4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
30 * SPA: Storage Pool Allocator
32 * This file contains all the routines used when modifying on-disk SPA state.
33 * This includes opening, importing, destroying, exporting a pool, and syncing a
37 #include <sys/zfs_context.h>
38 #include <sys/fm/fs/zfs.h>
39 #include <sys/spa_impl.h>
41 #include <sys/zio_checksum.h>
43 #include <sys/dmu_tx.h>
47 #include <sys/vdev_impl.h>
48 #include <sys/vdev_disk.h>
49 #include <sys/metaslab.h>
50 #include <sys/metaslab_impl.h>
51 #include <sys/uberblock_impl.h>
54 #include <sys/dmu_traverse.h>
55 #include <sys/dmu_objset.h>
56 #include <sys/unique.h>
57 #include <sys/dsl_pool.h>
58 #include <sys/dsl_dataset.h>
59 #include <sys/dsl_dir.h>
60 #include <sys/dsl_prop.h>
61 #include <sys/dsl_synctask.h>
62 #include <sys/fs/zfs.h>
64 #include <sys/callb.h>
65 #include <sys/systeminfo.h>
66 #include <sys/spa_boot.h>
67 #include <sys/zfs_ioctl.h>
68 #include <sys/dsl_scan.h>
69 #include <sys/zfeature.h>
70 #include <sys/dsl_destroy.h>
74 #include <sys/bootprops.h>
75 #include <sys/callb.h>
76 #include <sys/cpupart.h>
78 #include <sys/sysdc.h>
83 #include "zfs_comutil.h"
85 typedef enum zti_modes
{
86 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
87 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
88 ZTI_MODE_NULL
, /* don't create a taskq */
92 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
93 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
94 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
95 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
97 #define ZTI_N(n) ZTI_P(n, 1)
98 #define ZTI_ONE ZTI_N(1)
100 typedef struct zio_taskq_info
{
101 zti_modes_t zti_mode
;
106 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
107 "iss", "iss_h", "int", "int_h"
111 * This table defines the taskq settings for each ZFS I/O type. When
112 * initializing a pool, we use this table to create an appropriately sized
113 * taskq. Some operations are low volume and therefore have a small, static
114 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
115 * macros. Other operations process a large amount of data; the ZTI_BATCH
116 * macro causes us to create a taskq oriented for throughput. Some operations
117 * are so high frequency and short-lived that the taskq itself can become a a
118 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
119 * additional degree of parallelism specified by the number of threads per-
120 * taskq and the number of taskqs; when dispatching an event in this case, the
121 * particular taskq is chosen at random.
123 * The different taskq priorities are to handle the different contexts (issue
124 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
125 * need to be handled with minimum delay.
127 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
128 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
129 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
130 { ZTI_N(8), ZTI_NULL
, ZTI_BATCH
, ZTI_NULL
}, /* READ */
131 { ZTI_BATCH
, ZTI_N(5), ZTI_N(16), ZTI_N(5) }, /* WRITE */
132 { ZTI_P(4, 8), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
133 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
134 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
137 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
138 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
139 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
140 static inline int spa_load_impl(spa_t
*spa
, uint64_t, nvlist_t
*config
,
141 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
143 static void spa_vdev_resilver_done(spa_t
*spa
);
145 uint_t zio_taskq_batch_pct
= 75; /* 1 thread per cpu in pset */
146 id_t zio_taskq_psrset_bind
= PS_NONE
;
147 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
148 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
150 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
153 * This (illegal) pool name is used when temporarily importing a spa_t in order
154 * to get the vdev stats associated with the imported devices.
156 #define TRYIMPORT_NAME "$import"
159 * ==========================================================================
160 * SPA properties routines
161 * ==========================================================================
165 * Add a (source=src, propname=propval) list to an nvlist.
168 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
169 uint64_t intval
, zprop_source_t src
)
171 const char *propname
= zpool_prop_to_name(prop
);
174 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
175 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
178 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
180 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
182 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
183 nvlist_free(propval
);
187 * Get property values from the spa configuration.
190 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
192 vdev_t
*rvd
= spa
->spa_root_vdev
;
193 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
194 uint64_t size
, alloc
, cap
, version
;
195 zprop_source_t src
= ZPROP_SRC_NONE
;
196 spa_config_dirent_t
*dp
;
197 metaslab_class_t
*mc
= spa_normal_class(spa
);
199 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
202 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
203 size
= metaslab_class_get_space(spa_normal_class(spa
));
204 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
205 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
206 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
207 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
210 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
211 metaslab_class_fragmentation(mc
), src
);
212 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
213 metaslab_class_expandable_space(mc
), src
);
214 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
215 (spa_mode(spa
) == FREAD
), src
);
217 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
218 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
220 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
221 ddt_get_pool_dedup_ratio(spa
), src
);
223 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
224 rvd
->vdev_state
, src
);
226 version
= spa_version(spa
);
227 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
))
228 src
= ZPROP_SRC_DEFAULT
;
230 src
= ZPROP_SRC_LOCAL
;
231 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
236 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
237 * when opening pools before this version freedir will be NULL.
239 if (pool
->dp_free_dir
!= NULL
) {
240 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
241 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
244 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
248 if (pool
->dp_leak_dir
!= NULL
) {
249 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
250 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
253 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
258 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
260 if (spa
->spa_comment
!= NULL
) {
261 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
265 if (spa
->spa_root
!= NULL
)
266 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
269 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
270 if (dp
->scd_path
== NULL
) {
271 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
272 "none", 0, ZPROP_SRC_LOCAL
);
273 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
274 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
275 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
281 * Get zpool property values.
284 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
286 objset_t
*mos
= spa
->spa_meta_objset
;
291 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
295 mutex_enter(&spa
->spa_props_lock
);
298 * Get properties from the spa config.
300 spa_prop_get_config(spa
, nvp
);
302 /* If no pool property object, no more prop to get. */
303 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
304 mutex_exit(&spa
->spa_props_lock
);
309 * Get properties from the MOS pool property object.
311 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
312 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
313 zap_cursor_advance(&zc
)) {
316 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
319 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
322 switch (za
.za_integer_length
) {
324 /* integer property */
325 if (za
.za_first_integer
!=
326 zpool_prop_default_numeric(prop
))
327 src
= ZPROP_SRC_LOCAL
;
329 if (prop
== ZPOOL_PROP_BOOTFS
) {
331 dsl_dataset_t
*ds
= NULL
;
333 dp
= spa_get_dsl(spa
);
334 dsl_pool_config_enter(dp
, FTAG
);
335 if ((err
= dsl_dataset_hold_obj(dp
,
336 za
.za_first_integer
, FTAG
, &ds
))) {
337 dsl_pool_config_exit(dp
, FTAG
);
342 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1,
344 dsl_dataset_name(ds
, strval
);
345 dsl_dataset_rele(ds
, FTAG
);
346 dsl_pool_config_exit(dp
, FTAG
);
349 intval
= za
.za_first_integer
;
352 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
356 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1);
361 /* string property */
362 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
363 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
364 za
.za_name
, 1, za
.za_num_integers
, strval
);
366 kmem_free(strval
, za
.za_num_integers
);
369 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
370 kmem_free(strval
, za
.za_num_integers
);
377 zap_cursor_fini(&zc
);
378 mutex_exit(&spa
->spa_props_lock
);
380 if (err
&& err
!= ENOENT
) {
390 * Validate the given pool properties nvlist and modify the list
391 * for the property values to be set.
394 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
397 int error
= 0, reset_bootfs
= 0;
399 boolean_t has_feature
= B_FALSE
;
402 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
404 char *strval
, *slash
, *check
, *fname
;
405 const char *propname
= nvpair_name(elem
);
406 zpool_prop_t prop
= zpool_name_to_prop(propname
);
410 if (!zpool_prop_feature(propname
)) {
411 error
= SET_ERROR(EINVAL
);
416 * Sanitize the input.
418 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
419 error
= SET_ERROR(EINVAL
);
423 if (nvpair_value_uint64(elem
, &intval
) != 0) {
424 error
= SET_ERROR(EINVAL
);
429 error
= SET_ERROR(EINVAL
);
433 fname
= strchr(propname
, '@') + 1;
434 if (zfeature_lookup_name(fname
, NULL
) != 0) {
435 error
= SET_ERROR(EINVAL
);
439 has_feature
= B_TRUE
;
442 case ZPOOL_PROP_VERSION
:
443 error
= nvpair_value_uint64(elem
, &intval
);
445 (intval
< spa_version(spa
) ||
446 intval
> SPA_VERSION_BEFORE_FEATURES
||
448 error
= SET_ERROR(EINVAL
);
451 case ZPOOL_PROP_DELEGATION
:
452 case ZPOOL_PROP_AUTOREPLACE
:
453 case ZPOOL_PROP_LISTSNAPS
:
454 case ZPOOL_PROP_AUTOEXPAND
:
455 error
= nvpair_value_uint64(elem
, &intval
);
456 if (!error
&& intval
> 1)
457 error
= SET_ERROR(EINVAL
);
460 case ZPOOL_PROP_BOOTFS
:
462 * If the pool version is less than SPA_VERSION_BOOTFS,
463 * or the pool is still being created (version == 0),
464 * the bootfs property cannot be set.
466 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
467 error
= SET_ERROR(ENOTSUP
);
472 * Make sure the vdev config is bootable
474 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
475 error
= SET_ERROR(ENOTSUP
);
481 error
= nvpair_value_string(elem
, &strval
);
487 if (strval
== NULL
|| strval
[0] == '\0') {
488 objnum
= zpool_prop_default_numeric(
493 error
= dmu_objset_hold(strval
, FTAG
, &os
);
497 /* Must be ZPL and not gzip compressed. */
499 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
500 error
= SET_ERROR(ENOTSUP
);
502 dsl_prop_get_int_ds(dmu_objset_ds(os
),
503 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
505 !BOOTFS_COMPRESS_VALID(compress
)) {
506 error
= SET_ERROR(ENOTSUP
);
508 objnum
= dmu_objset_id(os
);
510 dmu_objset_rele(os
, FTAG
);
514 case ZPOOL_PROP_FAILUREMODE
:
515 error
= nvpair_value_uint64(elem
, &intval
);
516 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
517 intval
> ZIO_FAILURE_MODE_PANIC
))
518 error
= SET_ERROR(EINVAL
);
521 * This is a special case which only occurs when
522 * the pool has completely failed. This allows
523 * the user to change the in-core failmode property
524 * without syncing it out to disk (I/Os might
525 * currently be blocked). We do this by returning
526 * EIO to the caller (spa_prop_set) to trick it
527 * into thinking we encountered a property validation
530 if (!error
&& spa_suspended(spa
)) {
531 spa
->spa_failmode
= intval
;
532 error
= SET_ERROR(EIO
);
536 case ZPOOL_PROP_CACHEFILE
:
537 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
540 if (strval
[0] == '\0')
543 if (strcmp(strval
, "none") == 0)
546 if (strval
[0] != '/') {
547 error
= SET_ERROR(EINVAL
);
551 slash
= strrchr(strval
, '/');
552 ASSERT(slash
!= NULL
);
554 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
555 strcmp(slash
, "/..") == 0)
556 error
= SET_ERROR(EINVAL
);
559 case ZPOOL_PROP_COMMENT
:
560 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
562 for (check
= strval
; *check
!= '\0'; check
++) {
563 if (!isprint(*check
)) {
564 error
= SET_ERROR(EINVAL
);
569 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
570 error
= SET_ERROR(E2BIG
);
573 case ZPOOL_PROP_DEDUPDITTO
:
574 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
575 error
= SET_ERROR(ENOTSUP
);
577 error
= nvpair_value_uint64(elem
, &intval
);
579 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
580 error
= SET_ERROR(EINVAL
);
591 if (!error
&& reset_bootfs
) {
592 error
= nvlist_remove(props
,
593 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
596 error
= nvlist_add_uint64(props
,
597 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
605 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
608 spa_config_dirent_t
*dp
;
610 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
614 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
617 if (cachefile
[0] == '\0')
618 dp
->scd_path
= spa_strdup(spa_config_path
);
619 else if (strcmp(cachefile
, "none") == 0)
622 dp
->scd_path
= spa_strdup(cachefile
);
624 list_insert_head(&spa
->spa_config_list
, dp
);
626 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
630 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
633 nvpair_t
*elem
= NULL
;
634 boolean_t need_sync
= B_FALSE
;
636 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
639 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
640 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
642 if (prop
== ZPOOL_PROP_CACHEFILE
||
643 prop
== ZPOOL_PROP_ALTROOT
||
644 prop
== ZPOOL_PROP_READONLY
)
647 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPROP_INVAL
) {
650 if (prop
== ZPOOL_PROP_VERSION
) {
651 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
653 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
654 ver
= SPA_VERSION_FEATURES
;
658 /* Save time if the version is already set. */
659 if (ver
== spa_version(spa
))
663 * In addition to the pool directory object, we might
664 * create the pool properties object, the features for
665 * read object, the features for write object, or the
666 * feature descriptions object.
668 error
= dsl_sync_task(spa
->spa_name
, NULL
,
669 spa_sync_version
, &ver
,
670 6, ZFS_SPACE_CHECK_RESERVED
);
681 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
682 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
689 * If the bootfs property value is dsobj, clear it.
692 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
694 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
695 VERIFY(zap_remove(spa
->spa_meta_objset
,
696 spa
->spa_pool_props_object
,
697 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
704 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
706 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
707 vdev_t
*rvd
= spa
->spa_root_vdev
;
709 ASSERTV(uint64_t *newguid
= arg
);
711 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
712 vdev_state
= rvd
->vdev_state
;
713 spa_config_exit(spa
, SCL_STATE
, FTAG
);
715 if (vdev_state
!= VDEV_STATE_HEALTHY
)
716 return (SET_ERROR(ENXIO
));
718 ASSERT3U(spa_guid(spa
), !=, *newguid
);
724 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
726 uint64_t *newguid
= arg
;
727 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
729 vdev_t
*rvd
= spa
->spa_root_vdev
;
731 oldguid
= spa_guid(spa
);
733 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
734 rvd
->vdev_guid
= *newguid
;
735 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
736 vdev_config_dirty(rvd
);
737 spa_config_exit(spa
, SCL_STATE
, FTAG
);
739 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
744 * Change the GUID for the pool. This is done so that we can later
745 * re-import a pool built from a clone of our own vdevs. We will modify
746 * the root vdev's guid, our own pool guid, and then mark all of our
747 * vdevs dirty. Note that we must make sure that all our vdevs are
748 * online when we do this, or else any vdevs that weren't present
749 * would be orphaned from our pool. We are also going to issue a
750 * sysevent to update any watchers.
753 spa_change_guid(spa_t
*spa
)
758 mutex_enter(&spa
->spa_vdev_top_lock
);
759 mutex_enter(&spa_namespace_lock
);
760 guid
= spa_generate_guid(NULL
);
762 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
763 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
766 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
767 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_REGUID
);
770 mutex_exit(&spa_namespace_lock
);
771 mutex_exit(&spa
->spa_vdev_top_lock
);
777 * ==========================================================================
778 * SPA state manipulation (open/create/destroy/import/export)
779 * ==========================================================================
783 spa_error_entry_compare(const void *a
, const void *b
)
785 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
786 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
789 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
790 sizeof (zbookmark_phys_t
));
801 * Utility function which retrieves copies of the current logs and
802 * re-initializes them in the process.
805 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
807 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
809 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
810 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
812 avl_create(&spa
->spa_errlist_scrub
,
813 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
814 offsetof(spa_error_entry_t
, se_avl
));
815 avl_create(&spa
->spa_errlist_last
,
816 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
817 offsetof(spa_error_entry_t
, se_avl
));
821 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
823 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
824 enum zti_modes mode
= ztip
->zti_mode
;
825 uint_t value
= ztip
->zti_value
;
826 uint_t count
= ztip
->zti_count
;
827 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
830 boolean_t batch
= B_FALSE
;
832 if (mode
== ZTI_MODE_NULL
) {
834 tqs
->stqs_taskq
= NULL
;
838 ASSERT3U(count
, >, 0);
840 tqs
->stqs_count
= count
;
841 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
845 ASSERT3U(value
, >=, 1);
846 value
= MAX(value
, 1);
851 flags
|= TASKQ_THREADS_CPU_PCT
;
852 value
= zio_taskq_batch_pct
;
856 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
858 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
862 for (i
= 0; i
< count
; i
++) {
866 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
867 zio_type_name
[t
], zio_taskq_types
[q
], i
);
869 (void) snprintf(name
, sizeof (name
), "%s_%s",
870 zio_type_name
[t
], zio_taskq_types
[q
]);
873 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
875 flags
|= TASKQ_DC_BATCH
;
877 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
878 spa
->spa_proc
, zio_taskq_basedc
, flags
);
880 pri_t pri
= maxclsyspri
;
882 * The write issue taskq can be extremely CPU
883 * intensive. Run it at slightly lower priority
884 * than the other taskqs.
886 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
889 tq
= taskq_create_proc(name
, value
, pri
, 50,
890 INT_MAX
, spa
->spa_proc
, flags
);
893 tqs
->stqs_taskq
[i
] = tq
;
898 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
900 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
903 if (tqs
->stqs_taskq
== NULL
) {
904 ASSERT3U(tqs
->stqs_count
, ==, 0);
908 for (i
= 0; i
< tqs
->stqs_count
; i
++) {
909 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
910 taskq_destroy(tqs
->stqs_taskq
[i
]);
913 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
914 tqs
->stqs_taskq
= NULL
;
918 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
919 * Note that a type may have multiple discrete taskqs to avoid lock contention
920 * on the taskq itself. In that case we choose which taskq at random by using
921 * the low bits of gethrtime().
924 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
925 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
927 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
930 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
931 ASSERT3U(tqs
->stqs_count
, !=, 0);
933 if (tqs
->stqs_count
== 1) {
934 tq
= tqs
->stqs_taskq
[0];
936 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
939 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
943 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
946 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
947 task_func_t
*func
, void *arg
, uint_t flags
)
949 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
953 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
954 ASSERT3U(tqs
->stqs_count
, !=, 0);
956 if (tqs
->stqs_count
== 1) {
957 tq
= tqs
->stqs_taskq
[0];
959 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
962 id
= taskq_dispatch(tq
, func
, arg
, flags
);
964 taskq_wait_id(tq
, id
);
968 spa_create_zio_taskqs(spa_t
*spa
)
972 for (t
= 0; t
< ZIO_TYPES
; t
++) {
973 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
974 spa_taskqs_init(spa
, t
, q
);
979 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
981 spa_thread(void *arg
)
986 user_t
*pu
= PTOU(curproc
);
988 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
991 ASSERT(curproc
!= &p0
);
992 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
993 "zpool-%s", spa
->spa_name
);
994 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
996 /* bind this thread to the requested psrset */
997 if (zio_taskq_psrset_bind
!= PS_NONE
) {
999 mutex_enter(&cpu_lock
);
1000 mutex_enter(&pidlock
);
1001 mutex_enter(&curproc
->p_lock
);
1003 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1004 0, NULL
, NULL
) == 0) {
1005 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1008 "Couldn't bind process for zfs pool \"%s\" to "
1009 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1012 mutex_exit(&curproc
->p_lock
);
1013 mutex_exit(&pidlock
);
1014 mutex_exit(&cpu_lock
);
1018 if (zio_taskq_sysdc
) {
1019 sysdc_thread_enter(curthread
, 100, 0);
1022 spa
->spa_proc
= curproc
;
1023 spa
->spa_did
= curthread
->t_did
;
1025 spa_create_zio_taskqs(spa
);
1027 mutex_enter(&spa
->spa_proc_lock
);
1028 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1030 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1031 cv_broadcast(&spa
->spa_proc_cv
);
1033 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1034 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1035 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1036 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1038 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1039 spa
->spa_proc_state
= SPA_PROC_GONE
;
1040 spa
->spa_proc
= &p0
;
1041 cv_broadcast(&spa
->spa_proc_cv
);
1042 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1044 mutex_enter(&curproc
->p_lock
);
1050 * Activate an uninitialized pool.
1053 spa_activate(spa_t
*spa
, int mode
)
1055 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1057 spa
->spa_state
= POOL_STATE_ACTIVE
;
1058 spa
->spa_mode
= mode
;
1060 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1061 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1063 /* Try to create a covering process */
1064 mutex_enter(&spa
->spa_proc_lock
);
1065 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1066 ASSERT(spa
->spa_proc
== &p0
);
1069 #ifdef HAVE_SPA_THREAD
1070 /* Only create a process if we're going to be around a while. */
1071 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1072 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1074 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1075 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1076 cv_wait(&spa
->spa_proc_cv
,
1077 &spa
->spa_proc_lock
);
1079 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1080 ASSERT(spa
->spa_proc
!= &p0
);
1081 ASSERT(spa
->spa_did
!= 0);
1085 "Couldn't create process for zfs pool \"%s\"\n",
1090 #endif /* HAVE_SPA_THREAD */
1091 mutex_exit(&spa
->spa_proc_lock
);
1093 /* If we didn't create a process, we need to create our taskqs. */
1094 if (spa
->spa_proc
== &p0
) {
1095 spa_create_zio_taskqs(spa
);
1098 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1099 offsetof(vdev_t
, vdev_config_dirty_node
));
1100 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1101 offsetof(objset_t
, os_evicting_node
));
1102 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1103 offsetof(vdev_t
, vdev_state_dirty_node
));
1105 txg_list_create(&spa
->spa_vdev_txg_list
,
1106 offsetof(struct vdev
, vdev_txg_node
));
1108 avl_create(&spa
->spa_errlist_scrub
,
1109 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1110 offsetof(spa_error_entry_t
, se_avl
));
1111 avl_create(&spa
->spa_errlist_last
,
1112 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1113 offsetof(spa_error_entry_t
, se_avl
));
1117 * Opposite of spa_activate().
1120 spa_deactivate(spa_t
*spa
)
1124 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1125 ASSERT(spa
->spa_dsl_pool
== NULL
);
1126 ASSERT(spa
->spa_root_vdev
== NULL
);
1127 ASSERT(spa
->spa_async_zio_root
== NULL
);
1128 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1130 spa_evicting_os_wait(spa
);
1132 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1134 list_destroy(&spa
->spa_config_dirty_list
);
1135 list_destroy(&spa
->spa_evicting_os_list
);
1136 list_destroy(&spa
->spa_state_dirty_list
);
1138 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
1140 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1141 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1142 spa_taskqs_fini(spa
, t
, q
);
1146 metaslab_class_destroy(spa
->spa_normal_class
);
1147 spa
->spa_normal_class
= NULL
;
1149 metaslab_class_destroy(spa
->spa_log_class
);
1150 spa
->spa_log_class
= NULL
;
1153 * If this was part of an import or the open otherwise failed, we may
1154 * still have errors left in the queues. Empty them just in case.
1156 spa_errlog_drain(spa
);
1158 avl_destroy(&spa
->spa_errlist_scrub
);
1159 avl_destroy(&spa
->spa_errlist_last
);
1161 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1163 mutex_enter(&spa
->spa_proc_lock
);
1164 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1165 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1166 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1167 cv_broadcast(&spa
->spa_proc_cv
);
1168 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1169 ASSERT(spa
->spa_proc
!= &p0
);
1170 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1172 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1173 spa
->spa_proc_state
= SPA_PROC_NONE
;
1175 ASSERT(spa
->spa_proc
== &p0
);
1176 mutex_exit(&spa
->spa_proc_lock
);
1179 * We want to make sure spa_thread() has actually exited the ZFS
1180 * module, so that the module can't be unloaded out from underneath
1183 if (spa
->spa_did
!= 0) {
1184 thread_join(spa
->spa_did
);
1190 * Verify a pool configuration, and construct the vdev tree appropriately. This
1191 * will create all the necessary vdevs in the appropriate layout, with each vdev
1192 * in the CLOSED state. This will prep the pool before open/creation/import.
1193 * All vdev validation is done by the vdev_alloc() routine.
1196 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1197 uint_t id
, int atype
)
1204 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1207 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1210 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1213 if (error
== ENOENT
)
1219 return (SET_ERROR(EINVAL
));
1222 for (c
= 0; c
< children
; c
++) {
1224 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1232 ASSERT(*vdp
!= NULL
);
1238 * Opposite of spa_load().
1241 spa_unload(spa_t
*spa
)
1245 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1250 spa_async_suspend(spa
);
1255 if (spa
->spa_sync_on
) {
1256 txg_sync_stop(spa
->spa_dsl_pool
);
1257 spa
->spa_sync_on
= B_FALSE
;
1261 * Wait for any outstanding async I/O to complete.
1263 if (spa
->spa_async_zio_root
!= NULL
) {
1264 for (i
= 0; i
< max_ncpus
; i
++)
1265 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1266 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1267 spa
->spa_async_zio_root
= NULL
;
1270 bpobj_close(&spa
->spa_deferred_bpobj
);
1272 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1277 if (spa
->spa_root_vdev
)
1278 vdev_free(spa
->spa_root_vdev
);
1279 ASSERT(spa
->spa_root_vdev
== NULL
);
1282 * Close the dsl pool.
1284 if (spa
->spa_dsl_pool
) {
1285 dsl_pool_close(spa
->spa_dsl_pool
);
1286 spa
->spa_dsl_pool
= NULL
;
1287 spa
->spa_meta_objset
= NULL
;
1294 * Drop and purge level 2 cache
1296 spa_l2cache_drop(spa
);
1298 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1299 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1300 if (spa
->spa_spares
.sav_vdevs
) {
1301 kmem_free(spa
->spa_spares
.sav_vdevs
,
1302 spa
->spa_spares
.sav_count
* sizeof (void *));
1303 spa
->spa_spares
.sav_vdevs
= NULL
;
1305 if (spa
->spa_spares
.sav_config
) {
1306 nvlist_free(spa
->spa_spares
.sav_config
);
1307 spa
->spa_spares
.sav_config
= NULL
;
1309 spa
->spa_spares
.sav_count
= 0;
1311 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1312 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1313 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1315 if (spa
->spa_l2cache
.sav_vdevs
) {
1316 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1317 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1318 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1320 if (spa
->spa_l2cache
.sav_config
) {
1321 nvlist_free(spa
->spa_l2cache
.sav_config
);
1322 spa
->spa_l2cache
.sav_config
= NULL
;
1324 spa
->spa_l2cache
.sav_count
= 0;
1326 spa
->spa_async_suspended
= 0;
1328 if (spa
->spa_comment
!= NULL
) {
1329 spa_strfree(spa
->spa_comment
);
1330 spa
->spa_comment
= NULL
;
1333 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1337 * Load (or re-load) the current list of vdevs describing the active spares for
1338 * this pool. When this is called, we have some form of basic information in
1339 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1340 * then re-generate a more complete list including status information.
1343 spa_load_spares(spa_t
*spa
)
1350 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1353 * First, close and free any existing spare vdevs.
1355 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1356 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1358 /* Undo the call to spa_activate() below */
1359 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1360 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1361 spa_spare_remove(tvd
);
1366 if (spa
->spa_spares
.sav_vdevs
)
1367 kmem_free(spa
->spa_spares
.sav_vdevs
,
1368 spa
->spa_spares
.sav_count
* sizeof (void *));
1370 if (spa
->spa_spares
.sav_config
== NULL
)
1373 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1374 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1376 spa
->spa_spares
.sav_count
= (int)nspares
;
1377 spa
->spa_spares
.sav_vdevs
= NULL
;
1383 * Construct the array of vdevs, opening them to get status in the
1384 * process. For each spare, there is potentially two different vdev_t
1385 * structures associated with it: one in the list of spares (used only
1386 * for basic validation purposes) and one in the active vdev
1387 * configuration (if it's spared in). During this phase we open and
1388 * validate each vdev on the spare list. If the vdev also exists in the
1389 * active configuration, then we also mark this vdev as an active spare.
1391 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1393 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1394 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1395 VDEV_ALLOC_SPARE
) == 0);
1398 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1400 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1401 B_FALSE
)) != NULL
) {
1402 if (!tvd
->vdev_isspare
)
1406 * We only mark the spare active if we were successfully
1407 * able to load the vdev. Otherwise, importing a pool
1408 * with a bad active spare would result in strange
1409 * behavior, because multiple pool would think the spare
1410 * is actively in use.
1412 * There is a vulnerability here to an equally bizarre
1413 * circumstance, where a dead active spare is later
1414 * brought back to life (onlined or otherwise). Given
1415 * the rarity of this scenario, and the extra complexity
1416 * it adds, we ignore the possibility.
1418 if (!vdev_is_dead(tvd
))
1419 spa_spare_activate(tvd
);
1423 vd
->vdev_aux
= &spa
->spa_spares
;
1425 if (vdev_open(vd
) != 0)
1428 if (vdev_validate_aux(vd
) == 0)
1433 * Recompute the stashed list of spares, with status information
1436 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1437 DATA_TYPE_NVLIST_ARRAY
) == 0);
1439 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1441 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1442 spares
[i
] = vdev_config_generate(spa
,
1443 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1444 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1445 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1446 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1447 nvlist_free(spares
[i
]);
1448 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1452 * Load (or re-load) the current list of vdevs describing the active l2cache for
1453 * this pool. When this is called, we have some form of basic information in
1454 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1455 * then re-generate a more complete list including status information.
1456 * Devices which are already active have their details maintained, and are
1460 spa_load_l2cache(spa_t
*spa
)
1464 int i
, j
, oldnvdevs
;
1466 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1467 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1469 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1471 if (sav
->sav_config
!= NULL
) {
1472 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1473 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1474 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1480 oldvdevs
= sav
->sav_vdevs
;
1481 oldnvdevs
= sav
->sav_count
;
1482 sav
->sav_vdevs
= NULL
;
1486 * Process new nvlist of vdevs.
1488 for (i
= 0; i
< nl2cache
; i
++) {
1489 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1493 for (j
= 0; j
< oldnvdevs
; j
++) {
1495 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1497 * Retain previous vdev for add/remove ops.
1505 if (newvdevs
[i
] == NULL
) {
1509 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1510 VDEV_ALLOC_L2CACHE
) == 0);
1515 * Commit this vdev as an l2cache device,
1516 * even if it fails to open.
1518 spa_l2cache_add(vd
);
1523 spa_l2cache_activate(vd
);
1525 if (vdev_open(vd
) != 0)
1528 (void) vdev_validate_aux(vd
);
1530 if (!vdev_is_dead(vd
))
1531 l2arc_add_vdev(spa
, vd
);
1536 * Purge vdevs that were dropped
1538 for (i
= 0; i
< oldnvdevs
; i
++) {
1543 ASSERT(vd
->vdev_isl2cache
);
1545 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1546 pool
!= 0ULL && l2arc_vdev_present(vd
))
1547 l2arc_remove_vdev(vd
);
1548 vdev_clear_stats(vd
);
1554 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1556 if (sav
->sav_config
== NULL
)
1559 sav
->sav_vdevs
= newvdevs
;
1560 sav
->sav_count
= (int)nl2cache
;
1563 * Recompute the stashed list of l2cache devices, with status
1564 * information this time.
1566 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1567 DATA_TYPE_NVLIST_ARRAY
) == 0);
1569 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1570 for (i
= 0; i
< sav
->sav_count
; i
++)
1571 l2cache
[i
] = vdev_config_generate(spa
,
1572 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1573 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1574 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1576 for (i
= 0; i
< sav
->sav_count
; i
++)
1577 nvlist_free(l2cache
[i
]);
1579 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1583 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1586 char *packed
= NULL
;
1591 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1595 nvsize
= *(uint64_t *)db
->db_data
;
1596 dmu_buf_rele(db
, FTAG
);
1598 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1599 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1602 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1603 vmem_free(packed
, nvsize
);
1609 * Checks to see if the given vdev could not be opened, in which case we post a
1610 * sysevent to notify the autoreplace code that the device has been removed.
1613 spa_check_removed(vdev_t
*vd
)
1617 for (c
= 0; c
< vd
->vdev_children
; c
++)
1618 spa_check_removed(vd
->vdev_child
[c
]);
1620 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1622 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE
,
1623 vd
->vdev_spa
, vd
, NULL
, 0, 0);
1624 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_CHECK
);
1629 * Validate the current config against the MOS config
1632 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1634 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1638 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1640 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1641 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1643 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1646 * If we're doing a normal import, then build up any additional
1647 * diagnostic information about missing devices in this config.
1648 * We'll pass this up to the user for further processing.
1650 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1651 nvlist_t
**child
, *nv
;
1654 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1656 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1658 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1659 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1660 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1662 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1663 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1665 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1670 VERIFY(nvlist_add_nvlist_array(nv
,
1671 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1672 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1673 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1675 for (i
= 0; i
< idx
; i
++)
1676 nvlist_free(child
[i
]);
1679 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1683 * Compare the root vdev tree with the information we have
1684 * from the MOS config (mrvd). Check each top-level vdev
1685 * with the corresponding MOS config top-level (mtvd).
1687 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1688 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1689 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1692 * Resolve any "missing" vdevs in the current configuration.
1693 * If we find that the MOS config has more accurate information
1694 * about the top-level vdev then use that vdev instead.
1696 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1697 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1699 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1703 * Device specific actions.
1705 if (mtvd
->vdev_islog
) {
1706 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1709 * XXX - once we have 'readonly' pool
1710 * support we should be able to handle
1711 * missing data devices by transitioning
1712 * the pool to readonly.
1718 * Swap the missing vdev with the data we were
1719 * able to obtain from the MOS config.
1721 vdev_remove_child(rvd
, tvd
);
1722 vdev_remove_child(mrvd
, mtvd
);
1724 vdev_add_child(rvd
, mtvd
);
1725 vdev_add_child(mrvd
, tvd
);
1727 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1729 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1732 } else if (mtvd
->vdev_islog
) {
1734 * Load the slog device's state from the MOS config
1735 * since it's possible that the label does not
1736 * contain the most up-to-date information.
1738 vdev_load_log_state(tvd
, mtvd
);
1743 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1746 * Ensure we were able to validate the config.
1748 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1752 * Check for missing log devices
1755 spa_check_logs(spa_t
*spa
)
1757 boolean_t rv
= B_FALSE
;
1759 switch (spa
->spa_log_state
) {
1762 case SPA_LOG_MISSING
:
1763 /* need to recheck in case slog has been restored */
1764 case SPA_LOG_UNKNOWN
:
1765 rv
= (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
,
1766 NULL
, DS_FIND_CHILDREN
) != 0);
1768 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1775 spa_passivate_log(spa_t
*spa
)
1777 vdev_t
*rvd
= spa
->spa_root_vdev
;
1778 boolean_t slog_found
= B_FALSE
;
1781 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1783 if (!spa_has_slogs(spa
))
1786 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1787 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1788 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1790 if (tvd
->vdev_islog
) {
1791 metaslab_group_passivate(mg
);
1792 slog_found
= B_TRUE
;
1796 return (slog_found
);
1800 spa_activate_log(spa_t
*spa
)
1802 vdev_t
*rvd
= spa
->spa_root_vdev
;
1805 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1807 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1808 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1809 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1811 if (tvd
->vdev_islog
)
1812 metaslab_group_activate(mg
);
1817 spa_offline_log(spa_t
*spa
)
1821 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1822 NULL
, DS_FIND_CHILDREN
);
1825 * We successfully offlined the log device, sync out the
1826 * current txg so that the "stubby" block can be removed
1829 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1835 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1839 for (i
= 0; i
< sav
->sav_count
; i
++)
1840 spa_check_removed(sav
->sav_vdevs
[i
]);
1844 spa_claim_notify(zio_t
*zio
)
1846 spa_t
*spa
= zio
->io_spa
;
1851 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1852 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1853 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1854 mutex_exit(&spa
->spa_props_lock
);
1857 typedef struct spa_load_error
{
1858 uint64_t sle_meta_count
;
1859 uint64_t sle_data_count
;
1863 spa_load_verify_done(zio_t
*zio
)
1865 blkptr_t
*bp
= zio
->io_bp
;
1866 spa_load_error_t
*sle
= zio
->io_private
;
1867 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1868 int error
= zio
->io_error
;
1869 spa_t
*spa
= zio
->io_spa
;
1872 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1873 type
!= DMU_OT_INTENT_LOG
)
1874 atomic_add_64(&sle
->sle_meta_count
, 1);
1876 atomic_add_64(&sle
->sle_data_count
, 1);
1878 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1880 mutex_enter(&spa
->spa_scrub_lock
);
1881 spa
->spa_scrub_inflight
--;
1882 cv_broadcast(&spa
->spa_scrub_io_cv
);
1883 mutex_exit(&spa
->spa_scrub_lock
);
1887 * Maximum number of concurrent scrub i/os to create while verifying
1888 * a pool while importing it.
1890 int spa_load_verify_maxinflight
= 10000;
1891 int spa_load_verify_metadata
= B_TRUE
;
1892 int spa_load_verify_data
= B_TRUE
;
1896 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1897 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1903 if (BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
1906 * Note: normally this routine will not be called if
1907 * spa_load_verify_metadata is not set. However, it may be useful
1908 * to manually set the flag after the traversal has begun.
1910 if (!spa_load_verify_metadata
)
1912 if (BP_GET_BUFC_TYPE(bp
) == ARC_BUFC_DATA
&& !spa_load_verify_data
)
1916 size
= BP_GET_PSIZE(bp
);
1917 data
= zio_data_buf_alloc(size
);
1919 mutex_enter(&spa
->spa_scrub_lock
);
1920 while (spa
->spa_scrub_inflight
>= spa_load_verify_maxinflight
)
1921 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
1922 spa
->spa_scrub_inflight
++;
1923 mutex_exit(&spa
->spa_scrub_lock
);
1925 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1926 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1927 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1928 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1933 spa_load_verify(spa_t
*spa
)
1936 spa_load_error_t sle
= { 0 };
1937 zpool_rewind_policy_t policy
;
1938 boolean_t verify_ok
= B_FALSE
;
1941 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1943 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1946 rio
= zio_root(spa
, NULL
, &sle
,
1947 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1949 if (spa_load_verify_metadata
) {
1950 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1951 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
,
1952 spa_load_verify_cb
, rio
);
1955 (void) zio_wait(rio
);
1957 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1958 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1960 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1961 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1965 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1966 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1968 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1969 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1970 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1971 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1972 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1973 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1974 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1976 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1980 if (error
!= ENXIO
&& error
!= EIO
)
1981 error
= SET_ERROR(EIO
);
1985 return (verify_ok
? 0 : EIO
);
1989 * Find a value in the pool props object.
1992 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
1994 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
1995 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
1999 * Find a value in the pool directory object.
2002 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2004 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2005 name
, sizeof (uint64_t), 1, val
));
2009 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2011 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2016 * Fix up config after a partly-completed split. This is done with the
2017 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2018 * pool have that entry in their config, but only the splitting one contains
2019 * a list of all the guids of the vdevs that are being split off.
2021 * This function determines what to do with that list: either rejoin
2022 * all the disks to the pool, or complete the splitting process. To attempt
2023 * the rejoin, each disk that is offlined is marked online again, and
2024 * we do a reopen() call. If the vdev label for every disk that was
2025 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2026 * then we call vdev_split() on each disk, and complete the split.
2028 * Otherwise we leave the config alone, with all the vdevs in place in
2029 * the original pool.
2032 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2039 boolean_t attempt_reopen
;
2041 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2044 /* check that the config is complete */
2045 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2046 &glist
, &gcount
) != 0)
2049 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2051 /* attempt to online all the vdevs & validate */
2052 attempt_reopen
= B_TRUE
;
2053 for (i
= 0; i
< gcount
; i
++) {
2054 if (glist
[i
] == 0) /* vdev is hole */
2057 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2058 if (vd
[i
] == NULL
) {
2060 * Don't bother attempting to reopen the disks;
2061 * just do the split.
2063 attempt_reopen
= B_FALSE
;
2065 /* attempt to re-online it */
2066 vd
[i
]->vdev_offline
= B_FALSE
;
2070 if (attempt_reopen
) {
2071 vdev_reopen(spa
->spa_root_vdev
);
2073 /* check each device to see what state it's in */
2074 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2075 if (vd
[i
] != NULL
&&
2076 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2083 * If every disk has been moved to the new pool, or if we never
2084 * even attempted to look at them, then we split them off for
2087 if (!attempt_reopen
|| gcount
== extracted
) {
2088 for (i
= 0; i
< gcount
; i
++)
2091 vdev_reopen(spa
->spa_root_vdev
);
2094 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2098 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2099 boolean_t mosconfig
)
2101 nvlist_t
*config
= spa
->spa_config
;
2102 char *ereport
= FM_EREPORT_ZFS_POOL
;
2108 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2109 return (SET_ERROR(EINVAL
));
2111 ASSERT(spa
->spa_comment
== NULL
);
2112 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2113 spa
->spa_comment
= spa_strdup(comment
);
2116 * Versioning wasn't explicitly added to the label until later, so if
2117 * it's not present treat it as the initial version.
2119 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2120 &spa
->spa_ubsync
.ub_version
) != 0)
2121 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2123 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2124 &spa
->spa_config_txg
);
2126 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2127 spa_guid_exists(pool_guid
, 0)) {
2128 error
= SET_ERROR(EEXIST
);
2130 spa
->spa_config_guid
= pool_guid
;
2132 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2134 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2138 nvlist_free(spa
->spa_load_info
);
2139 spa
->spa_load_info
= fnvlist_alloc();
2141 gethrestime(&spa
->spa_loaded_ts
);
2142 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2143 mosconfig
, &ereport
);
2147 * Don't count references from objsets that are already closed
2148 * and are making their way through the eviction process.
2150 spa_evicting_os_wait(spa
);
2151 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2153 if (error
!= EEXIST
) {
2154 spa
->spa_loaded_ts
.tv_sec
= 0;
2155 spa
->spa_loaded_ts
.tv_nsec
= 0;
2157 if (error
!= EBADF
) {
2158 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2161 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2168 * Load an existing storage pool, using the pool's builtin spa_config as a
2169 * source of configuration information.
2171 __attribute__((always_inline
))
2173 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2174 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2178 nvlist_t
*nvroot
= NULL
;
2181 uberblock_t
*ub
= &spa
->spa_uberblock
;
2182 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2183 int orig_mode
= spa
->spa_mode
;
2186 boolean_t missing_feat_write
= B_FALSE
;
2189 * If this is an untrusted config, access the pool in read-only mode.
2190 * This prevents things like resilvering recently removed devices.
2193 spa
->spa_mode
= FREAD
;
2195 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2197 spa
->spa_load_state
= state
;
2199 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2200 return (SET_ERROR(EINVAL
));
2202 parse
= (type
== SPA_IMPORT_EXISTING
?
2203 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2206 * Create "The Godfather" zio to hold all async IOs
2208 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2210 for (i
= 0; i
< max_ncpus
; i
++) {
2211 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2212 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2213 ZIO_FLAG_GODFATHER
);
2217 * Parse the configuration into a vdev tree. We explicitly set the
2218 * value that will be returned by spa_version() since parsing the
2219 * configuration requires knowing the version number.
2221 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2222 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2223 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2228 ASSERT(spa
->spa_root_vdev
== rvd
);
2230 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2231 ASSERT(spa_guid(spa
) == pool_guid
);
2235 * Try to open all vdevs, loading each label in the process.
2237 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2238 error
= vdev_open(rvd
);
2239 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2244 * We need to validate the vdev labels against the configuration that
2245 * we have in hand, which is dependent on the setting of mosconfig. If
2246 * mosconfig is true then we're validating the vdev labels based on
2247 * that config. Otherwise, we're validating against the cached config
2248 * (zpool.cache) that was read when we loaded the zfs module, and then
2249 * later we will recursively call spa_load() and validate against
2252 * If we're assembling a new pool that's been split off from an
2253 * existing pool, the labels haven't yet been updated so we skip
2254 * validation for now.
2256 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2257 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2258 error
= vdev_validate(rvd
, mosconfig
);
2259 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2264 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2265 return (SET_ERROR(ENXIO
));
2269 * Find the best uberblock.
2271 vdev_uberblock_load(rvd
, ub
, &label
);
2274 * If we weren't able to find a single valid uberblock, return failure.
2276 if (ub
->ub_txg
== 0) {
2278 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2282 * If the pool has an unsupported version we can't open it.
2284 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2286 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2289 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2293 * If we weren't able to find what's necessary for reading the
2294 * MOS in the label, return failure.
2296 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2297 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2299 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2304 * Update our in-core representation with the definitive values
2307 nvlist_free(spa
->spa_label_features
);
2308 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2314 * Look through entries in the label nvlist's features_for_read. If
2315 * there is a feature listed there which we don't understand then we
2316 * cannot open a pool.
2318 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2319 nvlist_t
*unsup_feat
;
2322 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2325 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2327 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2328 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2329 VERIFY(nvlist_add_string(unsup_feat
,
2330 nvpair_name(nvp
), "") == 0);
2334 if (!nvlist_empty(unsup_feat
)) {
2335 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2336 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2337 nvlist_free(unsup_feat
);
2338 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2342 nvlist_free(unsup_feat
);
2346 * If the vdev guid sum doesn't match the uberblock, we have an
2347 * incomplete configuration. We first check to see if the pool
2348 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2349 * If it is, defer the vdev_guid_sum check till later so we
2350 * can handle missing vdevs.
2352 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2353 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2354 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2355 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2357 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2358 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2359 spa_try_repair(spa
, config
);
2360 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2361 nvlist_free(spa
->spa_config_splitting
);
2362 spa
->spa_config_splitting
= NULL
;
2366 * Initialize internal SPA structures.
2368 spa
->spa_state
= POOL_STATE_ACTIVE
;
2369 spa
->spa_ubsync
= spa
->spa_uberblock
;
2370 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2371 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2372 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2373 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2374 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2375 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2377 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2379 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2380 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2382 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2383 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2385 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2386 boolean_t missing_feat_read
= B_FALSE
;
2387 nvlist_t
*unsup_feat
, *enabled_feat
;
2390 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2391 &spa
->spa_feat_for_read_obj
) != 0) {
2392 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2395 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2396 &spa
->spa_feat_for_write_obj
) != 0) {
2397 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2400 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2401 &spa
->spa_feat_desc_obj
) != 0) {
2402 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2405 enabled_feat
= fnvlist_alloc();
2406 unsup_feat
= fnvlist_alloc();
2408 if (!spa_features_check(spa
, B_FALSE
,
2409 unsup_feat
, enabled_feat
))
2410 missing_feat_read
= B_TRUE
;
2412 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2413 if (!spa_features_check(spa
, B_TRUE
,
2414 unsup_feat
, enabled_feat
)) {
2415 missing_feat_write
= B_TRUE
;
2419 fnvlist_add_nvlist(spa
->spa_load_info
,
2420 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2422 if (!nvlist_empty(unsup_feat
)) {
2423 fnvlist_add_nvlist(spa
->spa_load_info
,
2424 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2427 fnvlist_free(enabled_feat
);
2428 fnvlist_free(unsup_feat
);
2430 if (!missing_feat_read
) {
2431 fnvlist_add_boolean(spa
->spa_load_info
,
2432 ZPOOL_CONFIG_CAN_RDONLY
);
2436 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2437 * twofold: to determine whether the pool is available for
2438 * import in read-write mode and (if it is not) whether the
2439 * pool is available for import in read-only mode. If the pool
2440 * is available for import in read-write mode, it is displayed
2441 * as available in userland; if it is not available for import
2442 * in read-only mode, it is displayed as unavailable in
2443 * userland. If the pool is available for import in read-only
2444 * mode but not read-write mode, it is displayed as unavailable
2445 * in userland with a special note that the pool is actually
2446 * available for open in read-only mode.
2448 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2449 * missing a feature for write, we must first determine whether
2450 * the pool can be opened read-only before returning to
2451 * userland in order to know whether to display the
2452 * abovementioned note.
2454 if (missing_feat_read
|| (missing_feat_write
&&
2455 spa_writeable(spa
))) {
2456 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2461 * Load refcounts for ZFS features from disk into an in-memory
2462 * cache during SPA initialization.
2464 for (i
= 0; i
< SPA_FEATURES
; i
++) {
2467 error
= feature_get_refcount_from_disk(spa
,
2468 &spa_feature_table
[i
], &refcount
);
2470 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2471 } else if (error
== ENOTSUP
) {
2472 spa
->spa_feat_refcount_cache
[i
] =
2473 SPA_FEATURE_DISABLED
;
2475 return (spa_vdev_err(rvd
,
2476 VDEV_AUX_CORRUPT_DATA
, EIO
));
2481 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2482 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2483 &spa
->spa_feat_enabled_txg_obj
) != 0)
2484 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2487 spa
->spa_is_initializing
= B_TRUE
;
2488 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2489 spa
->spa_is_initializing
= B_FALSE
;
2491 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2495 nvlist_t
*policy
= NULL
, *nvconfig
;
2497 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2498 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2500 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2501 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2503 unsigned long myhostid
= 0;
2505 VERIFY(nvlist_lookup_string(nvconfig
,
2506 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2509 myhostid
= zone_get_hostid(NULL
);
2512 * We're emulating the system's hostid in userland, so
2513 * we can't use zone_get_hostid().
2515 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2516 #endif /* _KERNEL */
2517 if (hostid
!= 0 && myhostid
!= 0 &&
2518 hostid
!= myhostid
) {
2519 nvlist_free(nvconfig
);
2520 cmn_err(CE_WARN
, "pool '%s' could not be "
2521 "loaded as it was last accessed by another "
2522 "system (host: %s hostid: 0x%lx). See: "
2523 "http://zfsonlinux.org/msg/ZFS-8000-EY",
2524 spa_name(spa
), hostname
,
2525 (unsigned long)hostid
);
2526 return (SET_ERROR(EBADF
));
2529 if (nvlist_lookup_nvlist(spa
->spa_config
,
2530 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2531 VERIFY(nvlist_add_nvlist(nvconfig
,
2532 ZPOOL_REWIND_POLICY
, policy
) == 0);
2534 spa_config_set(spa
, nvconfig
);
2536 spa_deactivate(spa
);
2537 spa_activate(spa
, orig_mode
);
2539 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2542 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2543 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2544 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2546 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2549 * Load the bit that tells us to use the new accounting function
2550 * (raid-z deflation). If we have an older pool, this will not
2553 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2554 if (error
!= 0 && error
!= ENOENT
)
2555 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2557 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2558 &spa
->spa_creation_version
);
2559 if (error
!= 0 && error
!= ENOENT
)
2560 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2563 * Load the persistent error log. If we have an older pool, this will
2566 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2567 if (error
!= 0 && error
!= ENOENT
)
2568 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2570 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2571 &spa
->spa_errlog_scrub
);
2572 if (error
!= 0 && error
!= ENOENT
)
2573 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2576 * Load the history object. If we have an older pool, this
2577 * will not be present.
2579 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2580 if (error
!= 0 && error
!= ENOENT
)
2581 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2584 * If we're assembling the pool from the split-off vdevs of
2585 * an existing pool, we don't want to attach the spares & cache
2590 * Load any hot spares for this pool.
2592 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2593 if (error
!= 0 && error
!= ENOENT
)
2594 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2595 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2596 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2597 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2598 &spa
->spa_spares
.sav_config
) != 0)
2599 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2601 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2602 spa_load_spares(spa
);
2603 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2604 } else if (error
== 0) {
2605 spa
->spa_spares
.sav_sync
= B_TRUE
;
2609 * Load any level 2 ARC devices for this pool.
2611 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2612 &spa
->spa_l2cache
.sav_object
);
2613 if (error
!= 0 && error
!= ENOENT
)
2614 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2615 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2616 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2617 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2618 &spa
->spa_l2cache
.sav_config
) != 0)
2619 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2621 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2622 spa_load_l2cache(spa
);
2623 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2624 } else if (error
== 0) {
2625 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2628 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2630 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2631 if (error
&& error
!= ENOENT
)
2632 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2635 uint64_t autoreplace
= 0;
2637 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2638 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2639 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2640 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2641 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2642 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2643 &spa
->spa_dedup_ditto
);
2645 spa
->spa_autoreplace
= (autoreplace
!= 0);
2649 * If the 'autoreplace' property is set, then post a resource notifying
2650 * the ZFS DE that it should not issue any faults for unopenable
2651 * devices. We also iterate over the vdevs, and post a sysevent for any
2652 * unopenable vdevs so that the normal autoreplace handler can take
2655 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2656 spa_check_removed(spa
->spa_root_vdev
);
2658 * For the import case, this is done in spa_import(), because
2659 * at this point we're using the spare definitions from
2660 * the MOS config, not necessarily from the userland config.
2662 if (state
!= SPA_LOAD_IMPORT
) {
2663 spa_aux_check_removed(&spa
->spa_spares
);
2664 spa_aux_check_removed(&spa
->spa_l2cache
);
2669 * Load the vdev state for all toplevel vdevs.
2674 * Propagate the leaf DTLs we just loaded all the way up the tree.
2676 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2677 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2678 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2681 * Load the DDTs (dedup tables).
2683 error
= ddt_load(spa
);
2685 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2687 spa_update_dspace(spa
);
2690 * Validate the config, using the MOS config to fill in any
2691 * information which might be missing. If we fail to validate
2692 * the config then declare the pool unfit for use. If we're
2693 * assembling a pool from a split, the log is not transferred
2696 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2699 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2700 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2702 if (!spa_config_valid(spa
, nvconfig
)) {
2703 nvlist_free(nvconfig
);
2704 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2707 nvlist_free(nvconfig
);
2710 * Now that we've validated the config, check the state of the
2711 * root vdev. If it can't be opened, it indicates one or
2712 * more toplevel vdevs are faulted.
2714 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2715 return (SET_ERROR(ENXIO
));
2717 if (spa_check_logs(spa
)) {
2718 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2719 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2723 if (missing_feat_write
) {
2724 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2727 * At this point, we know that we can open the pool in
2728 * read-only mode but not read-write mode. We now have enough
2729 * information and can return to userland.
2731 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2735 * We've successfully opened the pool, verify that we're ready
2736 * to start pushing transactions.
2738 if (state
!= SPA_LOAD_TRYIMPORT
) {
2739 if ((error
= spa_load_verify(spa
)))
2740 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2744 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2745 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2747 int need_update
= B_FALSE
;
2750 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2753 * Claim log blocks that haven't been committed yet.
2754 * This must all happen in a single txg.
2755 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2756 * invoked from zil_claim_log_block()'s i/o done callback.
2757 * Price of rollback is that we abandon the log.
2759 spa
->spa_claiming
= B_TRUE
;
2761 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
2762 spa_first_txg(spa
));
2763 (void) dmu_objset_find(spa_name(spa
),
2764 zil_claim
, tx
, DS_FIND_CHILDREN
);
2767 spa
->spa_claiming
= B_FALSE
;
2769 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2770 spa
->spa_sync_on
= B_TRUE
;
2771 txg_sync_start(spa
->spa_dsl_pool
);
2774 * Wait for all claims to sync. We sync up to the highest
2775 * claimed log block birth time so that claimed log blocks
2776 * don't appear to be from the future. spa_claim_max_txg
2777 * will have been set for us by either zil_check_log_chain()
2778 * (invoked from spa_check_logs()) or zil_claim() above.
2780 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2783 * If the config cache is stale, or we have uninitialized
2784 * metaslabs (see spa_vdev_add()), then update the config.
2786 * If this is a verbatim import, trust the current
2787 * in-core spa_config and update the disk labels.
2789 if (config_cache_txg
!= spa
->spa_config_txg
||
2790 state
== SPA_LOAD_IMPORT
||
2791 state
== SPA_LOAD_RECOVER
||
2792 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2793 need_update
= B_TRUE
;
2795 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2796 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2797 need_update
= B_TRUE
;
2800 * Update the config cache asychronously in case we're the
2801 * root pool, in which case the config cache isn't writable yet.
2804 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2807 * Check all DTLs to see if anything needs resilvering.
2809 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2810 vdev_resilver_needed(rvd
, NULL
, NULL
))
2811 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2814 * Log the fact that we booted up (so that we can detect if
2815 * we rebooted in the middle of an operation).
2817 spa_history_log_version(spa
, "open");
2820 * Delete any inconsistent datasets.
2822 (void) dmu_objset_find(spa_name(spa
),
2823 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2826 * Clean up any stale temporary dataset userrefs.
2828 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2835 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2837 int mode
= spa
->spa_mode
;
2840 spa_deactivate(spa
);
2842 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
2844 spa_activate(spa
, mode
);
2845 spa_async_suspend(spa
);
2847 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2851 * If spa_load() fails this function will try loading prior txg's. If
2852 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2853 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2854 * function will not rewind the pool and will return the same error as
2858 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2859 uint64_t max_request
, int rewind_flags
)
2861 nvlist_t
*loadinfo
= NULL
;
2862 nvlist_t
*config
= NULL
;
2863 int load_error
, rewind_error
;
2864 uint64_t safe_rewind_txg
;
2867 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2868 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2869 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2871 spa
->spa_load_max_txg
= max_request
;
2872 if (max_request
!= UINT64_MAX
)
2873 spa
->spa_extreme_rewind
= B_TRUE
;
2876 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2878 if (load_error
== 0)
2881 if (spa
->spa_root_vdev
!= NULL
)
2882 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2884 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2885 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2887 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2888 nvlist_free(config
);
2889 return (load_error
);
2892 if (state
== SPA_LOAD_RECOVER
) {
2893 /* Price of rolling back is discarding txgs, including log */
2894 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2897 * If we aren't rolling back save the load info from our first
2898 * import attempt so that we can restore it after attempting
2901 loadinfo
= spa
->spa_load_info
;
2902 spa
->spa_load_info
= fnvlist_alloc();
2905 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2906 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2907 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2908 TXG_INITIAL
: safe_rewind_txg
;
2911 * Continue as long as we're finding errors, we're still within
2912 * the acceptable rewind range, and we're still finding uberblocks
2914 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2915 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2916 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2917 spa
->spa_extreme_rewind
= B_TRUE
;
2918 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2921 spa
->spa_extreme_rewind
= B_FALSE
;
2922 spa
->spa_load_max_txg
= UINT64_MAX
;
2924 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2925 spa_config_set(spa
, config
);
2927 if (state
== SPA_LOAD_RECOVER
) {
2928 ASSERT3P(loadinfo
, ==, NULL
);
2929 return (rewind_error
);
2931 /* Store the rewind info as part of the initial load info */
2932 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
2933 spa
->spa_load_info
);
2935 /* Restore the initial load info */
2936 fnvlist_free(spa
->spa_load_info
);
2937 spa
->spa_load_info
= loadinfo
;
2939 return (load_error
);
2946 * The import case is identical to an open except that the configuration is sent
2947 * down from userland, instead of grabbed from the configuration cache. For the
2948 * case of an open, the pool configuration will exist in the
2949 * POOL_STATE_UNINITIALIZED state.
2951 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2952 * the same time open the pool, without having to keep around the spa_t in some
2956 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2960 spa_load_state_t state
= SPA_LOAD_OPEN
;
2962 int locked
= B_FALSE
;
2963 int firstopen
= B_FALSE
;
2968 * As disgusting as this is, we need to support recursive calls to this
2969 * function because dsl_dir_open() is called during spa_load(), and ends
2970 * up calling spa_open() again. The real fix is to figure out how to
2971 * avoid dsl_dir_open() calling this in the first place.
2973 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2974 mutex_enter(&spa_namespace_lock
);
2978 if ((spa
= spa_lookup(pool
)) == NULL
) {
2980 mutex_exit(&spa_namespace_lock
);
2981 return (SET_ERROR(ENOENT
));
2984 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
2985 zpool_rewind_policy_t policy
;
2989 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
2991 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
2992 state
= SPA_LOAD_RECOVER
;
2994 spa_activate(spa
, spa_mode_global
);
2996 if (state
!= SPA_LOAD_RECOVER
)
2997 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
2999 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3000 policy
.zrp_request
);
3002 if (error
== EBADF
) {
3004 * If vdev_validate() returns failure (indicated by
3005 * EBADF), it indicates that one of the vdevs indicates
3006 * that the pool has been exported or destroyed. If
3007 * this is the case, the config cache is out of sync and
3008 * we should remove the pool from the namespace.
3011 spa_deactivate(spa
);
3012 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3015 mutex_exit(&spa_namespace_lock
);
3016 return (SET_ERROR(ENOENT
));
3021 * We can't open the pool, but we still have useful
3022 * information: the state of each vdev after the
3023 * attempted vdev_open(). Return this to the user.
3025 if (config
!= NULL
&& spa
->spa_config
) {
3026 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3028 VERIFY(nvlist_add_nvlist(*config
,
3029 ZPOOL_CONFIG_LOAD_INFO
,
3030 spa
->spa_load_info
) == 0);
3033 spa_deactivate(spa
);
3034 spa
->spa_last_open_failed
= error
;
3036 mutex_exit(&spa_namespace_lock
);
3042 spa_open_ref(spa
, tag
);
3045 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3048 * If we've recovered the pool, pass back any information we
3049 * gathered while doing the load.
3051 if (state
== SPA_LOAD_RECOVER
) {
3052 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3053 spa
->spa_load_info
) == 0);
3057 spa
->spa_last_open_failed
= 0;
3058 spa
->spa_last_ubsync_txg
= 0;
3059 spa
->spa_load_txg
= 0;
3060 mutex_exit(&spa_namespace_lock
);
3065 zvol_create_minors(spa
->spa_name
);
3074 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3077 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3081 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3083 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3087 * Lookup the given spa_t, incrementing the inject count in the process,
3088 * preventing it from being exported or destroyed.
3091 spa_inject_addref(char *name
)
3095 mutex_enter(&spa_namespace_lock
);
3096 if ((spa
= spa_lookup(name
)) == NULL
) {
3097 mutex_exit(&spa_namespace_lock
);
3100 spa
->spa_inject_ref
++;
3101 mutex_exit(&spa_namespace_lock
);
3107 spa_inject_delref(spa_t
*spa
)
3109 mutex_enter(&spa_namespace_lock
);
3110 spa
->spa_inject_ref
--;
3111 mutex_exit(&spa_namespace_lock
);
3115 * Add spares device information to the nvlist.
3118 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3128 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3130 if (spa
->spa_spares
.sav_count
== 0)
3133 VERIFY(nvlist_lookup_nvlist(config
,
3134 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3135 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3136 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3138 VERIFY(nvlist_add_nvlist_array(nvroot
,
3139 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3140 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3141 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3144 * Go through and find any spares which have since been
3145 * repurposed as an active spare. If this is the case, update
3146 * their status appropriately.
3148 for (i
= 0; i
< nspares
; i
++) {
3149 VERIFY(nvlist_lookup_uint64(spares
[i
],
3150 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3151 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3153 VERIFY(nvlist_lookup_uint64_array(
3154 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3155 (uint64_t **)&vs
, &vsc
) == 0);
3156 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3157 vs
->vs_aux
= VDEV_AUX_SPARED
;
3164 * Add l2cache device information to the nvlist, including vdev stats.
3167 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3170 uint_t i
, j
, nl2cache
;
3177 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3179 if (spa
->spa_l2cache
.sav_count
== 0)
3182 VERIFY(nvlist_lookup_nvlist(config
,
3183 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3184 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3185 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3186 if (nl2cache
!= 0) {
3187 VERIFY(nvlist_add_nvlist_array(nvroot
,
3188 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3189 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3190 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3193 * Update level 2 cache device stats.
3196 for (i
= 0; i
< nl2cache
; i
++) {
3197 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3198 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3201 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3203 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3204 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3210 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3211 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3213 vdev_get_stats(vd
, vs
);
3219 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3224 if (spa
->spa_feat_for_read_obj
!= 0) {
3225 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3226 spa
->spa_feat_for_read_obj
);
3227 zap_cursor_retrieve(&zc
, &za
) == 0;
3228 zap_cursor_advance(&zc
)) {
3229 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3230 za
.za_num_integers
== 1);
3231 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3232 za
.za_first_integer
));
3234 zap_cursor_fini(&zc
);
3237 if (spa
->spa_feat_for_write_obj
!= 0) {
3238 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3239 spa
->spa_feat_for_write_obj
);
3240 zap_cursor_retrieve(&zc
, &za
) == 0;
3241 zap_cursor_advance(&zc
)) {
3242 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3243 za
.za_num_integers
== 1);
3244 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3245 za
.za_first_integer
));
3247 zap_cursor_fini(&zc
);
3252 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3256 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3257 zfeature_info_t feature
= spa_feature_table
[i
];
3260 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3263 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3268 * Store a list of pool features and their reference counts in the
3271 * The first time this is called on a spa, allocate a new nvlist, fetch
3272 * the pool features and reference counts from disk, then save the list
3273 * in the spa. In subsequent calls on the same spa use the saved nvlist
3274 * and refresh its values from the cached reference counts. This
3275 * ensures we don't block here on I/O on a suspended pool so 'zpool
3276 * clear' can resume the pool.
3279 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3283 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3285 mutex_enter(&spa
->spa_feat_stats_lock
);
3286 features
= spa
->spa_feat_stats
;
3288 if (features
!= NULL
) {
3289 spa_feature_stats_from_cache(spa
, features
);
3291 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3292 spa
->spa_feat_stats
= features
;
3293 spa_feature_stats_from_disk(spa
, features
);
3296 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3299 mutex_exit(&spa
->spa_feat_stats_lock
);
3303 spa_get_stats(const char *name
, nvlist_t
**config
,
3304 char *altroot
, size_t buflen
)
3310 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3314 * This still leaves a window of inconsistency where the spares
3315 * or l2cache devices could change and the config would be
3316 * self-inconsistent.
3318 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3320 if (*config
!= NULL
) {
3321 uint64_t loadtimes
[2];
3323 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3324 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3325 VERIFY(nvlist_add_uint64_array(*config
,
3326 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3328 VERIFY(nvlist_add_uint64(*config
,
3329 ZPOOL_CONFIG_ERRCOUNT
,
3330 spa_get_errlog_size(spa
)) == 0);
3332 if (spa_suspended(spa
))
3333 VERIFY(nvlist_add_uint64(*config
,
3334 ZPOOL_CONFIG_SUSPENDED
,
3335 spa
->spa_failmode
) == 0);
3337 spa_add_spares(spa
, *config
);
3338 spa_add_l2cache(spa
, *config
);
3339 spa_add_feature_stats(spa
, *config
);
3344 * We want to get the alternate root even for faulted pools, so we cheat
3345 * and call spa_lookup() directly.
3349 mutex_enter(&spa_namespace_lock
);
3350 spa
= spa_lookup(name
);
3352 spa_altroot(spa
, altroot
, buflen
);
3356 mutex_exit(&spa_namespace_lock
);
3358 spa_altroot(spa
, altroot
, buflen
);
3363 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3364 spa_close(spa
, FTAG
);
3371 * Validate that the auxiliary device array is well formed. We must have an
3372 * array of nvlists, each which describes a valid leaf vdev. If this is an
3373 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3374 * specified, as long as they are well-formed.
3377 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3378 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3379 vdev_labeltype_t label
)
3386 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3389 * It's acceptable to have no devs specified.
3391 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3395 return (SET_ERROR(EINVAL
));
3398 * Make sure the pool is formatted with a version that supports this
3401 if (spa_version(spa
) < version
)
3402 return (SET_ERROR(ENOTSUP
));
3405 * Set the pending device list so we correctly handle device in-use
3408 sav
->sav_pending
= dev
;
3409 sav
->sav_npending
= ndev
;
3411 for (i
= 0; i
< ndev
; i
++) {
3412 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3416 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3418 error
= SET_ERROR(EINVAL
);
3423 * The L2ARC currently only supports disk devices in
3424 * kernel context. For user-level testing, we allow it.
3427 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3428 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3429 error
= SET_ERROR(ENOTBLK
);
3436 if ((error
= vdev_open(vd
)) == 0 &&
3437 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3438 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3439 vd
->vdev_guid
) == 0);
3445 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3452 sav
->sav_pending
= NULL
;
3453 sav
->sav_npending
= 0;
3458 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3462 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3464 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3465 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3466 VDEV_LABEL_SPARE
)) != 0) {
3470 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3471 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3472 VDEV_LABEL_L2CACHE
));
3476 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3481 if (sav
->sav_config
!= NULL
) {
3487 * Generate new dev list by concatentating with the
3490 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3491 &olddevs
, &oldndevs
) == 0);
3493 newdevs
= kmem_alloc(sizeof (void *) *
3494 (ndevs
+ oldndevs
), KM_SLEEP
);
3495 for (i
= 0; i
< oldndevs
; i
++)
3496 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3498 for (i
= 0; i
< ndevs
; i
++)
3499 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3502 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3503 DATA_TYPE_NVLIST_ARRAY
) == 0);
3505 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3506 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3507 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3508 nvlist_free(newdevs
[i
]);
3509 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3512 * Generate a new dev list.
3514 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3516 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3522 * Stop and drop level 2 ARC devices
3525 spa_l2cache_drop(spa_t
*spa
)
3529 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3531 for (i
= 0; i
< sav
->sav_count
; i
++) {
3534 vd
= sav
->sav_vdevs
[i
];
3537 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3538 pool
!= 0ULL && l2arc_vdev_present(vd
))
3539 l2arc_remove_vdev(vd
);
3547 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3551 char *altroot
= NULL
;
3556 uint64_t txg
= TXG_INITIAL
;
3557 nvlist_t
**spares
, **l2cache
;
3558 uint_t nspares
, nl2cache
;
3559 uint64_t version
, obj
;
3560 boolean_t has_features
;
3566 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
3567 poolname
= (char *)pool
;
3570 * If this pool already exists, return failure.
3572 mutex_enter(&spa_namespace_lock
);
3573 if (spa_lookup(poolname
) != NULL
) {
3574 mutex_exit(&spa_namespace_lock
);
3575 return (SET_ERROR(EEXIST
));
3579 * Allocate a new spa_t structure.
3581 nvl
= fnvlist_alloc();
3582 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
3583 (void) nvlist_lookup_string(props
,
3584 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3585 spa
= spa_add(poolname
, nvl
, altroot
);
3587 spa_activate(spa
, spa_mode_global
);
3589 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3590 spa_deactivate(spa
);
3592 mutex_exit(&spa_namespace_lock
);
3597 * Temporary pool names should never be written to disk.
3599 if (poolname
!= pool
)
3600 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
3602 has_features
= B_FALSE
;
3603 for (elem
= nvlist_next_nvpair(props
, NULL
);
3604 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3605 if (zpool_prop_feature(nvpair_name(elem
)))
3606 has_features
= B_TRUE
;
3609 if (has_features
|| nvlist_lookup_uint64(props
,
3610 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3611 version
= SPA_VERSION
;
3613 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3615 spa
->spa_first_txg
= txg
;
3616 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3617 spa
->spa_uberblock
.ub_version
= version
;
3618 spa
->spa_ubsync
= spa
->spa_uberblock
;
3621 * Create "The Godfather" zio to hold all async IOs
3623 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3625 for (i
= 0; i
< max_ncpus
; i
++) {
3626 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3627 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3628 ZIO_FLAG_GODFATHER
);
3632 * Create the root vdev.
3634 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3636 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3638 ASSERT(error
!= 0 || rvd
!= NULL
);
3639 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3641 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3642 error
= SET_ERROR(EINVAL
);
3645 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3646 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3647 VDEV_ALLOC_ADD
)) == 0) {
3648 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
3649 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3650 vdev_expand(rvd
->vdev_child
[c
], txg
);
3654 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3658 spa_deactivate(spa
);
3660 mutex_exit(&spa_namespace_lock
);
3665 * Get the list of spares, if specified.
3667 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3668 &spares
, &nspares
) == 0) {
3669 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3671 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3672 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3673 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3674 spa_load_spares(spa
);
3675 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3676 spa
->spa_spares
.sav_sync
= B_TRUE
;
3680 * Get the list of level 2 cache devices, if specified.
3682 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3683 &l2cache
, &nl2cache
) == 0) {
3684 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3685 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3686 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3687 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3688 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3689 spa_load_l2cache(spa
);
3690 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3691 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3694 spa
->spa_is_initializing
= B_TRUE
;
3695 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3696 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3697 spa
->spa_is_initializing
= B_FALSE
;
3700 * Create DDTs (dedup tables).
3704 spa_update_dspace(spa
);
3706 tx
= dmu_tx_create_assigned(dp
, txg
);
3709 * Create the pool config object.
3711 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3712 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3713 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3715 if (zap_add(spa
->spa_meta_objset
,
3716 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3717 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3718 cmn_err(CE_PANIC
, "failed to add pool config");
3721 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3722 spa_feature_create_zap_objects(spa
, tx
);
3724 if (zap_add(spa
->spa_meta_objset
,
3725 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3726 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3727 cmn_err(CE_PANIC
, "failed to add pool version");
3730 /* Newly created pools with the right version are always deflated. */
3731 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3732 spa
->spa_deflate
= TRUE
;
3733 if (zap_add(spa
->spa_meta_objset
,
3734 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3735 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3736 cmn_err(CE_PANIC
, "failed to add deflate");
3741 * Create the deferred-free bpobj. Turn off compression
3742 * because sync-to-convergence takes longer if the blocksize
3745 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3746 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3747 ZIO_COMPRESS_OFF
, tx
);
3748 if (zap_add(spa
->spa_meta_objset
,
3749 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3750 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3751 cmn_err(CE_PANIC
, "failed to add bpobj");
3753 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3754 spa
->spa_meta_objset
, obj
));
3757 * Create the pool's history object.
3759 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3760 spa_history_create_obj(spa
, tx
);
3763 * Set pool properties.
3765 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3766 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3767 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3768 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3770 if (props
!= NULL
) {
3771 spa_configfile_set(spa
, props
, B_FALSE
);
3772 spa_sync_props(props
, tx
);
3777 spa
->spa_sync_on
= B_TRUE
;
3778 txg_sync_start(spa
->spa_dsl_pool
);
3781 * We explicitly wait for the first transaction to complete so that our
3782 * bean counters are appropriately updated.
3784 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3786 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3788 spa_history_log_version(spa
, "create");
3791 * Don't count references from objsets that are already closed
3792 * and are making their way through the eviction process.
3794 spa_evicting_os_wait(spa
);
3795 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3797 mutex_exit(&spa_namespace_lock
);
3804 * Get the root pool information from the root disk, then import the root pool
3805 * during the system boot up time.
3807 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3810 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3813 nvlist_t
*nvtop
, *nvroot
;
3816 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3820 * Add this top-level vdev to the child array.
3822 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3824 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3826 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3829 * Put this pool's top-level vdevs into a root vdev.
3831 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3832 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3833 VDEV_TYPE_ROOT
) == 0);
3834 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3835 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3836 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3840 * Replace the existing vdev_tree with the new root vdev in
3841 * this pool's configuration (remove the old, add the new).
3843 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3844 nvlist_free(nvroot
);
3849 * Walk the vdev tree and see if we can find a device with "better"
3850 * configuration. A configuration is "better" if the label on that
3851 * device has a more recent txg.
3854 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3858 for (c
= 0; c
< vd
->vdev_children
; c
++)
3859 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3861 if (vd
->vdev_ops
->vdev_op_leaf
) {
3865 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3869 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3873 * Do we have a better boot device?
3875 if (label_txg
> *txg
) {
3884 * Import a root pool.
3886 * For x86. devpath_list will consist of devid and/or physpath name of
3887 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3888 * The GRUB "findroot" command will return the vdev we should boot.
3890 * For Sparc, devpath_list consists the physpath name of the booting device
3891 * no matter the rootpool is a single device pool or a mirrored pool.
3893 * "/pci@1f,0/ide@d/disk@0,0:a"
3896 spa_import_rootpool(char *devpath
, char *devid
)
3899 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3900 nvlist_t
*config
, *nvtop
;
3906 * Read the label from the boot device and generate a configuration.
3908 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3909 #if defined(_OBP) && defined(_KERNEL)
3910 if (config
== NULL
) {
3911 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3913 get_iscsi_bootpath_phy(devpath
);
3914 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3918 if (config
== NULL
) {
3919 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
3921 return (SET_ERROR(EIO
));
3924 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3926 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3928 mutex_enter(&spa_namespace_lock
);
3929 if ((spa
= spa_lookup(pname
)) != NULL
) {
3931 * Remove the existing root pool from the namespace so that we
3932 * can replace it with the correct config we just read in.
3937 spa
= spa_add(pname
, config
, NULL
);
3938 spa
->spa_is_root
= B_TRUE
;
3939 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3942 * Build up a vdev tree based on the boot device's label config.
3944 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3946 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3947 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3948 VDEV_ALLOC_ROOTPOOL
);
3949 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3951 mutex_exit(&spa_namespace_lock
);
3952 nvlist_free(config
);
3953 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3959 * Get the boot vdev.
3961 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3962 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3963 (u_longlong_t
)guid
);
3964 error
= SET_ERROR(ENOENT
);
3969 * Determine if there is a better boot device.
3972 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3974 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3975 "try booting from '%s'", avd
->vdev_path
);
3976 error
= SET_ERROR(EINVAL
);
3981 * If the boot device is part of a spare vdev then ensure that
3982 * we're booting off the active spare.
3984 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3985 !bvd
->vdev_isspare
) {
3986 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
3987 "try booting from '%s'",
3989 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
3990 error
= SET_ERROR(EINVAL
);
3996 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3998 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3999 mutex_exit(&spa_namespace_lock
);
4001 nvlist_free(config
);
4008 * Import a non-root pool into the system.
4011 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4014 char *altroot
= NULL
;
4015 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4016 zpool_rewind_policy_t policy
;
4017 uint64_t mode
= spa_mode_global
;
4018 uint64_t readonly
= B_FALSE
;
4021 nvlist_t
**spares
, **l2cache
;
4022 uint_t nspares
, nl2cache
;
4025 * If a pool with this name exists, return failure.
4027 mutex_enter(&spa_namespace_lock
);
4028 if (spa_lookup(pool
) != NULL
) {
4029 mutex_exit(&spa_namespace_lock
);
4030 return (SET_ERROR(EEXIST
));
4034 * Create and initialize the spa structure.
4036 (void) nvlist_lookup_string(props
,
4037 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4038 (void) nvlist_lookup_uint64(props
,
4039 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4042 spa
= spa_add(pool
, config
, altroot
);
4043 spa
->spa_import_flags
= flags
;
4046 * Verbatim import - Take a pool and insert it into the namespace
4047 * as if it had been loaded at boot.
4049 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4051 spa_configfile_set(spa
, props
, B_FALSE
);
4053 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4055 mutex_exit(&spa_namespace_lock
);
4059 spa_activate(spa
, mode
);
4062 * Don't start async tasks until we know everything is healthy.
4064 spa_async_suspend(spa
);
4066 zpool_get_rewind_policy(config
, &policy
);
4067 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4068 state
= SPA_LOAD_RECOVER
;
4071 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4072 * because the user-supplied config is actually the one to trust when
4075 if (state
!= SPA_LOAD_RECOVER
)
4076 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4078 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4079 policy
.zrp_request
);
4082 * Propagate anything learned while loading the pool and pass it
4083 * back to caller (i.e. rewind info, missing devices, etc).
4085 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4086 spa
->spa_load_info
) == 0);
4088 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4090 * Toss any existing sparelist, as it doesn't have any validity
4091 * anymore, and conflicts with spa_has_spare().
4093 if (spa
->spa_spares
.sav_config
) {
4094 nvlist_free(spa
->spa_spares
.sav_config
);
4095 spa
->spa_spares
.sav_config
= NULL
;
4096 spa_load_spares(spa
);
4098 if (spa
->spa_l2cache
.sav_config
) {
4099 nvlist_free(spa
->spa_l2cache
.sav_config
);
4100 spa
->spa_l2cache
.sav_config
= NULL
;
4101 spa_load_l2cache(spa
);
4104 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4107 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4110 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4111 VDEV_ALLOC_L2CACHE
);
4112 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4115 spa_configfile_set(spa
, props
, B_FALSE
);
4117 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4118 (error
= spa_prop_set(spa
, props
)))) {
4120 spa_deactivate(spa
);
4122 mutex_exit(&spa_namespace_lock
);
4126 spa_async_resume(spa
);
4129 * Override any spares and level 2 cache devices as specified by
4130 * the user, as these may have correct device names/devids, etc.
4132 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4133 &spares
, &nspares
) == 0) {
4134 if (spa
->spa_spares
.sav_config
)
4135 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4136 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4138 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4139 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4140 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4141 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4142 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4143 spa_load_spares(spa
);
4144 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4145 spa
->spa_spares
.sav_sync
= B_TRUE
;
4147 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4148 &l2cache
, &nl2cache
) == 0) {
4149 if (spa
->spa_l2cache
.sav_config
)
4150 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4151 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4153 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4154 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4155 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4156 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4157 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4158 spa_load_l2cache(spa
);
4159 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4160 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4164 * Check for any removed devices.
4166 if (spa
->spa_autoreplace
) {
4167 spa_aux_check_removed(&spa
->spa_spares
);
4168 spa_aux_check_removed(&spa
->spa_l2cache
);
4171 if (spa_writeable(spa
)) {
4173 * Update the config cache to include the newly-imported pool.
4175 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4179 * It's possible that the pool was expanded while it was exported.
4180 * We kick off an async task to handle this for us.
4182 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4184 mutex_exit(&spa_namespace_lock
);
4185 spa_history_log_version(spa
, "import");
4188 zvol_create_minors(pool
);
4195 spa_tryimport(nvlist_t
*tryconfig
)
4197 nvlist_t
*config
= NULL
;
4203 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4206 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4210 * Create and initialize the spa structure.
4212 mutex_enter(&spa_namespace_lock
);
4213 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4214 spa_activate(spa
, FREAD
);
4217 * Pass off the heavy lifting to spa_load().
4218 * Pass TRUE for mosconfig because the user-supplied config
4219 * is actually the one to trust when doing an import.
4221 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4224 * If 'tryconfig' was at least parsable, return the current config.
4226 if (spa
->spa_root_vdev
!= NULL
) {
4227 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4228 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4230 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4232 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4233 spa
->spa_uberblock
.ub_timestamp
) == 0);
4234 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4235 spa
->spa_load_info
) == 0);
4236 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4237 spa
->spa_errata
) == 0);
4240 * If the bootfs property exists on this pool then we
4241 * copy it out so that external consumers can tell which
4242 * pools are bootable.
4244 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4245 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4248 * We have to play games with the name since the
4249 * pool was opened as TRYIMPORT_NAME.
4251 if (dsl_dsobj_to_dsname(spa_name(spa
),
4252 spa
->spa_bootfs
, tmpname
) == 0) {
4256 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4258 cp
= strchr(tmpname
, '/');
4260 (void) strlcpy(dsname
, tmpname
,
4263 (void) snprintf(dsname
, MAXPATHLEN
,
4264 "%s/%s", poolname
, ++cp
);
4266 VERIFY(nvlist_add_string(config
,
4267 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4268 kmem_free(dsname
, MAXPATHLEN
);
4270 kmem_free(tmpname
, MAXPATHLEN
);
4274 * Add the list of hot spares and level 2 cache devices.
4276 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4277 spa_add_spares(spa
, config
);
4278 spa_add_l2cache(spa
, config
);
4279 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4283 spa_deactivate(spa
);
4285 mutex_exit(&spa_namespace_lock
);
4291 * Pool export/destroy
4293 * The act of destroying or exporting a pool is very simple. We make sure there
4294 * is no more pending I/O and any references to the pool are gone. Then, we
4295 * update the pool state and sync all the labels to disk, removing the
4296 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4297 * we don't sync the labels or remove the configuration cache.
4300 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4301 boolean_t force
, boolean_t hardforce
)
4308 if (!(spa_mode_global
& FWRITE
))
4309 return (SET_ERROR(EROFS
));
4311 mutex_enter(&spa_namespace_lock
);
4312 if ((spa
= spa_lookup(pool
)) == NULL
) {
4313 mutex_exit(&spa_namespace_lock
);
4314 return (SET_ERROR(ENOENT
));
4318 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4319 * reacquire the namespace lock, and see if we can export.
4321 spa_open_ref(spa
, FTAG
);
4322 mutex_exit(&spa_namespace_lock
);
4323 spa_async_suspend(spa
);
4324 mutex_enter(&spa_namespace_lock
);
4325 spa_close(spa
, FTAG
);
4327 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4330 * The pool will be in core if it's openable, in which case we can
4331 * modify its state. Objsets may be open only because they're dirty,
4332 * so we have to force it to sync before checking spa_refcnt.
4334 if (spa
->spa_sync_on
) {
4335 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4336 spa_evicting_os_wait(spa
);
4340 * A pool cannot be exported or destroyed if there are active
4341 * references. If we are resetting a pool, allow references by
4342 * fault injection handlers.
4344 if (!spa_refcount_zero(spa
) ||
4345 (spa
->spa_inject_ref
!= 0 &&
4346 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4347 spa_async_resume(spa
);
4348 mutex_exit(&spa_namespace_lock
);
4349 return (SET_ERROR(EBUSY
));
4352 if (spa
->spa_sync_on
) {
4354 * A pool cannot be exported if it has an active shared spare.
4355 * This is to prevent other pools stealing the active spare
4356 * from an exported pool. At user's own will, such pool can
4357 * be forcedly exported.
4359 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4360 spa_has_active_shared_spare(spa
)) {
4361 spa_async_resume(spa
);
4362 mutex_exit(&spa_namespace_lock
);
4363 return (SET_ERROR(EXDEV
));
4367 * We want this to be reflected on every label,
4368 * so mark them all dirty. spa_unload() will do the
4369 * final sync that pushes these changes out.
4371 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4372 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4373 spa
->spa_state
= new_state
;
4374 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4376 vdev_config_dirty(spa
->spa_root_vdev
);
4377 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4382 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_DESTROY
);
4384 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4386 spa_deactivate(spa
);
4389 if (oldconfig
&& spa
->spa_config
)
4390 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4392 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4394 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4397 mutex_exit(&spa_namespace_lock
);
4403 * Destroy a storage pool.
4406 spa_destroy(char *pool
)
4408 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4413 * Export a storage pool.
4416 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4417 boolean_t hardforce
)
4419 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4424 * Similar to spa_export(), this unloads the spa_t without actually removing it
4425 * from the namespace in any way.
4428 spa_reset(char *pool
)
4430 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4435 * ==========================================================================
4436 * Device manipulation
4437 * ==========================================================================
4441 * Add a device to a storage pool.
4444 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4448 vdev_t
*rvd
= spa
->spa_root_vdev
;
4450 nvlist_t
**spares
, **l2cache
;
4451 uint_t nspares
, nl2cache
;
4454 ASSERT(spa_writeable(spa
));
4456 txg
= spa_vdev_enter(spa
);
4458 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4459 VDEV_ALLOC_ADD
)) != 0)
4460 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4462 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4464 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4468 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4472 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4473 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4475 if (vd
->vdev_children
!= 0 &&
4476 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4477 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4480 * We must validate the spares and l2cache devices after checking the
4481 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4483 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4484 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4487 * Transfer each new top-level vdev from vd to rvd.
4489 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4492 * Set the vdev id to the first hole, if one exists.
4494 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4495 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4496 vdev_free(rvd
->vdev_child
[id
]);
4500 tvd
= vd
->vdev_child
[c
];
4501 vdev_remove_child(vd
, tvd
);
4503 vdev_add_child(rvd
, tvd
);
4504 vdev_config_dirty(tvd
);
4508 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4509 ZPOOL_CONFIG_SPARES
);
4510 spa_load_spares(spa
);
4511 spa
->spa_spares
.sav_sync
= B_TRUE
;
4514 if (nl2cache
!= 0) {
4515 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4516 ZPOOL_CONFIG_L2CACHE
);
4517 spa_load_l2cache(spa
);
4518 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4522 * We have to be careful when adding new vdevs to an existing pool.
4523 * If other threads start allocating from these vdevs before we
4524 * sync the config cache, and we lose power, then upon reboot we may
4525 * fail to open the pool because there are DVAs that the config cache
4526 * can't translate. Therefore, we first add the vdevs without
4527 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4528 * and then let spa_config_update() initialize the new metaslabs.
4530 * spa_load() checks for added-but-not-initialized vdevs, so that
4531 * if we lose power at any point in this sequence, the remaining
4532 * steps will be completed the next time we load the pool.
4534 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4536 mutex_enter(&spa_namespace_lock
);
4537 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4538 mutex_exit(&spa_namespace_lock
);
4544 * Attach a device to a mirror. The arguments are the path to any device
4545 * in the mirror, and the nvroot for the new device. If the path specifies
4546 * a device that is not mirrored, we automatically insert the mirror vdev.
4548 * If 'replacing' is specified, the new device is intended to replace the
4549 * existing device; in this case the two devices are made into their own
4550 * mirror using the 'replacing' vdev, which is functionally identical to
4551 * the mirror vdev (it actually reuses all the same ops) but has a few
4552 * extra rules: you can't attach to it after it's been created, and upon
4553 * completion of resilvering, the first disk (the one being replaced)
4554 * is automatically detached.
4557 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4559 uint64_t txg
, dtl_max_txg
;
4560 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4562 char *oldvdpath
, *newvdpath
;
4565 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4567 ASSERT(spa_writeable(spa
));
4569 txg
= spa_vdev_enter(spa
);
4571 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4574 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4576 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4577 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4579 pvd
= oldvd
->vdev_parent
;
4581 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4582 VDEV_ALLOC_ATTACH
)) != 0)
4583 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4585 if (newrootvd
->vdev_children
!= 1)
4586 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4588 newvd
= newrootvd
->vdev_child
[0];
4590 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4591 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4593 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4594 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4597 * Spares can't replace logs
4599 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4600 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4604 * For attach, the only allowable parent is a mirror or the root
4607 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4608 pvd
->vdev_ops
!= &vdev_root_ops
)
4609 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4611 pvops
= &vdev_mirror_ops
;
4614 * Active hot spares can only be replaced by inactive hot
4617 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4618 oldvd
->vdev_isspare
&&
4619 !spa_has_spare(spa
, newvd
->vdev_guid
))
4620 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4623 * If the source is a hot spare, and the parent isn't already a
4624 * spare, then we want to create a new hot spare. Otherwise, we
4625 * want to create a replacing vdev. The user is not allowed to
4626 * attach to a spared vdev child unless the 'isspare' state is
4627 * the same (spare replaces spare, non-spare replaces
4630 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4631 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4632 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4633 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4634 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4635 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4638 if (newvd
->vdev_isspare
)
4639 pvops
= &vdev_spare_ops
;
4641 pvops
= &vdev_replacing_ops
;
4645 * Make sure the new device is big enough.
4647 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4648 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4651 * The new device cannot have a higher alignment requirement
4652 * than the top-level vdev.
4654 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4655 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4658 * If this is an in-place replacement, update oldvd's path and devid
4659 * to make it distinguishable from newvd, and unopenable from now on.
4661 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4662 spa_strfree(oldvd
->vdev_path
);
4663 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4665 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4666 newvd
->vdev_path
, "old");
4667 if (oldvd
->vdev_devid
!= NULL
) {
4668 spa_strfree(oldvd
->vdev_devid
);
4669 oldvd
->vdev_devid
= NULL
;
4673 /* mark the device being resilvered */
4674 newvd
->vdev_resilver_txg
= txg
;
4677 * If the parent is not a mirror, or if we're replacing, insert the new
4678 * mirror/replacing/spare vdev above oldvd.
4680 if (pvd
->vdev_ops
!= pvops
)
4681 pvd
= vdev_add_parent(oldvd
, pvops
);
4683 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4684 ASSERT(pvd
->vdev_ops
== pvops
);
4685 ASSERT(oldvd
->vdev_parent
== pvd
);
4688 * Extract the new device from its root and add it to pvd.
4690 vdev_remove_child(newrootvd
, newvd
);
4691 newvd
->vdev_id
= pvd
->vdev_children
;
4692 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4693 vdev_add_child(pvd
, newvd
);
4695 tvd
= newvd
->vdev_top
;
4696 ASSERT(pvd
->vdev_top
== tvd
);
4697 ASSERT(tvd
->vdev_parent
== rvd
);
4699 vdev_config_dirty(tvd
);
4702 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4703 * for any dmu_sync-ed blocks. It will propagate upward when
4704 * spa_vdev_exit() calls vdev_dtl_reassess().
4706 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4708 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4709 dtl_max_txg
- TXG_INITIAL
);
4711 if (newvd
->vdev_isspare
) {
4712 spa_spare_activate(newvd
);
4713 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_DEVICE_SPARE
);
4716 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4717 newvdpath
= spa_strdup(newvd
->vdev_path
);
4718 newvd_isspare
= newvd
->vdev_isspare
;
4721 * Mark newvd's DTL dirty in this txg.
4723 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4726 * Schedule the resilver to restart in the future. We do this to
4727 * ensure that dmu_sync-ed blocks have been stitched into the
4728 * respective datasets.
4730 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4735 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4737 spa_history_log_internal(spa
, "vdev attach", NULL
,
4738 "%s vdev=%s %s vdev=%s",
4739 replacing
&& newvd_isspare
? "spare in" :
4740 replacing
? "replace" : "attach", newvdpath
,
4741 replacing
? "for" : "to", oldvdpath
);
4743 spa_strfree(oldvdpath
);
4744 spa_strfree(newvdpath
);
4746 if (spa
->spa_bootfs
)
4747 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH
);
4753 * Detach a device from a mirror or replacing vdev.
4755 * If 'replace_done' is specified, only detach if the parent
4756 * is a replacing vdev.
4759 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4763 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4764 boolean_t unspare
= B_FALSE
;
4765 uint64_t unspare_guid
= 0;
4768 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4769 ASSERT(spa_writeable(spa
));
4771 txg
= spa_vdev_enter(spa
);
4773 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4776 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4778 if (!vd
->vdev_ops
->vdev_op_leaf
)
4779 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4781 pvd
= vd
->vdev_parent
;
4784 * If the parent/child relationship is not as expected, don't do it.
4785 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4786 * vdev that's replacing B with C. The user's intent in replacing
4787 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4788 * the replace by detaching C, the expected behavior is to end up
4789 * M(A,B). But suppose that right after deciding to detach C,
4790 * the replacement of B completes. We would have M(A,C), and then
4791 * ask to detach C, which would leave us with just A -- not what
4792 * the user wanted. To prevent this, we make sure that the
4793 * parent/child relationship hasn't changed -- in this example,
4794 * that C's parent is still the replacing vdev R.
4796 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4797 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4800 * Only 'replacing' or 'spare' vdevs can be replaced.
4802 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4803 pvd
->vdev_ops
!= &vdev_spare_ops
)
4804 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4806 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4807 spa_version(spa
) >= SPA_VERSION_SPARES
);
4810 * Only mirror, replacing, and spare vdevs support detach.
4812 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4813 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4814 pvd
->vdev_ops
!= &vdev_spare_ops
)
4815 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4818 * If this device has the only valid copy of some data,
4819 * we cannot safely detach it.
4821 if (vdev_dtl_required(vd
))
4822 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4824 ASSERT(pvd
->vdev_children
>= 2);
4827 * If we are detaching the second disk from a replacing vdev, then
4828 * check to see if we changed the original vdev's path to have "/old"
4829 * at the end in spa_vdev_attach(). If so, undo that change now.
4831 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4832 vd
->vdev_path
!= NULL
) {
4833 size_t len
= strlen(vd
->vdev_path
);
4835 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4836 cvd
= pvd
->vdev_child
[c
];
4838 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4841 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4842 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4843 spa_strfree(cvd
->vdev_path
);
4844 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4851 * If we are detaching the original disk from a spare, then it implies
4852 * that the spare should become a real disk, and be removed from the
4853 * active spare list for the pool.
4855 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4857 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4861 * Erase the disk labels so the disk can be used for other things.
4862 * This must be done after all other error cases are handled,
4863 * but before we disembowel vd (so we can still do I/O to it).
4864 * But if we can't do it, don't treat the error as fatal --
4865 * it may be that the unwritability of the disk is the reason
4866 * it's being detached!
4868 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4871 * Remove vd from its parent and compact the parent's children.
4873 vdev_remove_child(pvd
, vd
);
4874 vdev_compact_children(pvd
);
4877 * Remember one of the remaining children so we can get tvd below.
4879 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4882 * If we need to remove the remaining child from the list of hot spares,
4883 * do it now, marking the vdev as no longer a spare in the process.
4884 * We must do this before vdev_remove_parent(), because that can
4885 * change the GUID if it creates a new toplevel GUID. For a similar
4886 * reason, we must remove the spare now, in the same txg as the detach;
4887 * otherwise someone could attach a new sibling, change the GUID, and
4888 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4891 ASSERT(cvd
->vdev_isspare
);
4892 spa_spare_remove(cvd
);
4893 unspare_guid
= cvd
->vdev_guid
;
4894 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4895 cvd
->vdev_unspare
= B_TRUE
;
4899 * If the parent mirror/replacing vdev only has one child,
4900 * the parent is no longer needed. Remove it from the tree.
4902 if (pvd
->vdev_children
== 1) {
4903 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4904 cvd
->vdev_unspare
= B_FALSE
;
4905 vdev_remove_parent(cvd
);
4910 * We don't set tvd until now because the parent we just removed
4911 * may have been the previous top-level vdev.
4913 tvd
= cvd
->vdev_top
;
4914 ASSERT(tvd
->vdev_parent
== rvd
);
4917 * Reevaluate the parent vdev state.
4919 vdev_propagate_state(cvd
);
4922 * If the 'autoexpand' property is set on the pool then automatically
4923 * try to expand the size of the pool. For example if the device we
4924 * just detached was smaller than the others, it may be possible to
4925 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4926 * first so that we can obtain the updated sizes of the leaf vdevs.
4928 if (spa
->spa_autoexpand
) {
4930 vdev_expand(tvd
, txg
);
4933 vdev_config_dirty(tvd
);
4936 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4937 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4938 * But first make sure we're not on any *other* txg's DTL list, to
4939 * prevent vd from being accessed after it's freed.
4941 vdpath
= spa_strdup(vd
->vdev_path
);
4942 for (t
= 0; t
< TXG_SIZE
; t
++)
4943 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4944 vd
->vdev_detached
= B_TRUE
;
4945 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4947 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_REMOVE
);
4949 /* hang on to the spa before we release the lock */
4950 spa_open_ref(spa
, FTAG
);
4952 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4954 spa_history_log_internal(spa
, "detach", NULL
,
4956 spa_strfree(vdpath
);
4959 * If this was the removal of the original device in a hot spare vdev,
4960 * then we want to go through and remove the device from the hot spare
4961 * list of every other pool.
4964 spa_t
*altspa
= NULL
;
4966 mutex_enter(&spa_namespace_lock
);
4967 while ((altspa
= spa_next(altspa
)) != NULL
) {
4968 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4972 spa_open_ref(altspa
, FTAG
);
4973 mutex_exit(&spa_namespace_lock
);
4974 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4975 mutex_enter(&spa_namespace_lock
);
4976 spa_close(altspa
, FTAG
);
4978 mutex_exit(&spa_namespace_lock
);
4980 /* search the rest of the vdevs for spares to remove */
4981 spa_vdev_resilver_done(spa
);
4984 /* all done with the spa; OK to release */
4985 mutex_enter(&spa_namespace_lock
);
4986 spa_close(spa
, FTAG
);
4987 mutex_exit(&spa_namespace_lock
);
4993 * Split a set of devices from their mirrors, and create a new pool from them.
4996 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
4997 nvlist_t
*props
, boolean_t exp
)
5000 uint64_t txg
, *glist
;
5002 uint_t c
, children
, lastlog
;
5003 nvlist_t
**child
, *nvl
, *tmp
;
5005 char *altroot
= NULL
;
5006 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5007 boolean_t activate_slog
;
5009 ASSERT(spa_writeable(spa
));
5011 txg
= spa_vdev_enter(spa
);
5013 /* clear the log and flush everything up to now */
5014 activate_slog
= spa_passivate_log(spa
);
5015 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5016 error
= spa_offline_log(spa
);
5017 txg
= spa_vdev_config_enter(spa
);
5020 spa_activate_log(spa
);
5023 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5025 /* check new spa name before going any further */
5026 if (spa_lookup(newname
) != NULL
)
5027 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5030 * scan through all the children to ensure they're all mirrors
5032 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5033 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5035 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5037 /* first, check to ensure we've got the right child count */
5038 rvd
= spa
->spa_root_vdev
;
5040 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5041 vdev_t
*vd
= rvd
->vdev_child
[c
];
5043 /* don't count the holes & logs as children */
5044 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5052 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5053 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5055 /* next, ensure no spare or cache devices are part of the split */
5056 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5057 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5058 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5060 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5061 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5063 /* then, loop over each vdev and validate it */
5064 for (c
= 0; c
< children
; c
++) {
5065 uint64_t is_hole
= 0;
5067 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5071 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5072 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5075 error
= SET_ERROR(EINVAL
);
5080 /* which disk is going to be split? */
5081 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5083 error
= SET_ERROR(EINVAL
);
5087 /* look it up in the spa */
5088 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5089 if (vml
[c
] == NULL
) {
5090 error
= SET_ERROR(ENODEV
);
5094 /* make sure there's nothing stopping the split */
5095 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5096 vml
[c
]->vdev_islog
||
5097 vml
[c
]->vdev_ishole
||
5098 vml
[c
]->vdev_isspare
||
5099 vml
[c
]->vdev_isl2cache
||
5100 !vdev_writeable(vml
[c
]) ||
5101 vml
[c
]->vdev_children
!= 0 ||
5102 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5103 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5104 error
= SET_ERROR(EINVAL
);
5108 if (vdev_dtl_required(vml
[c
])) {
5109 error
= SET_ERROR(EBUSY
);
5113 /* we need certain info from the top level */
5114 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5115 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5116 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5117 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5118 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5119 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5120 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5121 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5125 kmem_free(vml
, children
* sizeof (vdev_t
*));
5126 kmem_free(glist
, children
* sizeof (uint64_t));
5127 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5130 /* stop writers from using the disks */
5131 for (c
= 0; c
< children
; c
++) {
5133 vml
[c
]->vdev_offline
= B_TRUE
;
5135 vdev_reopen(spa
->spa_root_vdev
);
5138 * Temporarily record the splitting vdevs in the spa config. This
5139 * will disappear once the config is regenerated.
5141 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5142 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5143 glist
, children
) == 0);
5144 kmem_free(glist
, children
* sizeof (uint64_t));
5146 mutex_enter(&spa
->spa_props_lock
);
5147 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5149 mutex_exit(&spa
->spa_props_lock
);
5150 spa
->spa_config_splitting
= nvl
;
5151 vdev_config_dirty(spa
->spa_root_vdev
);
5153 /* configure and create the new pool */
5154 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5155 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5156 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5157 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5158 spa_version(spa
)) == 0);
5159 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5160 spa
->spa_config_txg
) == 0);
5161 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5162 spa_generate_guid(NULL
)) == 0);
5163 (void) nvlist_lookup_string(props
,
5164 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5166 /* add the new pool to the namespace */
5167 newspa
= spa_add(newname
, config
, altroot
);
5168 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5169 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5171 /* release the spa config lock, retaining the namespace lock */
5172 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5174 if (zio_injection_enabled
)
5175 zio_handle_panic_injection(spa
, FTAG
, 1);
5177 spa_activate(newspa
, spa_mode_global
);
5178 spa_async_suspend(newspa
);
5180 /* create the new pool from the disks of the original pool */
5181 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5185 /* if that worked, generate a real config for the new pool */
5186 if (newspa
->spa_root_vdev
!= NULL
) {
5187 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5188 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5189 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5190 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5191 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5196 if (props
!= NULL
) {
5197 spa_configfile_set(newspa
, props
, B_FALSE
);
5198 error
= spa_prop_set(newspa
, props
);
5203 /* flush everything */
5204 txg
= spa_vdev_config_enter(newspa
);
5205 vdev_config_dirty(newspa
->spa_root_vdev
);
5206 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5208 if (zio_injection_enabled
)
5209 zio_handle_panic_injection(spa
, FTAG
, 2);
5211 spa_async_resume(newspa
);
5213 /* finally, update the original pool's config */
5214 txg
= spa_vdev_config_enter(spa
);
5215 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5216 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5219 for (c
= 0; c
< children
; c
++) {
5220 if (vml
[c
] != NULL
) {
5223 spa_history_log_internal(spa
, "detach", tx
,
5224 "vdev=%s", vml
[c
]->vdev_path
);
5228 vdev_config_dirty(spa
->spa_root_vdev
);
5229 spa
->spa_config_splitting
= NULL
;
5233 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5235 if (zio_injection_enabled
)
5236 zio_handle_panic_injection(spa
, FTAG
, 3);
5238 /* split is complete; log a history record */
5239 spa_history_log_internal(newspa
, "split", NULL
,
5240 "from pool %s", spa_name(spa
));
5242 kmem_free(vml
, children
* sizeof (vdev_t
*));
5244 /* if we're not going to mount the filesystems in userland, export */
5246 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5253 spa_deactivate(newspa
);
5256 txg
= spa_vdev_config_enter(spa
);
5258 /* re-online all offlined disks */
5259 for (c
= 0; c
< children
; c
++) {
5261 vml
[c
]->vdev_offline
= B_FALSE
;
5263 vdev_reopen(spa
->spa_root_vdev
);
5265 nvlist_free(spa
->spa_config_splitting
);
5266 spa
->spa_config_splitting
= NULL
;
5267 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5269 kmem_free(vml
, children
* sizeof (vdev_t
*));
5274 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5278 for (i
= 0; i
< count
; i
++) {
5281 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5284 if (guid
== target_guid
)
5292 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5293 nvlist_t
*dev_to_remove
)
5295 nvlist_t
**newdev
= NULL
;
5299 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5301 for (i
= 0, j
= 0; i
< count
; i
++) {
5302 if (dev
[i
] == dev_to_remove
)
5304 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5307 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5308 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5310 for (i
= 0; i
< count
- 1; i
++)
5311 nvlist_free(newdev
[i
]);
5314 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5318 * Evacuate the device.
5321 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5326 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5327 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5328 ASSERT(vd
== vd
->vdev_top
);
5331 * Evacuate the device. We don't hold the config lock as writer
5332 * since we need to do I/O but we do keep the
5333 * spa_namespace_lock held. Once this completes the device
5334 * should no longer have any blocks allocated on it.
5336 if (vd
->vdev_islog
) {
5337 if (vd
->vdev_stat
.vs_alloc
!= 0)
5338 error
= spa_offline_log(spa
);
5340 error
= SET_ERROR(ENOTSUP
);
5347 * The evacuation succeeded. Remove any remaining MOS metadata
5348 * associated with this vdev, and wait for these changes to sync.
5350 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5351 txg
= spa_vdev_config_enter(spa
);
5352 vd
->vdev_removing
= B_TRUE
;
5353 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5354 vdev_config_dirty(vd
);
5355 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5361 * Complete the removal by cleaning up the namespace.
5364 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5366 vdev_t
*rvd
= spa
->spa_root_vdev
;
5367 uint64_t id
= vd
->vdev_id
;
5368 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5370 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5371 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5372 ASSERT(vd
== vd
->vdev_top
);
5375 * Only remove any devices which are empty.
5377 if (vd
->vdev_stat
.vs_alloc
!= 0)
5380 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5382 if (list_link_active(&vd
->vdev_state_dirty_node
))
5383 vdev_state_clean(vd
);
5384 if (list_link_active(&vd
->vdev_config_dirty_node
))
5385 vdev_config_clean(vd
);
5390 vdev_compact_children(rvd
);
5392 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5393 vdev_add_child(rvd
, vd
);
5395 vdev_config_dirty(rvd
);
5398 * Reassess the health of our root vdev.
5404 * Remove a device from the pool -
5406 * Removing a device from the vdev namespace requires several steps
5407 * and can take a significant amount of time. As a result we use
5408 * the spa_vdev_config_[enter/exit] functions which allow us to
5409 * grab and release the spa_config_lock while still holding the namespace
5410 * lock. During each step the configuration is synced out.
5412 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5416 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5419 metaslab_group_t
*mg
;
5420 nvlist_t
**spares
, **l2cache
, *nv
;
5422 uint_t nspares
, nl2cache
;
5424 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5426 ASSERT(spa_writeable(spa
));
5429 txg
= spa_vdev_enter(spa
);
5431 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5433 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5434 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5435 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5436 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5438 * Only remove the hot spare if it's not currently in use
5441 if (vd
== NULL
|| unspare
) {
5442 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5443 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5444 spa_load_spares(spa
);
5445 spa
->spa_spares
.sav_sync
= B_TRUE
;
5447 error
= SET_ERROR(EBUSY
);
5449 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5450 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5451 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5452 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5454 * Cache devices can always be removed.
5456 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5457 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5458 spa_load_l2cache(spa
);
5459 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5460 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5462 ASSERT(vd
== vd
->vdev_top
);
5467 * Stop allocating from this vdev.
5469 metaslab_group_passivate(mg
);
5472 * Wait for the youngest allocations and frees to sync,
5473 * and then wait for the deferral of those frees to finish.
5475 spa_vdev_config_exit(spa
, NULL
,
5476 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5479 * Attempt to evacuate the vdev.
5481 error
= spa_vdev_remove_evacuate(spa
, vd
);
5483 txg
= spa_vdev_config_enter(spa
);
5486 * If we couldn't evacuate the vdev, unwind.
5489 metaslab_group_activate(mg
);
5490 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5494 * Clean up the vdev namespace.
5496 spa_vdev_remove_from_namespace(spa
, vd
);
5498 } else if (vd
!= NULL
) {
5500 * Normal vdevs cannot be removed (yet).
5502 error
= SET_ERROR(ENOTSUP
);
5505 * There is no vdev of any kind with the specified guid.
5507 error
= SET_ERROR(ENOENT
);
5511 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5517 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5518 * currently spared, so we can detach it.
5521 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5523 vdev_t
*newvd
, *oldvd
;
5526 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5527 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5533 * Check for a completed replacement. We always consider the first
5534 * vdev in the list to be the oldest vdev, and the last one to be
5535 * the newest (see spa_vdev_attach() for how that works). In
5536 * the case where the newest vdev is faulted, we will not automatically
5537 * remove it after a resilver completes. This is OK as it will require
5538 * user intervention to determine which disk the admin wishes to keep.
5540 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5541 ASSERT(vd
->vdev_children
> 1);
5543 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5544 oldvd
= vd
->vdev_child
[0];
5546 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5547 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5548 !vdev_dtl_required(oldvd
))
5553 * Check for a completed resilver with the 'unspare' flag set.
5555 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5556 vdev_t
*first
= vd
->vdev_child
[0];
5557 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5559 if (last
->vdev_unspare
) {
5562 } else if (first
->vdev_unspare
) {
5569 if (oldvd
!= NULL
&&
5570 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5571 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5572 !vdev_dtl_required(oldvd
))
5576 * If there are more than two spares attached to a disk,
5577 * and those spares are not required, then we want to
5578 * attempt to free them up now so that they can be used
5579 * by other pools. Once we're back down to a single
5580 * disk+spare, we stop removing them.
5582 if (vd
->vdev_children
> 2) {
5583 newvd
= vd
->vdev_child
[1];
5585 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5586 vdev_dtl_empty(last
, DTL_MISSING
) &&
5587 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5588 !vdev_dtl_required(newvd
))
5597 spa_vdev_resilver_done(spa_t
*spa
)
5599 vdev_t
*vd
, *pvd
, *ppvd
;
5600 uint64_t guid
, sguid
, pguid
, ppguid
;
5602 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5604 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5605 pvd
= vd
->vdev_parent
;
5606 ppvd
= pvd
->vdev_parent
;
5607 guid
= vd
->vdev_guid
;
5608 pguid
= pvd
->vdev_guid
;
5609 ppguid
= ppvd
->vdev_guid
;
5612 * If we have just finished replacing a hot spared device, then
5613 * we need to detach the parent's first child (the original hot
5616 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5617 ppvd
->vdev_children
== 2) {
5618 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5619 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5621 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5623 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5624 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5626 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5628 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5631 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5635 * Update the stored path or FRU for this vdev.
5638 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5642 boolean_t sync
= B_FALSE
;
5644 ASSERT(spa_writeable(spa
));
5646 spa_vdev_state_enter(spa
, SCL_ALL
);
5648 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5649 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5651 if (!vd
->vdev_ops
->vdev_op_leaf
)
5652 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5655 if (strcmp(value
, vd
->vdev_path
) != 0) {
5656 spa_strfree(vd
->vdev_path
);
5657 vd
->vdev_path
= spa_strdup(value
);
5661 if (vd
->vdev_fru
== NULL
) {
5662 vd
->vdev_fru
= spa_strdup(value
);
5664 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5665 spa_strfree(vd
->vdev_fru
);
5666 vd
->vdev_fru
= spa_strdup(value
);
5671 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5675 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5677 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5681 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5683 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5687 * ==========================================================================
5689 * ==========================================================================
5693 spa_scan_stop(spa_t
*spa
)
5695 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5696 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5697 return (SET_ERROR(EBUSY
));
5698 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5702 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5704 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5706 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5707 return (SET_ERROR(ENOTSUP
));
5710 * If a resilver was requested, but there is no DTL on a
5711 * writeable leaf device, we have nothing to do.
5713 if (func
== POOL_SCAN_RESILVER
&&
5714 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5715 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5719 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5723 * ==========================================================================
5724 * SPA async task processing
5725 * ==========================================================================
5729 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5733 if (vd
->vdev_remove_wanted
) {
5734 vd
->vdev_remove_wanted
= B_FALSE
;
5735 vd
->vdev_delayed_close
= B_FALSE
;
5736 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5739 * We want to clear the stats, but we don't want to do a full
5740 * vdev_clear() as that will cause us to throw away
5741 * degraded/faulted state as well as attempt to reopen the
5742 * device, all of which is a waste.
5744 vd
->vdev_stat
.vs_read_errors
= 0;
5745 vd
->vdev_stat
.vs_write_errors
= 0;
5746 vd
->vdev_stat
.vs_checksum_errors
= 0;
5748 vdev_state_dirty(vd
->vdev_top
);
5751 for (c
= 0; c
< vd
->vdev_children
; c
++)
5752 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5756 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5760 if (vd
->vdev_probe_wanted
) {
5761 vd
->vdev_probe_wanted
= B_FALSE
;
5762 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5765 for (c
= 0; c
< vd
->vdev_children
; c
++)
5766 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5770 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5774 if (!spa
->spa_autoexpand
)
5777 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5778 vdev_t
*cvd
= vd
->vdev_child
[c
];
5779 spa_async_autoexpand(spa
, cvd
);
5782 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5785 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND
);
5789 spa_async_thread(spa_t
*spa
)
5793 ASSERT(spa
->spa_sync_on
);
5795 mutex_enter(&spa
->spa_async_lock
);
5796 tasks
= spa
->spa_async_tasks
;
5797 spa
->spa_async_tasks
= 0;
5798 mutex_exit(&spa
->spa_async_lock
);
5801 * See if the config needs to be updated.
5803 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5804 uint64_t old_space
, new_space
;
5806 mutex_enter(&spa_namespace_lock
);
5807 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5808 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5809 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5810 mutex_exit(&spa_namespace_lock
);
5813 * If the pool grew as a result of the config update,
5814 * then log an internal history event.
5816 if (new_space
!= old_space
) {
5817 spa_history_log_internal(spa
, "vdev online", NULL
,
5818 "pool '%s' size: %llu(+%llu)",
5819 spa_name(spa
), new_space
, new_space
- old_space
);
5824 * See if any devices need to be marked REMOVED.
5826 if (tasks
& SPA_ASYNC_REMOVE
) {
5827 spa_vdev_state_enter(spa
, SCL_NONE
);
5828 spa_async_remove(spa
, spa
->spa_root_vdev
);
5829 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5830 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5831 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5832 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5833 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5836 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5837 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5838 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5839 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5843 * See if any devices need to be probed.
5845 if (tasks
& SPA_ASYNC_PROBE
) {
5846 spa_vdev_state_enter(spa
, SCL_NONE
);
5847 spa_async_probe(spa
, spa
->spa_root_vdev
);
5848 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5852 * If any devices are done replacing, detach them.
5854 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5855 spa_vdev_resilver_done(spa
);
5858 * Kick off a resilver.
5860 if (tasks
& SPA_ASYNC_RESILVER
)
5861 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5864 * Let the world know that we're done.
5866 mutex_enter(&spa
->spa_async_lock
);
5867 spa
->spa_async_thread
= NULL
;
5868 cv_broadcast(&spa
->spa_async_cv
);
5869 mutex_exit(&spa
->spa_async_lock
);
5874 spa_async_suspend(spa_t
*spa
)
5876 mutex_enter(&spa
->spa_async_lock
);
5877 spa
->spa_async_suspended
++;
5878 while (spa
->spa_async_thread
!= NULL
)
5879 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5880 mutex_exit(&spa
->spa_async_lock
);
5884 spa_async_resume(spa_t
*spa
)
5886 mutex_enter(&spa
->spa_async_lock
);
5887 ASSERT(spa
->spa_async_suspended
!= 0);
5888 spa
->spa_async_suspended
--;
5889 mutex_exit(&spa
->spa_async_lock
);
5893 spa_async_dispatch(spa_t
*spa
)
5895 mutex_enter(&spa
->spa_async_lock
);
5896 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5897 spa
->spa_async_thread
== NULL
&&
5898 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5899 spa
->spa_async_thread
= thread_create(NULL
, 0,
5900 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5901 mutex_exit(&spa
->spa_async_lock
);
5905 spa_async_request(spa_t
*spa
, int task
)
5907 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5908 mutex_enter(&spa
->spa_async_lock
);
5909 spa
->spa_async_tasks
|= task
;
5910 mutex_exit(&spa
->spa_async_lock
);
5914 * ==========================================================================
5915 * SPA syncing routines
5916 * ==========================================================================
5920 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5923 bpobj_enqueue(bpo
, bp
, tx
);
5928 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5932 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5938 * Note: this simple function is not inlined to make it easier to dtrace the
5939 * amount of time spent syncing frees.
5942 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
5944 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5945 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
5946 VERIFY(zio_wait(zio
) == 0);
5950 * Note: this simple function is not inlined to make it easier to dtrace the
5951 * amount of time spent syncing deferred frees.
5954 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
5956 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5957 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
5958 spa_free_sync_cb
, zio
, tx
), ==, 0);
5959 VERIFY0(zio_wait(zio
));
5963 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5965 char *packed
= NULL
;
5970 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5973 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5974 * information. This avoids the dmu_buf_will_dirty() path and
5975 * saves us a pre-read to get data we don't actually care about.
5977 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
5978 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
5980 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
5982 bzero(packed
+ nvsize
, bufsize
- nvsize
);
5984 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
5986 vmem_free(packed
, bufsize
);
5988 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
5989 dmu_buf_will_dirty(db
, tx
);
5990 *(uint64_t *)db
->db_data
= nvsize
;
5991 dmu_buf_rele(db
, FTAG
);
5995 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
5996 const char *config
, const char *entry
)
6006 * Update the MOS nvlist describing the list of available devices.
6007 * spa_validate_aux() will have already made sure this nvlist is
6008 * valid and the vdevs are labeled appropriately.
6010 if (sav
->sav_object
== 0) {
6011 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6012 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6013 sizeof (uint64_t), tx
);
6014 VERIFY(zap_update(spa
->spa_meta_objset
,
6015 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6016 &sav
->sav_object
, tx
) == 0);
6019 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6020 if (sav
->sav_count
== 0) {
6021 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6023 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6024 for (i
= 0; i
< sav
->sav_count
; i
++)
6025 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6026 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6027 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6028 sav
->sav_count
) == 0);
6029 for (i
= 0; i
< sav
->sav_count
; i
++)
6030 nvlist_free(list
[i
]);
6031 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6034 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6035 nvlist_free(nvroot
);
6037 sav
->sav_sync
= B_FALSE
;
6041 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6045 if (list_is_empty(&spa
->spa_config_dirty_list
))
6048 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6050 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6051 dmu_tx_get_txg(tx
), B_FALSE
);
6054 * If we're upgrading the spa version then make sure that
6055 * the config object gets updated with the correct version.
6057 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6058 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6059 spa
->spa_uberblock
.ub_version
);
6061 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6063 if (spa
->spa_config_syncing
)
6064 nvlist_free(spa
->spa_config_syncing
);
6065 spa
->spa_config_syncing
= config
;
6067 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6071 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6073 uint64_t *versionp
= arg
;
6074 uint64_t version
= *versionp
;
6075 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6078 * Setting the version is special cased when first creating the pool.
6080 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6082 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6083 ASSERT(version
>= spa_version(spa
));
6085 spa
->spa_uberblock
.ub_version
= version
;
6086 vdev_config_dirty(spa
->spa_root_vdev
);
6087 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6091 * Set zpool properties.
6094 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6096 nvlist_t
*nvp
= arg
;
6097 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6098 objset_t
*mos
= spa
->spa_meta_objset
;
6099 nvpair_t
*elem
= NULL
;
6101 mutex_enter(&spa
->spa_props_lock
);
6103 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6105 char *strval
, *fname
;
6107 const char *propname
;
6108 zprop_type_t proptype
;
6111 prop
= zpool_name_to_prop(nvpair_name(elem
));
6112 switch ((int)prop
) {
6115 * We checked this earlier in spa_prop_validate().
6117 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6119 fname
= strchr(nvpair_name(elem
), '@') + 1;
6120 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6122 spa_feature_enable(spa
, fid
, tx
);
6123 spa_history_log_internal(spa
, "set", tx
,
6124 "%s=enabled", nvpair_name(elem
));
6127 case ZPOOL_PROP_VERSION
:
6128 intval
= fnvpair_value_uint64(elem
);
6130 * The version is synced seperatly before other
6131 * properties and should be correct by now.
6133 ASSERT3U(spa_version(spa
), >=, intval
);
6136 case ZPOOL_PROP_ALTROOT
:
6138 * 'altroot' is a non-persistent property. It should
6139 * have been set temporarily at creation or import time.
6141 ASSERT(spa
->spa_root
!= NULL
);
6144 case ZPOOL_PROP_READONLY
:
6145 case ZPOOL_PROP_CACHEFILE
:
6147 * 'readonly' and 'cachefile' are also non-persisitent
6151 case ZPOOL_PROP_COMMENT
:
6152 strval
= fnvpair_value_string(elem
);
6153 if (spa
->spa_comment
!= NULL
)
6154 spa_strfree(spa
->spa_comment
);
6155 spa
->spa_comment
= spa_strdup(strval
);
6157 * We need to dirty the configuration on all the vdevs
6158 * so that their labels get updated. It's unnecessary
6159 * to do this for pool creation since the vdev's
6160 * configuratoin has already been dirtied.
6162 if (tx
->tx_txg
!= TXG_INITIAL
)
6163 vdev_config_dirty(spa
->spa_root_vdev
);
6164 spa_history_log_internal(spa
, "set", tx
,
6165 "%s=%s", nvpair_name(elem
), strval
);
6169 * Set pool property values in the poolprops mos object.
6171 if (spa
->spa_pool_props_object
== 0) {
6172 spa
->spa_pool_props_object
=
6173 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6174 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6178 /* normalize the property name */
6179 propname
= zpool_prop_to_name(prop
);
6180 proptype
= zpool_prop_get_type(prop
);
6182 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6183 ASSERT(proptype
== PROP_TYPE_STRING
);
6184 strval
= fnvpair_value_string(elem
);
6185 VERIFY0(zap_update(mos
,
6186 spa
->spa_pool_props_object
, propname
,
6187 1, strlen(strval
) + 1, strval
, tx
));
6188 spa_history_log_internal(spa
, "set", tx
,
6189 "%s=%s", nvpair_name(elem
), strval
);
6190 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6191 intval
= fnvpair_value_uint64(elem
);
6193 if (proptype
== PROP_TYPE_INDEX
) {
6195 VERIFY0(zpool_prop_index_to_string(
6196 prop
, intval
, &unused
));
6198 VERIFY0(zap_update(mos
,
6199 spa
->spa_pool_props_object
, propname
,
6200 8, 1, &intval
, tx
));
6201 spa_history_log_internal(spa
, "set", tx
,
6202 "%s=%lld", nvpair_name(elem
), intval
);
6204 ASSERT(0); /* not allowed */
6208 case ZPOOL_PROP_DELEGATION
:
6209 spa
->spa_delegation
= intval
;
6211 case ZPOOL_PROP_BOOTFS
:
6212 spa
->spa_bootfs
= intval
;
6214 case ZPOOL_PROP_FAILUREMODE
:
6215 spa
->spa_failmode
= intval
;
6217 case ZPOOL_PROP_AUTOEXPAND
:
6218 spa
->spa_autoexpand
= intval
;
6219 if (tx
->tx_txg
!= TXG_INITIAL
)
6220 spa_async_request(spa
,
6221 SPA_ASYNC_AUTOEXPAND
);
6223 case ZPOOL_PROP_DEDUPDITTO
:
6224 spa
->spa_dedup_ditto
= intval
;
6233 mutex_exit(&spa
->spa_props_lock
);
6237 * Perform one-time upgrade on-disk changes. spa_version() does not
6238 * reflect the new version this txg, so there must be no changes this
6239 * txg to anything that the upgrade code depends on after it executes.
6240 * Therefore this must be called after dsl_pool_sync() does the sync
6244 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6246 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6248 ASSERT(spa
->spa_sync_pass
== 1);
6250 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6252 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6253 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6254 dsl_pool_create_origin(dp
, tx
);
6256 /* Keeping the origin open increases spa_minref */
6257 spa
->spa_minref
+= 3;
6260 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6261 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6262 dsl_pool_upgrade_clones(dp
, tx
);
6265 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6266 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6267 dsl_pool_upgrade_dir_clones(dp
, tx
);
6269 /* Keeping the freedir open increases spa_minref */
6270 spa
->spa_minref
+= 3;
6273 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6274 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6275 spa_feature_create_zap_objects(spa
, tx
);
6279 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6280 * when possibility to use lz4 compression for metadata was added
6281 * Old pools that have this feature enabled must be upgraded to have
6282 * this feature active
6284 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6285 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6286 SPA_FEATURE_LZ4_COMPRESS
);
6287 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6288 SPA_FEATURE_LZ4_COMPRESS
);
6290 if (lz4_en
&& !lz4_ac
)
6291 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6293 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6297 * Sync the specified transaction group. New blocks may be dirtied as
6298 * part of the process, so we iterate until it converges.
6301 spa_sync(spa_t
*spa
, uint64_t txg
)
6303 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6304 objset_t
*mos
= spa
->spa_meta_objset
;
6305 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6306 vdev_t
*rvd
= spa
->spa_root_vdev
;
6312 VERIFY(spa_writeable(spa
));
6315 * Lock out configuration changes.
6317 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6319 spa
->spa_syncing_txg
= txg
;
6320 spa
->spa_sync_pass
= 0;
6323 * If there are any pending vdev state changes, convert them
6324 * into config changes that go out with this transaction group.
6326 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6327 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6329 * We need the write lock here because, for aux vdevs,
6330 * calling vdev_config_dirty() modifies sav_config.
6331 * This is ugly and will become unnecessary when we
6332 * eliminate the aux vdev wart by integrating all vdevs
6333 * into the root vdev tree.
6335 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6336 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6337 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6338 vdev_state_clean(vd
);
6339 vdev_config_dirty(vd
);
6341 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6342 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6344 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6346 tx
= dmu_tx_create_assigned(dp
, txg
);
6348 spa
->spa_sync_starttime
= gethrtime();
6349 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6350 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
6351 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6352 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6355 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6356 * set spa_deflate if we have no raid-z vdevs.
6358 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6359 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6362 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6363 vd
= rvd
->vdev_child
[i
];
6364 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6367 if (i
== rvd
->vdev_children
) {
6368 spa
->spa_deflate
= TRUE
;
6369 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6370 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6371 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6376 * If anything has changed in this txg, or if someone is waiting
6377 * for this txg to sync (eg, spa_vdev_remove()), push the
6378 * deferred frees from the previous txg. If not, leave them
6379 * alone so that we don't generate work on an otherwise idle
6382 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
6383 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
6384 !txg_list_empty(&dp
->dp_sync_tasks
, txg
) ||
6385 ((dsl_scan_active(dp
->dp_scan
) ||
6386 txg_sync_waiting(dp
)) && !spa_shutting_down(spa
))) {
6387 spa_sync_deferred_frees(spa
, tx
);
6391 * Iterate to convergence.
6394 int pass
= ++spa
->spa_sync_pass
;
6396 spa_sync_config_object(spa
, tx
);
6397 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6398 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6399 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6400 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6401 spa_errlog_sync(spa
, txg
);
6402 dsl_pool_sync(dp
, txg
);
6404 if (pass
< zfs_sync_pass_deferred_free
) {
6405 spa_sync_frees(spa
, free_bpl
, tx
);
6407 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6408 &spa
->spa_deferred_bpobj
, tx
);
6412 dsl_scan_sync(dp
, tx
);
6414 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6418 spa_sync_upgrades(spa
, tx
);
6420 } while (dmu_objset_is_dirty(mos
, txg
));
6423 * Rewrite the vdev configuration (which includes the uberblock)
6424 * to commit the transaction group.
6426 * If there are no dirty vdevs, we sync the uberblock to a few
6427 * random top-level vdevs that are known to be visible in the
6428 * config cache (see spa_vdev_add() for a complete description).
6429 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6433 * We hold SCL_STATE to prevent vdev open/close/etc.
6434 * while we're attempting to write the vdev labels.
6436 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6438 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6439 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6441 int children
= rvd
->vdev_children
;
6442 int c0
= spa_get_random(children
);
6444 for (c
= 0; c
< children
; c
++) {
6445 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6446 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6448 svd
[svdcount
++] = vd
;
6449 if (svdcount
== SPA_DVAS_PER_BP
)
6452 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
6454 error
= vdev_config_sync(svd
, svdcount
, txg
,
6457 error
= vdev_config_sync(rvd
->vdev_child
,
6458 rvd
->vdev_children
, txg
, B_FALSE
);
6460 error
= vdev_config_sync(rvd
->vdev_child
,
6461 rvd
->vdev_children
, txg
, B_TRUE
);
6465 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6467 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6471 zio_suspend(spa
, NULL
);
6472 zio_resume_wait(spa
);
6476 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6477 spa
->spa_deadman_tqid
= 0;
6480 * Clear the dirty config list.
6482 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6483 vdev_config_clean(vd
);
6486 * Now that the new config has synced transactionally,
6487 * let it become visible to the config cache.
6489 if (spa
->spa_config_syncing
!= NULL
) {
6490 spa_config_set(spa
, spa
->spa_config_syncing
);
6491 spa
->spa_config_txg
= txg
;
6492 spa
->spa_config_syncing
= NULL
;
6495 spa
->spa_ubsync
= spa
->spa_uberblock
;
6497 dsl_pool_sync_done(dp
, txg
);
6500 * Update usable space statistics.
6502 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6503 vdev_sync_done(vd
, txg
);
6505 spa_update_dspace(spa
);
6508 * It had better be the case that we didn't dirty anything
6509 * since vdev_config_sync().
6511 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6512 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6513 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6515 spa
->spa_sync_pass
= 0;
6517 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6519 spa_handle_ignored_writes(spa
);
6522 * If any async tasks have been requested, kick them off.
6524 spa_async_dispatch(spa
);
6528 * Sync all pools. We don't want to hold the namespace lock across these
6529 * operations, so we take a reference on the spa_t and drop the lock during the
6533 spa_sync_allpools(void)
6536 mutex_enter(&spa_namespace_lock
);
6537 while ((spa
= spa_next(spa
)) != NULL
) {
6538 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6539 !spa_writeable(spa
) || spa_suspended(spa
))
6541 spa_open_ref(spa
, FTAG
);
6542 mutex_exit(&spa_namespace_lock
);
6543 txg_wait_synced(spa_get_dsl(spa
), 0);
6544 mutex_enter(&spa_namespace_lock
);
6545 spa_close(spa
, FTAG
);
6547 mutex_exit(&spa_namespace_lock
);
6551 * ==========================================================================
6552 * Miscellaneous routines
6553 * ==========================================================================
6557 * Remove all pools in the system.
6565 * Remove all cached state. All pools should be closed now,
6566 * so every spa in the AVL tree should be unreferenced.
6568 mutex_enter(&spa_namespace_lock
);
6569 while ((spa
= spa_next(NULL
)) != NULL
) {
6571 * Stop async tasks. The async thread may need to detach
6572 * a device that's been replaced, which requires grabbing
6573 * spa_namespace_lock, so we must drop it here.
6575 spa_open_ref(spa
, FTAG
);
6576 mutex_exit(&spa_namespace_lock
);
6577 spa_async_suspend(spa
);
6578 mutex_enter(&spa_namespace_lock
);
6579 spa_close(spa
, FTAG
);
6581 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6583 spa_deactivate(spa
);
6587 mutex_exit(&spa_namespace_lock
);
6591 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6596 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6600 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6601 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6602 if (vd
->vdev_guid
== guid
)
6606 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6607 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6608 if (vd
->vdev_guid
== guid
)
6617 spa_upgrade(spa_t
*spa
, uint64_t version
)
6619 ASSERT(spa_writeable(spa
));
6621 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6624 * This should only be called for a non-faulted pool, and since a
6625 * future version would result in an unopenable pool, this shouldn't be
6628 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
6629 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
6631 spa
->spa_uberblock
.ub_version
= version
;
6632 vdev_config_dirty(spa
->spa_root_vdev
);
6634 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6636 txg_wait_synced(spa_get_dsl(spa
), 0);
6640 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6644 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6646 for (i
= 0; i
< sav
->sav_count
; i
++)
6647 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6650 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6651 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6652 &spareguid
) == 0 && spareguid
== guid
)
6660 * Check if a pool has an active shared spare device.
6661 * Note: reference count of an active spare is 2, as a spare and as a replace
6664 spa_has_active_shared_spare(spa_t
*spa
)
6668 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6670 for (i
= 0; i
< sav
->sav_count
; i
++) {
6671 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6672 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6681 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6682 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6683 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6684 * or zdb as real changes.
6687 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6690 zfs_ereport_post(name
, spa
, vd
, NULL
, 0, 0);
6694 #if defined(_KERNEL) && defined(HAVE_SPL)
6695 /* state manipulation functions */
6696 EXPORT_SYMBOL(spa_open
);
6697 EXPORT_SYMBOL(spa_open_rewind
);
6698 EXPORT_SYMBOL(spa_get_stats
);
6699 EXPORT_SYMBOL(spa_create
);
6700 EXPORT_SYMBOL(spa_import_rootpool
);
6701 EXPORT_SYMBOL(spa_import
);
6702 EXPORT_SYMBOL(spa_tryimport
);
6703 EXPORT_SYMBOL(spa_destroy
);
6704 EXPORT_SYMBOL(spa_export
);
6705 EXPORT_SYMBOL(spa_reset
);
6706 EXPORT_SYMBOL(spa_async_request
);
6707 EXPORT_SYMBOL(spa_async_suspend
);
6708 EXPORT_SYMBOL(spa_async_resume
);
6709 EXPORT_SYMBOL(spa_inject_addref
);
6710 EXPORT_SYMBOL(spa_inject_delref
);
6711 EXPORT_SYMBOL(spa_scan_stat_init
);
6712 EXPORT_SYMBOL(spa_scan_get_stats
);
6714 /* device maniion */
6715 EXPORT_SYMBOL(spa_vdev_add
);
6716 EXPORT_SYMBOL(spa_vdev_attach
);
6717 EXPORT_SYMBOL(spa_vdev_detach
);
6718 EXPORT_SYMBOL(spa_vdev_remove
);
6719 EXPORT_SYMBOL(spa_vdev_setpath
);
6720 EXPORT_SYMBOL(spa_vdev_setfru
);
6721 EXPORT_SYMBOL(spa_vdev_split_mirror
);
6723 /* spare statech is global across all pools) */
6724 EXPORT_SYMBOL(spa_spare_add
);
6725 EXPORT_SYMBOL(spa_spare_remove
);
6726 EXPORT_SYMBOL(spa_spare_exists
);
6727 EXPORT_SYMBOL(spa_spare_activate
);
6729 /* L2ARC statech is global across all pools) */
6730 EXPORT_SYMBOL(spa_l2cache_add
);
6731 EXPORT_SYMBOL(spa_l2cache_remove
);
6732 EXPORT_SYMBOL(spa_l2cache_exists
);
6733 EXPORT_SYMBOL(spa_l2cache_activate
);
6734 EXPORT_SYMBOL(spa_l2cache_drop
);
6737 EXPORT_SYMBOL(spa_scan
);
6738 EXPORT_SYMBOL(spa_scan_stop
);
6741 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
6742 EXPORT_SYMBOL(spa_sync_allpools
);
6745 EXPORT_SYMBOL(spa_prop_set
);
6746 EXPORT_SYMBOL(spa_prop_get
);
6747 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
6749 /* asynchronous event notification */
6750 EXPORT_SYMBOL(spa_event_notify
);
6753 #if defined(_KERNEL) && defined(HAVE_SPL)
6754 module_param(spa_load_verify_maxinflight
, int, 0644);
6755 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
6756 "Max concurrent traversal I/Os while verifying pool during import -X");
6758 module_param(spa_load_verify_metadata
, int, 0644);
6759 MODULE_PARM_DESC(spa_load_verify_metadata
,
6760 "Set to traverse metadata on pool import");
6762 module_param(spa_load_verify_data
, int, 0644);
6763 MODULE_PARM_DESC(spa_load_verify_data
,
6764 "Set to traverse data on pool import");