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 (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
270 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
271 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
273 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
274 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
277 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
278 if (dp
->scd_path
== NULL
) {
279 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
280 "none", 0, ZPROP_SRC_LOCAL
);
281 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
282 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
283 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
289 * Get zpool property values.
292 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
294 objset_t
*mos
= spa
->spa_meta_objset
;
299 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
303 mutex_enter(&spa
->spa_props_lock
);
306 * Get properties from the spa config.
308 spa_prop_get_config(spa
, nvp
);
310 /* If no pool property object, no more prop to get. */
311 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
312 mutex_exit(&spa
->spa_props_lock
);
317 * Get properties from the MOS pool property object.
319 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
320 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
321 zap_cursor_advance(&zc
)) {
324 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
327 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
330 switch (za
.za_integer_length
) {
332 /* integer property */
333 if (za
.za_first_integer
!=
334 zpool_prop_default_numeric(prop
))
335 src
= ZPROP_SRC_LOCAL
;
337 if (prop
== ZPOOL_PROP_BOOTFS
) {
339 dsl_dataset_t
*ds
= NULL
;
341 dp
= spa_get_dsl(spa
);
342 dsl_pool_config_enter(dp
, FTAG
);
343 if ((err
= dsl_dataset_hold_obj(dp
,
344 za
.za_first_integer
, FTAG
, &ds
))) {
345 dsl_pool_config_exit(dp
, FTAG
);
350 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1,
352 dsl_dataset_name(ds
, strval
);
353 dsl_dataset_rele(ds
, FTAG
);
354 dsl_pool_config_exit(dp
, FTAG
);
357 intval
= za
.za_first_integer
;
360 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
364 MAXNAMELEN
+ strlen(MOS_DIR_NAME
) + 1);
369 /* string property */
370 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
371 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
372 za
.za_name
, 1, za
.za_num_integers
, strval
);
374 kmem_free(strval
, za
.za_num_integers
);
377 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
378 kmem_free(strval
, za
.za_num_integers
);
385 zap_cursor_fini(&zc
);
386 mutex_exit(&spa
->spa_props_lock
);
388 if (err
&& err
!= ENOENT
) {
398 * Validate the given pool properties nvlist and modify the list
399 * for the property values to be set.
402 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
405 int error
= 0, reset_bootfs
= 0;
407 boolean_t has_feature
= B_FALSE
;
410 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
412 char *strval
, *slash
, *check
, *fname
;
413 const char *propname
= nvpair_name(elem
);
414 zpool_prop_t prop
= zpool_name_to_prop(propname
);
418 if (!zpool_prop_feature(propname
)) {
419 error
= SET_ERROR(EINVAL
);
424 * Sanitize the input.
426 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
427 error
= SET_ERROR(EINVAL
);
431 if (nvpair_value_uint64(elem
, &intval
) != 0) {
432 error
= SET_ERROR(EINVAL
);
437 error
= SET_ERROR(EINVAL
);
441 fname
= strchr(propname
, '@') + 1;
442 if (zfeature_lookup_name(fname
, NULL
) != 0) {
443 error
= SET_ERROR(EINVAL
);
447 has_feature
= B_TRUE
;
450 case ZPOOL_PROP_VERSION
:
451 error
= nvpair_value_uint64(elem
, &intval
);
453 (intval
< spa_version(spa
) ||
454 intval
> SPA_VERSION_BEFORE_FEATURES
||
456 error
= SET_ERROR(EINVAL
);
459 case ZPOOL_PROP_DELEGATION
:
460 case ZPOOL_PROP_AUTOREPLACE
:
461 case ZPOOL_PROP_LISTSNAPS
:
462 case ZPOOL_PROP_AUTOEXPAND
:
463 error
= nvpair_value_uint64(elem
, &intval
);
464 if (!error
&& intval
> 1)
465 error
= SET_ERROR(EINVAL
);
468 case ZPOOL_PROP_BOOTFS
:
470 * If the pool version is less than SPA_VERSION_BOOTFS,
471 * or the pool is still being created (version == 0),
472 * the bootfs property cannot be set.
474 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
475 error
= SET_ERROR(ENOTSUP
);
480 * Make sure the vdev config is bootable
482 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
483 error
= SET_ERROR(ENOTSUP
);
489 error
= nvpair_value_string(elem
, &strval
);
495 if (strval
== NULL
|| strval
[0] == '\0') {
496 objnum
= zpool_prop_default_numeric(
501 error
= dmu_objset_hold(strval
, FTAG
, &os
);
506 * Must be ZPL, and its property settings
507 * must be supported by GRUB (compression
508 * is not gzip, and large blocks are not used).
511 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
512 error
= SET_ERROR(ENOTSUP
);
514 dsl_prop_get_int_ds(dmu_objset_ds(os
),
515 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
517 !BOOTFS_COMPRESS_VALID(propval
)) {
518 error
= SET_ERROR(ENOTSUP
);
520 dsl_prop_get_int_ds(dmu_objset_ds(os
),
521 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
523 propval
> SPA_OLD_MAXBLOCKSIZE
) {
524 error
= SET_ERROR(ENOTSUP
);
526 objnum
= dmu_objset_id(os
);
528 dmu_objset_rele(os
, FTAG
);
532 case ZPOOL_PROP_FAILUREMODE
:
533 error
= nvpair_value_uint64(elem
, &intval
);
534 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
535 intval
> ZIO_FAILURE_MODE_PANIC
))
536 error
= SET_ERROR(EINVAL
);
539 * This is a special case which only occurs when
540 * the pool has completely failed. This allows
541 * the user to change the in-core failmode property
542 * without syncing it out to disk (I/Os might
543 * currently be blocked). We do this by returning
544 * EIO to the caller (spa_prop_set) to trick it
545 * into thinking we encountered a property validation
548 if (!error
&& spa_suspended(spa
)) {
549 spa
->spa_failmode
= intval
;
550 error
= SET_ERROR(EIO
);
554 case ZPOOL_PROP_CACHEFILE
:
555 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
558 if (strval
[0] == '\0')
561 if (strcmp(strval
, "none") == 0)
564 if (strval
[0] != '/') {
565 error
= SET_ERROR(EINVAL
);
569 slash
= strrchr(strval
, '/');
570 ASSERT(slash
!= NULL
);
572 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
573 strcmp(slash
, "/..") == 0)
574 error
= SET_ERROR(EINVAL
);
577 case ZPOOL_PROP_COMMENT
:
578 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
580 for (check
= strval
; *check
!= '\0'; check
++) {
581 if (!isprint(*check
)) {
582 error
= SET_ERROR(EINVAL
);
587 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
588 error
= SET_ERROR(E2BIG
);
591 case ZPOOL_PROP_DEDUPDITTO
:
592 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
593 error
= SET_ERROR(ENOTSUP
);
595 error
= nvpair_value_uint64(elem
, &intval
);
597 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
598 error
= SET_ERROR(EINVAL
);
609 if (!error
&& reset_bootfs
) {
610 error
= nvlist_remove(props
,
611 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
614 error
= nvlist_add_uint64(props
,
615 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
623 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
626 spa_config_dirent_t
*dp
;
628 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
632 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
635 if (cachefile
[0] == '\0')
636 dp
->scd_path
= spa_strdup(spa_config_path
);
637 else if (strcmp(cachefile
, "none") == 0)
640 dp
->scd_path
= spa_strdup(cachefile
);
642 list_insert_head(&spa
->spa_config_list
, dp
);
644 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
648 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
651 nvpair_t
*elem
= NULL
;
652 boolean_t need_sync
= B_FALSE
;
654 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
657 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
658 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
660 if (prop
== ZPOOL_PROP_CACHEFILE
||
661 prop
== ZPOOL_PROP_ALTROOT
||
662 prop
== ZPOOL_PROP_READONLY
)
665 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPROP_INVAL
) {
668 if (prop
== ZPOOL_PROP_VERSION
) {
669 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
671 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
672 ver
= SPA_VERSION_FEATURES
;
676 /* Save time if the version is already set. */
677 if (ver
== spa_version(spa
))
681 * In addition to the pool directory object, we might
682 * create the pool properties object, the features for
683 * read object, the features for write object, or the
684 * feature descriptions object.
686 error
= dsl_sync_task(spa
->spa_name
, NULL
,
687 spa_sync_version
, &ver
,
688 6, ZFS_SPACE_CHECK_RESERVED
);
699 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
700 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
707 * If the bootfs property value is dsobj, clear it.
710 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
712 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
713 VERIFY(zap_remove(spa
->spa_meta_objset
,
714 spa
->spa_pool_props_object
,
715 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
722 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
724 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
725 vdev_t
*rvd
= spa
->spa_root_vdev
;
727 ASSERTV(uint64_t *newguid
= arg
);
729 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
730 vdev_state
= rvd
->vdev_state
;
731 spa_config_exit(spa
, SCL_STATE
, FTAG
);
733 if (vdev_state
!= VDEV_STATE_HEALTHY
)
734 return (SET_ERROR(ENXIO
));
736 ASSERT3U(spa_guid(spa
), !=, *newguid
);
742 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
744 uint64_t *newguid
= arg
;
745 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
747 vdev_t
*rvd
= spa
->spa_root_vdev
;
749 oldguid
= spa_guid(spa
);
751 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
752 rvd
->vdev_guid
= *newguid
;
753 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
754 vdev_config_dirty(rvd
);
755 spa_config_exit(spa
, SCL_STATE
, FTAG
);
757 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
762 * Change the GUID for the pool. This is done so that we can later
763 * re-import a pool built from a clone of our own vdevs. We will modify
764 * the root vdev's guid, our own pool guid, and then mark all of our
765 * vdevs dirty. Note that we must make sure that all our vdevs are
766 * online when we do this, or else any vdevs that weren't present
767 * would be orphaned from our pool. We are also going to issue a
768 * sysevent to update any watchers.
771 spa_change_guid(spa_t
*spa
)
776 mutex_enter(&spa
->spa_vdev_top_lock
);
777 mutex_enter(&spa_namespace_lock
);
778 guid
= spa_generate_guid(NULL
);
780 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
781 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
784 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
785 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_REGUID
);
788 mutex_exit(&spa_namespace_lock
);
789 mutex_exit(&spa
->spa_vdev_top_lock
);
795 * ==========================================================================
796 * SPA state manipulation (open/create/destroy/import/export)
797 * ==========================================================================
801 spa_error_entry_compare(const void *a
, const void *b
)
803 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
804 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
807 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
808 sizeof (zbookmark_phys_t
));
819 * Utility function which retrieves copies of the current logs and
820 * re-initializes them in the process.
823 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
825 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
827 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
828 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
830 avl_create(&spa
->spa_errlist_scrub
,
831 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
832 offsetof(spa_error_entry_t
, se_avl
));
833 avl_create(&spa
->spa_errlist_last
,
834 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
835 offsetof(spa_error_entry_t
, se_avl
));
839 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
841 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
842 enum zti_modes mode
= ztip
->zti_mode
;
843 uint_t value
= ztip
->zti_value
;
844 uint_t count
= ztip
->zti_count
;
845 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
848 boolean_t batch
= B_FALSE
;
850 if (mode
== ZTI_MODE_NULL
) {
852 tqs
->stqs_taskq
= NULL
;
856 ASSERT3U(count
, >, 0);
858 tqs
->stqs_count
= count
;
859 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
863 ASSERT3U(value
, >=, 1);
864 value
= MAX(value
, 1);
869 flags
|= TASKQ_THREADS_CPU_PCT
;
870 value
= zio_taskq_batch_pct
;
874 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
876 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
880 for (i
= 0; i
< count
; i
++) {
884 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
885 zio_type_name
[t
], zio_taskq_types
[q
], i
);
887 (void) snprintf(name
, sizeof (name
), "%s_%s",
888 zio_type_name
[t
], zio_taskq_types
[q
]);
891 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
893 flags
|= TASKQ_DC_BATCH
;
895 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
896 spa
->spa_proc
, zio_taskq_basedc
, flags
);
898 pri_t pri
= maxclsyspri
;
900 * The write issue taskq can be extremely CPU
901 * intensive. Run it at slightly lower priority
902 * than the other taskqs.
904 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
907 tq
= taskq_create_proc(name
, value
, pri
, 50,
908 INT_MAX
, spa
->spa_proc
, flags
);
911 tqs
->stqs_taskq
[i
] = tq
;
916 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
918 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
921 if (tqs
->stqs_taskq
== NULL
) {
922 ASSERT3U(tqs
->stqs_count
, ==, 0);
926 for (i
= 0; i
< tqs
->stqs_count
; i
++) {
927 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
928 taskq_destroy(tqs
->stqs_taskq
[i
]);
931 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
932 tqs
->stqs_taskq
= NULL
;
936 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
937 * Note that a type may have multiple discrete taskqs to avoid lock contention
938 * on the taskq itself. In that case we choose which taskq at random by using
939 * the low bits of gethrtime().
942 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
943 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
945 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
948 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
949 ASSERT3U(tqs
->stqs_count
, !=, 0);
951 if (tqs
->stqs_count
== 1) {
952 tq
= tqs
->stqs_taskq
[0];
954 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
957 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
961 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
964 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
965 task_func_t
*func
, void *arg
, uint_t flags
)
967 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
971 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
972 ASSERT3U(tqs
->stqs_count
, !=, 0);
974 if (tqs
->stqs_count
== 1) {
975 tq
= tqs
->stqs_taskq
[0];
977 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
980 id
= taskq_dispatch(tq
, func
, arg
, flags
);
982 taskq_wait_id(tq
, id
);
986 spa_create_zio_taskqs(spa_t
*spa
)
990 for (t
= 0; t
< ZIO_TYPES
; t
++) {
991 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
992 spa_taskqs_init(spa
, t
, q
);
997 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
999 spa_thread(void *arg
)
1001 callb_cpr_t cprinfo
;
1004 user_t
*pu
= PTOU(curproc
);
1006 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1009 ASSERT(curproc
!= &p0
);
1010 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1011 "zpool-%s", spa
->spa_name
);
1012 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1014 /* bind this thread to the requested psrset */
1015 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1017 mutex_enter(&cpu_lock
);
1018 mutex_enter(&pidlock
);
1019 mutex_enter(&curproc
->p_lock
);
1021 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1022 0, NULL
, NULL
) == 0) {
1023 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1026 "Couldn't bind process for zfs pool \"%s\" to "
1027 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1030 mutex_exit(&curproc
->p_lock
);
1031 mutex_exit(&pidlock
);
1032 mutex_exit(&cpu_lock
);
1036 if (zio_taskq_sysdc
) {
1037 sysdc_thread_enter(curthread
, 100, 0);
1040 spa
->spa_proc
= curproc
;
1041 spa
->spa_did
= curthread
->t_did
;
1043 spa_create_zio_taskqs(spa
);
1045 mutex_enter(&spa
->spa_proc_lock
);
1046 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1048 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1049 cv_broadcast(&spa
->spa_proc_cv
);
1051 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1052 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1053 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1054 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1056 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1057 spa
->spa_proc_state
= SPA_PROC_GONE
;
1058 spa
->spa_proc
= &p0
;
1059 cv_broadcast(&spa
->spa_proc_cv
);
1060 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1062 mutex_enter(&curproc
->p_lock
);
1068 * Activate an uninitialized pool.
1071 spa_activate(spa_t
*spa
, int mode
)
1073 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1075 spa
->spa_state
= POOL_STATE_ACTIVE
;
1076 spa
->spa_mode
= mode
;
1078 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1079 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1081 /* Try to create a covering process */
1082 mutex_enter(&spa
->spa_proc_lock
);
1083 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1084 ASSERT(spa
->spa_proc
== &p0
);
1087 #ifdef HAVE_SPA_THREAD
1088 /* Only create a process if we're going to be around a while. */
1089 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1090 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1092 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1093 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1094 cv_wait(&spa
->spa_proc_cv
,
1095 &spa
->spa_proc_lock
);
1097 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1098 ASSERT(spa
->spa_proc
!= &p0
);
1099 ASSERT(spa
->spa_did
!= 0);
1103 "Couldn't create process for zfs pool \"%s\"\n",
1108 #endif /* HAVE_SPA_THREAD */
1109 mutex_exit(&spa
->spa_proc_lock
);
1111 /* If we didn't create a process, we need to create our taskqs. */
1112 if (spa
->spa_proc
== &p0
) {
1113 spa_create_zio_taskqs(spa
);
1116 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1117 offsetof(vdev_t
, vdev_config_dirty_node
));
1118 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1119 offsetof(objset_t
, os_evicting_node
));
1120 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1121 offsetof(vdev_t
, vdev_state_dirty_node
));
1123 txg_list_create(&spa
->spa_vdev_txg_list
,
1124 offsetof(struct vdev
, vdev_txg_node
));
1126 avl_create(&spa
->spa_errlist_scrub
,
1127 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1128 offsetof(spa_error_entry_t
, se_avl
));
1129 avl_create(&spa
->spa_errlist_last
,
1130 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1131 offsetof(spa_error_entry_t
, se_avl
));
1135 * Opposite of spa_activate().
1138 spa_deactivate(spa_t
*spa
)
1142 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1143 ASSERT(spa
->spa_dsl_pool
== NULL
);
1144 ASSERT(spa
->spa_root_vdev
== NULL
);
1145 ASSERT(spa
->spa_async_zio_root
== NULL
);
1146 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1148 spa_evicting_os_wait(spa
);
1150 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1152 list_destroy(&spa
->spa_config_dirty_list
);
1153 list_destroy(&spa
->spa_evicting_os_list
);
1154 list_destroy(&spa
->spa_state_dirty_list
);
1156 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
1158 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1159 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1160 spa_taskqs_fini(spa
, t
, q
);
1164 metaslab_class_destroy(spa
->spa_normal_class
);
1165 spa
->spa_normal_class
= NULL
;
1167 metaslab_class_destroy(spa
->spa_log_class
);
1168 spa
->spa_log_class
= NULL
;
1171 * If this was part of an import or the open otherwise failed, we may
1172 * still have errors left in the queues. Empty them just in case.
1174 spa_errlog_drain(spa
);
1176 avl_destroy(&spa
->spa_errlist_scrub
);
1177 avl_destroy(&spa
->spa_errlist_last
);
1179 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1181 mutex_enter(&spa
->spa_proc_lock
);
1182 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1183 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1184 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1185 cv_broadcast(&spa
->spa_proc_cv
);
1186 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1187 ASSERT(spa
->spa_proc
!= &p0
);
1188 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1190 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1191 spa
->spa_proc_state
= SPA_PROC_NONE
;
1193 ASSERT(spa
->spa_proc
== &p0
);
1194 mutex_exit(&spa
->spa_proc_lock
);
1197 * We want to make sure spa_thread() has actually exited the ZFS
1198 * module, so that the module can't be unloaded out from underneath
1201 if (spa
->spa_did
!= 0) {
1202 thread_join(spa
->spa_did
);
1208 * Verify a pool configuration, and construct the vdev tree appropriately. This
1209 * will create all the necessary vdevs in the appropriate layout, with each vdev
1210 * in the CLOSED state. This will prep the pool before open/creation/import.
1211 * All vdev validation is done by the vdev_alloc() routine.
1214 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1215 uint_t id
, int atype
)
1222 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1225 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1228 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1231 if (error
== ENOENT
)
1237 return (SET_ERROR(EINVAL
));
1240 for (c
= 0; c
< children
; c
++) {
1242 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1250 ASSERT(*vdp
!= NULL
);
1256 * Opposite of spa_load().
1259 spa_unload(spa_t
*spa
)
1263 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1268 spa_async_suspend(spa
);
1273 if (spa
->spa_sync_on
) {
1274 txg_sync_stop(spa
->spa_dsl_pool
);
1275 spa
->spa_sync_on
= B_FALSE
;
1279 * Wait for any outstanding async I/O to complete.
1281 if (spa
->spa_async_zio_root
!= NULL
) {
1282 for (i
= 0; i
< max_ncpus
; i
++)
1283 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1284 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1285 spa
->spa_async_zio_root
= NULL
;
1288 bpobj_close(&spa
->spa_deferred_bpobj
);
1290 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1295 if (spa
->spa_root_vdev
)
1296 vdev_free(spa
->spa_root_vdev
);
1297 ASSERT(spa
->spa_root_vdev
== NULL
);
1300 * Close the dsl pool.
1302 if (spa
->spa_dsl_pool
) {
1303 dsl_pool_close(spa
->spa_dsl_pool
);
1304 spa
->spa_dsl_pool
= NULL
;
1305 spa
->spa_meta_objset
= NULL
;
1312 * Drop and purge level 2 cache
1314 spa_l2cache_drop(spa
);
1316 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1317 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1318 if (spa
->spa_spares
.sav_vdevs
) {
1319 kmem_free(spa
->spa_spares
.sav_vdevs
,
1320 spa
->spa_spares
.sav_count
* sizeof (void *));
1321 spa
->spa_spares
.sav_vdevs
= NULL
;
1323 if (spa
->spa_spares
.sav_config
) {
1324 nvlist_free(spa
->spa_spares
.sav_config
);
1325 spa
->spa_spares
.sav_config
= NULL
;
1327 spa
->spa_spares
.sav_count
= 0;
1329 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1330 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1331 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1333 if (spa
->spa_l2cache
.sav_vdevs
) {
1334 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1335 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1336 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1338 if (spa
->spa_l2cache
.sav_config
) {
1339 nvlist_free(spa
->spa_l2cache
.sav_config
);
1340 spa
->spa_l2cache
.sav_config
= NULL
;
1342 spa
->spa_l2cache
.sav_count
= 0;
1344 spa
->spa_async_suspended
= 0;
1346 if (spa
->spa_comment
!= NULL
) {
1347 spa_strfree(spa
->spa_comment
);
1348 spa
->spa_comment
= NULL
;
1351 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1355 * Load (or re-load) the current list of vdevs describing the active spares for
1356 * this pool. When this is called, we have some form of basic information in
1357 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1358 * then re-generate a more complete list including status information.
1361 spa_load_spares(spa_t
*spa
)
1368 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1371 * First, close and free any existing spare vdevs.
1373 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1374 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1376 /* Undo the call to spa_activate() below */
1377 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1378 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1379 spa_spare_remove(tvd
);
1384 if (spa
->spa_spares
.sav_vdevs
)
1385 kmem_free(spa
->spa_spares
.sav_vdevs
,
1386 spa
->spa_spares
.sav_count
* sizeof (void *));
1388 if (spa
->spa_spares
.sav_config
== NULL
)
1391 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1392 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1394 spa
->spa_spares
.sav_count
= (int)nspares
;
1395 spa
->spa_spares
.sav_vdevs
= NULL
;
1401 * Construct the array of vdevs, opening them to get status in the
1402 * process. For each spare, there is potentially two different vdev_t
1403 * structures associated with it: one in the list of spares (used only
1404 * for basic validation purposes) and one in the active vdev
1405 * configuration (if it's spared in). During this phase we open and
1406 * validate each vdev on the spare list. If the vdev also exists in the
1407 * active configuration, then we also mark this vdev as an active spare.
1409 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1411 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1412 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1413 VDEV_ALLOC_SPARE
) == 0);
1416 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1418 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1419 B_FALSE
)) != NULL
) {
1420 if (!tvd
->vdev_isspare
)
1424 * We only mark the spare active if we were successfully
1425 * able to load the vdev. Otherwise, importing a pool
1426 * with a bad active spare would result in strange
1427 * behavior, because multiple pool would think the spare
1428 * is actively in use.
1430 * There is a vulnerability here to an equally bizarre
1431 * circumstance, where a dead active spare is later
1432 * brought back to life (onlined or otherwise). Given
1433 * the rarity of this scenario, and the extra complexity
1434 * it adds, we ignore the possibility.
1436 if (!vdev_is_dead(tvd
))
1437 spa_spare_activate(tvd
);
1441 vd
->vdev_aux
= &spa
->spa_spares
;
1443 if (vdev_open(vd
) != 0)
1446 if (vdev_validate_aux(vd
) == 0)
1451 * Recompute the stashed list of spares, with status information
1454 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1455 DATA_TYPE_NVLIST_ARRAY
) == 0);
1457 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1459 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1460 spares
[i
] = vdev_config_generate(spa
,
1461 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1462 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1463 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1464 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1465 nvlist_free(spares
[i
]);
1466 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1470 * Load (or re-load) the current list of vdevs describing the active l2cache for
1471 * this pool. When this is called, we have some form of basic information in
1472 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1473 * then re-generate a more complete list including status information.
1474 * Devices which are already active have their details maintained, and are
1478 spa_load_l2cache(spa_t
*spa
)
1482 int i
, j
, oldnvdevs
;
1484 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1485 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1487 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1489 if (sav
->sav_config
!= NULL
) {
1490 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1491 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1492 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1498 oldvdevs
= sav
->sav_vdevs
;
1499 oldnvdevs
= sav
->sav_count
;
1500 sav
->sav_vdevs
= NULL
;
1504 * Process new nvlist of vdevs.
1506 for (i
= 0; i
< nl2cache
; i
++) {
1507 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1511 for (j
= 0; j
< oldnvdevs
; j
++) {
1513 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1515 * Retain previous vdev for add/remove ops.
1523 if (newvdevs
[i
] == NULL
) {
1527 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1528 VDEV_ALLOC_L2CACHE
) == 0);
1533 * Commit this vdev as an l2cache device,
1534 * even if it fails to open.
1536 spa_l2cache_add(vd
);
1541 spa_l2cache_activate(vd
);
1543 if (vdev_open(vd
) != 0)
1546 (void) vdev_validate_aux(vd
);
1548 if (!vdev_is_dead(vd
))
1549 l2arc_add_vdev(spa
, vd
);
1554 * Purge vdevs that were dropped
1556 for (i
= 0; i
< oldnvdevs
; i
++) {
1561 ASSERT(vd
->vdev_isl2cache
);
1563 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1564 pool
!= 0ULL && l2arc_vdev_present(vd
))
1565 l2arc_remove_vdev(vd
);
1566 vdev_clear_stats(vd
);
1572 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1574 if (sav
->sav_config
== NULL
)
1577 sav
->sav_vdevs
= newvdevs
;
1578 sav
->sav_count
= (int)nl2cache
;
1581 * Recompute the stashed list of l2cache devices, with status
1582 * information this time.
1584 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1585 DATA_TYPE_NVLIST_ARRAY
) == 0);
1587 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1588 for (i
= 0; i
< sav
->sav_count
; i
++)
1589 l2cache
[i
] = vdev_config_generate(spa
,
1590 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1591 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1592 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1594 for (i
= 0; i
< sav
->sav_count
; i
++)
1595 nvlist_free(l2cache
[i
]);
1597 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1601 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1604 char *packed
= NULL
;
1609 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1613 nvsize
= *(uint64_t *)db
->db_data
;
1614 dmu_buf_rele(db
, FTAG
);
1616 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1617 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1620 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1621 vmem_free(packed
, nvsize
);
1627 * Checks to see if the given vdev could not be opened, in which case we post a
1628 * sysevent to notify the autoreplace code that the device has been removed.
1631 spa_check_removed(vdev_t
*vd
)
1635 for (c
= 0; c
< vd
->vdev_children
; c
++)
1636 spa_check_removed(vd
->vdev_child
[c
]);
1638 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1640 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE
,
1641 vd
->vdev_spa
, vd
, NULL
, 0, 0);
1642 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_CHECK
);
1647 * Validate the current config against the MOS config
1650 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1652 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1656 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1658 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1659 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1661 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1664 * If we're doing a normal import, then build up any additional
1665 * diagnostic information about missing devices in this config.
1666 * We'll pass this up to the user for further processing.
1668 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1669 nvlist_t
**child
, *nv
;
1672 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1674 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1676 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1677 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1678 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1680 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1681 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1683 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1688 VERIFY(nvlist_add_nvlist_array(nv
,
1689 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1690 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1691 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1693 for (i
= 0; i
< idx
; i
++)
1694 nvlist_free(child
[i
]);
1697 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1701 * Compare the root vdev tree with the information we have
1702 * from the MOS config (mrvd). Check each top-level vdev
1703 * with the corresponding MOS config top-level (mtvd).
1705 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1706 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1707 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1710 * Resolve any "missing" vdevs in the current configuration.
1711 * If we find that the MOS config has more accurate information
1712 * about the top-level vdev then use that vdev instead.
1714 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1715 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1717 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1721 * Device specific actions.
1723 if (mtvd
->vdev_islog
) {
1724 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1727 * XXX - once we have 'readonly' pool
1728 * support we should be able to handle
1729 * missing data devices by transitioning
1730 * the pool to readonly.
1736 * Swap the missing vdev with the data we were
1737 * able to obtain from the MOS config.
1739 vdev_remove_child(rvd
, tvd
);
1740 vdev_remove_child(mrvd
, mtvd
);
1742 vdev_add_child(rvd
, mtvd
);
1743 vdev_add_child(mrvd
, tvd
);
1745 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1747 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1750 } else if (mtvd
->vdev_islog
) {
1752 * Load the slog device's state from the MOS config
1753 * since it's possible that the label does not
1754 * contain the most up-to-date information.
1756 vdev_load_log_state(tvd
, mtvd
);
1761 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1764 * Ensure we were able to validate the config.
1766 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1770 * Check for missing log devices
1773 spa_check_logs(spa_t
*spa
)
1775 boolean_t rv
= B_FALSE
;
1777 switch (spa
->spa_log_state
) {
1780 case SPA_LOG_MISSING
:
1781 /* need to recheck in case slog has been restored */
1782 case SPA_LOG_UNKNOWN
:
1783 rv
= (dmu_objset_find(spa
->spa_name
, zil_check_log_chain
,
1784 NULL
, DS_FIND_CHILDREN
) != 0);
1786 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1793 spa_passivate_log(spa_t
*spa
)
1795 vdev_t
*rvd
= spa
->spa_root_vdev
;
1796 boolean_t slog_found
= B_FALSE
;
1799 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1801 if (!spa_has_slogs(spa
))
1804 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1805 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1806 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1808 if (tvd
->vdev_islog
) {
1809 metaslab_group_passivate(mg
);
1810 slog_found
= B_TRUE
;
1814 return (slog_found
);
1818 spa_activate_log(spa_t
*spa
)
1820 vdev_t
*rvd
= spa
->spa_root_vdev
;
1823 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1825 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1826 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1827 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1829 if (tvd
->vdev_islog
)
1830 metaslab_group_activate(mg
);
1835 spa_offline_log(spa_t
*spa
)
1839 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1840 NULL
, DS_FIND_CHILDREN
);
1843 * We successfully offlined the log device, sync out the
1844 * current txg so that the "stubby" block can be removed
1847 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1853 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1857 for (i
= 0; i
< sav
->sav_count
; i
++)
1858 spa_check_removed(sav
->sav_vdevs
[i
]);
1862 spa_claim_notify(zio_t
*zio
)
1864 spa_t
*spa
= zio
->io_spa
;
1869 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1870 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1871 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1872 mutex_exit(&spa
->spa_props_lock
);
1875 typedef struct spa_load_error
{
1876 uint64_t sle_meta_count
;
1877 uint64_t sle_data_count
;
1881 spa_load_verify_done(zio_t
*zio
)
1883 blkptr_t
*bp
= zio
->io_bp
;
1884 spa_load_error_t
*sle
= zio
->io_private
;
1885 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1886 int error
= zio
->io_error
;
1887 spa_t
*spa
= zio
->io_spa
;
1890 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1891 type
!= DMU_OT_INTENT_LOG
)
1892 atomic_add_64(&sle
->sle_meta_count
, 1);
1894 atomic_add_64(&sle
->sle_data_count
, 1);
1896 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1898 mutex_enter(&spa
->spa_scrub_lock
);
1899 spa
->spa_scrub_inflight
--;
1900 cv_broadcast(&spa
->spa_scrub_io_cv
);
1901 mutex_exit(&spa
->spa_scrub_lock
);
1905 * Maximum number of concurrent scrub i/os to create while verifying
1906 * a pool while importing it.
1908 int spa_load_verify_maxinflight
= 10000;
1909 int spa_load_verify_metadata
= B_TRUE
;
1910 int spa_load_verify_data
= B_TRUE
;
1914 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1915 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1921 if (BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
1924 * Note: normally this routine will not be called if
1925 * spa_load_verify_metadata is not set. However, it may be useful
1926 * to manually set the flag after the traversal has begun.
1928 if (!spa_load_verify_metadata
)
1930 if (BP_GET_BUFC_TYPE(bp
) == ARC_BUFC_DATA
&& !spa_load_verify_data
)
1934 size
= BP_GET_PSIZE(bp
);
1935 data
= zio_data_buf_alloc(size
);
1937 mutex_enter(&spa
->spa_scrub_lock
);
1938 while (spa
->spa_scrub_inflight
>= spa_load_verify_maxinflight
)
1939 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
1940 spa
->spa_scrub_inflight
++;
1941 mutex_exit(&spa
->spa_scrub_lock
);
1943 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1944 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1945 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1946 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1951 spa_load_verify(spa_t
*spa
)
1954 spa_load_error_t sle
= { 0 };
1955 zpool_rewind_policy_t policy
;
1956 boolean_t verify_ok
= B_FALSE
;
1959 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1961 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1964 rio
= zio_root(spa
, NULL
, &sle
,
1965 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1967 if (spa_load_verify_metadata
) {
1968 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1969 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
,
1970 spa_load_verify_cb
, rio
);
1973 (void) zio_wait(rio
);
1975 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1976 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1978 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1979 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1983 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1984 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1986 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1987 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1988 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1989 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1990 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1991 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1992 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1994 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
1998 if (error
!= ENXIO
&& error
!= EIO
)
1999 error
= SET_ERROR(EIO
);
2003 return (verify_ok
? 0 : EIO
);
2007 * Find a value in the pool props object.
2010 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2012 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2013 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2017 * Find a value in the pool directory object.
2020 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2022 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2023 name
, sizeof (uint64_t), 1, val
));
2027 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2029 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2034 * Fix up config after a partly-completed split. This is done with the
2035 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2036 * pool have that entry in their config, but only the splitting one contains
2037 * a list of all the guids of the vdevs that are being split off.
2039 * This function determines what to do with that list: either rejoin
2040 * all the disks to the pool, or complete the splitting process. To attempt
2041 * the rejoin, each disk that is offlined is marked online again, and
2042 * we do a reopen() call. If the vdev label for every disk that was
2043 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2044 * then we call vdev_split() on each disk, and complete the split.
2046 * Otherwise we leave the config alone, with all the vdevs in place in
2047 * the original pool.
2050 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2057 boolean_t attempt_reopen
;
2059 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2062 /* check that the config is complete */
2063 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2064 &glist
, &gcount
) != 0)
2067 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2069 /* attempt to online all the vdevs & validate */
2070 attempt_reopen
= B_TRUE
;
2071 for (i
= 0; i
< gcount
; i
++) {
2072 if (glist
[i
] == 0) /* vdev is hole */
2075 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2076 if (vd
[i
] == NULL
) {
2078 * Don't bother attempting to reopen the disks;
2079 * just do the split.
2081 attempt_reopen
= B_FALSE
;
2083 /* attempt to re-online it */
2084 vd
[i
]->vdev_offline
= B_FALSE
;
2088 if (attempt_reopen
) {
2089 vdev_reopen(spa
->spa_root_vdev
);
2091 /* check each device to see what state it's in */
2092 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2093 if (vd
[i
] != NULL
&&
2094 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2101 * If every disk has been moved to the new pool, or if we never
2102 * even attempted to look at them, then we split them off for
2105 if (!attempt_reopen
|| gcount
== extracted
) {
2106 for (i
= 0; i
< gcount
; i
++)
2109 vdev_reopen(spa
->spa_root_vdev
);
2112 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2116 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2117 boolean_t mosconfig
)
2119 nvlist_t
*config
= spa
->spa_config
;
2120 char *ereport
= FM_EREPORT_ZFS_POOL
;
2126 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2127 return (SET_ERROR(EINVAL
));
2129 ASSERT(spa
->spa_comment
== NULL
);
2130 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2131 spa
->spa_comment
= spa_strdup(comment
);
2134 * Versioning wasn't explicitly added to the label until later, so if
2135 * it's not present treat it as the initial version.
2137 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2138 &spa
->spa_ubsync
.ub_version
) != 0)
2139 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2141 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2142 &spa
->spa_config_txg
);
2144 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2145 spa_guid_exists(pool_guid
, 0)) {
2146 error
= SET_ERROR(EEXIST
);
2148 spa
->spa_config_guid
= pool_guid
;
2150 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2152 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2156 nvlist_free(spa
->spa_load_info
);
2157 spa
->spa_load_info
= fnvlist_alloc();
2159 gethrestime(&spa
->spa_loaded_ts
);
2160 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2161 mosconfig
, &ereport
);
2165 * Don't count references from objsets that are already closed
2166 * and are making their way through the eviction process.
2168 spa_evicting_os_wait(spa
);
2169 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2171 if (error
!= EEXIST
) {
2172 spa
->spa_loaded_ts
.tv_sec
= 0;
2173 spa
->spa_loaded_ts
.tv_nsec
= 0;
2175 if (error
!= EBADF
) {
2176 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2179 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2186 * Load an existing storage pool, using the pool's builtin spa_config as a
2187 * source of configuration information.
2189 __attribute__((always_inline
))
2191 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2192 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2196 nvlist_t
*nvroot
= NULL
;
2199 uberblock_t
*ub
= &spa
->spa_uberblock
;
2200 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2201 int orig_mode
= spa
->spa_mode
;
2204 boolean_t missing_feat_write
= B_FALSE
;
2207 * If this is an untrusted config, access the pool in read-only mode.
2208 * This prevents things like resilvering recently removed devices.
2211 spa
->spa_mode
= FREAD
;
2213 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2215 spa
->spa_load_state
= state
;
2217 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2218 return (SET_ERROR(EINVAL
));
2220 parse
= (type
== SPA_IMPORT_EXISTING
?
2221 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2224 * Create "The Godfather" zio to hold all async IOs
2226 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2228 for (i
= 0; i
< max_ncpus
; i
++) {
2229 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2230 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2231 ZIO_FLAG_GODFATHER
);
2235 * Parse the configuration into a vdev tree. We explicitly set the
2236 * value that will be returned by spa_version() since parsing the
2237 * configuration requires knowing the version number.
2239 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2240 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2241 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2246 ASSERT(spa
->spa_root_vdev
== rvd
);
2248 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2249 ASSERT(spa_guid(spa
) == pool_guid
);
2253 * Try to open all vdevs, loading each label in the process.
2255 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2256 error
= vdev_open(rvd
);
2257 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2262 * We need to validate the vdev labels against the configuration that
2263 * we have in hand, which is dependent on the setting of mosconfig. If
2264 * mosconfig is true then we're validating the vdev labels based on
2265 * that config. Otherwise, we're validating against the cached config
2266 * (zpool.cache) that was read when we loaded the zfs module, and then
2267 * later we will recursively call spa_load() and validate against
2270 * If we're assembling a new pool that's been split off from an
2271 * existing pool, the labels haven't yet been updated so we skip
2272 * validation for now.
2274 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2275 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2276 error
= vdev_validate(rvd
, mosconfig
);
2277 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2282 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2283 return (SET_ERROR(ENXIO
));
2287 * Find the best uberblock.
2289 vdev_uberblock_load(rvd
, ub
, &label
);
2292 * If we weren't able to find a single valid uberblock, return failure.
2294 if (ub
->ub_txg
== 0) {
2296 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2300 * If the pool has an unsupported version we can't open it.
2302 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2304 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2307 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2311 * If we weren't able to find what's necessary for reading the
2312 * MOS in the label, return failure.
2314 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2315 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2317 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2322 * Update our in-core representation with the definitive values
2325 nvlist_free(spa
->spa_label_features
);
2326 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2332 * Look through entries in the label nvlist's features_for_read. If
2333 * there is a feature listed there which we don't understand then we
2334 * cannot open a pool.
2336 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2337 nvlist_t
*unsup_feat
;
2340 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2343 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2345 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2346 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2347 VERIFY(nvlist_add_string(unsup_feat
,
2348 nvpair_name(nvp
), "") == 0);
2352 if (!nvlist_empty(unsup_feat
)) {
2353 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2354 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2355 nvlist_free(unsup_feat
);
2356 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2360 nvlist_free(unsup_feat
);
2364 * If the vdev guid sum doesn't match the uberblock, we have an
2365 * incomplete configuration. We first check to see if the pool
2366 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2367 * If it is, defer the vdev_guid_sum check till later so we
2368 * can handle missing vdevs.
2370 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2371 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2372 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2373 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2375 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2376 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2377 spa_try_repair(spa
, config
);
2378 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2379 nvlist_free(spa
->spa_config_splitting
);
2380 spa
->spa_config_splitting
= NULL
;
2384 * Initialize internal SPA structures.
2386 spa
->spa_state
= POOL_STATE_ACTIVE
;
2387 spa
->spa_ubsync
= spa
->spa_uberblock
;
2388 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2389 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2390 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2391 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2392 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2393 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2395 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2397 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2398 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2400 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2401 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2403 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2404 boolean_t missing_feat_read
= B_FALSE
;
2405 nvlist_t
*unsup_feat
, *enabled_feat
;
2408 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2409 &spa
->spa_feat_for_read_obj
) != 0) {
2410 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2413 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2414 &spa
->spa_feat_for_write_obj
) != 0) {
2415 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2418 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2419 &spa
->spa_feat_desc_obj
) != 0) {
2420 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2423 enabled_feat
= fnvlist_alloc();
2424 unsup_feat
= fnvlist_alloc();
2426 if (!spa_features_check(spa
, B_FALSE
,
2427 unsup_feat
, enabled_feat
))
2428 missing_feat_read
= B_TRUE
;
2430 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2431 if (!spa_features_check(spa
, B_TRUE
,
2432 unsup_feat
, enabled_feat
)) {
2433 missing_feat_write
= B_TRUE
;
2437 fnvlist_add_nvlist(spa
->spa_load_info
,
2438 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2440 if (!nvlist_empty(unsup_feat
)) {
2441 fnvlist_add_nvlist(spa
->spa_load_info
,
2442 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2445 fnvlist_free(enabled_feat
);
2446 fnvlist_free(unsup_feat
);
2448 if (!missing_feat_read
) {
2449 fnvlist_add_boolean(spa
->spa_load_info
,
2450 ZPOOL_CONFIG_CAN_RDONLY
);
2454 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2455 * twofold: to determine whether the pool is available for
2456 * import in read-write mode and (if it is not) whether the
2457 * pool is available for import in read-only mode. If the pool
2458 * is available for import in read-write mode, it is displayed
2459 * as available in userland; if it is not available for import
2460 * in read-only mode, it is displayed as unavailable in
2461 * userland. If the pool is available for import in read-only
2462 * mode but not read-write mode, it is displayed as unavailable
2463 * in userland with a special note that the pool is actually
2464 * available for open in read-only mode.
2466 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2467 * missing a feature for write, we must first determine whether
2468 * the pool can be opened read-only before returning to
2469 * userland in order to know whether to display the
2470 * abovementioned note.
2472 if (missing_feat_read
|| (missing_feat_write
&&
2473 spa_writeable(spa
))) {
2474 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2479 * Load refcounts for ZFS features from disk into an in-memory
2480 * cache during SPA initialization.
2482 for (i
= 0; i
< SPA_FEATURES
; i
++) {
2485 error
= feature_get_refcount_from_disk(spa
,
2486 &spa_feature_table
[i
], &refcount
);
2488 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2489 } else if (error
== ENOTSUP
) {
2490 spa
->spa_feat_refcount_cache
[i
] =
2491 SPA_FEATURE_DISABLED
;
2493 return (spa_vdev_err(rvd
,
2494 VDEV_AUX_CORRUPT_DATA
, EIO
));
2499 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2500 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2501 &spa
->spa_feat_enabled_txg_obj
) != 0)
2502 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2505 spa
->spa_is_initializing
= B_TRUE
;
2506 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2507 spa
->spa_is_initializing
= B_FALSE
;
2509 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2513 nvlist_t
*policy
= NULL
, *nvconfig
;
2515 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2516 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2518 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2519 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2521 unsigned long myhostid
= 0;
2523 VERIFY(nvlist_lookup_string(nvconfig
,
2524 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2527 myhostid
= zone_get_hostid(NULL
);
2530 * We're emulating the system's hostid in userland, so
2531 * we can't use zone_get_hostid().
2533 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2534 #endif /* _KERNEL */
2535 if (hostid
!= 0 && myhostid
!= 0 &&
2536 hostid
!= myhostid
) {
2537 nvlist_free(nvconfig
);
2538 cmn_err(CE_WARN
, "pool '%s' could not be "
2539 "loaded as it was last accessed by another "
2540 "system (host: %s hostid: 0x%lx). See: "
2541 "http://zfsonlinux.org/msg/ZFS-8000-EY",
2542 spa_name(spa
), hostname
,
2543 (unsigned long)hostid
);
2544 return (SET_ERROR(EBADF
));
2547 if (nvlist_lookup_nvlist(spa
->spa_config
,
2548 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2549 VERIFY(nvlist_add_nvlist(nvconfig
,
2550 ZPOOL_REWIND_POLICY
, policy
) == 0);
2552 spa_config_set(spa
, nvconfig
);
2554 spa_deactivate(spa
);
2555 spa_activate(spa
, orig_mode
);
2557 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2560 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2561 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2562 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2564 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2567 * Load the bit that tells us to use the new accounting function
2568 * (raid-z deflation). If we have an older pool, this will not
2571 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2572 if (error
!= 0 && error
!= ENOENT
)
2573 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2575 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2576 &spa
->spa_creation_version
);
2577 if (error
!= 0 && error
!= ENOENT
)
2578 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2581 * Load the persistent error log. If we have an older pool, this will
2584 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2585 if (error
!= 0 && error
!= ENOENT
)
2586 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2588 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2589 &spa
->spa_errlog_scrub
);
2590 if (error
!= 0 && error
!= ENOENT
)
2591 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2594 * Load the history object. If we have an older pool, this
2595 * will not be present.
2597 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2598 if (error
!= 0 && error
!= ENOENT
)
2599 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2602 * If we're assembling the pool from the split-off vdevs of
2603 * an existing pool, we don't want to attach the spares & cache
2608 * Load any hot spares for this pool.
2610 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2611 if (error
!= 0 && error
!= ENOENT
)
2612 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2613 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2614 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2615 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2616 &spa
->spa_spares
.sav_config
) != 0)
2617 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2619 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2620 spa_load_spares(spa
);
2621 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2622 } else if (error
== 0) {
2623 spa
->spa_spares
.sav_sync
= B_TRUE
;
2627 * Load any level 2 ARC devices for this pool.
2629 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2630 &spa
->spa_l2cache
.sav_object
);
2631 if (error
!= 0 && error
!= ENOENT
)
2632 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2633 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2634 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2635 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2636 &spa
->spa_l2cache
.sav_config
) != 0)
2637 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2639 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2640 spa_load_l2cache(spa
);
2641 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2642 } else if (error
== 0) {
2643 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2646 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2648 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2649 if (error
&& error
!= ENOENT
)
2650 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2653 uint64_t autoreplace
= 0;
2655 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2656 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2657 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2658 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2659 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2660 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2661 &spa
->spa_dedup_ditto
);
2663 spa
->spa_autoreplace
= (autoreplace
!= 0);
2667 * If the 'autoreplace' property is set, then post a resource notifying
2668 * the ZFS DE that it should not issue any faults for unopenable
2669 * devices. We also iterate over the vdevs, and post a sysevent for any
2670 * unopenable vdevs so that the normal autoreplace handler can take
2673 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2674 spa_check_removed(spa
->spa_root_vdev
);
2676 * For the import case, this is done in spa_import(), because
2677 * at this point we're using the spare definitions from
2678 * the MOS config, not necessarily from the userland config.
2680 if (state
!= SPA_LOAD_IMPORT
) {
2681 spa_aux_check_removed(&spa
->spa_spares
);
2682 spa_aux_check_removed(&spa
->spa_l2cache
);
2687 * Load the vdev state for all toplevel vdevs.
2692 * Propagate the leaf DTLs we just loaded all the way up the tree.
2694 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2695 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2696 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2699 * Load the DDTs (dedup tables).
2701 error
= ddt_load(spa
);
2703 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2705 spa_update_dspace(spa
);
2708 * Validate the config, using the MOS config to fill in any
2709 * information which might be missing. If we fail to validate
2710 * the config then declare the pool unfit for use. If we're
2711 * assembling a pool from a split, the log is not transferred
2714 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2717 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2718 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2720 if (!spa_config_valid(spa
, nvconfig
)) {
2721 nvlist_free(nvconfig
);
2722 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2725 nvlist_free(nvconfig
);
2728 * Now that we've validated the config, check the state of the
2729 * root vdev. If it can't be opened, it indicates one or
2730 * more toplevel vdevs are faulted.
2732 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2733 return (SET_ERROR(ENXIO
));
2735 if (spa_writeable(spa
) && spa_check_logs(spa
)) {
2736 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2737 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2741 if (missing_feat_write
) {
2742 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2745 * At this point, we know that we can open the pool in
2746 * read-only mode but not read-write mode. We now have enough
2747 * information and can return to userland.
2749 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2753 * We've successfully opened the pool, verify that we're ready
2754 * to start pushing transactions.
2756 if (state
!= SPA_LOAD_TRYIMPORT
) {
2757 if ((error
= spa_load_verify(spa
)))
2758 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2762 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2763 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2765 int need_update
= B_FALSE
;
2768 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2771 * Claim log blocks that haven't been committed yet.
2772 * This must all happen in a single txg.
2773 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2774 * invoked from zil_claim_log_block()'s i/o done callback.
2775 * Price of rollback is that we abandon the log.
2777 spa
->spa_claiming
= B_TRUE
;
2779 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
),
2780 spa_first_txg(spa
));
2781 (void) dmu_objset_find(spa_name(spa
),
2782 zil_claim
, tx
, DS_FIND_CHILDREN
);
2785 spa
->spa_claiming
= B_FALSE
;
2787 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2788 spa
->spa_sync_on
= B_TRUE
;
2789 txg_sync_start(spa
->spa_dsl_pool
);
2792 * Wait for all claims to sync. We sync up to the highest
2793 * claimed log block birth time so that claimed log blocks
2794 * don't appear to be from the future. spa_claim_max_txg
2795 * will have been set for us by either zil_check_log_chain()
2796 * (invoked from spa_check_logs()) or zil_claim() above.
2798 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2801 * If the config cache is stale, or we have uninitialized
2802 * metaslabs (see spa_vdev_add()), then update the config.
2804 * If this is a verbatim import, trust the current
2805 * in-core spa_config and update the disk labels.
2807 if (config_cache_txg
!= spa
->spa_config_txg
||
2808 state
== SPA_LOAD_IMPORT
||
2809 state
== SPA_LOAD_RECOVER
||
2810 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2811 need_update
= B_TRUE
;
2813 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2814 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2815 need_update
= B_TRUE
;
2818 * Update the config cache asychronously in case we're the
2819 * root pool, in which case the config cache isn't writable yet.
2822 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2825 * Check all DTLs to see if anything needs resilvering.
2827 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2828 vdev_resilver_needed(rvd
, NULL
, NULL
))
2829 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2832 * Log the fact that we booted up (so that we can detect if
2833 * we rebooted in the middle of an operation).
2835 spa_history_log_version(spa
, "open");
2838 * Delete any inconsistent datasets.
2840 (void) dmu_objset_find(spa_name(spa
),
2841 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2844 * Clean up any stale temporary dataset userrefs.
2846 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2853 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2855 int mode
= spa
->spa_mode
;
2858 spa_deactivate(spa
);
2860 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
2862 spa_activate(spa
, mode
);
2863 spa_async_suspend(spa
);
2865 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2869 * If spa_load() fails this function will try loading prior txg's. If
2870 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2871 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2872 * function will not rewind the pool and will return the same error as
2876 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2877 uint64_t max_request
, int rewind_flags
)
2879 nvlist_t
*loadinfo
= NULL
;
2880 nvlist_t
*config
= NULL
;
2881 int load_error
, rewind_error
;
2882 uint64_t safe_rewind_txg
;
2885 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2886 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2887 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2889 spa
->spa_load_max_txg
= max_request
;
2890 if (max_request
!= UINT64_MAX
)
2891 spa
->spa_extreme_rewind
= B_TRUE
;
2894 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2896 if (load_error
== 0)
2899 if (spa
->spa_root_vdev
!= NULL
)
2900 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2902 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2903 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2905 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2906 nvlist_free(config
);
2907 return (load_error
);
2910 if (state
== SPA_LOAD_RECOVER
) {
2911 /* Price of rolling back is discarding txgs, including log */
2912 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2915 * If we aren't rolling back save the load info from our first
2916 * import attempt so that we can restore it after attempting
2919 loadinfo
= spa
->spa_load_info
;
2920 spa
->spa_load_info
= fnvlist_alloc();
2923 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2924 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2925 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2926 TXG_INITIAL
: safe_rewind_txg
;
2929 * Continue as long as we're finding errors, we're still within
2930 * the acceptable rewind range, and we're still finding uberblocks
2932 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2933 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2934 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2935 spa
->spa_extreme_rewind
= B_TRUE
;
2936 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2939 spa
->spa_extreme_rewind
= B_FALSE
;
2940 spa
->spa_load_max_txg
= UINT64_MAX
;
2942 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2943 spa_config_set(spa
, config
);
2945 if (state
== SPA_LOAD_RECOVER
) {
2946 ASSERT3P(loadinfo
, ==, NULL
);
2947 return (rewind_error
);
2949 /* Store the rewind info as part of the initial load info */
2950 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
2951 spa
->spa_load_info
);
2953 /* Restore the initial load info */
2954 fnvlist_free(spa
->spa_load_info
);
2955 spa
->spa_load_info
= loadinfo
;
2957 return (load_error
);
2964 * The import case is identical to an open except that the configuration is sent
2965 * down from userland, instead of grabbed from the configuration cache. For the
2966 * case of an open, the pool configuration will exist in the
2967 * POOL_STATE_UNINITIALIZED state.
2969 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2970 * the same time open the pool, without having to keep around the spa_t in some
2974 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2978 spa_load_state_t state
= SPA_LOAD_OPEN
;
2980 int locked
= B_FALSE
;
2981 int firstopen
= B_FALSE
;
2986 * As disgusting as this is, we need to support recursive calls to this
2987 * function because dsl_dir_open() is called during spa_load(), and ends
2988 * up calling spa_open() again. The real fix is to figure out how to
2989 * avoid dsl_dir_open() calling this in the first place.
2991 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2992 mutex_enter(&spa_namespace_lock
);
2996 if ((spa
= spa_lookup(pool
)) == NULL
) {
2998 mutex_exit(&spa_namespace_lock
);
2999 return (SET_ERROR(ENOENT
));
3002 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3003 zpool_rewind_policy_t policy
;
3007 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3009 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3010 state
= SPA_LOAD_RECOVER
;
3012 spa_activate(spa
, spa_mode_global
);
3014 if (state
!= SPA_LOAD_RECOVER
)
3015 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3017 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3018 policy
.zrp_request
);
3020 if (error
== EBADF
) {
3022 * If vdev_validate() returns failure (indicated by
3023 * EBADF), it indicates that one of the vdevs indicates
3024 * that the pool has been exported or destroyed. If
3025 * this is the case, the config cache is out of sync and
3026 * we should remove the pool from the namespace.
3029 spa_deactivate(spa
);
3030 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3033 mutex_exit(&spa_namespace_lock
);
3034 return (SET_ERROR(ENOENT
));
3039 * We can't open the pool, but we still have useful
3040 * information: the state of each vdev after the
3041 * attempted vdev_open(). Return this to the user.
3043 if (config
!= NULL
&& spa
->spa_config
) {
3044 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3046 VERIFY(nvlist_add_nvlist(*config
,
3047 ZPOOL_CONFIG_LOAD_INFO
,
3048 spa
->spa_load_info
) == 0);
3051 spa_deactivate(spa
);
3052 spa
->spa_last_open_failed
= error
;
3054 mutex_exit(&spa_namespace_lock
);
3060 spa_open_ref(spa
, tag
);
3063 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3066 * If we've recovered the pool, pass back any information we
3067 * gathered while doing the load.
3069 if (state
== SPA_LOAD_RECOVER
) {
3070 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3071 spa
->spa_load_info
) == 0);
3075 spa
->spa_last_open_failed
= 0;
3076 spa
->spa_last_ubsync_txg
= 0;
3077 spa
->spa_load_txg
= 0;
3078 mutex_exit(&spa_namespace_lock
);
3083 zvol_create_minors(spa
->spa_name
);
3092 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3095 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3099 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3101 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3105 * Lookup the given spa_t, incrementing the inject count in the process,
3106 * preventing it from being exported or destroyed.
3109 spa_inject_addref(char *name
)
3113 mutex_enter(&spa_namespace_lock
);
3114 if ((spa
= spa_lookup(name
)) == NULL
) {
3115 mutex_exit(&spa_namespace_lock
);
3118 spa
->spa_inject_ref
++;
3119 mutex_exit(&spa_namespace_lock
);
3125 spa_inject_delref(spa_t
*spa
)
3127 mutex_enter(&spa_namespace_lock
);
3128 spa
->spa_inject_ref
--;
3129 mutex_exit(&spa_namespace_lock
);
3133 * Add spares device information to the nvlist.
3136 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3146 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3148 if (spa
->spa_spares
.sav_count
== 0)
3151 VERIFY(nvlist_lookup_nvlist(config
,
3152 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3153 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3154 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3156 VERIFY(nvlist_add_nvlist_array(nvroot
,
3157 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3158 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3159 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3162 * Go through and find any spares which have since been
3163 * repurposed as an active spare. If this is the case, update
3164 * their status appropriately.
3166 for (i
= 0; i
< nspares
; i
++) {
3167 VERIFY(nvlist_lookup_uint64(spares
[i
],
3168 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3169 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3171 VERIFY(nvlist_lookup_uint64_array(
3172 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3173 (uint64_t **)&vs
, &vsc
) == 0);
3174 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3175 vs
->vs_aux
= VDEV_AUX_SPARED
;
3182 * Add l2cache device information to the nvlist, including vdev stats.
3185 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3188 uint_t i
, j
, nl2cache
;
3195 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3197 if (spa
->spa_l2cache
.sav_count
== 0)
3200 VERIFY(nvlist_lookup_nvlist(config
,
3201 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3202 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3203 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3204 if (nl2cache
!= 0) {
3205 VERIFY(nvlist_add_nvlist_array(nvroot
,
3206 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3207 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3208 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3211 * Update level 2 cache device stats.
3214 for (i
= 0; i
< nl2cache
; i
++) {
3215 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3216 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3219 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3221 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3222 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3228 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3229 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3231 vdev_get_stats(vd
, vs
);
3237 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3242 if (spa
->spa_feat_for_read_obj
!= 0) {
3243 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3244 spa
->spa_feat_for_read_obj
);
3245 zap_cursor_retrieve(&zc
, &za
) == 0;
3246 zap_cursor_advance(&zc
)) {
3247 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3248 za
.za_num_integers
== 1);
3249 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3250 za
.za_first_integer
));
3252 zap_cursor_fini(&zc
);
3255 if (spa
->spa_feat_for_write_obj
!= 0) {
3256 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3257 spa
->spa_feat_for_write_obj
);
3258 zap_cursor_retrieve(&zc
, &za
) == 0;
3259 zap_cursor_advance(&zc
)) {
3260 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3261 za
.za_num_integers
== 1);
3262 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3263 za
.za_first_integer
));
3265 zap_cursor_fini(&zc
);
3270 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3274 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3275 zfeature_info_t feature
= spa_feature_table
[i
];
3278 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3281 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3286 * Store a list of pool features and their reference counts in the
3289 * The first time this is called on a spa, allocate a new nvlist, fetch
3290 * the pool features and reference counts from disk, then save the list
3291 * in the spa. In subsequent calls on the same spa use the saved nvlist
3292 * and refresh its values from the cached reference counts. This
3293 * ensures we don't block here on I/O on a suspended pool so 'zpool
3294 * clear' can resume the pool.
3297 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3301 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3303 mutex_enter(&spa
->spa_feat_stats_lock
);
3304 features
= spa
->spa_feat_stats
;
3306 if (features
!= NULL
) {
3307 spa_feature_stats_from_cache(spa
, features
);
3309 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3310 spa
->spa_feat_stats
= features
;
3311 spa_feature_stats_from_disk(spa
, features
);
3314 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3317 mutex_exit(&spa
->spa_feat_stats_lock
);
3321 spa_get_stats(const char *name
, nvlist_t
**config
,
3322 char *altroot
, size_t buflen
)
3328 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3332 * This still leaves a window of inconsistency where the spares
3333 * or l2cache devices could change and the config would be
3334 * self-inconsistent.
3336 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3338 if (*config
!= NULL
) {
3339 uint64_t loadtimes
[2];
3341 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3342 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3343 VERIFY(nvlist_add_uint64_array(*config
,
3344 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3346 VERIFY(nvlist_add_uint64(*config
,
3347 ZPOOL_CONFIG_ERRCOUNT
,
3348 spa_get_errlog_size(spa
)) == 0);
3350 if (spa_suspended(spa
))
3351 VERIFY(nvlist_add_uint64(*config
,
3352 ZPOOL_CONFIG_SUSPENDED
,
3353 spa
->spa_failmode
) == 0);
3355 spa_add_spares(spa
, *config
);
3356 spa_add_l2cache(spa
, *config
);
3357 spa_add_feature_stats(spa
, *config
);
3362 * We want to get the alternate root even for faulted pools, so we cheat
3363 * and call spa_lookup() directly.
3367 mutex_enter(&spa_namespace_lock
);
3368 spa
= spa_lookup(name
);
3370 spa_altroot(spa
, altroot
, buflen
);
3374 mutex_exit(&spa_namespace_lock
);
3376 spa_altroot(spa
, altroot
, buflen
);
3381 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3382 spa_close(spa
, FTAG
);
3389 * Validate that the auxiliary device array is well formed. We must have an
3390 * array of nvlists, each which describes a valid leaf vdev. If this is an
3391 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3392 * specified, as long as they are well-formed.
3395 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3396 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3397 vdev_labeltype_t label
)
3404 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3407 * It's acceptable to have no devs specified.
3409 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3413 return (SET_ERROR(EINVAL
));
3416 * Make sure the pool is formatted with a version that supports this
3419 if (spa_version(spa
) < version
)
3420 return (SET_ERROR(ENOTSUP
));
3423 * Set the pending device list so we correctly handle device in-use
3426 sav
->sav_pending
= dev
;
3427 sav
->sav_npending
= ndev
;
3429 for (i
= 0; i
< ndev
; i
++) {
3430 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3434 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3436 error
= SET_ERROR(EINVAL
);
3441 * The L2ARC currently only supports disk devices in
3442 * kernel context. For user-level testing, we allow it.
3445 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3446 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3447 error
= SET_ERROR(ENOTBLK
);
3454 if ((error
= vdev_open(vd
)) == 0 &&
3455 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3456 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3457 vd
->vdev_guid
) == 0);
3463 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3470 sav
->sav_pending
= NULL
;
3471 sav
->sav_npending
= 0;
3476 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3480 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3482 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3483 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3484 VDEV_LABEL_SPARE
)) != 0) {
3488 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3489 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3490 VDEV_LABEL_L2CACHE
));
3494 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3499 if (sav
->sav_config
!= NULL
) {
3505 * Generate new dev list by concatentating with the
3508 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3509 &olddevs
, &oldndevs
) == 0);
3511 newdevs
= kmem_alloc(sizeof (void *) *
3512 (ndevs
+ oldndevs
), KM_SLEEP
);
3513 for (i
= 0; i
< oldndevs
; i
++)
3514 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3516 for (i
= 0; i
< ndevs
; i
++)
3517 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3520 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3521 DATA_TYPE_NVLIST_ARRAY
) == 0);
3523 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3524 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3525 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3526 nvlist_free(newdevs
[i
]);
3527 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3530 * Generate a new dev list.
3532 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3534 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3540 * Stop and drop level 2 ARC devices
3543 spa_l2cache_drop(spa_t
*spa
)
3547 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3549 for (i
= 0; i
< sav
->sav_count
; i
++) {
3552 vd
= sav
->sav_vdevs
[i
];
3555 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3556 pool
!= 0ULL && l2arc_vdev_present(vd
))
3557 l2arc_remove_vdev(vd
);
3565 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3569 char *altroot
= NULL
;
3574 uint64_t txg
= TXG_INITIAL
;
3575 nvlist_t
**spares
, **l2cache
;
3576 uint_t nspares
, nl2cache
;
3577 uint64_t version
, obj
;
3578 boolean_t has_features
;
3584 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
3585 poolname
= (char *)pool
;
3588 * If this pool already exists, return failure.
3590 mutex_enter(&spa_namespace_lock
);
3591 if (spa_lookup(poolname
) != NULL
) {
3592 mutex_exit(&spa_namespace_lock
);
3593 return (SET_ERROR(EEXIST
));
3597 * Allocate a new spa_t structure.
3599 nvl
= fnvlist_alloc();
3600 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
3601 (void) nvlist_lookup_string(props
,
3602 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3603 spa
= spa_add(poolname
, nvl
, altroot
);
3605 spa_activate(spa
, spa_mode_global
);
3607 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3608 spa_deactivate(spa
);
3610 mutex_exit(&spa_namespace_lock
);
3615 * Temporary pool names should never be written to disk.
3617 if (poolname
!= pool
)
3618 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
3620 has_features
= B_FALSE
;
3621 for (elem
= nvlist_next_nvpair(props
, NULL
);
3622 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3623 if (zpool_prop_feature(nvpair_name(elem
)))
3624 has_features
= B_TRUE
;
3627 if (has_features
|| nvlist_lookup_uint64(props
,
3628 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3629 version
= SPA_VERSION
;
3631 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3633 spa
->spa_first_txg
= txg
;
3634 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3635 spa
->spa_uberblock
.ub_version
= version
;
3636 spa
->spa_ubsync
= spa
->spa_uberblock
;
3639 * Create "The Godfather" zio to hold all async IOs
3641 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3643 for (i
= 0; i
< max_ncpus
; i
++) {
3644 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3645 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3646 ZIO_FLAG_GODFATHER
);
3650 * Create the root vdev.
3652 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3654 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3656 ASSERT(error
!= 0 || rvd
!= NULL
);
3657 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3659 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3660 error
= SET_ERROR(EINVAL
);
3663 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3664 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3665 VDEV_ALLOC_ADD
)) == 0) {
3666 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
3667 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3668 vdev_expand(rvd
->vdev_child
[c
], txg
);
3672 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3676 spa_deactivate(spa
);
3678 mutex_exit(&spa_namespace_lock
);
3683 * Get the list of spares, if specified.
3685 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3686 &spares
, &nspares
) == 0) {
3687 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3689 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3690 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3691 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3692 spa_load_spares(spa
);
3693 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3694 spa
->spa_spares
.sav_sync
= B_TRUE
;
3698 * Get the list of level 2 cache devices, if specified.
3700 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3701 &l2cache
, &nl2cache
) == 0) {
3702 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3703 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3704 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3705 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3706 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3707 spa_load_l2cache(spa
);
3708 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3709 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3712 spa
->spa_is_initializing
= B_TRUE
;
3713 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3714 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3715 spa
->spa_is_initializing
= B_FALSE
;
3718 * Create DDTs (dedup tables).
3722 spa_update_dspace(spa
);
3724 tx
= dmu_tx_create_assigned(dp
, txg
);
3727 * Create the pool config object.
3729 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3730 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3731 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3733 if (zap_add(spa
->spa_meta_objset
,
3734 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3735 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3736 cmn_err(CE_PANIC
, "failed to add pool config");
3739 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3740 spa_feature_create_zap_objects(spa
, tx
);
3742 if (zap_add(spa
->spa_meta_objset
,
3743 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3744 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3745 cmn_err(CE_PANIC
, "failed to add pool version");
3748 /* Newly created pools with the right version are always deflated. */
3749 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3750 spa
->spa_deflate
= TRUE
;
3751 if (zap_add(spa
->spa_meta_objset
,
3752 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3753 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3754 cmn_err(CE_PANIC
, "failed to add deflate");
3759 * Create the deferred-free bpobj. Turn off compression
3760 * because sync-to-convergence takes longer if the blocksize
3763 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3764 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3765 ZIO_COMPRESS_OFF
, tx
);
3766 if (zap_add(spa
->spa_meta_objset
,
3767 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3768 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3769 cmn_err(CE_PANIC
, "failed to add bpobj");
3771 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3772 spa
->spa_meta_objset
, obj
));
3775 * Create the pool's history object.
3777 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3778 spa_history_create_obj(spa
, tx
);
3781 * Set pool properties.
3783 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3784 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3785 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3786 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3788 if (props
!= NULL
) {
3789 spa_configfile_set(spa
, props
, B_FALSE
);
3790 spa_sync_props(props
, tx
);
3795 spa
->spa_sync_on
= B_TRUE
;
3796 txg_sync_start(spa
->spa_dsl_pool
);
3799 * We explicitly wait for the first transaction to complete so that our
3800 * bean counters are appropriately updated.
3802 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3804 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3806 spa_history_log_version(spa
, "create");
3809 * Don't count references from objsets that are already closed
3810 * and are making their way through the eviction process.
3812 spa_evicting_os_wait(spa
);
3813 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3815 mutex_exit(&spa_namespace_lock
);
3822 * Get the root pool information from the root disk, then import the root pool
3823 * during the system boot up time.
3825 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3828 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3831 nvlist_t
*nvtop
, *nvroot
;
3834 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3838 * Add this top-level vdev to the child array.
3840 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3842 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3844 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3847 * Put this pool's top-level vdevs into a root vdev.
3849 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3850 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3851 VDEV_TYPE_ROOT
) == 0);
3852 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3853 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3854 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3858 * Replace the existing vdev_tree with the new root vdev in
3859 * this pool's configuration (remove the old, add the new).
3861 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3862 nvlist_free(nvroot
);
3867 * Walk the vdev tree and see if we can find a device with "better"
3868 * configuration. A configuration is "better" if the label on that
3869 * device has a more recent txg.
3872 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3876 for (c
= 0; c
< vd
->vdev_children
; c
++)
3877 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3879 if (vd
->vdev_ops
->vdev_op_leaf
) {
3883 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3887 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3891 * Do we have a better boot device?
3893 if (label_txg
> *txg
) {
3902 * Import a root pool.
3904 * For x86. devpath_list will consist of devid and/or physpath name of
3905 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3906 * The GRUB "findroot" command will return the vdev we should boot.
3908 * For Sparc, devpath_list consists the physpath name of the booting device
3909 * no matter the rootpool is a single device pool or a mirrored pool.
3911 * "/pci@1f,0/ide@d/disk@0,0:a"
3914 spa_import_rootpool(char *devpath
, char *devid
)
3917 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3918 nvlist_t
*config
, *nvtop
;
3924 * Read the label from the boot device and generate a configuration.
3926 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3927 #if defined(_OBP) && defined(_KERNEL)
3928 if (config
== NULL
) {
3929 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3931 get_iscsi_bootpath_phy(devpath
);
3932 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3936 if (config
== NULL
) {
3937 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
3939 return (SET_ERROR(EIO
));
3942 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3944 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3946 mutex_enter(&spa_namespace_lock
);
3947 if ((spa
= spa_lookup(pname
)) != NULL
) {
3949 * Remove the existing root pool from the namespace so that we
3950 * can replace it with the correct config we just read in.
3955 spa
= spa_add(pname
, config
, NULL
);
3956 spa
->spa_is_root
= B_TRUE
;
3957 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3960 * Build up a vdev tree based on the boot device's label config.
3962 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3964 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3965 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3966 VDEV_ALLOC_ROOTPOOL
);
3967 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3969 mutex_exit(&spa_namespace_lock
);
3970 nvlist_free(config
);
3971 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3977 * Get the boot vdev.
3979 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3980 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3981 (u_longlong_t
)guid
);
3982 error
= SET_ERROR(ENOENT
);
3987 * Determine if there is a better boot device.
3990 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3992 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3993 "try booting from '%s'", avd
->vdev_path
);
3994 error
= SET_ERROR(EINVAL
);
3999 * If the boot device is part of a spare vdev then ensure that
4000 * we're booting off the active spare.
4002 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4003 !bvd
->vdev_isspare
) {
4004 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
4005 "try booting from '%s'",
4007 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
4008 error
= SET_ERROR(EINVAL
);
4014 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4016 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4017 mutex_exit(&spa_namespace_lock
);
4019 nvlist_free(config
);
4026 * Import a non-root pool into the system.
4029 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4032 char *altroot
= NULL
;
4033 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4034 zpool_rewind_policy_t policy
;
4035 uint64_t mode
= spa_mode_global
;
4036 uint64_t readonly
= B_FALSE
;
4039 nvlist_t
**spares
, **l2cache
;
4040 uint_t nspares
, nl2cache
;
4043 * If a pool with this name exists, return failure.
4045 mutex_enter(&spa_namespace_lock
);
4046 if (spa_lookup(pool
) != NULL
) {
4047 mutex_exit(&spa_namespace_lock
);
4048 return (SET_ERROR(EEXIST
));
4052 * Create and initialize the spa structure.
4054 (void) nvlist_lookup_string(props
,
4055 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4056 (void) nvlist_lookup_uint64(props
,
4057 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4060 spa
= spa_add(pool
, config
, altroot
);
4061 spa
->spa_import_flags
= flags
;
4064 * Verbatim import - Take a pool and insert it into the namespace
4065 * as if it had been loaded at boot.
4067 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4069 spa_configfile_set(spa
, props
, B_FALSE
);
4071 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4073 mutex_exit(&spa_namespace_lock
);
4077 spa_activate(spa
, mode
);
4080 * Don't start async tasks until we know everything is healthy.
4082 spa_async_suspend(spa
);
4084 zpool_get_rewind_policy(config
, &policy
);
4085 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4086 state
= SPA_LOAD_RECOVER
;
4089 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4090 * because the user-supplied config is actually the one to trust when
4093 if (state
!= SPA_LOAD_RECOVER
)
4094 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4096 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4097 policy
.zrp_request
);
4100 * Propagate anything learned while loading the pool and pass it
4101 * back to caller (i.e. rewind info, missing devices, etc).
4103 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4104 spa
->spa_load_info
) == 0);
4106 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4108 * Toss any existing sparelist, as it doesn't have any validity
4109 * anymore, and conflicts with spa_has_spare().
4111 if (spa
->spa_spares
.sav_config
) {
4112 nvlist_free(spa
->spa_spares
.sav_config
);
4113 spa
->spa_spares
.sav_config
= NULL
;
4114 spa_load_spares(spa
);
4116 if (spa
->spa_l2cache
.sav_config
) {
4117 nvlist_free(spa
->spa_l2cache
.sav_config
);
4118 spa
->spa_l2cache
.sav_config
= NULL
;
4119 spa_load_l2cache(spa
);
4122 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4125 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4128 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4129 VDEV_ALLOC_L2CACHE
);
4130 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4133 spa_configfile_set(spa
, props
, B_FALSE
);
4135 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4136 (error
= spa_prop_set(spa
, props
)))) {
4138 spa_deactivate(spa
);
4140 mutex_exit(&spa_namespace_lock
);
4144 spa_async_resume(spa
);
4147 * Override any spares and level 2 cache devices as specified by
4148 * the user, as these may have correct device names/devids, etc.
4150 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4151 &spares
, &nspares
) == 0) {
4152 if (spa
->spa_spares
.sav_config
)
4153 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4154 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4156 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4157 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4158 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4159 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4160 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4161 spa_load_spares(spa
);
4162 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4163 spa
->spa_spares
.sav_sync
= B_TRUE
;
4165 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4166 &l2cache
, &nl2cache
) == 0) {
4167 if (spa
->spa_l2cache
.sav_config
)
4168 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4169 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4171 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4172 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4173 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4174 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4175 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4176 spa_load_l2cache(spa
);
4177 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4178 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4182 * Check for any removed devices.
4184 if (spa
->spa_autoreplace
) {
4185 spa_aux_check_removed(&spa
->spa_spares
);
4186 spa_aux_check_removed(&spa
->spa_l2cache
);
4189 if (spa_writeable(spa
)) {
4191 * Update the config cache to include the newly-imported pool.
4193 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4197 * It's possible that the pool was expanded while it was exported.
4198 * We kick off an async task to handle this for us.
4200 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4202 mutex_exit(&spa_namespace_lock
);
4203 spa_history_log_version(spa
, "import");
4206 zvol_create_minors(pool
);
4213 spa_tryimport(nvlist_t
*tryconfig
)
4215 nvlist_t
*config
= NULL
;
4221 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4224 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4228 * Create and initialize the spa structure.
4230 mutex_enter(&spa_namespace_lock
);
4231 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4232 spa_activate(spa
, FREAD
);
4235 * Pass off the heavy lifting to spa_load().
4236 * Pass TRUE for mosconfig because the user-supplied config
4237 * is actually the one to trust when doing an import.
4239 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4242 * If 'tryconfig' was at least parsable, return the current config.
4244 if (spa
->spa_root_vdev
!= NULL
) {
4245 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4246 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4248 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4250 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4251 spa
->spa_uberblock
.ub_timestamp
) == 0);
4252 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4253 spa
->spa_load_info
) == 0);
4254 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4255 spa
->spa_errata
) == 0);
4258 * If the bootfs property exists on this pool then we
4259 * copy it out so that external consumers can tell which
4260 * pools are bootable.
4262 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4263 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4266 * We have to play games with the name since the
4267 * pool was opened as TRYIMPORT_NAME.
4269 if (dsl_dsobj_to_dsname(spa_name(spa
),
4270 spa
->spa_bootfs
, tmpname
) == 0) {
4274 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4276 cp
= strchr(tmpname
, '/');
4278 (void) strlcpy(dsname
, tmpname
,
4281 (void) snprintf(dsname
, MAXPATHLEN
,
4282 "%s/%s", poolname
, ++cp
);
4284 VERIFY(nvlist_add_string(config
,
4285 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4286 kmem_free(dsname
, MAXPATHLEN
);
4288 kmem_free(tmpname
, MAXPATHLEN
);
4292 * Add the list of hot spares and level 2 cache devices.
4294 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4295 spa_add_spares(spa
, config
);
4296 spa_add_l2cache(spa
, config
);
4297 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4301 spa_deactivate(spa
);
4303 mutex_exit(&spa_namespace_lock
);
4309 * Pool export/destroy
4311 * The act of destroying or exporting a pool is very simple. We make sure there
4312 * is no more pending I/O and any references to the pool are gone. Then, we
4313 * update the pool state and sync all the labels to disk, removing the
4314 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4315 * we don't sync the labels or remove the configuration cache.
4318 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4319 boolean_t force
, boolean_t hardforce
)
4326 if (!(spa_mode_global
& FWRITE
))
4327 return (SET_ERROR(EROFS
));
4329 mutex_enter(&spa_namespace_lock
);
4330 if ((spa
= spa_lookup(pool
)) == NULL
) {
4331 mutex_exit(&spa_namespace_lock
);
4332 return (SET_ERROR(ENOENT
));
4336 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4337 * reacquire the namespace lock, and see if we can export.
4339 spa_open_ref(spa
, FTAG
);
4340 mutex_exit(&spa_namespace_lock
);
4341 spa_async_suspend(spa
);
4342 mutex_enter(&spa_namespace_lock
);
4343 spa_close(spa
, FTAG
);
4345 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4348 * The pool will be in core if it's openable, in which case we can
4349 * modify its state. Objsets may be open only because they're dirty,
4350 * so we have to force it to sync before checking spa_refcnt.
4352 if (spa
->spa_sync_on
) {
4353 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4354 spa_evicting_os_wait(spa
);
4358 * A pool cannot be exported or destroyed if there are active
4359 * references. If we are resetting a pool, allow references by
4360 * fault injection handlers.
4362 if (!spa_refcount_zero(spa
) ||
4363 (spa
->spa_inject_ref
!= 0 &&
4364 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4365 spa_async_resume(spa
);
4366 mutex_exit(&spa_namespace_lock
);
4367 return (SET_ERROR(EBUSY
));
4370 if (spa
->spa_sync_on
) {
4372 * A pool cannot be exported if it has an active shared spare.
4373 * This is to prevent other pools stealing the active spare
4374 * from an exported pool. At user's own will, such pool can
4375 * be forcedly exported.
4377 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4378 spa_has_active_shared_spare(spa
)) {
4379 spa_async_resume(spa
);
4380 mutex_exit(&spa_namespace_lock
);
4381 return (SET_ERROR(EXDEV
));
4385 * We want this to be reflected on every label,
4386 * so mark them all dirty. spa_unload() will do the
4387 * final sync that pushes these changes out.
4389 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4390 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4391 spa
->spa_state
= new_state
;
4392 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4394 vdev_config_dirty(spa
->spa_root_vdev
);
4395 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4400 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_DESTROY
);
4402 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4404 spa_deactivate(spa
);
4407 if (oldconfig
&& spa
->spa_config
)
4408 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4410 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4412 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4415 mutex_exit(&spa_namespace_lock
);
4421 * Destroy a storage pool.
4424 spa_destroy(char *pool
)
4426 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4431 * Export a storage pool.
4434 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4435 boolean_t hardforce
)
4437 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4442 * Similar to spa_export(), this unloads the spa_t without actually removing it
4443 * from the namespace in any way.
4446 spa_reset(char *pool
)
4448 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4453 * ==========================================================================
4454 * Device manipulation
4455 * ==========================================================================
4459 * Add a device to a storage pool.
4462 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4466 vdev_t
*rvd
= spa
->spa_root_vdev
;
4468 nvlist_t
**spares
, **l2cache
;
4469 uint_t nspares
, nl2cache
;
4472 ASSERT(spa_writeable(spa
));
4474 txg
= spa_vdev_enter(spa
);
4476 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4477 VDEV_ALLOC_ADD
)) != 0)
4478 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4480 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4482 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4486 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4490 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4491 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4493 if (vd
->vdev_children
!= 0 &&
4494 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4495 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4498 * We must validate the spares and l2cache devices after checking the
4499 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4501 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4502 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4505 * Transfer each new top-level vdev from vd to rvd.
4507 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4510 * Set the vdev id to the first hole, if one exists.
4512 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4513 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4514 vdev_free(rvd
->vdev_child
[id
]);
4518 tvd
= vd
->vdev_child
[c
];
4519 vdev_remove_child(vd
, tvd
);
4521 vdev_add_child(rvd
, tvd
);
4522 vdev_config_dirty(tvd
);
4526 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4527 ZPOOL_CONFIG_SPARES
);
4528 spa_load_spares(spa
);
4529 spa
->spa_spares
.sav_sync
= B_TRUE
;
4532 if (nl2cache
!= 0) {
4533 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4534 ZPOOL_CONFIG_L2CACHE
);
4535 spa_load_l2cache(spa
);
4536 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4540 * We have to be careful when adding new vdevs to an existing pool.
4541 * If other threads start allocating from these vdevs before we
4542 * sync the config cache, and we lose power, then upon reboot we may
4543 * fail to open the pool because there are DVAs that the config cache
4544 * can't translate. Therefore, we first add the vdevs without
4545 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4546 * and then let spa_config_update() initialize the new metaslabs.
4548 * spa_load() checks for added-but-not-initialized vdevs, so that
4549 * if we lose power at any point in this sequence, the remaining
4550 * steps will be completed the next time we load the pool.
4552 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4554 mutex_enter(&spa_namespace_lock
);
4555 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4556 mutex_exit(&spa_namespace_lock
);
4562 * Attach a device to a mirror. The arguments are the path to any device
4563 * in the mirror, and the nvroot for the new device. If the path specifies
4564 * a device that is not mirrored, we automatically insert the mirror vdev.
4566 * If 'replacing' is specified, the new device is intended to replace the
4567 * existing device; in this case the two devices are made into their own
4568 * mirror using the 'replacing' vdev, which is functionally identical to
4569 * the mirror vdev (it actually reuses all the same ops) but has a few
4570 * extra rules: you can't attach to it after it's been created, and upon
4571 * completion of resilvering, the first disk (the one being replaced)
4572 * is automatically detached.
4575 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4577 uint64_t txg
, dtl_max_txg
;
4578 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4580 char *oldvdpath
, *newvdpath
;
4583 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4585 ASSERT(spa_writeable(spa
));
4587 txg
= spa_vdev_enter(spa
);
4589 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4592 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4594 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4595 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4597 pvd
= oldvd
->vdev_parent
;
4599 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4600 VDEV_ALLOC_ATTACH
)) != 0)
4601 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4603 if (newrootvd
->vdev_children
!= 1)
4604 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4606 newvd
= newrootvd
->vdev_child
[0];
4608 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4609 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4611 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4612 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4615 * Spares can't replace logs
4617 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4618 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4622 * For attach, the only allowable parent is a mirror or the root
4625 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4626 pvd
->vdev_ops
!= &vdev_root_ops
)
4627 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4629 pvops
= &vdev_mirror_ops
;
4632 * Active hot spares can only be replaced by inactive hot
4635 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4636 oldvd
->vdev_isspare
&&
4637 !spa_has_spare(spa
, newvd
->vdev_guid
))
4638 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4641 * If the source is a hot spare, and the parent isn't already a
4642 * spare, then we want to create a new hot spare. Otherwise, we
4643 * want to create a replacing vdev. The user is not allowed to
4644 * attach to a spared vdev child unless the 'isspare' state is
4645 * the same (spare replaces spare, non-spare replaces
4648 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4649 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4650 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4651 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4652 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4653 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4656 if (newvd
->vdev_isspare
)
4657 pvops
= &vdev_spare_ops
;
4659 pvops
= &vdev_replacing_ops
;
4663 * Make sure the new device is big enough.
4665 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4666 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4669 * The new device cannot have a higher alignment requirement
4670 * than the top-level vdev.
4672 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4673 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4676 * If this is an in-place replacement, update oldvd's path and devid
4677 * to make it distinguishable from newvd, and unopenable from now on.
4679 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4680 spa_strfree(oldvd
->vdev_path
);
4681 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4683 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4684 newvd
->vdev_path
, "old");
4685 if (oldvd
->vdev_devid
!= NULL
) {
4686 spa_strfree(oldvd
->vdev_devid
);
4687 oldvd
->vdev_devid
= NULL
;
4691 /* mark the device being resilvered */
4692 newvd
->vdev_resilver_txg
= txg
;
4695 * If the parent is not a mirror, or if we're replacing, insert the new
4696 * mirror/replacing/spare vdev above oldvd.
4698 if (pvd
->vdev_ops
!= pvops
)
4699 pvd
= vdev_add_parent(oldvd
, pvops
);
4701 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4702 ASSERT(pvd
->vdev_ops
== pvops
);
4703 ASSERT(oldvd
->vdev_parent
== pvd
);
4706 * Extract the new device from its root and add it to pvd.
4708 vdev_remove_child(newrootvd
, newvd
);
4709 newvd
->vdev_id
= pvd
->vdev_children
;
4710 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4711 vdev_add_child(pvd
, newvd
);
4713 tvd
= newvd
->vdev_top
;
4714 ASSERT(pvd
->vdev_top
== tvd
);
4715 ASSERT(tvd
->vdev_parent
== rvd
);
4717 vdev_config_dirty(tvd
);
4720 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4721 * for any dmu_sync-ed blocks. It will propagate upward when
4722 * spa_vdev_exit() calls vdev_dtl_reassess().
4724 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4726 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4727 dtl_max_txg
- TXG_INITIAL
);
4729 if (newvd
->vdev_isspare
) {
4730 spa_spare_activate(newvd
);
4731 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_DEVICE_SPARE
);
4734 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4735 newvdpath
= spa_strdup(newvd
->vdev_path
);
4736 newvd_isspare
= newvd
->vdev_isspare
;
4739 * Mark newvd's DTL dirty in this txg.
4741 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4744 * Schedule the resilver to restart in the future. We do this to
4745 * ensure that dmu_sync-ed blocks have been stitched into the
4746 * respective datasets.
4748 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4753 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4755 spa_history_log_internal(spa
, "vdev attach", NULL
,
4756 "%s vdev=%s %s vdev=%s",
4757 replacing
&& newvd_isspare
? "spare in" :
4758 replacing
? "replace" : "attach", newvdpath
,
4759 replacing
? "for" : "to", oldvdpath
);
4761 spa_strfree(oldvdpath
);
4762 spa_strfree(newvdpath
);
4764 if (spa
->spa_bootfs
)
4765 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH
);
4771 * Detach a device from a mirror or replacing vdev.
4773 * If 'replace_done' is specified, only detach if the parent
4774 * is a replacing vdev.
4777 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4781 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4782 boolean_t unspare
= B_FALSE
;
4783 uint64_t unspare_guid
= 0;
4786 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4787 ASSERT(spa_writeable(spa
));
4789 txg
= spa_vdev_enter(spa
);
4791 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4794 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4796 if (!vd
->vdev_ops
->vdev_op_leaf
)
4797 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4799 pvd
= vd
->vdev_parent
;
4802 * If the parent/child relationship is not as expected, don't do it.
4803 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4804 * vdev that's replacing B with C. The user's intent in replacing
4805 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4806 * the replace by detaching C, the expected behavior is to end up
4807 * M(A,B). But suppose that right after deciding to detach C,
4808 * the replacement of B completes. We would have M(A,C), and then
4809 * ask to detach C, which would leave us with just A -- not what
4810 * the user wanted. To prevent this, we make sure that the
4811 * parent/child relationship hasn't changed -- in this example,
4812 * that C's parent is still the replacing vdev R.
4814 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4815 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4818 * Only 'replacing' or 'spare' vdevs can be replaced.
4820 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4821 pvd
->vdev_ops
!= &vdev_spare_ops
)
4822 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4824 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4825 spa_version(spa
) >= SPA_VERSION_SPARES
);
4828 * Only mirror, replacing, and spare vdevs support detach.
4830 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4831 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4832 pvd
->vdev_ops
!= &vdev_spare_ops
)
4833 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4836 * If this device has the only valid copy of some data,
4837 * we cannot safely detach it.
4839 if (vdev_dtl_required(vd
))
4840 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4842 ASSERT(pvd
->vdev_children
>= 2);
4845 * If we are detaching the second disk from a replacing vdev, then
4846 * check to see if we changed the original vdev's path to have "/old"
4847 * at the end in spa_vdev_attach(). If so, undo that change now.
4849 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4850 vd
->vdev_path
!= NULL
) {
4851 size_t len
= strlen(vd
->vdev_path
);
4853 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4854 cvd
= pvd
->vdev_child
[c
];
4856 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4859 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4860 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4861 spa_strfree(cvd
->vdev_path
);
4862 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4869 * If we are detaching the original disk from a spare, then it implies
4870 * that the spare should become a real disk, and be removed from the
4871 * active spare list for the pool.
4873 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4875 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4879 * Erase the disk labels so the disk can be used for other things.
4880 * This must be done after all other error cases are handled,
4881 * but before we disembowel vd (so we can still do I/O to it).
4882 * But if we can't do it, don't treat the error as fatal --
4883 * it may be that the unwritability of the disk is the reason
4884 * it's being detached!
4886 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4889 * Remove vd from its parent and compact the parent's children.
4891 vdev_remove_child(pvd
, vd
);
4892 vdev_compact_children(pvd
);
4895 * Remember one of the remaining children so we can get tvd below.
4897 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4900 * If we need to remove the remaining child from the list of hot spares,
4901 * do it now, marking the vdev as no longer a spare in the process.
4902 * We must do this before vdev_remove_parent(), because that can
4903 * change the GUID if it creates a new toplevel GUID. For a similar
4904 * reason, we must remove the spare now, in the same txg as the detach;
4905 * otherwise someone could attach a new sibling, change the GUID, and
4906 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4909 ASSERT(cvd
->vdev_isspare
);
4910 spa_spare_remove(cvd
);
4911 unspare_guid
= cvd
->vdev_guid
;
4912 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4913 cvd
->vdev_unspare
= B_TRUE
;
4917 * If the parent mirror/replacing vdev only has one child,
4918 * the parent is no longer needed. Remove it from the tree.
4920 if (pvd
->vdev_children
== 1) {
4921 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4922 cvd
->vdev_unspare
= B_FALSE
;
4923 vdev_remove_parent(cvd
);
4928 * We don't set tvd until now because the parent we just removed
4929 * may have been the previous top-level vdev.
4931 tvd
= cvd
->vdev_top
;
4932 ASSERT(tvd
->vdev_parent
== rvd
);
4935 * Reevaluate the parent vdev state.
4937 vdev_propagate_state(cvd
);
4940 * If the 'autoexpand' property is set on the pool then automatically
4941 * try to expand the size of the pool. For example if the device we
4942 * just detached was smaller than the others, it may be possible to
4943 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4944 * first so that we can obtain the updated sizes of the leaf vdevs.
4946 if (spa
->spa_autoexpand
) {
4948 vdev_expand(tvd
, txg
);
4951 vdev_config_dirty(tvd
);
4954 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4955 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4956 * But first make sure we're not on any *other* txg's DTL list, to
4957 * prevent vd from being accessed after it's freed.
4959 vdpath
= spa_strdup(vd
->vdev_path
);
4960 for (t
= 0; t
< TXG_SIZE
; t
++)
4961 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4962 vd
->vdev_detached
= B_TRUE
;
4963 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4965 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_REMOVE
);
4967 /* hang on to the spa before we release the lock */
4968 spa_open_ref(spa
, FTAG
);
4970 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4972 spa_history_log_internal(spa
, "detach", NULL
,
4974 spa_strfree(vdpath
);
4977 * If this was the removal of the original device in a hot spare vdev,
4978 * then we want to go through and remove the device from the hot spare
4979 * list of every other pool.
4982 spa_t
*altspa
= NULL
;
4984 mutex_enter(&spa_namespace_lock
);
4985 while ((altspa
= spa_next(altspa
)) != NULL
) {
4986 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4990 spa_open_ref(altspa
, FTAG
);
4991 mutex_exit(&spa_namespace_lock
);
4992 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4993 mutex_enter(&spa_namespace_lock
);
4994 spa_close(altspa
, FTAG
);
4996 mutex_exit(&spa_namespace_lock
);
4998 /* search the rest of the vdevs for spares to remove */
4999 spa_vdev_resilver_done(spa
);
5002 /* all done with the spa; OK to release */
5003 mutex_enter(&spa_namespace_lock
);
5004 spa_close(spa
, FTAG
);
5005 mutex_exit(&spa_namespace_lock
);
5011 * Split a set of devices from their mirrors, and create a new pool from them.
5014 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5015 nvlist_t
*props
, boolean_t exp
)
5018 uint64_t txg
, *glist
;
5020 uint_t c
, children
, lastlog
;
5021 nvlist_t
**child
, *nvl
, *tmp
;
5023 char *altroot
= NULL
;
5024 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5025 boolean_t activate_slog
;
5027 ASSERT(spa_writeable(spa
));
5029 txg
= spa_vdev_enter(spa
);
5031 /* clear the log and flush everything up to now */
5032 activate_slog
= spa_passivate_log(spa
);
5033 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5034 error
= spa_offline_log(spa
);
5035 txg
= spa_vdev_config_enter(spa
);
5038 spa_activate_log(spa
);
5041 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5043 /* check new spa name before going any further */
5044 if (spa_lookup(newname
) != NULL
)
5045 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5048 * scan through all the children to ensure they're all mirrors
5050 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5051 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5053 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5055 /* first, check to ensure we've got the right child count */
5056 rvd
= spa
->spa_root_vdev
;
5058 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5059 vdev_t
*vd
= rvd
->vdev_child
[c
];
5061 /* don't count the holes & logs as children */
5062 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5070 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5071 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5073 /* next, ensure no spare or cache devices are part of the split */
5074 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5075 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5076 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5078 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5079 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5081 /* then, loop over each vdev and validate it */
5082 for (c
= 0; c
< children
; c
++) {
5083 uint64_t is_hole
= 0;
5085 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5089 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5090 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5093 error
= SET_ERROR(EINVAL
);
5098 /* which disk is going to be split? */
5099 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5101 error
= SET_ERROR(EINVAL
);
5105 /* look it up in the spa */
5106 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5107 if (vml
[c
] == NULL
) {
5108 error
= SET_ERROR(ENODEV
);
5112 /* make sure there's nothing stopping the split */
5113 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5114 vml
[c
]->vdev_islog
||
5115 vml
[c
]->vdev_ishole
||
5116 vml
[c
]->vdev_isspare
||
5117 vml
[c
]->vdev_isl2cache
||
5118 !vdev_writeable(vml
[c
]) ||
5119 vml
[c
]->vdev_children
!= 0 ||
5120 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5121 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5122 error
= SET_ERROR(EINVAL
);
5126 if (vdev_dtl_required(vml
[c
])) {
5127 error
= SET_ERROR(EBUSY
);
5131 /* we need certain info from the top level */
5132 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5133 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5134 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5135 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5136 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5137 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5138 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5139 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5143 kmem_free(vml
, children
* sizeof (vdev_t
*));
5144 kmem_free(glist
, children
* sizeof (uint64_t));
5145 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5148 /* stop writers from using the disks */
5149 for (c
= 0; c
< children
; c
++) {
5151 vml
[c
]->vdev_offline
= B_TRUE
;
5153 vdev_reopen(spa
->spa_root_vdev
);
5156 * Temporarily record the splitting vdevs in the spa config. This
5157 * will disappear once the config is regenerated.
5159 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5160 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5161 glist
, children
) == 0);
5162 kmem_free(glist
, children
* sizeof (uint64_t));
5164 mutex_enter(&spa
->spa_props_lock
);
5165 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5167 mutex_exit(&spa
->spa_props_lock
);
5168 spa
->spa_config_splitting
= nvl
;
5169 vdev_config_dirty(spa
->spa_root_vdev
);
5171 /* configure and create the new pool */
5172 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5173 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5174 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5175 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5176 spa_version(spa
)) == 0);
5177 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5178 spa
->spa_config_txg
) == 0);
5179 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5180 spa_generate_guid(NULL
)) == 0);
5181 (void) nvlist_lookup_string(props
,
5182 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5184 /* add the new pool to the namespace */
5185 newspa
= spa_add(newname
, config
, altroot
);
5186 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5187 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5189 /* release the spa config lock, retaining the namespace lock */
5190 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5192 if (zio_injection_enabled
)
5193 zio_handle_panic_injection(spa
, FTAG
, 1);
5195 spa_activate(newspa
, spa_mode_global
);
5196 spa_async_suspend(newspa
);
5198 /* create the new pool from the disks of the original pool */
5199 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5203 /* if that worked, generate a real config for the new pool */
5204 if (newspa
->spa_root_vdev
!= NULL
) {
5205 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5206 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5207 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5208 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5209 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5214 if (props
!= NULL
) {
5215 spa_configfile_set(newspa
, props
, B_FALSE
);
5216 error
= spa_prop_set(newspa
, props
);
5221 /* flush everything */
5222 txg
= spa_vdev_config_enter(newspa
);
5223 vdev_config_dirty(newspa
->spa_root_vdev
);
5224 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5226 if (zio_injection_enabled
)
5227 zio_handle_panic_injection(spa
, FTAG
, 2);
5229 spa_async_resume(newspa
);
5231 /* finally, update the original pool's config */
5232 txg
= spa_vdev_config_enter(spa
);
5233 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5234 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5237 for (c
= 0; c
< children
; c
++) {
5238 if (vml
[c
] != NULL
) {
5241 spa_history_log_internal(spa
, "detach", tx
,
5242 "vdev=%s", vml
[c
]->vdev_path
);
5246 vdev_config_dirty(spa
->spa_root_vdev
);
5247 spa
->spa_config_splitting
= NULL
;
5251 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5253 if (zio_injection_enabled
)
5254 zio_handle_panic_injection(spa
, FTAG
, 3);
5256 /* split is complete; log a history record */
5257 spa_history_log_internal(newspa
, "split", NULL
,
5258 "from pool %s", spa_name(spa
));
5260 kmem_free(vml
, children
* sizeof (vdev_t
*));
5262 /* if we're not going to mount the filesystems in userland, export */
5264 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5271 spa_deactivate(newspa
);
5274 txg
= spa_vdev_config_enter(spa
);
5276 /* re-online all offlined disks */
5277 for (c
= 0; c
< children
; c
++) {
5279 vml
[c
]->vdev_offline
= B_FALSE
;
5281 vdev_reopen(spa
->spa_root_vdev
);
5283 nvlist_free(spa
->spa_config_splitting
);
5284 spa
->spa_config_splitting
= NULL
;
5285 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5287 kmem_free(vml
, children
* sizeof (vdev_t
*));
5292 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5296 for (i
= 0; i
< count
; i
++) {
5299 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5302 if (guid
== target_guid
)
5310 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5311 nvlist_t
*dev_to_remove
)
5313 nvlist_t
**newdev
= NULL
;
5317 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5319 for (i
= 0, j
= 0; i
< count
; i
++) {
5320 if (dev
[i
] == dev_to_remove
)
5322 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5325 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5326 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5328 for (i
= 0; i
< count
- 1; i
++)
5329 nvlist_free(newdev
[i
]);
5332 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5336 * Evacuate the device.
5339 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5344 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5345 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5346 ASSERT(vd
== vd
->vdev_top
);
5349 * Evacuate the device. We don't hold the config lock as writer
5350 * since we need to do I/O but we do keep the
5351 * spa_namespace_lock held. Once this completes the device
5352 * should no longer have any blocks allocated on it.
5354 if (vd
->vdev_islog
) {
5355 if (vd
->vdev_stat
.vs_alloc
!= 0)
5356 error
= spa_offline_log(spa
);
5358 error
= SET_ERROR(ENOTSUP
);
5365 * The evacuation succeeded. Remove any remaining MOS metadata
5366 * associated with this vdev, and wait for these changes to sync.
5368 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5369 txg
= spa_vdev_config_enter(spa
);
5370 vd
->vdev_removing
= B_TRUE
;
5371 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5372 vdev_config_dirty(vd
);
5373 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5379 * Complete the removal by cleaning up the namespace.
5382 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5384 vdev_t
*rvd
= spa
->spa_root_vdev
;
5385 uint64_t id
= vd
->vdev_id
;
5386 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5388 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5389 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5390 ASSERT(vd
== vd
->vdev_top
);
5393 * Only remove any devices which are empty.
5395 if (vd
->vdev_stat
.vs_alloc
!= 0)
5398 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5400 if (list_link_active(&vd
->vdev_state_dirty_node
))
5401 vdev_state_clean(vd
);
5402 if (list_link_active(&vd
->vdev_config_dirty_node
))
5403 vdev_config_clean(vd
);
5408 vdev_compact_children(rvd
);
5410 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5411 vdev_add_child(rvd
, vd
);
5413 vdev_config_dirty(rvd
);
5416 * Reassess the health of our root vdev.
5422 * Remove a device from the pool -
5424 * Removing a device from the vdev namespace requires several steps
5425 * and can take a significant amount of time. As a result we use
5426 * the spa_vdev_config_[enter/exit] functions which allow us to
5427 * grab and release the spa_config_lock while still holding the namespace
5428 * lock. During each step the configuration is synced out.
5430 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5434 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5437 metaslab_group_t
*mg
;
5438 nvlist_t
**spares
, **l2cache
, *nv
;
5440 uint_t nspares
, nl2cache
;
5442 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5444 ASSERT(spa_writeable(spa
));
5447 txg
= spa_vdev_enter(spa
);
5449 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5451 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5452 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5453 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5454 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5456 * Only remove the hot spare if it's not currently in use
5459 if (vd
== NULL
|| unspare
) {
5460 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5461 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5462 spa_load_spares(spa
);
5463 spa
->spa_spares
.sav_sync
= B_TRUE
;
5465 error
= SET_ERROR(EBUSY
);
5467 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5468 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5469 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5470 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5472 * Cache devices can always be removed.
5474 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5475 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5476 spa_load_l2cache(spa
);
5477 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5478 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5480 ASSERT(vd
== vd
->vdev_top
);
5485 * Stop allocating from this vdev.
5487 metaslab_group_passivate(mg
);
5490 * Wait for the youngest allocations and frees to sync,
5491 * and then wait for the deferral of those frees to finish.
5493 spa_vdev_config_exit(spa
, NULL
,
5494 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5497 * Attempt to evacuate the vdev.
5499 error
= spa_vdev_remove_evacuate(spa
, vd
);
5501 txg
= spa_vdev_config_enter(spa
);
5504 * If we couldn't evacuate the vdev, unwind.
5507 metaslab_group_activate(mg
);
5508 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5512 * Clean up the vdev namespace.
5514 spa_vdev_remove_from_namespace(spa
, vd
);
5516 } else if (vd
!= NULL
) {
5518 * Normal vdevs cannot be removed (yet).
5520 error
= SET_ERROR(ENOTSUP
);
5523 * There is no vdev of any kind with the specified guid.
5525 error
= SET_ERROR(ENOENT
);
5529 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5535 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5536 * currently spared, so we can detach it.
5539 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5541 vdev_t
*newvd
, *oldvd
;
5544 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5545 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5551 * Check for a completed replacement. We always consider the first
5552 * vdev in the list to be the oldest vdev, and the last one to be
5553 * the newest (see spa_vdev_attach() for how that works). In
5554 * the case where the newest vdev is faulted, we will not automatically
5555 * remove it after a resilver completes. This is OK as it will require
5556 * user intervention to determine which disk the admin wishes to keep.
5558 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5559 ASSERT(vd
->vdev_children
> 1);
5561 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5562 oldvd
= vd
->vdev_child
[0];
5564 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5565 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5566 !vdev_dtl_required(oldvd
))
5571 * Check for a completed resilver with the 'unspare' flag set.
5573 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5574 vdev_t
*first
= vd
->vdev_child
[0];
5575 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5577 if (last
->vdev_unspare
) {
5580 } else if (first
->vdev_unspare
) {
5587 if (oldvd
!= NULL
&&
5588 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5589 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5590 !vdev_dtl_required(oldvd
))
5594 * If there are more than two spares attached to a disk,
5595 * and those spares are not required, then we want to
5596 * attempt to free them up now so that they can be used
5597 * by other pools. Once we're back down to a single
5598 * disk+spare, we stop removing them.
5600 if (vd
->vdev_children
> 2) {
5601 newvd
= vd
->vdev_child
[1];
5603 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5604 vdev_dtl_empty(last
, DTL_MISSING
) &&
5605 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5606 !vdev_dtl_required(newvd
))
5615 spa_vdev_resilver_done(spa_t
*spa
)
5617 vdev_t
*vd
, *pvd
, *ppvd
;
5618 uint64_t guid
, sguid
, pguid
, ppguid
;
5620 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5622 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5623 pvd
= vd
->vdev_parent
;
5624 ppvd
= pvd
->vdev_parent
;
5625 guid
= vd
->vdev_guid
;
5626 pguid
= pvd
->vdev_guid
;
5627 ppguid
= ppvd
->vdev_guid
;
5630 * If we have just finished replacing a hot spared device, then
5631 * we need to detach the parent's first child (the original hot
5634 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5635 ppvd
->vdev_children
== 2) {
5636 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5637 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5639 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5641 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5642 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5644 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5646 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5649 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5653 * Update the stored path or FRU for this vdev.
5656 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5660 boolean_t sync
= B_FALSE
;
5662 ASSERT(spa_writeable(spa
));
5664 spa_vdev_state_enter(spa
, SCL_ALL
);
5666 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5667 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5669 if (!vd
->vdev_ops
->vdev_op_leaf
)
5670 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5673 if (strcmp(value
, vd
->vdev_path
) != 0) {
5674 spa_strfree(vd
->vdev_path
);
5675 vd
->vdev_path
= spa_strdup(value
);
5679 if (vd
->vdev_fru
== NULL
) {
5680 vd
->vdev_fru
= spa_strdup(value
);
5682 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5683 spa_strfree(vd
->vdev_fru
);
5684 vd
->vdev_fru
= spa_strdup(value
);
5689 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5693 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5695 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5699 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5701 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5705 * ==========================================================================
5707 * ==========================================================================
5711 spa_scan_stop(spa_t
*spa
)
5713 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5714 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5715 return (SET_ERROR(EBUSY
));
5716 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5720 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5722 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5724 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5725 return (SET_ERROR(ENOTSUP
));
5728 * If a resilver was requested, but there is no DTL on a
5729 * writeable leaf device, we have nothing to do.
5731 if (func
== POOL_SCAN_RESILVER
&&
5732 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5733 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5737 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5741 * ==========================================================================
5742 * SPA async task processing
5743 * ==========================================================================
5747 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5751 if (vd
->vdev_remove_wanted
) {
5752 vd
->vdev_remove_wanted
= B_FALSE
;
5753 vd
->vdev_delayed_close
= B_FALSE
;
5754 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5757 * We want to clear the stats, but we don't want to do a full
5758 * vdev_clear() as that will cause us to throw away
5759 * degraded/faulted state as well as attempt to reopen the
5760 * device, all of which is a waste.
5762 vd
->vdev_stat
.vs_read_errors
= 0;
5763 vd
->vdev_stat
.vs_write_errors
= 0;
5764 vd
->vdev_stat
.vs_checksum_errors
= 0;
5766 vdev_state_dirty(vd
->vdev_top
);
5769 for (c
= 0; c
< vd
->vdev_children
; c
++)
5770 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5774 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5778 if (vd
->vdev_probe_wanted
) {
5779 vd
->vdev_probe_wanted
= B_FALSE
;
5780 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5783 for (c
= 0; c
< vd
->vdev_children
; c
++)
5784 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5788 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5792 if (!spa
->spa_autoexpand
)
5795 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5796 vdev_t
*cvd
= vd
->vdev_child
[c
];
5797 spa_async_autoexpand(spa
, cvd
);
5800 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5803 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND
);
5807 spa_async_thread(spa_t
*spa
)
5811 ASSERT(spa
->spa_sync_on
);
5813 mutex_enter(&spa
->spa_async_lock
);
5814 tasks
= spa
->spa_async_tasks
;
5815 spa
->spa_async_tasks
= 0;
5816 mutex_exit(&spa
->spa_async_lock
);
5819 * See if the config needs to be updated.
5821 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5822 uint64_t old_space
, new_space
;
5824 mutex_enter(&spa_namespace_lock
);
5825 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5826 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5827 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5828 mutex_exit(&spa_namespace_lock
);
5831 * If the pool grew as a result of the config update,
5832 * then log an internal history event.
5834 if (new_space
!= old_space
) {
5835 spa_history_log_internal(spa
, "vdev online", NULL
,
5836 "pool '%s' size: %llu(+%llu)",
5837 spa_name(spa
), new_space
, new_space
- old_space
);
5842 * See if any devices need to be marked REMOVED.
5844 if (tasks
& SPA_ASYNC_REMOVE
) {
5845 spa_vdev_state_enter(spa
, SCL_NONE
);
5846 spa_async_remove(spa
, spa
->spa_root_vdev
);
5847 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5848 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5849 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5850 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5851 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5854 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5855 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5856 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5857 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5861 * See if any devices need to be probed.
5863 if (tasks
& SPA_ASYNC_PROBE
) {
5864 spa_vdev_state_enter(spa
, SCL_NONE
);
5865 spa_async_probe(spa
, spa
->spa_root_vdev
);
5866 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5870 * If any devices are done replacing, detach them.
5872 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5873 spa_vdev_resilver_done(spa
);
5876 * Kick off a resilver.
5878 if (tasks
& SPA_ASYNC_RESILVER
)
5879 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5882 * Let the world know that we're done.
5884 mutex_enter(&spa
->spa_async_lock
);
5885 spa
->spa_async_thread
= NULL
;
5886 cv_broadcast(&spa
->spa_async_cv
);
5887 mutex_exit(&spa
->spa_async_lock
);
5892 spa_async_suspend(spa_t
*spa
)
5894 mutex_enter(&spa
->spa_async_lock
);
5895 spa
->spa_async_suspended
++;
5896 while (spa
->spa_async_thread
!= NULL
)
5897 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5898 mutex_exit(&spa
->spa_async_lock
);
5902 spa_async_resume(spa_t
*spa
)
5904 mutex_enter(&spa
->spa_async_lock
);
5905 ASSERT(spa
->spa_async_suspended
!= 0);
5906 spa
->spa_async_suspended
--;
5907 mutex_exit(&spa
->spa_async_lock
);
5911 spa_async_dispatch(spa_t
*spa
)
5913 mutex_enter(&spa
->spa_async_lock
);
5914 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5915 spa
->spa_async_thread
== NULL
&&
5916 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5917 spa
->spa_async_thread
= thread_create(NULL
, 0,
5918 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5919 mutex_exit(&spa
->spa_async_lock
);
5923 spa_async_request(spa_t
*spa
, int task
)
5925 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5926 mutex_enter(&spa
->spa_async_lock
);
5927 spa
->spa_async_tasks
|= task
;
5928 mutex_exit(&spa
->spa_async_lock
);
5932 * ==========================================================================
5933 * SPA syncing routines
5934 * ==========================================================================
5938 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5941 bpobj_enqueue(bpo
, bp
, tx
);
5946 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5950 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5956 * Note: this simple function is not inlined to make it easier to dtrace the
5957 * amount of time spent syncing frees.
5960 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
5962 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5963 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
5964 VERIFY(zio_wait(zio
) == 0);
5968 * Note: this simple function is not inlined to make it easier to dtrace the
5969 * amount of time spent syncing deferred frees.
5972 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
5974 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5975 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
5976 spa_free_sync_cb
, zio
, tx
), ==, 0);
5977 VERIFY0(zio_wait(zio
));
5981 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5983 char *packed
= NULL
;
5988 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5991 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5992 * information. This avoids the dmu_buf_will_dirty() path and
5993 * saves us a pre-read to get data we don't actually care about.
5995 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
5996 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
5998 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6000 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6002 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6004 vmem_free(packed
, bufsize
);
6006 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6007 dmu_buf_will_dirty(db
, tx
);
6008 *(uint64_t *)db
->db_data
= nvsize
;
6009 dmu_buf_rele(db
, FTAG
);
6013 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6014 const char *config
, const char *entry
)
6024 * Update the MOS nvlist describing the list of available devices.
6025 * spa_validate_aux() will have already made sure this nvlist is
6026 * valid and the vdevs are labeled appropriately.
6028 if (sav
->sav_object
== 0) {
6029 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6030 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6031 sizeof (uint64_t), tx
);
6032 VERIFY(zap_update(spa
->spa_meta_objset
,
6033 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6034 &sav
->sav_object
, tx
) == 0);
6037 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6038 if (sav
->sav_count
== 0) {
6039 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6041 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6042 for (i
= 0; i
< sav
->sav_count
; i
++)
6043 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6044 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6045 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6046 sav
->sav_count
) == 0);
6047 for (i
= 0; i
< sav
->sav_count
; i
++)
6048 nvlist_free(list
[i
]);
6049 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6052 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6053 nvlist_free(nvroot
);
6055 sav
->sav_sync
= B_FALSE
;
6059 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6063 if (list_is_empty(&spa
->spa_config_dirty_list
))
6066 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6068 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6069 dmu_tx_get_txg(tx
), B_FALSE
);
6072 * If we're upgrading the spa version then make sure that
6073 * the config object gets updated with the correct version.
6075 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6076 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6077 spa
->spa_uberblock
.ub_version
);
6079 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6081 if (spa
->spa_config_syncing
)
6082 nvlist_free(spa
->spa_config_syncing
);
6083 spa
->spa_config_syncing
= config
;
6085 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6089 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6091 uint64_t *versionp
= arg
;
6092 uint64_t version
= *versionp
;
6093 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6096 * Setting the version is special cased when first creating the pool.
6098 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6100 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6101 ASSERT(version
>= spa_version(spa
));
6103 spa
->spa_uberblock
.ub_version
= version
;
6104 vdev_config_dirty(spa
->spa_root_vdev
);
6105 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6109 * Set zpool properties.
6112 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6114 nvlist_t
*nvp
= arg
;
6115 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6116 objset_t
*mos
= spa
->spa_meta_objset
;
6117 nvpair_t
*elem
= NULL
;
6119 mutex_enter(&spa
->spa_props_lock
);
6121 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6123 char *strval
, *fname
;
6125 const char *propname
;
6126 zprop_type_t proptype
;
6129 prop
= zpool_name_to_prop(nvpair_name(elem
));
6130 switch ((int)prop
) {
6133 * We checked this earlier in spa_prop_validate().
6135 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6137 fname
= strchr(nvpair_name(elem
), '@') + 1;
6138 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6140 spa_feature_enable(spa
, fid
, tx
);
6141 spa_history_log_internal(spa
, "set", tx
,
6142 "%s=enabled", nvpair_name(elem
));
6145 case ZPOOL_PROP_VERSION
:
6146 intval
= fnvpair_value_uint64(elem
);
6148 * The version is synced seperatly before other
6149 * properties and should be correct by now.
6151 ASSERT3U(spa_version(spa
), >=, intval
);
6154 case ZPOOL_PROP_ALTROOT
:
6156 * 'altroot' is a non-persistent property. It should
6157 * have been set temporarily at creation or import time.
6159 ASSERT(spa
->spa_root
!= NULL
);
6162 case ZPOOL_PROP_READONLY
:
6163 case ZPOOL_PROP_CACHEFILE
:
6165 * 'readonly' and 'cachefile' are also non-persisitent
6169 case ZPOOL_PROP_COMMENT
:
6170 strval
= fnvpair_value_string(elem
);
6171 if (spa
->spa_comment
!= NULL
)
6172 spa_strfree(spa
->spa_comment
);
6173 spa
->spa_comment
= spa_strdup(strval
);
6175 * We need to dirty the configuration on all the vdevs
6176 * so that their labels get updated. It's unnecessary
6177 * to do this for pool creation since the vdev's
6178 * configuratoin has already been dirtied.
6180 if (tx
->tx_txg
!= TXG_INITIAL
)
6181 vdev_config_dirty(spa
->spa_root_vdev
);
6182 spa_history_log_internal(spa
, "set", tx
,
6183 "%s=%s", nvpair_name(elem
), strval
);
6187 * Set pool property values in the poolprops mos object.
6189 if (spa
->spa_pool_props_object
== 0) {
6190 spa
->spa_pool_props_object
=
6191 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6192 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6196 /* normalize the property name */
6197 propname
= zpool_prop_to_name(prop
);
6198 proptype
= zpool_prop_get_type(prop
);
6200 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6201 ASSERT(proptype
== PROP_TYPE_STRING
);
6202 strval
= fnvpair_value_string(elem
);
6203 VERIFY0(zap_update(mos
,
6204 spa
->spa_pool_props_object
, propname
,
6205 1, strlen(strval
) + 1, strval
, tx
));
6206 spa_history_log_internal(spa
, "set", tx
,
6207 "%s=%s", nvpair_name(elem
), strval
);
6208 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6209 intval
= fnvpair_value_uint64(elem
);
6211 if (proptype
== PROP_TYPE_INDEX
) {
6213 VERIFY0(zpool_prop_index_to_string(
6214 prop
, intval
, &unused
));
6216 VERIFY0(zap_update(mos
,
6217 spa
->spa_pool_props_object
, propname
,
6218 8, 1, &intval
, tx
));
6219 spa_history_log_internal(spa
, "set", tx
,
6220 "%s=%lld", nvpair_name(elem
), intval
);
6222 ASSERT(0); /* not allowed */
6226 case ZPOOL_PROP_DELEGATION
:
6227 spa
->spa_delegation
= intval
;
6229 case ZPOOL_PROP_BOOTFS
:
6230 spa
->spa_bootfs
= intval
;
6232 case ZPOOL_PROP_FAILUREMODE
:
6233 spa
->spa_failmode
= intval
;
6235 case ZPOOL_PROP_AUTOEXPAND
:
6236 spa
->spa_autoexpand
= intval
;
6237 if (tx
->tx_txg
!= TXG_INITIAL
)
6238 spa_async_request(spa
,
6239 SPA_ASYNC_AUTOEXPAND
);
6241 case ZPOOL_PROP_DEDUPDITTO
:
6242 spa
->spa_dedup_ditto
= intval
;
6251 mutex_exit(&spa
->spa_props_lock
);
6255 * Perform one-time upgrade on-disk changes. spa_version() does not
6256 * reflect the new version this txg, so there must be no changes this
6257 * txg to anything that the upgrade code depends on after it executes.
6258 * Therefore this must be called after dsl_pool_sync() does the sync
6262 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6264 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6266 ASSERT(spa
->spa_sync_pass
== 1);
6268 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6270 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6271 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6272 dsl_pool_create_origin(dp
, tx
);
6274 /* Keeping the origin open increases spa_minref */
6275 spa
->spa_minref
+= 3;
6278 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6279 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6280 dsl_pool_upgrade_clones(dp
, tx
);
6283 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6284 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6285 dsl_pool_upgrade_dir_clones(dp
, tx
);
6287 /* Keeping the freedir open increases spa_minref */
6288 spa
->spa_minref
+= 3;
6291 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6292 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6293 spa_feature_create_zap_objects(spa
, tx
);
6297 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6298 * when possibility to use lz4 compression for metadata was added
6299 * Old pools that have this feature enabled must be upgraded to have
6300 * this feature active
6302 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6303 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6304 SPA_FEATURE_LZ4_COMPRESS
);
6305 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6306 SPA_FEATURE_LZ4_COMPRESS
);
6308 if (lz4_en
&& !lz4_ac
)
6309 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6311 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6315 * Sync the specified transaction group. New blocks may be dirtied as
6316 * part of the process, so we iterate until it converges.
6319 spa_sync(spa_t
*spa
, uint64_t txg
)
6321 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6322 objset_t
*mos
= spa
->spa_meta_objset
;
6323 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6324 vdev_t
*rvd
= spa
->spa_root_vdev
;
6330 VERIFY(spa_writeable(spa
));
6333 * Lock out configuration changes.
6335 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6337 spa
->spa_syncing_txg
= txg
;
6338 spa
->spa_sync_pass
= 0;
6341 * If there are any pending vdev state changes, convert them
6342 * into config changes that go out with this transaction group.
6344 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6345 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6347 * We need the write lock here because, for aux vdevs,
6348 * calling vdev_config_dirty() modifies sav_config.
6349 * This is ugly and will become unnecessary when we
6350 * eliminate the aux vdev wart by integrating all vdevs
6351 * into the root vdev tree.
6353 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6354 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6355 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6356 vdev_state_clean(vd
);
6357 vdev_config_dirty(vd
);
6359 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6360 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6362 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6364 tx
= dmu_tx_create_assigned(dp
, txg
);
6366 spa
->spa_sync_starttime
= gethrtime();
6367 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6368 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
6369 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6370 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6373 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6374 * set spa_deflate if we have no raid-z vdevs.
6376 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6377 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6380 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6381 vd
= rvd
->vdev_child
[i
];
6382 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6385 if (i
== rvd
->vdev_children
) {
6386 spa
->spa_deflate
= TRUE
;
6387 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6388 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6389 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6394 * If anything has changed in this txg, or if someone is waiting
6395 * for this txg to sync (eg, spa_vdev_remove()), push the
6396 * deferred frees from the previous txg. If not, leave them
6397 * alone so that we don't generate work on an otherwise idle
6400 if (!txg_list_empty(&dp
->dp_dirty_datasets
, txg
) ||
6401 !txg_list_empty(&dp
->dp_dirty_dirs
, txg
) ||
6402 !txg_list_empty(&dp
->dp_sync_tasks
, txg
) ||
6403 ((dsl_scan_active(dp
->dp_scan
) ||
6404 txg_sync_waiting(dp
)) && !spa_shutting_down(spa
))) {
6405 spa_sync_deferred_frees(spa
, tx
);
6409 * Iterate to convergence.
6412 int pass
= ++spa
->spa_sync_pass
;
6414 spa_sync_config_object(spa
, tx
);
6415 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6416 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6417 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6418 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6419 spa_errlog_sync(spa
, txg
);
6420 dsl_pool_sync(dp
, txg
);
6422 if (pass
< zfs_sync_pass_deferred_free
) {
6423 spa_sync_frees(spa
, free_bpl
, tx
);
6425 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6426 &spa
->spa_deferred_bpobj
, tx
);
6430 dsl_scan_sync(dp
, tx
);
6432 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6436 spa_sync_upgrades(spa
, tx
);
6438 } while (dmu_objset_is_dirty(mos
, txg
));
6441 * Rewrite the vdev configuration (which includes the uberblock)
6442 * to commit the transaction group.
6444 * If there are no dirty vdevs, we sync the uberblock to a few
6445 * random top-level vdevs that are known to be visible in the
6446 * config cache (see spa_vdev_add() for a complete description).
6447 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6451 * We hold SCL_STATE to prevent vdev open/close/etc.
6452 * while we're attempting to write the vdev labels.
6454 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6456 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6457 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6459 int children
= rvd
->vdev_children
;
6460 int c0
= spa_get_random(children
);
6462 for (c
= 0; c
< children
; c
++) {
6463 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6464 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6466 svd
[svdcount
++] = vd
;
6467 if (svdcount
== SPA_DVAS_PER_BP
)
6470 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
6472 error
= vdev_config_sync(svd
, svdcount
, txg
,
6475 error
= vdev_config_sync(rvd
->vdev_child
,
6476 rvd
->vdev_children
, txg
, B_FALSE
);
6478 error
= vdev_config_sync(rvd
->vdev_child
,
6479 rvd
->vdev_children
, txg
, B_TRUE
);
6483 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6485 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6489 zio_suspend(spa
, NULL
);
6490 zio_resume_wait(spa
);
6494 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6495 spa
->spa_deadman_tqid
= 0;
6498 * Clear the dirty config list.
6500 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6501 vdev_config_clean(vd
);
6504 * Now that the new config has synced transactionally,
6505 * let it become visible to the config cache.
6507 if (spa
->spa_config_syncing
!= NULL
) {
6508 spa_config_set(spa
, spa
->spa_config_syncing
);
6509 spa
->spa_config_txg
= txg
;
6510 spa
->spa_config_syncing
= NULL
;
6513 spa
->spa_ubsync
= spa
->spa_uberblock
;
6515 dsl_pool_sync_done(dp
, txg
);
6518 * Update usable space statistics.
6520 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6521 vdev_sync_done(vd
, txg
);
6523 spa_update_dspace(spa
);
6526 * It had better be the case that we didn't dirty anything
6527 * since vdev_config_sync().
6529 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6530 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6531 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6533 spa
->spa_sync_pass
= 0;
6535 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6537 spa_handle_ignored_writes(spa
);
6540 * If any async tasks have been requested, kick them off.
6542 spa_async_dispatch(spa
);
6546 * Sync all pools. We don't want to hold the namespace lock across these
6547 * operations, so we take a reference on the spa_t and drop the lock during the
6551 spa_sync_allpools(void)
6554 mutex_enter(&spa_namespace_lock
);
6555 while ((spa
= spa_next(spa
)) != NULL
) {
6556 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6557 !spa_writeable(spa
) || spa_suspended(spa
))
6559 spa_open_ref(spa
, FTAG
);
6560 mutex_exit(&spa_namespace_lock
);
6561 txg_wait_synced(spa_get_dsl(spa
), 0);
6562 mutex_enter(&spa_namespace_lock
);
6563 spa_close(spa
, FTAG
);
6565 mutex_exit(&spa_namespace_lock
);
6569 * ==========================================================================
6570 * Miscellaneous routines
6571 * ==========================================================================
6575 * Remove all pools in the system.
6583 * Remove all cached state. All pools should be closed now,
6584 * so every spa in the AVL tree should be unreferenced.
6586 mutex_enter(&spa_namespace_lock
);
6587 while ((spa
= spa_next(NULL
)) != NULL
) {
6589 * Stop async tasks. The async thread may need to detach
6590 * a device that's been replaced, which requires grabbing
6591 * spa_namespace_lock, so we must drop it here.
6593 spa_open_ref(spa
, FTAG
);
6594 mutex_exit(&spa_namespace_lock
);
6595 spa_async_suspend(spa
);
6596 mutex_enter(&spa_namespace_lock
);
6597 spa_close(spa
, FTAG
);
6599 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6601 spa_deactivate(spa
);
6605 mutex_exit(&spa_namespace_lock
);
6609 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6614 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6618 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6619 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6620 if (vd
->vdev_guid
== guid
)
6624 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6625 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6626 if (vd
->vdev_guid
== guid
)
6635 spa_upgrade(spa_t
*spa
, uint64_t version
)
6637 ASSERT(spa_writeable(spa
));
6639 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6642 * This should only be called for a non-faulted pool, and since a
6643 * future version would result in an unopenable pool, this shouldn't be
6646 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
6647 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
6649 spa
->spa_uberblock
.ub_version
= version
;
6650 vdev_config_dirty(spa
->spa_root_vdev
);
6652 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6654 txg_wait_synced(spa_get_dsl(spa
), 0);
6658 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6662 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6664 for (i
= 0; i
< sav
->sav_count
; i
++)
6665 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6668 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6669 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6670 &spareguid
) == 0 && spareguid
== guid
)
6678 * Check if a pool has an active shared spare device.
6679 * Note: reference count of an active spare is 2, as a spare and as a replace
6682 spa_has_active_shared_spare(spa_t
*spa
)
6686 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6688 for (i
= 0; i
< sav
->sav_count
; i
++) {
6689 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6690 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6699 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6700 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6701 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6702 * or zdb as real changes.
6705 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6708 zfs_ereport_post(name
, spa
, vd
, NULL
, 0, 0);
6712 #if defined(_KERNEL) && defined(HAVE_SPL)
6713 /* state manipulation functions */
6714 EXPORT_SYMBOL(spa_open
);
6715 EXPORT_SYMBOL(spa_open_rewind
);
6716 EXPORT_SYMBOL(spa_get_stats
);
6717 EXPORT_SYMBOL(spa_create
);
6718 EXPORT_SYMBOL(spa_import_rootpool
);
6719 EXPORT_SYMBOL(spa_import
);
6720 EXPORT_SYMBOL(spa_tryimport
);
6721 EXPORT_SYMBOL(spa_destroy
);
6722 EXPORT_SYMBOL(spa_export
);
6723 EXPORT_SYMBOL(spa_reset
);
6724 EXPORT_SYMBOL(spa_async_request
);
6725 EXPORT_SYMBOL(spa_async_suspend
);
6726 EXPORT_SYMBOL(spa_async_resume
);
6727 EXPORT_SYMBOL(spa_inject_addref
);
6728 EXPORT_SYMBOL(spa_inject_delref
);
6729 EXPORT_SYMBOL(spa_scan_stat_init
);
6730 EXPORT_SYMBOL(spa_scan_get_stats
);
6732 /* device maniion */
6733 EXPORT_SYMBOL(spa_vdev_add
);
6734 EXPORT_SYMBOL(spa_vdev_attach
);
6735 EXPORT_SYMBOL(spa_vdev_detach
);
6736 EXPORT_SYMBOL(spa_vdev_remove
);
6737 EXPORT_SYMBOL(spa_vdev_setpath
);
6738 EXPORT_SYMBOL(spa_vdev_setfru
);
6739 EXPORT_SYMBOL(spa_vdev_split_mirror
);
6741 /* spare statech is global across all pools) */
6742 EXPORT_SYMBOL(spa_spare_add
);
6743 EXPORT_SYMBOL(spa_spare_remove
);
6744 EXPORT_SYMBOL(spa_spare_exists
);
6745 EXPORT_SYMBOL(spa_spare_activate
);
6747 /* L2ARC statech is global across all pools) */
6748 EXPORT_SYMBOL(spa_l2cache_add
);
6749 EXPORT_SYMBOL(spa_l2cache_remove
);
6750 EXPORT_SYMBOL(spa_l2cache_exists
);
6751 EXPORT_SYMBOL(spa_l2cache_activate
);
6752 EXPORT_SYMBOL(spa_l2cache_drop
);
6755 EXPORT_SYMBOL(spa_scan
);
6756 EXPORT_SYMBOL(spa_scan_stop
);
6759 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
6760 EXPORT_SYMBOL(spa_sync_allpools
);
6763 EXPORT_SYMBOL(spa_prop_set
);
6764 EXPORT_SYMBOL(spa_prop_get
);
6765 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
6767 /* asynchronous event notification */
6768 EXPORT_SYMBOL(spa_event_notify
);
6771 #if defined(_KERNEL) && defined(HAVE_SPL)
6772 module_param(spa_load_verify_maxinflight
, int, 0644);
6773 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
6774 "Max concurrent traversal I/Os while verifying pool during import -X");
6776 module_param(spa_load_verify_metadata
, int, 0644);
6777 MODULE_PARM_DESC(spa_load_verify_metadata
,
6778 "Set to traverse metadata on pool import");
6780 module_param(spa_load_verify_data
, int, 0644);
6781 MODULE_PARM_DESC(spa_load_verify_data
,
6782 "Set to traverse data on pool import");