*nvroot
= NULL
;
1018 uberblock_t
*ub
= &spa
->spa_uberblock
;
1019 uint64_t config_cache_txg
= spa
->spa_config_txg
;
1022 uint64_t autoreplace
= 0;
1023 int orig_mode
= spa
->spa_mode
;
1024 char *ereport
= FM_EREPORT_ZFS_POOL
;
1027 * If this is an untrusted config, access the pool in read-only mode.
1028 * This prevents things like resilvering recently removed devices.
1031 spa
->spa_mode
= FREAD
;
1033 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1035 spa
->spa_load_state
= state
;
1037 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) ||
1038 nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
1044 * Versioning wasn't explicitly added to the label until later, so if
1045 * it's not present treat it as the initial version.
1047 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
, &version
) != 0)
1048 version
= SPA_VERSION_INITIAL
;
1050 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
1051 &spa
->spa_config_txg
);
1053 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
1054 spa_guid_exists(pool_guid
, 0)) {
1059 spa
->spa_load_guid
= pool_guid
;
1062 * Parse the configuration into a vdev tree. We explicitly set the
1063 * value that will be returned by spa_version() since parsing the
1064 * configuration requires knowing the version number.
1066 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1067 spa
->spa_ubsync
.ub_version
= version
;
1068 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_LOAD
);
1069 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1074 ASSERT(spa
->spa_root_vdev
== rvd
);
1075 ASSERT(spa_guid(spa
) == pool_guid
);
1078 * Try to open all vdevs, loading each label in the process.
1080 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1081 error
= vdev_open(rvd
);
1082 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1087 * Validate the labels for all leaf vdevs. We need to grab the config
1088 * lock because all label I/O is done with ZIO_FLAG_CONFIG_WRITER.
1091 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1092 error
= vdev_validate(rvd
);
1093 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1098 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
1104 * Find the best uberblock.
1106 vdev_uberblock_load(NULL
, rvd
, ub
);
1109 * If we weren't able to find a single valid uberblock, return failure.
1111 if (ub
->ub_txg
== 0) {
1112 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1113 VDEV_AUX_CORRUPT_DATA
);
1119 * If the pool is newer than the code, we can't open it.
1121 if (ub
->ub_version
> SPA_VERSION
) {
1122 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1123 VDEV_AUX_VERSION_NEWER
);
1129 * If the vdev guid sum doesn't match the uberblock, we have an
1130 * incomplete configuration.
1132 if (rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
&& mosconfig
) {
1133 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1134 VDEV_AUX_BAD_GUID_SUM
);
1140 * Initialize internal SPA structures.
1142 spa
->spa_state
= POOL_STATE_ACTIVE
;
1143 spa
->spa_ubsync
= spa
->spa_uberblock
;
1144 spa
->spa_first_txg
= spa_last_synced_txg(spa
) + 1;
1145 error
= dsl_pool_open(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
1147 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1148 VDEV_AUX_CORRUPT_DATA
);
1151 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
1153 if (zap_lookup(spa
->spa_meta_objset
,
1154 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
1155 sizeof (uint64_t), 1, &spa
->spa_config_object
) != 0) {
1156 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1157 VDEV_AUX_CORRUPT_DATA
);
1163 nvlist_t
*newconfig
;
1166 if (load_nvlist(spa
, spa
->spa_config_object
, &newconfig
) != 0) {
1167 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1168 VDEV_AUX_CORRUPT_DATA
);
1173 if (!spa_is_root(spa
) && nvlist_lookup_uint64(newconfig
,
1174 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
1176 unsigned long myhostid
= 0;
1178 VERIFY(nvlist_lookup_string(newconfig
,
1179 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
1181 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
1182 if (hostid
!= 0 && myhostid
!= 0 &&
1183 (unsigned long)hostid
!= myhostid
) {
1184 cmn_err(CE_WARN
, "pool '%s' could not be "
1185 "loaded as it was last accessed by "
1186 "another system (host: %s hostid: 0x%lx). "
1187 "See: http://www.sun.com/msg/ZFS-8000-EY",
1188 spa_name(spa
), hostname
,
1189 (unsigned long)hostid
);
1195 spa_config_set(spa
, newconfig
);
1197 spa_deactivate(spa
);
1198 spa_activate(spa
, orig_mode
);
1200 return (spa_load(spa
, newconfig
, state
, B_TRUE
));
1203 if (zap_lookup(spa
->spa_meta_objset
,
1204 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPLIST
,
1205 sizeof (uint64_t), 1, &spa
->spa_sync_bplist_obj
) != 0) {
1206 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1207 VDEV_AUX_CORRUPT_DATA
);
1213 * Load the bit that tells us to use the new accounting function
1214 * (raid-z deflation). If we have an older pool, this will not
1217 error
= zap_lookup(spa
->spa_meta_objset
,
1218 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
1219 sizeof (uint64_t), 1, &spa
->spa_deflate
);
1220 if (error
!= 0 && error
!= ENOENT
) {
1221 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1222 VDEV_AUX_CORRUPT_DATA
);
1228 * Load the persistent error log. If we have an older pool, this will
1231 error
= zap_lookup(spa
->spa_meta_objset
,
1232 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ERRLOG_LAST
,
1233 sizeof (uint64_t), 1, &spa
->spa_errlog_last
);
1234 if (error
!= 0 && error
!= ENOENT
) {
1235 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1236 VDEV_AUX_CORRUPT_DATA
);
1241 error
= zap_lookup(spa
->spa_meta_objset
,
1242 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_ERRLOG_SCRUB
,
1243 sizeof (uint64_t), 1, &spa
->spa_errlog_scrub
);
1244 if (error
!= 0 && error
!= ENOENT
) {
1245 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1246 VDEV_AUX_CORRUPT_DATA
);
1252 * Load the history object. If we have an older pool, this
1253 * will not be present.
1255 error
= zap_lookup(spa
->spa_meta_objset
,
1256 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_HISTORY
,
1257 sizeof (uint64_t), 1, &spa
->spa_history
);
1258 if (error
!= 0 && error
!= ENOENT
) {
1259 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1260 VDEV_AUX_CORRUPT_DATA
);
1266 * Load any hot spares for this pool.
1268 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1269 DMU_POOL_SPARES
, sizeof (uint64_t), 1, &spa
->spa_spares
.sav_object
);
1270 if (error
!= 0 && error
!= ENOENT
) {
1271 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1272 VDEV_AUX_CORRUPT_DATA
);
1277 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
1278 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
1279 &spa
->spa_spares
.sav_config
) != 0) {
1280 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1281 VDEV_AUX_CORRUPT_DATA
);
1286 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1287 spa_load_spares(spa
);
1288 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1292 * Load any level 2 ARC devices for this pool.
1294 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1295 DMU_POOL_L2CACHE
, sizeof (uint64_t), 1,
1296 &spa
->spa_l2cache
.sav_object
);
1297 if (error
!= 0 && error
!= ENOENT
) {
1298 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1299 VDEV_AUX_CORRUPT_DATA
);
1304 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
1305 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
1306 &spa
->spa_l2cache
.sav_config
) != 0) {
1307 vdev_set_state(rvd
, B_TRUE
,
1308 VDEV_STATE_CANT_OPEN
,
1309 VDEV_AUX_CORRUPT_DATA
);
1314 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1315 spa_load_l2cache(spa
);
1316 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1319 if (spa_check_logs(spa
)) {
1320 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1323 ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
1328 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
1330 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1331 DMU_POOL_PROPS
, sizeof (uint64_t), 1, &spa
->spa_pool_props_object
);
1333 if (error
&& error
!= ENOENT
) {
1334 vdev_set_state(rvd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1335 VDEV_AUX_CORRUPT_DATA
);
1341 (void) zap_lookup(spa
->spa_meta_objset
,
1342 spa
->spa_pool_props_object
,
1343 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
),
1344 sizeof (uint64_t), 1, &spa
->spa_bootfs
);
1345 (void) zap_lookup(spa
->spa_meta_objset
,
1346 spa
->spa_pool_props_object
,
1347 zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE
),
1348 sizeof (uint64_t), 1, &autoreplace
);
1349 (void) zap_lookup(spa
->spa_meta_objset
,
1350 spa
->spa_pool_props_object
,
1351 zpool_prop_to_name(ZPOOL_PROP_DELEGATION
),
1352 sizeof (uint64_t), 1, &spa
->spa_delegation
);
1353 (void) zap_lookup(spa
->spa_meta_objset
,
1354 spa
->spa_pool_props_object
,
1355 zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE
),
1356 sizeof (uint64_t), 1, &spa
->spa_failmode
);
1360 * If the 'autoreplace' property is set, then post a resource notifying
1361 * the ZFS DE that it should not issue any faults for unopenable
1362 * devices. We also iterate over the vdevs, and post a sysevent for any
1363 * unopenable vdevs so that the normal autoreplace handler can take
1366 if (autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
)
1367 spa_check_removed(spa
->spa_root_vdev
);
1370 * Load the vdev state for all toplevel vdevs.
1375 * Propagate the leaf DTLs we just loaded all the way up the tree.
1377 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1378 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
1379 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1382 * Check the state of the root vdev. If it can't be opened, it
1383 * indicates one or more toplevel vdevs are faulted.
1385 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
1390 if (spa_writeable(spa
)) {
1392 int need_update
= B_FALSE
;
1394 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
1397 * Claim log blocks that haven't been committed yet.
1398 * This must all happen in a single txg.
1400 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
1401 spa_first_txg(spa
));
1402 (void) dmu_objset_find(spa_name(spa
),
1403 zil_claim
, tx
, DS_FIND_CHILDREN
);
1406 spa
->spa_sync_on
= B_TRUE
;
1407 txg_sync_start(spa
->spa_dsl_pool
);
1410 * Wait for all claims to sync.
1412 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1415 * If the config cache is stale, or we have uninitialized
1416 * metaslabs (see spa_vdev_add()), then update the config.
1418 if (config_cache_txg
!= spa
->spa_config_txg
||
1419 state
== SPA_LOAD_IMPORT
)
1420 need_update
= B_TRUE
;
1422 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
1423 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
1424 need_update
= B_TRUE
;
1427 * Update the config cache asychronously in case we're the
1428 * root pool, in which case the config cache isn't writable yet.
1431 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
1434 * Check all DTLs to see if anything needs resilvering.
1436 if (vdev_resilver_needed(rvd
, NULL
, NULL
))
1437 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1442 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
1443 if (error
&& error
!= EBADF
)
1444 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
1445 spa
->spa_load_state
= SPA_LOAD_NONE
;
1454 * The import case is identical to an open except that the configuration is sent
1455 * down from userland, instead of grabbed from the configuration cache. For the
1456 * case of an open, the pool configuration will exist in the
1457 * POOL_STATE_UNINITIALIZED state.
1459 * The stats information (gen/count/ustats) is used to gather vdev statistics at
1460 * the same time open the pool, without having to keep around the spa_t in some
1464 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
**config
)
1468 int locked
= B_FALSE
;
1473 * As disgusting as this is, we need to support recursive calls to this
1474 * function because dsl_dir_open() is called during spa_load(), and ends
1475 * up calling spa_open() again. The real fix is to figure out how to
1476 * avoid dsl_dir_open() calling this in the first place.
1478 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
1479 mutex_enter(&spa_namespace_lock
);
1483 if ((spa
= spa_lookup(pool
)) == NULL
) {
1485 mutex_exit(&spa_namespace_lock
);
1488 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
1490 spa_activate(spa
, spa_mode_global
);
1492 error
= spa_load(spa
, spa
->spa_config
, SPA_LOAD_OPEN
, B_FALSE
);
1494 if (error
== EBADF
) {
1496 * If vdev_validate() returns failure (indicated by
1497 * EBADF), it indicates that one of the vdevs indicates
1498 * that the pool has been exported or destroyed. If
1499 * this is the case, the config cache is out of sync and
1500 * we should remove the pool from the namespace.
1503 spa_deactivate(spa
);
1504 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
1507 mutex_exit(&spa_namespace_lock
);
1513 * We can't open the pool, but we still have useful
1514 * information: the state of each vdev after the
1515 * attempted vdev_open(). Return this to the user.
1517 if (config
!= NULL
&& spa
->spa_root_vdev
!= NULL
)
1518 *config
= spa_config_generate(spa
, NULL
, -1ULL,
1521 spa_deactivate(spa
);
1522 spa
->spa_last_open_failed
= B_TRUE
;
1524 mutex_exit(&spa_namespace_lock
);
1528 spa
->spa_last_open_failed
= B_FALSE
;
1532 spa_open_ref(spa
, tag
);
1535 mutex_exit(&spa_namespace_lock
);
1540 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
1546 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
1548 return (spa_open_common(name
, spapp
, tag
, NULL
));
1552 * Lookup the given spa_t, incrementing the inject count in the process,
1553 * preventing it from being exported or destroyed.
1556 spa_inject_addref(char *name
)
1560 mutex_enter(&spa_namespace_lock
);
1561 if ((spa
= spa_lookup(name
)) == NULL
) {
1562 mutex_exit(&spa_namespace_lock
);
1565 spa
->spa_inject_ref
++;
1566 mutex_exit(&spa_namespace_lock
);
1572 spa_inject_delref(spa_t
*spa
)
1574 mutex_enter(&spa_namespace_lock
);
1575 spa
->spa_inject_ref
--;
1576 mutex_exit(&spa_namespace_lock
);
1580 * Add spares device information to the nvlist.
1583 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
1593 if (spa
->spa_spares
.sav_count
== 0)
1596 VERIFY(nvlist_lookup_nvlist(config
,
1597 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
1598 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1599 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1601 VERIFY(nvlist_add_nvlist_array(nvroot
,
1602 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
1603 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
1604 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1607 * Go through and find any spares which have since been
1608 * repurposed as an active spare. If this is the case, update
1609 * their status appropriately.
1611 for (i
= 0; i
< nspares
; i
++) {
1612 VERIFY(nvlist_lookup_uint64(spares
[i
],
1613 ZPOOL_CONFIG_GUID
, &guid
) == 0);
1614 if (spa_spare_exists(guid
, &pool
, NULL
) &&
1616 VERIFY(nvlist_lookup_uint64_array(
1617 spares
[i
], ZPOOL_CONFIG_STATS
,
1618 (uint64_t **)&vs
, &vsc
) == 0);
1619 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
1620 vs
->vs_aux
= VDEV_AUX_SPARED
;
1627 * Add l2cache device information to the nvlist, including vdev stats.
1630 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
1633 uint_t i
, j
, nl2cache
;
1640 if (spa
->spa_l2cache
.sav_count
== 0)
1643 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
1645 VERIFY(nvlist_lookup_nvlist(config
,
1646 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
1647 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
1648 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1649 if (nl2cache
!= 0) {
1650 VERIFY(nvlist_add_nvlist_array(nvroot
,
1651 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
1652 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
1653 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1656 * Update level 2 cache device stats.
1659 for (i
= 0; i
< nl2cache
; i
++) {
1660 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
1661 ZPOOL_CONFIG_GUID
, &guid
) == 0);
1664 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
1666 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
1667 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
1673 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
1674 ZPOOL_CONFIG_STATS
, (uint64_t **)&vs
, &vsc
) == 0);
1675 vdev_get_stats(vd
, vs
);
1679 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
1683 spa_get_stats(const char *name
, nvlist_t
**config
, char *altroot
, size_t buflen
)
1689 error
= spa_open_common(name
, &spa
, FTAG
, config
);
1691 if (spa
&& *config
!= NULL
) {
1692 VERIFY(nvlist_add_uint64(*config
, ZPOOL_CONFIG_ERRCOUNT
,
1693 spa_get_errlog_size(spa
)) == 0);
1695 if (spa_suspended(spa
))
1696 VERIFY(nvlist_add_uint64(*config
,
1697 ZPOOL_CONFIG_SUSPENDED
, spa
->spa_failmode
) == 0);
1699 spa_add_spares(spa
, *config
);
1700 spa_add_l2cache(spa
, *config
);
1704 * We want to get the alternate root even for faulted pools, so we cheat
1705 * and call spa_lookup() directly.
1709 mutex_enter(&spa_namespace_lock
);
1710 spa
= spa_lookup(name
);
1712 spa_altroot(spa
, altroot
, buflen
);
1716 mutex_exit(&spa_namespace_lock
);
1718 spa_altroot(spa
, altroot
, buflen
);
1723 spa_close(spa
, FTAG
);
1729 * Validate that the auxiliary device array is well formed. We must have an
1730 * array of nvlists, each which describes a valid leaf vdev. If this is an
1731 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
1732 * specified, as long as they are well-formed.
1735 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
1736 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
1737 vdev_labeltype_t label
)
1744 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1747 * It's acceptable to have no devs specified.
1749 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
1756 * Make sure the pool is formatted with a version that supports this
1759 if (spa_version(spa
) < version
)
1763 * Set the pending device list so we correctly handle device in-use
1766 sav
->sav_pending
= dev
;
1767 sav
->sav_npending
= ndev
;
1769 for (i
= 0; i
< ndev
; i
++) {
1770 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
1774 if (!vd
->vdev_ops
->vdev_op_leaf
) {
1781 * The L2ARC currently only supports disk devices in
1782 * kernel context. For user-level testing, we allow it.
1785 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
1786 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
1793 if ((error
= vdev_open(vd
)) == 0 &&
1794 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
1795 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
1796 vd
->vdev_guid
) == 0);
1802 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
1809 sav
->sav_pending
= NULL
;
1810 sav
->sav_npending
= 0;
1815 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
1819 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1821 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
1822 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
1823 VDEV_LABEL_SPARE
)) != 0) {
1827 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
1828 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
1829 VDEV_LABEL_L2CACHE
));
1833 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
1838 if (sav
->sav_config
!= NULL
) {
1844 * Generate new dev list by concatentating with the
1847 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
1848 &olddevs
, &oldndevs
) == 0);
1850 newdevs
= kmem_alloc(sizeof (void *) *
1851 (ndevs
+ oldndevs
), KM_SLEEP
);
1852 for (i
= 0; i
< oldndevs
; i
++)
1853 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
1855 for (i
= 0; i
< ndevs
; i
++)
1856 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
1859 VERIFY(nvlist_remove(sav
->sav_config
, config
,
1860 DATA_TYPE_NVLIST_ARRAY
) == 0);
1862 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1863 config
, newdevs
, ndevs
+ oldndevs
) == 0);
1864 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
1865 nvlist_free(newdevs
[i
]);
1866 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
1869 * Generate a new dev list.
1871 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
1873 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
1879 * Stop and drop level 2 ARC devices
1882 spa_l2cache_drop(spa_t
*spa
)
1886 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1888 for (i
= 0; i
< sav
->sav_count
; i
++) {
1891 vd
= sav
->sav_vdevs
[i
];
1894 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1895 pool
!= 0ULL && l2arc_vdev_present(vd
))
1896 l2arc_remove_vdev(vd
);
1897 if (vd
->vdev_isl2cache
)
1898 spa_l2cache_remove(vd
);
1899 vdev_clear_stats(vd
);
1900 (void) vdev_close(vd
);
1908 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
1909 const char *history_str
, nvlist_t
*zplprops
)
1912 char *altroot
= NULL
;
1917 uint64_t txg
= TXG_INITIAL
;
1918 nvlist_t
**spares
, **l2cache
;
1919 uint_t nspares
, nl2cache
;
1923 * If this pool already exists, return failure.
1925 mutex_enter(&spa_namespace_lock
);
1926 if (spa_lookup(pool
) != NULL
) {
1927 mutex_exit(&spa_namespace_lock
);
1932 * Allocate a new spa_t structure.
1934 (void) nvlist_lookup_string(props
,
1935 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
1936 spa
= spa_add(pool
, altroot
);
1937 spa_activate(spa
, spa_mode_global
);
1939 spa
->spa_uberblock
.ub_txg
= txg
- 1;
1941 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
1943 spa_deactivate(spa
);
1945 mutex_exit(&spa_namespace_lock
);
1949 if (nvlist_lookup_uint64(props
, zpool_prop_to_name(ZPOOL_PROP_VERSION
),
1951 version
= SPA_VERSION
;
1952 ASSERT(version
<= SPA_VERSION
);
1953 spa
->spa_uberblock
.ub_version
= version
;
1954 spa
->spa_ubsync
= spa
->spa_uberblock
;
1957 * Create the root vdev.
1959 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1961 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
1963 ASSERT(error
!= 0 || rvd
!= NULL
);
1964 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
1966 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
1970 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
1971 (error
= spa_validate_aux(spa
, nvroot
, txg
,
1972 VDEV_ALLOC_ADD
)) == 0) {
1973 for (c
= 0; c
< rvd
->vdev_children
; c
++)
1974 vdev_init(rvd
->vdev_child
[c
], txg
);
1975 vdev_config_dirty(rvd
);
1978 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1982 spa_deactivate(spa
);
1984 mutex_exit(&spa_namespace_lock
);
1989 * Get the list of spares, if specified.
1991 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
1992 &spares
, &nspares
) == 0) {
1993 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
1995 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1996 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
1997 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1998 spa_load_spares(spa
);
1999 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2000 spa
->spa_spares
.sav_sync
= B_TRUE
;
2004 * Get the list of level 2 cache devices, if specified.
2006 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
2007 &l2cache
, &nl2cache
) == 0) {
2008 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
2009 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2010 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
2011 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
2012 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2013 spa_load_l2cache(spa
);
2014 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2015 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2018 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
2019 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
2021 tx
= dmu_tx_create_assigned(dp
, txg
);
2024 * Create the pool config object.
2026 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
2027 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
2028 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
2030 if (zap_add(spa
->spa_meta_objset
,
2031 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
2032 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
2033 cmn_err(CE_PANIC
, "failed to add pool config");
2036 /* Newly created pools with the right version are always deflated. */
2037 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
2038 spa
->spa_deflate
= TRUE
;
2039 if (zap_add(spa
->spa_meta_objset
,
2040 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
2041 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
2042 cmn_err(CE_PANIC
, "failed to add deflate");
2047 * Create the deferred-free bplist object. Turn off compression
2048 * because sync-to-convergence takes longer if the blocksize
2051 spa
->spa_sync_bplist_obj
= bplist_create(spa
->spa_meta_objset
,
2053 dmu_object_set_compress(spa
->spa_meta_objset
, spa
->spa_sync_bplist_obj
,
2054 ZIO_COMPRESS_OFF
, tx
);
2056 if (zap_add(spa
->spa_meta_objset
,
2057 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPLIST
,
2058 sizeof (uint64_t), 1, &spa
->spa_sync_bplist_obj
, tx
) != 0) {
2059 cmn_err(CE_PANIC
, "failed to add bplist");
2063 * Create the pool's history object.
2065 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
2066 spa_history_create_obj(spa
, tx
);
2069 * Set pool properties.
2071 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
2072 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2073 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
2075 spa_sync_props(spa
, props
, CRED(), tx
);
2079 spa
->spa_sync_on
= B_TRUE
;
2080 txg_sync_start(spa
->spa_dsl_pool
);
2083 * We explicitly wait for the first transaction to complete so that our
2084 * bean counters are appropriately updated.
2086 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
2088 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
2090 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& history_str
!= NULL
)
2091 (void) spa_history_log(spa
, history_str
, LOG_CMD_POOL_CREATE
);
2093 mutex_exit(&spa_namespace_lock
);
2095 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2101 * Import the given pool into the system. We set up the necessary spa_t and
2102 * then call spa_load() to do the dirty work.
2105 spa_import_common(const char *pool
, nvlist_t
*config
, nvlist_t
*props
,
2106 boolean_t isroot
, boolean_t allowfaulted
)
2109 char *altroot
= NULL
;
2112 nvlist_t
**spares
, **l2cache
;
2113 uint_t nspares
, nl2cache
;
2116 * If a pool with this name exists, return failure.
2118 mutex_enter(&spa_namespace_lock
);
2119 if ((spa
= spa_lookup(pool
)) != NULL
) {
2122 * Remove the existing root pool from the
2123 * namespace so that we can replace it with
2124 * the correct config we just read in.
2126 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
2129 mutex_exit(&spa_namespace_lock
);
2135 * Create and initialize the spa structure.
2137 (void) nvlist_lookup_string(props
,
2138 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
2139 spa
= spa_add(pool
, altroot
);
2140 spa_activate(spa
, spa_mode_global
);
2143 spa
->spa_import_faulted
= B_TRUE
;
2144 spa
->spa_is_root
= isroot
;
2147 * Pass off the heavy lifting to spa_load().
2148 * Pass TRUE for mosconfig (unless this is a root pool) because
2149 * the user-supplied config is actually the one to trust when
2152 loaderr
= error
= spa_load(spa
, config
, SPA_LOAD_IMPORT
, !isroot
);
2154 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2156 * Toss any existing sparelist, as it doesn't have any validity anymore,
2157 * and conflicts with spa_has_spare().
2159 if (!isroot
&& spa
->spa_spares
.sav_config
) {
2160 nvlist_free(spa
->spa_spares
.sav_config
);
2161 spa
->spa_spares
.sav_config
= NULL
;
2162 spa_load_spares(spa
);
2164 if (!isroot
&& spa
->spa_l2cache
.sav_config
) {
2165 nvlist_free(spa
->spa_l2cache
.sav_config
);
2166 spa
->spa_l2cache
.sav_config
= NULL
;
2167 spa_load_l2cache(spa
);
2170 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
2173 error
= spa_validate_aux(spa
, nvroot
, -1ULL, VDEV_ALLOC_SPARE
);
2175 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
2176 VDEV_ALLOC_L2CACHE
);
2177 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2179 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
2180 (error
= spa_prop_set(spa
, props
)))) {
2181 if (loaderr
!= 0 && loaderr
!= EINVAL
&& allowfaulted
) {
2183 * If we failed to load the pool, but 'allowfaulted' is
2184 * set, then manually set the config as if the config
2185 * passed in was specified in the cache file.
2188 spa
->spa_import_faulted
= B_FALSE
;
2189 if (spa
->spa_config
== NULL
)
2190 spa
->spa_config
= spa_config_generate(spa
,
2191 NULL
, -1ULL, B_TRUE
);
2193 spa_deactivate(spa
);
2194 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
2197 spa_deactivate(spa
);
2200 mutex_exit(&spa_namespace_lock
);
2205 * Override any spares and level 2 cache devices as specified by
2206 * the user, as these may have correct device names/devids, etc.
2208 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
2209 &spares
, &nspares
) == 0) {
2210 if (spa
->spa_spares
.sav_config
)
2211 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
2212 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
2214 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
2215 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2216 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
2217 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
2218 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2219 spa_load_spares(spa
);
2220 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2221 spa
->spa_spares
.sav_sync
= B_TRUE
;
2223 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
2224 &l2cache
, &nl2cache
) == 0) {
2225 if (spa
->spa_l2cache
.sav_config
)
2226 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
2227 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
2229 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
2230 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2231 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
2232 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
2233 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2234 spa_load_l2cache(spa
);
2235 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2236 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2239 if (spa_writeable(spa
)) {
2241 * Update the config cache to include the newly-imported pool.
2243 spa_config_update_common(spa
, SPA_CONFIG_UPDATE_POOL
, isroot
);
2246 spa
->spa_import_faulted
= B_FALSE
;
2247 mutex_exit(&spa_namespace_lock
);
2254 * Build a "root" vdev for a top level vdev read in from a rootpool
2258 spa_build_rootpool_config(nvlist_t
*config
)
2260 nvlist_t
*nvtop
, *nvroot
;
2264 * Add this top-level vdev to the child array.
2266 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtop
)
2268 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pgid
)
2272 * Put this pool's top-level vdevs into a root vdev.
2274 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2275 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
)
2277 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
2278 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
2279 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
2283 * Replace the existing vdev_tree with the new root vdev in
2284 * this pool's configuration (remove the old, add the new).
2286 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
2287 nvlist_free(nvroot
);
2291 * Get the root pool information from the root disk, then import the root pool
2292 * during the system boot up time.
2294 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
2297 spa_check_rootconf(char *devpath
, char *devid
, nvlist_t
**bestconf
,
2304 if (error
= vdev_disk_read_rootlabel(devpath
, devid
, &config
))
2307 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
2309 if (bestconf
!= NULL
)
2312 nvlist_free(config
);
2318 spa_rootdev_validate(nvlist_t
*nv
)
2322 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
, &ival
) == 0 ||
2323 nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
, &ival
) == 0 ||
2324 nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
, &ival
) == 0)
2332 * Given the boot device's physical path or devid, check if the device
2333 * is in a valid state. If so, return the configuration from the vdev
2337 spa_get_rootconf(char *devpath
, char *devid
, nvlist_t
**bestconf
)
2339 nvlist_t
*conf
= NULL
;
2341 nvlist_t
*nvtop
, **child
;
2343 char *bootpath
= NULL
;
2348 if (devpath
&& ((tmp
= strchr(devpath
, ' ')) != NULL
))
2350 if (error
= spa_check_rootconf(devpath
, devid
, &conf
, &txg
)) {
2351 cmn_err(CE_NOTE
, "error reading device label");
2355 cmn_err(CE_NOTE
, "this device is detached");
2360 VERIFY(nvlist_lookup_nvlist(conf
, ZPOOL_CONFIG_VDEV_TREE
,
2362 VERIFY(nvlist_lookup_string(nvtop
, ZPOOL_CONFIG_TYPE
, &type
) == 0);
2364 if (strcmp(type
, VDEV_TYPE_DISK
) == 0) {
2365 if (spa_rootdev_validate(nvtop
)) {
2373 ASSERT(strcmp(type
, VDEV_TYPE_MIRROR
) == 0);
2375 VERIFY(nvlist_lookup_nvlist_array(nvtop
, ZPOOL_CONFIG_CHILDREN
,
2376 &child
, &children
) == 0);
2379 * Go thru vdevs in the mirror to see if the given device
2380 * has the most recent txg. Only the device with the most
2381 * recent txg has valid information and should be booted.
2383 for (c
= 0; c
< children
; c
++) {
2384 char *cdevid
, *cpath
;
2389 if (nvlist_lookup_string(child
[c
], ZPOOL_CONFIG_PHYS_PATH
,
2390 &cpath
) != 0 && nvlist_lookup_string(child
[c
],
2391 ZPOOL_CONFIG_DEVID
, &cdevid
) != 0)
2393 if ((spa_check_rootconf(cpath
, cdevid
, NULL
,
2394 &tmptxg
) == 0) && (tmptxg
> txg
)) {
2396 VERIFY(nvlist_lookup_string(child
[c
],
2397 ZPOOL_CONFIG_PATH
, &bootpath
) == 0);
2401 /* Does the best device match the one we've booted from? */
2403 cmn_err(CE_NOTE
, "try booting from '%s'", bootpath
);
2412 * Import a root pool.
2414 * For x86. devpath_list will consist of devid and/or physpath name of
2415 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2416 * The GRUB "findroot" command will return the vdev we should boot.
2418 * For Sparc, devpath_list consists the physpath name of the booting device
2419 * no matter the rootpool is a single device pool or a mirrored pool.
2421 * "/pci@1f,0/ide@d/disk@0,0:a"
2424 spa_import_rootpool(char *devpath
, char *devid
)
2426 nvlist_t
*conf
= NULL
;
2431 * Get the vdev pathname and configuation from the most
2432 * recently updated vdev (highest txg).
2434 if (error
= spa_get_rootconf(devpath
, devid
, &conf
))
2438 * Add type "root" vdev to the config.
2440 spa_build_rootpool_config(conf
);
2442 VERIFY(nvlist_lookup_string(conf
, ZPOOL_CONFIG_POOL_NAME
, &pname
) == 0);
2445 * We specify 'allowfaulted' for this to be treated like spa_open()
2446 * instead of spa_import(). This prevents us from marking vdevs as
2447 * persistently unavailable, and generates FMA ereports as if it were a
2448 * pool open, not import.
2450 error
= spa_import_common(pname
, conf
, NULL
, B_TRUE
, B_TRUE
);
2451 ASSERT(error
!= EEXIST
);
2457 cmn_err(CE_NOTE
, "\n"
2458 " *************************************************** \n"
2459 " * This device is not bootable! * \n"
2460 " * It is either offlined or detached or faulted. * \n"
2461 " * Please try to boot from a different device. * \n"
2462 " *************************************************** ");
2469 * Import a non-root pool into the system.
2472 spa_import(const char *pool
, nvlist_t
*config
, nvlist_t
*props
)
2474 return (spa_import_common(pool
, config
, props
, B_FALSE
, B_FALSE
));
2478 spa_import_faulted(const char *pool
, nvlist_t
*config
, nvlist_t
*props
)
2480 return (spa_import_common(pool
, config
, props
, B_FALSE
, B_TRUE
));
2485 * This (illegal) pool name is used when temporarily importing a spa_t in order
2486 * to get the vdev stats associated with the imported devices.
2488 #define TRYIMPORT_NAME "$import"
2491 spa_tryimport(nvlist_t
*tryconfig
)
2493 nvlist_t
*config
= NULL
;
2498 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
2501 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
2505 * Create and initialize the spa structure.
2507 mutex_enter(&spa_namespace_lock
);
2508 spa
= spa_add(TRYIMPORT_NAME
, NULL
);
2509 spa_activate(spa
, FREAD
);
2512 * Pass off the heavy lifting to spa_load().
2513 * Pass TRUE for mosconfig because the user-supplied config
2514 * is actually the one to trust when doing an import.
2516 (void) spa_load(spa
, tryconfig
, SPA_LOAD_TRYIMPORT
, B_TRUE
);
2519 * If 'tryconfig' was at least parsable, return the current config.
2521 if (spa
->spa_root_vdev
!= NULL
) {
2522 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2523 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
2525 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
2527 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2528 spa
->spa_uberblock
.ub_timestamp
) == 0);
2531 * If the bootfs property exists on this pool then we
2532 * copy it out so that external consumers can tell which
2533 * pools are bootable.
2535 if (spa
->spa_bootfs
) {
2536 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
2539 * We have to play games with the name since the
2540 * pool was opened as TRYIMPORT_NAME.
2542 if (dsl_dsobj_to_dsname(spa_name(spa
),
2543 spa
->spa_bootfs
, tmpname
) == 0) {
2545 char *dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
2547 cp
= strchr(tmpname
, '/');
2549 (void) strlcpy(dsname
, tmpname
,
2552 (void) snprintf(dsname
, MAXPATHLEN
,
2553 "%s/%s", poolname
, ++cp
);
2555 VERIFY(nvlist_add_string(config
,
2556 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
2557 kmem_free(dsname
, MAXPATHLEN
);
2559 kmem_free(tmpname
, MAXPATHLEN
);
2563 * Add the list of hot spares and level 2 cache devices.
2565 spa_add_spares(spa
, config
);
2566 spa_add_l2cache(spa
, config
);
2570 spa_deactivate(spa
);
2572 mutex_exit(&spa_namespace_lock
);
2578 * Pool export/destroy
2580 * The act of destroying or exporting a pool is very simple. We make sure there
2581 * is no more pending I/O and any references to the pool are gone. Then, we
2582 * update the pool state and sync all the labels to disk, removing the
2583 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
2584 * we don't sync the labels or remove the configuration cache.
2587 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
2588 boolean_t force
, boolean_t hardforce
)
2595 if (!(spa_mode_global
& FWRITE
))
2598 mutex_enter(&spa_namespace_lock
);
2599 if ((spa
= spa_lookup(pool
)) == NULL
) {
2600 mutex_exit(&spa_namespace_lock
);
2605 * Put a hold on the pool, drop the namespace lock, stop async tasks,
2606 * reacquire the namespace lock, and see if we can export.
2608 spa_open_ref(spa
, FTAG
);
2609 mutex_exit(&spa_namespace_lock
);
2610 spa_async_suspend(spa
);
2611 mutex_enter(&spa_namespace_lock
);
2612 spa_close(spa
, FTAG
);
2615 * The pool will be in core if it's openable,
2616 * in which case we can modify its state.
2618 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
&& spa
->spa_sync_on
) {
2620 * Objsets may be open only because they're dirty, so we
2621 * have to force it to sync before checking spa_refcnt.
2623 txg_wait_synced(spa
->spa_dsl_pool
, 0);
2626 * A pool cannot be exported or destroyed if there are active
2627 * references. If we are resetting a pool, allow references by
2628 * fault injection handlers.
2630 if (!spa_refcount_zero(spa
) ||
2631 (spa
->spa_inject_ref
!= 0 &&
2632 new_state
!= POOL_STATE_UNINITIALIZED
)) {
2633 spa_async_resume(spa
);
2634 mutex_exit(&spa_namespace_lock
);
2639 * A pool cannot be exported if it has an active shared spare.
2640 * This is to prevent other pools stealing the active spare
2641 * from an exported pool. At user's own will, such pool can
2642 * be forcedly exported.
2644 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
2645 spa_has_active_shared_spare(spa
)) {
2646 spa_async_resume(spa
);
2647 mutex_exit(&spa_namespace_lock
);
2652 * We want this to be reflected on every label,
2653 * so mark them all dirty. spa_unload() will do the
2654 * final sync that pushes these changes out.
2656 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
2657 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2658 spa
->spa_state
= new_state
;
2659 spa
->spa_final_txg
= spa_last_synced_txg(spa
) + 1;
2660 vdev_config_dirty(spa
->spa_root_vdev
);
2661 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2665 spa_event_notify(spa
, NULL
, ESC_ZFS_POOL_DESTROY
);
2667 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
2669 spa_deactivate(spa
);
2672 if (oldconfig
&& spa
->spa_config
)
2673 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
2675 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
2677 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
2680 mutex_exit(&spa_namespace_lock
);
2686 * Destroy a storage pool.
2689 spa_destroy(char *pool
)
2691 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
2696 * Export a storage pool.
2699 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
2700 boolean_t hardforce
)
2702 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
2707 * Similar to spa_export(), this unloads the spa_t without actually removing it
2708 * from the namespace in any way.
2711 spa_reset(char *pool
)
2713 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
2718 * ==========================================================================
2719 * Device manipulation
2720 * ==========================================================================
2724 * Add a device to a storage pool.
2727 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
2731 vdev_t
*rvd
= spa
->spa_root_vdev
;
2733 nvlist_t
**spares
, **l2cache
;
2734 uint_t nspares
, nl2cache
;
2736 txg
= spa_vdev_enter(spa
);
2738 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
2739 VDEV_ALLOC_ADD
)) != 0)
2740 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
2742 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
2744 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
2748 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
2752 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
2753 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
2755 if (vd
->vdev_children
!= 0 &&
2756 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
2757 return (spa_vdev_exit(spa
, vd
, txg
, error
));
2760 * We must validate the spares and l2cache devices after checking the
2761 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
2763 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
2764 return (spa_vdev_exit(spa
, vd
, txg
, error
));
2767 * Transfer each new top-level vdev from vd to rvd.
2769 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2770 tvd
= vd
->vdev_child
[c
];
2771 vdev_remove_child(vd
, tvd
);
2772 tvd
->vdev_id
= rvd
->vdev_children
;
2773 vdev_add_child(rvd
, tvd
);
2774 vdev_config_dirty(tvd
);
2778 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
2779 ZPOOL_CONFIG_SPARES
);
2780 spa_load_spares(spa
);
2781 spa
->spa_spares
.sav_sync
= B_TRUE
;
2784 if (nl2cache
!= 0) {
2785 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
2786 ZPOOL_CONFIG_L2CACHE
);
2787 spa_load_l2cache(spa
);
2788 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2792 * We have to be careful when adding new vdevs to an existing pool.
2793 * If other threads start allocating from these vdevs before we
2794 * sync the config cache, and we lose power, then upon reboot we may
2795 * fail to open the pool because there are DVAs that the config cache
2796 * can't translate. Therefore, we first add the vdevs without
2797 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
2798 * and then let spa_config_update() initialize the new metaslabs.
2800 * spa_load() checks for added-but-not-initialized vdevs, so that
2801 * if we lose power at any point in this sequence, the remaining
2802 * steps will be completed the next time we load the pool.
2804 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
2806 mutex_enter(&spa_namespace_lock
);
2807 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
2808 mutex_exit(&spa_namespace_lock
);
2814 * Attach a device to a mirror. The arguments are the path to any device
2815 * in the mirror, and the nvroot for the new device. If the path specifies
2816 * a device that is not mirrored, we automatically insert the mirror vdev.
2818 * If 'replacing' is specified, the new device is intended to replace the
2819 * existing device; in this case the two devices are made into their own
2820 * mirror using the 'replacing' vdev, which is functionally identical to
2821 * the mirror vdev (it actually reuses all the same ops) but has a few
2822 * extra rules: you can't attach to it after it's been created, and upon
2823 * completion of resilvering, the first disk (the one being replaced)
2824 * is automatically detached.
2827 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
2829 uint64_t txg
, open_txg
;
2830 vdev_t
*rvd
= spa
->spa_root_vdev
;
2831 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
2834 char *oldvdpath
, *newvdpath
;
2838 txg
= spa_vdev_enter(spa
);
2840 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
2843 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
2845 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
2846 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
2848 pvd
= oldvd
->vdev_parent
;
2850 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
2851 VDEV_ALLOC_ADD
)) != 0)
2852 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
2854 if (newrootvd
->vdev_children
!= 1)
2855 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
2857 newvd
= newrootvd
->vdev_child
[0];
2859 if (!newvd
->vdev_ops
->vdev_op_leaf
)
2860 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
2862 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
2863 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
2866 * Spares can't replace logs
2868 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
2869 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
2873 * For attach, the only allowable parent is a mirror or the root
2876 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
2877 pvd
->vdev_ops
!= &vdev_root_ops
)
2878 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
2880 pvops
= &vdev_mirror_ops
;
2883 * Active hot spares can only be replaced by inactive hot
2886 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
2887 pvd
->vdev_child
[1] == oldvd
&&
2888 !spa_has_spare(spa
, newvd
->vdev_guid
))
2889 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
2892 * If the source is a hot spare, and the parent isn't already a
2893 * spare, then we want to create a new hot spare. Otherwise, we
2894 * want to create a replacing vdev. The user is not allowed to
2895 * attach to a spared vdev child unless the 'isspare' state is
2896 * the same (spare replaces spare, non-spare replaces
2899 if (pvd
->vdev_ops
== &vdev_replacing_ops
)
2900 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
2901 else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
2902 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
)
2903 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
2904 else if (pvd
->vdev_ops
!= &vdev_spare_ops
&&
2905 newvd
->vdev_isspare
)
2906 pvops
= &vdev_spare_ops
;
2908 pvops
= &vdev_replacing_ops
;
2912 * Compare the new device size with the replaceable/attachable
2915 if (newvd
->vdev_psize
< vdev_get_rsize(oldvd
))
2916 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
2919 * The new device cannot have a higher alignment requirement
2920 * than the top-level vdev.
2922 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
2923 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
2926 * If this is an in-place replacement, update oldvd's path and devid
2927 * to make it distinguishable from newvd, and unopenable from now on.
2929 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
2930 spa_strfree(oldvd
->vdev_path
);
2931 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
2933 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
2934 newvd
->vdev_path
, "old");
2935 if (oldvd
->vdev_devid
!= NULL
) {
2936 spa_strfree(oldvd
->vdev_devid
);
2937 oldvd
->vdev_devid
= NULL
;
2942 * If the parent is not a mirror, or if we're replacing, insert the new
2943 * mirror/replacing/spare vdev above oldvd.
2945 if (pvd
->vdev_ops
!= pvops
)
2946 pvd
= vdev_add_parent(oldvd
, pvops
);
2948 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
2949 ASSERT(pvd
->vdev_ops
== pvops
);
2950 ASSERT(oldvd
->vdev_parent
== pvd
);
2953 * Extract the new device from its root and add it to pvd.
2955 vdev_remove_child(newrootvd
, newvd
);
2956 newvd
->vdev_id
= pvd
->vdev_children
;
2957 vdev_add_child(pvd
, newvd
);
2960 * If newvd is smaller than oldvd, but larger than its rsize,
2961 * the addition of newvd may have decreased our parent's asize.
2963 pvd
->vdev_asize
= MIN(pvd
->vdev_asize
, newvd
->vdev_asize
);
2965 tvd
= newvd
->vdev_top
;
2966 ASSERT(pvd
->vdev_top
== tvd
);
2967 ASSERT(tvd
->vdev_parent
== rvd
);
2969 vdev_config_dirty(tvd
);
2972 * Set newvd's DTL to [TXG_INITIAL, open_txg]. It will propagate
2973 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
2975 open_txg
= txg
+ TXG_CONCURRENT_STATES
- 1;
2977 vdev_dtl_dirty(newvd
, DTL_MISSING
,
2978 TXG_INITIAL
, open_txg
- TXG_INITIAL
+ 1);
2980 if (newvd
->vdev_isspare
)
2981 spa_spare_activate(newvd
);
2982 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
2983 newvdpath
= spa_strdup(newvd
->vdev_path
);
2984 newvd_isspare
= newvd
->vdev_isspare
;
2987 * Mark newvd's DTL dirty in this txg.
2989 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
2991 (void) spa_vdev_exit(spa
, newrootvd
, open_txg
, 0);
2993 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
2994 if (dmu_tx_assign(tx
, TXG_WAIT
) == 0) {
2995 spa_history_internal_log(LOG_POOL_VDEV_ATTACH
, spa
, tx
,
2996 CRED(), "%s vdev=%s %s vdev=%s",
2997 replacing
&& newvd_isspare
? "spare in" :
2998 replacing
? "replace" : "attach", newvdpath
,
2999 replacing
? "for" : "to", oldvdpath
);
3005 spa_strfree(oldvdpath
);
3006 spa_strfree(newvdpath
);
3009 * Kick off a resilver to update newvd.
3011 VERIFY3U(spa_scrub(spa
, POOL_SCRUB_RESILVER
), ==, 0);
3017 * Detach a device from a mirror or replacing vdev.
3018 * If 'replace_done' is specified, only detach if the parent
3019 * is a replacing vdev.
3022 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
3026 vdev_t
*rvd
= spa
->spa_root_vdev
;
3027 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
3028 boolean_t unspare
= B_FALSE
;
3029 uint64_t unspare_guid
;
3032 txg
= spa_vdev_enter(spa
);
3034 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
3037 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
3039 if (!vd
->vdev_ops
->vdev_op_leaf
)
3040 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3042 pvd
= vd
->vdev_parent
;
3045 * If the parent/child relationship is not as expected, don't do it.
3046 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3047 * vdev that's replacing B with C. The user's intent in replacing
3048 * is to go from M(A,B) to M(A,C). If the user decides to cancel
3049 * the replace by detaching C, the expected behavior is to end up
3050 * M(A,B). But suppose that right after deciding to detach C,
3051 * the replacement of B completes. We would have M(A,C), and then
3052 * ask to detach C, which would leave us with just A -- not what
3053 * the user wanted. To prevent this, we make sure that the
3054 * parent/child relationship hasn't changed -- in this example,
3055 * that C's parent is still the replacing vdev R.
3057 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
3058 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
3061 * If replace_done is specified, only remove this device if it's
3062 * the first child of a replacing vdev. For the 'spare' vdev, either
3063 * disk can be removed.
3066 if (pvd
->vdev_ops
== &vdev_replacing_ops
) {
3067 if (vd
->vdev_id
!= 0)
3068 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3069 } else if (pvd
->vdev_ops
!= &vdev_spare_ops
) {
3070 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3074 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
3075 spa_version(spa
) >= SPA_VERSION_SPARES
);
3078 * Only mirror, replacing, and spare vdevs support detach.
3080 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
3081 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
3082 pvd
->vdev_ops
!= &vdev_spare_ops
)
3083 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3086 * If this device has the only valid copy of some data,
3087 * we cannot safely detach it.
3089 if (vdev_dtl_required(vd
))
3090 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
3092 ASSERT(pvd
->vdev_children
>= 2);
3095 * If we are detaching the second disk from a replacing vdev, then
3096 * check to see if we changed the original vdev's path to have "/old"
3097 * at the end in spa_vdev_attach(). If so, undo that change now.
3099 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
== 1 &&
3100 pvd
->vdev_child
[0]->vdev_path
!= NULL
&&
3101 pvd
->vdev_child
[1]->vdev_path
!= NULL
) {
3102 ASSERT(pvd
->vdev_child
[1] == vd
);
3103 cvd
= pvd
->vdev_child
[0];
3104 len
= strlen(vd
->vdev_path
);
3105 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
3106 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
3107 spa_strfree(cvd
->vdev_path
);
3108 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
3113 * If we are detaching the original disk from a spare, then it implies
3114 * that the spare should become a real disk, and be removed from the
3115 * active spare list for the pool.
3117 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
3118 vd
->vdev_id
== 0 && pvd
->vdev_child
[1]->vdev_isspare
)
3122 * Erase the disk labels so the disk can be used for other things.
3123 * This must be done after all other error cases are handled,
3124 * but before we disembowel vd (so we can still do I/O to it).
3125 * But if we can't do it, don't treat the error as fatal --
3126 * it may be that the unwritability of the disk is the reason
3127 * it's being detached!
3129 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
3132 * Remove vd from its parent and compact the parent's children.
3134 vdev_remove_child(pvd
, vd
);
3135 vdev_compact_children(pvd
);
3138 * Remember one of the remaining children so we can get tvd below.
3140 cvd
= pvd
->vdev_child
[0];
3143 * If we need to remove the remaining child from the list of hot spares,
3144 * do it now, marking the vdev as no longer a spare in the process.
3145 * We must do this before vdev_remove_parent(), because that can
3146 * change the GUID if it creates a new toplevel GUID. For a similar
3147 * reason, we must remove the spare now, in the same txg as the detach;
3148 * otherwise someone could attach a new sibling, change the GUID, and
3149 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3152 ASSERT(cvd
->vdev_isspare
);
3153 spa_spare_remove(cvd
);
3154 unspare_guid
= cvd
->vdev_guid
;
3155 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
3159 * If the parent mirror/replacing vdev only has one child,
3160 * the parent is no longer needed. Remove it from the tree.
3162 if (pvd
->vdev_children
== 1)
3163 vdev_remove_parent(cvd
);
3166 * We don't set tvd until now because the parent we just removed
3167 * may have been the previous top-level vdev.
3169 tvd
= cvd
->vdev_top
;
3170 ASSERT(tvd
->vdev_parent
== rvd
);
3173 * Reevaluate the parent vdev state.
3175 vdev_propagate_state(cvd
);
3178 * If the device we just detached was smaller than the others, it may be
3179 * possible to add metaslabs (i.e. grow the pool). vdev_metaslab_init()
3180 * can't fail because the existing metaslabs are already in core, so
3181 * there's nothing to read from disk.
3183 VERIFY(vdev_metaslab_init(tvd
, txg
) == 0);
3185 vdev_config_dirty(tvd
);
3188 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
3189 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3190 * But first make sure we're not on any *other* txg's DTL list, to
3191 * prevent vd from being accessed after it's freed.
3193 for (int t
= 0; t
< TXG_SIZE
; t
++)
3194 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
3195 vd
->vdev_detached
= B_TRUE
;
3196 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
3198 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_REMOVE
);
3200 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
3203 * If this was the removal of the original device in a hot spare vdev,
3204 * then we want to go through and remove the device from the hot spare
3205 * list of every other pool.
3210 mutex_enter(&spa_namespace_lock
);
3211 while ((spa
= spa_next(spa
)) != NULL
) {
3212 if (spa
->spa_state
!= POOL_STATE_ACTIVE
)
3216 spa_open_ref(spa
, FTAG
);
3217 mutex_exit(&spa_namespace_lock
);
3218 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
3219 mutex_enter(&spa_namespace_lock
);
3220 spa_close(spa
, FTAG
);
3222 mutex_exit(&spa_namespace_lock
);
3229 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
3231 for (int i
= 0; i
< count
; i
++) {
3234 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
3237 if (guid
== target_guid
)
3245 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
3246 nvlist_t
*dev_to_remove
)
3248 nvlist_t
**newdev
= NULL
;
3251 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
3253 for (int i
= 0, j
= 0; i
< count
; i
++) {
3254 if (dev
[i
] == dev_to_remove
)
3256 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
3259 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3260 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
3262 for (int i
= 0; i
< count
- 1; i
++)
3263 nvlist_free(newdev
[i
]);
3266 kmem_free(newdev
, (count
- 1) * sizeof (void *));
3270 * Remove a device from the pool. Currently, this supports removing only hot
3271 * spares and level 2 ARC devices.
3274 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
3277 nvlist_t
**spares
, **l2cache
, *nv
;
3278 uint_t nspares
, nl2cache
;
3281 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
3284 txg
= spa_vdev_enter(spa
);
3286 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
3288 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
3289 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3290 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
3291 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
3293 * Only remove the hot spare if it's not currently in use
3296 if (vd
== NULL
|| unspare
) {
3297 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
3298 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
3299 spa_load_spares(spa
);
3300 spa
->spa_spares
.sav_sync
= B_TRUE
;
3304 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
3305 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3306 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
3307 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
3309 * Cache devices can always be removed.
3311 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
3312 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
3313 spa_load_l2cache(spa
);
3314 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3315 } else if (vd
!= NULL
) {
3317 * Normal vdevs cannot be removed (yet).
3322 * There is no vdev of any kind with the specified guid.
3328 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
3334 * Find any device that's done replacing, or a vdev marked 'unspare' that's
3335 * current spared, so we can detach it.
3338 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
3340 vdev_t
*newvd
, *oldvd
;
3343 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3344 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
3350 * Check for a completed replacement.
3352 if (vd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_children
== 2) {
3353 oldvd
= vd
->vdev_child
[0];
3354 newvd
= vd
->vdev_child
[1];
3356 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
3357 !vdev_dtl_required(oldvd
))
3362 * Check for a completed resilver with the 'unspare' flag set.
3364 if (vd
->vdev_ops
== &vdev_spare_ops
&& vd
->vdev_children
== 2) {
3365 newvd
= vd
->vdev_child
[0];
3366 oldvd
= vd
->vdev_child
[1];
3368 if (newvd
->vdev_unspare
&&
3369 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
3370 !vdev_dtl_required(oldvd
)) {
3371 newvd
->vdev_unspare
= 0;
3380 spa_vdev_resilver_done(spa_t
*spa
)
3382 vdev_t
*vd
, *pvd
, *ppvd
;
3383 uint64_t guid
, sguid
, pguid
, ppguid
;
3385 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3387 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
3388 pvd
= vd
->vdev_parent
;
3389 ppvd
= pvd
->vdev_parent
;
3390 guid
= vd
->vdev_guid
;
3391 pguid
= pvd
->vdev_guid
;
3392 ppguid
= ppvd
->vdev_guid
;
3395 * If we have just finished replacing a hot spared device, then
3396 * we need to detach the parent's first child (the original hot
3399 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0) {
3400 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
3401 ASSERT(ppvd
->vdev_children
== 2);
3402 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
3404 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3405 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
3407 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
3409 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3412 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3416 * Update the stored path for this vdev. Dirty the vdev configuration, relying
3417 * on spa_vdev_enter/exit() to synchronize the labels and cache.
3420 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
3425 txg
= spa_vdev_enter(spa
);
3427 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
) {
3429 * Determine if this is a reference to a hot spare device. If
3430 * it is, update the path manually as there is no associated
3431 * vdev_t that can be synced to disk.
3436 if (spa
->spa_spares
.sav_config
!= NULL
) {
3437 VERIFY(nvlist_lookup_nvlist_array(
3438 spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
3439 &spares
, &nspares
) == 0);
3440 for (i
= 0; i
< nspares
; i
++) {
3442 VERIFY(nvlist_lookup_uint64(spares
[i
],
3443 ZPOOL_CONFIG_GUID
, &theguid
) == 0);
3444 if (theguid
== guid
) {
3445 VERIFY(nvlist_add_string(spares
[i
],
3446 ZPOOL_CONFIG_PATH
, newpath
) == 0);
3447 spa_load_spares(spa
);
3448 spa
->spa_spares
.sav_sync
= B_TRUE
;
3449 return (spa_vdev_exit(spa
, NULL
, txg
,
3455 return (spa_vdev_exit(spa
, NULL
, txg
, ENOENT
));
3458 if (!vd
->vdev_ops
->vdev_op_leaf
)
3459 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3461 spa_strfree(vd
->vdev_path
);
3462 vd
->vdev_path
= spa_strdup(newpath
);
3464 vdev_config_dirty(vd
->vdev_top
);
3466 return (spa_vdev_exit(spa
, NULL
, txg
, 0));
3470 * ==========================================================================
3472 * ==========================================================================
3476 spa_scrub(spa_t
*spa
, pool_scrub_type_t type
)
3478 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
3480 if ((uint_t
)type
>= POOL_SCRUB_TYPES
)
3484 * If a resilver was requested, but there is no DTL on a
3485 * writeable leaf device, we have nothing to do.
3487 if (type
== POOL_SCRUB_RESILVER
&&
3488 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
3489 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
3493 if (type
== POOL_SCRUB_EVERYTHING
&&
3494 spa
->spa_dsl_pool
->dp_scrub_func
!= SCRUB_FUNC_NONE
&&
3495 spa
->spa_dsl_pool
->dp_scrub_isresilver
)
3498 if (type
== POOL_SCRUB_EVERYTHING
|| type
== POOL_SCRUB_RESILVER
) {
3499 return (dsl_pool_scrub_clean(spa
->spa_dsl_pool
));
3500 } else if (type
== POOL_SCRUB_NONE
) {
3501 return (dsl_pool_scrub_cancel(spa
->spa_dsl_pool
));
3508 * ==========================================================================
3509 * SPA async task processing
3510 * ==========================================================================
3514 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
3516 if (vd
->vdev_remove_wanted
) {
3517 vd
->vdev_remove_wanted
= 0;
3518 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
3519 vdev_clear(spa
, vd
);
3520 vdev_state_dirty(vd
->vdev_top
);
3523 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3524 spa_async_remove(spa
, vd
->vdev_child
[c
]);
3528 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
3530 if (vd
->vdev_probe_wanted
) {
3531 vd
->vdev_probe_wanted
= 0;
3532 vdev_reopen(vd
); /* vdev_open() does the actual probe */
3535 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3536 spa_async_probe(spa
, vd
->vdev_child
[c
]);
3540 spa_async_thread(spa_t
*spa
)
3544 ASSERT(spa
->spa_sync_on
);
3546 mutex_enter(&spa
->spa_async_lock
);
3547 tasks
= spa
->spa_async_tasks
;
3548 spa
->spa_async_tasks
= 0;
3549 mutex_exit(&spa
->spa_async_lock
);
3552 * See if the config needs to be updated.
3554 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
3555 mutex_enter(&spa_namespace_lock
);
3556 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
3557 mutex_exit(&spa_namespace_lock
);
3561 * See if any devices need to be marked REMOVED.
3563 if (tasks
& SPA_ASYNC_REMOVE
) {
3564 spa_vdev_state_enter(spa
);
3565 spa_async_remove(spa
, spa
->spa_root_vdev
);
3566 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
3567 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
3568 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
3569 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
3570 (void) spa_vdev_state_exit(spa
, NULL
, 0);
3574 * See if any devices need to be probed.
3576 if (tasks
& SPA_ASYNC_PROBE
) {
3577 spa_vdev_state_enter(spa
);
3578 spa_async_probe(spa
, spa
->spa_root_vdev
);
3579 (void) spa_vdev_state_exit(spa
, NULL
, 0);
3583 * If any devices are done replacing, detach them.
3585 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
3586 spa_vdev_resilver_done(spa
);
3589 * Kick off a resilver.
3591 if (tasks
& SPA_ASYNC_RESILVER
)
3592 VERIFY(spa_scrub(spa
, POOL_SCRUB_RESILVER
) == 0);
3595 * Let the world know that we're done.
3597 mutex_enter(&spa
->spa_async_lock
);
3598 spa
->spa_async_thread
= NULL
;
3599 cv_broadcast(&spa
->spa_async_cv
);
3600 mutex_exit(&spa
->spa_async_lock
);
3605 spa_async_suspend(spa_t
*spa
)
3607 mutex_enter(&spa
->spa_async_lock
);
3608 spa
->spa_async_suspended
++;
3609 while (spa
->spa_async_thread
!= NULL
)
3610 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
3611 mutex_exit(&spa
->spa_async_lock
);
3615 spa_async_resume(spa_t
*spa
)
3617 mutex_enter(&spa
->spa_async_lock
);
3618 ASSERT(spa
->spa_async_suspended
!= 0);
3619 spa
->spa_async_suspended
--;
3620 mutex_exit(&spa
->spa_async_lock
);
3624 spa_async_dispatch(spa_t
*spa
)
3626 mutex_enter(&spa
->spa_async_lock
);
3627 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
3628 spa
->spa_async_thread
== NULL
&&
3629 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
3630 spa
->spa_async_thread
= thread_create(NULL
, 0,
3631 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
3632 mutex_exit(&spa
->spa_async_lock
);
3636 spa_async_request(spa_t
*spa
, int task
)
3638 mutex_enter(&spa
->spa_async_lock
);
3639 spa
->spa_async_tasks
|= task
;
3640 mutex_exit(&spa
->spa_async_lock
);
3644 * ==========================================================================
3645 * SPA syncing routines
3646 * ==========================================================================
3650 spa_sync_deferred_frees(spa_t
*spa
, uint64_t txg
)
3652 bplist_t
*bpl
= &spa
->spa_sync_bplist
;
3660 zio
= zio_root(spa
, NULL
, NULL
, ZIO_FLAG_CANFAIL
);
3662 while (bplist_iterate(bpl
, &itor
, &blk
) == 0) {
3663 ASSERT(blk
.blk_birth
< txg
);
3664 zio_nowait(zio_free(zio
, spa
, txg
, &blk
, NULL
, NULL
,
3665 ZIO_FLAG_MUSTSUCCEED
));
3668 error
= zio_wait(zio
);
3669 ASSERT3U(error
, ==, 0);
3671 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3672 bplist_vacate(bpl
, tx
);
3675 * Pre-dirty the first block so we sync to convergence faster.
3676 * (Usually only the first block is needed.)
3678 dmu_write(spa
->spa_meta_objset
, spa
->spa_sync_bplist_obj
, 0, 1, &c
, tx
);
3683 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
3685 char *packed
= NULL
;
3690 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
3693 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
3694 * information. This avoids the dbuf_will_dirty() path and
3695 * saves us a pre-read to get data we don't actually care about.
3697 bufsize
= P2ROUNDUP(nvsize
, SPA_CONFIG_BLOCKSIZE
);
3698 packed
= kmem_alloc(bufsize
, KM_SLEEP
);
3700 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
3702 bzero(packed
+ nvsize
, bufsize
- nvsize
);
3704 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
3706 kmem_free(packed
, bufsize
);
3708 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
3709 dmu_buf_will_dirty(db
, tx
);
3710 *(uint64_t *)db
->db_data
= nvsize
;
3711 dmu_buf_rele(db
, FTAG
);
3715 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
3716 const char *config
, const char *entry
)
3726 * Update the MOS nvlist describing the list of available devices.
3727 * spa_validate_aux() will have already made sure this nvlist is
3728 * valid and the vdevs are labeled appropriately.
3730 if (sav
->sav_object
== 0) {
3731 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3732 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
3733 sizeof (uint64_t), tx
);
3734 VERIFY(zap_update(spa
->spa_meta_objset
,
3735 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
3736 &sav
->sav_object
, tx
) == 0);
3739 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3740 if (sav
->sav_count
== 0) {
3741 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
3743 list
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
3744 for (i
= 0; i
< sav
->sav_count
; i
++)
3745 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
3746 B_FALSE
, B_FALSE
, B_TRUE
);
3747 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
3748 sav
->sav_count
) == 0);
3749 for (i
= 0; i
< sav
->sav_count
; i
++)
3750 nvlist_free(list
[i
]);
3751 kmem_free(list
, sav
->sav_count
* sizeof (void *));
3754 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
3755 nvlist_free(nvroot
);
3757 sav
->sav_sync
= B_FALSE
;
3761 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
3765 if (list_is_empty(&spa
->spa_config_dirty_list
))
3768 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
3770 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
3771 dmu_tx_get_txg(tx
), B_FALSE
);
3773 spa_config_exit(spa
, SCL_STATE
, FTAG
);
3775 if (spa
->spa_config_syncing
)
3776 nvlist_free(spa
->spa_config_syncing
);
3777 spa
->spa_config_syncing
= config
;
3779 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
3783 * Set zpool properties.
3786 spa_sync_props(void *arg1
, void *arg2
, cred_t
*cr
, dmu_tx_t
*tx
)
3789 objset_t
*mos
= spa
->spa_meta_objset
;
3790 nvlist_t
*nvp
= arg2
;
3795 const char *propname
;
3796 zprop_type_t proptype
;
3797 spa_config_dirent_t
*dp
;
3799 mutex_enter(&spa
->spa_props_lock
);
3802 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
3803 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
3804 case ZPOOL_PROP_VERSION
:
3806 * Only set version for non-zpool-creation cases
3807 * (set/import). spa_create() needs special care
3808 * for version setting.
3810 if (tx
->tx_txg
!= TXG_INITIAL
) {
3811 VERIFY(nvpair_value_uint64(elem
,
3813 ASSERT(intval
<= SPA_VERSION
);
3814 ASSERT(intval
>= spa_version(spa
));
3815 spa
->spa_uberblock
.ub_version
= intval
;
3816 vdev_config_dirty(spa
->spa_root_vdev
);
3820 case ZPOOL_PROP_ALTROOT
:
3822 * 'altroot' is a non-persistent property. It should
3823 * have been set temporarily at creation or import time.
3825 ASSERT(spa
->spa_root
!= NULL
);
3828 case ZPOOL_PROP_CACHEFILE
:
3830 * 'cachefile' is a non-persistent property, but note
3831 * an async request that the config cache needs to be
3834 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
3836 dp
= kmem_alloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
3838 if (strval
[0] == '\0')
3839 dp
->scd_path
= spa_strdup(spa_config_path
);
3840 else if (strcmp(strval
, "none") == 0)
3841 dp
->scd_path
= NULL
;
3843 dp
->scd_path
= spa_strdup(strval
);
3845 list_insert_head(&spa
->spa_config_list
, dp
);
3846 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3850 * Set pool property values in the poolprops mos object.
3852 if (spa
->spa_pool_props_object
== 0) {
3853 objset_t
*mos
= spa
->spa_meta_objset
;
3855 VERIFY((spa
->spa_pool_props_object
=
3856 zap_create(mos
, DMU_OT_POOL_PROPS
,
3857 DMU_OT_NONE
, 0, tx
)) > 0);
3859 VERIFY(zap_update(mos
,
3860 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
3861 8, 1, &spa
->spa_pool_props_object
, tx
)
3865 /* normalize the property name */
3866 propname
= zpool_prop_to_name(prop
);
3867 proptype
= zpool_prop_get_type(prop
);
3869 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
3870 ASSERT(proptype
== PROP_TYPE_STRING
);
3871 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
3872 VERIFY(zap_update(mos
,
3873 spa
->spa_pool_props_object
, propname
,
3874 1, strlen(strval
) + 1, strval
, tx
) == 0);
3876 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
3877 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
3879 if (proptype
== PROP_TYPE_INDEX
) {
3881 VERIFY(zpool_prop_index_to_string(
3882 prop
, intval
, &unused
) == 0);
3884 VERIFY(zap_update(mos
,
3885 spa
->spa_pool_props_object
, propname
,
3886 8, 1, &intval
, tx
) == 0);
3888 ASSERT(0); /* not allowed */
3892 case ZPOOL_PROP_DELEGATION
:
3893 spa
->spa_delegation
= intval
;
3895 case ZPOOL_PROP_BOOTFS
:
3896 spa
->spa_bootfs
= intval
;
3898 case ZPOOL_PROP_FAILUREMODE
:
3899 spa
->spa_failmode
= intval
;
3906 /* log internal history if this is not a zpool create */
3907 if (spa_version(spa
) >= SPA_VERSION_ZPOOL_HISTORY
&&
3908 tx
->tx_txg
!= TXG_INITIAL
) {
3909 spa_history_internal_log(LOG_POOL_PROPSET
,
3910 spa
, tx
, cr
, "%s %lld %s",
3911 nvpair_name(elem
), intval
, spa_name(spa
));
3915 mutex_exit(&spa
->spa_props_lock
);
3919 * Sync the specified transaction group. New blocks may be dirtied as
3920 * part of the process, so we iterate until it converges.
3923 spa_sync(spa_t
*spa
, uint64_t txg
)
3925 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
3926 objset_t
*mos
= spa
->spa_meta_objset
;
3927 bplist_t
*bpl
= &spa
->spa_sync_bplist
;
3928 vdev_t
*rvd
= spa
->spa_root_vdev
;
3935 * Lock out configuration changes.
3937 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3939 spa
->spa_syncing_txg
= txg
;
3940 spa
->spa_sync_pass
= 0;
3943 * If there are any pending vdev state changes, convert them
3944 * into config changes that go out with this transaction group.
3946 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
3947 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
3949 * We need the write lock here because, for aux vdevs,
3950 * calling vdev_config_dirty() modifies sav_config.
3951 * This is ugly and will become unnecessary when we
3952 * eliminate the aux vdev wart by integrating all vdevs
3953 * into the root vdev tree.
3955 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
3956 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
3957 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
3958 vdev_state_clean(vd
);
3959 vdev_config_dirty(vd
);
3961 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
3962 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
3964 spa_config_exit(spa
, SCL_STATE
, FTAG
);
3966 VERIFY(0 == bplist_open(bpl
, mos
, spa
->spa_sync_bplist_obj
));
3968 tx
= dmu_tx_create_assigned(dp
, txg
);
3971 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
3972 * set spa_deflate if we have no raid-z vdevs.
3974 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
3975 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3978 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
3979 vd
= rvd
->vdev_child
[i
];
3980 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
3983 if (i
== rvd
->vdev_children
) {
3984 spa
->spa_deflate
= TRUE
;
3985 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
3986 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3987 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
3991 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
3992 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
3993 dsl_pool_create_origin(dp
, tx
);
3995 /* Keeping the origin open increases spa_minref */
3996 spa
->spa_minref
+= 3;
3999 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
4000 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
4001 dsl_pool_upgrade_clones(dp
, tx
);
4005 * If anything has changed in this txg, push the deferred frees
4006 * from the previous txg. If not, leave them alone so that we
4007 * don't generate work on an otherwise idle system.
4009 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
4010 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
4011 !txg_list_empty(&dp
->dp_sync_tasks
, txg
))
4012 spa_sync_deferred_frees(spa
, txg
);
4015 * Iterate to convergence.
4018 spa
->spa_sync_pass
++;
4020 spa_sync_config_object(spa
, tx
);
4021 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
4022 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
4023 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
4024 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
4025 spa_errlog_sync(spa
, txg
);
4026 dsl_pool_sync(dp
, txg
);
4029 while (vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)) {
4034 bplist_sync(bpl
, tx
);
4035 } while (dirty_vdevs
);
4039 dprintf("txg %llu passes %d\n", txg
, spa
->spa_sync_pass
);
4042 * Rewrite the vdev configuration (which includes the uberblock)
4043 * to commit the transaction group.
4045 * If there are no dirty vdevs, we sync the uberblock to a few
4046 * random top-level vdevs that are known to be visible in the
4047 * config cache (see spa_vdev_add() for a complete description).
4048 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4052 * We hold SCL_STATE to prevent vdev open/close/etc.
4053 * while we're attempting to write the vdev labels.
4055 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4057 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
4058 vdev_t
*svd
[SPA_DVAS_PER_BP
];
4060 int children
= rvd
->vdev_children
;
4061 int c0
= spa_get_random(children
);
4064 for (c
= 0; c
< children
; c
++) {
4065 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
4066 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
4068 svd
[svdcount
++] = vd
;
4069 if (svdcount
== SPA_DVAS_PER_BP
)
4072 error
= vdev_config_sync(svd
, svdcount
, txg
);
4074 error
= vdev_config_sync(rvd
->vdev_child
,
4075 rvd
->vdev_children
, txg
);
4078 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4082 zio_suspend(spa
, NULL
);
4083 zio_resume_wait(spa
);
4088 * Clear the dirty config list.
4090 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
4091 vdev_config_clean(vd
);
4094 * Now that the new config has synced transactionally,
4095 * let it become visible to the config cache.
4097 if (spa
->spa_config_syncing
!= NULL
) {
4098 spa_config_set(spa
, spa
->spa_config_syncing
);
4099 spa
->spa_config_txg
= txg
;
4100 spa
->spa_config_syncing
= NULL
;
4103 spa
->spa_ubsync
= spa
->spa_uberblock
;
4106 * Clean up the ZIL records for the synced txg.
4108 dsl_pool_zil_clean(dp
);
4111 * Update usable space statistics.
4113 while (vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
4114 vdev_sync_done(vd
, txg
);
4117 * It had better be the case that we didn't dirty anything
4118 * since vdev_config_sync().
4120 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
4121 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
4122 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
4123 ASSERT(bpl
->bpl_queue
== NULL
);
4125 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4128 * If any async tasks have been requested, kick them off.
4130 spa_async_dispatch(spa
);
4134 * Sync all pools. We don't want to hold the namespace lock across these
4135 * operations, so we take a reference on the spa_t and drop the lock during the
4139 spa_sync_allpools(void)
4142 mutex_enter(&spa_namespace_lock
);
4143 while ((spa
= spa_next(spa
)) != NULL
) {
4144 if (spa_state(spa
) != POOL_STATE_ACTIVE
|| spa_suspended(spa
))
4146 spa_open_ref(spa
, FTAG
);
4147 mutex_exit(&spa_namespace_lock
);
4148 txg_wait_synced(spa_get_dsl(spa
), 0);
4149 mutex_enter(&spa_namespace_lock
);
4150 spa_close(spa
, FTAG
);
4152 mutex_exit(&spa_namespace_lock
);
4156 * ==========================================================================
4157 * Miscellaneous routines
4158 * ==========================================================================
4162 * Remove all pools in the system.
4170 * Remove all cached state. All pools should be closed now,
4171 * so every spa in the AVL tree should be unreferenced.
4173 mutex_enter(&spa_namespace_lock
);
4174 while ((spa
= spa_next(NULL
)) != NULL
) {
4176 * Stop async tasks. The async thread may need to detach
4177 * a device that's been replaced, which requires grabbing
4178 * spa_namespace_lock, so we must drop it here.
4180 spa_open_ref(spa
, FTAG
);
4181 mutex_exit(&spa_namespace_lock
);
4182 spa_async_suspend(spa
);
4183 mutex_enter(&spa_namespace_lock
);
4184 spa_close(spa
, FTAG
);
4186 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4188 spa_deactivate(spa
);
4192 mutex_exit(&spa_namespace_lock
);
4196 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t l2cache
)
4201 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
4205 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
4206 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
4207 if (vd
->vdev_guid
== guid
)
4216 spa_upgrade(spa_t
*spa
, uint64_t version
)
4218 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4221 * This should only be called for a non-faulted pool, and since a
4222 * future version would result in an unopenable pool, this shouldn't be
4225 ASSERT(spa
->spa_uberblock
.ub_version
<= SPA_VERSION
);
4226 ASSERT(version
>= spa
->spa_uberblock
.ub_version
);
4228 spa
->spa_uberblock
.ub_version
= version
;
4229 vdev_config_dirty(spa
->spa_root_vdev
);
4231 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4233 txg_wait_synced(spa_get_dsl(spa
), 0);
4237 spa_has_spare(spa_t
*spa
, uint64_t guid
)
4241 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
4243 for (i
= 0; i
< sav
->sav_count
; i
++)
4244 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
4247 for (i
= 0; i
< sav
->sav_npending
; i
++) {
4248 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
4249 &spareguid
) == 0 && spareguid
== guid
)
4257 * Check if a pool has an active shared spare device.
4258 * Note: reference count of an active spare is 2, as a spare and as a replace
4261 spa_has_active_shared_spare(spa_t
*spa
)
4265 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
4267 for (i
= 0; i
< sav
->sav_count
; i
++) {
4268 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
4269 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
4278 * Post a sysevent corresponding to the given event. The 'name' must be one of
4279 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
4280 * filled in from the spa and (optionally) the vdev. This doesn't do anything
4281 * in the userland libzpool, as we don't want consumers to misinterpret ztest
4282 * or zdb as real changes.
4285 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
4289 sysevent_attr_list_t
*attr
= NULL
;
4290 sysevent_value_t value
;
4293 ev
= sysevent_alloc(EC_ZFS
, (char *)name
, SUNW_KERN_PUB
"zfs",
4296 value
.value_type
= SE_DATA_TYPE_STRING
;
4297 value
.value
.sv_string
= spa_name(spa
);
4298 if (sysevent_add_attr(&attr
, ZFS_EV_POOL_NAME
, &value
, SE_SLEEP
) != 0)
4301 value
.value_type
= SE_DATA_TYPE_UINT64
;
4302 value
.value
.sv_uint64
= spa_guid(spa
);
4303 if (sysevent_add_attr(&attr
, ZFS_EV_POOL_GUID
, &value
, SE_SLEEP
) != 0)
4307 value
.value_type
= SE_DATA_TYPE_UINT64
;
4308 value
.value
.sv_uint64
= vd
->vdev_guid
;
4309 if (sysevent_add_attr(&attr
, ZFS_EV_VDEV_GUID
, &value
,
4313 if (vd
->vdev_path
) {
4314 value
.value_type
= SE_DATA_TYPE_STRING
;
4315 value
.value
.sv_string
= vd
->vdev_path
;
4316 if (sysevent_add_attr(&attr
, ZFS_EV_VDEV_PATH
,
4317 &value
, SE_SLEEP
) != 0)
4322 if (sysevent_attach_attributes(ev
, attr
) != 0)
4326 (void) log_sysevent(ev
, SE_SLEEP
, &eid
);
4330 sysevent_free_attr(attr
);