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_P(12, 8), ZTI_NULL
}, /* READ */
131 { ZTI_BATCH
, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */
132 { ZTI_P(12, 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
];
847 uint_t i
, flags
= TASKQ_DYNAMIC
;
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
= MIN(zio_taskq_batch_pct
, 100);
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 less important
902 * priority than the other taskqs. Under Linux this
903 * means incrementing the priority value on platforms
904 * like illumos it should be decremented.
906 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
909 tq
= taskq_create_proc(name
, value
, pri
, 50,
910 INT_MAX
, spa
->spa_proc
, flags
);
913 tqs
->stqs_taskq
[i
] = tq
;
918 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
920 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
923 if (tqs
->stqs_taskq
== NULL
) {
924 ASSERT3U(tqs
->stqs_count
, ==, 0);
928 for (i
= 0; i
< tqs
->stqs_count
; i
++) {
929 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
930 taskq_destroy(tqs
->stqs_taskq
[i
]);
933 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
934 tqs
->stqs_taskq
= NULL
;
938 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
939 * Note that a type may have multiple discrete taskqs to avoid lock contention
940 * on the taskq itself. In that case we choose which taskq at random by using
941 * the low bits of gethrtime().
944 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
945 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
947 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
950 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
951 ASSERT3U(tqs
->stqs_count
, !=, 0);
953 if (tqs
->stqs_count
== 1) {
954 tq
= tqs
->stqs_taskq
[0];
956 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
959 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
963 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
966 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
967 task_func_t
*func
, void *arg
, uint_t flags
)
969 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
973 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
974 ASSERT3U(tqs
->stqs_count
, !=, 0);
976 if (tqs
->stqs_count
== 1) {
977 tq
= tqs
->stqs_taskq
[0];
979 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
982 id
= taskq_dispatch(tq
, func
, arg
, flags
);
984 taskq_wait_id(tq
, id
);
988 spa_create_zio_taskqs(spa_t
*spa
)
992 for (t
= 0; t
< ZIO_TYPES
; t
++) {
993 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
994 spa_taskqs_init(spa
, t
, q
);
999 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1001 spa_thread(void *arg
)
1003 callb_cpr_t cprinfo
;
1006 user_t
*pu
= PTOU(curproc
);
1008 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1011 ASSERT(curproc
!= &p0
);
1012 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1013 "zpool-%s", spa
->spa_name
);
1014 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1016 /* bind this thread to the requested psrset */
1017 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1019 mutex_enter(&cpu_lock
);
1020 mutex_enter(&pidlock
);
1021 mutex_enter(&curproc
->p_lock
);
1023 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1024 0, NULL
, NULL
) == 0) {
1025 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1028 "Couldn't bind process for zfs pool \"%s\" to "
1029 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1032 mutex_exit(&curproc
->p_lock
);
1033 mutex_exit(&pidlock
);
1034 mutex_exit(&cpu_lock
);
1038 if (zio_taskq_sysdc
) {
1039 sysdc_thread_enter(curthread
, 100, 0);
1042 spa
->spa_proc
= curproc
;
1043 spa
->spa_did
= curthread
->t_did
;
1045 spa_create_zio_taskqs(spa
);
1047 mutex_enter(&spa
->spa_proc_lock
);
1048 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1050 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1051 cv_broadcast(&spa
->spa_proc_cv
);
1053 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1054 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1055 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1056 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1058 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1059 spa
->spa_proc_state
= SPA_PROC_GONE
;
1060 spa
->spa_proc
= &p0
;
1061 cv_broadcast(&spa
->spa_proc_cv
);
1062 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1064 mutex_enter(&curproc
->p_lock
);
1070 * Activate an uninitialized pool.
1073 spa_activate(spa_t
*spa
, int mode
)
1075 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1077 spa
->spa_state
= POOL_STATE_ACTIVE
;
1078 spa
->spa_mode
= mode
;
1080 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1081 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1083 /* Try to create a covering process */
1084 mutex_enter(&spa
->spa_proc_lock
);
1085 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1086 ASSERT(spa
->spa_proc
== &p0
);
1089 #ifdef HAVE_SPA_THREAD
1090 /* Only create a process if we're going to be around a while. */
1091 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1092 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1094 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1095 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1096 cv_wait(&spa
->spa_proc_cv
,
1097 &spa
->spa_proc_lock
);
1099 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1100 ASSERT(spa
->spa_proc
!= &p0
);
1101 ASSERT(spa
->spa_did
!= 0);
1105 "Couldn't create process for zfs pool \"%s\"\n",
1110 #endif /* HAVE_SPA_THREAD */
1111 mutex_exit(&spa
->spa_proc_lock
);
1113 /* If we didn't create a process, we need to create our taskqs. */
1114 if (spa
->spa_proc
== &p0
) {
1115 spa_create_zio_taskqs(spa
);
1118 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1119 offsetof(vdev_t
, vdev_config_dirty_node
));
1120 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1121 offsetof(objset_t
, os_evicting_node
));
1122 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1123 offsetof(vdev_t
, vdev_state_dirty_node
));
1125 txg_list_create(&spa
->spa_vdev_txg_list
,
1126 offsetof(struct vdev
, vdev_txg_node
));
1128 avl_create(&spa
->spa_errlist_scrub
,
1129 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1130 offsetof(spa_error_entry_t
, se_avl
));
1131 avl_create(&spa
->spa_errlist_last
,
1132 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1133 offsetof(spa_error_entry_t
, se_avl
));
1137 * Opposite of spa_activate().
1140 spa_deactivate(spa_t
*spa
)
1144 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1145 ASSERT(spa
->spa_dsl_pool
== NULL
);
1146 ASSERT(spa
->spa_root_vdev
== NULL
);
1147 ASSERT(spa
->spa_async_zio_root
== NULL
);
1148 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1150 spa_evicting_os_wait(spa
);
1152 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1154 list_destroy(&spa
->spa_config_dirty_list
);
1155 list_destroy(&spa
->spa_evicting_os_list
);
1156 list_destroy(&spa
->spa_state_dirty_list
);
1158 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
1160 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1161 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1162 spa_taskqs_fini(spa
, t
, q
);
1166 metaslab_class_destroy(spa
->spa_normal_class
);
1167 spa
->spa_normal_class
= NULL
;
1169 metaslab_class_destroy(spa
->spa_log_class
);
1170 spa
->spa_log_class
= NULL
;
1173 * If this was part of an import or the open otherwise failed, we may
1174 * still have errors left in the queues. Empty them just in case.
1176 spa_errlog_drain(spa
);
1178 avl_destroy(&spa
->spa_errlist_scrub
);
1179 avl_destroy(&spa
->spa_errlist_last
);
1181 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1183 mutex_enter(&spa
->spa_proc_lock
);
1184 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1185 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1186 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1187 cv_broadcast(&spa
->spa_proc_cv
);
1188 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1189 ASSERT(spa
->spa_proc
!= &p0
);
1190 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1192 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1193 spa
->spa_proc_state
= SPA_PROC_NONE
;
1195 ASSERT(spa
->spa_proc
== &p0
);
1196 mutex_exit(&spa
->spa_proc_lock
);
1199 * We want to make sure spa_thread() has actually exited the ZFS
1200 * module, so that the module can't be unloaded out from underneath
1203 if (spa
->spa_did
!= 0) {
1204 thread_join(spa
->spa_did
);
1210 * Verify a pool configuration, and construct the vdev tree appropriately. This
1211 * will create all the necessary vdevs in the appropriate layout, with each vdev
1212 * in the CLOSED state. This will prep the pool before open/creation/import.
1213 * All vdev validation is done by the vdev_alloc() routine.
1216 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1217 uint_t id
, int atype
)
1224 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1227 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1230 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1233 if (error
== ENOENT
)
1239 return (SET_ERROR(EINVAL
));
1242 for (c
= 0; c
< children
; c
++) {
1244 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1252 ASSERT(*vdp
!= NULL
);
1258 * Opposite of spa_load().
1261 spa_unload(spa_t
*spa
)
1265 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1270 spa_async_suspend(spa
);
1275 if (spa
->spa_sync_on
) {
1276 txg_sync_stop(spa
->spa_dsl_pool
);
1277 spa
->spa_sync_on
= B_FALSE
;
1281 * Wait for any outstanding async I/O to complete.
1283 if (spa
->spa_async_zio_root
!= NULL
) {
1284 for (i
= 0; i
< max_ncpus
; i
++)
1285 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1286 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1287 spa
->spa_async_zio_root
= NULL
;
1290 bpobj_close(&spa
->spa_deferred_bpobj
);
1292 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1297 if (spa
->spa_root_vdev
)
1298 vdev_free(spa
->spa_root_vdev
);
1299 ASSERT(spa
->spa_root_vdev
== NULL
);
1302 * Close the dsl pool.
1304 if (spa
->spa_dsl_pool
) {
1305 dsl_pool_close(spa
->spa_dsl_pool
);
1306 spa
->spa_dsl_pool
= NULL
;
1307 spa
->spa_meta_objset
= NULL
;
1314 * Drop and purge level 2 cache
1316 spa_l2cache_drop(spa
);
1318 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1319 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1320 if (spa
->spa_spares
.sav_vdevs
) {
1321 kmem_free(spa
->spa_spares
.sav_vdevs
,
1322 spa
->spa_spares
.sav_count
* sizeof (void *));
1323 spa
->spa_spares
.sav_vdevs
= NULL
;
1325 if (spa
->spa_spares
.sav_config
) {
1326 nvlist_free(spa
->spa_spares
.sav_config
);
1327 spa
->spa_spares
.sav_config
= NULL
;
1329 spa
->spa_spares
.sav_count
= 0;
1331 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1332 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1333 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1335 if (spa
->spa_l2cache
.sav_vdevs
) {
1336 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1337 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1338 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1340 if (spa
->spa_l2cache
.sav_config
) {
1341 nvlist_free(spa
->spa_l2cache
.sav_config
);
1342 spa
->spa_l2cache
.sav_config
= NULL
;
1344 spa
->spa_l2cache
.sav_count
= 0;
1346 spa
->spa_async_suspended
= 0;
1348 if (spa
->spa_comment
!= NULL
) {
1349 spa_strfree(spa
->spa_comment
);
1350 spa
->spa_comment
= NULL
;
1353 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1357 * Load (or re-load) the current list of vdevs describing the active spares for
1358 * this pool. When this is called, we have some form of basic information in
1359 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1360 * then re-generate a more complete list including status information.
1363 spa_load_spares(spa_t
*spa
)
1370 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1373 * First, close and free any existing spare vdevs.
1375 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1376 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1378 /* Undo the call to spa_activate() below */
1379 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1380 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1381 spa_spare_remove(tvd
);
1386 if (spa
->spa_spares
.sav_vdevs
)
1387 kmem_free(spa
->spa_spares
.sav_vdevs
,
1388 spa
->spa_spares
.sav_count
* sizeof (void *));
1390 if (spa
->spa_spares
.sav_config
== NULL
)
1393 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1394 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1396 spa
->spa_spares
.sav_count
= (int)nspares
;
1397 spa
->spa_spares
.sav_vdevs
= NULL
;
1403 * Construct the array of vdevs, opening them to get status in the
1404 * process. For each spare, there is potentially two different vdev_t
1405 * structures associated with it: one in the list of spares (used only
1406 * for basic validation purposes) and one in the active vdev
1407 * configuration (if it's spared in). During this phase we open and
1408 * validate each vdev on the spare list. If the vdev also exists in the
1409 * active configuration, then we also mark this vdev as an active spare.
1411 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1413 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1414 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1415 VDEV_ALLOC_SPARE
) == 0);
1418 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1420 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1421 B_FALSE
)) != NULL
) {
1422 if (!tvd
->vdev_isspare
)
1426 * We only mark the spare active if we were successfully
1427 * able to load the vdev. Otherwise, importing a pool
1428 * with a bad active spare would result in strange
1429 * behavior, because multiple pool would think the spare
1430 * is actively in use.
1432 * There is a vulnerability here to an equally bizarre
1433 * circumstance, where a dead active spare is later
1434 * brought back to life (onlined or otherwise). Given
1435 * the rarity of this scenario, and the extra complexity
1436 * it adds, we ignore the possibility.
1438 if (!vdev_is_dead(tvd
))
1439 spa_spare_activate(tvd
);
1443 vd
->vdev_aux
= &spa
->spa_spares
;
1445 if (vdev_open(vd
) != 0)
1448 if (vdev_validate_aux(vd
) == 0)
1453 * Recompute the stashed list of spares, with status information
1456 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1457 DATA_TYPE_NVLIST_ARRAY
) == 0);
1459 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1461 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1462 spares
[i
] = vdev_config_generate(spa
,
1463 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1464 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1465 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1466 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1467 nvlist_free(spares
[i
]);
1468 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1472 * Load (or re-load) the current list of vdevs describing the active l2cache for
1473 * this pool. When this is called, we have some form of basic information in
1474 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1475 * then re-generate a more complete list including status information.
1476 * Devices which are already active have their details maintained, and are
1480 spa_load_l2cache(spa_t
*spa
)
1484 int i
, j
, oldnvdevs
;
1486 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1487 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1489 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1491 if (sav
->sav_config
!= NULL
) {
1492 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1493 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1494 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1500 oldvdevs
= sav
->sav_vdevs
;
1501 oldnvdevs
= sav
->sav_count
;
1502 sav
->sav_vdevs
= NULL
;
1506 * Process new nvlist of vdevs.
1508 for (i
= 0; i
< nl2cache
; i
++) {
1509 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1513 for (j
= 0; j
< oldnvdevs
; j
++) {
1515 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1517 * Retain previous vdev for add/remove ops.
1525 if (newvdevs
[i
] == NULL
) {
1529 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1530 VDEV_ALLOC_L2CACHE
) == 0);
1535 * Commit this vdev as an l2cache device,
1536 * even if it fails to open.
1538 spa_l2cache_add(vd
);
1543 spa_l2cache_activate(vd
);
1545 if (vdev_open(vd
) != 0)
1548 (void) vdev_validate_aux(vd
);
1550 if (!vdev_is_dead(vd
))
1551 l2arc_add_vdev(spa
, vd
);
1556 * Purge vdevs that were dropped
1558 for (i
= 0; i
< oldnvdevs
; i
++) {
1563 ASSERT(vd
->vdev_isl2cache
);
1565 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1566 pool
!= 0ULL && l2arc_vdev_present(vd
))
1567 l2arc_remove_vdev(vd
);
1568 vdev_clear_stats(vd
);
1574 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1576 if (sav
->sav_config
== NULL
)
1579 sav
->sav_vdevs
= newvdevs
;
1580 sav
->sav_count
= (int)nl2cache
;
1583 * Recompute the stashed list of l2cache devices, with status
1584 * information this time.
1586 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1587 DATA_TYPE_NVLIST_ARRAY
) == 0);
1589 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1590 for (i
= 0; i
< sav
->sav_count
; i
++)
1591 l2cache
[i
] = vdev_config_generate(spa
,
1592 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1593 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1594 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1596 for (i
= 0; i
< sav
->sav_count
; i
++)
1597 nvlist_free(l2cache
[i
]);
1599 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1603 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1606 char *packed
= NULL
;
1611 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1615 nvsize
= *(uint64_t *)db
->db_data
;
1616 dmu_buf_rele(db
, FTAG
);
1618 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1619 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1622 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1623 vmem_free(packed
, nvsize
);
1629 * Checks to see if the given vdev could not be opened, in which case we post a
1630 * sysevent to notify the autoreplace code that the device has been removed.
1633 spa_check_removed(vdev_t
*vd
)
1637 for (c
= 0; c
< vd
->vdev_children
; c
++)
1638 spa_check_removed(vd
->vdev_child
[c
]);
1640 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1642 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE
,
1643 vd
->vdev_spa
, vd
, NULL
, 0, 0);
1644 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_CHECK
);
1649 * Validate the current config against the MOS config
1652 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1654 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1658 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1660 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1661 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1663 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1666 * If we're doing a normal import, then build up any additional
1667 * diagnostic information about missing devices in this config.
1668 * We'll pass this up to the user for further processing.
1670 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1671 nvlist_t
**child
, *nv
;
1674 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1676 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1678 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1679 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1680 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1682 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1683 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1685 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1690 VERIFY(nvlist_add_nvlist_array(nv
,
1691 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1692 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1693 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1695 for (i
= 0; i
< idx
; i
++)
1696 nvlist_free(child
[i
]);
1699 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1703 * Compare the root vdev tree with the information we have
1704 * from the MOS config (mrvd). Check each top-level vdev
1705 * with the corresponding MOS config top-level (mtvd).
1707 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1708 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1709 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1712 * Resolve any "missing" vdevs in the current configuration.
1713 * If we find that the MOS config has more accurate information
1714 * about the top-level vdev then use that vdev instead.
1716 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1717 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1719 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1723 * Device specific actions.
1725 if (mtvd
->vdev_islog
) {
1726 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1729 * XXX - once we have 'readonly' pool
1730 * support we should be able to handle
1731 * missing data devices by transitioning
1732 * the pool to readonly.
1738 * Swap the missing vdev with the data we were
1739 * able to obtain from the MOS config.
1741 vdev_remove_child(rvd
, tvd
);
1742 vdev_remove_child(mrvd
, mtvd
);
1744 vdev_add_child(rvd
, mtvd
);
1745 vdev_add_child(mrvd
, tvd
);
1747 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1749 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1752 } else if (mtvd
->vdev_islog
) {
1754 * Load the slog device's state from the MOS config
1755 * since it's possible that the label does not
1756 * contain the most up-to-date information.
1758 vdev_load_log_state(tvd
, mtvd
);
1763 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1766 * Ensure we were able to validate the config.
1768 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1772 * Check for missing log devices
1775 spa_check_logs(spa_t
*spa
)
1777 boolean_t rv
= B_FALSE
;
1778 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1780 switch (spa
->spa_log_state
) {
1783 case SPA_LOG_MISSING
:
1784 /* need to recheck in case slog has been restored */
1785 case SPA_LOG_UNKNOWN
:
1786 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1787 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1789 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1796 spa_passivate_log(spa_t
*spa
)
1798 vdev_t
*rvd
= spa
->spa_root_vdev
;
1799 boolean_t slog_found
= B_FALSE
;
1802 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1804 if (!spa_has_slogs(spa
))
1807 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1808 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1809 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1811 if (tvd
->vdev_islog
) {
1812 metaslab_group_passivate(mg
);
1813 slog_found
= B_TRUE
;
1817 return (slog_found
);
1821 spa_activate_log(spa_t
*spa
)
1823 vdev_t
*rvd
= spa
->spa_root_vdev
;
1826 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1828 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1829 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1830 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1832 if (tvd
->vdev_islog
)
1833 metaslab_group_activate(mg
);
1838 spa_offline_log(spa_t
*spa
)
1842 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1843 NULL
, DS_FIND_CHILDREN
);
1846 * We successfully offlined the log device, sync out the
1847 * current txg so that the "stubby" block can be removed
1850 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1856 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1860 for (i
= 0; i
< sav
->sav_count
; i
++)
1861 spa_check_removed(sav
->sav_vdevs
[i
]);
1865 spa_claim_notify(zio_t
*zio
)
1867 spa_t
*spa
= zio
->io_spa
;
1872 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1873 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1874 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1875 mutex_exit(&spa
->spa_props_lock
);
1878 typedef struct spa_load_error
{
1879 uint64_t sle_meta_count
;
1880 uint64_t sle_data_count
;
1884 spa_load_verify_done(zio_t
*zio
)
1886 blkptr_t
*bp
= zio
->io_bp
;
1887 spa_load_error_t
*sle
= zio
->io_private
;
1888 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1889 int error
= zio
->io_error
;
1890 spa_t
*spa
= zio
->io_spa
;
1893 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1894 type
!= DMU_OT_INTENT_LOG
)
1895 atomic_add_64(&sle
->sle_meta_count
, 1);
1897 atomic_add_64(&sle
->sle_data_count
, 1);
1899 zio_data_buf_free(zio
->io_data
, zio
->io_size
);
1901 mutex_enter(&spa
->spa_scrub_lock
);
1902 spa
->spa_scrub_inflight
--;
1903 cv_broadcast(&spa
->spa_scrub_io_cv
);
1904 mutex_exit(&spa
->spa_scrub_lock
);
1908 * Maximum number of concurrent scrub i/os to create while verifying
1909 * a pool while importing it.
1911 int spa_load_verify_maxinflight
= 10000;
1912 int spa_load_verify_metadata
= B_TRUE
;
1913 int spa_load_verify_data
= B_TRUE
;
1917 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1918 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1924 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
1927 * Note: normally this routine will not be called if
1928 * spa_load_verify_metadata is not set. However, it may be useful
1929 * to manually set the flag after the traversal has begun.
1931 if (!spa_load_verify_metadata
)
1933 if (BP_GET_BUFC_TYPE(bp
) == ARC_BUFC_DATA
&& !spa_load_verify_data
)
1937 size
= BP_GET_PSIZE(bp
);
1938 data
= zio_data_buf_alloc(size
);
1940 mutex_enter(&spa
->spa_scrub_lock
);
1941 while (spa
->spa_scrub_inflight
>= spa_load_verify_maxinflight
)
1942 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
1943 spa
->spa_scrub_inflight
++;
1944 mutex_exit(&spa
->spa_scrub_lock
);
1946 zio_nowait(zio_read(rio
, spa
, bp
, data
, size
,
1947 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1948 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1949 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1954 spa_load_verify(spa_t
*spa
)
1957 spa_load_error_t sle
= { 0 };
1958 zpool_rewind_policy_t policy
;
1959 boolean_t verify_ok
= B_FALSE
;
1962 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1964 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1967 rio
= zio_root(spa
, NULL
, &sle
,
1968 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1970 if (spa_load_verify_metadata
) {
1971 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
1972 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
,
1973 spa_load_verify_cb
, rio
);
1976 (void) zio_wait(rio
);
1978 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
1979 spa
->spa_load_data_errors
= sle
.sle_data_count
;
1981 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
1982 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
1986 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
1987 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
1989 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
1990 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1991 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
1992 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
1993 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
1994 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
1995 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
1997 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2001 if (error
!= ENXIO
&& error
!= EIO
)
2002 error
= SET_ERROR(EIO
);
2006 return (verify_ok
? 0 : EIO
);
2010 * Find a value in the pool props object.
2013 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2015 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2016 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2020 * Find a value in the pool directory object.
2023 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2025 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2026 name
, sizeof (uint64_t), 1, val
));
2030 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2032 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2037 * Fix up config after a partly-completed split. This is done with the
2038 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2039 * pool have that entry in their config, but only the splitting one contains
2040 * a list of all the guids of the vdevs that are being split off.
2042 * This function determines what to do with that list: either rejoin
2043 * all the disks to the pool, or complete the splitting process. To attempt
2044 * the rejoin, each disk that is offlined is marked online again, and
2045 * we do a reopen() call. If the vdev label for every disk that was
2046 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2047 * then we call vdev_split() on each disk, and complete the split.
2049 * Otherwise we leave the config alone, with all the vdevs in place in
2050 * the original pool.
2053 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2060 boolean_t attempt_reopen
;
2062 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2065 /* check that the config is complete */
2066 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2067 &glist
, &gcount
) != 0)
2070 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2072 /* attempt to online all the vdevs & validate */
2073 attempt_reopen
= B_TRUE
;
2074 for (i
= 0; i
< gcount
; i
++) {
2075 if (glist
[i
] == 0) /* vdev is hole */
2078 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2079 if (vd
[i
] == NULL
) {
2081 * Don't bother attempting to reopen the disks;
2082 * just do the split.
2084 attempt_reopen
= B_FALSE
;
2086 /* attempt to re-online it */
2087 vd
[i
]->vdev_offline
= B_FALSE
;
2091 if (attempt_reopen
) {
2092 vdev_reopen(spa
->spa_root_vdev
);
2094 /* check each device to see what state it's in */
2095 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2096 if (vd
[i
] != NULL
&&
2097 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2104 * If every disk has been moved to the new pool, or if we never
2105 * even attempted to look at them, then we split them off for
2108 if (!attempt_reopen
|| gcount
== extracted
) {
2109 for (i
= 0; i
< gcount
; i
++)
2112 vdev_reopen(spa
->spa_root_vdev
);
2115 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2119 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2120 boolean_t mosconfig
)
2122 nvlist_t
*config
= spa
->spa_config
;
2123 char *ereport
= FM_EREPORT_ZFS_POOL
;
2129 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2130 return (SET_ERROR(EINVAL
));
2132 ASSERT(spa
->spa_comment
== NULL
);
2133 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2134 spa
->spa_comment
= spa_strdup(comment
);
2137 * Versioning wasn't explicitly added to the label until later, so if
2138 * it's not present treat it as the initial version.
2140 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2141 &spa
->spa_ubsync
.ub_version
) != 0)
2142 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2144 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2145 &spa
->spa_config_txg
);
2147 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2148 spa_guid_exists(pool_guid
, 0)) {
2149 error
= SET_ERROR(EEXIST
);
2151 spa
->spa_config_guid
= pool_guid
;
2153 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2155 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2159 nvlist_free(spa
->spa_load_info
);
2160 spa
->spa_load_info
= fnvlist_alloc();
2162 gethrestime(&spa
->spa_loaded_ts
);
2163 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2164 mosconfig
, &ereport
);
2168 * Don't count references from objsets that are already closed
2169 * and are making their way through the eviction process.
2171 spa_evicting_os_wait(spa
);
2172 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2174 if (error
!= EEXIST
) {
2175 spa
->spa_loaded_ts
.tv_sec
= 0;
2176 spa
->spa_loaded_ts
.tv_nsec
= 0;
2178 if (error
!= EBADF
) {
2179 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2182 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2189 * Load an existing storage pool, using the pool's builtin spa_config as a
2190 * source of configuration information.
2192 __attribute__((always_inline
))
2194 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2195 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2199 nvlist_t
*nvroot
= NULL
;
2202 uberblock_t
*ub
= &spa
->spa_uberblock
;
2203 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2204 int orig_mode
= spa
->spa_mode
;
2207 boolean_t missing_feat_write
= B_FALSE
;
2210 * If this is an untrusted config, access the pool in read-only mode.
2211 * This prevents things like resilvering recently removed devices.
2214 spa
->spa_mode
= FREAD
;
2216 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2218 spa
->spa_load_state
= state
;
2220 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2221 return (SET_ERROR(EINVAL
));
2223 parse
= (type
== SPA_IMPORT_EXISTING
?
2224 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2227 * Create "The Godfather" zio to hold all async IOs
2229 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2231 for (i
= 0; i
< max_ncpus
; i
++) {
2232 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2233 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2234 ZIO_FLAG_GODFATHER
);
2238 * Parse the configuration into a vdev tree. We explicitly set the
2239 * value that will be returned by spa_version() since parsing the
2240 * configuration requires knowing the version number.
2242 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2243 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2244 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2249 ASSERT(spa
->spa_root_vdev
== rvd
);
2250 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2251 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2253 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2254 ASSERT(spa_guid(spa
) == pool_guid
);
2258 * Try to open all vdevs, loading each label in the process.
2260 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2261 error
= vdev_open(rvd
);
2262 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2267 * We need to validate the vdev labels against the configuration that
2268 * we have in hand, which is dependent on the setting of mosconfig. If
2269 * mosconfig is true then we're validating the vdev labels based on
2270 * that config. Otherwise, we're validating against the cached config
2271 * (zpool.cache) that was read when we loaded the zfs module, and then
2272 * later we will recursively call spa_load() and validate against
2275 * If we're assembling a new pool that's been split off from an
2276 * existing pool, the labels haven't yet been updated so we skip
2277 * validation for now.
2279 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2280 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2281 error
= vdev_validate(rvd
, mosconfig
);
2282 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2287 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2288 return (SET_ERROR(ENXIO
));
2292 * Find the best uberblock.
2294 vdev_uberblock_load(rvd
, ub
, &label
);
2297 * If we weren't able to find a single valid uberblock, return failure.
2299 if (ub
->ub_txg
== 0) {
2301 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2305 * If the pool has an unsupported version we can't open it.
2307 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2309 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2312 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2316 * If we weren't able to find what's necessary for reading the
2317 * MOS in the label, return failure.
2319 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2320 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2322 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2327 * Update our in-core representation with the definitive values
2330 nvlist_free(spa
->spa_label_features
);
2331 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2337 * Look through entries in the label nvlist's features_for_read. If
2338 * there is a feature listed there which we don't understand then we
2339 * cannot open a pool.
2341 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2342 nvlist_t
*unsup_feat
;
2345 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2348 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2350 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2351 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2352 VERIFY(nvlist_add_string(unsup_feat
,
2353 nvpair_name(nvp
), "") == 0);
2357 if (!nvlist_empty(unsup_feat
)) {
2358 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2359 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2360 nvlist_free(unsup_feat
);
2361 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2365 nvlist_free(unsup_feat
);
2369 * If the vdev guid sum doesn't match the uberblock, we have an
2370 * incomplete configuration. We first check to see if the pool
2371 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2372 * If it is, defer the vdev_guid_sum check till later so we
2373 * can handle missing vdevs.
2375 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2376 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2377 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2378 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2380 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2381 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2382 spa_try_repair(spa
, config
);
2383 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2384 nvlist_free(spa
->spa_config_splitting
);
2385 spa
->spa_config_splitting
= NULL
;
2389 * Initialize internal SPA structures.
2391 spa
->spa_state
= POOL_STATE_ACTIVE
;
2392 spa
->spa_ubsync
= spa
->spa_uberblock
;
2393 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2394 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2395 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2396 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2397 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2398 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2400 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2402 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2403 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2405 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2406 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2408 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2409 boolean_t missing_feat_read
= B_FALSE
;
2410 nvlist_t
*unsup_feat
, *enabled_feat
;
2413 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2414 &spa
->spa_feat_for_read_obj
) != 0) {
2415 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2418 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2419 &spa
->spa_feat_for_write_obj
) != 0) {
2420 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2423 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2424 &spa
->spa_feat_desc_obj
) != 0) {
2425 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2428 enabled_feat
= fnvlist_alloc();
2429 unsup_feat
= fnvlist_alloc();
2431 if (!spa_features_check(spa
, B_FALSE
,
2432 unsup_feat
, enabled_feat
))
2433 missing_feat_read
= B_TRUE
;
2435 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2436 if (!spa_features_check(spa
, B_TRUE
,
2437 unsup_feat
, enabled_feat
)) {
2438 missing_feat_write
= B_TRUE
;
2442 fnvlist_add_nvlist(spa
->spa_load_info
,
2443 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2445 if (!nvlist_empty(unsup_feat
)) {
2446 fnvlist_add_nvlist(spa
->spa_load_info
,
2447 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2450 fnvlist_free(enabled_feat
);
2451 fnvlist_free(unsup_feat
);
2453 if (!missing_feat_read
) {
2454 fnvlist_add_boolean(spa
->spa_load_info
,
2455 ZPOOL_CONFIG_CAN_RDONLY
);
2459 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2460 * twofold: to determine whether the pool is available for
2461 * import in read-write mode and (if it is not) whether the
2462 * pool is available for import in read-only mode. If the pool
2463 * is available for import in read-write mode, it is displayed
2464 * as available in userland; if it is not available for import
2465 * in read-only mode, it is displayed as unavailable in
2466 * userland. If the pool is available for import in read-only
2467 * mode but not read-write mode, it is displayed as unavailable
2468 * in userland with a special note that the pool is actually
2469 * available for open in read-only mode.
2471 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2472 * missing a feature for write, we must first determine whether
2473 * the pool can be opened read-only before returning to
2474 * userland in order to know whether to display the
2475 * abovementioned note.
2477 if (missing_feat_read
|| (missing_feat_write
&&
2478 spa_writeable(spa
))) {
2479 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2484 * Load refcounts for ZFS features from disk into an in-memory
2485 * cache during SPA initialization.
2487 for (i
= 0; i
< SPA_FEATURES
; i
++) {
2490 error
= feature_get_refcount_from_disk(spa
,
2491 &spa_feature_table
[i
], &refcount
);
2493 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2494 } else if (error
== ENOTSUP
) {
2495 spa
->spa_feat_refcount_cache
[i
] =
2496 SPA_FEATURE_DISABLED
;
2498 return (spa_vdev_err(rvd
,
2499 VDEV_AUX_CORRUPT_DATA
, EIO
));
2504 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2505 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2506 &spa
->spa_feat_enabled_txg_obj
) != 0)
2507 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2510 spa
->spa_is_initializing
= B_TRUE
;
2511 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2512 spa
->spa_is_initializing
= B_FALSE
;
2514 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2518 nvlist_t
*policy
= NULL
, *nvconfig
;
2520 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2521 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2523 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2524 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2526 unsigned long myhostid
= 0;
2528 VERIFY(nvlist_lookup_string(nvconfig
,
2529 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2532 myhostid
= zone_get_hostid(NULL
);
2535 * We're emulating the system's hostid in userland, so
2536 * we can't use zone_get_hostid().
2538 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2539 #endif /* _KERNEL */
2540 if (hostid
!= 0 && myhostid
!= 0 &&
2541 hostid
!= myhostid
) {
2542 nvlist_free(nvconfig
);
2543 cmn_err(CE_WARN
, "pool '%s' could not be "
2544 "loaded as it was last accessed by another "
2545 "system (host: %s hostid: 0x%lx). See: "
2546 "http://zfsonlinux.org/msg/ZFS-8000-EY",
2547 spa_name(spa
), hostname
,
2548 (unsigned long)hostid
);
2549 return (SET_ERROR(EBADF
));
2552 if (nvlist_lookup_nvlist(spa
->spa_config
,
2553 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2554 VERIFY(nvlist_add_nvlist(nvconfig
,
2555 ZPOOL_REWIND_POLICY
, policy
) == 0);
2557 spa_config_set(spa
, nvconfig
);
2559 spa_deactivate(spa
);
2560 spa_activate(spa
, orig_mode
);
2562 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2565 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2566 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2567 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2569 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2572 * Load the bit that tells us to use the new accounting function
2573 * (raid-z deflation). If we have an older pool, this will not
2576 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2577 if (error
!= 0 && error
!= ENOENT
)
2578 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2580 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2581 &spa
->spa_creation_version
);
2582 if (error
!= 0 && error
!= ENOENT
)
2583 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2586 * Load the persistent error log. If we have an older pool, this will
2589 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2590 if (error
!= 0 && error
!= ENOENT
)
2591 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2593 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2594 &spa
->spa_errlog_scrub
);
2595 if (error
!= 0 && error
!= ENOENT
)
2596 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2599 * Load the history object. If we have an older pool, this
2600 * will not be present.
2602 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2603 if (error
!= 0 && error
!= ENOENT
)
2604 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2607 * If we're assembling the pool from the split-off vdevs of
2608 * an existing pool, we don't want to attach the spares & cache
2613 * Load any hot spares for this pool.
2615 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
2616 if (error
!= 0 && error
!= ENOENT
)
2617 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2618 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2619 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
2620 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
2621 &spa
->spa_spares
.sav_config
) != 0)
2622 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2624 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2625 spa_load_spares(spa
);
2626 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2627 } else if (error
== 0) {
2628 spa
->spa_spares
.sav_sync
= B_TRUE
;
2632 * Load any level 2 ARC devices for this pool.
2634 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
2635 &spa
->spa_l2cache
.sav_object
);
2636 if (error
!= 0 && error
!= ENOENT
)
2637 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2638 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
2639 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
2640 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
2641 &spa
->spa_l2cache
.sav_config
) != 0)
2642 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2644 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2645 spa_load_l2cache(spa
);
2646 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2647 } else if (error
== 0) {
2648 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
2651 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
2653 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
2654 if (error
&& error
!= ENOENT
)
2655 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2658 uint64_t autoreplace
= 0;
2660 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
2661 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
2662 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
2663 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
2664 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
2665 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
2666 &spa
->spa_dedup_ditto
);
2668 spa
->spa_autoreplace
= (autoreplace
!= 0);
2672 * If the 'autoreplace' property is set, then post a resource notifying
2673 * the ZFS DE that it should not issue any faults for unopenable
2674 * devices. We also iterate over the vdevs, and post a sysevent for any
2675 * unopenable vdevs so that the normal autoreplace handler can take
2678 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
2679 spa_check_removed(spa
->spa_root_vdev
);
2681 * For the import case, this is done in spa_import(), because
2682 * at this point we're using the spare definitions from
2683 * the MOS config, not necessarily from the userland config.
2685 if (state
!= SPA_LOAD_IMPORT
) {
2686 spa_aux_check_removed(&spa
->spa_spares
);
2687 spa_aux_check_removed(&spa
->spa_l2cache
);
2692 * Load the vdev state for all toplevel vdevs.
2697 * Propagate the leaf DTLs we just loaded all the way up the tree.
2699 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2700 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
2701 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2704 * Load the DDTs (dedup tables).
2706 error
= ddt_load(spa
);
2708 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2710 spa_update_dspace(spa
);
2713 * Validate the config, using the MOS config to fill in any
2714 * information which might be missing. If we fail to validate
2715 * the config then declare the pool unfit for use. If we're
2716 * assembling a pool from a split, the log is not transferred
2719 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2722 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2723 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2725 if (!spa_config_valid(spa
, nvconfig
)) {
2726 nvlist_free(nvconfig
);
2727 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2730 nvlist_free(nvconfig
);
2733 * Now that we've validated the config, check the state of the
2734 * root vdev. If it can't be opened, it indicates one or
2735 * more toplevel vdevs are faulted.
2737 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2738 return (SET_ERROR(ENXIO
));
2740 if (spa_writeable(spa
) && spa_check_logs(spa
)) {
2741 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
2742 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
2746 if (missing_feat_write
) {
2747 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
2750 * At this point, we know that we can open the pool in
2751 * read-only mode but not read-write mode. We now have enough
2752 * information and can return to userland.
2754 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
2758 * We've successfully opened the pool, verify that we're ready
2759 * to start pushing transactions.
2761 if (state
!= SPA_LOAD_TRYIMPORT
) {
2762 if ((error
= spa_load_verify(spa
)))
2763 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2767 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
2768 spa
->spa_load_max_txg
== UINT64_MAX
)) {
2770 int need_update
= B_FALSE
;
2771 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2774 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
2777 * Claim log blocks that haven't been committed yet.
2778 * This must all happen in a single txg.
2779 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2780 * invoked from zil_claim_log_block()'s i/o done callback.
2781 * Price of rollback is that we abandon the log.
2783 spa
->spa_claiming
= B_TRUE
;
2785 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
2786 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
2787 zil_claim
, tx
, DS_FIND_CHILDREN
);
2790 spa
->spa_claiming
= B_FALSE
;
2792 spa_set_log_state(spa
, SPA_LOG_GOOD
);
2793 spa
->spa_sync_on
= B_TRUE
;
2794 txg_sync_start(spa
->spa_dsl_pool
);
2797 * Wait for all claims to sync. We sync up to the highest
2798 * claimed log block birth time so that claimed log blocks
2799 * don't appear to be from the future. spa_claim_max_txg
2800 * will have been set for us by either zil_check_log_chain()
2801 * (invoked from spa_check_logs()) or zil_claim() above.
2803 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
2806 * If the config cache is stale, or we have uninitialized
2807 * metaslabs (see spa_vdev_add()), then update the config.
2809 * If this is a verbatim import, trust the current
2810 * in-core spa_config and update the disk labels.
2812 if (config_cache_txg
!= spa
->spa_config_txg
||
2813 state
== SPA_LOAD_IMPORT
||
2814 state
== SPA_LOAD_RECOVER
||
2815 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
2816 need_update
= B_TRUE
;
2818 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2819 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
2820 need_update
= B_TRUE
;
2823 * Update the config cache asychronously in case we're the
2824 * root pool, in which case the config cache isn't writable yet.
2827 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2830 * Check all DTLs to see if anything needs resilvering.
2832 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
2833 vdev_resilver_needed(rvd
, NULL
, NULL
))
2834 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2837 * Log the fact that we booted up (so that we can detect if
2838 * we rebooted in the middle of an operation).
2840 spa_history_log_version(spa
, "open");
2843 * Delete any inconsistent datasets.
2845 (void) dmu_objset_find(spa_name(spa
),
2846 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
2849 * Clean up any stale temporary dataset userrefs.
2851 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
2858 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
2860 int mode
= spa
->spa_mode
;
2863 spa_deactivate(spa
);
2865 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
2867 spa_activate(spa
, mode
);
2868 spa_async_suspend(spa
);
2870 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
2874 * If spa_load() fails this function will try loading prior txg's. If
2875 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2876 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2877 * function will not rewind the pool and will return the same error as
2881 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
2882 uint64_t max_request
, int rewind_flags
)
2884 nvlist_t
*loadinfo
= NULL
;
2885 nvlist_t
*config
= NULL
;
2886 int load_error
, rewind_error
;
2887 uint64_t safe_rewind_txg
;
2890 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
2891 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
2892 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2894 spa
->spa_load_max_txg
= max_request
;
2895 if (max_request
!= UINT64_MAX
)
2896 spa
->spa_extreme_rewind
= B_TRUE
;
2899 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
2901 if (load_error
== 0)
2904 if (spa
->spa_root_vdev
!= NULL
)
2905 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
2907 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
2908 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2910 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
2911 nvlist_free(config
);
2912 return (load_error
);
2915 if (state
== SPA_LOAD_RECOVER
) {
2916 /* Price of rolling back is discarding txgs, including log */
2917 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
2920 * If we aren't rolling back save the load info from our first
2921 * import attempt so that we can restore it after attempting
2924 loadinfo
= spa
->spa_load_info
;
2925 spa
->spa_load_info
= fnvlist_alloc();
2928 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
2929 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
2930 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
2931 TXG_INITIAL
: safe_rewind_txg
;
2934 * Continue as long as we're finding errors, we're still within
2935 * the acceptable rewind range, and we're still finding uberblocks
2937 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
2938 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
2939 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
2940 spa
->spa_extreme_rewind
= B_TRUE
;
2941 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
2944 spa
->spa_extreme_rewind
= B_FALSE
;
2945 spa
->spa_load_max_txg
= UINT64_MAX
;
2947 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
2948 spa_config_set(spa
, config
);
2950 if (state
== SPA_LOAD_RECOVER
) {
2951 ASSERT3P(loadinfo
, ==, NULL
);
2952 return (rewind_error
);
2954 /* Store the rewind info as part of the initial load info */
2955 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
2956 spa
->spa_load_info
);
2958 /* Restore the initial load info */
2959 fnvlist_free(spa
->spa_load_info
);
2960 spa
->spa_load_info
= loadinfo
;
2962 return (load_error
);
2969 * The import case is identical to an open except that the configuration is sent
2970 * down from userland, instead of grabbed from the configuration cache. For the
2971 * case of an open, the pool configuration will exist in the
2972 * POOL_STATE_UNINITIALIZED state.
2974 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2975 * the same time open the pool, without having to keep around the spa_t in some
2979 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
2983 spa_load_state_t state
= SPA_LOAD_OPEN
;
2985 int locked
= B_FALSE
;
2986 int firstopen
= B_FALSE
;
2991 * As disgusting as this is, we need to support recursive calls to this
2992 * function because dsl_dir_open() is called during spa_load(), and ends
2993 * up calling spa_open() again. The real fix is to figure out how to
2994 * avoid dsl_dir_open() calling this in the first place.
2996 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
2997 mutex_enter(&spa_namespace_lock
);
3001 if ((spa
= spa_lookup(pool
)) == NULL
) {
3003 mutex_exit(&spa_namespace_lock
);
3004 return (SET_ERROR(ENOENT
));
3007 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3008 zpool_rewind_policy_t policy
;
3012 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3014 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3015 state
= SPA_LOAD_RECOVER
;
3017 spa_activate(spa
, spa_mode_global
);
3019 if (state
!= SPA_LOAD_RECOVER
)
3020 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3022 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3023 policy
.zrp_request
);
3025 if (error
== EBADF
) {
3027 * If vdev_validate() returns failure (indicated by
3028 * EBADF), it indicates that one of the vdevs indicates
3029 * that the pool has been exported or destroyed. If
3030 * this is the case, the config cache is out of sync and
3031 * we should remove the pool from the namespace.
3034 spa_deactivate(spa
);
3035 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3038 mutex_exit(&spa_namespace_lock
);
3039 return (SET_ERROR(ENOENT
));
3044 * We can't open the pool, but we still have useful
3045 * information: the state of each vdev after the
3046 * attempted vdev_open(). Return this to the user.
3048 if (config
!= NULL
&& spa
->spa_config
) {
3049 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3051 VERIFY(nvlist_add_nvlist(*config
,
3052 ZPOOL_CONFIG_LOAD_INFO
,
3053 spa
->spa_load_info
) == 0);
3056 spa_deactivate(spa
);
3057 spa
->spa_last_open_failed
= error
;
3059 mutex_exit(&spa_namespace_lock
);
3065 spa_open_ref(spa
, tag
);
3068 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3071 * If we've recovered the pool, pass back any information we
3072 * gathered while doing the load.
3074 if (state
== SPA_LOAD_RECOVER
) {
3075 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3076 spa
->spa_load_info
) == 0);
3080 spa
->spa_last_open_failed
= 0;
3081 spa
->spa_last_ubsync_txg
= 0;
3082 spa
->spa_load_txg
= 0;
3083 mutex_exit(&spa_namespace_lock
);
3088 zvol_create_minors(spa
->spa_name
);
3097 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3100 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3104 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3106 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3110 * Lookup the given spa_t, incrementing the inject count in the process,
3111 * preventing it from being exported or destroyed.
3114 spa_inject_addref(char *name
)
3118 mutex_enter(&spa_namespace_lock
);
3119 if ((spa
= spa_lookup(name
)) == NULL
) {
3120 mutex_exit(&spa_namespace_lock
);
3123 spa
->spa_inject_ref
++;
3124 mutex_exit(&spa_namespace_lock
);
3130 spa_inject_delref(spa_t
*spa
)
3132 mutex_enter(&spa_namespace_lock
);
3133 spa
->spa_inject_ref
--;
3134 mutex_exit(&spa_namespace_lock
);
3138 * Add spares device information to the nvlist.
3141 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3151 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3153 if (spa
->spa_spares
.sav_count
== 0)
3156 VERIFY(nvlist_lookup_nvlist(config
,
3157 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3158 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3159 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3161 VERIFY(nvlist_add_nvlist_array(nvroot
,
3162 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3163 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3164 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3167 * Go through and find any spares which have since been
3168 * repurposed as an active spare. If this is the case, update
3169 * their status appropriately.
3171 for (i
= 0; i
< nspares
; i
++) {
3172 VERIFY(nvlist_lookup_uint64(spares
[i
],
3173 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3174 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3176 VERIFY(nvlist_lookup_uint64_array(
3177 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3178 (uint64_t **)&vs
, &vsc
) == 0);
3179 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3180 vs
->vs_aux
= VDEV_AUX_SPARED
;
3187 * Add l2cache device information to the nvlist, including vdev stats.
3190 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3193 uint_t i
, j
, nl2cache
;
3200 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3202 if (spa
->spa_l2cache
.sav_count
== 0)
3205 VERIFY(nvlist_lookup_nvlist(config
,
3206 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3207 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3208 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3209 if (nl2cache
!= 0) {
3210 VERIFY(nvlist_add_nvlist_array(nvroot
,
3211 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3212 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3213 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3216 * Update level 2 cache device stats.
3219 for (i
= 0; i
< nl2cache
; i
++) {
3220 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3221 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3224 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3226 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3227 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3233 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3234 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3236 vdev_get_stats(vd
, vs
);
3242 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3247 if (spa
->spa_feat_for_read_obj
!= 0) {
3248 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3249 spa
->spa_feat_for_read_obj
);
3250 zap_cursor_retrieve(&zc
, &za
) == 0;
3251 zap_cursor_advance(&zc
)) {
3252 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3253 za
.za_num_integers
== 1);
3254 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3255 za
.za_first_integer
));
3257 zap_cursor_fini(&zc
);
3260 if (spa
->spa_feat_for_write_obj
!= 0) {
3261 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3262 spa
->spa_feat_for_write_obj
);
3263 zap_cursor_retrieve(&zc
, &za
) == 0;
3264 zap_cursor_advance(&zc
)) {
3265 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3266 za
.za_num_integers
== 1);
3267 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3268 za
.za_first_integer
));
3270 zap_cursor_fini(&zc
);
3275 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3279 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3280 zfeature_info_t feature
= spa_feature_table
[i
];
3283 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3286 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3291 * Store a list of pool features and their reference counts in the
3294 * The first time this is called on a spa, allocate a new nvlist, fetch
3295 * the pool features and reference counts from disk, then save the list
3296 * in the spa. In subsequent calls on the same spa use the saved nvlist
3297 * and refresh its values from the cached reference counts. This
3298 * ensures we don't block here on I/O on a suspended pool so 'zpool
3299 * clear' can resume the pool.
3302 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3306 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3308 mutex_enter(&spa
->spa_feat_stats_lock
);
3309 features
= spa
->spa_feat_stats
;
3311 if (features
!= NULL
) {
3312 spa_feature_stats_from_cache(spa
, features
);
3314 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3315 spa
->spa_feat_stats
= features
;
3316 spa_feature_stats_from_disk(spa
, features
);
3319 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3322 mutex_exit(&spa
->spa_feat_stats_lock
);
3326 spa_get_stats(const char *name
, nvlist_t
**config
,
3327 char *altroot
, size_t buflen
)
3333 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3337 * This still leaves a window of inconsistency where the spares
3338 * or l2cache devices could change and the config would be
3339 * self-inconsistent.
3341 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3343 if (*config
!= NULL
) {
3344 uint64_t loadtimes
[2];
3346 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3347 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3348 VERIFY(nvlist_add_uint64_array(*config
,
3349 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3351 VERIFY(nvlist_add_uint64(*config
,
3352 ZPOOL_CONFIG_ERRCOUNT
,
3353 spa_get_errlog_size(spa
)) == 0);
3355 if (spa_suspended(spa
))
3356 VERIFY(nvlist_add_uint64(*config
,
3357 ZPOOL_CONFIG_SUSPENDED
,
3358 spa
->spa_failmode
) == 0);
3360 spa_add_spares(spa
, *config
);
3361 spa_add_l2cache(spa
, *config
);
3362 spa_add_feature_stats(spa
, *config
);
3367 * We want to get the alternate root even for faulted pools, so we cheat
3368 * and call spa_lookup() directly.
3372 mutex_enter(&spa_namespace_lock
);
3373 spa
= spa_lookup(name
);
3375 spa_altroot(spa
, altroot
, buflen
);
3379 mutex_exit(&spa_namespace_lock
);
3381 spa_altroot(spa
, altroot
, buflen
);
3386 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3387 spa_close(spa
, FTAG
);
3394 * Validate that the auxiliary device array is well formed. We must have an
3395 * array of nvlists, each which describes a valid leaf vdev. If this is an
3396 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3397 * specified, as long as they are well-formed.
3400 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3401 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3402 vdev_labeltype_t label
)
3409 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3412 * It's acceptable to have no devs specified.
3414 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3418 return (SET_ERROR(EINVAL
));
3421 * Make sure the pool is formatted with a version that supports this
3424 if (spa_version(spa
) < version
)
3425 return (SET_ERROR(ENOTSUP
));
3428 * Set the pending device list so we correctly handle device in-use
3431 sav
->sav_pending
= dev
;
3432 sav
->sav_npending
= ndev
;
3434 for (i
= 0; i
< ndev
; i
++) {
3435 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3439 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3441 error
= SET_ERROR(EINVAL
);
3446 * The L2ARC currently only supports disk devices in
3447 * kernel context. For user-level testing, we allow it.
3450 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
3451 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
3452 error
= SET_ERROR(ENOTBLK
);
3459 if ((error
= vdev_open(vd
)) == 0 &&
3460 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3461 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3462 vd
->vdev_guid
) == 0);
3468 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3475 sav
->sav_pending
= NULL
;
3476 sav
->sav_npending
= 0;
3481 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3485 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3487 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3488 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3489 VDEV_LABEL_SPARE
)) != 0) {
3493 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3494 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3495 VDEV_LABEL_L2CACHE
));
3499 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3504 if (sav
->sav_config
!= NULL
) {
3510 * Generate new dev list by concatentating with the
3513 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3514 &olddevs
, &oldndevs
) == 0);
3516 newdevs
= kmem_alloc(sizeof (void *) *
3517 (ndevs
+ oldndevs
), KM_SLEEP
);
3518 for (i
= 0; i
< oldndevs
; i
++)
3519 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3521 for (i
= 0; i
< ndevs
; i
++)
3522 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3525 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3526 DATA_TYPE_NVLIST_ARRAY
) == 0);
3528 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3529 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3530 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3531 nvlist_free(newdevs
[i
]);
3532 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3535 * Generate a new dev list.
3537 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3539 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3545 * Stop and drop level 2 ARC devices
3548 spa_l2cache_drop(spa_t
*spa
)
3552 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3554 for (i
= 0; i
< sav
->sav_count
; i
++) {
3557 vd
= sav
->sav_vdevs
[i
];
3560 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3561 pool
!= 0ULL && l2arc_vdev_present(vd
))
3562 l2arc_remove_vdev(vd
);
3570 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3574 char *altroot
= NULL
;
3579 uint64_t txg
= TXG_INITIAL
;
3580 nvlist_t
**spares
, **l2cache
;
3581 uint_t nspares
, nl2cache
;
3582 uint64_t version
, obj
;
3583 boolean_t has_features
;
3589 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
3590 poolname
= (char *)pool
;
3593 * If this pool already exists, return failure.
3595 mutex_enter(&spa_namespace_lock
);
3596 if (spa_lookup(poolname
) != NULL
) {
3597 mutex_exit(&spa_namespace_lock
);
3598 return (SET_ERROR(EEXIST
));
3602 * Allocate a new spa_t structure.
3604 nvl
= fnvlist_alloc();
3605 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
3606 (void) nvlist_lookup_string(props
,
3607 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
3608 spa
= spa_add(poolname
, nvl
, altroot
);
3610 spa_activate(spa
, spa_mode_global
);
3612 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
3613 spa_deactivate(spa
);
3615 mutex_exit(&spa_namespace_lock
);
3620 * Temporary pool names should never be written to disk.
3622 if (poolname
!= pool
)
3623 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
3625 has_features
= B_FALSE
;
3626 for (elem
= nvlist_next_nvpair(props
, NULL
);
3627 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
3628 if (zpool_prop_feature(nvpair_name(elem
)))
3629 has_features
= B_TRUE
;
3632 if (has_features
|| nvlist_lookup_uint64(props
,
3633 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
3634 version
= SPA_VERSION
;
3636 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
3638 spa
->spa_first_txg
= txg
;
3639 spa
->spa_uberblock
.ub_txg
= txg
- 1;
3640 spa
->spa_uberblock
.ub_version
= version
;
3641 spa
->spa_ubsync
= spa
->spa_uberblock
;
3644 * Create "The Godfather" zio to hold all async IOs
3646 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
3648 for (i
= 0; i
< max_ncpus
; i
++) {
3649 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
3650 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
3651 ZIO_FLAG_GODFATHER
);
3655 * Create the root vdev.
3657 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3659 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
3661 ASSERT(error
!= 0 || rvd
!= NULL
);
3662 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
3664 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
3665 error
= SET_ERROR(EINVAL
);
3668 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
3669 (error
= spa_validate_aux(spa
, nvroot
, txg
,
3670 VDEV_ALLOC_ADD
)) == 0) {
3671 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
3672 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
3673 vdev_expand(rvd
->vdev_child
[c
], txg
);
3677 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3681 spa_deactivate(spa
);
3683 mutex_exit(&spa_namespace_lock
);
3688 * Get the list of spares, if specified.
3690 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
3691 &spares
, &nspares
) == 0) {
3692 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
3694 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
3695 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3696 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3697 spa_load_spares(spa
);
3698 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3699 spa
->spa_spares
.sav_sync
= B_TRUE
;
3703 * Get the list of level 2 cache devices, if specified.
3705 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
3706 &l2cache
, &nl2cache
) == 0) {
3707 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
3708 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3709 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
3710 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3711 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3712 spa_load_l2cache(spa
);
3713 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3714 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3717 spa
->spa_is_initializing
= B_TRUE
;
3718 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
3719 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
3720 spa
->spa_is_initializing
= B_FALSE
;
3723 * Create DDTs (dedup tables).
3727 spa_update_dspace(spa
);
3729 tx
= dmu_tx_create_assigned(dp
, txg
);
3732 * Create the pool config object.
3734 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
3735 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
3736 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
3738 if (zap_add(spa
->spa_meta_objset
,
3739 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
3740 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
3741 cmn_err(CE_PANIC
, "failed to add pool config");
3744 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
3745 spa_feature_create_zap_objects(spa
, tx
);
3747 if (zap_add(spa
->spa_meta_objset
,
3748 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
3749 sizeof (uint64_t), 1, &version
, tx
) != 0) {
3750 cmn_err(CE_PANIC
, "failed to add pool version");
3753 /* Newly created pools with the right version are always deflated. */
3754 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
3755 spa
->spa_deflate
= TRUE
;
3756 if (zap_add(spa
->spa_meta_objset
,
3757 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
3758 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
3759 cmn_err(CE_PANIC
, "failed to add deflate");
3764 * Create the deferred-free bpobj. Turn off compression
3765 * because sync-to-convergence takes longer if the blocksize
3768 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
3769 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
3770 ZIO_COMPRESS_OFF
, tx
);
3771 if (zap_add(spa
->spa_meta_objset
,
3772 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
3773 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
3774 cmn_err(CE_PANIC
, "failed to add bpobj");
3776 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
3777 spa
->spa_meta_objset
, obj
));
3780 * Create the pool's history object.
3782 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
3783 spa_history_create_obj(spa
, tx
);
3786 * Set pool properties.
3788 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
3789 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3790 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
3791 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
3793 if (props
!= NULL
) {
3794 spa_configfile_set(spa
, props
, B_FALSE
);
3795 spa_sync_props(props
, tx
);
3800 spa
->spa_sync_on
= B_TRUE
;
3801 txg_sync_start(spa
->spa_dsl_pool
);
3804 * We explicitly wait for the first transaction to complete so that our
3805 * bean counters are appropriately updated.
3807 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
3809 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
3811 spa_history_log_version(spa
, "create");
3814 * Don't count references from objsets that are already closed
3815 * and are making their way through the eviction process.
3817 spa_evicting_os_wait(spa
);
3818 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
3820 mutex_exit(&spa_namespace_lock
);
3827 * Get the root pool information from the root disk, then import the root pool
3828 * during the system boot up time.
3830 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
3833 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
3836 nvlist_t
*nvtop
, *nvroot
;
3839 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
3843 * Add this top-level vdev to the child array.
3845 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3847 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
3849 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
3852 * Put this pool's top-level vdevs into a root vdev.
3854 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3855 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
3856 VDEV_TYPE_ROOT
) == 0);
3857 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
3858 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
3859 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
3863 * Replace the existing vdev_tree with the new root vdev in
3864 * this pool's configuration (remove the old, add the new).
3866 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
3867 nvlist_free(nvroot
);
3872 * Walk the vdev tree and see if we can find a device with "better"
3873 * configuration. A configuration is "better" if the label on that
3874 * device has a more recent txg.
3877 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
3881 for (c
= 0; c
< vd
->vdev_children
; c
++)
3882 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
3884 if (vd
->vdev_ops
->vdev_op_leaf
) {
3888 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
3892 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
3896 * Do we have a better boot device?
3898 if (label_txg
> *txg
) {
3907 * Import a root pool.
3909 * For x86. devpath_list will consist of devid and/or physpath name of
3910 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3911 * The GRUB "findroot" command will return the vdev we should boot.
3913 * For Sparc, devpath_list consists the physpath name of the booting device
3914 * no matter the rootpool is a single device pool or a mirrored pool.
3916 * "/pci@1f,0/ide@d/disk@0,0:a"
3919 spa_import_rootpool(char *devpath
, char *devid
)
3922 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
3923 nvlist_t
*config
, *nvtop
;
3929 * Read the label from the boot device and generate a configuration.
3931 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3932 #if defined(_OBP) && defined(_KERNEL)
3933 if (config
== NULL
) {
3934 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
3936 get_iscsi_bootpath_phy(devpath
);
3937 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
3941 if (config
== NULL
) {
3942 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
3944 return (SET_ERROR(EIO
));
3947 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
3949 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
3951 mutex_enter(&spa_namespace_lock
);
3952 if ((spa
= spa_lookup(pname
)) != NULL
) {
3954 * Remove the existing root pool from the namespace so that we
3955 * can replace it with the correct config we just read in.
3960 spa
= spa_add(pname
, config
, NULL
);
3961 spa
->spa_is_root
= B_TRUE
;
3962 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
3965 * Build up a vdev tree based on the boot device's label config.
3967 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
3969 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3970 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
3971 VDEV_ALLOC_ROOTPOOL
);
3972 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3974 mutex_exit(&spa_namespace_lock
);
3975 nvlist_free(config
);
3976 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
3982 * Get the boot vdev.
3984 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
3985 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
3986 (u_longlong_t
)guid
);
3987 error
= SET_ERROR(ENOENT
);
3992 * Determine if there is a better boot device.
3995 spa_alt_rootvdev(rvd
, &avd
, &txg
);
3997 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
3998 "try booting from '%s'", avd
->vdev_path
);
3999 error
= SET_ERROR(EINVAL
);
4004 * If the boot device is part of a spare vdev then ensure that
4005 * we're booting off the active spare.
4007 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4008 !bvd
->vdev_isspare
) {
4009 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
4010 "try booting from '%s'",
4012 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
4013 error
= SET_ERROR(EINVAL
);
4019 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4021 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4022 mutex_exit(&spa_namespace_lock
);
4024 nvlist_free(config
);
4031 * Import a non-root pool into the system.
4034 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4037 char *altroot
= NULL
;
4038 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4039 zpool_rewind_policy_t policy
;
4040 uint64_t mode
= spa_mode_global
;
4041 uint64_t readonly
= B_FALSE
;
4044 nvlist_t
**spares
, **l2cache
;
4045 uint_t nspares
, nl2cache
;
4048 * If a pool with this name exists, return failure.
4050 mutex_enter(&spa_namespace_lock
);
4051 if (spa_lookup(pool
) != NULL
) {
4052 mutex_exit(&spa_namespace_lock
);
4053 return (SET_ERROR(EEXIST
));
4057 * Create and initialize the spa structure.
4059 (void) nvlist_lookup_string(props
,
4060 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4061 (void) nvlist_lookup_uint64(props
,
4062 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4065 spa
= spa_add(pool
, config
, altroot
);
4066 spa
->spa_import_flags
= flags
;
4069 * Verbatim import - Take a pool and insert it into the namespace
4070 * as if it had been loaded at boot.
4072 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4074 spa_configfile_set(spa
, props
, B_FALSE
);
4076 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4078 mutex_exit(&spa_namespace_lock
);
4082 spa_activate(spa
, mode
);
4085 * Don't start async tasks until we know everything is healthy.
4087 spa_async_suspend(spa
);
4089 zpool_get_rewind_policy(config
, &policy
);
4090 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4091 state
= SPA_LOAD_RECOVER
;
4094 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4095 * because the user-supplied config is actually the one to trust when
4098 if (state
!= SPA_LOAD_RECOVER
)
4099 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4101 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4102 policy
.zrp_request
);
4105 * Propagate anything learned while loading the pool and pass it
4106 * back to caller (i.e. rewind info, missing devices, etc).
4108 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4109 spa
->spa_load_info
) == 0);
4111 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4113 * Toss any existing sparelist, as it doesn't have any validity
4114 * anymore, and conflicts with spa_has_spare().
4116 if (spa
->spa_spares
.sav_config
) {
4117 nvlist_free(spa
->spa_spares
.sav_config
);
4118 spa
->spa_spares
.sav_config
= NULL
;
4119 spa_load_spares(spa
);
4121 if (spa
->spa_l2cache
.sav_config
) {
4122 nvlist_free(spa
->spa_l2cache
.sav_config
);
4123 spa
->spa_l2cache
.sav_config
= NULL
;
4124 spa_load_l2cache(spa
);
4127 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4130 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4133 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4134 VDEV_ALLOC_L2CACHE
);
4135 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4138 spa_configfile_set(spa
, props
, B_FALSE
);
4140 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4141 (error
= spa_prop_set(spa
, props
)))) {
4143 spa_deactivate(spa
);
4145 mutex_exit(&spa_namespace_lock
);
4149 spa_async_resume(spa
);
4152 * Override any spares and level 2 cache devices as specified by
4153 * the user, as these may have correct device names/devids, etc.
4155 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4156 &spares
, &nspares
) == 0) {
4157 if (spa
->spa_spares
.sav_config
)
4158 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4159 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4161 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4162 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4163 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4164 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4165 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4166 spa_load_spares(spa
);
4167 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4168 spa
->spa_spares
.sav_sync
= B_TRUE
;
4170 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4171 &l2cache
, &nl2cache
) == 0) {
4172 if (spa
->spa_l2cache
.sav_config
)
4173 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4174 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4176 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4177 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4178 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4179 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4180 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4181 spa_load_l2cache(spa
);
4182 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4183 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4187 * Check for any removed devices.
4189 if (spa
->spa_autoreplace
) {
4190 spa_aux_check_removed(&spa
->spa_spares
);
4191 spa_aux_check_removed(&spa
->spa_l2cache
);
4194 if (spa_writeable(spa
)) {
4196 * Update the config cache to include the newly-imported pool.
4198 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4202 * It's possible that the pool was expanded while it was exported.
4203 * We kick off an async task to handle this for us.
4205 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4207 mutex_exit(&spa_namespace_lock
);
4208 spa_history_log_version(spa
, "import");
4211 zvol_create_minors(pool
);
4218 spa_tryimport(nvlist_t
*tryconfig
)
4220 nvlist_t
*config
= NULL
;
4226 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4229 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4233 * Create and initialize the spa structure.
4235 mutex_enter(&spa_namespace_lock
);
4236 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4237 spa_activate(spa
, FREAD
);
4240 * Pass off the heavy lifting to spa_load().
4241 * Pass TRUE for mosconfig because the user-supplied config
4242 * is actually the one to trust when doing an import.
4244 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4247 * If 'tryconfig' was at least parsable, return the current config.
4249 if (spa
->spa_root_vdev
!= NULL
) {
4250 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4251 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4253 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4255 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4256 spa
->spa_uberblock
.ub_timestamp
) == 0);
4257 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4258 spa
->spa_load_info
) == 0);
4259 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4260 spa
->spa_errata
) == 0);
4263 * If the bootfs property exists on this pool then we
4264 * copy it out so that external consumers can tell which
4265 * pools are bootable.
4267 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4268 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4271 * We have to play games with the name since the
4272 * pool was opened as TRYIMPORT_NAME.
4274 if (dsl_dsobj_to_dsname(spa_name(spa
),
4275 spa
->spa_bootfs
, tmpname
) == 0) {
4279 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4281 cp
= strchr(tmpname
, '/');
4283 (void) strlcpy(dsname
, tmpname
,
4286 (void) snprintf(dsname
, MAXPATHLEN
,
4287 "%s/%s", poolname
, ++cp
);
4289 VERIFY(nvlist_add_string(config
,
4290 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4291 kmem_free(dsname
, MAXPATHLEN
);
4293 kmem_free(tmpname
, MAXPATHLEN
);
4297 * Add the list of hot spares and level 2 cache devices.
4299 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4300 spa_add_spares(spa
, config
);
4301 spa_add_l2cache(spa
, config
);
4302 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4306 spa_deactivate(spa
);
4308 mutex_exit(&spa_namespace_lock
);
4314 * Pool export/destroy
4316 * The act of destroying or exporting a pool is very simple. We make sure there
4317 * is no more pending I/O and any references to the pool are gone. Then, we
4318 * update the pool state and sync all the labels to disk, removing the
4319 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4320 * we don't sync the labels or remove the configuration cache.
4323 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4324 boolean_t force
, boolean_t hardforce
)
4331 if (!(spa_mode_global
& FWRITE
))
4332 return (SET_ERROR(EROFS
));
4334 mutex_enter(&spa_namespace_lock
);
4335 if ((spa
= spa_lookup(pool
)) == NULL
) {
4336 mutex_exit(&spa_namespace_lock
);
4337 return (SET_ERROR(ENOENT
));
4341 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4342 * reacquire the namespace lock, and see if we can export.
4344 spa_open_ref(spa
, FTAG
);
4345 mutex_exit(&spa_namespace_lock
);
4346 spa_async_suspend(spa
);
4347 mutex_enter(&spa_namespace_lock
);
4348 spa_close(spa
, FTAG
);
4350 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4353 * The pool will be in core if it's openable, in which case we can
4354 * modify its state. Objsets may be open only because they're dirty,
4355 * so we have to force it to sync before checking spa_refcnt.
4357 if (spa
->spa_sync_on
) {
4358 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4359 spa_evicting_os_wait(spa
);
4363 * A pool cannot be exported or destroyed if there are active
4364 * references. If we are resetting a pool, allow references by
4365 * fault injection handlers.
4367 if (!spa_refcount_zero(spa
) ||
4368 (spa
->spa_inject_ref
!= 0 &&
4369 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4370 spa_async_resume(spa
);
4371 mutex_exit(&spa_namespace_lock
);
4372 return (SET_ERROR(EBUSY
));
4375 if (spa
->spa_sync_on
) {
4377 * A pool cannot be exported if it has an active shared spare.
4378 * This is to prevent other pools stealing the active spare
4379 * from an exported pool. At user's own will, such pool can
4380 * be forcedly exported.
4382 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4383 spa_has_active_shared_spare(spa
)) {
4384 spa_async_resume(spa
);
4385 mutex_exit(&spa_namespace_lock
);
4386 return (SET_ERROR(EXDEV
));
4390 * We want this to be reflected on every label,
4391 * so mark them all dirty. spa_unload() will do the
4392 * final sync that pushes these changes out.
4394 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4395 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4396 spa
->spa_state
= new_state
;
4397 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4399 vdev_config_dirty(spa
->spa_root_vdev
);
4400 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4405 spa_event_notify(spa
, NULL
, FM_EREPORT_ZFS_POOL_DESTROY
);
4407 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4409 spa_deactivate(spa
);
4412 if (oldconfig
&& spa
->spa_config
)
4413 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4415 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4417 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4420 mutex_exit(&spa_namespace_lock
);
4426 * Destroy a storage pool.
4429 spa_destroy(char *pool
)
4431 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4436 * Export a storage pool.
4439 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4440 boolean_t hardforce
)
4442 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4447 * Similar to spa_export(), this unloads the spa_t without actually removing it
4448 * from the namespace in any way.
4451 spa_reset(char *pool
)
4453 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4458 * ==========================================================================
4459 * Device manipulation
4460 * ==========================================================================
4464 * Add a device to a storage pool.
4467 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4471 vdev_t
*rvd
= spa
->spa_root_vdev
;
4473 nvlist_t
**spares
, **l2cache
;
4474 uint_t nspares
, nl2cache
;
4477 ASSERT(spa_writeable(spa
));
4479 txg
= spa_vdev_enter(spa
);
4481 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4482 VDEV_ALLOC_ADD
)) != 0)
4483 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4485 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4487 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4491 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4495 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4496 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4498 if (vd
->vdev_children
!= 0 &&
4499 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4500 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4503 * We must validate the spares and l2cache devices after checking the
4504 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4506 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4507 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4510 * Transfer each new top-level vdev from vd to rvd.
4512 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4515 * Set the vdev id to the first hole, if one exists.
4517 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4518 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4519 vdev_free(rvd
->vdev_child
[id
]);
4523 tvd
= vd
->vdev_child
[c
];
4524 vdev_remove_child(vd
, tvd
);
4526 vdev_add_child(rvd
, tvd
);
4527 vdev_config_dirty(tvd
);
4531 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4532 ZPOOL_CONFIG_SPARES
);
4533 spa_load_spares(spa
);
4534 spa
->spa_spares
.sav_sync
= B_TRUE
;
4537 if (nl2cache
!= 0) {
4538 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4539 ZPOOL_CONFIG_L2CACHE
);
4540 spa_load_l2cache(spa
);
4541 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4545 * We have to be careful when adding new vdevs to an existing pool.
4546 * If other threads start allocating from these vdevs before we
4547 * sync the config cache, and we lose power, then upon reboot we may
4548 * fail to open the pool because there are DVAs that the config cache
4549 * can't translate. Therefore, we first add the vdevs without
4550 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4551 * and then let spa_config_update() initialize the new metaslabs.
4553 * spa_load() checks for added-but-not-initialized vdevs, so that
4554 * if we lose power at any point in this sequence, the remaining
4555 * steps will be completed the next time we load the pool.
4557 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4559 mutex_enter(&spa_namespace_lock
);
4560 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4561 mutex_exit(&spa_namespace_lock
);
4567 * Attach a device to a mirror. The arguments are the path to any device
4568 * in the mirror, and the nvroot for the new device. If the path specifies
4569 * a device that is not mirrored, we automatically insert the mirror vdev.
4571 * If 'replacing' is specified, the new device is intended to replace the
4572 * existing device; in this case the two devices are made into their own
4573 * mirror using the 'replacing' vdev, which is functionally identical to
4574 * the mirror vdev (it actually reuses all the same ops) but has a few
4575 * extra rules: you can't attach to it after it's been created, and upon
4576 * completion of resilvering, the first disk (the one being replaced)
4577 * is automatically detached.
4580 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4582 uint64_t txg
, dtl_max_txg
;
4583 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4585 char *oldvdpath
, *newvdpath
;
4588 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4590 ASSERT(spa_writeable(spa
));
4592 txg
= spa_vdev_enter(spa
);
4594 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4597 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4599 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4600 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4602 pvd
= oldvd
->vdev_parent
;
4604 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4605 VDEV_ALLOC_ATTACH
)) != 0)
4606 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4608 if (newrootvd
->vdev_children
!= 1)
4609 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4611 newvd
= newrootvd
->vdev_child
[0];
4613 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4614 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4616 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4617 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4620 * Spares can't replace logs
4622 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4623 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4627 * For attach, the only allowable parent is a mirror or the root
4630 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4631 pvd
->vdev_ops
!= &vdev_root_ops
)
4632 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4634 pvops
= &vdev_mirror_ops
;
4637 * Active hot spares can only be replaced by inactive hot
4640 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4641 oldvd
->vdev_isspare
&&
4642 !spa_has_spare(spa
, newvd
->vdev_guid
))
4643 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4646 * If the source is a hot spare, and the parent isn't already a
4647 * spare, then we want to create a new hot spare. Otherwise, we
4648 * want to create a replacing vdev. The user is not allowed to
4649 * attach to a spared vdev child unless the 'isspare' state is
4650 * the same (spare replaces spare, non-spare replaces
4653 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4654 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4655 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4656 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4657 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4658 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4661 if (newvd
->vdev_isspare
)
4662 pvops
= &vdev_spare_ops
;
4664 pvops
= &vdev_replacing_ops
;
4668 * Make sure the new device is big enough.
4670 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4671 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4674 * The new device cannot have a higher alignment requirement
4675 * than the top-level vdev.
4677 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4678 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4681 * If this is an in-place replacement, update oldvd's path and devid
4682 * to make it distinguishable from newvd, and unopenable from now on.
4684 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4685 spa_strfree(oldvd
->vdev_path
);
4686 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4688 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4689 newvd
->vdev_path
, "old");
4690 if (oldvd
->vdev_devid
!= NULL
) {
4691 spa_strfree(oldvd
->vdev_devid
);
4692 oldvd
->vdev_devid
= NULL
;
4696 /* mark the device being resilvered */
4697 newvd
->vdev_resilver_txg
= txg
;
4700 * If the parent is not a mirror, or if we're replacing, insert the new
4701 * mirror/replacing/spare vdev above oldvd.
4703 if (pvd
->vdev_ops
!= pvops
)
4704 pvd
= vdev_add_parent(oldvd
, pvops
);
4706 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4707 ASSERT(pvd
->vdev_ops
== pvops
);
4708 ASSERT(oldvd
->vdev_parent
== pvd
);
4711 * Extract the new device from its root and add it to pvd.
4713 vdev_remove_child(newrootvd
, newvd
);
4714 newvd
->vdev_id
= pvd
->vdev_children
;
4715 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4716 vdev_add_child(pvd
, newvd
);
4718 tvd
= newvd
->vdev_top
;
4719 ASSERT(pvd
->vdev_top
== tvd
);
4720 ASSERT(tvd
->vdev_parent
== rvd
);
4722 vdev_config_dirty(tvd
);
4725 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4726 * for any dmu_sync-ed blocks. It will propagate upward when
4727 * spa_vdev_exit() calls vdev_dtl_reassess().
4729 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4731 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4732 dtl_max_txg
- TXG_INITIAL
);
4734 if (newvd
->vdev_isspare
) {
4735 spa_spare_activate(newvd
);
4736 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_DEVICE_SPARE
);
4739 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4740 newvdpath
= spa_strdup(newvd
->vdev_path
);
4741 newvd_isspare
= newvd
->vdev_isspare
;
4744 * Mark newvd's DTL dirty in this txg.
4746 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4749 * Schedule the resilver to restart in the future. We do this to
4750 * ensure that dmu_sync-ed blocks have been stitched into the
4751 * respective datasets.
4753 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4758 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4760 spa_history_log_internal(spa
, "vdev attach", NULL
,
4761 "%s vdev=%s %s vdev=%s",
4762 replacing
&& newvd_isspare
? "spare in" :
4763 replacing
? "replace" : "attach", newvdpath
,
4764 replacing
? "for" : "to", oldvdpath
);
4766 spa_strfree(oldvdpath
);
4767 spa_strfree(newvdpath
);
4769 if (spa
->spa_bootfs
)
4770 spa_event_notify(spa
, newvd
, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH
);
4776 * Detach a device from a mirror or replacing vdev.
4778 * If 'replace_done' is specified, only detach if the parent
4779 * is a replacing vdev.
4782 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
4786 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
4787 boolean_t unspare
= B_FALSE
;
4788 uint64_t unspare_guid
= 0;
4791 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4792 ASSERT(spa_writeable(spa
));
4794 txg
= spa_vdev_enter(spa
);
4796 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4799 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4801 if (!vd
->vdev_ops
->vdev_op_leaf
)
4802 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4804 pvd
= vd
->vdev_parent
;
4807 * If the parent/child relationship is not as expected, don't do it.
4808 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4809 * vdev that's replacing B with C. The user's intent in replacing
4810 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4811 * the replace by detaching C, the expected behavior is to end up
4812 * M(A,B). But suppose that right after deciding to detach C,
4813 * the replacement of B completes. We would have M(A,C), and then
4814 * ask to detach C, which would leave us with just A -- not what
4815 * the user wanted. To prevent this, we make sure that the
4816 * parent/child relationship hasn't changed -- in this example,
4817 * that C's parent is still the replacing vdev R.
4819 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
4820 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4823 * Only 'replacing' or 'spare' vdevs can be replaced.
4825 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4826 pvd
->vdev_ops
!= &vdev_spare_ops
)
4827 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4829 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
4830 spa_version(spa
) >= SPA_VERSION_SPARES
);
4833 * Only mirror, replacing, and spare vdevs support detach.
4835 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
4836 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4837 pvd
->vdev_ops
!= &vdev_spare_ops
)
4838 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4841 * If this device has the only valid copy of some data,
4842 * we cannot safely detach it.
4844 if (vdev_dtl_required(vd
))
4845 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
4847 ASSERT(pvd
->vdev_children
>= 2);
4850 * If we are detaching the second disk from a replacing vdev, then
4851 * check to see if we changed the original vdev's path to have "/old"
4852 * at the end in spa_vdev_attach(). If so, undo that change now.
4854 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
4855 vd
->vdev_path
!= NULL
) {
4856 size_t len
= strlen(vd
->vdev_path
);
4858 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
4859 cvd
= pvd
->vdev_child
[c
];
4861 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
4864 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
4865 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
4866 spa_strfree(cvd
->vdev_path
);
4867 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
4874 * If we are detaching the original disk from a spare, then it implies
4875 * that the spare should become a real disk, and be removed from the
4876 * active spare list for the pool.
4878 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4880 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
4884 * Erase the disk labels so the disk can be used for other things.
4885 * This must be done after all other error cases are handled,
4886 * but before we disembowel vd (so we can still do I/O to it).
4887 * But if we can't do it, don't treat the error as fatal --
4888 * it may be that the unwritability of the disk is the reason
4889 * it's being detached!
4891 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
4894 * Remove vd from its parent and compact the parent's children.
4896 vdev_remove_child(pvd
, vd
);
4897 vdev_compact_children(pvd
);
4900 * Remember one of the remaining children so we can get tvd below.
4902 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
4905 * If we need to remove the remaining child from the list of hot spares,
4906 * do it now, marking the vdev as no longer a spare in the process.
4907 * We must do this before vdev_remove_parent(), because that can
4908 * change the GUID if it creates a new toplevel GUID. For a similar
4909 * reason, we must remove the spare now, in the same txg as the detach;
4910 * otherwise someone could attach a new sibling, change the GUID, and
4911 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4914 ASSERT(cvd
->vdev_isspare
);
4915 spa_spare_remove(cvd
);
4916 unspare_guid
= cvd
->vdev_guid
;
4917 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
4918 cvd
->vdev_unspare
= B_TRUE
;
4922 * If the parent mirror/replacing vdev only has one child,
4923 * the parent is no longer needed. Remove it from the tree.
4925 if (pvd
->vdev_children
== 1) {
4926 if (pvd
->vdev_ops
== &vdev_spare_ops
)
4927 cvd
->vdev_unspare
= B_FALSE
;
4928 vdev_remove_parent(cvd
);
4933 * We don't set tvd until now because the parent we just removed
4934 * may have been the previous top-level vdev.
4936 tvd
= cvd
->vdev_top
;
4937 ASSERT(tvd
->vdev_parent
== rvd
);
4940 * Reevaluate the parent vdev state.
4942 vdev_propagate_state(cvd
);
4945 * If the 'autoexpand' property is set on the pool then automatically
4946 * try to expand the size of the pool. For example if the device we
4947 * just detached was smaller than the others, it may be possible to
4948 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4949 * first so that we can obtain the updated sizes of the leaf vdevs.
4951 if (spa
->spa_autoexpand
) {
4953 vdev_expand(tvd
, txg
);
4956 vdev_config_dirty(tvd
);
4959 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4960 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4961 * But first make sure we're not on any *other* txg's DTL list, to
4962 * prevent vd from being accessed after it's freed.
4964 vdpath
= spa_strdup(vd
->vdev_path
);
4965 for (t
= 0; t
< TXG_SIZE
; t
++)
4966 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
4967 vd
->vdev_detached
= B_TRUE
;
4968 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
4970 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_REMOVE
);
4972 /* hang on to the spa before we release the lock */
4973 spa_open_ref(spa
, FTAG
);
4975 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
4977 spa_history_log_internal(spa
, "detach", NULL
,
4979 spa_strfree(vdpath
);
4982 * If this was the removal of the original device in a hot spare vdev,
4983 * then we want to go through and remove the device from the hot spare
4984 * list of every other pool.
4987 spa_t
*altspa
= NULL
;
4989 mutex_enter(&spa_namespace_lock
);
4990 while ((altspa
= spa_next(altspa
)) != NULL
) {
4991 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
4995 spa_open_ref(altspa
, FTAG
);
4996 mutex_exit(&spa_namespace_lock
);
4997 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
4998 mutex_enter(&spa_namespace_lock
);
4999 spa_close(altspa
, FTAG
);
5001 mutex_exit(&spa_namespace_lock
);
5003 /* search the rest of the vdevs for spares to remove */
5004 spa_vdev_resilver_done(spa
);
5007 /* all done with the spa; OK to release */
5008 mutex_enter(&spa_namespace_lock
);
5009 spa_close(spa
, FTAG
);
5010 mutex_exit(&spa_namespace_lock
);
5016 * Split a set of devices from their mirrors, and create a new pool from them.
5019 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5020 nvlist_t
*props
, boolean_t exp
)
5023 uint64_t txg
, *glist
;
5025 uint_t c
, children
, lastlog
;
5026 nvlist_t
**child
, *nvl
, *tmp
;
5028 char *altroot
= NULL
;
5029 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5030 boolean_t activate_slog
;
5032 ASSERT(spa_writeable(spa
));
5034 txg
= spa_vdev_enter(spa
);
5036 /* clear the log and flush everything up to now */
5037 activate_slog
= spa_passivate_log(spa
);
5038 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5039 error
= spa_offline_log(spa
);
5040 txg
= spa_vdev_config_enter(spa
);
5043 spa_activate_log(spa
);
5046 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5048 /* check new spa name before going any further */
5049 if (spa_lookup(newname
) != NULL
)
5050 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5053 * scan through all the children to ensure they're all mirrors
5055 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5056 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5058 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5060 /* first, check to ensure we've got the right child count */
5061 rvd
= spa
->spa_root_vdev
;
5063 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5064 vdev_t
*vd
= rvd
->vdev_child
[c
];
5066 /* don't count the holes & logs as children */
5067 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5075 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5076 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5078 /* next, ensure no spare or cache devices are part of the split */
5079 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5080 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5081 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5083 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5084 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5086 /* then, loop over each vdev and validate it */
5087 for (c
= 0; c
< children
; c
++) {
5088 uint64_t is_hole
= 0;
5090 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5094 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5095 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5098 error
= SET_ERROR(EINVAL
);
5103 /* which disk is going to be split? */
5104 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5106 error
= SET_ERROR(EINVAL
);
5110 /* look it up in the spa */
5111 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5112 if (vml
[c
] == NULL
) {
5113 error
= SET_ERROR(ENODEV
);
5117 /* make sure there's nothing stopping the split */
5118 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5119 vml
[c
]->vdev_islog
||
5120 vml
[c
]->vdev_ishole
||
5121 vml
[c
]->vdev_isspare
||
5122 vml
[c
]->vdev_isl2cache
||
5123 !vdev_writeable(vml
[c
]) ||
5124 vml
[c
]->vdev_children
!= 0 ||
5125 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5126 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5127 error
= SET_ERROR(EINVAL
);
5131 if (vdev_dtl_required(vml
[c
])) {
5132 error
= SET_ERROR(EBUSY
);
5136 /* we need certain info from the top level */
5137 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5138 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5139 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5140 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5141 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5142 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5143 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5144 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5148 kmem_free(vml
, children
* sizeof (vdev_t
*));
5149 kmem_free(glist
, children
* sizeof (uint64_t));
5150 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5153 /* stop writers from using the disks */
5154 for (c
= 0; c
< children
; c
++) {
5156 vml
[c
]->vdev_offline
= B_TRUE
;
5158 vdev_reopen(spa
->spa_root_vdev
);
5161 * Temporarily record the splitting vdevs in the spa config. This
5162 * will disappear once the config is regenerated.
5164 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5165 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5166 glist
, children
) == 0);
5167 kmem_free(glist
, children
* sizeof (uint64_t));
5169 mutex_enter(&spa
->spa_props_lock
);
5170 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5172 mutex_exit(&spa
->spa_props_lock
);
5173 spa
->spa_config_splitting
= nvl
;
5174 vdev_config_dirty(spa
->spa_root_vdev
);
5176 /* configure and create the new pool */
5177 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5178 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5179 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5180 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5181 spa_version(spa
)) == 0);
5182 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5183 spa
->spa_config_txg
) == 0);
5184 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5185 spa_generate_guid(NULL
)) == 0);
5186 (void) nvlist_lookup_string(props
,
5187 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5189 /* add the new pool to the namespace */
5190 newspa
= spa_add(newname
, config
, altroot
);
5191 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5192 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5194 /* release the spa config lock, retaining the namespace lock */
5195 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5197 if (zio_injection_enabled
)
5198 zio_handle_panic_injection(spa
, FTAG
, 1);
5200 spa_activate(newspa
, spa_mode_global
);
5201 spa_async_suspend(newspa
);
5203 /* create the new pool from the disks of the original pool */
5204 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5208 /* if that worked, generate a real config for the new pool */
5209 if (newspa
->spa_root_vdev
!= NULL
) {
5210 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5211 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5212 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5213 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5214 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5219 if (props
!= NULL
) {
5220 spa_configfile_set(newspa
, props
, B_FALSE
);
5221 error
= spa_prop_set(newspa
, props
);
5226 /* flush everything */
5227 txg
= spa_vdev_config_enter(newspa
);
5228 vdev_config_dirty(newspa
->spa_root_vdev
);
5229 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5231 if (zio_injection_enabled
)
5232 zio_handle_panic_injection(spa
, FTAG
, 2);
5234 spa_async_resume(newspa
);
5236 /* finally, update the original pool's config */
5237 txg
= spa_vdev_config_enter(spa
);
5238 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5239 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5242 for (c
= 0; c
< children
; c
++) {
5243 if (vml
[c
] != NULL
) {
5246 spa_history_log_internal(spa
, "detach", tx
,
5247 "vdev=%s", vml
[c
]->vdev_path
);
5251 vdev_config_dirty(spa
->spa_root_vdev
);
5252 spa
->spa_config_splitting
= NULL
;
5256 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5258 if (zio_injection_enabled
)
5259 zio_handle_panic_injection(spa
, FTAG
, 3);
5261 /* split is complete; log a history record */
5262 spa_history_log_internal(newspa
, "split", NULL
,
5263 "from pool %s", spa_name(spa
));
5265 kmem_free(vml
, children
* sizeof (vdev_t
*));
5267 /* if we're not going to mount the filesystems in userland, export */
5269 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5276 spa_deactivate(newspa
);
5279 txg
= spa_vdev_config_enter(spa
);
5281 /* re-online all offlined disks */
5282 for (c
= 0; c
< children
; c
++) {
5284 vml
[c
]->vdev_offline
= B_FALSE
;
5286 vdev_reopen(spa
->spa_root_vdev
);
5288 nvlist_free(spa
->spa_config_splitting
);
5289 spa
->spa_config_splitting
= NULL
;
5290 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5292 kmem_free(vml
, children
* sizeof (vdev_t
*));
5297 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5301 for (i
= 0; i
< count
; i
++) {
5304 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5307 if (guid
== target_guid
)
5315 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5316 nvlist_t
*dev_to_remove
)
5318 nvlist_t
**newdev
= NULL
;
5322 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5324 for (i
= 0, j
= 0; i
< count
; i
++) {
5325 if (dev
[i
] == dev_to_remove
)
5327 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5330 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5331 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5333 for (i
= 0; i
< count
- 1; i
++)
5334 nvlist_free(newdev
[i
]);
5337 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5341 * Evacuate the device.
5344 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5349 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5350 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5351 ASSERT(vd
== vd
->vdev_top
);
5354 * Evacuate the device. We don't hold the config lock as writer
5355 * since we need to do I/O but we do keep the
5356 * spa_namespace_lock held. Once this completes the device
5357 * should no longer have any blocks allocated on it.
5359 if (vd
->vdev_islog
) {
5360 if (vd
->vdev_stat
.vs_alloc
!= 0)
5361 error
= spa_offline_log(spa
);
5363 error
= SET_ERROR(ENOTSUP
);
5370 * The evacuation succeeded. Remove any remaining MOS metadata
5371 * associated with this vdev, and wait for these changes to sync.
5373 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5374 txg
= spa_vdev_config_enter(spa
);
5375 vd
->vdev_removing
= B_TRUE
;
5376 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5377 vdev_config_dirty(vd
);
5378 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5384 * Complete the removal by cleaning up the namespace.
5387 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5389 vdev_t
*rvd
= spa
->spa_root_vdev
;
5390 uint64_t id
= vd
->vdev_id
;
5391 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5393 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5394 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5395 ASSERT(vd
== vd
->vdev_top
);
5398 * Only remove any devices which are empty.
5400 if (vd
->vdev_stat
.vs_alloc
!= 0)
5403 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5405 if (list_link_active(&vd
->vdev_state_dirty_node
))
5406 vdev_state_clean(vd
);
5407 if (list_link_active(&vd
->vdev_config_dirty_node
))
5408 vdev_config_clean(vd
);
5413 vdev_compact_children(rvd
);
5415 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5416 vdev_add_child(rvd
, vd
);
5418 vdev_config_dirty(rvd
);
5421 * Reassess the health of our root vdev.
5427 * Remove a device from the pool -
5429 * Removing a device from the vdev namespace requires several steps
5430 * and can take a significant amount of time. As a result we use
5431 * the spa_vdev_config_[enter/exit] functions which allow us to
5432 * grab and release the spa_config_lock while still holding the namespace
5433 * lock. During each step the configuration is synced out.
5435 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5439 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5442 metaslab_group_t
*mg
;
5443 nvlist_t
**spares
, **l2cache
, *nv
;
5445 uint_t nspares
, nl2cache
;
5447 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5449 ASSERT(spa_writeable(spa
));
5452 txg
= spa_vdev_enter(spa
);
5454 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5456 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5457 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5458 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5459 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5461 * Only remove the hot spare if it's not currently in use
5464 if (vd
== NULL
|| unspare
) {
5465 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5466 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5467 spa_load_spares(spa
);
5468 spa
->spa_spares
.sav_sync
= B_TRUE
;
5470 error
= SET_ERROR(EBUSY
);
5472 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5473 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5474 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5475 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5477 * Cache devices can always be removed.
5479 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5480 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5481 spa_load_l2cache(spa
);
5482 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5483 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5485 ASSERT(vd
== vd
->vdev_top
);
5490 * Stop allocating from this vdev.
5492 metaslab_group_passivate(mg
);
5495 * Wait for the youngest allocations and frees to sync,
5496 * and then wait for the deferral of those frees to finish.
5498 spa_vdev_config_exit(spa
, NULL
,
5499 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5502 * Attempt to evacuate the vdev.
5504 error
= spa_vdev_remove_evacuate(spa
, vd
);
5506 txg
= spa_vdev_config_enter(spa
);
5509 * If we couldn't evacuate the vdev, unwind.
5512 metaslab_group_activate(mg
);
5513 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5517 * Clean up the vdev namespace.
5519 spa_vdev_remove_from_namespace(spa
, vd
);
5521 } else if (vd
!= NULL
) {
5523 * Normal vdevs cannot be removed (yet).
5525 error
= SET_ERROR(ENOTSUP
);
5528 * There is no vdev of any kind with the specified guid.
5530 error
= SET_ERROR(ENOENT
);
5534 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5540 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5541 * currently spared, so we can detach it.
5544 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5546 vdev_t
*newvd
, *oldvd
;
5549 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5550 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5556 * Check for a completed replacement. We always consider the first
5557 * vdev in the list to be the oldest vdev, and the last one to be
5558 * the newest (see spa_vdev_attach() for how that works). In
5559 * the case where the newest vdev is faulted, we will not automatically
5560 * remove it after a resilver completes. This is OK as it will require
5561 * user intervention to determine which disk the admin wishes to keep.
5563 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5564 ASSERT(vd
->vdev_children
> 1);
5566 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5567 oldvd
= vd
->vdev_child
[0];
5569 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5570 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5571 !vdev_dtl_required(oldvd
))
5576 * Check for a completed resilver with the 'unspare' flag set.
5578 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5579 vdev_t
*first
= vd
->vdev_child
[0];
5580 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5582 if (last
->vdev_unspare
) {
5585 } else if (first
->vdev_unspare
) {
5592 if (oldvd
!= NULL
&&
5593 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5594 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5595 !vdev_dtl_required(oldvd
))
5599 * If there are more than two spares attached to a disk,
5600 * and those spares are not required, then we want to
5601 * attempt to free them up now so that they can be used
5602 * by other pools. Once we're back down to a single
5603 * disk+spare, we stop removing them.
5605 if (vd
->vdev_children
> 2) {
5606 newvd
= vd
->vdev_child
[1];
5608 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5609 vdev_dtl_empty(last
, DTL_MISSING
) &&
5610 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5611 !vdev_dtl_required(newvd
))
5620 spa_vdev_resilver_done(spa_t
*spa
)
5622 vdev_t
*vd
, *pvd
, *ppvd
;
5623 uint64_t guid
, sguid
, pguid
, ppguid
;
5625 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5627 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5628 pvd
= vd
->vdev_parent
;
5629 ppvd
= pvd
->vdev_parent
;
5630 guid
= vd
->vdev_guid
;
5631 pguid
= pvd
->vdev_guid
;
5632 ppguid
= ppvd
->vdev_guid
;
5635 * If we have just finished replacing a hot spared device, then
5636 * we need to detach the parent's first child (the original hot
5639 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5640 ppvd
->vdev_children
== 2) {
5641 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5642 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5644 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5646 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5647 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5649 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5651 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5654 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5658 * Update the stored path or FRU for this vdev.
5661 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5665 boolean_t sync
= B_FALSE
;
5667 ASSERT(spa_writeable(spa
));
5669 spa_vdev_state_enter(spa
, SCL_ALL
);
5671 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5672 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5674 if (!vd
->vdev_ops
->vdev_op_leaf
)
5675 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5678 if (strcmp(value
, vd
->vdev_path
) != 0) {
5679 spa_strfree(vd
->vdev_path
);
5680 vd
->vdev_path
= spa_strdup(value
);
5684 if (vd
->vdev_fru
== NULL
) {
5685 vd
->vdev_fru
= spa_strdup(value
);
5687 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5688 spa_strfree(vd
->vdev_fru
);
5689 vd
->vdev_fru
= spa_strdup(value
);
5694 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5698 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5700 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5704 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5706 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5710 * ==========================================================================
5712 * ==========================================================================
5716 spa_scan_stop(spa_t
*spa
)
5718 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5719 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5720 return (SET_ERROR(EBUSY
));
5721 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5725 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5727 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5729 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5730 return (SET_ERROR(ENOTSUP
));
5733 * If a resilver was requested, but there is no DTL on a
5734 * writeable leaf device, we have nothing to do.
5736 if (func
== POOL_SCAN_RESILVER
&&
5737 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5738 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
5742 return (dsl_scan(spa
->spa_dsl_pool
, func
));
5746 * ==========================================================================
5747 * SPA async task processing
5748 * ==========================================================================
5752 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
5756 if (vd
->vdev_remove_wanted
) {
5757 vd
->vdev_remove_wanted
= B_FALSE
;
5758 vd
->vdev_delayed_close
= B_FALSE
;
5759 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
5762 * We want to clear the stats, but we don't want to do a full
5763 * vdev_clear() as that will cause us to throw away
5764 * degraded/faulted state as well as attempt to reopen the
5765 * device, all of which is a waste.
5767 vd
->vdev_stat
.vs_read_errors
= 0;
5768 vd
->vdev_stat
.vs_write_errors
= 0;
5769 vd
->vdev_stat
.vs_checksum_errors
= 0;
5771 vdev_state_dirty(vd
->vdev_top
);
5774 for (c
= 0; c
< vd
->vdev_children
; c
++)
5775 spa_async_remove(spa
, vd
->vdev_child
[c
]);
5779 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
5783 if (vd
->vdev_probe_wanted
) {
5784 vd
->vdev_probe_wanted
= B_FALSE
;
5785 vdev_reopen(vd
); /* vdev_open() does the actual probe */
5788 for (c
= 0; c
< vd
->vdev_children
; c
++)
5789 spa_async_probe(spa
, vd
->vdev_child
[c
]);
5793 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
5797 if (!spa
->spa_autoexpand
)
5800 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5801 vdev_t
*cvd
= vd
->vdev_child
[c
];
5802 spa_async_autoexpand(spa
, cvd
);
5805 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
5808 spa_event_notify(vd
->vdev_spa
, vd
, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND
);
5812 spa_async_thread(spa_t
*spa
)
5816 ASSERT(spa
->spa_sync_on
);
5818 mutex_enter(&spa
->spa_async_lock
);
5819 tasks
= spa
->spa_async_tasks
;
5820 spa
->spa_async_tasks
= 0;
5821 mutex_exit(&spa
->spa_async_lock
);
5824 * See if the config needs to be updated.
5826 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
5827 uint64_t old_space
, new_space
;
5829 mutex_enter(&spa_namespace_lock
);
5830 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
5831 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5832 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
5833 mutex_exit(&spa_namespace_lock
);
5836 * If the pool grew as a result of the config update,
5837 * then log an internal history event.
5839 if (new_space
!= old_space
) {
5840 spa_history_log_internal(spa
, "vdev online", NULL
,
5841 "pool '%s' size: %llu(+%llu)",
5842 spa_name(spa
), new_space
, new_space
- old_space
);
5847 * See if any devices need to be marked REMOVED.
5849 if (tasks
& SPA_ASYNC_REMOVE
) {
5850 spa_vdev_state_enter(spa
, SCL_NONE
);
5851 spa_async_remove(spa
, spa
->spa_root_vdev
);
5852 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
5853 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
5854 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
5855 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
5856 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5859 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
5860 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5861 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
5862 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5866 * See if any devices need to be probed.
5868 if (tasks
& SPA_ASYNC_PROBE
) {
5869 spa_vdev_state_enter(spa
, SCL_NONE
);
5870 spa_async_probe(spa
, spa
->spa_root_vdev
);
5871 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5875 * If any devices are done replacing, detach them.
5877 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
5878 spa_vdev_resilver_done(spa
);
5881 * Kick off a resilver.
5883 if (tasks
& SPA_ASYNC_RESILVER
)
5884 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
5887 * Let the world know that we're done.
5889 mutex_enter(&spa
->spa_async_lock
);
5890 spa
->spa_async_thread
= NULL
;
5891 cv_broadcast(&spa
->spa_async_cv
);
5892 mutex_exit(&spa
->spa_async_lock
);
5897 spa_async_suspend(spa_t
*spa
)
5899 mutex_enter(&spa
->spa_async_lock
);
5900 spa
->spa_async_suspended
++;
5901 while (spa
->spa_async_thread
!= NULL
)
5902 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
5903 mutex_exit(&spa
->spa_async_lock
);
5907 spa_async_resume(spa_t
*spa
)
5909 mutex_enter(&spa
->spa_async_lock
);
5910 ASSERT(spa
->spa_async_suspended
!= 0);
5911 spa
->spa_async_suspended
--;
5912 mutex_exit(&spa
->spa_async_lock
);
5916 spa_async_dispatch(spa_t
*spa
)
5918 mutex_enter(&spa
->spa_async_lock
);
5919 if (spa
->spa_async_tasks
&& !spa
->spa_async_suspended
&&
5920 spa
->spa_async_thread
== NULL
&&
5921 rootdir
!= NULL
&& !vn_is_readonly(rootdir
))
5922 spa
->spa_async_thread
= thread_create(NULL
, 0,
5923 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
5924 mutex_exit(&spa
->spa_async_lock
);
5928 spa_async_request(spa_t
*spa
, int task
)
5930 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
5931 mutex_enter(&spa
->spa_async_lock
);
5932 spa
->spa_async_tasks
|= task
;
5933 mutex_exit(&spa
->spa_async_lock
);
5937 * ==========================================================================
5938 * SPA syncing routines
5939 * ==========================================================================
5943 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5946 bpobj_enqueue(bpo
, bp
, tx
);
5951 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
5955 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
5961 * Note: this simple function is not inlined to make it easier to dtrace the
5962 * amount of time spent syncing frees.
5965 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
5967 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5968 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
5969 VERIFY(zio_wait(zio
) == 0);
5973 * Note: this simple function is not inlined to make it easier to dtrace the
5974 * amount of time spent syncing deferred frees.
5977 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
5979 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
5980 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
5981 spa_free_sync_cb
, zio
, tx
), ==, 0);
5982 VERIFY0(zio_wait(zio
));
5986 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
5988 char *packed
= NULL
;
5993 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
5996 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5997 * information. This avoids the dmu_buf_will_dirty() path and
5998 * saves us a pre-read to get data we don't actually care about.
6000 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6001 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6003 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6005 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6007 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6009 vmem_free(packed
, bufsize
);
6011 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6012 dmu_buf_will_dirty(db
, tx
);
6013 *(uint64_t *)db
->db_data
= nvsize
;
6014 dmu_buf_rele(db
, FTAG
);
6018 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6019 const char *config
, const char *entry
)
6029 * Update the MOS nvlist describing the list of available devices.
6030 * spa_validate_aux() will have already made sure this nvlist is
6031 * valid and the vdevs are labeled appropriately.
6033 if (sav
->sav_object
== 0) {
6034 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6035 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6036 sizeof (uint64_t), tx
);
6037 VERIFY(zap_update(spa
->spa_meta_objset
,
6038 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6039 &sav
->sav_object
, tx
) == 0);
6042 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6043 if (sav
->sav_count
== 0) {
6044 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6046 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6047 for (i
= 0; i
< sav
->sav_count
; i
++)
6048 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6049 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6050 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6051 sav
->sav_count
) == 0);
6052 for (i
= 0; i
< sav
->sav_count
; i
++)
6053 nvlist_free(list
[i
]);
6054 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6057 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6058 nvlist_free(nvroot
);
6060 sav
->sav_sync
= B_FALSE
;
6064 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6068 if (list_is_empty(&spa
->spa_config_dirty_list
))
6071 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6073 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6074 dmu_tx_get_txg(tx
), B_FALSE
);
6077 * If we're upgrading the spa version then make sure that
6078 * the config object gets updated with the correct version.
6080 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6081 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6082 spa
->spa_uberblock
.ub_version
);
6084 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6086 if (spa
->spa_config_syncing
)
6087 nvlist_free(spa
->spa_config_syncing
);
6088 spa
->spa_config_syncing
= config
;
6090 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6094 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6096 uint64_t *versionp
= arg
;
6097 uint64_t version
= *versionp
;
6098 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6101 * Setting the version is special cased when first creating the pool.
6103 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6105 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6106 ASSERT(version
>= spa_version(spa
));
6108 spa
->spa_uberblock
.ub_version
= version
;
6109 vdev_config_dirty(spa
->spa_root_vdev
);
6110 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6114 * Set zpool properties.
6117 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6119 nvlist_t
*nvp
= arg
;
6120 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6121 objset_t
*mos
= spa
->spa_meta_objset
;
6122 nvpair_t
*elem
= NULL
;
6124 mutex_enter(&spa
->spa_props_lock
);
6126 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6128 char *strval
, *fname
;
6130 const char *propname
;
6131 zprop_type_t proptype
;
6134 prop
= zpool_name_to_prop(nvpair_name(elem
));
6135 switch ((int)prop
) {
6138 * We checked this earlier in spa_prop_validate().
6140 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6142 fname
= strchr(nvpair_name(elem
), '@') + 1;
6143 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6145 spa_feature_enable(spa
, fid
, tx
);
6146 spa_history_log_internal(spa
, "set", tx
,
6147 "%s=enabled", nvpair_name(elem
));
6150 case ZPOOL_PROP_VERSION
:
6151 intval
= fnvpair_value_uint64(elem
);
6153 * The version is synced seperatly before other
6154 * properties and should be correct by now.
6156 ASSERT3U(spa_version(spa
), >=, intval
);
6159 case ZPOOL_PROP_ALTROOT
:
6161 * 'altroot' is a non-persistent property. It should
6162 * have been set temporarily at creation or import time.
6164 ASSERT(spa
->spa_root
!= NULL
);
6167 case ZPOOL_PROP_READONLY
:
6168 case ZPOOL_PROP_CACHEFILE
:
6170 * 'readonly' and 'cachefile' are also non-persisitent
6174 case ZPOOL_PROP_COMMENT
:
6175 strval
= fnvpair_value_string(elem
);
6176 if (spa
->spa_comment
!= NULL
)
6177 spa_strfree(spa
->spa_comment
);
6178 spa
->spa_comment
= spa_strdup(strval
);
6180 * We need to dirty the configuration on all the vdevs
6181 * so that their labels get updated. It's unnecessary
6182 * to do this for pool creation since the vdev's
6183 * configuratoin has already been dirtied.
6185 if (tx
->tx_txg
!= TXG_INITIAL
)
6186 vdev_config_dirty(spa
->spa_root_vdev
);
6187 spa_history_log_internal(spa
, "set", tx
,
6188 "%s=%s", nvpair_name(elem
), strval
);
6192 * Set pool property values in the poolprops mos object.
6194 if (spa
->spa_pool_props_object
== 0) {
6195 spa
->spa_pool_props_object
=
6196 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6197 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6201 /* normalize the property name */
6202 propname
= zpool_prop_to_name(prop
);
6203 proptype
= zpool_prop_get_type(prop
);
6205 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6206 ASSERT(proptype
== PROP_TYPE_STRING
);
6207 strval
= fnvpair_value_string(elem
);
6208 VERIFY0(zap_update(mos
,
6209 spa
->spa_pool_props_object
, propname
,
6210 1, strlen(strval
) + 1, strval
, tx
));
6211 spa_history_log_internal(spa
, "set", tx
,
6212 "%s=%s", nvpair_name(elem
), strval
);
6213 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6214 intval
= fnvpair_value_uint64(elem
);
6216 if (proptype
== PROP_TYPE_INDEX
) {
6218 VERIFY0(zpool_prop_index_to_string(
6219 prop
, intval
, &unused
));
6221 VERIFY0(zap_update(mos
,
6222 spa
->spa_pool_props_object
, propname
,
6223 8, 1, &intval
, tx
));
6224 spa_history_log_internal(spa
, "set", tx
,
6225 "%s=%lld", nvpair_name(elem
), intval
);
6227 ASSERT(0); /* not allowed */
6231 case ZPOOL_PROP_DELEGATION
:
6232 spa
->spa_delegation
= intval
;
6234 case ZPOOL_PROP_BOOTFS
:
6235 spa
->spa_bootfs
= intval
;
6237 case ZPOOL_PROP_FAILUREMODE
:
6238 spa
->spa_failmode
= intval
;
6240 case ZPOOL_PROP_AUTOEXPAND
:
6241 spa
->spa_autoexpand
= intval
;
6242 if (tx
->tx_txg
!= TXG_INITIAL
)
6243 spa_async_request(spa
,
6244 SPA_ASYNC_AUTOEXPAND
);
6246 case ZPOOL_PROP_DEDUPDITTO
:
6247 spa
->spa_dedup_ditto
= intval
;
6256 mutex_exit(&spa
->spa_props_lock
);
6260 * Perform one-time upgrade on-disk changes. spa_version() does not
6261 * reflect the new version this txg, so there must be no changes this
6262 * txg to anything that the upgrade code depends on after it executes.
6263 * Therefore this must be called after dsl_pool_sync() does the sync
6267 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6269 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6271 ASSERT(spa
->spa_sync_pass
== 1);
6273 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6275 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6276 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6277 dsl_pool_create_origin(dp
, tx
);
6279 /* Keeping the origin open increases spa_minref */
6280 spa
->spa_minref
+= 3;
6283 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6284 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6285 dsl_pool_upgrade_clones(dp
, tx
);
6288 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6289 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6290 dsl_pool_upgrade_dir_clones(dp
, tx
);
6292 /* Keeping the freedir open increases spa_minref */
6293 spa
->spa_minref
+= 3;
6296 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6297 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6298 spa_feature_create_zap_objects(spa
, tx
);
6302 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6303 * when possibility to use lz4 compression for metadata was added
6304 * Old pools that have this feature enabled must be upgraded to have
6305 * this feature active
6307 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6308 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6309 SPA_FEATURE_LZ4_COMPRESS
);
6310 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6311 SPA_FEATURE_LZ4_COMPRESS
);
6313 if (lz4_en
&& !lz4_ac
)
6314 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6316 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6320 * Sync the specified transaction group. New blocks may be dirtied as
6321 * part of the process, so we iterate until it converges.
6324 spa_sync(spa_t
*spa
, uint64_t txg
)
6326 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6327 objset_t
*mos
= spa
->spa_meta_objset
;
6328 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6329 vdev_t
*rvd
= spa
->spa_root_vdev
;
6335 VERIFY(spa_writeable(spa
));
6338 * Lock out configuration changes.
6340 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6342 spa
->spa_syncing_txg
= txg
;
6343 spa
->spa_sync_pass
= 0;
6346 * If there are any pending vdev state changes, convert them
6347 * into config changes that go out with this transaction group.
6349 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6350 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6352 * We need the write lock here because, for aux vdevs,
6353 * calling vdev_config_dirty() modifies sav_config.
6354 * This is ugly and will become unnecessary when we
6355 * eliminate the aux vdev wart by integrating all vdevs
6356 * into the root vdev tree.
6358 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6359 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6360 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6361 vdev_state_clean(vd
);
6362 vdev_config_dirty(vd
);
6364 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6365 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6367 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6369 tx
= dmu_tx_create_assigned(dp
, txg
);
6371 spa
->spa_sync_starttime
= gethrtime();
6372 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6373 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
6374 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6375 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6378 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6379 * set spa_deflate if we have no raid-z vdevs.
6381 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6382 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6385 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6386 vd
= rvd
->vdev_child
[i
];
6387 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6390 if (i
== rvd
->vdev_children
) {
6391 spa
->spa_deflate
= TRUE
;
6392 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6393 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6394 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6399 * Iterate to convergence.
6402 int pass
= ++spa
->spa_sync_pass
;
6404 spa_sync_config_object(spa
, tx
);
6405 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6406 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6407 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6408 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6409 spa_errlog_sync(spa
, txg
);
6410 dsl_pool_sync(dp
, txg
);
6412 if (pass
< zfs_sync_pass_deferred_free
) {
6413 spa_sync_frees(spa
, free_bpl
, tx
);
6416 * We can not defer frees in pass 1, because
6417 * we sync the deferred frees later in pass 1.
6419 ASSERT3U(pass
, >, 1);
6420 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6421 &spa
->spa_deferred_bpobj
, tx
);
6425 dsl_scan_sync(dp
, tx
);
6427 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6431 spa_sync_upgrades(spa
, tx
);
6433 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
6435 * Note: We need to check if the MOS is dirty
6436 * because we could have marked the MOS dirty
6437 * without updating the uberblock (e.g. if we
6438 * have sync tasks but no dirty user data). We
6439 * need to check the uberblock's rootbp because
6440 * it is updated if we have synced out dirty
6441 * data (though in this case the MOS will most
6442 * likely also be dirty due to second order
6443 * effects, we don't want to rely on that here).
6445 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
6446 !dmu_objset_is_dirty(mos
, txg
)) {
6448 * Nothing changed on the first pass,
6449 * therefore this TXG is a no-op. Avoid
6450 * syncing deferred frees, so that we
6451 * can keep this TXG as a no-op.
6453 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
6455 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6456 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
6459 spa_sync_deferred_frees(spa
, tx
);
6462 } while (dmu_objset_is_dirty(mos
, txg
));
6465 * Rewrite the vdev configuration (which includes the uberblock)
6466 * to commit the transaction group.
6468 * If there are no dirty vdevs, we sync the uberblock to a few
6469 * random top-level vdevs that are known to be visible in the
6470 * config cache (see spa_vdev_add() for a complete description).
6471 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6475 * We hold SCL_STATE to prevent vdev open/close/etc.
6476 * while we're attempting to write the vdev labels.
6478 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6480 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6481 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6483 int children
= rvd
->vdev_children
;
6484 int c0
= spa_get_random(children
);
6486 for (c
= 0; c
< children
; c
++) {
6487 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6488 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6490 svd
[svdcount
++] = vd
;
6491 if (svdcount
== SPA_DVAS_PER_BP
)
6494 error
= vdev_config_sync(svd
, svdcount
, txg
, B_FALSE
);
6496 error
= vdev_config_sync(svd
, svdcount
, txg
,
6499 error
= vdev_config_sync(rvd
->vdev_child
,
6500 rvd
->vdev_children
, txg
, B_FALSE
);
6502 error
= vdev_config_sync(rvd
->vdev_child
,
6503 rvd
->vdev_children
, txg
, B_TRUE
);
6507 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6509 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6513 zio_suspend(spa
, NULL
);
6514 zio_resume_wait(spa
);
6518 taskq_cancel_id(system_taskq
, spa
->spa_deadman_tqid
);
6519 spa
->spa_deadman_tqid
= 0;
6522 * Clear the dirty config list.
6524 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6525 vdev_config_clean(vd
);
6528 * Now that the new config has synced transactionally,
6529 * let it become visible to the config cache.
6531 if (spa
->spa_config_syncing
!= NULL
) {
6532 spa_config_set(spa
, spa
->spa_config_syncing
);
6533 spa
->spa_config_txg
= txg
;
6534 spa
->spa_config_syncing
= NULL
;
6537 spa
->spa_ubsync
= spa
->spa_uberblock
;
6539 dsl_pool_sync_done(dp
, txg
);
6542 * Update usable space statistics.
6544 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6545 vdev_sync_done(vd
, txg
);
6547 spa_update_dspace(spa
);
6550 * It had better be the case that we didn't dirty anything
6551 * since vdev_config_sync().
6553 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
6554 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6555 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
6557 spa
->spa_sync_pass
= 0;
6559 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6561 spa_handle_ignored_writes(spa
);
6564 * If any async tasks have been requested, kick them off.
6566 spa_async_dispatch(spa
);
6570 * Sync all pools. We don't want to hold the namespace lock across these
6571 * operations, so we take a reference on the spa_t and drop the lock during the
6575 spa_sync_allpools(void)
6578 mutex_enter(&spa_namespace_lock
);
6579 while ((spa
= spa_next(spa
)) != NULL
) {
6580 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
6581 !spa_writeable(spa
) || spa_suspended(spa
))
6583 spa_open_ref(spa
, FTAG
);
6584 mutex_exit(&spa_namespace_lock
);
6585 txg_wait_synced(spa_get_dsl(spa
), 0);
6586 mutex_enter(&spa_namespace_lock
);
6587 spa_close(spa
, FTAG
);
6589 mutex_exit(&spa_namespace_lock
);
6593 * ==========================================================================
6594 * Miscellaneous routines
6595 * ==========================================================================
6599 * Remove all pools in the system.
6607 * Remove all cached state. All pools should be closed now,
6608 * so every spa in the AVL tree should be unreferenced.
6610 mutex_enter(&spa_namespace_lock
);
6611 while ((spa
= spa_next(NULL
)) != NULL
) {
6613 * Stop async tasks. The async thread may need to detach
6614 * a device that's been replaced, which requires grabbing
6615 * spa_namespace_lock, so we must drop it here.
6617 spa_open_ref(spa
, FTAG
);
6618 mutex_exit(&spa_namespace_lock
);
6619 spa_async_suspend(spa
);
6620 mutex_enter(&spa_namespace_lock
);
6621 spa_close(spa
, FTAG
);
6623 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
6625 spa_deactivate(spa
);
6629 mutex_exit(&spa_namespace_lock
);
6633 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
6638 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
6642 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
6643 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
6644 if (vd
->vdev_guid
== guid
)
6648 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
6649 vd
= spa
->spa_spares
.sav_vdevs
[i
];
6650 if (vd
->vdev_guid
== guid
)
6659 spa_upgrade(spa_t
*spa
, uint64_t version
)
6661 ASSERT(spa_writeable(spa
));
6663 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6666 * This should only be called for a non-faulted pool, and since a
6667 * future version would result in an unopenable pool, this shouldn't be
6670 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
6671 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
6673 spa
->spa_uberblock
.ub_version
= version
;
6674 vdev_config_dirty(spa
->spa_root_vdev
);
6676 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6678 txg_wait_synced(spa_get_dsl(spa
), 0);
6682 spa_has_spare(spa_t
*spa
, uint64_t guid
)
6686 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6688 for (i
= 0; i
< sav
->sav_count
; i
++)
6689 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
6692 for (i
= 0; i
< sav
->sav_npending
; i
++) {
6693 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
6694 &spareguid
) == 0 && spareguid
== guid
)
6702 * Check if a pool has an active shared spare device.
6703 * Note: reference count of an active spare is 2, as a spare and as a replace
6706 spa_has_active_shared_spare(spa_t
*spa
)
6710 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
6712 for (i
= 0; i
< sav
->sav_count
; i
++) {
6713 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
6714 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
6723 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
6724 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6725 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6726 * or zdb as real changes.
6729 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, const char *name
)
6732 zfs_ereport_post(name
, spa
, vd
, NULL
, 0, 0);
6736 #if defined(_KERNEL) && defined(HAVE_SPL)
6737 /* state manipulation functions */
6738 EXPORT_SYMBOL(spa_open
);
6739 EXPORT_SYMBOL(spa_open_rewind
);
6740 EXPORT_SYMBOL(spa_get_stats
);
6741 EXPORT_SYMBOL(spa_create
);
6742 EXPORT_SYMBOL(spa_import_rootpool
);
6743 EXPORT_SYMBOL(spa_import
);
6744 EXPORT_SYMBOL(spa_tryimport
);
6745 EXPORT_SYMBOL(spa_destroy
);
6746 EXPORT_SYMBOL(spa_export
);
6747 EXPORT_SYMBOL(spa_reset
);
6748 EXPORT_SYMBOL(spa_async_request
);
6749 EXPORT_SYMBOL(spa_async_suspend
);
6750 EXPORT_SYMBOL(spa_async_resume
);
6751 EXPORT_SYMBOL(spa_inject_addref
);
6752 EXPORT_SYMBOL(spa_inject_delref
);
6753 EXPORT_SYMBOL(spa_scan_stat_init
);
6754 EXPORT_SYMBOL(spa_scan_get_stats
);
6756 /* device maniion */
6757 EXPORT_SYMBOL(spa_vdev_add
);
6758 EXPORT_SYMBOL(spa_vdev_attach
);
6759 EXPORT_SYMBOL(spa_vdev_detach
);
6760 EXPORT_SYMBOL(spa_vdev_remove
);
6761 EXPORT_SYMBOL(spa_vdev_setpath
);
6762 EXPORT_SYMBOL(spa_vdev_setfru
);
6763 EXPORT_SYMBOL(spa_vdev_split_mirror
);
6765 /* spare statech is global across all pools) */
6766 EXPORT_SYMBOL(spa_spare_add
);
6767 EXPORT_SYMBOL(spa_spare_remove
);
6768 EXPORT_SYMBOL(spa_spare_exists
);
6769 EXPORT_SYMBOL(spa_spare_activate
);
6771 /* L2ARC statech is global across all pools) */
6772 EXPORT_SYMBOL(spa_l2cache_add
);
6773 EXPORT_SYMBOL(spa_l2cache_remove
);
6774 EXPORT_SYMBOL(spa_l2cache_exists
);
6775 EXPORT_SYMBOL(spa_l2cache_activate
);
6776 EXPORT_SYMBOL(spa_l2cache_drop
);
6779 EXPORT_SYMBOL(spa_scan
);
6780 EXPORT_SYMBOL(spa_scan_stop
);
6783 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
6784 EXPORT_SYMBOL(spa_sync_allpools
);
6787 EXPORT_SYMBOL(spa_prop_set
);
6788 EXPORT_SYMBOL(spa_prop_get
);
6789 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
6791 /* asynchronous event notification */
6792 EXPORT_SYMBOL(spa_event_notify
);
6795 #if defined(_KERNEL) && defined(HAVE_SPL)
6796 module_param(spa_load_verify_maxinflight
, int, 0644);
6797 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
6798 "Max concurrent traversal I/Os while verifying pool during import -X");
6800 module_param(spa_load_verify_metadata
, int, 0644);
6801 MODULE_PARM_DESC(spa_load_verify_metadata
,
6802 "Set to traverse metadata on pool import");
6804 module_param(spa_load_verify_data
, int, 0644);
6805 MODULE_PARM_DESC(spa_load_verify_data
,
6806 "Set to traverse data on pool import");
6808 module_param(zio_taskq_batch_pct
, uint
, 0444);
6809 MODULE_PARM_DESC(zio_taskq_batch_pct
,
6810 "Percentage of CPUs to run an IO worker thread");