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
, 6);
680 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
688 * If the bootfs property value is dsobj, clear it.
691 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
693 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
694 VERIFY(zap_remove(spa
->spa_meta_objset
,
695 spa
->spa_pool_props_object
,
696 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
703 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
705 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
706 vdev_t
*rvd
= spa
->spa_root_vdev
;
708 ASSERTV(uint64_t *newguid
= arg
);
710 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
711 vdev_state
= rvd
->vdev_state
;
712 spa_config_exit(spa
, SCL_STATE
, FTAG
);
714 if (vdev_state
!= VDEV_STATE_HEALTHY
)
715 return (SET_ERROR(ENXIO
));
717 ASSERT3U(spa_guid(spa
), !=, *newguid
);
723 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
725 uint64_t *newguid
= arg
;
726 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
728 vdev_t
*rvd
= spa
->spa_root_vdev
;
730 oldguid
= spa_guid(spa
);
732 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
733 rvd
->vdev_guid
= *newguid
;
734 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
735 vdev_config_dirty(rvd
);
736 spa_config_exit(spa
, SCL_STATE
, FTAG
);
738 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
743 * Change the GUID for the pool. This is done so that we can later
744 * re-import a pool built from a clone of our own vdevs. We will modify
745 * the root vdev's guid, our own pool guid, and then mark all of our
746 * vdevs dirty. Note that we must make sure that all our vdevs are
747 * online when we do this, or else any vdevs that weren't present
748 * would be orphaned from our pool. We are also going to issue a
749 * sysevent to update any watchers.
752 spa_change_guid(spa_t
*spa
)
757 mutex_enter(&spa
->spa_vdev_top_lock
);
758 mutex_enter(&spa_namespace_lock
);
759 guid
= spa_generate_guid(NULL
);
761 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
762 spa_change_guid_sync
, &guid
, 5);
765 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
766 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_REGUID
);
769 mutex_exit(&spa_namespace_lock
);
770 mutex_exit(&spa
->spa_vdev_top_lock
);
776 * ==========================================================================
777 * SPA state manipulation (open/create/destroy/import/export)
778 * ==========================================================================
782 spa_error_entry_compare(const void *a
, const void *b
)
784 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
785 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
788 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
789 sizeof (zbookmark_phys_t
));
800 * Utility function which retrieves copies of the current logs and
801 * re-initializes them in the process.
804 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
806 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
808 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
809 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
811 avl_create(&spa
->spa_errlist_scrub
,
812 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
813 offsetof(spa_error_entry_t
, se_avl
));
814 avl_create(&spa
->spa_errlist_last
,
815 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
816 offsetof(spa_error_entry_t
, se_avl
));
820 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
822 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
823 enum zti_modes mode
= ztip
->zti_mode
;
824 uint_t value
= ztip
->zti_value
;
825 uint_t count
= ztip
->zti_count
;
826 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
829 boolean_t batch
= B_FALSE
;
831 if (mode
== ZTI_MODE_NULL
) {
833 tqs
->stqs_taskq
= NULL
;
837 ASSERT3U(count
, >, 0);
839 tqs
->stqs_count
= count
;
840 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
844 ASSERT3U(value
, >=, 1);
845 value
= MAX(value
, 1);
850 flags
|= TASKQ_THREADS_CPU_PCT
;
851 value
= zio_taskq_batch_pct
;
855 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
857 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
861 for (i
= 0; i
< count
; i
++) {
865 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
866 zio_type_name
[t
], zio_taskq_types
[q
], i
);
868 (void) snprintf(name
, sizeof (name
), "%s_%s",
869 zio_type_name
[t
], zio_taskq_types
[q
]);
872 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
874 flags
|= TASKQ_DC_BATCH
;
876 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
877 spa
->spa_proc
, zio_taskq_basedc
, flags
);
879 pri_t pri
= maxclsyspri
;
881 * The write issue taskq can be extremely CPU
882 * intensive. Run it at slightly lower priority
883 * than the other taskqs.
885 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
888 tq
= taskq_create_proc(name
, value
, pri
, 50,
889 INT_MAX
, spa
->spa_proc
, flags
);
892 tqs
->stqs_taskq
[i
] = tq
;
897 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
899 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
902 if (tqs
->stqs_taskq
== NULL
) {
903 ASSERT3U(tqs
->stqs_count
, ==, 0);
907 for (i
= 0; i
< tqs
->stqs_count
; i
++) {
908 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
909 taskq_destroy(tqs
->stqs_taskq
[i
]);
912 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
913 tqs
->stqs_taskq
= NULL
;
917 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
918 * Note that a type may have multiple discrete taskqs to avoid lock contention
919 * on the taskq itself. In that case we choose which taskq at random by using
920 * the low bits of gethrtime().
923 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
924 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
926 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
929 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
930 ASSERT3U(tqs
->stqs_count
, !=, 0);
932 if (tqs
->stqs_count
== 1) {
933 tq
= tqs
->stqs_taskq
[0];
935 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
938 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
942 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
945 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
946 task_func_t
*func
, void *arg
, uint_t flags
)
948 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
952 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
953 ASSERT3U(tqs
->stqs_count
, !=, 0);
955 if (tqs
->stqs_count
== 1) {
956 tq
= tqs
->stqs_taskq
[0];
958 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
961 id
= taskq_dispatch(tq
, func
, arg
, flags
);
963 taskq_wait_id(tq
, id
);
967 spa_create_zio_taskqs(spa_t
*spa
)
971 for (t
= 0; t
< ZIO_TYPES
; t
++) {
972 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
973 spa_taskqs_init(spa
, t
, q
);
978 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
980 spa_thread(void *arg
)
985 user_t
*pu
= PTOU(curproc
);
987 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
990 ASSERT(curproc
!= &p0
);
991 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
992 "zpool-%s", spa
->spa_name
);
993 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
995 /* bind this thread to the requested psrset */
996 if (zio_taskq_psrset_bind
!= PS_NONE
) {
998 mutex_enter(&cpu_lock
);
999 mutex_enter(&pidlock
);
1000 mutex_enter(&curproc
->p_lock
);
1002 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1003 0, NULL
, NULL
) == 0) {
1004 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1007 "Couldn't bind process for zfs pool \"%s\" to "
1008 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1011 mutex_exit(&curproc
->p_lock
);
1012 mutex_exit(&pidlock
);
1013 mutex_exit(&cpu_lock
);
1017 if (zio_taskq_sysdc
) {
1018 sysdc_thread_enter(curthread
, 100, 0);
1021 spa
->spa_proc
= curproc
;
1022 spa
->spa_did
= curthread
->t_did
;
1024 spa_create_zio_taskqs(spa
);
1026 mutex_enter(&spa
->spa_proc_lock
);
1027 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1029 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1030 cv_broadcast(&spa
->spa_proc_cv
);
1032 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1033 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1034 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1035 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1037 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1038 spa
->spa_proc_state
= SPA_PROC_GONE
;
1039 spa
->spa_proc
= &p0
;
1040 cv_broadcast(&spa
->spa_proc_cv
);
1041 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1043 mutex_enter(&curproc
->p_lock
);
1049 * Activate an uninitialized pool.
1052 spa_activate(spa_t
*spa
, int mode
)
1054 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1056 spa
->spa_state
= POOL_STATE_ACTIVE
;
1057 spa
->spa_mode
= mode
;
1059 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1060 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1062 /* Try to create a covering process */
1063 mutex_enter(&spa
->spa_proc_lock
);
1064 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1065 ASSERT(spa
->spa_proc
== &p0
);
1068 #ifdef HAVE_SPA_THREAD
1069 /* Only create a process if we're going to be around a while. */
1070 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1071 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1073 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1074 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1075 cv_wait(&spa
->spa_proc_cv
,
1076 &spa
->spa_proc_lock
);
1078 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1079 ASSERT(spa
->spa_proc
!= &p0
);
1080 ASSERT(spa
->spa_did
!= 0);
1084 "Couldn't create process for zfs pool \"%s\"\n",
1089 #endif /* HAVE_SPA_THREAD */
1090 mutex_exit(&spa
->spa_proc_lock
);
1092 /* If we didn't create a process, we need to create our taskqs. */
1093 if (spa
->spa_proc
== &p0
) {
1094 spa_create_zio_taskqs(spa
);
1097 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1098 offsetof(vdev_t
, vdev_config_dirty_node
));
1099 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1100 offsetof(objset_t
, os_evicting_node
));
1101 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1102 offsetof(vdev_t
, vdev_state_dirty_node
));
1104 txg_list_create(&spa
->spa_vdev_txg_list
,
1105 offsetof(struct vdev
, vdev_txg_node
));
1107 avl_create(&spa
->spa_errlist_scrub
,
1108 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1109 offsetof(spa_error_entry_t
, se_avl
));
1110 avl_create(&spa
->spa_errlist_last
,
1111 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1112 offsetof(spa_error_entry_t
, se_avl
));
1116 * Opposite of spa_activate().
1119 spa_deactivate(spa_t
*spa
)
1123 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1124 ASSERT(spa
->spa_dsl_pool
== NULL
);
1125 ASSERT(spa
->spa_root_vdev
== NULL
);
1126 ASSERT(spa
->spa_async_zio_root
== NULL
);
1127 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1129 spa_evicting_os_wait(spa
);
1131 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1133 list_destroy(&spa
->spa_config_dirty_list
);
1134 list_destroy(&spa
->spa_evicting_os_list
);
1135 list_destroy(&spa
->spa_state_dirty_list
);
1137 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
1139 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1140 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1141 spa_taskqs_fini(spa
, t
, q
);
1145 metaslab_class_destroy(spa
->spa_normal_class
);
1146 spa
->spa_normal_class
= NULL
;
1148 metaslab_class_destroy(spa
->spa_log_class
);
1149 spa
->spa_log_class
= NULL
;
1152 * If this was part of an import or the open otherwise failed, we may
1153 * still have errors left in the queues. Empty them just in case.
1155 spa_errlog_drain(spa
);
1157 avl_destroy(&spa
->spa_errlist_scrub
);
1158 avl_destroy(&spa
->spa_errlist_last
);
1160 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1162 mutex_enter(&spa
->spa_proc_lock
);
1163 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1164 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1165 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1166 cv_broadcast(&spa
->spa_proc_cv
);
1167 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1168 ASSERT(spa
->spa_proc
!= &p0
);
1169 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1171 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1172 spa
->spa_proc_state
= SPA_PROC_NONE
;
1174 ASSERT(spa
->spa_proc
== &p0
);
1175 mutex_exit(&spa
->spa_proc_lock
);
1178 * We want to make sure spa_thread() has actually exited the ZFS
1179 * module, so that the module can't be unloaded out from underneath
1182 if (spa
->spa_did
!= 0) {
1183 thread_join(spa
->spa_did
);
1189 * Verify a pool configuration, and construct the vdev tree appropriately. This
1190 * will create all the necessary vdevs in the appropriate layout, with each vdev
1191 * in the CLOSED state. This will prep the pool before open/creation/import.
1192 * All vdev validation is done by the vdev_alloc() routine.
1195 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1196 uint_t id
, int atype
)
1203 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1206 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1209 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1212 if (error
== ENOENT
)
1218 return (SET_ERROR(EINVAL
));
1221 for (c
= 0; c
< children
; c
++) {
1223 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1231 ASSERT(*vdp
!= NULL
);
1237 * Opposite of spa_load().
1240 spa_unload(spa_t
*spa
)
1244 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1249 spa_async_suspend(spa
);
1254 if (spa
->spa_sync_on
) {
1255 txg_sync_stop(spa
->spa_dsl_pool
);
1256 spa
->spa_sync_on
= B_FALSE
;
1260 * Wait for any outstanding async I/O to complete.
1262 if (spa
->spa_async_zio_root
!= NULL
) {
1263 for (i
= 0; i
< max_ncpus
; i
++)
1264 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1265 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1266 spa
->spa_async_zio_root
= NULL
;
1269 bpobj_close(&spa
->spa_deferred_bpobj
);
1271 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1276 if (spa
->spa_root_vdev
)
1277 vdev_free(spa
->spa_root_vdev
);
1278 ASSERT(spa
->spa_root_vdev
== NULL
);
1281 * Close the dsl pool.
1283 if (spa
->spa_dsl_pool
) {
1284 dsl_pool_close(spa
->spa_dsl_pool
);
1285 spa
->spa_dsl_pool
= NULL
;
1286 spa
->spa_meta_objset
= NULL
;
1293 * Drop and purge level 2 cache
1295 spa_l2cache_drop(spa
);
1297 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1298 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1299 if (spa
->spa_spares
.sav_vdevs
) {
1300 kmem_free(spa
->spa_spares
.sav_vdevs
,
1301 spa
->spa_spares
.sav_count
* sizeof (void *));
1302 spa
->spa_spares
.sav_vdevs
= NULL
;
1304 if (spa
->spa_spares
.sav_config
) {
1305 nvlist_free(spa
->spa_spares
.sav_config
);
1306 spa
->spa_spares
.sav_config
= NULL
;
1308 spa
->spa_spares
.sav_count
= 0;
1310 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1311 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1312 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1314 if (spa
->spa_l2cache
.sav_vdevs
) {
1315 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1316 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1317 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1319 if (spa
->spa_l2cache
.sav_config
) {
1320 nvlist_free(spa
->spa_l2cache
.sav_config
);
1321 spa
->spa_l2cache
.sav_config
= NULL
;
1323 spa
->spa_l2cache
.sav_count
= 0;
1325 spa
->spa_async_suspended
= 0;
1327 if (spa
->spa_comment
!= NULL
) {
1328 spa_strfree(spa
->spa_comment
);
1329 spa
->spa_comment
= NULL
;
1332 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1336 * Load (or re-load) the current list of vdevs describing the active spares for
1337 * this pool. When this is called, we have some form of basic information in
1338 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1339 * then re-generate a more complete list including status information.
1342 spa_load_spares(spa_t
*spa
)
1349 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1352 * First, close and free any existing spare vdevs.
1354 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1355 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1357 /* Undo the call to spa_activate() below */
1358 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1359 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1360 spa_spare_remove(tvd
);
1365 if (spa
->spa_spares
.sav_vdevs
)
1366 kmem_free(spa
->spa_spares
.sav_vdevs
,
1367 spa
->spa_spares
.sav_count
* sizeof (void *));
1369 if (spa
->spa_spares
.sav_config
== NULL
)
1372 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1373 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1375 spa
->spa_spares
.sav_count
= (int)nspares
;
1376 spa
->spa_spares
.sav_vdevs
= NULL
;
1382 * Construct the array of vdevs, opening them to get status in the
1383 * process. For each spare, there is potentially two different vdev_t
1384 * structures associated with it: one in the list of spares (used only
1385 * for basic validation purposes) and one in the active vdev
1386 * configuration (if it's spared in). During this phase we open and
1387 * validate each vdev on the spare list. If the vdev also exists in the
1388 * active configuration, then we also mark this vdev as an active spare.
1390 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1392 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1393 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1394 VDEV_ALLOC_SPARE
) == 0);
1397 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1399 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1400 B_FALSE
)) != NULL
) {
1401 if (!tvd
->vdev_isspare
)
1405 * We only mark the spare active if we were successfully
1406 * able to load the vdev. Otherwise, importing a pool
1407 * with a bad active spare would result in strange
1408 * behavior, because multiple pool would think the spare
1409 * is actively in use.
1411 * There is a vulnerability here to an equally bizarre
1412 * circumstance, where a dead active spare is later
1413 * brought back to life (onlined or otherwise). Given
1414 * the rarity of this scenario, and the extra complexity
1415 * it adds, we ignore the possibility.
1417 if (!vdev_is_dead(tvd
))
1418 spa_spare_activate(tvd
);
1422 vd
->vdev_aux
= &spa
->spa_spares
;
1424 if (vdev_open(vd
) != 0)
1427 if (vdev_validate_aux(vd
) == 0)
1432 * Recompute the stashed list of spares, with status information
1435 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1436 DATA_TYPE_NVLIST_ARRAY
) == 0);
1438 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1440 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1441 spares
[i
] = vdev_config_generate(spa
,
1442 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1443 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1444 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1445 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1446 nvlist_free(spares
[i
]);
1447 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1451 * Load (or re-load) the current list of vdevs describing the active l2cache for
1452 * this pool. When this is called, we have some form of basic information in
1453 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1454 * then re-generate a more complete list including status information.
1455 * Devices which are already active have their details maintained, and are
1459 spa_load_l2cache(spa_t
*spa
)
1463 int i
, j
, oldnvdevs
;
1465 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1466 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1468 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1470 if (sav
->sav_config
!= NULL
) {
1471 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1472 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1473 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1479 oldvdevs
= sav
->sav_vdevs
;
1480 oldnvdevs
= sav
->sav_count
;
1481 sav
->sav_vdevs
= NULL
;
1485 * Process new nvlist of vdevs.
1487 for (i
= 0; i
< nl2cache
; i
++) {
1488 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1492 for (j
= 0; j
< oldnvdevs
; j
++) {
1494 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1496 * Retain previous vdev for add/remove ops.
1504 if (newvdevs
[i
] == NULL
) {
1508 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1509 VDEV_ALLOC_L2CACHE
) == 0);
1514 * Commit this vdev as an l2cache device,
1515 * even if it fails to open.
1517 spa_l2cache_add(vd
);
1522 spa_l2cache_activate(vd
);
1524 if (vdev_open(vd
) != 0)
1527 (void) vdev_validate_aux(vd
);
1529 if (!vdev_is_dead(vd
))
1530 l2arc_add_vdev(spa
, vd
);
1535 * Purge vdevs that were dropped
1537 for (i
= 0; i
< oldnvdevs
; i
++) {
1542 ASSERT(vd
->vdev_isl2cache
);
1544 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1545 pool
!= 0ULL && l2arc_vdev_present(vd
))
1546 l2arc_remove_vdev(vd
);
1547 vdev_clear_stats(vd
);
1553 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1555 if (sav
->sav_config
== NULL
)
1558 sav
->sav_vdevs
= newvdevs
;
1559 sav
->sav_count
= (int)nl2cache
;
1562 * Recompute the stashed list of l2cache devices, with status
1563 * information this time.
1565 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1566 DATA_TYPE_NVLIST_ARRAY
) == 0);
1568 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1569 for (i
= 0; i
< sav
->sav_count
; i
++)
1570 l2cache
[i
] = vdev_config_generate(spa
,
1571 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1572 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1573 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1575 for (i
= 0; i
< sav
->sav_count
; i
++)
1576 nvlist_free(l2cache
[i
]);
1578 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1582 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1585 char *packed
= NULL
;
1590 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1594 nvsize
= *(uint64_t *)db
->db_data
;
1595 dmu_buf_rele(db
, FTAG
);
1597 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1598 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1601 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1602 vmem_free(packed
, nvsize
);
1608 * Checks to see if the given vdev could not be opened, in which case we post a
1609 * sysevent to notify the autoreplace code that the device has been removed.
1612 spa_check_removed(vdev_t
*vd
)
1616 for (c
= 0; c
< vd
->vdev_children
; c
++)
1617 spa_check_removed(vd
->vdev_child
[c
]);
1619 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1621 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE
,
1622 vd
->vdev_spa
, vd
, NULL
, 0, 0);
1623 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_CHECK
);
1628 * Validate the current config against the MOS config
1631 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1633 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1637 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1639 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1640 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1642 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1645 * If we're doing a normal import, then build up any additional
1646 * diagnostic information about missing devices in this config.
1647 * We'll pass this up to the user for further processing.
1649 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1650 nvlist_t
**child
, *nv
;
1653 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1655 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1657 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1658 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1659 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1661 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1662 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1664 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1669 VERIFY(nvlist_add_nvlist_array(nv
,
1670 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1671 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1672 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1674 for (i
= 0; i
< idx
; i
++)
1675 nvlist_free(child
[i
]);
1678 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1682 * Compare the root vdev tree with the information we have
1683 * from the MOS config (mrvd). Check each top-level vdev
1684 * with the corresponding MOS config top-level (mtvd).
1686 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1687 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1688 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1691 * Resolve any "missing" vdevs in the current configuration.
1692 * If we find that the MOS config has more accurate information
1693 * about the top-level vdev then use that vdev instead.
1695 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1696 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1698 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1702 * Device specific actions.
1704 if (mtvd
->vdev_islog
) {
1705 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1708 * XXX - once we have 'readonly' pool
1709 * support we should be able to handle
1710 * missing data devices by transitioning
1711 * the pool to readonly.
1717 * Swap the missing vdev with the data we were
1718 * able to obtain from the MOS config.
1720 vdev_remove_child(rvd
, tvd
);
1721 vdev_remove_child(mrvd
, mtvd
);
1723 vdev_add_child(rvd
, mtvd
);
1724 vdev_add_child(mrvd
, tvd
);
1726 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1728 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1731 } else if (mtvd
->vdev_islog
) {
1733 * Load the slog device's state from the MOS config
1734 * since it's possible that the label does not
1735 * contain the most up-to-date information.
1737 vdev_load_log_state(tvd
, mtvd
);
1742 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1745 * Ensure we were able to validate the config.
1747 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1751 * Check for missing log devices
1754 spa_check_logs(spa_t
*spa
)
1756 boolean_t rv
= B_FALSE
;
1758 switch (spa
->spa_log_state
) {
1761 case SPA_LOG_MISSING
:
1762 /* need to recheck in case slog has been restored */
1763 case SPA_LOG_UNKNOWN
:
1764 rv
= (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
,
1765 NULL
, DS_FIND_CHILDREN
) != 0);
1767 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1774 spa_passivate_log(spa_t
*spa
)
1776 vdev_t
*rvd
= spa
->spa_root_vdev
;
1777 boolean_t slog_found
= B_FALSE
;
1780 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1782 if (!spa_has_slogs(spa
))
1785 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1786 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1787 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1789 if (tvd
->vdev_islog
) {
1790 metaslab_group_passivate(mg
);
1791 slog_found
= B_TRUE
;
1795 return (slog_found
);
1799 spa_activate_log(spa_t
*spa
)
1801 vdev_t
*rvd
= spa
->spa_root_vdev
;
1804 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1806 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1807 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1808 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1810 if (tvd
->vdev_islog
)
1811 metaslab_group_activate(mg
);
1816 spa_offline_log(spa_t
*spa
)
1820 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1821 NULL
, DS_FIND_CHILDREN
);
1824 * We successfully offlined the log device, sync out the
1825 * current txg so that the "stubby" block can be removed
1828 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1834 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1838 for (i
= 0; i
< sav
->sav_count
; i
++)
1839 spa_check_removed(sav
->sav_vdevs
[i
]);
1843 spa_claim_notify(zio_t
*zio
)
1845 spa_t
*spa
= zio
->io_spa
;
1850 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1851 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1852 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1853 mutex_exit(&spa
->spa_props_lock
);
1856 typedef struct spa_load_error
{
1857 uint64_t sle_meta_count
;
1858 uint64_t sle_data_count
;
1862 spa_load_verify_done(zio_t
*zio
)
1864 blkptr_t
*bp
= zio
->io_bp
;
1865 spa_load_error_t
*sle
= zio
->io_private
;
1866 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1867 int error
= zio
->io_error
;
1868 spa_t
*spa
= zio
->io_spa
;
1871 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1872 type
!= DMU_OT_INTENT_LOG
)
1873 atomic_add_64(&sle
->sle_meta_count
, 1);
1875 atomic_add_64(&sle
->sle_data_count
, 1);
1877 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1879 mutex_enter(&spa
->spa_scrub_lock
);
1880 spa
->spa_scrub_inflight
--;
1881 cv_broadcast(&spa
->spa_scrub_io_cv
);
1882 mutex_exit(&spa
->spa_scrub_lock
);
1886 * Maximum number of concurrent scrub i/os to create while verifying
1887 * a pool while importing it.
1889 int spa_load_verify_maxinflight
= 10000;
1890 int spa_load_verify_metadata
= B_TRUE
;
1891 int spa_load_verify_data
= B_TRUE
;
1895 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1896 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1902 if (BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
1905 * Note: normally this routine will not be called if
1906 * spa_load_verify_metadata is not set. However, it may be useful
1907 * to manually set the flag after the traversal has begun.
1909 if (!spa_load_verify_metadata
)
1911 if (BP_GET_BUFC_TYPE(bp
) == ARC_BUFC_DATA
&& !spa_load_verify_data
)
1915 size
= BP_GET_PSIZE(bp
);
1916 data
= zio_data_buf_alloc(size
);
1918 mutex_enter(&spa
->spa_scrub_lock
);
1919 while (spa
->spa_scrub_inflight
>= spa_load_verify_maxinflight
)
1920 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
1921 spa
->spa_scrub_inflight
++;
1922 mutex_exit(&spa
->spa_scrub_lock
);
1924 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1925 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1926 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1927 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1932 spa_load_verify(spa_t
*spa
)
1935 spa_load_error_t sle
= { 0 };
1936 zpool_rewind_policy_t policy
;
1937 boolean_t verify_ok
= B_FALSE
;
1940 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1942 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1945 rio
= zio_root(spa
, NULL
, &sle
,
1946 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1948 if (spa_load_verify_metadata
) {
1949 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1950 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
,
1951 spa_load_verify_cb
, rio
);
1954 (void) zio_wait(rio
);
1956 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1957 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1959 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1960 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1964 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1965 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1967 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1968 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1969 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1970 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1971 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1972 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1973 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1975 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1979 if (error
!= ENXIO
&& error
!= EIO
)
1980 error
= SET_ERROR(EIO
);
1984 return (verify_ok
? 0 : EIO
);
1988 * Find a value in the pool props object.
1991 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
1993 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
1994 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
1998 * Find a value in the pool directory object.
2001 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2003 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2004 name
, sizeof (uint64_t), 1, val
));
2008 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2010 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2015 * Fix up config after a partly-completed split. This is done with the
2016 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2017 * pool have that entry in their config, but only the splitting one contains
2018 * a list of all the guids of the vdevs that are being split off.
2020 * This function determines what to do with that list: either rejoin
2021 * all the disks to the pool, or complete the splitting process. To attempt
2022 * the rejoin, each disk that is offlined is marked online again, and
2023 * we do a reopen() call. If the vdev label for every disk that was
2024 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2025 * then we call vdev_split() on each disk, and complete the split.
2027 * Otherwise we leave the config alone, with all the vdevs in place in
2028 * the original pool.
2031 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2038 boolean_t attempt_reopen
;
2040 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2043 /* check that the config is complete */
2044 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2045 &glist
, &gcount
) != 0)
2048 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2050 /* attempt to online all the vdevs & validate */
2051 attempt_reopen
= B_TRUE
;
2052 for (i
= 0; i
< gcount
; i
++) {
2053 if (glist
[i
] == 0) /* vdev is hole */
2056 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2057 if (vd
[i
] == NULL
) {
2059 * Don't bother attempting to reopen the disks;
2060 * just do the split.
2062 attempt_reopen
= B_FALSE
;
2064 /* attempt to re-online it */
2065 vd
[i
]->vdev_offline
= B_FALSE
;
2069 if (attempt_reopen
) {
2070 vdev_reopen(spa
->spa_root_vdev
);
2072 /* check each device to see what state it's in */
2073 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2074 if (vd
[i
] != NULL
&&
2075 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2082 * If every disk has been moved to the new pool, or if we never
2083 * even attempted to look at them, then we split them off for
2086 if (!attempt_reopen
|| gcount
== extracted
) {
2087 for (i
= 0; i
< gcount
; i
++)
2090 vdev_reopen(spa
->spa_root_vdev
);
2093 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2097 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2098 boolean_t mosconfig
)
2100 nvlist_t
*config
= spa
->spa_config
;
2101 char *ereport
= FM_EREPORT_ZFS_POOL
;
2107 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2108 return (SET_ERROR(EINVAL
));
2110 ASSERT(spa
->spa_comment
== NULL
);
2111 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2112 spa
->spa_comment
= spa_strdup(comment
);
2115 * Versioning wasn't explicitly added to the label until later, so if
2116 * it's not present treat it as the initial version.
2118 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2119 &spa
->spa_ubsync
.ub_version
) != 0)
2120 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2122 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2123 &spa
->spa_config_txg
);
2125 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2126 spa_guid_exists(pool_guid
, 0)) {
2127 error
= SET_ERROR(EEXIST
);
2129 spa
->spa_config_guid
= pool_guid
;
2131 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2133 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2137 nvlist_free(spa
->spa_load_info
);
2138 spa
->spa_load_info
= fnvlist_alloc();
2140 gethrestime(&spa
->spa_loaded_ts
);
2141 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2142 mosconfig
, &ereport
);
2146 * Don't count references from objsets that are already closed
2147 * and are making their way through the eviction process.
2149 spa_evicting_os_wait(spa
);
2150 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2152 if (error
!= EEXIST
) {
2153 spa
->spa_loaded_ts
.tv_sec
= 0;
2154 spa
->spa_loaded_ts
.tv_nsec
= 0;
2156 if (error
!= EBADF
) {
2157 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2160 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2167 * Load an existing storage pool, using the pool's builtin spa_config as a
2168 * source of configuration information.
2170 __attribute__((always_inline
))
2172 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2173 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2177 nvlist_t
*nvroot
= NULL
;
2180 uberblock_t
*ub
= &spa
->spa_uberblock
;
2181 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2182 int orig_mode
= spa
->spa_mode
;
2185 boolean_t missing_feat_write
= B_FALSE
;
2188 * If this is an untrusted config, access the pool in read-only mode.
2189 * This prevents things like resilvering recently removed devices.
2192 spa
->spa_mode
= FREAD
;
2194 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2196 spa
->spa_load_state
= state
;
2198 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2199 return (SET_ERROR(EINVAL
));
2201 parse
= (type
== SPA_IMPORT_EXISTING
?
2202 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2205 * Create "The Godfather" zio to hold all async IOs
2207 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2209 for (i
= 0; i
< max_ncpus
; i
++) {
2210 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2211 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2212 ZIO_FLAG_GODFATHER
);
2216 * Parse the configuration into a vdev tree. We explicitly set the
2217 * value that will be returned by spa_version() since parsing the
2218 * configuration requires knowing the version number.
2220 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2221 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2222 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2227 ASSERT(spa
->spa_root_vdev
== rvd
);
2229 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2230 ASSERT(spa_guid(spa
) == pool_guid
);
2234 * Try to open all vdevs, loading each label in the process.
2236 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2237 error
= vdev_open(rvd
);
2238 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2243 * We need to validate the vdev labels against the configuration that
2244 * we have in hand, which is dependent on the setting of mosconfig. If
2245 * mosconfig is true then we're validating the vdev labels based on
2246 * that config. Otherwise, we're validating against the cached config
2247 * (zpool.cache) that was read when we loaded the zfs module, and then
2248 * later we will recursively call spa_load() and validate against
2251 * If we're assembling a new pool that's been split off from an
2252 * existing pool, the labels haven't yet been updated so we skip
2253 * validation for now.
2255 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2256 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2257 error
= vdev_validate(rvd
, mosconfig
);
2258 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2263 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2264 return (SET_ERROR(ENXIO
));
2268 * Find the best uberblock.
2270 vdev_uberblock_load(rvd
, ub
, &label
);
2273 * If we weren't able to find a single valid uberblock, return failure.
2275 if (ub
->ub_txg
== 0) {
2277 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2281 * If the pool has an unsupported version we can't open it.
2283 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2285 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2288 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2292 * If we weren't able to find what's necessary for reading the
2293 * MOS in the label, return failure.
2295 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2296 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2298 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2303 * Update our in-core representation with the definitive values
2306 nvlist_free(spa
->spa_label_features
);
2307 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2313 * Look through entries in the label nvlist's features_for_read. If
2314 * there is a feature listed there which we don't understand then we
2315 * cannot open a pool.
2317 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2318 nvlist_t
*unsup_feat
;
2321 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2324 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2326 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2327 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2328 VERIFY(nvlist_add_string(unsup_feat
,
2329 nvpair_name(nvp
), "") == 0);
2333 if (!nvlist_empty(unsup_feat
)) {
2334 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2335 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2336 nvlist_free(unsup_feat
);
2337 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2341 nvlist_free(unsup_feat
);
2345 * If the vdev guid sum doesn't match the uberblock, we have an
2346 * incomplete configuration. We first check to see if the pool
2347 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2348 * If it is, defer the vdev_guid_sum check till later so we
2349 * can handle missing vdevs.
2351 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2352 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2353 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2354 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2356 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2357 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2358 spa_try_repair(spa
, config
);
2359 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2360 nvlist_free(spa
->spa_config_splitting
);
2361 spa
->spa_config_splitting
= NULL
;
2365 * Initialize internal SPA structures.
2367 spa
->spa_state
= POOL_STATE_ACTIVE
;
2368 spa
->spa_ubsync
= spa
->spa_uberblock
;
2369 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2370 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2371 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2372 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2373 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2374 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2376 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2378 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2379 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2381 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2382 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2384 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2385 boolean_t missing_feat_read
= B_FALSE
;
2386 nvlist_t
*unsup_feat
, *enabled_feat
;
2389 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2390 &spa
->spa_feat_for_read_obj
) != 0) {
2391 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2394 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2395 &spa
->spa_feat_for_write_obj
) != 0) {
2396 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2399 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2400 &spa
->spa_feat_desc_obj
) != 0) {
2401 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2404 enabled_feat
= fnvlist_alloc();
2405 unsup_feat
= fnvlist_alloc();
2407 if (!spa_features_check(spa
, B_FALSE
,
2408 unsup_feat
, enabled_feat
))
2409 missing_feat_read
= B_TRUE
;
2411 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2412 if (!spa_features_check(spa
, B_TRUE
,
2413 unsup_feat
, enabled_feat
)) {
2414 missing_feat_write
= B_TRUE
;
2418 fnvlist_add_nvlist(spa
->spa_load_info
,
2419 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2421 if (!nvlist_empty(unsup_feat
)) {
2422 fnvlist_add_nvlist(spa
->spa_load_info
,
2423 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2426 fnvlist_free(enabled_feat
);
2427 fnvlist_free(unsup_feat
);
2429 if (!missing_feat_read
) {
2430 fnvlist_add_boolean(spa
->spa_load_info
,
2431 ZPOOL_CONFIG_CAN_RDONLY
);
2435 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2436 * twofold: to determine whether the pool is available for
2437 * import in read-write mode and (if it is not) whether the
2438 * pool is available for import in read-only mode. If the pool
2439 * is available for import in read-write mode, it is displayed
2440 * as available in userland; if it is not available for import
2441 * in read-only mode, it is displayed as unavailable in
2442 * userland. If the pool is available for import in read-only
2443 * mode but not read-write mode, it is displayed as unavailable
2444 * in userland with a special note that the pool is actually
2445 * available for open in read-only mode.
2447 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2448 * missing a feature for write, we must first determine whether
2449 * the pool can be opened read-only before returning to
2450 * userland in order to know whether to display the
2451 * abovementioned note.
2453 if (missing_feat_read
|| (missing_feat_write
&&
2454 spa_writeable(spa
))) {
2455 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2460 * Load refcounts for ZFS features from disk into an in-memory
2461 * cache during SPA initialization.
2463 for (i
= 0; i
< SPA_FEATURES
; i
++) {
2466 error
= feature_get_refcount_from_disk(spa
,
2467 &spa_feature_table
[i
], &refcount
);
2469 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2470 } else if (error
== ENOTSUP
) {
2471 spa
->spa_feat_refcount_cache
[i
] =
2472 SPA_FEATURE_DISABLED
;
2474 return (spa_vdev_err(rvd
,
2475 VDEV_AUX_CORRUPT_DATA
, EIO
));
2480 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2481 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2482 &spa
->spa_feat_enabled_txg_obj
) != 0)
2483 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2486 spa
->spa_is_initializing
= B_TRUE
;
2487 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2488 spa
->spa_is_initializing
= B_FALSE
;
2490 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2494 nvlist_t
*policy
= NULL
, *nvconfig
;
2496 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2497 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2499 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2500 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2502 unsigned long myhostid
= 0;
2504 VERIFY(nvlist_lookup_string(nvconfig
,
2505 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2508 myhostid
= zone_get_hostid(NULL
);
2511 * We're emulating the system's hostid in userland, so
2512 * we can't use zone_get_hostid().
2514 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2515 #endif /* _KERNEL */
2516 if (hostid
!= 0 && myhostid
!= 0 &&
2517 hostid
!= myhostid
) {
2518 nvlist_free(nvconfig
);
2519 cmn_err(CE_WARN
, "pool '%s' could not be "
2520 "loaded as it was last accessed by another "
2521 "system (host: %s hostid: 0x%lx). See: "
2522 "http://zfsonlinux.org/msg/ZFS-8000-EY",
2523 spa_name(spa
), hostname
,
2524 (unsigned long)hostid
);
2525 return (SET_ERROR(EBADF
));
2528 if (nvlist_lookup_nvlist(spa
->spa_config
,
2529 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2530 VERIFY(nvlist_add_nvlist(nvconfig
,
2531 ZPOOL_REWIND_POLICY
, policy
) == 0);
2533 spa_config_set(spa
, nvconfig
);
2535 spa_deactivate(spa
);
2536 spa_activate(spa
, orig_mode
);
2538 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2541 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2542 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2543 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2545 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2548 * Load the bit that tells us to use the new accounting function
2549 * (raid-z deflation). If we have an older pool, this will not
2552 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2553 if (error
!= 0 && error
!= ENOENT
)
2554 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2556 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2557 &spa
->spa_creation_version
);
2558 if (error
!= 0 && error
!= ENOENT
)
2559 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2562 * Load the persistent error log. If we have an older pool, this will
2565 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2566 if (error
!= 0 && error
!= ENOENT
)
2567 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2569 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2570 &spa
->spa_errlog_scrub
);
2571 if (error
!= 0 && error
!= ENOENT
)
2572 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2575 * Load the history object. If we have an older pool, this
2576 * will not be present.
2578 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2579 if (error
!= 0 && error
!= ENOENT
)
2580 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2583 * If we're assembling the pool from the split-off vdevs of
2584 * an existing pool, we don't want to attach the spares & cache
2589 * Load any hot spares for this pool.
2591 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2592 if (error
!= 0 && error
!= ENOENT
)
2593 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2594 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2595 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2596 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2597 &spa
->spa_spares
.sav_config
) != 0)
2598 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2600 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2601 spa_load_spares(spa
);
2602 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2603 } else if (error
== 0) {
2604 spa
->spa_spares
.sav_sync
= B_TRUE
;
2608 * Load any level 2 ARC devices for this pool.
2610 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2611 &spa
->spa_l2cache
.sav_object
);
2612 if (error
!= 0 && error
!= ENOENT
)
2613 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2614 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2615 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2616 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2617 &spa
->spa_l2cache
.sav_config
) != 0)
2618 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2620 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2621 spa_load_l2cache(spa
);
2622 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2623 } else if (error
== 0) {
2624 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2627 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2629 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2630 if (error
&& error
!= ENOENT
)
2631 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2634 uint64_t autoreplace
= 0;
2636 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2637 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2638 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2639 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2640 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2641 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2642 &spa
->spa_dedup_ditto
);
2644 spa
->spa_autoreplace
= (autoreplace
!= 0);
2648 * If the 'autoreplace' property is set, then post a resource notifying
2649 * the ZFS DE that it should not issue any faults for unopenable
2650 * devices. We also iterate over the vdevs, and post a sysevent for any
2651 * unopenable vdevs so that the normal autoreplace handler can take
2654 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2655 spa_check_removed(spa
->spa_root_vdev
);
2657 * For the import case, this is done in spa_import(), because
2658 * at this point we're using the spare definitions from
2659 * the MOS config, not necessarily from the userland config.
2661 if (state
!= SPA_LOAD_IMPORT
) {
2662 spa_aux_check_removed(&spa
->spa_spares
);
2663 spa_aux_check_removed(&spa
->spa_l2cache
);
2668 * Load the vdev state for all toplevel vdevs.
2673 * Propagate the leaf DTLs we just loaded all the way up the tree.
2675 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2676 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2677 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2680 * Load the DDTs (dedup tables).
2682 error
= ddt_load(spa
);
2684 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2686 spa_update_dspace(spa
);
2689 * Validate the config, using the MOS config to fill in any
2690 * information which might be missing. If we fail to validate
2691 * the config then declare the pool unfit for use. If we're
2692 * assembling a pool from a split, the log is not transferred
2695 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2698 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2699 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2701 if (!spa_config_valid(spa
, nvconfig
)) {
2702 nvlist_free(nvconfig
);
2703 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2706 nvlist_free(nvconfig
);
2709 * Now that we've validated the config, check the state of the
2710 * root vdev. If it can't be opened, it indicates one or
2711 * more toplevel vdevs are faulted.
2713 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2714 return (SET_ERROR(ENXIO
));
2716 if (spa_check_logs(spa
)) {
2717 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2718 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2722 if (missing_feat_write
) {
2723 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2726 * At this point, we know that we can open the pool in
2727 * read-only mode but not read-write mode. We now have enough
2728 * information and can return to userland.
2730 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2734 * We've successfully opened the pool, verify that we're ready
2735 * to start pushing transactions.
2737 if (state
!= SPA_LOAD_TRYIMPORT
) {
2738 if ((error
= spa_load_verify(spa
)))
2739 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2743 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2744 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2746 int need_update
= B_FALSE
;
2749 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2752 * Claim log blocks that haven't been committed yet.
2753 * This must all happen in a single txg.
2754 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2755 * invoked from zil_claim_log_block()'s i/o done callback.
2756 * Price of rollback is that we abandon the log.
2758 spa
->spa_claiming
= B_TRUE
;
2760 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
2761 spa_first_txg(spa
));
2762 (void) dmu_objset_find(spa_name(spa
),
2763 zil_claim
, tx
, DS_FIND_CHILDREN
);
2766 spa
->spa_claiming
= B_FALSE
;
2768 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2769 spa
->spa_sync_on
= B_TRUE
;
2770 txg_sync_start(spa
->spa_dsl_pool
);
2773 * Wait for all claims to sync. We sync up to the highest
2774 * claimed log block birth time so that claimed log blocks
2775 * don't appear to be from the future. spa_claim_max_txg
2776 * will have been set for us by either zil_check_log_chain()
2777 * (invoked from spa_check_logs()) or zil_claim() above.
2779 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2782 * If the config cache is stale, or we have uninitialized
2783 * metaslabs (see spa_vdev_add()), then update the config.
2785 * If this is a verbatim import, trust the current
2786 * in-core spa_config and update the disk labels.
2788 if (config_cache_txg
!= spa
->spa_config_txg
||
2789 state
== SPA_LOAD_IMPORT
||
2790 state
== SPA_LOAD_RECOVER
||
2791 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2792 need_update
= B_TRUE
;
2794 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2795 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2796 need_update
= B_TRUE
;
2799 * Update the config cache asychronously in case we're the
2800 * root pool, in which case the config cache isn't writable yet.
2803 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2806 * Check all DTLs to see if anything needs resilvering.
2808 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2809 vdev_resilver_needed(rvd
, NULL
, NULL
))
2810 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2813 * Log the fact that we booted up (so that we can detect if
2814 * we rebooted in the middle of an operation).
2816 spa_history_log_version(spa
, "open");
2819 * Delete any inconsistent datasets.
2821 (void) dmu_objset_find(spa_name(spa
),
2822 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2825 * Clean up any stale temporary dataset userrefs.
2827 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2834 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2836 int mode
= spa
->spa_mode
;
2839 spa_deactivate(spa
);
2841 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
2843 spa_activate(spa
, mode
);
2844 spa_async_suspend(spa
);
2846 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2850 * If spa_load() fails this function will try loading prior txg's. If
2851 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2852 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2853 * function will not rewind the pool and will return the same error as
2857 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2858 uint64_t max_request
, int rewind_flags
)
2860 nvlist_t
*loadinfo
= NULL
;
2861 nvlist_t
*config
= NULL
;
2862 int load_error
, rewind_error
;
2863 uint64_t safe_rewind_txg
;
2866 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2867 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2868 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2870 spa
->spa_load_max_txg
= max_request
;
2871 if (max_request
!= UINT64_MAX
)
2872 spa
->spa_extreme_rewind
= B_TRUE
;
2875 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2877 if (load_error
== 0)
2880 if (spa
->spa_root_vdev
!= NULL
)
2881 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2883 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2884 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2886 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2887 nvlist_free(config
);
2888 return (load_error
);
2891 if (state
== SPA_LOAD_RECOVER
) {
2892 /* Price of rolling back is discarding txgs, including log */
2893 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2896 * If we aren't rolling back save the load info from our first
2897 * import attempt so that we can restore it after attempting
2900 loadinfo
= spa
->spa_load_info
;
2901 spa
->spa_load_info
= fnvlist_alloc();
2904 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2905 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2906 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2907 TXG_INITIAL
: safe_rewind_txg
;
2910 * Continue as long as we're finding errors, we're still within
2911 * the acceptable rewind range, and we're still finding uberblocks
2913 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2914 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2915 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2916 spa
->spa_extreme_rewind
= B_TRUE
;
2917 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2920 spa
->spa_extreme_rewind
= B_FALSE
;
2921 spa
->spa_load_max_txg
= UINT64_MAX
;
2923 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2924 spa_config_set(spa
, config
);
2926 if (state
== SPA_LOAD_RECOVER
) {
2927 ASSERT3P(loadinfo
, ==, NULL
);
2928 return (rewind_error
);
2930 /* Store the rewind info as part of the initial load info */
2931 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
2932 spa
->spa_load_info
);
2934 /* Restore the initial load info */
2935 fnvlist_free(spa
->spa_load_info
);
2936 spa
->spa_load_info
= loadinfo
;
2938 return (load_error
);
2945 * The import case is identical to an open except that the configuration is sent
2946 * down from userland, instead of grabbed from the configuration cache. For the
2947 * case of an open, the pool configuration will exist in the
2948 * POOL_STATE_UNINITIALIZED state.
2950 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2951 * the same time open the pool, without having to keep around the spa_t in some
2955 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2959 spa_load_state_t state
= SPA_LOAD_OPEN
;
2961 int locked
= B_FALSE
;
2962 int firstopen
= B_FALSE
;
2967 * As disgusting as this is, we need to support recursive calls to this
2968 * function because dsl_dir_open() is called during spa_load(), and ends
2969 * up calling spa_open() again. The real fix is to figure out how to
2970 * avoid dsl_dir_open() calling this in the first place.
2972 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2973 mutex_enter(&spa_namespace_lock
);
2977 if ((spa
= spa_lookup(pool
)) == NULL
) {
2979 mutex_exit(&spa_namespace_lock
);
2980 return (SET_ERROR(ENOENT
));
2983 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
2984 zpool_rewind_policy_t policy
;
2988 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
2990 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
2991 state
= SPA_LOAD_RECOVER
;
2993 spa_activate(spa
, spa_mode_global
);
2995 if (state
!= SPA_LOAD_RECOVER
)
2996 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
2998 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
2999 policy
.zrp_request
);
3001 if (error
== EBADF
) {
3003 * If vdev_validate() returns failure (indicated by
3004 * EBADF), it indicates that one of the vdevs indicates
3005 * that the pool has been exported or destroyed. If
3006 * this is the case, the config cache is out of sync and
3007 * we should remove the pool from the namespace.
3010 spa_deactivate(spa
);
3011 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3014 mutex_exit(&spa_namespace_lock
);
3015 return (SET_ERROR(ENOENT
));
3020 * We can't open the pool, but we still have useful
3021 * information: the state of each vdev after the
3022 * attempted vdev_open(). Return this to the user.
3024 if (config
!= NULL
&& spa
->spa_config
) {
3025 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3027 VERIFY(nvlist_add_nvlist(*config
,
3028 ZPOOL_CONFIG_LOAD_INFO
,
3029 spa
->spa_load_info
) == 0);
3032 spa_deactivate(spa
);
3033 spa
->spa_last_open_failed
= error
;
3035 mutex_exit(&spa_namespace_lock
);
3041 spa_open_ref(spa
, tag
);
3044 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3047 * If we've recovered the pool, pass back any information we
3048 * gathered while doing the load.
3050 if (state
== SPA_LOAD_RECOVER
) {
3051 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3052 spa
->spa_load_info
) == 0);
3056 spa
->spa_last_open_failed
= 0;
3057 spa
->spa_last_ubsync_txg
= 0;
3058 spa
->spa_load_txg
= 0;
3059 mutex_exit(&spa_namespace_lock
);
3064 zvol_create_minors(spa
->spa_name
);
3073 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3076 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3080 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3082 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3086 * Lookup the given spa_t, incrementing the inject count in the process,
3087 * preventing it from being exported or destroyed.
3090 spa_inject_addref(char *name
)
3094 mutex_enter(&spa_namespace_lock
);
3095 if ((spa
= spa_lookup(name
)) == NULL
) {
3096 mutex_exit(&spa_namespace_lock
);
3099 spa
->spa_inject_ref
++;
3100 mutex_exit(&spa_namespace_lock
);
3106 spa_inject_delref(spa_t
*spa
)
3108 mutex_enter(&spa_namespace_lock
);
3109 spa
->spa_inject_ref
--;
3110 mutex_exit(&spa_namespace_lock
);
3114 * Add spares device information to the nvlist.
3117 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3127 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3129 if (spa
->spa_spares
.sav_count
== 0)
3132 VERIFY(nvlist_lookup_nvlist(config
,
3133 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3134 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3135 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3137 VERIFY(nvlist_add_nvlist_array(nvroot
,
3138 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3139 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3140 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3143 * Go through and find any spares which have since been
3144 * repurposed as an active spare. If this is the case, update
3145 * their status appropriately.
3147 for (i
= 0; i
< nspares
; i
++) {
3148 VERIFY(nvlist_lookup_uint64(spares
[i
],
3149 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3150 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3152 VERIFY(nvlist_lookup_uint64_array(
3153 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3154 (uint64_t **)&vs
, &vsc
) == 0);
3155 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3156 vs
->vs_aux
= VDEV_AUX_SPARED
;
3163 * Add l2cache device information to the nvlist, including vdev stats.
3166 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3169 uint_t i
, j
, nl2cache
;
3176 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3178 if (spa
->spa_l2cache
.sav_count
== 0)
3181 VERIFY(nvlist_lookup_nvlist(config
,
3182 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3183 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3184 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3185 if (nl2cache
!= 0) {
3186 VERIFY(nvlist_add_nvlist_array(nvroot
,
3187 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3188 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3189 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3192 * Update level 2 cache device stats.
3195 for (i
= 0; i
< nl2cache
; i
++) {
3196 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3197 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3200 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3202 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3203 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3209 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3210 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3212 vdev_get_stats(vd
, vs
);
3218 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3223 if (spa
->spa_feat_for_read_obj
!= 0) {
3224 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3225 spa
->spa_feat_for_read_obj
);
3226 zap_cursor_retrieve(&zc
, &za
) == 0;
3227 zap_cursor_advance(&zc
)) {
3228 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3229 za
.za_num_integers
== 1);
3230 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3231 za
.za_first_integer
));
3233 zap_cursor_fini(&zc
);
3236 if (spa
->spa_feat_for_write_obj
!= 0) {
3237 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3238 spa
->spa_feat_for_write_obj
);
3239 zap_cursor_retrieve(&zc
, &za
) == 0;
3240 zap_cursor_advance(&zc
)) {
3241 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3242 za
.za_num_integers
== 1);
3243 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3244 za
.za_first_integer
));
3246 zap_cursor_fini(&zc
);
3251 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3255 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3256 zfeature_info_t feature
= spa_feature_table
[i
];
3259 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3262 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3267 * Store a list of pool features and their reference counts in the
3270 * The first time this is called on a spa, allocate a new nvlist, fetch
3271 * the pool features and reference counts from disk, then save the list
3272 * in the spa. In subsequent calls on the same spa use the saved nvlist
3273 * and refresh its values from the cached reference counts. This
3274 * ensures we don't block here on I/O on a suspended pool so 'zpool
3275 * clear' can resume the pool.
3278 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3282 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3284 mutex_enter(&spa
->spa_feat_stats_lock
);
3285 features
= spa
->spa_feat_stats
;
3287 if (features
!= NULL
) {
3288 spa_feature_stats_from_cache(spa
, features
);
3290 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3291 spa
->spa_feat_stats
= features
;
3292 spa_feature_stats_from_disk(spa
, features
);
3295 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3298 mutex_exit(&spa
->spa_feat_stats_lock
);
3302 spa_get_stats(const char *name
, nvlist_t
**config
,
3303 char *altroot
, size_t buflen
)
3309 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3313 * This still leaves a window of inconsistency where the spares
3314 * or l2cache devices could change and the config would be
3315 * self-inconsistent.
3317 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3319 if (*config
!= NULL
) {
3320 uint64_t loadtimes
[2];
3322 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3323 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3324 VERIFY(nvlist_add_uint64_array(*config
,
3325 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3327 VERIFY(nvlist_add_uint64(*config
,
3328 ZPOOL_CONFIG_ERRCOUNT
,
3329 spa_get_errlog_size(spa
)) == 0);
3331 if (spa_suspended(spa
))
3332 VERIFY(nvlist_add_uint64(*config
,
3333 ZPOOL_CONFIG_SUSPENDED
,
3334 spa
->spa_failmode
) == 0);
3336 spa_add_spares(spa
, *config
);
3337 spa_add_l2cache(spa
, *config
);
3338 spa_add_feature_stats(spa
, *config
);
3343 * We want to get the alternate root even for faulted pools, so we cheat
3344 * and call spa_lookup() directly.
3348 mutex_enter(&spa_namespace_lock
);
3349 spa
= spa_lookup(name
);
3351 spa_altroot(spa
, altroot
, buflen
);
3355 mutex_exit(&spa_namespace_lock
);
3357 spa_altroot(spa
, altroot
, buflen
);
3362 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3363 spa_close(spa
, FTAG
);
3370 * Validate that the auxiliary device array is well formed. We must have an
3371 * array of nvlists, each which describes a valid leaf vdev. If this is an
3372 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3373 * specified, as long as they are well-formed.
3376 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3377 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3378 vdev_labeltype_t label
)
3385 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3388 * It's acceptable to have no devs specified.
3390 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3394 return (SET_ERROR(EINVAL
));
3397 * Make sure the pool is formatted with a version that supports this
3400 if (spa_version(spa
) < version
)
3401 return (SET_ERROR(ENOTSUP
));
3404 * Set the pending device list so we correctly handle device in-use
3407 sav
->sav_pending
= dev
;
3408 sav
->sav_npending
= ndev
;
3410 for (i
= 0; i
< ndev
; i
++) {
3411 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3415 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3417 error
= SET_ERROR(EINVAL
);
3422 * The L2ARC currently only supports disk devices in
3423 * kernel context. For user-level testing, we allow it.
3426 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3427 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3428 error
= SET_ERROR(ENOTBLK
);
3435 if ((error
= vdev_open(vd
)) == 0 &&
3436 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3437 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3438 vd
->vdev_guid
) == 0);
3444 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3451 sav
->sav_pending
= NULL
;
3452 sav
->sav_npending
= 0;
3457 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3461 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3463 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3464 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3465 VDEV_LABEL_SPARE
)) != 0) {
3469 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3470 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3471 VDEV_LABEL_L2CACHE
));
3475 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3480 if (sav
->sav_config
!= NULL
) {
3486 * Generate new dev list by concatentating with the
3489 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3490 &olddevs
, &oldndevs
) == 0);
3492 newdevs
= kmem_alloc(sizeof (void *) *
3493 (ndevs
+ oldndevs
), KM_SLEEP
);
3494 for (i
= 0; i
< oldndevs
; i
++)
3495 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3497 for (i
= 0; i
< ndevs
; i
++)
3498 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3501 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3502 DATA_TYPE_NVLIST_ARRAY
) == 0);
3504 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3505 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3506 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3507 nvlist_free(newdevs
[i
]);
3508 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3511 * Generate a new dev list.
3513 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3515 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3521 * Stop and drop level 2 ARC devices
3524 spa_l2cache_drop(spa_t
*spa
)
3528 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3530 for (i
= 0; i
< sav
->sav_count
; i
++) {
3533 vd
= sav
->sav_vdevs
[i
];
3536 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3537 pool
!= 0ULL && l2arc_vdev_present(vd
))
3538 l2arc_remove_vdev(vd
);
3546 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3550 char *altroot
= NULL
;
3555 uint64_t txg
= TXG_INITIAL
;
3556 nvlist_t
**spares
, **l2cache
;
3557 uint_t nspares
, nl2cache
;
3558 uint64_t version
, obj
;
3559 boolean_t has_features
;
3565 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
3566 poolname
= (char *)pool
;
3569 * If this pool already exists, return failure.
3571 mutex_enter(&spa_namespace_lock
);
3572 if (spa_lookup(poolname
) != NULL
) {
3573 mutex_exit(&spa_namespace_lock
);
3574 return (SET_ERROR(EEXIST
));
3578 * Allocate a new spa_t structure.
3580 nvl
= fnvlist_alloc();
3581 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
3582 (void) nvlist_lookup_string(props
,
3583 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3584 spa
= spa_add(poolname
, nvl
, altroot
);
3586 spa_activate(spa
, spa_mode_global
);
3588 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3589 spa_deactivate(spa
);
3591 mutex_exit(&spa_namespace_lock
);
3596 * Temporary pool names should never be written to disk.
3598 if (poolname
!= pool
)
3599 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
3601 has_features
= B_FALSE
;
3602 for (elem
= nvlist_next_nvpair(props
, NULL
);
3603 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3604 if (zpool_prop_feature(nvpair_name(elem
)))
3605 has_features
= B_TRUE
;
3608 if (has_features
|| nvlist_lookup_uint64(props
,
3609 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3610 version
= SPA_VERSION
;
3612 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3614 spa
->spa_first_txg
= txg
;
3615 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3616 spa
->spa_uberblock
.ub_version
= version
;
3617 spa
->spa_ubsync
= spa
->spa_uberblock
;
3620 * Create "The Godfather" zio to hold all async IOs
3622 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3624 for (i
= 0; i
< max_ncpus
; i
++) {
3625 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3626 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3627 ZIO_FLAG_GODFATHER
);
3631 * Create the root vdev.
3633 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3635 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3637 ASSERT(error
!= 0 || rvd
!= NULL
);
3638 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3640 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3641 error
= SET_ERROR(EINVAL
);
3644 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3645 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3646 VDEV_ALLOC_ADD
)) == 0) {
3647 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
3648 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3649 vdev_expand(rvd
->vdev_child
[c
], txg
);
3653 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3657 spa_deactivate(spa
);
3659 mutex_exit(&spa_namespace_lock
);
3664 * Get the list of spares, if specified.
3666 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3667 &spares
, &nspares
) == 0) {
3668 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3670 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3671 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3672 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3673 spa_load_spares(spa
);
3674 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3675 spa
->spa_spares
.sav_sync
= B_TRUE
;
3679 * Get the list of level 2 cache devices, if specified.
3681 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3682 &l2cache
, &nl2cache
) == 0) {
3683 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3684 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3685 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3686 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3687 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3688 spa_load_l2cache(spa
);
3689 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3690 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3693 spa
->spa_is_initializing
= B_TRUE
;
3694 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3695 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3696 spa
->spa_is_initializing
= B_FALSE
;
3699 * Create DDTs (dedup tables).
3703 spa_update_dspace(spa
);
3705 tx
= dmu_tx_create_assigned(dp
, txg
);
3708 * Create the pool config object.
3710 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3711 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3712 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3714 if (zap_add(spa
->spa_meta_objset
,
3715 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3716 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3717 cmn_err(CE_PANIC
, "failed to add pool config");
3720 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3721 spa_feature_create_zap_objects(spa
, tx
);
3723 if (zap_add(spa
->spa_meta_objset
,
3724 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3725 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3726 cmn_err(CE_PANIC
, "failed to add pool version");
3729 /* Newly created pools with the right version are always deflated. */
3730 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3731 spa
->spa_deflate
= TRUE
;
3732 if (zap_add(spa
->spa_meta_objset
,
3733 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3734 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3735 cmn_err(CE_PANIC
, "failed to add deflate");
3740 * Create the deferred-free bpobj. Turn off compression
3741 * because sync-to-convergence takes longer if the blocksize
3744 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3745 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3746 ZIO_COMPRESS_OFF
, tx
);
3747 if (zap_add(spa
->spa_meta_objset
,
3748 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3749 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3750 cmn_err(CE_PANIC
, "failed to add bpobj");
3752 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3753 spa
->spa_meta_objset
, obj
));
3756 * Create the pool's history object.
3758 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3759 spa_history_create_obj(spa
, tx
);
3762 * Set pool properties.
3764 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3765 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3766 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3767 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3769 if (props
!= NULL
) {
3770 spa_configfile_set(spa
, props
, B_FALSE
);
3771 spa_sync_props(props
, tx
);
3776 spa
->spa_sync_on
= B_TRUE
;
3777 txg_sync_start(spa
->spa_dsl_pool
);
3780 * We explicitly wait for the first transaction to complete so that our
3781 * bean counters are appropriately updated.
3783 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3785 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3787 spa_history_log_version(spa
, "create");
3790 * Don't count references from objsets that are already closed
3791 * and are making their way through the eviction process.
3793 spa_evicting_os_wait(spa
);
3794 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3796 mutex_exit(&spa_namespace_lock
);
3803 * Get the root pool information from the root disk, then import the root pool
3804 * during the system boot up time.
3806 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3809 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3812 nvlist_t
*nvtop
, *nvroot
;
3815 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3819 * Add this top-level vdev to the child array.
3821 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3823 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3825 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3828 * Put this pool's top-level vdevs into a root vdev.
3830 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3831 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3832 VDEV_TYPE_ROOT
) == 0);
3833 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3834 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3835 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3839 * Replace the existing vdev_tree with the new root vdev in
3840 * this pool's configuration (remove the old, add the new).
3842 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3843 nvlist_free(nvroot
);
3848 * Walk the vdev tree and see if we can find a device with "better"
3849 * configuration. A configuration is "better" if the label on that
3850 * device has a more recent txg.
3853 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3857 for (c
= 0; c
< vd
->vdev_children
; c
++)
3858 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3860 if (vd
->vdev_ops
->vdev_op_leaf
) {
3864 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3868 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3872 * Do we have a better boot device?
3874 if (label_txg
> *txg
) {
3883 * Import a root pool.
3885 * For x86. devpath_list will consist of devid and/or physpath name of
3886 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3887 * The GRUB "findroot" command will return the vdev we should boot.
3889 * For Sparc, devpath_list consists the physpath name of the booting device
3890 * no matter the rootpool is a single device pool or a mirrored pool.
3892 * "/pci@1f,0/ide@d/disk@0,0:a"
3895 spa_import_rootpool(char *devpath
, char *devid
)
3898 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3899 nvlist_t
*config
, *nvtop
;
3905 * Read the label from the boot device and generate a configuration.
3907 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3908 #if defined(_OBP) && defined(_KERNEL)
3909 if (config
== NULL
) {
3910 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3912 get_iscsi_bootpath_phy(devpath
);
3913 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3917 if (config
== NULL
) {
3918 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
3920 return (SET_ERROR(EIO
));
3923 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3925 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3927 mutex_enter(&spa_namespace_lock
);
3928 if ((spa
= spa_lookup(pname
)) != NULL
) {
3930 * Remove the existing root pool from the namespace so that we
3931 * can replace it with the correct config we just read in.
3936 spa
= spa_add(pname
, config
, NULL
);
3937 spa
->spa_is_root
= B_TRUE
;
3938 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3941 * Build up a vdev tree based on the boot device's label config.
3943 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3945 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3946 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3947 VDEV_ALLOC_ROOTPOOL
);
3948 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3950 mutex_exit(&spa_namespace_lock
);
3951 nvlist_free(config
);
3952 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3958 * Get the boot vdev.
3960 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3961 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3962 (u_longlong_t
)guid
);
3963 error
= SET_ERROR(ENOENT
);
3968 * Determine if there is a better boot device.
3971 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3973 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3974 "try booting from '%s'", avd
->vdev_path
);
3975 error
= SET_ERROR(EINVAL
);
3980 * If the boot device is part of a spare vdev then ensure that
3981 * we're booting off the active spare.
3983 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3984 !bvd
->vdev_isspare
) {
3985 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
3986 "try booting from '%s'",
3988 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
3989 error
= SET_ERROR(EINVAL
);
3995 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3997 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3998 mutex_exit(&spa_namespace_lock
);
4000 nvlist_free(config
);
4007 * Import a non-root pool into the system.
4010 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4013 char *altroot
= NULL
;
4014 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4015 zpool_rewind_policy_t policy
;
4016 uint64_t mode
= spa_mode_global
;
4017 uint64_t readonly
= B_FALSE
;
4020 nvlist_t
**spares
, **l2cache
;
4021 uint_t nspares
, nl2cache
;
4024 * If a pool with this name exists, return failure.
4026 mutex_enter(&spa_namespace_lock
);
4027 if (spa_lookup(pool
) != NULL
) {
4028 mutex_exit(&spa_namespace_lock
);
4029 return (SET_ERROR(EEXIST
));
4033 * Create and initialize the spa structure.
4035 (void) nvlist_lookup_string(props
,
4036 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4037 (void) nvlist_lookup_uint64(props
,
4038 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4041 spa
= spa_add(pool
, config
, altroot
);
4042 spa
->spa_import_flags
= flags
;
4045 * Verbatim import - Take a pool and insert it into the namespace
4046 * as if it had been loaded at boot.
4048 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4050 spa_configfile_set(spa
, props
, B_FALSE
);
4052 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4054 mutex_exit(&spa_namespace_lock
);
4058 spa_activate(spa
, mode
);
4061 * Don't start async tasks until we know everything is healthy.
4063 spa_async_suspend(spa
);
4065 zpool_get_rewind_policy(config
, &policy
);
4066 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4067 state
= SPA_LOAD_RECOVER
;
4070 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4071 * because the user-supplied config is actually the one to trust when
4074 if (state
!= SPA_LOAD_RECOVER
)
4075 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4077 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4078 policy
.zrp_request
);
4081 * Propagate anything learned while loading the pool and pass it
4082 * back to caller (i.e. rewind info, missing devices, etc).
4084 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4085 spa
->spa_load_info
) == 0);
4087 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4089 * Toss any existing sparelist, as it doesn't have any validity
4090 * anymore, and conflicts with spa_has_spare().
4092 if (spa
->spa_spares
.sav_config
) {
4093 nvlist_free(spa
->spa_spares
.sav_config
);
4094 spa
->spa_spares
.sav_config
= NULL
;
4095 spa_load_spares(spa
);
4097 if (spa
->spa_l2cache
.sav_config
) {
4098 nvlist_free(spa
->spa_l2cache
.sav_config
);
4099 spa
->spa_l2cache
.sav_config
= NULL
;
4100 spa_load_l2cache(spa
);
4103 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4106 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4109 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4110 VDEV_ALLOC_L2CACHE
);
4111 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4114 spa_configfile_set(spa
, props
, B_FALSE
);
4116 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4117 (error
= spa_prop_set(spa
, props
)))) {
4119 spa_deactivate(spa
);
4121 mutex_exit(&spa_namespace_lock
);
4125 spa_async_resume(spa
);
4128 * Override any spares and level 2 cache devices as specified by
4129 * the user, as these may have correct device names/devids, etc.
4131 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4132 &spares
, &nspares
) == 0) {
4133 if (spa
->spa_spares
.sav_config
)
4134 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4135 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4137 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4138 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4139 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4140 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4141 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4142 spa_load_spares(spa
);
4143 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4144 spa
->spa_spares
.sav_sync
= B_TRUE
;
4146 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4147 &l2cache
, &nl2cache
) == 0) {
4148 if (spa
->spa_l2cache
.sav_config
)
4149 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4150 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4152 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4153 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4154 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4155 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4156 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4157 spa_load_l2cache(spa
);
4158 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4159 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4163 * Check for any removed devices.
4165 if (spa
->spa_autoreplace
) {
4166 spa_aux_check_removed(&spa
->spa_spares
);
4167 spa_aux_check_removed(&spa
->spa_l2cache
);
4170 if (spa_writeable(spa
)) {
4172 * Update the config cache to include the newly-imported pool.
4174 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4178 * It's possible that the pool was expanded while it was exported.
4179 * We kick off an async task to handle this for us.
4181 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4183 mutex_exit(&spa_namespace_lock
);
4184 spa_history_log_version(spa
, "import");
4187 zvol_create_minors(pool
);
4194 spa_tryimport(nvlist_t
*tryconfig
)
4196 nvlist_t
*config
= NULL
;
4202 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4205 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4209 * Create and initialize the spa structure.
4211 mutex_enter(&spa_namespace_lock
);
4212 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4213 spa_activate(spa
, FREAD
);
4216 * Pass off the heavy lifting to spa_load().
4217 * Pass TRUE for mosconfig because the user-supplied config
4218 * is actually the one to trust when doing an import.
4220 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4223 * If 'tryconfig' was at least parsable, return the current config.
4225 if (spa
->spa_root_vdev
!= NULL
) {
4226 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4227 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4229 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4231 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4232 spa
->spa_uberblock
.ub_timestamp
) == 0);
4233 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4234 spa
->spa_load_info
) == 0);
4235 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4236 spa
->spa_errata
) == 0);
4239 * If the bootfs property exists on this pool then we
4240 * copy it out so that external consumers can tell which
4241 * pools are bootable.
4243 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4244 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4247 * We have to play games with the name since the
4248 * pool was opened as TRYIMPORT_NAME.
4250 if (dsl_dsobj_to_dsname(spa_name(spa
),
4251 spa
->spa_bootfs
, tmpname
) == 0) {
4255 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4257 cp
= strchr(tmpname
, '/');
4259 (void) strlcpy(dsname
, tmpname
,
4262 (void) snprintf(dsname
, MAXPATHLEN
,
4263 "%s/%s", poolname
, ++cp
);
4265 VERIFY(nvlist_add_string(config
,
4266 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4267 kmem_free(dsname
, MAXPATHLEN
);
4269 kmem_free(tmpname
, MAXPATHLEN
);
4273 * Add the list of hot spares and level 2 cache devices.
4275 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4276 spa_add_spares(spa
, config
);
4277 spa_add_l2cache(spa
, config
);
4278 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4282 spa_deactivate(spa
);
4284 mutex_exit(&spa_namespace_lock
);
4290 * Pool export/destroy
4292 * The act of destroying or exporting a pool is very simple. We make sure there
4293 * is no more pending I/O and any references to the pool are gone. Then, we
4294 * update the pool state and sync all the labels to disk, removing the
4295 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4296 * we don't sync the labels or remove the configuration cache.
4299 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4300 boolean_t force
, boolean_t hardforce
)
4307 if (!(spa_mode_global
& FWRITE
))
4308 return (SET_ERROR(EROFS
));
4310 mutex_enter(&spa_namespace_lock
);
4311 if ((spa
= spa_lookup(pool
)) == NULL
) {
4312 mutex_exit(&spa_namespace_lock
);
4313 return (SET_ERROR(ENOENT
));
4317 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4318 * reacquire the namespace lock, and see if we can export.
4320 spa_open_ref(spa
, FTAG
);
4321 mutex_exit(&spa_namespace_lock
);
4322 spa_async_suspend(spa
);
4323 mutex_enter(&spa_namespace_lock
);
4324 spa_close(spa
, FTAG
);
4326 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4329 * The pool will be in core if it's openable, in which case we can
4330 * modify its state. Objsets may be open only because they're dirty,
4331 * so we have to force it to sync before checking spa_refcnt.
4333 if (spa
->spa_sync_on
) {
4334 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4335 spa_evicting_os_wait(spa
);
4339 * A pool cannot be exported or destroyed if there are active
4340 * references. If we are resetting a pool, allow references by
4341 * fault injection handlers.
4343 if (!spa_refcount_zero(spa
) ||
4344 (spa
->spa_inject_ref
!= 0 &&
4345 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4346 spa_async_resume(spa
);
4347 mutex_exit(&spa_namespace_lock
);
4348 return (SET_ERROR(EBUSY
));
4351 if (spa
->spa_sync_on
) {
4353 * A pool cannot be exported if it has an active shared spare.
4354 * This is to prevent other pools stealing the active spare
4355 * from an exported pool. At user's own will, such pool can
4356 * be forcedly exported.
4358 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4359 spa_has_active_shared_spare(spa
)) {
4360 spa_async_resume(spa
);
4361 mutex_exit(&spa_namespace_lock
);
4362 return (SET_ERROR(EXDEV
));
4366 * We want this to be reflected on every label,
4367 * so mark them all dirty. spa_unload() will do the
4368 * final sync that pushes these changes out.
4370 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4371 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4372 spa
->spa_state
= new_state
;
4373 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4375 vdev_config_dirty(spa
->spa_root_vdev
);
4376 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4381 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_DESTROY
);
4383 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4385 spa_deactivate(spa
);
4388 if (oldconfig
&& spa
->spa_config
)
4389 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4391 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4393 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4396 mutex_exit(&spa_namespace_lock
);
4402 * Destroy a storage pool.
4405 spa_destroy(char *pool
)
4407 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4412 * Export a storage pool.
4415 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4416 boolean_t hardforce
)
4418 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4423 * Similar to spa_export(), this unloads the spa_t without actually removing it
4424 * from the namespace in any way.
4427 spa_reset(char *pool
)
4429 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4434 * ==========================================================================
4435 * Device manipulation
4436 * ==========================================================================
4440 * Add a device to a storage pool.
4443 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4447 vdev_t
*rvd
= spa
->spa_root_vdev
;
4449 nvlist_t
**spares
, **l2cache
;
4450 uint_t nspares
, nl2cache
;
4453 ASSERT(spa_writeable(spa
));
4455 txg
= spa_vdev_enter(spa
);
4457 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4458 VDEV_ALLOC_ADD
)) != 0)
4459 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4461 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4463 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4467 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4471 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4472 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4474 if (vd
->vdev_children
!= 0 &&
4475 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4476 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4479 * We must validate the spares and l2cache devices after checking the
4480 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4482 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4483 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4486 * Transfer each new top-level vdev from vd to rvd.
4488 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4491 * Set the vdev id to the first hole, if one exists.
4493 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4494 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4495 vdev_free(rvd
->vdev_child
[id
]);
4499 tvd
= vd
->vdev_child
[c
];
4500 vdev_remove_child(vd
, tvd
);
4502 vdev_add_child(rvd
, tvd
);
4503 vdev_config_dirty(tvd
);
4507 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4508 ZPOOL_CONFIG_SPARES
);
4509 spa_load_spares(spa
);
4510 spa
->spa_spares
.sav_sync
= B_TRUE
;
4513 if (nl2cache
!= 0) {
4514 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4515 ZPOOL_CONFIG_L2CACHE
);
4516 spa_load_l2cache(spa
);
4517 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4521 * We have to be careful when adding new vdevs to an existing pool.
4522 * If other threads start allocating from these vdevs before we
4523 * sync the config cache, and we lose power, then upon reboot we may
4524 * fail to open the pool because there are DVAs that the config cache
4525 * can't translate. Therefore, we first add the vdevs without
4526 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4527 * and then let spa_config_update() initialize the new metaslabs.
4529 * spa_load() checks for added-but-not-initialized vdevs, so that
4530 * if we lose power at any point in this sequence, the remaining
4531 * steps will be completed the next time we load the pool.
4533 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4535 mutex_enter(&spa_namespace_lock
);
4536 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4537 mutex_exit(&spa_namespace_lock
);
4543 * Attach a device to a mirror. The arguments are the path to any device
4544 * in the mirror, and the nvroot for the new device. If the path specifies
4545 * a device that is not mirrored, we automatically insert the mirror vdev.
4547 * If 'replacing' is specified, the new device is intended to replace the
4548 * existing device; in this case the two devices are made into their own
4549 * mirror using the 'replacing' vdev, which is functionally identical to
4550 * the mirror vdev (it actually reuses all the same ops) but has a few
4551 * extra rules: you can't attach to it after it's been created, and upon
4552 * completion of resilvering, the first disk (the one being replaced)
4553 * is automatically detached.
4556 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4558 uint64_t txg
, dtl_max_txg
;
4559 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4561 char *oldvdpath
, *newvdpath
;
4564 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4566 ASSERT(spa_writeable(spa
));
4568 txg
= spa_vdev_enter(spa
);
4570 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4573 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4575 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4576 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4578 pvd
= oldvd
->vdev_parent
;
4580 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4581 VDEV_ALLOC_ATTACH
)) != 0)
4582 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4584 if (newrootvd
->vdev_children
!= 1)
4585 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4587 newvd
= newrootvd
->vdev_child
[0];
4589 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4590 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4592 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4593 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4596 * Spares can't replace logs
4598 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4599 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4603 * For attach, the only allowable parent is a mirror or the root
4606 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4607 pvd
->vdev_ops
!= &vdev_root_ops
)
4608 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4610 pvops
= &vdev_mirror_ops
;
4613 * Active hot spares can only be replaced by inactive hot
4616 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4617 oldvd
->vdev_isspare
&&
4618 !spa_has_spare(spa
, newvd
->vdev_guid
))
4619 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4622 * If the source is a hot spare, and the parent isn't already a
4623 * spare, then we want to create a new hot spare. Otherwise, we
4624 * want to create a replacing vdev. The user is not allowed to
4625 * attach to a spared vdev child unless the 'isspare' state is
4626 * the same (spare replaces spare, non-spare replaces
4629 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4630 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4631 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4632 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4633 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4634 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4637 if (newvd
->vdev_isspare
)
4638 pvops
= &vdev_spare_ops
;
4640 pvops
= &vdev_replacing_ops
;
4644 * Make sure the new device is big enough.
4646 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4647 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4650 * The new device cannot have a higher alignment requirement
4651 * than the top-level vdev.
4653 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4654 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4657 * If this is an in-place replacement, update oldvd's path and devid
4658 * to make it distinguishable from newvd, and unopenable from now on.
4660 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4661 spa_strfree(oldvd
->vdev_path
);
4662 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4664 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4665 newvd
->vdev_path
, "old");
4666 if (oldvd
->vdev_devid
!= NULL
) {
4667 spa_strfree(oldvd
->vdev_devid
);
4668 oldvd
->vdev_devid
= NULL
;
4672 /* mark the device being resilvered */
4673 newvd
->vdev_resilver_txg
= txg
;
4676 * If the parent is not a mirror, or if we're replacing, insert the new
4677 * mirror/replacing/spare vdev above oldvd.
4679 if (pvd
->vdev_ops
!= pvops
)
4680 pvd
= vdev_add_parent(oldvd
, pvops
);
4682 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4683 ASSERT(pvd
->vdev_ops
== pvops
);
4684 ASSERT(oldvd
->vdev_parent
== pvd
);
4687 * Extract the new device from its root and add it to pvd.
4689 vdev_remove_child(newrootvd
, newvd
);
4690 newvd
->vdev_id
= pvd
->vdev_children
;
4691 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4692 vdev_add_child(pvd
, newvd
);
4694 tvd
= newvd
->vdev_top
;
4695 ASSERT(pvd
->vdev_top
== tvd
);
4696 ASSERT(tvd
->vdev_parent
== rvd
);
4698 vdev_config_dirty(tvd
);
4701 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4702 * for any dmu_sync-ed blocks. It will propagate upward when
4703 * spa_vdev_exit() calls vdev_dtl_reassess().
4705 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4707 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4708 dtl_max_txg
- TXG_INITIAL
);
4710 if (newvd
->vdev_isspare
) {
4711 spa_spare_activate(newvd
);
4712 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_DEVICE_SPARE
);
4715 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4716 newvdpath
= spa_strdup(newvd
->vdev_path
);
4717 newvd_isspare
= newvd
->vdev_isspare
;
4720 * Mark newvd's DTL dirty in this txg.
4722 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4725 * Schedule the resilver to restart in the future. We do this to
4726 * ensure that dmu_sync-ed blocks have been stitched into the
4727 * respective datasets.
4729 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4734 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4736 spa_history_log_internal(spa
, "vdev attach", NULL
,
4737 "%s vdev=%s %s vdev=%s",
4738 replacing
&& newvd_isspare
? "spare in" :
4739 replacing
? "replace" : "attach", newvdpath
,
4740 replacing
? "for" : "to", oldvdpath
);
4742 spa_strfree(oldvdpath
);
4743 spa_strfree(newvdpath
);
4745 if (spa
->spa_bootfs
)
4746 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH
);
4752 * Detach a device from a mirror or replacing vdev.
4754 * If 'replace_done' is specified, only detach if the parent
4755 * is a replacing vdev.
4758 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4762 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4763 boolean_t unspare
= B_FALSE
;
4764 uint64_t unspare_guid
= 0;
4767 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4768 ASSERT(spa_writeable(spa
));
4770 txg
= spa_vdev_enter(spa
);
4772 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4775 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4777 if (!vd
->vdev_ops
->vdev_op_leaf
)
4778 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4780 pvd
= vd
->vdev_parent
;
4783 * If the parent/child relationship is not as expected, don't do it.
4784 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4785 * vdev that's replacing B with C. The user's intent in replacing
4786 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4787 * the replace by detaching C, the expected behavior is to end up
4788 * M(A,B). But suppose that right after deciding to detach C,
4789 * the replacement of B completes. We would have M(A,C), and then
4790 * ask to detach C, which would leave us with just A -- not what
4791 * the user wanted. To prevent this, we make sure that the
4792 * parent/child relationship hasn't changed -- in this example,
4793 * that C's parent is still the replacing vdev R.
4795 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4796 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4799 * Only 'replacing' or 'spare' vdevs can be replaced.
4801 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4802 pvd
->vdev_ops
!= &vdev_spare_ops
)
4803 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4805 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4806 spa_version(spa
) >= SPA_VERSION_SPARES
);
4809 * Only mirror, replacing, and spare vdevs support detach.
4811 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4812 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4813 pvd
->vdev_ops
!= &vdev_spare_ops
)
4814 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4817 * If this device has the only valid copy of some data,
4818 * we cannot safely detach it.
4820 if (vdev_dtl_required(vd
))
4821 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4823 ASSERT(pvd
->vdev_children
>= 2);
4826 * If we are detaching the second disk from a replacing vdev, then
4827 * check to see if we changed the original vdev's path to have "/old"
4828 * at the end in spa_vdev_attach(). If so, undo that change now.
4830 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4831 vd
->vdev_path
!= NULL
) {
4832 size_t len
= strlen(vd
->vdev_path
);
4834 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4835 cvd
= pvd
->vdev_child
[c
];
4837 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4840 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4841 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4842 spa_strfree(cvd
->vdev_path
);
4843 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4850 * If we are detaching the original disk from a spare, then it implies
4851 * that the spare should become a real disk, and be removed from the
4852 * active spare list for the pool.
4854 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4856 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4860 * Erase the disk labels so the disk can be used for other things.
4861 * This must be done after all other error cases are handled,
4862 * but before we disembowel vd (so we can still do I/O to it).
4863 * But if we can't do it, don't treat the error as fatal --
4864 * it may be that the unwritability of the disk is the reason
4865 * it's being detached!
4867 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4870 * Remove vd from its parent and compact the parent's children.
4872 vdev_remove_child(pvd
, vd
);
4873 vdev_compact_children(pvd
);
4876 * Remember one of the remaining children so we can get tvd below.
4878 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4881 * If we need to remove the remaining child from the list of hot spares,
4882 * do it now, marking the vdev as no longer a spare in the process.
4883 * We must do this before vdev_remove_parent(), because that can
4884 * change the GUID if it creates a new toplevel GUID. For a similar
4885 * reason, we must remove the spare now, in the same txg as the detach;
4886 * otherwise someone could attach a new sibling, change the GUID, and
4887 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4890 ASSERT(cvd
->vdev_isspare
);
4891 spa_spare_remove(cvd
);
4892 unspare_guid
= cvd
->vdev_guid
;
4893 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4894 cvd
->vdev_unspare
= B_TRUE
;
4898 * If the parent mirror/replacing vdev only has one child,
4899 * the parent is no longer needed. Remove it from the tree.
4901 if (pvd
->vdev_children
== 1) {
4902 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4903 cvd
->vdev_unspare
= B_FALSE
;
4904 vdev_remove_parent(cvd
);
4909 * We don't set tvd until now because the parent we just removed
4910 * may have been the previous top-level vdev.
4912 tvd
= cvd
->vdev_top
;
4913 ASSERT(tvd
->vdev_parent
== rvd
);
4916 * Reevaluate the parent vdev state.
4918 vdev_propagate_state(cvd
);
4921 * If the 'autoexpand' property is set on the pool then automatically
4922 * try to expand the size of the pool. For example if the device we
4923 * just detached was smaller than the others, it may be possible to
4924 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4925 * first so that we can obtain the updated sizes of the leaf vdevs.
4927 if (spa
->spa_autoexpand
) {
4929 vdev_expand(tvd
, txg
);
4932 vdev_config_dirty(tvd
);
4935 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4936 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4937 * But first make sure we're not on any *other* txg's DTL list, to
4938 * prevent vd from being accessed after it's freed.
4940 vdpath
= spa_strdup(vd
->vdev_path
);
4941 for (t
= 0; t
< TXG_SIZE
; t
++)
4942 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4943 vd
->vdev_detached
= B_TRUE
;
4944 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4946 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_REMOVE
);
4948 /* hang on to the spa before we release the lock */
4949 spa_open_ref(spa
, FTAG
);
4951 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4953 spa_history_log_internal(spa
, "detach", NULL
,
4955 spa_strfree(vdpath
);
4958 * If this was the removal of the original device in a hot spare vdev,
4959 * then we want to go through and remove the device from the hot spare
4960 * list of every other pool.
4963 spa_t
*altspa
= NULL
;
4965 mutex_enter(&spa_namespace_lock
);
4966 while ((altspa
= spa_next(altspa
)) != NULL
) {
4967 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4971 spa_open_ref(altspa
, FTAG
);
4972 mutex_exit(&spa_namespace_lock
);
4973 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4974 mutex_enter(&spa_namespace_lock
);
4975 spa_close(altspa
, FTAG
);
4977 mutex_exit(&spa_namespace_lock
);
4979 /* search the rest of the vdevs for spares to remove */
4980 spa_vdev_resilver_done(spa
);
4983 /* all done with the spa; OK to release */
4984 mutex_enter(&spa_namespace_lock
);
4985 spa_close(spa
, FTAG
);
4986 mutex_exit(&spa_namespace_lock
);
4992 * Split a set of devices from their mirrors, and create a new pool from them.
4995 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
4996 nvlist_t
*props
, boolean_t exp
)
4999 uint64_t txg
, *glist
;
5001 uint_t c
, children
, lastlog
;
5002 nvlist_t
**child
, *nvl
, *tmp
;
5004 char *altroot
= NULL
;
5005 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5006 boolean_t activate_slog
;
5008 ASSERT(spa_writeable(spa
));
5010 txg
= spa_vdev_enter(spa
);
5012 /* clear the log and flush everything up to now */
5013 activate_slog
= spa_passivate_log(spa
);
5014 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5015 error
= spa_offline_log(spa
);
5016 txg
= spa_vdev_config_enter(spa
);
5019 spa_activate_log(spa
);
5022 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5024 /* check new spa name before going any further */
5025 if (spa_lookup(newname
) != NULL
)
5026 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5029 * scan through all the children to ensure they're all mirrors
5031 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5032 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5034 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5036 /* first, check to ensure we've got the right child count */
5037 rvd
= spa
->spa_root_vdev
;
5039 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5040 vdev_t
*vd
= rvd
->vdev_child
[c
];
5042 /* don't count the holes & logs as children */
5043 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5051 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5052 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5054 /* next, ensure no spare or cache devices are part of the split */
5055 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5056 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5057 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5059 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5060 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5062 /* then, loop over each vdev and validate it */
5063 for (c
= 0; c
< children
; c
++) {
5064 uint64_t is_hole
= 0;
5066 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5070 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5071 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5074 error
= SET_ERROR(EINVAL
);
5079 /* which disk is going to be split? */
5080 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5082 error
= SET_ERROR(EINVAL
);
5086 /* look it up in the spa */
5087 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5088 if (vml
[c
] == NULL
) {
5089 error
= SET_ERROR(ENODEV
);
5093 /* make sure there's nothing stopping the split */
5094 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5095 vml
[c
]->vdev_islog
||
5096 vml
[c
]->vdev_ishole
||
5097 vml
[c
]->vdev_isspare
||
5098 vml
[c
]->vdev_isl2cache
||
5099 !vdev_writeable(vml
[c
]) ||
5100 vml
[c
]->vdev_children
!= 0 ||
5101 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5102 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5103 error
= SET_ERROR(EINVAL
);
5107 if (vdev_dtl_required(vml
[c
])) {
5108 error
= SET_ERROR(EBUSY
);
5112 /* we need certain info from the top level */
5113 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5114 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5115 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5116 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5117 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5118 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5119 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5120 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5124 kmem_free(vml
, children
* sizeof (vdev_t
*));
5125 kmem_free(glist
, children
* sizeof (uint64_t));
5126 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5129 /* stop writers from using the disks */
5130 for (c
= 0; c
< children
; c
++) {
5132 vml
[c
]->vdev_offline
= B_TRUE
;
5134 vdev_reopen(spa
->spa_root_vdev
);
5137 * Temporarily record the splitting vdevs in the spa config. This
5138 * will disappear once the config is regenerated.
5140 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5141 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5142 glist
, children
) == 0);
5143 kmem_free(glist
, children
* sizeof (uint64_t));
5145 mutex_enter(&spa
->spa_props_lock
);
5146 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5148 mutex_exit(&spa
->spa_props_lock
);
5149 spa
->spa_config_splitting
= nvl
;
5150 vdev_config_dirty(spa
->spa_root_vdev
);
5152 /* configure and create the new pool */
5153 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5154 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5155 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5156 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5157 spa_version(spa
)) == 0);
5158 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5159 spa
->spa_config_txg
) == 0);
5160 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5161 spa_generate_guid(NULL
)) == 0);
5162 (void) nvlist_lookup_string(props
,
5163 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5165 /* add the new pool to the namespace */
5166 newspa
= spa_add(newname
, config
, altroot
);
5167 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5168 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5170 /* release the spa config lock, retaining the namespace lock */
5171 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5173 if (zio_injection_enabled
)
5174 zio_handle_panic_injection(spa
, FTAG
, 1);
5176 spa_activate(newspa
, spa_mode_global
);
5177 spa_async_suspend(newspa
);
5179 /* create the new pool from the disks of the original pool */
5180 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5184 /* if that worked, generate a real config for the new pool */
5185 if (newspa
->spa_root_vdev
!= NULL
) {
5186 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5187 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5188 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5189 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5190 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5195 if (props
!= NULL
) {
5196 spa_configfile_set(newspa
, props
, B_FALSE
);
5197 error
= spa_prop_set(newspa
, props
);
5202 /* flush everything */
5203 txg
= spa_vdev_config_enter(newspa
);
5204 vdev_config_dirty(newspa
->spa_root_vdev
);
5205 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5207 if (zio_injection_enabled
)
5208 zio_handle_panic_injection(spa
, FTAG
, 2);
5210 spa_async_resume(newspa
);
5212 /* finally, update the original pool's config */
5213 txg
= spa_vdev_config_enter(spa
);
5214 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5215 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5218 for (c
= 0; c
< children
; c
++) {
5219 if (vml
[c
] != NULL
) {
5222 spa_history_log_internal(spa
, "detach", tx
,
5223 "vdev=%s", vml
[c
]->vdev_path
);
5227 vdev_config_dirty(spa
->spa_root_vdev
);
5228 spa
->spa_config_splitting
= NULL
;
5232 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5234 if (zio_injection_enabled
)
5235 zio_handle_panic_injection(spa
, FTAG
, 3);
5237 /* split is complete; log a history record */
5238 spa_history_log_internal(newspa
, "split", NULL
,
5239 "from pool %s", spa_name(spa
));
5241 kmem_free(vml
, children
* sizeof (vdev_t
*));
5243 /* if we're not going to mount the filesystems in userland, export */
5245 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5252 spa_deactivate(newspa
);
5255 txg
= spa_vdev_config_enter(spa
);
5257 /* re-online all offlined disks */
5258 for (c
= 0; c
< children
; c
++) {
5260 vml
[c
]->vdev_offline
= B_FALSE
;
5262 vdev_reopen(spa
->spa_root_vdev
);
5264 nvlist_free(spa
->spa_config_splitting
);
5265 spa
->spa_config_splitting
= NULL
;
5266 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5268 kmem_free(vml
, children
* sizeof (vdev_t
*));
5273 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5277 for (i
= 0; i
< count
; i
++) {
5280 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5283 if (guid
== target_guid
)
5291 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5292 nvlist_t
*dev_to_remove
)
5294 nvlist_t
**newdev
= NULL
;
5298 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5300 for (i
= 0, j
= 0; i
< count
; i
++) {
5301 if (dev
[i
] == dev_to_remove
)
5303 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5306 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5307 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5309 for (i
= 0; i
< count
- 1; i
++)
5310 nvlist_free(newdev
[i
]);
5313 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5317 * Evacuate the device.
5320 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5325 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5326 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5327 ASSERT(vd
== vd
->vdev_top
);
5330 * Evacuate the device. We don't hold the config lock as writer
5331 * since we need to do I/O but we do keep the
5332 * spa_namespace_lock held. Once this completes the device
5333 * should no longer have any blocks allocated on it.
5335 if (vd
->vdev_islog
) {
5336 if (vd
->vdev_stat
.vs_alloc
!= 0)
5337 error
= spa_offline_log(spa
);
5339 error
= SET_ERROR(ENOTSUP
);
5346 * The evacuation succeeded. Remove any remaining MOS metadata
5347 * associated with this vdev, and wait for these changes to sync.
5349 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5350 txg
= spa_vdev_config_enter(spa
);
5351 vd
->vdev_removing
= B_TRUE
;
5352 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5353 vdev_config_dirty(vd
);
5354 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5360 * Complete the removal by cleaning up the namespace.
5363 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5365 vdev_t
*rvd
= spa
->spa_root_vdev
;
5366 uint64_t id
= vd
->vdev_id
;
5367 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5369 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5370 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5371 ASSERT(vd
== vd
->vdev_top
);
5374 * Only remove any devices which are empty.
5376 if (vd
->vdev_stat
.vs_alloc
!= 0)
5379 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5381 if (list_link_active(&vd
->vdev_state_dirty_node
))
5382 vdev_state_clean(vd
);
5383 if (list_link_active(&vd
->vdev_config_dirty_node
))
5384 vdev_config_clean(vd
);
5389 vdev_compact_children(rvd
);
5391 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5392 vdev_add_child(rvd
, vd
);
5394 vdev_config_dirty(rvd
);
5397 * Reassess the health of our root vdev.
5403 * Remove a device from the pool -
5405 * Removing a device from the vdev namespace requires several steps
5406 * and can take a significant amount of time. As a result we use
5407 * the spa_vdev_config_[enter/exit] functions which allow us to
5408 * grab and release the spa_config_lock while still holding the namespace
5409 * lock. During each step the configuration is synced out.
5411 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5415 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5418 metaslab_group_t
*mg
;
5419 nvlist_t
**spares
, **l2cache
, *nv
;
5421 uint_t nspares
, nl2cache
;
5423 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5425 ASSERT(spa_writeable(spa
));
5428 txg
= spa_vdev_enter(spa
);
5430 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5432 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5433 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5434 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5435 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5437 * Only remove the hot spare if it's not currently in use
5440 if (vd
== NULL
|| unspare
) {
5441 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5442 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5443 spa_load_spares(spa
);
5444 spa
->spa_spares
.sav_sync
= B_TRUE
;
5446 error
= SET_ERROR(EBUSY
);
5448 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5449 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5450 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5451 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5453 * Cache devices can always be removed.
5455 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5456 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5457 spa_load_l2cache(spa
);
5458 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5459 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5461 ASSERT(vd
== vd
->vdev_top
);
5466 * Stop allocating from this vdev.
5468 metaslab_group_passivate(mg
);
5471 * Wait for the youngest allocations and frees to sync,
5472 * and then wait for the deferral of those frees to finish.
5474 spa_vdev_config_exit(spa
, NULL
,
5475 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5478 * Attempt to evacuate the vdev.
5480 error
= spa_vdev_remove_evacuate(spa
, vd
);
5482 txg
= spa_vdev_config_enter(spa
);
5485 * If we couldn't evacuate the vdev, unwind.
5488 metaslab_group_activate(mg
);
5489 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5493 * Clean up the vdev namespace.
5495 spa_vdev_remove_from_namespace(spa
, vd
);
5497 } else if (vd
!= NULL
) {
5499 * Normal vdevs cannot be removed (yet).
5501 error
= SET_ERROR(ENOTSUP
);
5504 * There is no vdev of any kind with the specified guid.
5506 error
= SET_ERROR(ENOENT
);
5510 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5516 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5517 * currently spared, so we can detach it.
5520 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5522 vdev_t
*newvd
, *oldvd
;
5525 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5526 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5532 * Check for a completed replacement. We always consider the first
5533 * vdev in the list to be the oldest vdev, and the last one to be
5534 * the newest (see spa_vdev_attach() for how that works). In
5535 * the case where the newest vdev is faulted, we will not automatically
5536 * remove it after a resilver completes. This is OK as it will require
5537 * user intervention to determine which disk the admin wishes to keep.
5539 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5540 ASSERT(vd
->vdev_children
> 1);
5542 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5543 oldvd
= vd
->vdev_child
[0];
5545 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5546 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5547 !vdev_dtl_required(oldvd
))
5552 * Check for a completed resilver with the 'unspare' flag set.
5554 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5555 vdev_t
*first
= vd
->vdev_child
[0];
5556 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5558 if (last
->vdev_unspare
) {
5561 } else if (first
->vdev_unspare
) {
5568 if (oldvd
!= NULL
&&
5569 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5570 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5571 !vdev_dtl_required(oldvd
))
5575 * If there are more than two spares attached to a disk,
5576 * and those spares are not required, then we want to
5577 * attempt to free them up now so that they can be used
5578 * by other pools. Once we're back down to a single
5579 * disk+spare, we stop removing them.
5581 if (vd
->vdev_children
> 2) {
5582 newvd
= vd
->vdev_child
[1];
5584 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5585 vdev_dtl_empty(last
, DTL_MISSING
) &&
5586 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5587 !vdev_dtl_required(newvd
))
5596 spa_vdev_resilver_done(spa_t
*spa
)
5598 vdev_t
*vd
, *pvd
, *ppvd
;
5599 uint64_t guid
, sguid
, pguid
, ppguid
;
5601 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5603 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5604 pvd
= vd
->vdev_parent
;
5605 ppvd
= pvd
->vdev_parent
;
5606 guid
= vd
->vdev_guid
;
5607 pguid
= pvd
->vdev_guid
;
5608 ppguid
= ppvd
->vdev_guid
;
5611 * If we have just finished replacing a hot spared device, then
5612 * we need to detach the parent's first child (the original hot
5615 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5616 ppvd
->vdev_children
== 2) {
5617 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5618 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5620 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5622 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5623 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5625 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5627 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5630 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5634 * Update the stored path or FRU for this vdev.
5637 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5641 boolean_t sync
= B_FALSE
;
5643 ASSERT(spa_writeable(spa
));
5645 spa_vdev_state_enter(spa
, SCL_ALL
);
5647 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5648 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5650 if (!vd
->vdev_ops
->vdev_op_leaf
)
5651 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5654 if (strcmp(value
, vd
->vdev_path
) != 0) {
5655 spa_strfree(vd
->vdev_path
);
5656 vd
->vdev_path
= spa_strdup(value
);
5660 if (vd
->vdev_fru
== NULL
) {
5661 vd
->vdev_fru
= spa_strdup(value
);
5663 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5664 spa_strfree(vd
->vdev_fru
);
5665 vd
->vdev_fru
= spa_strdup(value
);
5670 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5674 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5676 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5680 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5682 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5686 * ==========================================================================
5688 * ==========================================================================
5692 spa_scan_stop(spa_t
*spa
)
5694 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5695 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5696 return (SET_ERROR(EBUSY
));
5697 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5701 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5703 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5705 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5706 return (SET_ERROR(ENOTSUP
));
5709 * If a resilver was requested, but there is no DTL on a
5710 * writeable leaf device, we have nothing to do.
5712 if (func
== POOL_SCAN_RESILVER
&&
5713 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5714 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5718 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5722 * ==========================================================================
5723 * SPA async task processing
5724 * ==========================================================================
5728 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5732 if (vd
->vdev_remove_wanted
) {
5733 vd
->vdev_remove_wanted
= B_FALSE
;
5734 vd
->vdev_delayed_close
= B_FALSE
;
5735 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5738 * We want to clear the stats, but we don't want to do a full
5739 * vdev_clear() as that will cause us to throw away
5740 * degraded/faulted state as well as attempt to reopen the
5741 * device, all of which is a waste.
5743 vd
->vdev_stat
.vs_read_errors
= 0;
5744 vd
->vdev_stat
.vs_write_errors
= 0;
5745 vd
->vdev_stat
.vs_checksum_errors
= 0;
5747 vdev_state_dirty(vd
->vdev_top
);
5750 for (c
= 0; c
< vd
->vdev_children
; c
++)
5751 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5755 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5759 if (vd
->vdev_probe_wanted
) {
5760 vd
->vdev_probe_wanted
= B_FALSE
;
5761 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5764 for (c
= 0; c
< vd
->vdev_children
; c
++)
5765 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5769 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5773 if (!spa
->spa_autoexpand
)
5776 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5777 vdev_t
*cvd
= vd
->vdev_child
[c
];
5778 spa_async_autoexpand(spa
, cvd
);
5781 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5784 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND
);
5788 spa_async_thread(spa_t
*spa
)
5792 ASSERT(spa
->spa_sync_on
);
5794 mutex_enter(&spa
->spa_async_lock
);
5795 tasks
= spa
->spa_async_tasks
;
5796 spa
->spa_async_tasks
= 0;
5797 mutex_exit(&spa
->spa_async_lock
);
5800 * See if the config needs to be updated.
5802 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5803 uint64_t old_space
, new_space
;
5805 mutex_enter(&spa_namespace_lock
);
5806 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5807 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5808 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5809 mutex_exit(&spa_namespace_lock
);
5812 * If the pool grew as a result of the config update,
5813 * then log an internal history event.
5815 if (new_space
!= old_space
) {
5816 spa_history_log_internal(spa
, "vdev online", NULL
,
5817 "pool '%s' size: %llu(+%llu)",
5818 spa_name(spa
), new_space
, new_space
- old_space
);
5823 * See if any devices need to be marked REMOVED.
5825 if (tasks
& SPA_ASYNC_REMOVE
) {
5826 spa_vdev_state_enter(spa
, SCL_NONE
);
5827 spa_async_remove(spa
, spa
->spa_root_vdev
);
5828 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5829 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5830 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5831 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5832 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5835 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5836 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5837 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5838 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5842 * See if any devices need to be probed.
5844 if (tasks
& SPA_ASYNC_PROBE
) {
5845 spa_vdev_state_enter(spa
, SCL_NONE
);
5846 spa_async_probe(spa
, spa
->spa_root_vdev
);
5847 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5851 * If any devices are done replacing, detach them.
5853 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5854 spa_vdev_resilver_done(spa
);
5857 * Kick off a resilver.
5859 if (tasks
& SPA_ASYNC_RESILVER
)
5860 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5863 * Let the world know that we're done.
5865 mutex_enter(&spa
->spa_async_lock
);
5866 spa
->spa_async_thread
= NULL
;
5867 cv_broadcast(&spa
->spa_async_cv
);
5868 mutex_exit(&spa
->spa_async_lock
);
5873 spa_async_suspend(spa_t
*spa
)
5875 mutex_enter(&spa
->spa_async_lock
);
5876 spa
->spa_async_suspended
++;
5877 while (spa
->spa_async_thread
!= NULL
)
5878 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5879 mutex_exit(&spa
->spa_async_lock
);
5883 spa_async_resume(spa_t
*spa
)
5885 mutex_enter(&spa
->spa_async_lock
);
5886 ASSERT(spa
->spa_async_suspended
!= 0);
5887 spa
->spa_async_suspended
--;
5888 mutex_exit(&spa
->spa_async_lock
);
5892 spa_async_dispatch(spa_t
*spa
)
5894 mutex_enter(&spa
->spa_async_lock
);
5895 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5896 spa
->spa_async_thread
== NULL
&&
5897 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5898 spa
->spa_async_thread
= thread_create(NULL
, 0,
5899 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5900 mutex_exit(&spa
->spa_async_lock
);
5904 spa_async_request(spa_t
*spa
, int task
)
5906 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5907 mutex_enter(&spa
->spa_async_lock
);
5908 spa
->spa_async_tasks
|= task
;
5909 mutex_exit(&spa
->spa_async_lock
);
5913 * ==========================================================================
5914 * SPA syncing routines
5915 * ==========================================================================
5919 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5922 bpobj_enqueue(bpo
, bp
, tx
);
5927 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5931 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5937 * Note: this simple function is not inlined to make it easier to dtrace the
5938 * amount of time spent syncing frees.
5941 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
5943 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5944 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
5945 VERIFY(zio_wait(zio
) == 0);
5949 * Note: this simple function is not inlined to make it easier to dtrace the
5950 * amount of time spent syncing deferred frees.
5953 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
5955 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5956 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
5957 spa_free_sync_cb
, zio
, tx
), ==, 0);
5958 VERIFY0(zio_wait(zio
));
5962 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5964 char *packed
= NULL
;
5969 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5972 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5973 * information. This avoids the dmu_buf_will_dirty() path and
5974 * saves us a pre-read to get data we don't actually care about.
5976 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
5977 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
5979 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
5981 bzero(packed
+ nvsize
, bufsize
- nvsize
);
5983 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
5985 vmem_free(packed
, bufsize
);
5987 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
5988 dmu_buf_will_dirty(db
, tx
);
5989 *(uint64_t *)db
->db_data
= nvsize
;
5990 dmu_buf_rele(db
, FTAG
);
5994 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
5995 const char *config
, const char *entry
)
6005 * Update the MOS nvlist describing the list of available devices.
6006 * spa_validate_aux() will have already made sure this nvlist is
6007 * valid and the vdevs are labeled appropriately.
6009 if (sav
->sav_object
== 0) {
6010 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6011 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6012 sizeof (uint64_t), tx
);
6013 VERIFY(zap_update(spa
->spa_meta_objset
,
6014 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6015 &sav
->sav_object
, tx
) == 0);
6018 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6019 if (sav
->sav_count
== 0) {
6020 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6022 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6023 for (i
= 0; i
< sav
->sav_count
; i
++)
6024 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6025 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6026 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6027 sav
->sav_count
) == 0);
6028 for (i
= 0; i
< sav
->sav_count
; i
++)
6029 nvlist_free(list
[i
]);
6030 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6033 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6034 nvlist_free(nvroot
);
6036 sav
->sav_sync
= B_FALSE
;
6040 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6044 if (list_is_empty(&spa
->spa_config_dirty_list
))
6047 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6049 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6050 dmu_tx_get_txg(tx
), B_FALSE
);
6053 * If we're upgrading the spa version then make sure that
6054 * the config object gets updated with the correct version.
6056 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6057 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6058 spa
->spa_uberblock
.ub_version
);
6060 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6062 if (spa
->spa_config_syncing
)
6063 nvlist_free(spa
->spa_config_syncing
);
6064 spa
->spa_config_syncing
= config
;
6066 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6070 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6072 uint64_t *versionp
= arg
;
6073 uint64_t version
= *versionp
;
6074 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6077 * Setting the version is special cased when first creating the pool.
6079 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6081 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6082 ASSERT(version
>= spa_version(spa
));
6084 spa
->spa_uberblock
.ub_version
= version
;
6085 vdev_config_dirty(spa
->spa_root_vdev
);
6086 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6090 * Set zpool properties.
6093 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6095 nvlist_t
*nvp
= arg
;
6096 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6097 objset_t
*mos
= spa
->spa_meta_objset
;
6098 nvpair_t
*elem
= NULL
;
6100 mutex_enter(&spa
->spa_props_lock
);
6102 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6104 char *strval
, *fname
;
6106 const char *propname
;
6107 zprop_type_t proptype
;
6110 prop
= zpool_name_to_prop(nvpair_name(elem
));
6111 switch ((int)prop
) {
6114 * We checked this earlier in spa_prop_validate().
6116 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6118 fname
= strchr(nvpair_name(elem
), '@') + 1;
6119 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6121 spa_feature_enable(spa
, fid
, tx
);
6122 spa_history_log_internal(spa
, "set", tx
,
6123 "%s=enabled", nvpair_name(elem
));
6126 case ZPOOL_PROP_VERSION
:
6127 intval
= fnvpair_value_uint64(elem
);
6129 * The version is synced seperatly before other
6130 * properties and should be correct by now.
6132 ASSERT3U(spa_version(spa
), >=, intval
);
6135 case ZPOOL_PROP_ALTROOT
:
6137 * 'altroot' is a non-persistent property. It should
6138 * have been set temporarily at creation or import time.
6140 ASSERT(spa
->spa_root
!= NULL
);
6143 case ZPOOL_PROP_READONLY
:
6144 case ZPOOL_PROP_CACHEFILE
:
6146 * 'readonly' and 'cachefile' are also non-persisitent
6150 case ZPOOL_PROP_COMMENT
:
6151 strval
= fnvpair_value_string(elem
);
6152 if (spa
->spa_comment
!= NULL
)
6153 spa_strfree(spa
->spa_comment
);
6154 spa
->spa_comment
= spa_strdup(strval
);
6156 * We need to dirty the configuration on all the vdevs
6157 * so that their labels get updated. It's unnecessary
6158 * to do this for pool creation since the vdev's
6159 * configuratoin has already been dirtied.
6161 if (tx
->tx_txg
!= TXG_INITIAL
)
6162 vdev_config_dirty(spa
->spa_root_vdev
);
6163 spa_history_log_internal(spa
, "set", tx
,
6164 "%s=%s", nvpair_name(elem
), strval
);
6168 * Set pool property values in the poolprops mos object.
6170 if (spa
->spa_pool_props_object
== 0) {
6171 spa
->spa_pool_props_object
=
6172 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6173 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6177 /* normalize the property name */
6178 propname
= zpool_prop_to_name(prop
);
6179 proptype
= zpool_prop_get_type(prop
);
6181 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6182 ASSERT(proptype
== PROP_TYPE_STRING
);
6183 strval
= fnvpair_value_string(elem
);
6184 VERIFY0(zap_update(mos
,
6185 spa
->spa_pool_props_object
, propname
,
6186 1, strlen(strval
) + 1, strval
, tx
));
6187 spa_history_log_internal(spa
, "set", tx
,
6188 "%s=%s", nvpair_name(elem
), strval
);
6189 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6190 intval
= fnvpair_value_uint64(elem
);
6192 if (proptype
== PROP_TYPE_INDEX
) {
6194 VERIFY0(zpool_prop_index_to_string(
6195 prop
, intval
, &unused
));
6197 VERIFY0(zap_update(mos
,
6198 spa
->spa_pool_props_object
, propname
,
6199 8, 1, &intval
, tx
));
6200 spa_history_log_internal(spa
, "set", tx
,
6201 "%s=%lld", nvpair_name(elem
), intval
);
6203 ASSERT(0); /* not allowed */
6207 case ZPOOL_PROP_DELEGATION
:
6208 spa
->spa_delegation
= intval
;
6210 case ZPOOL_PROP_BOOTFS
:
6211 spa
->spa_bootfs
= intval
;
6213 case ZPOOL_PROP_FAILUREMODE
:
6214 spa
->spa_failmode
= intval
;
6216 case ZPOOL_PROP_AUTOEXPAND
:
6217 spa
->spa_autoexpand
= intval
;
6218 if (tx
->tx_txg
!= TXG_INITIAL
)
6219 spa_async_request(spa
,
6220 SPA_ASYNC_AUTOEXPAND
);
6222 case ZPOOL_PROP_DEDUPDITTO
:
6223 spa
->spa_dedup_ditto
= intval
;
6232 mutex_exit(&spa
->spa_props_lock
);
6236 * Perform one-time upgrade on-disk changes. spa_version() does not
6237 * reflect the new version this txg, so there must be no changes this
6238 * txg to anything that the upgrade code depends on after it executes.
6239 * Therefore this must be called after dsl_pool_sync() does the sync
6243 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6245 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6247 ASSERT(spa
->spa_sync_pass
== 1);
6249 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6251 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6252 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6253 dsl_pool_create_origin(dp
, tx
);
6255 /* Keeping the origin open increases spa_minref */
6256 spa
->spa_minref
+= 3;
6259 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6260 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6261 dsl_pool_upgrade_clones(dp
, tx
);
6264 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6265 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6266 dsl_pool_upgrade_dir_clones(dp
, tx
);
6268 /* Keeping the freedir open increases spa_minref */
6269 spa
->spa_minref
+= 3;
6272 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6273 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6274 spa_feature_create_zap_objects(spa
, tx
);
6278 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6279 * when possibility to use lz4 compression for metadata was added
6280 * Old pools that have this feature enabled must be upgraded to have
6281 * this feature active
6283 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6284 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6285 SPA_FEATURE_LZ4_COMPRESS
);
6286 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6287 SPA_FEATURE_LZ4_COMPRESS
);
6289 if (lz4_en
&& !lz4_ac
)
6290 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6292 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6296 * Sync the specified transaction group. New blocks may be dirtied as
6297 * part of the process, so we iterate until it converges.
6300 spa_sync(spa_t
*spa
, uint64_t txg
)
6302 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6303 objset_t
*mos
= spa
->spa_meta_objset
;
6304 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6305 vdev_t
*rvd
= spa
->spa_root_vdev
;
6311 VERIFY(spa_writeable(spa
));
6314 * Lock out configuration changes.
6316 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6318 spa
->spa_syncing_txg
= txg
;
6319 spa
->spa_sync_pass
= 0;
6322 * If there are any pending vdev state changes, convert them
6323 * into config changes that go out with this transaction group.
6325 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6326 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6328 * We need the write lock here because, for aux vdevs,
6329 * calling vdev_config_dirty() modifies sav_config.
6330 * This is ugly and will become unnecessary when we
6331 * eliminate the aux vdev wart by integrating all vdevs
6332 * into the root vdev tree.
6334 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6335 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6336 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6337 vdev_state_clean(vd
);
6338 vdev_config_dirty(vd
);
6340 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6341 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6343 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6345 tx
= dmu_tx_create_assigned(dp
, txg
);
6347 spa
->spa_sync_starttime
= gethrtime();
6348 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6349 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
6350 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6351 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6354 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6355 * set spa_deflate if we have no raid-z vdevs.
6357 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6358 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6361 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6362 vd
= rvd
->vdev_child
[i
];
6363 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6366 if (i
== rvd
->vdev_children
) {
6367 spa
->spa_deflate
= TRUE
;
6368 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6369 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6370 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6375 * If anything has changed in this txg, or if someone is waiting
6376 * for this txg to sync (eg, spa_vdev_remove()), push the
6377 * deferred frees from the previous txg. If not, leave them
6378 * alone so that we don't generate work on an otherwise idle
6381 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
6382 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
6383 !txg_list_empty(&dp
->dp_sync_tasks
, txg
) ||
6384 ((dsl_scan_active(dp
->dp_scan
) ||
6385 txg_sync_waiting(dp
)) && !spa_shutting_down(spa
))) {
6386 spa_sync_deferred_frees(spa
, tx
);
6390 * Iterate to convergence.
6393 int pass
= ++spa
->spa_sync_pass
;
6395 spa_sync_config_object(spa
, tx
);
6396 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6397 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6398 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6399 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6400 spa_errlog_sync(spa
, txg
);
6401 dsl_pool_sync(dp
, txg
);
6403 if (pass
< zfs_sync_pass_deferred_free
) {
6404 spa_sync_frees(spa
, free_bpl
, tx
);
6406 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6407 &spa
->spa_deferred_bpobj
, tx
);
6411 dsl_scan_sync(dp
, tx
);
6413 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6417 spa_sync_upgrades(spa
, tx
);
6419 } while (dmu_objset_is_dirty(mos
, txg
));
6422 * Rewrite the vdev configuration (which includes the uberblock)
6423 * to commit the transaction group.
6425 * If there are no dirty vdevs, we sync the uberblock to a few
6426 * random top-level vdevs that are known to be visible in the
6427 * config cache (see spa_vdev_add() for a complete description).
6428 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6432 * We hold SCL_STATE to prevent vdev open/close/etc.
6433 * while we're attempting to write the vdev labels.
6435 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6437 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6438 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6440 int children
= rvd
->vdev_children
;
6441 int c0
= spa_get_random(children
);
6443 for (c
= 0; c
< children
; c
++) {
6444 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6445 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6447 svd
[svdcount
++] = vd
;
6448 if (svdcount
== SPA_DVAS_PER_BP
)
6451 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
6453 error
= vdev_config_sync(svd
, svdcount
, txg
,
6456 error
= vdev_config_sync(rvd
->vdev_child
,
6457 rvd
->vdev_children
, txg
, B_FALSE
);
6459 error
= vdev_config_sync(rvd
->vdev_child
,
6460 rvd
->vdev_children
, txg
, B_TRUE
);
6464 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6466 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6470 zio_suspend(spa
, NULL
);
6471 zio_resume_wait(spa
);
6475 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6476 spa
->spa_deadman_tqid
= 0;
6479 * Clear the dirty config list.
6481 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6482 vdev_config_clean(vd
);
6485 * Now that the new config has synced transactionally,
6486 * let it become visible to the config cache.
6488 if (spa
->spa_config_syncing
!= NULL
) {
6489 spa_config_set(spa
, spa
->spa_config_syncing
);
6490 spa
->spa_config_txg
= txg
;
6491 spa
->spa_config_syncing
= NULL
;
6494 spa
->spa_ubsync
= spa
->spa_uberblock
;
6496 dsl_pool_sync_done(dp
, txg
);
6499 * Update usable space statistics.
6501 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6502 vdev_sync_done(vd
, txg
);
6504 spa_update_dspace(spa
);
6507 * It had better be the case that we didn't dirty anything
6508 * since vdev_config_sync().
6510 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6511 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6512 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6514 spa
->spa_sync_pass
= 0;
6516 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6518 spa_handle_ignored_writes(spa
);
6521 * If any async tasks have been requested, kick them off.
6523 spa_async_dispatch(spa
);
6527 * Sync all pools. We don't want to hold the namespace lock across these
6528 * operations, so we take a reference on the spa_t and drop the lock during the
6532 spa_sync_allpools(void)
6535 mutex_enter(&spa_namespace_lock
);
6536 while ((spa
= spa_next(spa
)) != NULL
) {
6537 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6538 !spa_writeable(spa
) || spa_suspended(spa
))
6540 spa_open_ref(spa
, FTAG
);
6541 mutex_exit(&spa_namespace_lock
);
6542 txg_wait_synced(spa_get_dsl(spa
), 0);
6543 mutex_enter(&spa_namespace_lock
);
6544 spa_close(spa
, FTAG
);
6546 mutex_exit(&spa_namespace_lock
);
6550 * ==========================================================================
6551 * Miscellaneous routines
6552 * ==========================================================================
6556 * Remove all pools in the system.
6564 * Remove all cached state. All pools should be closed now,
6565 * so every spa in the AVL tree should be unreferenced.
6567 mutex_enter(&spa_namespace_lock
);
6568 while ((spa
= spa_next(NULL
)) != NULL
) {
6570 * Stop async tasks. The async thread may need to detach
6571 * a device that's been replaced, which requires grabbing
6572 * spa_namespace_lock, so we must drop it here.
6574 spa_open_ref(spa
, FTAG
);
6575 mutex_exit(&spa_namespace_lock
);
6576 spa_async_suspend(spa
);
6577 mutex_enter(&spa_namespace_lock
);
6578 spa_close(spa
, FTAG
);
6580 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6582 spa_deactivate(spa
);
6586 mutex_exit(&spa_namespace_lock
);
6590 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6595 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6599 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6600 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6601 if (vd
->vdev_guid
== guid
)
6605 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6606 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6607 if (vd
->vdev_guid
== guid
)
6616 spa_upgrade(spa_t
*spa
, uint64_t version
)
6618 ASSERT(spa_writeable(spa
));
6620 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6623 * This should only be called for a non-faulted pool, and since a
6624 * future version would result in an unopenable pool, this shouldn't be
6627 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
6628 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
6630 spa
->spa_uberblock
.ub_version
= version
;
6631 vdev_config_dirty(spa
->spa_root_vdev
);
6633 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6635 txg_wait_synced(spa_get_dsl(spa
), 0);
6639 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6643 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6645 for (i
= 0; i
< sav
->sav_count
; i
++)
6646 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6649 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6650 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6651 &spareguid
) == 0 && spareguid
== guid
)
6659 * Check if a pool has an active shared spare device.
6660 * Note: reference count of an active spare is 2, as a spare and as a replace
6663 spa_has_active_shared_spare(spa_t
*spa
)
6667 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6669 for (i
= 0; i
< sav
->sav_count
; i
++) {
6670 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6671 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6680 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6681 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6682 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6683 * or zdb as real changes.
6686 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6689 zfs_ereport_post(name
, spa
, vd
, NULL
, 0, 0);
6693 #if defined(_KERNEL) && defined(HAVE_SPL)
6694 /* state manipulation functions */
6695 EXPORT_SYMBOL(spa_open
);
6696 EXPORT_SYMBOL(spa_open_rewind
);
6697 EXPORT_SYMBOL(spa_get_stats
);
6698 EXPORT_SYMBOL(spa_create
);
6699 EXPORT_SYMBOL(spa_import_rootpool
);
6700 EXPORT_SYMBOL(spa_import
);
6701 EXPORT_SYMBOL(spa_tryimport
);
6702 EXPORT_SYMBOL(spa_destroy
);
6703 EXPORT_SYMBOL(spa_export
);
6704 EXPORT_SYMBOL(spa_reset
);
6705 EXPORT_SYMBOL(spa_async_request
);
6706 EXPORT_SYMBOL(spa_async_suspend
);
6707 EXPORT_SYMBOL(spa_async_resume
);
6708 EXPORT_SYMBOL(spa_inject_addref
);
6709 EXPORT_SYMBOL(spa_inject_delref
);
6710 EXPORT_SYMBOL(spa_scan_stat_init
);
6711 EXPORT_SYMBOL(spa_scan_get_stats
);
6713 /* device maniion */
6714 EXPORT_SYMBOL(spa_vdev_add
);
6715 EXPORT_SYMBOL(spa_vdev_attach
);
6716 EXPORT_SYMBOL(spa_vdev_detach
);
6717 EXPORT_SYMBOL(spa_vdev_remove
);
6718 EXPORT_SYMBOL(spa_vdev_setpath
);
6719 EXPORT_SYMBOL(spa_vdev_setfru
);
6720 EXPORT_SYMBOL(spa_vdev_split_mirror
);
6722 /* spare statech is global across all pools) */
6723 EXPORT_SYMBOL(spa_spare_add
);
6724 EXPORT_SYMBOL(spa_spare_remove
);
6725 EXPORT_SYMBOL(spa_spare_exists
);
6726 EXPORT_SYMBOL(spa_spare_activate
);
6728 /* L2ARC statech is global across all pools) */
6729 EXPORT_SYMBOL(spa_l2cache_add
);
6730 EXPORT_SYMBOL(spa_l2cache_remove
);
6731 EXPORT_SYMBOL(spa_l2cache_exists
);
6732 EXPORT_SYMBOL(spa_l2cache_activate
);
6733 EXPORT_SYMBOL(spa_l2cache_drop
);
6736 EXPORT_SYMBOL(spa_scan
);
6737 EXPORT_SYMBOL(spa_scan_stop
);
6740 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
6741 EXPORT_SYMBOL(spa_sync_allpools
);
6744 EXPORT_SYMBOL(spa_prop_set
);
6745 EXPORT_SYMBOL(spa_prop_get
);
6746 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
6748 /* asynchronous event notification */
6749 EXPORT_SYMBOL(spa_event_notify
);
6752 #if defined(_KERNEL) && defined(HAVE_SPL)
6753 module_param(spa_load_verify_maxinflight
, int, 0644);
6754 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
6755 "Max concurrent traversal I/Os while verifying pool during import -X");
6757 module_param(spa_load_verify_metadata
, int, 0644);
6758 MODULE_PARM_DESC(spa_load_verify_metadata
,
6759 "Set to traverse metadata on pool import");
6761 module_param(spa_load_verify_data
, int, 0644);
6762 MODULE_PARM_DESC(spa_load_verify_data
,
6763 "Set to traverse data on pool import");