4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
27 * This file contains all the routines used when modifying on-disk SPA state.
28 * This includes opening, importing, destroying, exporting a pool, and syncing a
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
36 #include <sys/zio_checksum.h>
38 #include <sys/dmu_tx.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/vdev_disk.h>
44 #include <sys/metaslab.h>
45 #include <sys/metaslab_impl.h>
46 #include <sys/uberblock_impl.h>
49 #include <sys/dmu_traverse.h>
50 #include <sys/dmu_objset.h>
51 #include <sys/unique.h>
52 #include <sys/dsl_pool.h>
53 #include <sys/dsl_dataset.h>
54 #include <sys/dsl_dir.h>
55 #include <sys/dsl_prop.h>
56 #include <sys/dsl_synctask.h>
57 #include <sys/fs/zfs.h>
59 #include <sys/callb.h>
60 #include <sys/systeminfo.h>
61 #include <sys/spa_boot.h>
62 #include <sys/zfs_ioctl.h>
63 #include <sys/dsl_scan.h>
66 #include <sys/bootprops.h>
67 #include <sys/callb.h>
68 #include <sys/cpupart.h>
70 #include <sys/sysdc.h>
75 #include "zfs_comutil.h"
77 typedef enum zti_modes
{
78 zti_mode_fixed
, /* value is # of threads (min 1) */
79 zti_mode_online_percent
, /* value is % of online CPUs */
80 zti_mode_batch
, /* cpu-intensive; value is ignored */
81 zti_mode_null
, /* don't create a taskq */
85 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
86 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
87 #define ZTI_BATCH { zti_mode_batch, 0 }
88 #define ZTI_NULL { zti_mode_null, 0 }
90 #define ZTI_ONE ZTI_FIX(1)
92 typedef struct zio_taskq_info
{
93 enum zti_modes zti_mode
;
97 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
98 "issue", "issue_high", "intr", "intr_high"
102 * Define the taskq threads for the following I/O types:
103 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
105 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
106 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
107 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
108 { ZTI_FIX(8), ZTI_NULL
, ZTI_BATCH
, ZTI_NULL
},
109 { ZTI_BATCH
, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) },
110 { ZTI_FIX(100), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
111 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
112 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
},
115 static dsl_syncfunc_t spa_sync_props
;
116 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
117 static inline int spa_load_impl(spa_t
*spa
, uint64_t, nvlist_t
*config
,
118 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
120 static void spa_vdev_resilver_done(spa_t
*spa
);
122 uint_t zio_taskq_batch_pct
= 100; /* 1 thread per cpu in pset */
123 id_t zio_taskq_psrset_bind
= PS_NONE
;
124 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
125 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
127 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
130 * This (illegal) pool name is used when temporarily importing a spa_t in order
131 * to get the vdev stats associated with the imported devices.
133 #define TRYIMPORT_NAME "$import"
136 * ==========================================================================
137 * SPA properties routines
138 * ==========================================================================
142 * Add a (source=src, propname=propval) list to an nvlist.
145 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
146 uint64_t intval
, zprop_source_t src
)
148 const char *propname
= zpool_prop_to_name(prop
);
151 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
152 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
155 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
157 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
159 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
160 nvlist_free(propval
);
164 * Get property values from the spa configuration.
167 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
171 uint64_t cap
, version
;
172 zprop_source_t src
= ZPROP_SRC_NONE
;
173 spa_config_dirent_t
*dp
;
175 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
177 if (spa
->spa_root_vdev
!= NULL
) {
178 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
179 size
= metaslab_class_get_space(spa_normal_class(spa
));
180 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
181 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
182 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
183 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
185 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
186 (spa_mode(spa
) == FREAD
), src
);
188 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
189 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
191 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
192 ddt_get_pool_dedup_ratio(spa
), src
);
194 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
195 spa
->spa_root_vdev
->vdev_state
, src
);
197 version
= spa_version(spa
);
198 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
))
199 src
= ZPROP_SRC_DEFAULT
;
201 src
= ZPROP_SRC_LOCAL
;
202 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
205 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
207 if (spa
->spa_root
!= NULL
)
208 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
211 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
212 if (dp
->scd_path
== NULL
) {
213 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
214 "none", 0, ZPROP_SRC_LOCAL
);
215 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
216 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
217 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
223 * Get zpool property values.
226 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
228 objset_t
*mos
= spa
->spa_meta_objset
;
233 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
237 mutex_enter(&spa
->spa_props_lock
);
240 * Get properties from the spa config.
242 spa_prop_get_config(spa
, nvp
);
244 /* If no pool property object, no more prop to get. */
245 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
246 mutex_exit(&spa
->spa_props_lock
);
251 * Get properties from the MOS pool property object.
253 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
254 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
255 zap_cursor_advance(&zc
)) {
258 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
261 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
264 switch (za
.za_integer_length
) {
266 /* integer property */
267 if (za
.za_first_integer
!=
268 zpool_prop_default_numeric(prop
))
269 src
= ZPROP_SRC_LOCAL
;
271 if (prop
== ZPOOL_PROP_BOOTFS
) {
273 dsl_dataset_t
*ds
= NULL
;
275 dp
= spa_get_dsl(spa
);
276 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
277 if ((err
= dsl_dataset_hold_obj(dp
,
278 za
.za_first_integer
, FTAG
, &ds
))) {
279 rw_exit(&dp
->dp_config_rwlock
);
284 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1,
286 dsl_dataset_name(ds
, strval
);
287 dsl_dataset_rele(ds
, FTAG
);
288 rw_exit(&dp
->dp_config_rwlock
);
291 intval
= za
.za_first_integer
;
294 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
298 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1);
303 /* string property */
304 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
305 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
306 za
.za_name
, 1, za
.za_num_integers
, strval
);
308 kmem_free(strval
, za
.za_num_integers
);
311 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
312 kmem_free(strval
, za
.za_num_integers
);
319 zap_cursor_fini(&zc
);
320 mutex_exit(&spa
->spa_props_lock
);
322 if (err
&& err
!= ENOENT
) {
332 * Validate the given pool properties nvlist and modify the list
333 * for the property values to be set.
336 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
339 int error
= 0, reset_bootfs
= 0;
343 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
345 char *propname
, *strval
;
350 propname
= nvpair_name(elem
);
352 if ((prop
= zpool_name_to_prop(propname
)) == ZPROP_INVAL
)
356 case ZPOOL_PROP_VERSION
:
357 error
= nvpair_value_uint64(elem
, &intval
);
359 (intval
< spa_version(spa
) || intval
> SPA_VERSION
))
363 case ZPOOL_PROP_DELEGATION
:
364 case ZPOOL_PROP_AUTOREPLACE
:
365 case ZPOOL_PROP_LISTSNAPS
:
366 case ZPOOL_PROP_AUTOEXPAND
:
367 error
= nvpair_value_uint64(elem
, &intval
);
368 if (!error
&& intval
> 1)
372 case ZPOOL_PROP_BOOTFS
:
374 * If the pool version is less than SPA_VERSION_BOOTFS,
375 * or the pool is still being created (version == 0),
376 * the bootfs property cannot be set.
378 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
384 * Make sure the vdev config is bootable
386 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
393 error
= nvpair_value_string(elem
, &strval
);
398 if (strval
== NULL
|| strval
[0] == '\0') {
399 objnum
= zpool_prop_default_numeric(
404 if ((error
= dmu_objset_hold(strval
,FTAG
,&os
)))
407 /* Must be ZPL and not gzip compressed. */
409 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
411 } else if ((error
= dsl_prop_get_integer(strval
,
412 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
413 &compress
, NULL
)) == 0 &&
414 !BOOTFS_COMPRESS_VALID(compress
)) {
417 objnum
= dmu_objset_id(os
);
419 dmu_objset_rele(os
, FTAG
);
423 case ZPOOL_PROP_FAILUREMODE
:
424 error
= nvpair_value_uint64(elem
, &intval
);
425 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
426 intval
> ZIO_FAILURE_MODE_PANIC
))
430 * This is a special case which only occurs when
431 * the pool has completely failed. This allows
432 * the user to change the in-core failmode property
433 * without syncing it out to disk (I/Os might
434 * currently be blocked). We do this by returning
435 * EIO to the caller (spa_prop_set) to trick it
436 * into thinking we encountered a property validation
439 if (!error
&& spa_suspended(spa
)) {
440 spa
->spa_failmode
= intval
;
445 case ZPOOL_PROP_CACHEFILE
:
446 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
449 if (strval
[0] == '\0')
452 if (strcmp(strval
, "none") == 0)
455 if (strval
[0] != '/') {
460 slash
= strrchr(strval
, '/');
461 ASSERT(slash
!= NULL
);
463 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
464 strcmp(slash
, "/..") == 0)
468 case ZPOOL_PROP_DEDUPDITTO
:
469 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
472 error
= nvpair_value_uint64(elem
, &intval
);
474 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
486 if (!error
&& reset_bootfs
) {
487 error
= nvlist_remove(props
,
488 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
491 error
= nvlist_add_uint64(props
,
492 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
500 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
503 spa_config_dirent_t
*dp
;
505 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
509 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
512 if (cachefile
[0] == '\0')
513 dp
->scd_path
= spa_strdup(spa_config_path
);
514 else if (strcmp(cachefile
, "none") == 0)
517 dp
->scd_path
= spa_strdup(cachefile
);
519 list_insert_head(&spa
->spa_config_list
, dp
);
521 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
525 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
529 boolean_t need_sync
= B_FALSE
;
532 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
536 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
537 if ((prop
= zpool_name_to_prop(
538 nvpair_name(elem
))) == ZPROP_INVAL
)
541 if (prop
== ZPOOL_PROP_CACHEFILE
||
542 prop
== ZPOOL_PROP_ALTROOT
||
543 prop
== ZPOOL_PROP_READONLY
)
551 return (dsl_sync_task_do(spa_get_dsl(spa
), NULL
, spa_sync_props
,
558 * If the bootfs property value is dsobj, clear it.
561 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
563 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
564 VERIFY(zap_remove(spa
->spa_meta_objset
,
565 spa
->spa_pool_props_object
,
566 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
572 * ==========================================================================
573 * SPA state manipulation (open/create/destroy/import/export)
574 * ==========================================================================
578 spa_error_entry_compare(const void *a
, const void *b
)
580 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
581 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
584 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
585 sizeof (zbookmark_t
));
596 * Utility function which retrieves copies of the current logs and
597 * re-initializes them in the process.
600 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
602 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
604 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
605 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
607 avl_create(&spa
->spa_errlist_scrub
,
608 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
609 offsetof(spa_error_entry_t
, se_avl
));
610 avl_create(&spa
->spa_errlist_last
,
611 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
612 offsetof(spa_error_entry_t
, se_avl
));
616 spa_taskq_create(spa_t
*spa
, const char *name
, enum zti_modes mode
,
619 uint_t flags
= TASKQ_PREPOPULATE
;
620 boolean_t batch
= B_FALSE
;
624 return (NULL
); /* no taskq needed */
627 ASSERT3U(value
, >=, 1);
628 value
= MAX(value
, 1);
633 flags
|= TASKQ_THREADS_CPU_PCT
;
634 value
= zio_taskq_batch_pct
;
637 case zti_mode_online_percent
:
638 flags
|= TASKQ_THREADS_CPU_PCT
;
642 panic("unrecognized mode for %s taskq (%u:%u) in "
648 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
650 flags
|= TASKQ_DC_BATCH
;
652 return (taskq_create_sysdc(name
, value
, 50, INT_MAX
,
653 spa
->spa_proc
, zio_taskq_basedc
, flags
));
655 return (taskq_create_proc(name
, value
, maxclsyspri
, 50, INT_MAX
,
656 spa
->spa_proc
, flags
));
660 spa_create_zio_taskqs(spa_t
*spa
)
664 for (t
= 0; t
< ZIO_TYPES
; t
++) {
665 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
666 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
667 enum zti_modes mode
= ztip
->zti_mode
;
668 uint_t value
= ztip
->zti_value
;
671 (void) snprintf(name
, sizeof (name
),
672 "%s_%s", zio_type_name
[t
], zio_taskq_types
[q
]);
674 spa
->spa_zio_taskq
[t
][q
] =
675 spa_taskq_create(spa
, name
, mode
, value
);
682 spa_thread(void *arg
)
687 user_t
*pu
= PTOU(curproc
);
689 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
692 ASSERT(curproc
!= &p0
);
693 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
694 "zpool-%s", spa
->spa_name
);
695 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
697 /* bind this thread to the requested psrset */
698 if (zio_taskq_psrset_bind
!= PS_NONE
) {
700 mutex_enter(&cpu_lock
);
701 mutex_enter(&pidlock
);
702 mutex_enter(&curproc
->p_lock
);
704 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
705 0, NULL
, NULL
) == 0) {
706 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
709 "Couldn't bind process for zfs pool \"%s\" to "
710 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
713 mutex_exit(&curproc
->p_lock
);
714 mutex_exit(&pidlock
);
715 mutex_exit(&cpu_lock
);
719 if (zio_taskq_sysdc
) {
720 sysdc_thread_enter(curthread
, 100, 0);
723 spa
->spa_proc
= curproc
;
724 spa
->spa_did
= curthread
->t_did
;
726 spa_create_zio_taskqs(spa
);
728 mutex_enter(&spa
->spa_proc_lock
);
729 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
731 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
732 cv_broadcast(&spa
->spa_proc_cv
);
734 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
735 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
736 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
737 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
739 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
740 spa
->spa_proc_state
= SPA_PROC_GONE
;
742 cv_broadcast(&spa
->spa_proc_cv
);
743 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
745 mutex_enter(&curproc
->p_lock
);
751 * Activate an uninitialized pool.
754 spa_activate(spa_t
*spa
, int mode
)
756 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
758 spa
->spa_state
= POOL_STATE_ACTIVE
;
759 spa
->spa_mode
= mode
;
761 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
762 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
764 /* Try to create a covering process */
765 mutex_enter(&spa
->spa_proc_lock
);
766 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
767 ASSERT(spa
->spa_proc
== &p0
);
770 /* Only create a process if we're going to be around a while. */
771 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
772 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
774 spa
->spa_proc_state
= SPA_PROC_CREATED
;
775 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
776 cv_wait(&spa
->spa_proc_cv
,
777 &spa
->spa_proc_lock
);
779 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
780 ASSERT(spa
->spa_proc
!= &p0
);
781 ASSERT(spa
->spa_did
!= 0);
785 "Couldn't create process for zfs pool \"%s\"\n",
790 mutex_exit(&spa
->spa_proc_lock
);
792 /* If we didn't create a process, we need to create our taskqs. */
793 if (spa
->spa_proc
== &p0
) {
794 spa_create_zio_taskqs(spa
);
797 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
798 offsetof(vdev_t
, vdev_config_dirty_node
));
799 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
800 offsetof(vdev_t
, vdev_state_dirty_node
));
802 txg_list_create(&spa
->spa_vdev_txg_list
,
803 offsetof(struct vdev
, vdev_txg_node
));
805 avl_create(&spa
->spa_errlist_scrub
,
806 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
807 offsetof(spa_error_entry_t
, se_avl
));
808 avl_create(&spa
->spa_errlist_last
,
809 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
810 offsetof(spa_error_entry_t
, se_avl
));
814 * Opposite of spa_activate().
817 spa_deactivate(spa_t
*spa
)
821 ASSERT(spa
->spa_sync_on
== B_FALSE
);
822 ASSERT(spa
->spa_dsl_pool
== NULL
);
823 ASSERT(spa
->spa_root_vdev
== NULL
);
824 ASSERT(spa
->spa_async_zio_root
== NULL
);
825 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
827 txg_list_destroy(&spa
->spa_vdev_txg_list
);
829 list_destroy(&spa
->spa_config_dirty_list
);
830 list_destroy(&spa
->spa_state_dirty_list
);
832 for (t
= 0; t
< ZIO_TYPES
; t
++) {
833 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
834 if (spa
->spa_zio_taskq
[t
][q
] != NULL
)
835 taskq_destroy(spa
->spa_zio_taskq
[t
][q
]);
836 spa
->spa_zio_taskq
[t
][q
] = NULL
;
840 metaslab_class_destroy(spa
->spa_normal_class
);
841 spa
->spa_normal_class
= NULL
;
843 metaslab_class_destroy(spa
->spa_log_class
);
844 spa
->spa_log_class
= NULL
;
847 * If this was part of an import or the open otherwise failed, we may
848 * still have errors left in the queues. Empty them just in case.
850 spa_errlog_drain(spa
);
852 avl_destroy(&spa
->spa_errlist_scrub
);
853 avl_destroy(&spa
->spa_errlist_last
);
855 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
857 mutex_enter(&spa
->spa_proc_lock
);
858 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
859 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
860 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
861 cv_broadcast(&spa
->spa_proc_cv
);
862 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
863 ASSERT(spa
->spa_proc
!= &p0
);
864 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
866 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
867 spa
->spa_proc_state
= SPA_PROC_NONE
;
869 ASSERT(spa
->spa_proc
== &p0
);
870 mutex_exit(&spa
->spa_proc_lock
);
873 * We want to make sure spa_thread() has actually exited the ZFS
874 * module, so that the module can't be unloaded out from underneath
877 if (spa
->spa_did
!= 0) {
878 thread_join(spa
->spa_did
);
884 * Verify a pool configuration, and construct the vdev tree appropriately. This
885 * will create all the necessary vdevs in the appropriate layout, with each vdev
886 * in the CLOSED state. This will prep the pool before open/creation/import.
887 * All vdev validation is done by the vdev_alloc() routine.
890 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
891 uint_t id
, int atype
)
898 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
901 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
904 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
916 for (c
= 0; c
< children
; c
++) {
918 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
926 ASSERT(*vdp
!= NULL
);
932 * Opposite of spa_load().
935 spa_unload(spa_t
*spa
)
939 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
944 spa_async_suspend(spa
);
949 if (spa
->spa_sync_on
) {
950 txg_sync_stop(spa
->spa_dsl_pool
);
951 spa
->spa_sync_on
= B_FALSE
;
955 * Wait for any outstanding async I/O to complete.
957 if (spa
->spa_async_zio_root
!= NULL
) {
958 (void) zio_wait(spa
->spa_async_zio_root
);
959 spa
->spa_async_zio_root
= NULL
;
962 bpobj_close(&spa
->spa_deferred_bpobj
);
965 * Close the dsl pool.
967 if (spa
->spa_dsl_pool
) {
968 dsl_pool_close(spa
->spa_dsl_pool
);
969 spa
->spa_dsl_pool
= NULL
;
970 spa
->spa_meta_objset
= NULL
;
975 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
978 * Drop and purge level 2 cache
980 spa_l2cache_drop(spa
);
985 if (spa
->spa_root_vdev
)
986 vdev_free(spa
->spa_root_vdev
);
987 ASSERT(spa
->spa_root_vdev
== NULL
);
989 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
990 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
991 if (spa
->spa_spares
.sav_vdevs
) {
992 kmem_free(spa
->spa_spares
.sav_vdevs
,
993 spa
->spa_spares
.sav_count
* sizeof (void *));
994 spa
->spa_spares
.sav_vdevs
= NULL
;
996 if (spa
->spa_spares
.sav_config
) {
997 nvlist_free(spa
->spa_spares
.sav_config
);
998 spa
->spa_spares
.sav_config
= NULL
;
1000 spa
->spa_spares
.sav_count
= 0;
1002 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
1003 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1004 if (spa
->spa_l2cache
.sav_vdevs
) {
1005 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1006 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1007 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1009 if (spa
->spa_l2cache
.sav_config
) {
1010 nvlist_free(spa
->spa_l2cache
.sav_config
);
1011 spa
->spa_l2cache
.sav_config
= NULL
;
1013 spa
->spa_l2cache
.sav_count
= 0;
1015 spa
->spa_async_suspended
= 0;
1017 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1021 * Load (or re-load) the current list of vdevs describing the active spares for
1022 * this pool. When this is called, we have some form of basic information in
1023 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1024 * then re-generate a more complete list including status information.
1027 spa_load_spares(spa_t
*spa
)
1034 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1037 * First, close and free any existing spare vdevs.
1039 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1040 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1042 /* Undo the call to spa_activate() below */
1043 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1044 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1045 spa_spare_remove(tvd
);
1050 if (spa
->spa_spares
.sav_vdevs
)
1051 kmem_free(spa
->spa_spares
.sav_vdevs
,
1052 spa
->spa_spares
.sav_count
* sizeof (void *));
1054 if (spa
->spa_spares
.sav_config
== NULL
)
1057 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1058 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1060 spa
->spa_spares
.sav_count
= (int)nspares
;
1061 spa
->spa_spares
.sav_vdevs
= NULL
;
1067 * Construct the array of vdevs, opening them to get status in the
1068 * process. For each spare, there is potentially two different vdev_t
1069 * structures associated with it: one in the list of spares (used only
1070 * for basic validation purposes) and one in the active vdev
1071 * configuration (if it's spared in). During this phase we open and
1072 * validate each vdev on the spare list. If the vdev also exists in the
1073 * active configuration, then we also mark this vdev as an active spare.
1075 spa
->spa_spares
.sav_vdevs
= kmem_alloc(nspares
* sizeof (void *),
1077 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1078 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1079 VDEV_ALLOC_SPARE
) == 0);
1082 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1084 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1085 B_FALSE
)) != NULL
) {
1086 if (!tvd
->vdev_isspare
)
1090 * We only mark the spare active if we were successfully
1091 * able to load the vdev. Otherwise, importing a pool
1092 * with a bad active spare would result in strange
1093 * behavior, because multiple pool would think the spare
1094 * is actively in use.
1096 * There is a vulnerability here to an equally bizarre
1097 * circumstance, where a dead active spare is later
1098 * brought back to life (onlined or otherwise). Given
1099 * the rarity of this scenario, and the extra complexity
1100 * it adds, we ignore the possibility.
1102 if (!vdev_is_dead(tvd
))
1103 spa_spare_activate(tvd
);
1107 vd
->vdev_aux
= &spa
->spa_spares
;
1109 if (vdev_open(vd
) != 0)
1112 if (vdev_validate_aux(vd
) == 0)
1117 * Recompute the stashed list of spares, with status information
1120 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1121 DATA_TYPE_NVLIST_ARRAY
) == 0);
1123 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1125 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1126 spares
[i
] = vdev_config_generate(spa
,
1127 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1128 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1129 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1130 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1131 nvlist_free(spares
[i
]);
1132 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1136 * Load (or re-load) the current list of vdevs describing the active l2cache for
1137 * this pool. When this is called, we have some form of basic information in
1138 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1139 * then re-generate a more complete list including status information.
1140 * Devices which are already active have their details maintained, and are
1144 spa_load_l2cache(spa_t
*spa
)
1148 int i
, j
, oldnvdevs
;
1150 vdev_t
*vd
, **oldvdevs
, **newvdevs
= NULL
;
1151 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1153 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1155 if (sav
->sav_config
!= NULL
) {
1156 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1157 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1158 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1163 oldvdevs
= sav
->sav_vdevs
;
1164 oldnvdevs
= sav
->sav_count
;
1165 sav
->sav_vdevs
= NULL
;
1169 * Process new nvlist of vdevs.
1171 for (i
= 0; i
< nl2cache
; i
++) {
1172 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1176 for (j
= 0; j
< oldnvdevs
; j
++) {
1178 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1180 * Retain previous vdev for add/remove ops.
1188 if (newvdevs
[i
] == NULL
) {
1192 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1193 VDEV_ALLOC_L2CACHE
) == 0);
1198 * Commit this vdev as an l2cache device,
1199 * even if it fails to open.
1201 spa_l2cache_add(vd
);
1206 spa_l2cache_activate(vd
);
1208 if (vdev_open(vd
) != 0)
1211 (void) vdev_validate_aux(vd
);
1213 if (!vdev_is_dead(vd
))
1214 l2arc_add_vdev(spa
, vd
);
1219 * Purge vdevs that were dropped
1221 for (i
= 0; i
< oldnvdevs
; i
++) {
1226 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1227 pool
!= 0ULL && l2arc_vdev_present(vd
))
1228 l2arc_remove_vdev(vd
);
1229 (void) vdev_close(vd
);
1230 spa_l2cache_remove(vd
);
1235 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1237 if (sav
->sav_config
== NULL
)
1240 sav
->sav_vdevs
= newvdevs
;
1241 sav
->sav_count
= (int)nl2cache
;
1244 * Recompute the stashed list of l2cache devices, with status
1245 * information this time.
1247 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1248 DATA_TYPE_NVLIST_ARRAY
) == 0);
1250 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1251 for (i
= 0; i
< sav
->sav_count
; i
++)
1252 l2cache
[i
] = vdev_config_generate(spa
,
1253 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1254 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1255 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1257 for (i
= 0; i
< sav
->sav_count
; i
++)
1258 nvlist_free(l2cache
[i
]);
1260 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1264 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1267 char *packed
= NULL
;
1272 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
1273 nvsize
= *(uint64_t *)db
->db_data
;
1274 dmu_buf_rele(db
, FTAG
);
1276 packed
= kmem_alloc(nvsize
, KM_SLEEP
| KM_NODEBUG
);
1277 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1280 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1281 kmem_free(packed
, nvsize
);
1287 * Checks to see if the given vdev could not be opened, in which case we post a
1288 * sysevent to notify the autoreplace code that the device has been removed.
1291 spa_check_removed(vdev_t
*vd
)
1295 for (c
= 0; c
< vd
->vdev_children
; c
++)
1296 spa_check_removed(vd
->vdev_child
[c
]);
1298 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
)) {
1299 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE
,
1300 vd
->vdev_spa
, vd
, NULL
, 0, 0);
1301 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_CHECK
);
1306 * Validate the current config against the MOS config
1309 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1311 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1315 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1317 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1318 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1320 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1323 * If we're doing a normal import, then build up any additional
1324 * diagnostic information about missing devices in this config.
1325 * We'll pass this up to the user for further processing.
1327 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1328 nvlist_t
**child
, *nv
;
1331 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1333 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1335 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1336 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1337 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1339 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1340 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1342 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1347 VERIFY(nvlist_add_nvlist_array(nv
,
1348 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1349 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1350 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1352 for (i
= 0; i
< idx
; i
++)
1353 nvlist_free(child
[i
]);
1356 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1360 * Compare the root vdev tree with the information we have
1361 * from the MOS config (mrvd). Check each top-level vdev
1362 * with the corresponding MOS config top-level (mtvd).
1364 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1365 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1366 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1369 * Resolve any "missing" vdevs in the current configuration.
1370 * If we find that the MOS config has more accurate information
1371 * about the top-level vdev then use that vdev instead.
1373 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1374 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1376 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1380 * Device specific actions.
1382 if (mtvd
->vdev_islog
) {
1383 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1386 * XXX - once we have 'readonly' pool
1387 * support we should be able to handle
1388 * missing data devices by transitioning
1389 * the pool to readonly.
1395 * Swap the missing vdev with the data we were
1396 * able to obtain from the MOS config.
1398 vdev_remove_child(rvd
, tvd
);
1399 vdev_remove_child(mrvd
, mtvd
);
1401 vdev_add_child(rvd
, mtvd
);
1402 vdev_add_child(mrvd
, tvd
);
1404 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1406 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1409 } else if (mtvd
->vdev_islog
) {
1411 * Load the slog device's state from the MOS config
1412 * since it's possible that the label does not
1413 * contain the most up-to-date information.
1415 vdev_load_log_state(tvd
, mtvd
);
1420 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1423 * Ensure we were able to validate the config.
1425 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1429 * Check for missing log devices
1432 spa_check_logs(spa_t
*spa
)
1434 switch (spa
->spa_log_state
) {
1437 case SPA_LOG_MISSING
:
1438 /* need to recheck in case slog has been restored */
1439 case SPA_LOG_UNKNOWN
:
1440 if (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
, NULL
,
1441 DS_FIND_CHILDREN
)) {
1442 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1451 spa_passivate_log(spa_t
*spa
)
1453 vdev_t
*rvd
= spa
->spa_root_vdev
;
1454 boolean_t slog_found
= B_FALSE
;
1457 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1459 if (!spa_has_slogs(spa
))
1462 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1463 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1464 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1466 if (tvd
->vdev_islog
) {
1467 metaslab_group_passivate(mg
);
1468 slog_found
= B_TRUE
;
1472 return (slog_found
);
1476 spa_activate_log(spa_t
*spa
)
1478 vdev_t
*rvd
= spa
->spa_root_vdev
;
1481 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1483 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1484 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1485 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1487 if (tvd
->vdev_islog
)
1488 metaslab_group_activate(mg
);
1493 spa_offline_log(spa_t
*spa
)
1497 if ((error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1498 NULL
, DS_FIND_CHILDREN
)) == 0) {
1501 * We successfully offlined the log device, sync out the
1502 * current txg so that the "stubby" block can be removed
1505 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1511 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1515 for (i
= 0; i
< sav
->sav_count
; i
++)
1516 spa_check_removed(sav
->sav_vdevs
[i
]);
1520 spa_claim_notify(zio_t
*zio
)
1522 spa_t
*spa
= zio
->io_spa
;
1527 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1528 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1529 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1530 mutex_exit(&spa
->spa_props_lock
);
1533 typedef struct spa_load_error
{
1534 uint64_t sle_meta_count
;
1535 uint64_t sle_data_count
;
1539 spa_load_verify_done(zio_t
*zio
)
1541 blkptr_t
*bp
= zio
->io_bp
;
1542 spa_load_error_t
*sle
= zio
->io_private
;
1543 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1544 int error
= zio
->io_error
;
1547 if ((BP_GET_LEVEL(bp
) != 0 || dmu_ot
[type
].ot_metadata
) &&
1548 type
!= DMU_OT_INTENT_LOG
)
1549 atomic_add_64(&sle
->sle_meta_count
, 1);
1551 atomic_add_64(&sle
->sle_data_count
, 1);
1553 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1558 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1559 arc_buf_t
*pbuf
, const zbookmark_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1563 size_t size
= BP_GET_PSIZE(bp
);
1564 void *data
= zio_data_buf_alloc(size
);
1566 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1567 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1568 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1569 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1575 spa_load_verify(spa_t
*spa
)
1578 spa_load_error_t sle
= { 0 };
1579 zpool_rewind_policy_t policy
;
1580 boolean_t verify_ok
= B_FALSE
;
1583 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1585 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1588 rio
= zio_root(spa
, NULL
, &sle
,
1589 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1591 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1592 TRAVERSE_PRE
| TRAVERSE_PREFETCH
, spa_load_verify_cb
, rio
);
1594 (void) zio_wait(rio
);
1596 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1597 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1599 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1600 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1604 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1605 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1607 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1608 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1609 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1610 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1611 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1612 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1613 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1615 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1619 if (error
!= ENXIO
&& error
!= EIO
)
1624 return (verify_ok
? 0 : EIO
);
1628 * Find a value in the pool props object.
1631 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
1633 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
1634 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
1638 * Find a value in the pool directory object.
1641 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
1643 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
1644 name
, sizeof (uint64_t), 1, val
));
1648 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
1650 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
1655 * Fix up config after a partly-completed split. This is done with the
1656 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1657 * pool have that entry in their config, but only the splitting one contains
1658 * a list of all the guids of the vdevs that are being split off.
1660 * This function determines what to do with that list: either rejoin
1661 * all the disks to the pool, or complete the splitting process. To attempt
1662 * the rejoin, each disk that is offlined is marked online again, and
1663 * we do a reopen() call. If the vdev label for every disk that was
1664 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1665 * then we call vdev_split() on each disk, and complete the split.
1667 * Otherwise we leave the config alone, with all the vdevs in place in
1668 * the original pool.
1671 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
1678 boolean_t attempt_reopen
;
1680 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
1683 /* check that the config is complete */
1684 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
1685 &glist
, &gcount
) != 0)
1688 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
1690 /* attempt to online all the vdevs & validate */
1691 attempt_reopen
= B_TRUE
;
1692 for (i
= 0; i
< gcount
; i
++) {
1693 if (glist
[i
] == 0) /* vdev is hole */
1696 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
1697 if (vd
[i
] == NULL
) {
1699 * Don't bother attempting to reopen the disks;
1700 * just do the split.
1702 attempt_reopen
= B_FALSE
;
1704 /* attempt to re-online it */
1705 vd
[i
]->vdev_offline
= B_FALSE
;
1709 if (attempt_reopen
) {
1710 vdev_reopen(spa
->spa_root_vdev
);
1712 /* check each device to see what state it's in */
1713 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
1714 if (vd
[i
] != NULL
&&
1715 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
1722 * If every disk has been moved to the new pool, or if we never
1723 * even attempted to look at them, then we split them off for
1726 if (!attempt_reopen
|| gcount
== extracted
) {
1727 for (i
= 0; i
< gcount
; i
++)
1730 vdev_reopen(spa
->spa_root_vdev
);
1733 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
1737 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
1738 boolean_t mosconfig
)
1740 nvlist_t
*config
= spa
->spa_config
;
1741 char *ereport
= FM_EREPORT_ZFS_POOL
;
1746 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
1750 * Versioning wasn't explicitly added to the label until later, so if
1751 * it's not present treat it as the initial version.
1753 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
1754 &spa
->spa_ubsync
.ub_version
) != 0)
1755 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
1757 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
1758 &spa
->spa_config_txg
);
1760 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
1761 spa_guid_exists(pool_guid
, 0)) {
1764 spa
->spa_load_guid
= pool_guid
;
1766 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
1768 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
1772 gethrestime(&spa
->spa_loaded_ts
);
1773 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
1774 mosconfig
, &ereport
);
1777 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
1779 if (error
!= EEXIST
) {
1780 spa
->spa_loaded_ts
.tv_sec
= 0;
1781 spa
->spa_loaded_ts
.tv_nsec
= 0;
1783 if (error
!= EBADF
) {
1784 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
1787 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
1794 * Load an existing storage pool, using the pool's builtin spa_config as a
1795 * source of configuration information.
1797 __attribute__((always_inline
))
1799 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
1800 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
1804 nvlist_t
*nvroot
= NULL
;
1806 uberblock_t
*ub
= &spa
->spa_uberblock
;
1807 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
1808 int orig_mode
= spa
->spa_mode
;
1813 * If this is an untrusted config, access the pool in read-only mode.
1814 * This prevents things like resilvering recently removed devices.
1817 spa
->spa_mode
= FREAD
;
1819 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1821 spa
->spa_load_state
= state
;
1823 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
1826 parse
= (type
== SPA_IMPORT_EXISTING
?
1827 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
1830 * Create "The Godfather" zio to hold all async IOs
1832 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
1833 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
1836 * Parse the configuration into a vdev tree. We explicitly set the
1837 * value that will be returned by spa_version() since parsing the
1838 * configuration requires knowing the version number.
1840 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1841 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
1842 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1847 ASSERT(spa
->spa_root_vdev
== rvd
);
1849 if (type
!= SPA_IMPORT_ASSEMBLE
) {
1850 ASSERT(spa_guid(spa
) == pool_guid
);
1854 * Try to open all vdevs, loading each label in the process.
1856 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1857 error
= vdev_open(rvd
);
1858 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1863 * We need to validate the vdev labels against the configuration that
1864 * we have in hand, which is dependent on the setting of mosconfig. If
1865 * mosconfig is true then we're validating the vdev labels based on
1866 * that config. Otherwise, we're validating against the cached config
1867 * (zpool.cache) that was read when we loaded the zfs module, and then
1868 * later we will recursively call spa_load() and validate against
1871 * If we're assembling a new pool that's been split off from an
1872 * existing pool, the labels haven't yet been updated so we skip
1873 * validation for now.
1875 if (type
!= SPA_IMPORT_ASSEMBLE
) {
1876 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1877 error
= vdev_validate(rvd
);
1878 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1883 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
1888 * Find the best uberblock.
1890 vdev_uberblock_load(NULL
, rvd
, ub
);
1893 * If we weren't able to find a single valid uberblock, return failure.
1895 if (ub
->ub_txg
== 0)
1896 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
1899 * If the pool is newer than the code, we can't open it.
1901 if (ub
->ub_version
> SPA_VERSION
)
1902 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
1905 * If the vdev guid sum doesn't match the uberblock, we have an
1906 * incomplete configuration. We first check to see if the pool
1907 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1908 * If it is, defer the vdev_guid_sum check till later so we
1909 * can handle missing vdevs.
1911 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
1912 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
1913 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
1914 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
1916 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
1917 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1918 spa_try_repair(spa
, config
);
1919 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1920 nvlist_free(spa
->spa_config_splitting
);
1921 spa
->spa_config_splitting
= NULL
;
1925 * Initialize internal SPA structures.
1927 spa
->spa_state
= POOL_STATE_ACTIVE
;
1928 spa
->spa_ubsync
= spa
->spa_uberblock
;
1929 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
1930 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
1931 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
1932 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
1933 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
1934 spa
->spa_prev_software_version
= ub
->ub_software_version
;
1936 error
= dsl_pool_open(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
1938 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
1939 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
1941 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
1942 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
1946 nvlist_t
*policy
= NULL
, *nvconfig
;
1948 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
1949 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
1951 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
1952 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
1954 unsigned long myhostid
= 0;
1956 VERIFY(nvlist_lookup_string(nvconfig
,
1957 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
1960 myhostid
= zone_get_hostid(NULL
);
1963 * We're emulating the system's hostid in userland, so
1964 * we can't use zone_get_hostid().
1966 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
1967 #endif /* _KERNEL */
1968 if (hostid
!= 0 && myhostid
!= 0 &&
1969 hostid
!= myhostid
) {
1970 nvlist_free(nvconfig
);
1971 cmn_err(CE_WARN
, "pool '%s' could not be "
1972 "loaded as it was last accessed by "
1973 "another system (host: %s hostid: 0x%lx). "
1974 "See: http://www.sun.com/msg/ZFS-8000-EY",
1975 spa_name(spa
), hostname
,
1976 (unsigned long)hostid
);
1980 if (nvlist_lookup_nvlist(spa
->spa_config
,
1981 ZPOOL_REWIND_POLICY
, &policy
) == 0)
1982 VERIFY(nvlist_add_nvlist(nvconfig
,
1983 ZPOOL_REWIND_POLICY
, policy
) == 0);
1985 spa_config_set(spa
, nvconfig
);
1987 spa_deactivate(spa
);
1988 spa_activate(spa
, orig_mode
);
1990 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
1993 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
1994 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
1995 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
1997 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2000 * Load the bit that tells us to use the new accounting function
2001 * (raid-z deflation). If we have an older pool, this will not
2004 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2005 if (error
!= 0 && error
!= ENOENT
)
2006 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2008 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2009 &spa
->spa_creation_version
);
2010 if (error
!= 0 && error
!= ENOENT
)
2011 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2014 * Load the persistent error log. If we have an older pool, this will
2017 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2018 if (error
!= 0 && error
!= ENOENT
)
2019 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2021 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2022 &spa
->spa_errlog_scrub
);
2023 if (error
!= 0 && error
!= ENOENT
)
2024 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2027 * Load the history object. If we have an older pool, this
2028 * will not be present.
2030 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2031 if (error
!= 0 && error
!= ENOENT
)
2032 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2035 * If we're assembling the pool from the split-off vdevs of
2036 * an existing pool, we don't want to attach the spares & cache
2041 * Load any hot spares for this pool.
2043 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2044 if (error
!= 0 && error
!= ENOENT
)
2045 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2046 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2047 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2048 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2049 &spa
->spa_spares
.sav_config
) != 0)
2050 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2052 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2053 spa_load_spares(spa
);
2054 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2055 } else if (error
== 0) {
2056 spa
->spa_spares
.sav_sync
= B_TRUE
;
2060 * Load any level 2 ARC devices for this pool.
2062 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2063 &spa
->spa_l2cache
.sav_object
);
2064 if (error
!= 0 && error
!= ENOENT
)
2065 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2066 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2067 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2068 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2069 &spa
->spa_l2cache
.sav_config
) != 0)
2070 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2072 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2073 spa_load_l2cache(spa
);
2074 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2075 } else if (error
== 0) {
2076 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2079 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2081 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2082 if (error
&& error
!= ENOENT
)
2083 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2086 uint64_t autoreplace
;
2088 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2089 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2090 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2091 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2092 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2093 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2094 &spa
->spa_dedup_ditto
);
2096 spa
->spa_autoreplace
= (autoreplace
!= 0);
2100 * If the 'autoreplace' property is set, then post a resource notifying
2101 * the ZFS DE that it should not issue any faults for unopenable
2102 * devices. We also iterate over the vdevs, and post a sysevent for any
2103 * unopenable vdevs so that the normal autoreplace handler can take
2106 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2107 spa_check_removed(spa
->spa_root_vdev
);
2109 * For the import case, this is done in spa_import(), because
2110 * at this point we're using the spare definitions from
2111 * the MOS config, not necessarily from the userland config.
2113 if (state
!= SPA_LOAD_IMPORT
) {
2114 spa_aux_check_removed(&spa
->spa_spares
);
2115 spa_aux_check_removed(&spa
->spa_l2cache
);
2120 * Load the vdev state for all toplevel vdevs.
2125 * Propagate the leaf DTLs we just loaded all the way up the tree.
2127 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2128 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2129 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2132 * Load the DDTs (dedup tables).
2134 error
= ddt_load(spa
);
2136 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2138 spa_update_dspace(spa
);
2141 * Validate the config, using the MOS config to fill in any
2142 * information which might be missing. If we fail to validate
2143 * the config then declare the pool unfit for use. If we're
2144 * assembling a pool from a split, the log is not transferred
2147 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2150 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2151 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2153 if (!spa_config_valid(spa
, nvconfig
)) {
2154 nvlist_free(nvconfig
);
2155 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2158 nvlist_free(nvconfig
);
2161 * Now that we've validate the config, check the state of the
2162 * root vdev. If it can't be opened, it indicates one or
2163 * more toplevel vdevs are faulted.
2165 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2168 if (spa_check_logs(spa
)) {
2169 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2170 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2175 * We've successfully opened the pool, verify that we're ready
2176 * to start pushing transactions.
2178 if (state
!= SPA_LOAD_TRYIMPORT
) {
2179 if ((error
= spa_load_verify(spa
)))
2180 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2184 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2185 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2187 int need_update
= B_FALSE
;
2190 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2193 * Claim log blocks that haven't been committed yet.
2194 * This must all happen in a single txg.
2195 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2196 * invoked from zil_claim_log_block()'s i/o done callback.
2197 * Price of rollback is that we abandon the log.
2199 spa
->spa_claiming
= B_TRUE
;
2201 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
2202 spa_first_txg(spa
));
2203 (void) dmu_objset_find(spa_name(spa
),
2204 zil_claim
, tx
, DS_FIND_CHILDREN
);
2207 spa
->spa_claiming
= B_FALSE
;
2209 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2210 spa
->spa_sync_on
= B_TRUE
;
2211 txg_sync_start(spa
->spa_dsl_pool
);
2214 * Wait for all claims to sync. We sync up to the highest
2215 * claimed log block birth time so that claimed log blocks
2216 * don't appear to be from the future. spa_claim_max_txg
2217 * will have been set for us by either zil_check_log_chain()
2218 * (invoked from spa_check_logs()) or zil_claim() above.
2220 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2223 * If the config cache is stale, or we have uninitialized
2224 * metaslabs (see spa_vdev_add()), then update the config.
2226 * If this is a verbatim import, trust the current
2227 * in-core spa_config and update the disk labels.
2229 if (config_cache_txg
!= spa
->spa_config_txg
||
2230 state
== SPA_LOAD_IMPORT
||
2231 state
== SPA_LOAD_RECOVER
||
2232 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2233 need_update
= B_TRUE
;
2235 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2236 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2237 need_update
= B_TRUE
;
2240 * Update the config cache asychronously in case we're the
2241 * root pool, in which case the config cache isn't writable yet.
2244 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2247 * Check all DTLs to see if anything needs resilvering.
2249 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2250 vdev_resilver_needed(rvd
, NULL
, NULL
))
2251 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2254 * Delete any inconsistent datasets.
2256 (void) dmu_objset_find(spa_name(spa
),
2257 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2260 * Clean up any stale temporary dataset userrefs.
2262 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2269 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2271 int mode
= spa
->spa_mode
;
2274 spa_deactivate(spa
);
2276 spa
->spa_load_max_txg
--;
2278 spa_activate(spa
, mode
);
2279 spa_async_suspend(spa
);
2281 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2285 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2286 uint64_t max_request
, int rewind_flags
)
2288 nvlist_t
*config
= NULL
;
2289 int load_error
, rewind_error
;
2290 uint64_t safe_rewind_txg
;
2293 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2294 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2295 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2297 spa
->spa_load_max_txg
= max_request
;
2300 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2302 if (load_error
== 0)
2305 if (spa
->spa_root_vdev
!= NULL
)
2306 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2308 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2309 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2311 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2312 nvlist_free(config
);
2313 return (load_error
);
2316 /* Price of rolling back is discarding txgs, including log */
2317 if (state
== SPA_LOAD_RECOVER
)
2318 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2320 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2321 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2322 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2323 TXG_INITIAL
: safe_rewind_txg
;
2326 * Continue as long as we're finding errors, we're still within
2327 * the acceptable rewind range, and we're still finding uberblocks
2329 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2330 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2331 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2332 spa
->spa_extreme_rewind
= B_TRUE
;
2333 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2336 spa
->spa_extreme_rewind
= B_FALSE
;
2337 spa
->spa_load_max_txg
= UINT64_MAX
;
2339 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2340 spa_config_set(spa
, config
);
2342 return (state
== SPA_LOAD_RECOVER
? rewind_error
: load_error
);
2348 * The import case is identical to an open except that the configuration is sent
2349 * down from userland, instead of grabbed from the configuration cache. For the
2350 * case of an open, the pool configuration will exist in the
2351 * POOL_STATE_UNINITIALIZED state.
2353 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2354 * the same time open the pool, without having to keep around the spa_t in some
2358 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2362 spa_load_state_t state
= SPA_LOAD_OPEN
;
2364 int locked
= B_FALSE
;
2369 * As disgusting as this is, we need to support recursive calls to this
2370 * function because dsl_dir_open() is called during spa_load(), and ends
2371 * up calling spa_open() again. The real fix is to figure out how to
2372 * avoid dsl_dir_open() calling this in the first place.
2374 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2375 mutex_enter(&spa_namespace_lock
);
2379 if ((spa
= spa_lookup(pool
)) == NULL
) {
2381 mutex_exit(&spa_namespace_lock
);
2385 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
2386 zpool_rewind_policy_t policy
;
2388 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
2390 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
2391 state
= SPA_LOAD_RECOVER
;
2393 spa_activate(spa
, spa_mode_global
);
2395 if (state
!= SPA_LOAD_RECOVER
)
2396 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
2398 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
2399 policy
.zrp_request
);
2401 if (error
== EBADF
) {
2403 * If vdev_validate() returns failure (indicated by
2404 * EBADF), it indicates that one of the vdevs indicates
2405 * that the pool has been exported or destroyed. If
2406 * this is the case, the config cache is out of sync and
2407 * we should remove the pool from the namespace.
2410 spa_deactivate(spa
);
2411 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
2414 mutex_exit(&spa_namespace_lock
);
2420 * We can't open the pool, but we still have useful
2421 * information: the state of each vdev after the
2422 * attempted vdev_open(). Return this to the user.
2424 if (config
!= NULL
&& spa
->spa_config
) {
2425 VERIFY(nvlist_dup(spa
->spa_config
, config
,
2427 VERIFY(nvlist_add_nvlist(*config
,
2428 ZPOOL_CONFIG_LOAD_INFO
,
2429 spa
->spa_load_info
) == 0);
2432 spa_deactivate(spa
);
2433 spa
->spa_last_open_failed
= error
;
2435 mutex_exit(&spa_namespace_lock
);
2441 spa_open_ref(spa
, tag
);
2444 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2447 * If we've recovered the pool, pass back any information we
2448 * gathered while doing the load.
2450 if (state
== SPA_LOAD_RECOVER
) {
2451 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
2452 spa
->spa_load_info
) == 0);
2456 spa
->spa_last_open_failed
= 0;
2457 spa
->spa_last_ubsync_txg
= 0;
2458 spa
->spa_load_txg
= 0;
2459 mutex_exit(&spa_namespace_lock
);
2468 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
2471 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
2475 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
2477 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
2481 * Lookup the given spa_t, incrementing the inject count in the process,
2482 * preventing it from being exported or destroyed.
2485 spa_inject_addref(char *name
)
2489 mutex_enter(&spa_namespace_lock
);
2490 if ((spa
= spa_lookup(name
)) == NULL
) {
2491 mutex_exit(&spa_namespace_lock
);
2494 spa
->spa_inject_ref
++;
2495 mutex_exit(&spa_namespace_lock
);
2501 spa_inject_delref(spa_t
*spa
)
2503 mutex_enter(&spa_namespace_lock
);
2504 spa
->spa_inject_ref
--;
2505 mutex_exit(&spa_namespace_lock
);
2509 * Add spares device information to the nvlist.
2512 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
2522 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2524 if (spa
->spa_spares
.sav_count
== 0)
2527 VERIFY(nvlist_lookup_nvlist(config
,
2528 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
2529 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
2530 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
2532 VERIFY(nvlist_add_nvlist_array(nvroot
,
2533 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
2534 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
2535 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
2538 * Go through and find any spares which have since been
2539 * repurposed as an active spare. If this is the case, update
2540 * their status appropriately.
2542 for (i
= 0; i
< nspares
; i
++) {
2543 VERIFY(nvlist_lookup_uint64(spares
[i
],
2544 ZPOOL_CONFIG_GUID
, &guid
) == 0);
2545 if (spa_spare_exists(guid
, &pool
, NULL
) &&
2547 VERIFY(nvlist_lookup_uint64_array(
2548 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
2549 (uint64_t **)&vs
, &vsc
) == 0);
2550 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
2551 vs
->vs_aux
= VDEV_AUX_SPARED
;
2558 * Add l2cache device information to the nvlist, including vdev stats.
2561 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
2564 uint_t i
, j
, nl2cache
;
2571 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
2573 if (spa
->spa_l2cache
.sav_count
== 0)
2576 VERIFY(nvlist_lookup_nvlist(config
,
2577 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
2578 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
2579 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
2580 if (nl2cache
!= 0) {
2581 VERIFY(nvlist_add_nvlist_array(nvroot
,
2582 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
2583 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
2584 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
2587 * Update level 2 cache device stats.
2590 for (i
= 0; i
< nl2cache
; i
++) {
2591 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
2592 ZPOOL_CONFIG_GUID
, &guid
) == 0);
2595 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
2597 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
2598 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
2604 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
2605 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
2607 vdev_get_stats(vd
, vs
);
2613 spa_get_stats(const char *name
, nvlist_t
**config
, char *altroot
, size_t buflen
)
2619 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
2623 * This still leaves a window of inconsistency where the spares
2624 * or l2cache devices could change and the config would be
2625 * self-inconsistent.
2627 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
2629 if (*config
!= NULL
) {
2630 uint64_t loadtimes
[2];
2632 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
2633 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
2634 VERIFY(nvlist_add_uint64_array(*config
,
2635 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
2637 VERIFY(nvlist_add_uint64(*config
,
2638 ZPOOL_CONFIG_ERRCOUNT
,
2639 spa_get_errlog_size(spa
)) == 0);
2641 if (spa_suspended(spa
))
2642 VERIFY(nvlist_add_uint64(*config
,
2643 ZPOOL_CONFIG_SUSPENDED
,
2644 spa
->spa_failmode
) == 0);
2646 spa_add_spares(spa
, *config
);
2647 spa_add_l2cache(spa
, *config
);
2652 * We want to get the alternate root even for faulted pools, so we cheat
2653 * and call spa_lookup() directly.
2657 mutex_enter(&spa_namespace_lock
);
2658 spa
= spa_lookup(name
);
2660 spa_altroot(spa
, altroot
, buflen
);
2664 mutex_exit(&spa_namespace_lock
);
2666 spa_altroot(spa
, altroot
, buflen
);
2671 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
2672 spa_close(spa
, FTAG
);
2679 * Validate that the auxiliary device array is well formed. We must have an
2680 * array of nvlists, each which describes a valid leaf vdev. If this is an
2681 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2682 * specified, as long as they are well-formed.
2685 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
2686 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
2687 vdev_labeltype_t label
)
2694 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
2697 * It's acceptable to have no devs specified.
2699 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
2706 * Make sure the pool is formatted with a version that supports this
2709 if (spa_version(spa
) < version
)
2713 * Set the pending device list so we correctly handle device in-use
2716 sav
->sav_pending
= dev
;
2717 sav
->sav_npending
= ndev
;
2719 for (i
= 0; i
< ndev
; i
++) {
2720 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
2724 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2731 * The L2ARC currently only supports disk devices in
2732 * kernel context. For user-level testing, we allow it.
2735 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
2736 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
2743 if ((error
= vdev_open(vd
)) == 0 &&
2744 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
2745 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
2746 vd
->vdev_guid
) == 0);
2752 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
2759 sav
->sav_pending
= NULL
;
2760 sav
->sav_npending
= 0;
2765 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
2769 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
2771 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
2772 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
2773 VDEV_LABEL_SPARE
)) != 0) {
2777 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
2778 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
2779 VDEV_LABEL_L2CACHE
));
2783 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
2788 if (sav
->sav_config
!= NULL
) {
2794 * Generate new dev list by concatentating with the
2797 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
2798 &olddevs
, &oldndevs
) == 0);
2800 newdevs
= kmem_alloc(sizeof (void *) *
2801 (ndevs
+ oldndevs
), KM_SLEEP
);
2802 for (i
= 0; i
< oldndevs
; i
++)
2803 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
2805 for (i
= 0; i
< ndevs
; i
++)
2806 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
2809 VERIFY(nvlist_remove(sav
->sav_config
, config
,
2810 DATA_TYPE_NVLIST_ARRAY
) == 0);
2812 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
2813 config
, newdevs
, ndevs
+ oldndevs
) == 0);
2814 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
2815 nvlist_free(newdevs
[i
]);
2816 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
2819 * Generate a new dev list.
2821 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
2823 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
2829 * Stop and drop level 2 ARC devices
2832 spa_l2cache_drop(spa_t
*spa
)
2836 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
2838 for (i
= 0; i
< sav
->sav_count
; i
++) {
2841 vd
= sav
->sav_vdevs
[i
];
2844 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
2845 pool
!= 0ULL && l2arc_vdev_present(vd
))
2846 l2arc_remove_vdev(vd
);
2847 if (vd
->vdev_isl2cache
)
2848 spa_l2cache_remove(vd
);
2849 vdev_clear_stats(vd
);
2850 (void) vdev_close(vd
);
2858 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
2859 const char *history_str
, nvlist_t
*zplprops
)
2862 char *altroot
= NULL
;
2867 uint64_t txg
= TXG_INITIAL
;
2868 nvlist_t
**spares
, **l2cache
;
2869 uint_t nspares
, nl2cache
;
2870 uint64_t version
, obj
;
2874 * If this pool already exists, return failure.
2876 mutex_enter(&spa_namespace_lock
);
2877 if (spa_lookup(pool
) != NULL
) {
2878 mutex_exit(&spa_namespace_lock
);
2883 * Allocate a new spa_t structure.
2885 (void) nvlist_lookup_string(props
,
2886 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
2887 spa
= spa_add(pool
, NULL
, altroot
);
2888 spa_activate(spa
, spa_mode_global
);
2890 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
2891 spa_deactivate(spa
);
2893 mutex_exit(&spa_namespace_lock
);
2897 if (nvlist_lookup_uint64(props
, zpool_prop_to_name(ZPOOL_PROP_VERSION
),
2899 version
= SPA_VERSION
;
2900 ASSERT(version
<= SPA_VERSION
);
2902 spa
->spa_first_txg
= txg
;
2903 spa
->spa_uberblock
.ub_txg
= txg
- 1;
2904 spa
->spa_uberblock
.ub_version
= version
;
2905 spa
->spa_ubsync
= spa
->spa_uberblock
;
2908 * Create "The Godfather" zio to hold all async IOs
2910 spa
->spa_async_zio_root
= zio_root(spa
, NULL
, NULL
,
2911 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_GODFATHER
);
2914 * Create the root vdev.
2916 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2918 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
2920 ASSERT(error
!= 0 || rvd
!= NULL
);
2921 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
2923 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
2927 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
2928 (error
= spa_validate_aux(spa
, nvroot
, txg
,
2929 VDEV_ALLOC_ADD
)) == 0) {
2930 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2931 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
2932 vdev_expand(rvd
->vdev_child
[c
], txg
);
2936 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2940 spa_deactivate(spa
);
2942 mutex_exit(&spa_namespace_lock
);
2947 * Get the list of spares, if specified.
2949 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
2950 &spares
, &nspares
) == 0) {
2951 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
2953 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
2954 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
2955 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2956 spa_load_spares(spa
);
2957 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2958 spa
->spa_spares
.sav_sync
= B_TRUE
;
2962 * Get the list of level 2 cache devices, if specified.
2964 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
2965 &l2cache
, &nl2cache
) == 0) {
2966 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
2967 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
2968 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
2969 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
2970 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2971 spa_load_l2cache(spa
);
2972 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2973 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2976 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
2977 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
2980 * Create DDTs (dedup tables).
2984 spa_update_dspace(spa
);
2986 tx
= dmu_tx_create_assigned(dp
, txg
);
2989 * Create the pool config object.
2991 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
2992 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
2993 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
2995 if (zap_add(spa
->spa_meta_objset
,
2996 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
2997 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
2998 cmn_err(CE_PANIC
, "failed to add pool config");
3001 if (zap_add(spa
->spa_meta_objset
,
3002 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3003 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3004 cmn_err(CE_PANIC
, "failed to add pool version");
3007 /* Newly created pools with the right version are always deflated. */
3008 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3009 spa
->spa_deflate
= TRUE
;
3010 if (zap_add(spa
->spa_meta_objset
,
3011 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3012 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3013 cmn_err(CE_PANIC
, "failed to add deflate");
3018 * Create the deferred-free bpobj. Turn off compression
3019 * because sync-to-convergence takes longer if the blocksize
3022 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3023 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3024 ZIO_COMPRESS_OFF
, tx
);
3025 if (zap_add(spa
->spa_meta_objset
,
3026 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3027 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3028 cmn_err(CE_PANIC
, "failed to add bpobj");
3030 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3031 spa
->spa_meta_objset
, obj
));
3034 * Create the pool's history object.
3036 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3037 spa_history_create_obj(spa
, tx
);
3040 * Set pool properties.
3042 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3043 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3044 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3045 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3047 if (props
!= NULL
) {
3048 spa_configfile_set(spa
, props
, B_FALSE
);
3049 spa_sync_props(spa
, props
, tx
);
3054 spa
->spa_sync_on
= B_TRUE
;
3055 txg_sync_start(spa
->spa_dsl_pool
);
3058 * We explicitly wait for the first transaction to complete so that our
3059 * bean counters are appropriately updated.
3061 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3063 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3065 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& history_str
!= NULL
)
3066 (void) spa_history_log(spa
, history_str
, LOG_CMD_POOL_CREATE
);
3067 spa_history_log_version(spa
, LOG_POOL_CREATE
);
3069 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3071 mutex_exit(&spa_namespace_lock
);
3078 * Get the root pool information from the root disk, then import the root pool
3079 * during the system boot up time.
3081 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3084 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3087 nvlist_t
*nvtop
, *nvroot
;
3090 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3094 * Add this top-level vdev to the child array.
3096 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3098 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3100 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3103 * Put this pool's top-level vdevs into a root vdev.
3105 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3106 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3107 VDEV_TYPE_ROOT
) == 0);
3108 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3109 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3110 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3114 * Replace the existing vdev_tree with the new root vdev in
3115 * this pool's configuration (remove the old, add the new).
3117 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3118 nvlist_free(nvroot
);
3123 * Walk the vdev tree and see if we can find a device with "better"
3124 * configuration. A configuration is "better" if the label on that
3125 * device has a more recent txg.
3128 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3132 for (c
= 0; c
< vd
->vdev_children
; c
++)
3133 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3135 if (vd
->vdev_ops
->vdev_op_leaf
) {
3139 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3143 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3147 * Do we have a better boot device?
3149 if (label_txg
> *txg
) {
3158 * Import a root pool.
3160 * For x86. devpath_list will consist of devid and/or physpath name of
3161 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3162 * The GRUB "findroot" command will return the vdev we should boot.
3164 * For Sparc, devpath_list consists the physpath name of the booting device
3165 * no matter the rootpool is a single device pool or a mirrored pool.
3167 * "/pci@1f,0/ide@d/disk@0,0:a"
3170 spa_import_rootpool(char *devpath
, char *devid
)
3173 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3174 nvlist_t
*config
, *nvtop
;
3180 * Read the label from the boot device and generate a configuration.
3182 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3183 #if defined(_OBP) && defined(_KERNEL)
3184 if (config
== NULL
) {
3185 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3187 get_iscsi_bootpath_phy(devpath
);
3188 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3192 if (config
== NULL
) {
3193 cmn_err(CE_NOTE
, "Can not read the pool label from '%s'",
3198 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3200 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3202 mutex_enter(&spa_namespace_lock
);
3203 if ((spa
= spa_lookup(pname
)) != NULL
) {
3205 * Remove the existing root pool from the namespace so that we
3206 * can replace it with the correct config we just read in.
3211 spa
= spa_add(pname
, config
, NULL
);
3212 spa
->spa_is_root
= B_TRUE
;
3213 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3216 * Build up a vdev tree based on the boot device's label config.
3218 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3220 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3221 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3222 VDEV_ALLOC_ROOTPOOL
);
3223 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3225 mutex_exit(&spa_namespace_lock
);
3226 nvlist_free(config
);
3227 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3233 * Get the boot vdev.
3235 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3236 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3237 (u_longlong_t
)guid
);
3243 * Determine if there is a better boot device.
3246 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3248 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3249 "try booting from '%s'", avd
->vdev_path
);
3255 * If the boot device is part of a spare vdev then ensure that
3256 * we're booting off the active spare.
3258 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3259 !bvd
->vdev_isspare
) {
3260 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
3261 "try booting from '%s'",
3263 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
3269 spa_history_log_version(spa
, LOG_POOL_IMPORT
);
3271 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3273 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3274 mutex_exit(&spa_namespace_lock
);
3276 nvlist_free(config
);
3283 * Import a non-root pool into the system.
3286 spa_import(const char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
3289 char *altroot
= NULL
;
3290 spa_load_state_t state
= SPA_LOAD_IMPORT
;
3291 zpool_rewind_policy_t policy
;
3292 uint64_t mode
= spa_mode_global
;
3293 uint64_t readonly
= B_FALSE
;
3296 nvlist_t
**spares
, **l2cache
;
3297 uint_t nspares
, nl2cache
;
3300 * If a pool with this name exists, return failure.
3302 mutex_enter(&spa_namespace_lock
);
3303 if (spa_lookup(pool
) != NULL
) {
3304 mutex_exit(&spa_namespace_lock
);
3309 * Create and initialize the spa structure.
3311 (void) nvlist_lookup_string(props
,
3312 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3313 (void) nvlist_lookup_uint64(props
,
3314 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
3317 spa
= spa_add(pool
, config
, altroot
);
3318 spa
->spa_import_flags
= flags
;
3321 * Verbatim import - Take a pool and insert it into the namespace
3322 * as if it had been loaded at boot.
3324 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
3326 spa_configfile_set(spa
, props
, B_FALSE
);
3328 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3330 mutex_exit(&spa_namespace_lock
);
3331 spa_history_log_version(spa
, LOG_POOL_IMPORT
);
3336 spa_activate(spa
, mode
);
3339 * Don't start async tasks until we know everything is healthy.
3341 spa_async_suspend(spa
);
3343 zpool_get_rewind_policy(config
, &policy
);
3344 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3345 state
= SPA_LOAD_RECOVER
;
3348 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3349 * because the user-supplied config is actually the one to trust when
3352 if (state
!= SPA_LOAD_RECOVER
)
3353 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3355 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
3356 policy
.zrp_request
);
3359 * Propagate anything learned while loading the pool and pass it
3360 * back to caller (i.e. rewind info, missing devices, etc).
3362 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
3363 spa
->spa_load_info
) == 0);
3365 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3367 * Toss any existing sparelist, as it doesn't have any validity
3368 * anymore, and conflicts with spa_has_spare().
3370 if (spa
->spa_spares
.sav_config
) {
3371 nvlist_free(spa
->spa_spares
.sav_config
);
3372 spa
->spa_spares
.sav_config
= NULL
;
3373 spa_load_spares(spa
);
3375 if (spa
->spa_l2cache
.sav_config
) {
3376 nvlist_free(spa
->spa_l2cache
.sav_config
);
3377 spa
->spa_l2cache
.sav_config
= NULL
;
3378 spa_load_l2cache(spa
);
3381 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3384 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3387 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
3388 VDEV_ALLOC_L2CACHE
);
3389 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3392 spa_configfile_set(spa
, props
, B_FALSE
);
3394 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
3395 (error
= spa_prop_set(spa
, props
)))) {
3397 spa_deactivate(spa
);
3399 mutex_exit(&spa_namespace_lock
);
3403 spa_async_resume(spa
);
3406 * Override any spares and level 2 cache devices as specified by
3407 * the user, as these may have correct device names/devids, etc.
3409 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3410 &spares
, &nspares
) == 0) {
3411 if (spa
->spa_spares
.sav_config
)
3412 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
3413 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3415 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
3416 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3417 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3418 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3419 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3420 spa_load_spares(spa
);
3421 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3422 spa
->spa_spares
.sav_sync
= B_TRUE
;
3424 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3425 &l2cache
, &nl2cache
) == 0) {
3426 if (spa
->spa_l2cache
.sav_config
)
3427 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
3428 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
3430 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3431 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3432 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3433 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3434 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3435 spa_load_l2cache(spa
);
3436 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3437 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3441 * Check for any removed devices.
3443 if (spa
->spa_autoreplace
) {
3444 spa_aux_check_removed(&spa
->spa_spares
);
3445 spa_aux_check_removed(&spa
->spa_l2cache
);
3448 if (spa_writeable(spa
)) {
3450 * Update the config cache to include the newly-imported pool.
3452 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
3456 * It's possible that the pool was expanded while it was exported.
3457 * We kick off an async task to handle this for us.
3459 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
3461 mutex_exit(&spa_namespace_lock
);
3462 spa_history_log_version(spa
, LOG_POOL_IMPORT
);
3468 spa_tryimport(nvlist_t
*tryconfig
)
3470 nvlist_t
*config
= NULL
;
3476 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
3479 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
3483 * Create and initialize the spa structure.
3485 mutex_enter(&spa_namespace_lock
);
3486 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
3487 spa_activate(spa
, FREAD
);
3490 * Pass off the heavy lifting to spa_load().
3491 * Pass TRUE for mosconfig because the user-supplied config
3492 * is actually the one to trust when doing an import.
3494 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
3497 * If 'tryconfig' was at least parsable, return the current config.
3499 if (spa
->spa_root_vdev
!= NULL
) {
3500 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3501 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3503 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
3505 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
3506 spa
->spa_uberblock
.ub_timestamp
) == 0);
3509 * If the bootfs property exists on this pool then we
3510 * copy it out so that external consumers can tell which
3511 * pools are bootable.
3513 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
3514 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
3517 * We have to play games with the name since the
3518 * pool was opened as TRYIMPORT_NAME.
3520 if (dsl_dsobj_to_dsname(spa_name(spa
),
3521 spa
->spa_bootfs
, tmpname
) == 0) {
3523 char *dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
3525 cp
= strchr(tmpname
, '/');
3527 (void) strlcpy(dsname
, tmpname
,
3530 (void) snprintf(dsname
, MAXPATHLEN
,
3531 "%s/%s", poolname
, ++cp
);
3533 VERIFY(nvlist_add_string(config
,
3534 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
3535 kmem_free(dsname
, MAXPATHLEN
);
3537 kmem_free(tmpname
, MAXPATHLEN
);
3541 * Add the list of hot spares and level 2 cache devices.
3543 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3544 spa_add_spares(spa
, config
);
3545 spa_add_l2cache(spa
, config
);
3546 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3550 spa_deactivate(spa
);
3552 mutex_exit(&spa_namespace_lock
);
3558 * Pool export/destroy
3560 * The act of destroying or exporting a pool is very simple. We make sure there
3561 * is no more pending I/O and any references to the pool are gone. Then, we
3562 * update the pool state and sync all the labels to disk, removing the
3563 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3564 * we don't sync the labels or remove the configuration cache.
3567 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
3568 boolean_t force
, boolean_t hardforce
)
3575 if (!(spa_mode_global
& FWRITE
))
3578 mutex_enter(&spa_namespace_lock
);
3579 if ((spa
= spa_lookup(pool
)) == NULL
) {
3580 mutex_exit(&spa_namespace_lock
);
3585 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3586 * reacquire the namespace lock, and see if we can export.
3588 spa_open_ref(spa
, FTAG
);
3589 mutex_exit(&spa_namespace_lock
);
3590 spa_async_suspend(spa
);
3591 mutex_enter(&spa_namespace_lock
);
3592 spa_close(spa
, FTAG
);
3595 * The pool will be in core if it's openable,
3596 * in which case we can modify its state.
3598 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
&& spa
->spa_sync_on
) {
3600 * Objsets may be open only because they're dirty, so we
3601 * have to force it to sync before checking spa_refcnt.
3603 txg_wait_synced(spa
->spa_dsl_pool
, 0);
3606 * A pool cannot be exported or destroyed if there are active
3607 * references. If we are resetting a pool, allow references by
3608 * fault injection handlers.
3610 if (!spa_refcount_zero(spa
) ||
3611 (spa
->spa_inject_ref
!= 0 &&
3612 new_state
!= POOL_STATE_UNINITIALIZED
)) {
3613 spa_async_resume(spa
);
3614 mutex_exit(&spa_namespace_lock
);
3619 * A pool cannot be exported if it has an active shared spare.
3620 * This is to prevent other pools stealing the active spare
3621 * from an exported pool. At user's own will, such pool can
3622 * be forcedly exported.
3624 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
3625 spa_has_active_shared_spare(spa
)) {
3626 spa_async_resume(spa
);
3627 mutex_exit(&spa_namespace_lock
);
3632 * We want this to be reflected on every label,
3633 * so mark them all dirty. spa_unload() will do the
3634 * final sync that pushes these changes out.
3636 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
3637 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3638 spa
->spa_state
= new_state
;
3639 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
3641 vdev_config_dirty(spa
->spa_root_vdev
);
3642 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3646 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_DESTROY
);
3648 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
3650 spa_deactivate(spa
);
3653 if (oldconfig
&& spa
->spa_config
)
3654 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
3656 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
3658 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3661 mutex_exit(&spa_namespace_lock
);
3667 * Destroy a storage pool.
3670 spa_destroy(char *pool
)
3672 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
3677 * Export a storage pool.
3680 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
3681 boolean_t hardforce
)
3683 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
3688 * Similar to spa_export(), this unloads the spa_t without actually removing it
3689 * from the namespace in any way.
3692 spa_reset(char *pool
)
3694 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
3699 * ==========================================================================
3700 * Device manipulation
3701 * ==========================================================================
3705 * Add a device to a storage pool.
3708 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
3712 vdev_t
*rvd
= spa
->spa_root_vdev
;
3714 nvlist_t
**spares
, **l2cache
;
3715 uint_t nspares
, nl2cache
;
3718 ASSERT(spa_writeable(spa
));
3720 txg
= spa_vdev_enter(spa
);
3722 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
3723 VDEV_ALLOC_ADD
)) != 0)
3724 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
3726 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
3728 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
3732 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
3736 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
3737 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
3739 if (vd
->vdev_children
!= 0 &&
3740 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
3741 return (spa_vdev_exit(spa
, vd
, txg
, error
));
3744 * We must validate the spares and l2cache devices after checking the
3745 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3747 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
3748 return (spa_vdev_exit(spa
, vd
, txg
, error
));
3751 * Transfer each new top-level vdev from vd to rvd.
3753 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3756 * Set the vdev id to the first hole, if one exists.
3758 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
3759 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
3760 vdev_free(rvd
->vdev_child
[id
]);
3764 tvd
= vd
->vdev_child
[c
];
3765 vdev_remove_child(vd
, tvd
);
3767 vdev_add_child(rvd
, tvd
);
3768 vdev_config_dirty(tvd
);
3772 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
3773 ZPOOL_CONFIG_SPARES
);
3774 spa_load_spares(spa
);
3775 spa
->spa_spares
.sav_sync
= B_TRUE
;
3778 if (nl2cache
!= 0) {
3779 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
3780 ZPOOL_CONFIG_L2CACHE
);
3781 spa_load_l2cache(spa
);
3782 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3786 * We have to be careful when adding new vdevs to an existing pool.
3787 * If other threads start allocating from these vdevs before we
3788 * sync the config cache, and we lose power, then upon reboot we may
3789 * fail to open the pool because there are DVAs that the config cache
3790 * can't translate. Therefore, we first add the vdevs without
3791 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3792 * and then let spa_config_update() initialize the new metaslabs.
3794 * spa_load() checks for added-but-not-initialized vdevs, so that
3795 * if we lose power at any point in this sequence, the remaining
3796 * steps will be completed the next time we load the pool.
3798 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
3800 mutex_enter(&spa_namespace_lock
);
3801 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
3802 mutex_exit(&spa_namespace_lock
);
3808 * Attach a device to a mirror. The arguments are the path to any device
3809 * in the mirror, and the nvroot for the new device. If the path specifies
3810 * a device that is not mirrored, we automatically insert the mirror vdev.
3812 * If 'replacing' is specified, the new device is intended to replace the
3813 * existing device; in this case the two devices are made into their own
3814 * mirror using the 'replacing' vdev, which is functionally identical to
3815 * the mirror vdev (it actually reuses all the same ops) but has a few
3816 * extra rules: you can't attach to it after it's been created, and upon
3817 * completion of resilvering, the first disk (the one being replaced)
3818 * is automatically detached.
3821 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
3823 uint64_t txg
, dtl_max_txg
;
3824 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
;)
3825 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
3827 char *oldvdpath
, *newvdpath
;
3831 ASSERT(spa_writeable(spa
));
3833 txg
= spa_vdev_enter(spa
);
3835 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
3838 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
3840 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
3841 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
3843 pvd
= oldvd
->vdev_parent
;
3845 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
3846 VDEV_ALLOC_ADD
)) != 0)
3847 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
3849 if (newrootvd
->vdev_children
!= 1)
3850 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
3852 newvd
= newrootvd
->vdev_child
[0];
3854 if (!newvd
->vdev_ops
->vdev_op_leaf
)
3855 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
3857 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
3858 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
3861 * Spares can't replace logs
3863 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
3864 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3868 * For attach, the only allowable parent is a mirror or the root
3871 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
3872 pvd
->vdev_ops
!= &vdev_root_ops
)
3873 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3875 pvops
= &vdev_mirror_ops
;
3878 * Active hot spares can only be replaced by inactive hot
3881 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
3882 oldvd
->vdev_isspare
&&
3883 !spa_has_spare(spa
, newvd
->vdev_guid
))
3884 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3887 * If the source is a hot spare, and the parent isn't already a
3888 * spare, then we want to create a new hot spare. Otherwise, we
3889 * want to create a replacing vdev. The user is not allowed to
3890 * attach to a spared vdev child unless the 'isspare' state is
3891 * the same (spare replaces spare, non-spare replaces
3894 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
3895 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
3896 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3897 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
3898 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
3899 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
3902 if (newvd
->vdev_isspare
)
3903 pvops
= &vdev_spare_ops
;
3905 pvops
= &vdev_replacing_ops
;
3909 * Make sure the new device is big enough.
3911 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
3912 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
3915 * The new device cannot have a higher alignment requirement
3916 * than the top-level vdev.
3918 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
3919 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
3922 * If this is an in-place replacement, update oldvd's path and devid
3923 * to make it distinguishable from newvd, and unopenable from now on.
3925 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
3926 spa_strfree(oldvd
->vdev_path
);
3927 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
3929 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
3930 newvd
->vdev_path
, "old");
3931 if (oldvd
->vdev_devid
!= NULL
) {
3932 spa_strfree(oldvd
->vdev_devid
);
3933 oldvd
->vdev_devid
= NULL
;
3937 /* mark the device being resilvered */
3938 newvd
->vdev_resilvering
= B_TRUE
;
3941 * If the parent is not a mirror, or if we're replacing, insert the new
3942 * mirror/replacing/spare vdev above oldvd.
3944 if (pvd
->vdev_ops
!= pvops
)
3945 pvd
= vdev_add_parent(oldvd
, pvops
);
3947 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
3948 ASSERT(pvd
->vdev_ops
== pvops
);
3949 ASSERT(oldvd
->vdev_parent
== pvd
);
3952 * Extract the new device from its root and add it to pvd.
3954 vdev_remove_child(newrootvd
, newvd
);
3955 newvd
->vdev_id
= pvd
->vdev_children
;
3956 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
3957 vdev_add_child(pvd
, newvd
);
3959 tvd
= newvd
->vdev_top
;
3960 ASSERT(pvd
->vdev_top
== tvd
);
3961 ASSERT(tvd
->vdev_parent
== rvd
);
3963 vdev_config_dirty(tvd
);
3966 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3967 * for any dmu_sync-ed blocks. It will propagate upward when
3968 * spa_vdev_exit() calls vdev_dtl_reassess().
3970 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
3972 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
3973 dtl_max_txg
- TXG_INITIAL
);
3975 if (newvd
->vdev_isspare
) {
3976 spa_spare_activate(newvd
);
3977 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_DEVICE_SPARE
);
3980 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
3981 newvdpath
= spa_strdup(newvd
->vdev_path
);
3982 newvd_isspare
= newvd
->vdev_isspare
;
3985 * Mark newvd's DTL dirty in this txg.
3987 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
3990 * Restart the resilver
3992 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
3997 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
3999 spa_history_log_internal(LOG_POOL_VDEV_ATTACH
, spa
, NULL
,
4000 "%s vdev=%s %s vdev=%s",
4001 replacing
&& newvd_isspare
? "spare in" :
4002 replacing
? "replace" : "attach", newvdpath
,
4003 replacing
? "for" : "to", oldvdpath
);
4005 spa_strfree(oldvdpath
);
4006 spa_strfree(newvdpath
);
4008 if (spa
->spa_bootfs
)
4009 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH
);
4015 * Detach a device from a mirror or replacing vdev.
4016 * If 'replace_done' is specified, only detach if the parent
4017 * is a replacing vdev.
4020 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4024 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
;)
4025 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4026 boolean_t unspare
= B_FALSE
;
4027 uint64_t unspare_guid
= 0;
4031 ASSERT(spa_writeable(spa
));
4033 txg
= spa_vdev_enter(spa
);
4035 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4038 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4040 if (!vd
->vdev_ops
->vdev_op_leaf
)
4041 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4043 pvd
= vd
->vdev_parent
;
4046 * If the parent/child relationship is not as expected, don't do it.
4047 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4048 * vdev that's replacing B with C. The user's intent in replacing
4049 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4050 * the replace by detaching C, the expected behavior is to end up
4051 * M(A,B). But suppose that right after deciding to detach C,
4052 * the replacement of B completes. We would have M(A,C), and then
4053 * ask to detach C, which would leave us with just A -- not what
4054 * the user wanted. To prevent this, we make sure that the
4055 * parent/child relationship hasn't changed -- in this example,
4056 * that C's parent is still the replacing vdev R.
4058 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4059 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4062 * Only 'replacing' or 'spare' vdevs can be replaced.
4064 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4065 pvd
->vdev_ops
!= &vdev_spare_ops
)
4066 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4068 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4069 spa_version(spa
) >= SPA_VERSION_SPARES
);
4072 * Only mirror, replacing, and spare vdevs support detach.
4074 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4075 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4076 pvd
->vdev_ops
!= &vdev_spare_ops
)
4077 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4080 * If this device has the only valid copy of some data,
4081 * we cannot safely detach it.
4083 if (vdev_dtl_required(vd
))
4084 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4086 ASSERT(pvd
->vdev_children
>= 2);
4089 * If we are detaching the second disk from a replacing vdev, then
4090 * check to see if we changed the original vdev's path to have "/old"
4091 * at the end in spa_vdev_attach(). If so, undo that change now.
4093 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4094 vd
->vdev_path
!= NULL
) {
4095 size_t len
= strlen(vd
->vdev_path
);
4097 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4098 cvd
= pvd
->vdev_child
[c
];
4100 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4103 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4104 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4105 spa_strfree(cvd
->vdev_path
);
4106 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4113 * If we are detaching the original disk from a spare, then it implies
4114 * that the spare should become a real disk, and be removed from the
4115 * active spare list for the pool.
4117 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4119 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4123 * Erase the disk labels so the disk can be used for other things.
4124 * This must be done after all other error cases are handled,
4125 * but before we disembowel vd (so we can still do I/O to it).
4126 * But if we can't do it, don't treat the error as fatal --
4127 * it may be that the unwritability of the disk is the reason
4128 * it's being detached!
4130 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4133 * Remove vd from its parent and compact the parent's children.
4135 vdev_remove_child(pvd
, vd
);
4136 vdev_compact_children(pvd
);
4139 * Remember one of the remaining children so we can get tvd below.
4141 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4144 * If we need to remove the remaining child from the list of hot spares,
4145 * do it now, marking the vdev as no longer a spare in the process.
4146 * We must do this before vdev_remove_parent(), because that can
4147 * change the GUID if it creates a new toplevel GUID. For a similar
4148 * reason, we must remove the spare now, in the same txg as the detach;
4149 * otherwise someone could attach a new sibling, change the GUID, and
4150 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4153 ASSERT(cvd
->vdev_isspare
);
4154 spa_spare_remove(cvd
);
4155 unspare_guid
= cvd
->vdev_guid
;
4156 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4157 cvd
->vdev_unspare
= B_TRUE
;
4161 * If the parent mirror/replacing vdev only has one child,
4162 * the parent is no longer needed. Remove it from the tree.
4164 if (pvd
->vdev_children
== 1) {
4165 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4166 cvd
->vdev_unspare
= B_FALSE
;
4167 vdev_remove_parent(cvd
);
4168 cvd
->vdev_resilvering
= B_FALSE
;
4173 * We don't set tvd until now because the parent we just removed
4174 * may have been the previous top-level vdev.
4176 tvd
= cvd
->vdev_top
;
4177 ASSERT(tvd
->vdev_parent
== rvd
);
4180 * Reevaluate the parent vdev state.
4182 vdev_propagate_state(cvd
);
4185 * If the 'autoexpand' property is set on the pool then automatically
4186 * try to expand the size of the pool. For example if the device we
4187 * just detached was smaller than the others, it may be possible to
4188 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4189 * first so that we can obtain the updated sizes of the leaf vdevs.
4191 if (spa
->spa_autoexpand
) {
4193 vdev_expand(tvd
, txg
);
4196 vdev_config_dirty(tvd
);
4199 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4200 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4201 * But first make sure we're not on any *other* txg's DTL list, to
4202 * prevent vd from being accessed after it's freed.
4204 vdpath
= spa_strdup(vd
->vdev_path
);
4205 for (t
= 0; t
< TXG_SIZE
; t
++)
4206 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4207 vd
->vdev_detached
= B_TRUE
;
4208 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4210 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_REMOVE
);
4212 /* hang on to the spa before we release the lock */
4213 spa_open_ref(spa
, FTAG
);
4215 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4217 spa_history_log_internal(LOG_POOL_VDEV_DETACH
, spa
, NULL
,
4219 spa_strfree(vdpath
);
4222 * If this was the removal of the original device in a hot spare vdev,
4223 * then we want to go through and remove the device from the hot spare
4224 * list of every other pool.
4227 spa_t
*altspa
= NULL
;
4229 mutex_enter(&spa_namespace_lock
);
4230 while ((altspa
= spa_next(altspa
)) != NULL
) {
4231 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4235 spa_open_ref(altspa
, FTAG
);
4236 mutex_exit(&spa_namespace_lock
);
4237 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4238 mutex_enter(&spa_namespace_lock
);
4239 spa_close(altspa
, FTAG
);
4241 mutex_exit(&spa_namespace_lock
);
4243 /* search the rest of the vdevs for spares to remove */
4244 spa_vdev_resilver_done(spa
);
4247 /* all done with the spa; OK to release */
4248 mutex_enter(&spa_namespace_lock
);
4249 spa_close(spa
, FTAG
);
4250 mutex_exit(&spa_namespace_lock
);
4256 * Split a set of devices from their mirrors, and create a new pool from them.
4259 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
4260 nvlist_t
*props
, boolean_t exp
)
4263 uint64_t txg
, *glist
;
4265 uint_t c
, children
, lastlog
;
4266 nvlist_t
**child
, *nvl
, *tmp
;
4268 char *altroot
= NULL
;
4269 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
4270 boolean_t activate_slog
;
4272 ASSERT(spa_writeable(spa
));
4274 txg
= spa_vdev_enter(spa
);
4276 /* clear the log and flush everything up to now */
4277 activate_slog
= spa_passivate_log(spa
);
4278 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4279 error
= spa_offline_log(spa
);
4280 txg
= spa_vdev_config_enter(spa
);
4283 spa_activate_log(spa
);
4286 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4288 /* check new spa name before going any further */
4289 if (spa_lookup(newname
) != NULL
)
4290 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
4293 * scan through all the children to ensure they're all mirrors
4295 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
4296 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
4298 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4300 /* first, check to ensure we've got the right child count */
4301 rvd
= spa
->spa_root_vdev
;
4303 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
4304 vdev_t
*vd
= rvd
->vdev_child
[c
];
4306 /* don't count the holes & logs as children */
4307 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
4315 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
4316 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4318 /* next, ensure no spare or cache devices are part of the split */
4319 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
4320 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
4321 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4323 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
4324 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
4326 /* then, loop over each vdev and validate it */
4327 for (c
= 0; c
< children
; c
++) {
4328 uint64_t is_hole
= 0;
4330 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
4334 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
4335 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
4343 /* which disk is going to be split? */
4344 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
4350 /* look it up in the spa */
4351 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
4352 if (vml
[c
] == NULL
) {
4357 /* make sure there's nothing stopping the split */
4358 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
4359 vml
[c
]->vdev_islog
||
4360 vml
[c
]->vdev_ishole
||
4361 vml
[c
]->vdev_isspare
||
4362 vml
[c
]->vdev_isl2cache
||
4363 !vdev_writeable(vml
[c
]) ||
4364 vml
[c
]->vdev_children
!= 0 ||
4365 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
4366 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
4371 if (vdev_dtl_required(vml
[c
])) {
4376 /* we need certain info from the top level */
4377 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
4378 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
4379 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
4380 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
4381 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
4382 vml
[c
]->vdev_top
->vdev_asize
) == 0);
4383 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
4384 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
4388 kmem_free(vml
, children
* sizeof (vdev_t
*));
4389 kmem_free(glist
, children
* sizeof (uint64_t));
4390 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4393 /* stop writers from using the disks */
4394 for (c
= 0; c
< children
; c
++) {
4396 vml
[c
]->vdev_offline
= B_TRUE
;
4398 vdev_reopen(spa
->spa_root_vdev
);
4401 * Temporarily record the splitting vdevs in the spa config. This
4402 * will disappear once the config is regenerated.
4404 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4405 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
4406 glist
, children
) == 0);
4407 kmem_free(glist
, children
* sizeof (uint64_t));
4409 mutex_enter(&spa
->spa_props_lock
);
4410 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
4412 mutex_exit(&spa
->spa_props_lock
);
4413 spa
->spa_config_splitting
= nvl
;
4414 vdev_config_dirty(spa
->spa_root_vdev
);
4416 /* configure and create the new pool */
4417 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
4418 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4419 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
4420 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
4421 spa_version(spa
)) == 0);
4422 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
4423 spa
->spa_config_txg
) == 0);
4424 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
4425 spa_generate_guid(NULL
)) == 0);
4426 (void) nvlist_lookup_string(props
,
4427 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4429 /* add the new pool to the namespace */
4430 newspa
= spa_add(newname
, config
, altroot
);
4431 newspa
->spa_config_txg
= spa
->spa_config_txg
;
4432 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
4434 /* release the spa config lock, retaining the namespace lock */
4435 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4437 if (zio_injection_enabled
)
4438 zio_handle_panic_injection(spa
, FTAG
, 1);
4440 spa_activate(newspa
, spa_mode_global
);
4441 spa_async_suspend(newspa
);
4443 /* create the new pool from the disks of the original pool */
4444 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
4448 /* if that worked, generate a real config for the new pool */
4449 if (newspa
->spa_root_vdev
!= NULL
) {
4450 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
4451 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4452 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
4453 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
4454 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
4459 if (props
!= NULL
) {
4460 spa_configfile_set(newspa
, props
, B_FALSE
);
4461 error
= spa_prop_set(newspa
, props
);
4466 /* flush everything */
4467 txg
= spa_vdev_config_enter(newspa
);
4468 vdev_config_dirty(newspa
->spa_root_vdev
);
4469 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
4471 if (zio_injection_enabled
)
4472 zio_handle_panic_injection(spa
, FTAG
, 2);
4474 spa_async_resume(newspa
);
4476 /* finally, update the original pool's config */
4477 txg
= spa_vdev_config_enter(spa
);
4478 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
4479 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4482 for (c
= 0; c
< children
; c
++) {
4483 if (vml
[c
] != NULL
) {
4486 spa_history_log_internal(LOG_POOL_VDEV_DETACH
,
4492 vdev_config_dirty(spa
->spa_root_vdev
);
4493 spa
->spa_config_splitting
= NULL
;
4497 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
4499 if (zio_injection_enabled
)
4500 zio_handle_panic_injection(spa
, FTAG
, 3);
4502 /* split is complete; log a history record */
4503 spa_history_log_internal(LOG_POOL_SPLIT
, newspa
, NULL
,
4504 "split new pool %s from pool %s", newname
, spa_name(spa
));
4506 kmem_free(vml
, children
* sizeof (vdev_t
*));
4508 /* if we're not going to mount the filesystems in userland, export */
4510 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
4517 spa_deactivate(newspa
);
4520 txg
= spa_vdev_config_enter(spa
);
4522 /* re-online all offlined disks */
4523 for (c
= 0; c
< children
; c
++) {
4525 vml
[c
]->vdev_offline
= B_FALSE
;
4527 vdev_reopen(spa
->spa_root_vdev
);
4529 nvlist_free(spa
->spa_config_splitting
);
4530 spa
->spa_config_splitting
= NULL
;
4531 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
4533 kmem_free(vml
, children
* sizeof (vdev_t
*));
4538 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
4542 for (i
= 0; i
< count
; i
++) {
4545 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
4548 if (guid
== target_guid
)
4556 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
4557 nvlist_t
*dev_to_remove
)
4559 nvlist_t
**newdev
= NULL
;
4563 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
4565 for (i
= 0, j
= 0; i
< count
; i
++) {
4566 if (dev
[i
] == dev_to_remove
)
4568 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
4571 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4572 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
4574 for (i
= 0; i
< count
- 1; i
++)
4575 nvlist_free(newdev
[i
]);
4578 kmem_free(newdev
, (count
- 1) * sizeof (void *));
4582 * Evacuate the device.
4585 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
4590 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4591 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
4592 ASSERT(vd
== vd
->vdev_top
);
4595 * Evacuate the device. We don't hold the config lock as writer
4596 * since we need to do I/O but we do keep the
4597 * spa_namespace_lock held. Once this completes the device
4598 * should no longer have any blocks allocated on it.
4600 if (vd
->vdev_islog
) {
4601 if (vd
->vdev_stat
.vs_alloc
!= 0)
4602 error
= spa_offline_log(spa
);
4611 * The evacuation succeeded. Remove any remaining MOS metadata
4612 * associated with this vdev, and wait for these changes to sync.
4614 ASSERT3U(vd
->vdev_stat
.vs_alloc
, ==, 0);
4615 txg
= spa_vdev_config_enter(spa
);
4616 vd
->vdev_removing
= B_TRUE
;
4617 vdev_dirty(vd
, 0, NULL
, txg
);
4618 vdev_config_dirty(vd
);
4619 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
4625 * Complete the removal by cleaning up the namespace.
4628 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
4630 vdev_t
*rvd
= spa
->spa_root_vdev
;
4631 uint64_t id
= vd
->vdev_id
;
4632 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
4634 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
4635 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4636 ASSERT(vd
== vd
->vdev_top
);
4639 * Only remove any devices which are empty.
4641 if (vd
->vdev_stat
.vs_alloc
!= 0)
4644 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4646 if (list_link_active(&vd
->vdev_state_dirty_node
))
4647 vdev_state_clean(vd
);
4648 if (list_link_active(&vd
->vdev_config_dirty_node
))
4649 vdev_config_clean(vd
);
4654 vdev_compact_children(rvd
);
4656 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
4657 vdev_add_child(rvd
, vd
);
4659 vdev_config_dirty(rvd
);
4662 * Reassess the health of our root vdev.
4668 * Remove a device from the pool -
4670 * Removing a device from the vdev namespace requires several steps
4671 * and can take a significant amount of time. As a result we use
4672 * the spa_vdev_config_[enter/exit] functions which allow us to
4673 * grab and release the spa_config_lock while still holding the namespace
4674 * lock. During each step the configuration is synced out.
4678 * Remove a device from the pool. Currently, this supports removing only hot
4679 * spares, slogs, and level 2 ARC devices.
4682 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
4685 metaslab_group_t
*mg
;
4686 nvlist_t
**spares
, **l2cache
, *nv
;
4688 uint_t nspares
, nl2cache
;
4690 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
4692 ASSERT(spa_writeable(spa
));
4695 txg
= spa_vdev_enter(spa
);
4697 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4699 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
4700 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
4701 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
4702 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
4704 * Only remove the hot spare if it's not currently in use
4707 if (vd
== NULL
|| unspare
) {
4708 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
4709 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
4710 spa_load_spares(spa
);
4711 spa
->spa_spares
.sav_sync
= B_TRUE
;
4715 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
4716 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
4717 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
4718 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
4720 * Cache devices can always be removed.
4722 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
4723 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
4724 spa_load_l2cache(spa
);
4725 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4726 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
4728 ASSERT(vd
== vd
->vdev_top
);
4731 * XXX - Once we have bp-rewrite this should
4732 * become the common case.
4738 * Stop allocating from this vdev.
4740 metaslab_group_passivate(mg
);
4743 * Wait for the youngest allocations and frees to sync,
4744 * and then wait for the deferral of those frees to finish.
4746 spa_vdev_config_exit(spa
, NULL
,
4747 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
4750 * Attempt to evacuate the vdev.
4752 error
= spa_vdev_remove_evacuate(spa
, vd
);
4754 txg
= spa_vdev_config_enter(spa
);
4757 * If we couldn't evacuate the vdev, unwind.
4760 metaslab_group_activate(mg
);
4761 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4765 * Clean up the vdev namespace.
4767 spa_vdev_remove_from_namespace(spa
, vd
);
4769 } else if (vd
!= NULL
) {
4771 * Normal vdevs cannot be removed (yet).
4776 * There is no vdev of any kind with the specified guid.
4782 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4788 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4789 * current spared, so we can detach it.
4792 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
4794 vdev_t
*newvd
, *oldvd
;
4797 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4798 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
4804 * Check for a completed replacement. We always consider the first
4805 * vdev in the list to be the oldest vdev, and the last one to be
4806 * the newest (see spa_vdev_attach() for how that works). In
4807 * the case where the newest vdev is faulted, we will not automatically
4808 * remove it after a resilver completes. This is OK as it will require
4809 * user intervention to determine which disk the admin wishes to keep.
4811 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
4812 ASSERT(vd
->vdev_children
> 1);
4814 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
4815 oldvd
= vd
->vdev_child
[0];
4817 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
4818 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
4819 !vdev_dtl_required(oldvd
))
4824 * Check for a completed resilver with the 'unspare' flag set.
4826 if (vd
->vdev_ops
== &vdev_spare_ops
) {
4827 vdev_t
*first
= vd
->vdev_child
[0];
4828 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
4830 if (last
->vdev_unspare
) {
4833 } else if (first
->vdev_unspare
) {
4840 if (oldvd
!= NULL
&&
4841 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
4842 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
4843 !vdev_dtl_required(oldvd
))
4847 * If there are more than two spares attached to a disk,
4848 * and those spares are not required, then we want to
4849 * attempt to free them up now so that they can be used
4850 * by other pools. Once we're back down to a single
4851 * disk+spare, we stop removing them.
4853 if (vd
->vdev_children
> 2) {
4854 newvd
= vd
->vdev_child
[1];
4856 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
4857 vdev_dtl_empty(last
, DTL_MISSING
) &&
4858 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
4859 !vdev_dtl_required(newvd
))
4868 spa_vdev_resilver_done(spa_t
*spa
)
4870 vdev_t
*vd
, *pvd
, *ppvd
;
4871 uint64_t guid
, sguid
, pguid
, ppguid
;
4873 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4875 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
4876 pvd
= vd
->vdev_parent
;
4877 ppvd
= pvd
->vdev_parent
;
4878 guid
= vd
->vdev_guid
;
4879 pguid
= pvd
->vdev_guid
;
4880 ppguid
= ppvd
->vdev_guid
;
4883 * If we have just finished replacing a hot spared device, then
4884 * we need to detach the parent's first child (the original hot
4887 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
4888 ppvd
->vdev_children
== 2) {
4889 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
4890 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
4892 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4893 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
4895 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
4897 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4900 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4904 * Update the stored path or FRU for this vdev.
4907 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
4911 boolean_t sync
= B_FALSE
;
4913 ASSERT(spa_writeable(spa
));
4915 spa_vdev_state_enter(spa
, SCL_ALL
);
4917 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4918 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
4920 if (!vd
->vdev_ops
->vdev_op_leaf
)
4921 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
4924 if (strcmp(value
, vd
->vdev_path
) != 0) {
4925 spa_strfree(vd
->vdev_path
);
4926 vd
->vdev_path
= spa_strdup(value
);
4930 if (vd
->vdev_fru
== NULL
) {
4931 vd
->vdev_fru
= spa_strdup(value
);
4933 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
4934 spa_strfree(vd
->vdev_fru
);
4935 vd
->vdev_fru
= spa_strdup(value
);
4940 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
4944 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
4946 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
4950 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
4952 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
4956 * ==========================================================================
4958 * ==========================================================================
4962 spa_scan_stop(spa_t
*spa
)
4964 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
4965 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
4967 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
4971 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
4973 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
4975 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
4979 * If a resilver was requested, but there is no DTL on a
4980 * writeable leaf device, we have nothing to do.
4982 if (func
== POOL_SCAN_RESILVER
&&
4983 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
4984 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
4988 return (dsl_scan(spa
->spa_dsl_pool
, func
));
4992 * ==========================================================================
4993 * SPA async task processing
4994 * ==========================================================================
4998 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5002 if (vd
->vdev_remove_wanted
) {
5003 vd
->vdev_remove_wanted
= B_FALSE
;
5004 vd
->vdev_delayed_close
= B_FALSE
;
5005 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5008 * We want to clear the stats, but we don't want to do a full
5009 * vdev_clear() as that will cause us to throw away
5010 * degraded/faulted state as well as attempt to reopen the
5011 * device, all of which is a waste.
5013 vd
->vdev_stat
.vs_read_errors
= 0;
5014 vd
->vdev_stat
.vs_write_errors
= 0;
5015 vd
->vdev_stat
.vs_checksum_errors
= 0;
5017 vdev_state_dirty(vd
->vdev_top
);
5020 for (c
= 0; c
< vd
->vdev_children
; c
++)
5021 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5025 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5029 if (vd
->vdev_probe_wanted
) {
5030 vd
->vdev_probe_wanted
= B_FALSE
;
5031 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5034 for (c
= 0; c
< vd
->vdev_children
; c
++)
5035 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5039 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5043 if (!spa
->spa_autoexpand
)
5046 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5047 vdev_t
*cvd
= vd
->vdev_child
[c
];
5048 spa_async_autoexpand(spa
, cvd
);
5051 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5054 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND
);
5058 spa_async_thread(spa_t
*spa
)
5062 ASSERT(spa
->spa_sync_on
);
5064 mutex_enter(&spa
->spa_async_lock
);
5065 tasks
= spa
->spa_async_tasks
;
5066 spa
->spa_async_tasks
= 0;
5067 mutex_exit(&spa
->spa_async_lock
);
5070 * See if the config needs to be updated.
5072 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5073 uint64_t old_space
, new_space
;
5075 mutex_enter(&spa_namespace_lock
);
5076 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5077 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5078 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5079 mutex_exit(&spa_namespace_lock
);
5082 * If the pool grew as a result of the config update,
5083 * then log an internal history event.
5085 if (new_space
!= old_space
) {
5086 spa_history_log_internal(LOG_POOL_VDEV_ONLINE
,
5088 "pool '%s' size: %llu(+%llu)",
5089 spa_name(spa
), new_space
, new_space
- old_space
);
5094 * See if any devices need to be marked REMOVED.
5096 if (tasks
& SPA_ASYNC_REMOVE
) {
5097 spa_vdev_state_enter(spa
, SCL_NONE
);
5098 spa_async_remove(spa
, spa
->spa_root_vdev
);
5099 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5100 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5101 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5102 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5103 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5106 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5107 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5108 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5109 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5113 * See if any devices need to be probed.
5115 if (tasks
& SPA_ASYNC_PROBE
) {
5116 spa_vdev_state_enter(spa
, SCL_NONE
);
5117 spa_async_probe(spa
, spa
->spa_root_vdev
);
5118 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5122 * If any devices are done replacing, detach them.
5124 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5125 spa_vdev_resilver_done(spa
);
5128 * Kick off a resilver.
5130 if (tasks
& SPA_ASYNC_RESILVER
)
5131 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5134 * Let the world know that we're done.
5136 mutex_enter(&spa
->spa_async_lock
);
5137 spa
->spa_async_thread
= NULL
;
5138 cv_broadcast(&spa
->spa_async_cv
);
5139 mutex_exit(&spa
->spa_async_lock
);
5144 spa_async_suspend(spa_t
*spa
)
5146 mutex_enter(&spa
->spa_async_lock
);
5147 spa
->spa_async_suspended
++;
5148 while (spa
->spa_async_thread
!= NULL
)
5149 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5150 mutex_exit(&spa
->spa_async_lock
);
5154 spa_async_resume(spa_t
*spa
)
5156 mutex_enter(&spa
->spa_async_lock
);
5157 ASSERT(spa
->spa_async_suspended
!= 0);
5158 spa
->spa_async_suspended
--;
5159 mutex_exit(&spa
->spa_async_lock
);
5163 spa_async_dispatch(spa_t
*spa
)
5165 mutex_enter(&spa
->spa_async_lock
);
5166 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5167 spa
->spa_async_thread
== NULL
&&
5168 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5169 spa
->spa_async_thread
= thread_create(NULL
, 0,
5170 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5171 mutex_exit(&spa
->spa_async_lock
);
5175 spa_async_request(spa_t
*spa
, int task
)
5177 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5178 mutex_enter(&spa
->spa_async_lock
);
5179 spa
->spa_async_tasks
|= task
;
5180 mutex_exit(&spa
->spa_async_lock
);
5184 * ==========================================================================
5185 * SPA syncing routines
5186 * ==========================================================================
5190 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5193 bpobj_enqueue(bpo
, bp
, tx
);
5198 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5202 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5208 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5210 char *packed
= NULL
;
5215 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5218 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5219 * information. This avoids the dbuf_will_dirty() path and
5220 * saves us a pre-read to get data we don't actually care about.
5222 bufsize
= P2ROUNDUP(nvsize
, SPA_CONFIG_BLOCKSIZE
);
5223 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
5225 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
5227 bzero(packed
+ nvsize
, bufsize
- nvsize
);
5229 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
5231 vmem_free(packed
, bufsize
);
5233 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
5234 dmu_buf_will_dirty(db
, tx
);
5235 *(uint64_t *)db
->db_data
= nvsize
;
5236 dmu_buf_rele(db
, FTAG
);
5240 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
5241 const char *config
, const char *entry
)
5251 * Update the MOS nvlist describing the list of available devices.
5252 * spa_validate_aux() will have already made sure this nvlist is
5253 * valid and the vdevs are labeled appropriately.
5255 if (sav
->sav_object
== 0) {
5256 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
5257 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
5258 sizeof (uint64_t), tx
);
5259 VERIFY(zap_update(spa
->spa_meta_objset
,
5260 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
5261 &sav
->sav_object
, tx
) == 0);
5264 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5265 if (sav
->sav_count
== 0) {
5266 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
5268 list
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
5269 for (i
= 0; i
< sav
->sav_count
; i
++)
5270 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
5271 B_FALSE
, VDEV_CONFIG_L2CACHE
);
5272 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
5273 sav
->sav_count
) == 0);
5274 for (i
= 0; i
< sav
->sav_count
; i
++)
5275 nvlist_free(list
[i
]);
5276 kmem_free(list
, sav
->sav_count
* sizeof (void *));
5279 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
5280 nvlist_free(nvroot
);
5282 sav
->sav_sync
= B_FALSE
;
5286 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
5290 if (list_is_empty(&spa
->spa_config_dirty_list
))
5293 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5295 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
5296 dmu_tx_get_txg(tx
), B_FALSE
);
5298 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5300 if (spa
->spa_config_syncing
)
5301 nvlist_free(spa
->spa_config_syncing
);
5302 spa
->spa_config_syncing
= config
;
5304 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
5308 * Set zpool properties.
5311 spa_sync_props(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
5314 objset_t
*mos
= spa
->spa_meta_objset
;
5315 nvlist_t
*nvp
= arg2
;
5320 const char *propname
;
5321 zprop_type_t proptype
;
5323 mutex_enter(&spa
->spa_props_lock
);
5326 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
5327 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
5328 case ZPOOL_PROP_VERSION
:
5330 * Only set version for non-zpool-creation cases
5331 * (set/import). spa_create() needs special care
5332 * for version setting.
5334 if (tx
->tx_txg
!= TXG_INITIAL
) {
5335 VERIFY(nvpair_value_uint64(elem
,
5337 ASSERT(intval
<= SPA_VERSION
);
5338 ASSERT(intval
>= spa_version(spa
));
5339 spa
->spa_uberblock
.ub_version
= intval
;
5340 vdev_config_dirty(spa
->spa_root_vdev
);
5344 case ZPOOL_PROP_ALTROOT
:
5346 * 'altroot' is a non-persistent property. It should
5347 * have been set temporarily at creation or import time.
5349 ASSERT(spa
->spa_root
!= NULL
);
5352 case ZPOOL_PROP_READONLY
:
5353 case ZPOOL_PROP_CACHEFILE
:
5355 * 'readonly' and 'cachefile' are also non-persisitent
5361 * Set pool property values in the poolprops mos object.
5363 if (spa
->spa_pool_props_object
== 0) {
5364 VERIFY((spa
->spa_pool_props_object
=
5365 zap_create(mos
, DMU_OT_POOL_PROPS
,
5366 DMU_OT_NONE
, 0, tx
)) > 0);
5368 VERIFY(zap_update(mos
,
5369 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
5370 8, 1, &spa
->spa_pool_props_object
, tx
)
5374 /* normalize the property name */
5375 propname
= zpool_prop_to_name(prop
);
5376 proptype
= zpool_prop_get_type(prop
);
5378 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
5379 ASSERT(proptype
== PROP_TYPE_STRING
);
5380 VERIFY(nvpair_value_string(elem
, &strval
) == 0);
5381 VERIFY(zap_update(mos
,
5382 spa
->spa_pool_props_object
, propname
,
5383 1, strlen(strval
) + 1, strval
, tx
) == 0);
5385 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
5386 VERIFY(nvpair_value_uint64(elem
, &intval
) == 0);
5388 if (proptype
== PROP_TYPE_INDEX
) {
5390 VERIFY(zpool_prop_index_to_string(
5391 prop
, intval
, &unused
) == 0);
5393 VERIFY(zap_update(mos
,
5394 spa
->spa_pool_props_object
, propname
,
5395 8, 1, &intval
, tx
) == 0);
5397 ASSERT(0); /* not allowed */
5401 case ZPOOL_PROP_DELEGATION
:
5402 spa
->spa_delegation
= intval
;
5404 case ZPOOL_PROP_BOOTFS
:
5405 spa
->spa_bootfs
= intval
;
5407 case ZPOOL_PROP_FAILUREMODE
:
5408 spa
->spa_failmode
= intval
;
5410 case ZPOOL_PROP_AUTOEXPAND
:
5411 spa
->spa_autoexpand
= intval
;
5412 if (tx
->tx_txg
!= TXG_INITIAL
)
5413 spa_async_request(spa
,
5414 SPA_ASYNC_AUTOEXPAND
);
5416 case ZPOOL_PROP_DEDUPDITTO
:
5417 spa
->spa_dedup_ditto
= intval
;
5424 /* log internal history if this is not a zpool create */
5425 if (spa_version(spa
) >= SPA_VERSION_ZPOOL_HISTORY
&&
5426 tx
->tx_txg
!= TXG_INITIAL
) {
5427 spa_history_log_internal(LOG_POOL_PROPSET
,
5428 spa
, tx
, "%s %lld %s",
5429 nvpair_name(elem
), intval
, spa_name(spa
));
5433 mutex_exit(&spa
->spa_props_lock
);
5437 * Perform one-time upgrade on-disk changes. spa_version() does not
5438 * reflect the new version this txg, so there must be no changes this
5439 * txg to anything that the upgrade code depends on after it executes.
5440 * Therefore this must be called after dsl_pool_sync() does the sync
5444 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
5446 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
5448 ASSERT(spa
->spa_sync_pass
== 1);
5450 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
5451 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
5452 dsl_pool_create_origin(dp
, tx
);
5454 /* Keeping the origin open increases spa_minref */
5455 spa
->spa_minref
+= 3;
5458 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
5459 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
5460 dsl_pool_upgrade_clones(dp
, tx
);
5463 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
5464 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
5465 dsl_pool_upgrade_dir_clones(dp
, tx
);
5467 /* Keeping the freedir open increases spa_minref */
5468 spa
->spa_minref
+= 3;
5473 * Sync the specified transaction group. New blocks may be dirtied as
5474 * part of the process, so we iterate until it converges.
5477 spa_sync(spa_t
*spa
, uint64_t txg
)
5479 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
5480 objset_t
*mos
= spa
->spa_meta_objset
;
5481 bpobj_t
*defer_bpo
= &spa
->spa_deferred_bpobj
;
5482 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
5483 vdev_t
*rvd
= spa
->spa_root_vdev
;
5489 VERIFY(spa_writeable(spa
));
5492 * Lock out configuration changes.
5494 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5496 spa
->spa_syncing_txg
= txg
;
5497 spa
->spa_sync_pass
= 0;
5500 * If there are any pending vdev state changes, convert them
5501 * into config changes that go out with this transaction group.
5503 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5504 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
5506 * We need the write lock here because, for aux vdevs,
5507 * calling vdev_config_dirty() modifies sav_config.
5508 * This is ugly and will become unnecessary when we
5509 * eliminate the aux vdev wart by integrating all vdevs
5510 * into the root vdev tree.
5512 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
5513 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
5514 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
5515 vdev_state_clean(vd
);
5516 vdev_config_dirty(vd
);
5518 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
5519 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
5521 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5523 tx
= dmu_tx_create_assigned(dp
, txg
);
5526 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5527 * set spa_deflate if we have no raid-z vdevs.
5529 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
5530 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
5533 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
5534 vd
= rvd
->vdev_child
[i
];
5535 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
5538 if (i
== rvd
->vdev_children
) {
5539 spa
->spa_deflate
= TRUE
;
5540 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
5541 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
5542 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
5547 * If anything has changed in this txg, or if someone is waiting
5548 * for this txg to sync (eg, spa_vdev_remove()), push the
5549 * deferred frees from the previous txg. If not, leave them
5550 * alone so that we don't generate work on an otherwise idle
5553 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
5554 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
5555 !txg_list_empty(&dp
->dp_sync_tasks
, txg
) ||
5556 ((dsl_scan_active(dp
->dp_scan
) ||
5557 txg_sync_waiting(dp
)) && !spa_shutting_down(spa
))) {
5558 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5559 VERIFY3U(bpobj_iterate(defer_bpo
,
5560 spa_free_sync_cb
, zio
, tx
), ==, 0);
5561 VERIFY3U(zio_wait(zio
), ==, 0);
5565 * Iterate to convergence.
5568 int pass
= ++spa
->spa_sync_pass
;
5570 spa_sync_config_object(spa
, tx
);
5571 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
5572 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
5573 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
5574 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
5575 spa_errlog_sync(spa
, txg
);
5576 dsl_pool_sync(dp
, txg
);
5578 if (pass
<= SYNC_PASS_DEFERRED_FREE
) {
5579 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5580 bplist_iterate(free_bpl
, spa_free_sync_cb
,
5582 VERIFY(zio_wait(zio
) == 0);
5584 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
5589 dsl_scan_sync(dp
, tx
);
5591 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
5595 spa_sync_upgrades(spa
, tx
);
5597 } while (dmu_objset_is_dirty(mos
, txg
));
5600 * Rewrite the vdev configuration (which includes the uberblock)
5601 * to commit the transaction group.
5603 * If there are no dirty vdevs, we sync the uberblock to a few
5604 * random top-level vdevs that are known to be visible in the
5605 * config cache (see spa_vdev_add() for a complete description).
5606 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5610 * We hold SCL_STATE to prevent vdev open/close/etc.
5611 * while we're attempting to write the vdev labels.
5613 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5615 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
5616 vdev_t
*svd
[SPA_DVAS_PER_BP
];
5618 int children
= rvd
->vdev_children
;
5619 int c0
= spa_get_random(children
);
5621 for (c
= 0; c
< children
; c
++) {
5622 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
5623 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
5625 svd
[svdcount
++] = vd
;
5626 if (svdcount
== SPA_DVAS_PER_BP
)
5629 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
5631 error
= vdev_config_sync(svd
, svdcount
, txg
,
5634 error
= vdev_config_sync(rvd
->vdev_child
,
5635 rvd
->vdev_children
, txg
, B_FALSE
);
5637 error
= vdev_config_sync(rvd
->vdev_child
,
5638 rvd
->vdev_children
, txg
, B_TRUE
);
5641 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5645 zio_suspend(spa
, NULL
);
5646 zio_resume_wait(spa
);
5651 * Clear the dirty config list.
5653 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
5654 vdev_config_clean(vd
);
5657 * Now that the new config has synced transactionally,
5658 * let it become visible to the config cache.
5660 if (spa
->spa_config_syncing
!= NULL
) {
5661 spa_config_set(spa
, spa
->spa_config_syncing
);
5662 spa
->spa_config_txg
= txg
;
5663 spa
->spa_config_syncing
= NULL
;
5666 spa
->spa_ubsync
= spa
->spa_uberblock
;
5668 dsl_pool_sync_done(dp
, txg
);
5671 * Update usable space statistics.
5673 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
5674 vdev_sync_done(vd
, txg
);
5676 spa_update_dspace(spa
);
5679 * It had better be the case that we didn't dirty anything
5680 * since vdev_config_sync().
5682 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
5683 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
5684 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
5686 spa
->spa_sync_pass
= 0;
5688 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5690 spa_handle_ignored_writes(spa
);
5693 * If any async tasks have been requested, kick them off.
5695 spa_async_dispatch(spa
);
5699 * Sync all pools. We don't want to hold the namespace lock across these
5700 * operations, so we take a reference on the spa_t and drop the lock during the
5704 spa_sync_allpools(void)
5707 mutex_enter(&spa_namespace_lock
);
5708 while ((spa
= spa_next(spa
)) != NULL
) {
5709 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
5710 !spa_writeable(spa
) || spa_suspended(spa
))
5712 spa_open_ref(spa
, FTAG
);
5713 mutex_exit(&spa_namespace_lock
);
5714 txg_wait_synced(spa_get_dsl(spa
), 0);
5715 mutex_enter(&spa_namespace_lock
);
5716 spa_close(spa
, FTAG
);
5718 mutex_exit(&spa_namespace_lock
);
5722 * ==========================================================================
5723 * Miscellaneous routines
5724 * ==========================================================================
5728 * Remove all pools in the system.
5736 * Remove all cached state. All pools should be closed now,
5737 * so every spa in the AVL tree should be unreferenced.
5739 mutex_enter(&spa_namespace_lock
);
5740 while ((spa
= spa_next(NULL
)) != NULL
) {
5742 * Stop async tasks. The async thread may need to detach
5743 * a device that's been replaced, which requires grabbing
5744 * spa_namespace_lock, so we must drop it here.
5746 spa_open_ref(spa
, FTAG
);
5747 mutex_exit(&spa_namespace_lock
);
5748 spa_async_suspend(spa
);
5749 mutex_enter(&spa_namespace_lock
);
5750 spa_close(spa
, FTAG
);
5752 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
5754 spa_deactivate(spa
);
5758 mutex_exit(&spa_namespace_lock
);
5762 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
5767 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
5771 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
5772 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
5773 if (vd
->vdev_guid
== guid
)
5777 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
5778 vd
= spa
->spa_spares
.sav_vdevs
[i
];
5779 if (vd
->vdev_guid
== guid
)
5788 spa_upgrade(spa_t
*spa
, uint64_t version
)
5790 ASSERT(spa_writeable(spa
));
5792 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5795 * This should only be called for a non-faulted pool, and since a
5796 * future version would result in an unopenable pool, this shouldn't be
5799 ASSERT(spa
->spa_uberblock
.ub_version
<= SPA_VERSION
);
5800 ASSERT(version
>= spa
->spa_uberblock
.ub_version
);
5802 spa
->spa_uberblock
.ub_version
= version
;
5803 vdev_config_dirty(spa
->spa_root_vdev
);
5805 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5807 txg_wait_synced(spa_get_dsl(spa
), 0);
5811 spa_has_spare(spa_t
*spa
, uint64_t guid
)
5815 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
5817 for (i
= 0; i
< sav
->sav_count
; i
++)
5818 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
5821 for (i
= 0; i
< sav
->sav_npending
; i
++) {
5822 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
5823 &spareguid
) == 0 && spareguid
== guid
)
5831 * Check if a pool has an active shared spare device.
5832 * Note: reference count of an active spare is 2, as a spare and as a replace
5835 spa_has_active_shared_spare(spa_t
*spa
)
5839 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
5841 for (i
= 0; i
< sav
->sav_count
; i
++) {
5842 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
5843 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
5852 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
5853 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5854 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5855 * or zdb as real changes.
5858 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
5861 zfs_ereport_post(name
, spa
, vd
, NULL
, 0, 0);
5865 #if defined(_KERNEL) && defined(HAVE_SPL)
5866 /* state manipulation functions */
5867 EXPORT_SYMBOL(spa_open
);
5868 EXPORT_SYMBOL(spa_open_rewind
);
5869 EXPORT_SYMBOL(spa_get_stats
);
5870 EXPORT_SYMBOL(spa_create
);
5871 EXPORT_SYMBOL(spa_import_rootpool
);
5872 EXPORT_SYMBOL(spa_import
);
5873 EXPORT_SYMBOL(spa_tryimport
);
5874 EXPORT_SYMBOL(spa_destroy
);
5875 EXPORT_SYMBOL(spa_export
);
5876 EXPORT_SYMBOL(spa_reset
);
5877 EXPORT_SYMBOL(spa_async_request
);
5878 EXPORT_SYMBOL(spa_async_suspend
);
5879 EXPORT_SYMBOL(spa_async_resume
);
5880 EXPORT_SYMBOL(spa_inject_addref
);
5881 EXPORT_SYMBOL(spa_inject_delref
);
5882 EXPORT_SYMBOL(spa_scan_stat_init
);
5883 EXPORT_SYMBOL(spa_scan_get_stats
);
5885 /* device maniion */
5886 EXPORT_SYMBOL(spa_vdev_add
);
5887 EXPORT_SYMBOL(spa_vdev_attach
);
5888 EXPORT_SYMBOL(spa_vdev_detach
);
5889 EXPORT_SYMBOL(spa_vdev_remove
);
5890 EXPORT_SYMBOL(spa_vdev_setpath
);
5891 EXPORT_SYMBOL(spa_vdev_setfru
);
5892 EXPORT_SYMBOL(spa_vdev_split_mirror
);
5894 /* spare statech is global across all pools) */
5895 EXPORT_SYMBOL(spa_spare_add
);
5896 EXPORT_SYMBOL(spa_spare_remove
);
5897 EXPORT_SYMBOL(spa_spare_exists
);
5898 EXPORT_SYMBOL(spa_spare_activate
);
5900 /* L2ARC statech is global across all pools) */
5901 EXPORT_SYMBOL(spa_l2cache_add
);
5902 EXPORT_SYMBOL(spa_l2cache_remove
);
5903 EXPORT_SYMBOL(spa_l2cache_exists
);
5904 EXPORT_SYMBOL(spa_l2cache_activate
);
5905 EXPORT_SYMBOL(spa_l2cache_drop
);
5908 EXPORT_SYMBOL(spa_scan
);
5909 EXPORT_SYMBOL(spa_scan_stop
);
5912 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
5913 EXPORT_SYMBOL(spa_sync_allpools
);
5916 EXPORT_SYMBOL(spa_prop_set
);
5917 EXPORT_SYMBOL(spa_prop_get
);
5918 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
5920 /* asynchronous event notification */
5921 EXPORT_SYMBOL(spa_event_notify
);