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) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
32 * Copyright (c) 2017 Datto Inc.
33 * Copyright 2017 Joyent, Inc.
37 * SPA: Storage Pool Allocator
39 * This file contains all the routines used when modifying on-disk SPA state.
40 * This includes opening, importing, destroying, exporting a pool, and syncing a
44 #include <sys/zfs_context.h>
45 #include <sys/fm/fs/zfs.h>
46 #include <sys/spa_impl.h>
48 #include <sys/zio_checksum.h>
50 #include <sys/dmu_tx.h>
54 #include <sys/vdev_impl.h>
55 #include <sys/vdev_disk.h>
56 #include <sys/metaslab.h>
57 #include <sys/metaslab_impl.h>
59 #include <sys/uberblock_impl.h>
62 #include <sys/dmu_traverse.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/unique.h>
65 #include <sys/dsl_pool.h>
66 #include <sys/dsl_dataset.h>
67 #include <sys/dsl_dir.h>
68 #include <sys/dsl_prop.h>
69 #include <sys/dsl_synctask.h>
70 #include <sys/fs/zfs.h>
72 #include <sys/callb.h>
73 #include <sys/systeminfo.h>
74 #include <sys/spa_boot.h>
75 #include <sys/zfs_ioctl.h>
76 #include <sys/dsl_scan.h>
77 #include <sys/zfeature.h>
78 #include <sys/dsl_destroy.h>
82 #include <sys/fm/protocol.h>
83 #include <sys/fm/util.h>
84 #include <sys/bootprops.h>
85 #include <sys/callb.h>
86 #include <sys/cpupart.h>
88 #include <sys/sysdc.h>
93 #include "zfs_comutil.h"
96 * The interval, in seconds, at which failed configuration cache file writes
99 static int zfs_ccw_retry_interval
= 300;
101 typedef enum zti_modes
{
102 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
103 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
104 ZTI_MODE_NULL
, /* don't create a taskq */
108 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
109 #define ZTI_PCT(n) { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
110 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
111 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
113 #define ZTI_N(n) ZTI_P(n, 1)
114 #define ZTI_ONE ZTI_N(1)
116 typedef struct zio_taskq_info
{
117 zti_modes_t zti_mode
;
122 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
123 "iss", "iss_h", "int", "int_h"
127 * This table defines the taskq settings for each ZFS I/O type. When
128 * initializing a pool, we use this table to create an appropriately sized
129 * taskq. Some operations are low volume and therefore have a small, static
130 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
131 * macros. Other operations process a large amount of data; the ZTI_BATCH
132 * macro causes us to create a taskq oriented for throughput. Some operations
133 * are so high frequency and short-lived that the taskq itself can become a a
134 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
135 * additional degree of parallelism specified by the number of threads per-
136 * taskq and the number of taskqs; when dispatching an event in this case, the
137 * particular taskq is chosen at random.
139 * The different taskq priorities are to handle the different contexts (issue
140 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
141 * need to be handled with minimum delay.
143 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
144 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
145 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
146 { ZTI_N(8), ZTI_NULL
, ZTI_P(12, 8), ZTI_NULL
}, /* READ */
147 { ZTI_BATCH
, ZTI_N(5), ZTI_P(12, 8), ZTI_N(5) }, /* WRITE */
148 { ZTI_P(12, 8), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
149 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
150 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
153 static sysevent_t
*spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
,
155 static void spa_event_post(sysevent_t
*ev
);
156 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
157 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
158 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
159 static inline int spa_load_impl(spa_t
*spa
, uint64_t, nvlist_t
*config
,
160 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
162 static void spa_vdev_resilver_done(spa_t
*spa
);
164 uint_t zio_taskq_batch_pct
= 75; /* 1 thread per cpu in pset */
165 id_t zio_taskq_psrset_bind
= PS_NONE
;
166 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
167 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
169 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
172 * This (illegal) pool name is used when temporarily importing a spa_t in order
173 * to get the vdev stats associated with the imported devices.
175 #define TRYIMPORT_NAME "$import"
178 * ==========================================================================
179 * SPA properties routines
180 * ==========================================================================
184 * Add a (source=src, propname=propval) list to an nvlist.
187 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
188 uint64_t intval
, zprop_source_t src
)
190 const char *propname
= zpool_prop_to_name(prop
);
193 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
194 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
197 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
199 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
201 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
202 nvlist_free(propval
);
206 * Get property values from the spa configuration.
209 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
211 vdev_t
*rvd
= spa
->spa_root_vdev
;
212 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
213 uint64_t size
, alloc
, cap
, version
;
214 const zprop_source_t src
= ZPROP_SRC_NONE
;
215 spa_config_dirent_t
*dp
;
216 metaslab_class_t
*mc
= spa_normal_class(spa
);
218 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
221 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
222 size
= metaslab_class_get_space(spa_normal_class(spa
));
223 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
224 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
225 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
226 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
229 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
230 metaslab_class_fragmentation(mc
), src
);
231 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
232 metaslab_class_expandable_space(mc
), src
);
233 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
234 (spa_mode(spa
) == FREAD
), src
);
236 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
237 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
239 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
240 ddt_get_pool_dedup_ratio(spa
), src
);
242 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
243 rvd
->vdev_state
, src
);
245 version
= spa_version(spa
);
246 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
)) {
247 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
248 version
, ZPROP_SRC_DEFAULT
);
250 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
251 version
, ZPROP_SRC_LOCAL
);
257 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
258 * when opening pools before this version freedir will be NULL.
260 if (pool
->dp_free_dir
!= NULL
) {
261 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
262 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
265 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
269 if (pool
->dp_leak_dir
!= NULL
) {
270 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
271 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
274 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
279 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
281 if (spa
->spa_comment
!= NULL
) {
282 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
286 if (spa
->spa_root
!= NULL
)
287 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
290 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
291 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
292 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
294 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
295 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
298 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
)) {
299 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
300 DNODE_MAX_SIZE
, ZPROP_SRC_NONE
);
302 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
303 DNODE_MIN_SIZE
, ZPROP_SRC_NONE
);
306 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
307 if (dp
->scd_path
== NULL
) {
308 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
309 "none", 0, ZPROP_SRC_LOCAL
);
310 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
311 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
312 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
318 * Get zpool property values.
321 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
323 objset_t
*mos
= spa
->spa_meta_objset
;
328 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
332 mutex_enter(&spa
->spa_props_lock
);
335 * Get properties from the spa config.
337 spa_prop_get_config(spa
, nvp
);
339 /* If no pool property object, no more prop to get. */
340 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
341 mutex_exit(&spa
->spa_props_lock
);
346 * Get properties from the MOS pool property object.
348 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
349 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
350 zap_cursor_advance(&zc
)) {
353 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
356 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPROP_INVAL
)
359 switch (za
.za_integer_length
) {
361 /* integer property */
362 if (za
.za_first_integer
!=
363 zpool_prop_default_numeric(prop
))
364 src
= ZPROP_SRC_LOCAL
;
366 if (prop
== ZPOOL_PROP_BOOTFS
) {
368 dsl_dataset_t
*ds
= NULL
;
370 dp
= spa_get_dsl(spa
);
371 dsl_pool_config_enter(dp
, FTAG
);
372 if ((err
= dsl_dataset_hold_obj(dp
,
373 za
.za_first_integer
, FTAG
, &ds
))) {
374 dsl_pool_config_exit(dp
, FTAG
);
378 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
380 dsl_dataset_name(ds
, strval
);
381 dsl_dataset_rele(ds
, FTAG
);
382 dsl_pool_config_exit(dp
, FTAG
);
385 intval
= za
.za_first_integer
;
388 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
391 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
396 /* string property */
397 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
398 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
399 za
.za_name
, 1, za
.za_num_integers
, strval
);
401 kmem_free(strval
, za
.za_num_integers
);
404 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
405 kmem_free(strval
, za
.za_num_integers
);
412 zap_cursor_fini(&zc
);
413 mutex_exit(&spa
->spa_props_lock
);
415 if (err
&& err
!= ENOENT
) {
425 * Validate the given pool properties nvlist and modify the list
426 * for the property values to be set.
429 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
432 int error
= 0, reset_bootfs
= 0;
434 boolean_t has_feature
= B_FALSE
;
437 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
439 char *strval
, *slash
, *check
, *fname
;
440 const char *propname
= nvpair_name(elem
);
441 zpool_prop_t prop
= zpool_name_to_prop(propname
);
445 if (!zpool_prop_feature(propname
)) {
446 error
= SET_ERROR(EINVAL
);
451 * Sanitize the input.
453 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
454 error
= SET_ERROR(EINVAL
);
458 if (nvpair_value_uint64(elem
, &intval
) != 0) {
459 error
= SET_ERROR(EINVAL
);
464 error
= SET_ERROR(EINVAL
);
468 fname
= strchr(propname
, '@') + 1;
469 if (zfeature_lookup_name(fname
, NULL
) != 0) {
470 error
= SET_ERROR(EINVAL
);
474 has_feature
= B_TRUE
;
477 case ZPOOL_PROP_VERSION
:
478 error
= nvpair_value_uint64(elem
, &intval
);
480 (intval
< spa_version(spa
) ||
481 intval
> SPA_VERSION_BEFORE_FEATURES
||
483 error
= SET_ERROR(EINVAL
);
486 case ZPOOL_PROP_DELEGATION
:
487 case ZPOOL_PROP_AUTOREPLACE
:
488 case ZPOOL_PROP_LISTSNAPS
:
489 case ZPOOL_PROP_AUTOEXPAND
:
490 error
= nvpair_value_uint64(elem
, &intval
);
491 if (!error
&& intval
> 1)
492 error
= SET_ERROR(EINVAL
);
495 case ZPOOL_PROP_MULTIHOST
:
496 error
= nvpair_value_uint64(elem
, &intval
);
497 if (!error
&& intval
> 1)
498 error
= SET_ERROR(EINVAL
);
500 if (!error
&& !spa_get_hostid())
501 error
= SET_ERROR(ENOTSUP
);
505 case ZPOOL_PROP_BOOTFS
:
507 * If the pool version is less than SPA_VERSION_BOOTFS,
508 * or the pool is still being created (version == 0),
509 * the bootfs property cannot be set.
511 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
512 error
= SET_ERROR(ENOTSUP
);
517 * Make sure the vdev config is bootable
519 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
520 error
= SET_ERROR(ENOTSUP
);
526 error
= nvpair_value_string(elem
, &strval
);
532 if (strval
== NULL
|| strval
[0] == '\0') {
533 objnum
= zpool_prop_default_numeric(
538 error
= dmu_objset_hold(strval
, FTAG
, &os
);
543 * Must be ZPL, and its property settings
544 * must be supported by GRUB (compression
545 * is not gzip, and large blocks or large
546 * dnodes are not used).
549 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
550 error
= SET_ERROR(ENOTSUP
);
552 dsl_prop_get_int_ds(dmu_objset_ds(os
),
553 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
555 !BOOTFS_COMPRESS_VALID(propval
)) {
556 error
= SET_ERROR(ENOTSUP
);
558 dsl_prop_get_int_ds(dmu_objset_ds(os
),
559 zfs_prop_to_name(ZFS_PROP_DNODESIZE
),
561 propval
!= ZFS_DNSIZE_LEGACY
) {
562 error
= SET_ERROR(ENOTSUP
);
564 objnum
= dmu_objset_id(os
);
566 dmu_objset_rele(os
, FTAG
);
570 case ZPOOL_PROP_FAILUREMODE
:
571 error
= nvpair_value_uint64(elem
, &intval
);
572 if (!error
&& intval
> ZIO_FAILURE_MODE_PANIC
)
573 error
= SET_ERROR(EINVAL
);
576 * This is a special case which only occurs when
577 * the pool has completely failed. This allows
578 * the user to change the in-core failmode property
579 * without syncing it out to disk (I/Os might
580 * currently be blocked). We do this by returning
581 * EIO to the caller (spa_prop_set) to trick it
582 * into thinking we encountered a property validation
585 if (!error
&& spa_suspended(spa
)) {
586 spa
->spa_failmode
= intval
;
587 error
= SET_ERROR(EIO
);
591 case ZPOOL_PROP_CACHEFILE
:
592 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
595 if (strval
[0] == '\0')
598 if (strcmp(strval
, "none") == 0)
601 if (strval
[0] != '/') {
602 error
= SET_ERROR(EINVAL
);
606 slash
= strrchr(strval
, '/');
607 ASSERT(slash
!= NULL
);
609 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
610 strcmp(slash
, "/..") == 0)
611 error
= SET_ERROR(EINVAL
);
614 case ZPOOL_PROP_COMMENT
:
615 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
617 for (check
= strval
; *check
!= '\0'; check
++) {
618 if (!isprint(*check
)) {
619 error
= SET_ERROR(EINVAL
);
623 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
624 error
= SET_ERROR(E2BIG
);
627 case ZPOOL_PROP_DEDUPDITTO
:
628 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
629 error
= SET_ERROR(ENOTSUP
);
631 error
= nvpair_value_uint64(elem
, &intval
);
633 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
634 error
= SET_ERROR(EINVAL
);
645 if (!error
&& reset_bootfs
) {
646 error
= nvlist_remove(props
,
647 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
650 error
= nvlist_add_uint64(props
,
651 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
659 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
662 spa_config_dirent_t
*dp
;
664 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
668 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
671 if (cachefile
[0] == '\0')
672 dp
->scd_path
= spa_strdup(spa_config_path
);
673 else if (strcmp(cachefile
, "none") == 0)
676 dp
->scd_path
= spa_strdup(cachefile
);
678 list_insert_head(&spa
->spa_config_list
, dp
);
680 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
684 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
687 nvpair_t
*elem
= NULL
;
688 boolean_t need_sync
= B_FALSE
;
690 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
693 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
694 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
696 if (prop
== ZPOOL_PROP_CACHEFILE
||
697 prop
== ZPOOL_PROP_ALTROOT
||
698 prop
== ZPOOL_PROP_READONLY
)
701 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPROP_INVAL
) {
704 if (prop
== ZPOOL_PROP_VERSION
) {
705 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
707 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
708 ver
= SPA_VERSION_FEATURES
;
712 /* Save time if the version is already set. */
713 if (ver
== spa_version(spa
))
717 * In addition to the pool directory object, we might
718 * create the pool properties object, the features for
719 * read object, the features for write object, or the
720 * feature descriptions object.
722 error
= dsl_sync_task(spa
->spa_name
, NULL
,
723 spa_sync_version
, &ver
,
724 6, ZFS_SPACE_CHECK_RESERVED
);
735 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
736 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
743 * If the bootfs property value is dsobj, clear it.
746 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
748 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
749 VERIFY(zap_remove(spa
->spa_meta_objset
,
750 spa
->spa_pool_props_object
,
751 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
758 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
760 ASSERTV(uint64_t *newguid
= arg
);
761 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
762 vdev_t
*rvd
= spa
->spa_root_vdev
;
765 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
766 vdev_state
= rvd
->vdev_state
;
767 spa_config_exit(spa
, SCL_STATE
, FTAG
);
769 if (vdev_state
!= VDEV_STATE_HEALTHY
)
770 return (SET_ERROR(ENXIO
));
772 ASSERT3U(spa_guid(spa
), !=, *newguid
);
778 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
780 uint64_t *newguid
= arg
;
781 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
783 vdev_t
*rvd
= spa
->spa_root_vdev
;
785 oldguid
= spa_guid(spa
);
787 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
788 rvd
->vdev_guid
= *newguid
;
789 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
790 vdev_config_dirty(rvd
);
791 spa_config_exit(spa
, SCL_STATE
, FTAG
);
793 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
798 * Change the GUID for the pool. This is done so that we can later
799 * re-import a pool built from a clone of our own vdevs. We will modify
800 * the root vdev's guid, our own pool guid, and then mark all of our
801 * vdevs dirty. Note that we must make sure that all our vdevs are
802 * online when we do this, or else any vdevs that weren't present
803 * would be orphaned from our pool. We are also going to issue a
804 * sysevent to update any watchers.
807 spa_change_guid(spa_t
*spa
)
812 mutex_enter(&spa
->spa_vdev_top_lock
);
813 mutex_enter(&spa_namespace_lock
);
814 guid
= spa_generate_guid(NULL
);
816 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
817 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
820 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
821 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
824 mutex_exit(&spa_namespace_lock
);
825 mutex_exit(&spa
->spa_vdev_top_lock
);
831 * ==========================================================================
832 * SPA state manipulation (open/create/destroy/import/export)
833 * ==========================================================================
837 spa_error_entry_compare(const void *a
, const void *b
)
839 const spa_error_entry_t
*sa
= (const spa_error_entry_t
*)a
;
840 const spa_error_entry_t
*sb
= (const spa_error_entry_t
*)b
;
843 ret
= memcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
844 sizeof (zbookmark_phys_t
));
846 return (AVL_ISIGN(ret
));
850 * Utility function which retrieves copies of the current logs and
851 * re-initializes them in the process.
854 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
856 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
858 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
859 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
861 avl_create(&spa
->spa_errlist_scrub
,
862 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
863 offsetof(spa_error_entry_t
, se_avl
));
864 avl_create(&spa
->spa_errlist_last
,
865 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
866 offsetof(spa_error_entry_t
, se_avl
));
870 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
872 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
873 enum zti_modes mode
= ztip
->zti_mode
;
874 uint_t value
= ztip
->zti_value
;
875 uint_t count
= ztip
->zti_count
;
876 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
879 boolean_t batch
= B_FALSE
;
881 if (mode
== ZTI_MODE_NULL
) {
883 tqs
->stqs_taskq
= NULL
;
887 ASSERT3U(count
, >, 0);
889 tqs
->stqs_count
= count
;
890 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
894 ASSERT3U(value
, >=, 1);
895 value
= MAX(value
, 1);
896 flags
|= TASKQ_DYNAMIC
;
901 flags
|= TASKQ_THREADS_CPU_PCT
;
902 value
= MIN(zio_taskq_batch_pct
, 100);
906 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
908 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
912 for (uint_t i
= 0; i
< count
; i
++) {
916 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
917 zio_type_name
[t
], zio_taskq_types
[q
], i
);
919 (void) snprintf(name
, sizeof (name
), "%s_%s",
920 zio_type_name
[t
], zio_taskq_types
[q
]);
923 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
925 flags
|= TASKQ_DC_BATCH
;
927 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
928 spa
->spa_proc
, zio_taskq_basedc
, flags
);
930 pri_t pri
= maxclsyspri
;
932 * The write issue taskq can be extremely CPU
933 * intensive. Run it at slightly less important
934 * priority than the other taskqs. Under Linux this
935 * means incrementing the priority value on platforms
936 * like illumos it should be decremented.
938 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
941 tq
= taskq_create_proc(name
, value
, pri
, 50,
942 INT_MAX
, spa
->spa_proc
, flags
);
945 tqs
->stqs_taskq
[i
] = tq
;
950 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
952 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
954 if (tqs
->stqs_taskq
== NULL
) {
955 ASSERT3U(tqs
->stqs_count
, ==, 0);
959 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
960 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
961 taskq_destroy(tqs
->stqs_taskq
[i
]);
964 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
965 tqs
->stqs_taskq
= NULL
;
969 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
970 * Note that a type may have multiple discrete taskqs to avoid lock contention
971 * on the taskq itself. In that case we choose which taskq at random by using
972 * the low bits of gethrtime().
975 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
976 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
978 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
981 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
982 ASSERT3U(tqs
->stqs_count
, !=, 0);
984 if (tqs
->stqs_count
== 1) {
985 tq
= tqs
->stqs_taskq
[0];
987 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
990 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
994 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
997 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
998 task_func_t
*func
, void *arg
, uint_t flags
)
1000 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1004 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1005 ASSERT3U(tqs
->stqs_count
, !=, 0);
1007 if (tqs
->stqs_count
== 1) {
1008 tq
= tqs
->stqs_taskq
[0];
1010 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1013 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1015 taskq_wait_id(tq
, id
);
1019 spa_create_zio_taskqs(spa_t
*spa
)
1021 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1022 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1023 spa_taskqs_init(spa
, t
, q
);
1029 * Disabled until spa_thread() can be adapted for Linux.
1031 #undef HAVE_SPA_THREAD
1033 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1035 spa_thread(void *arg
)
1037 callb_cpr_t cprinfo
;
1040 user_t
*pu
= PTOU(curproc
);
1042 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1045 ASSERT(curproc
!= &p0
);
1046 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1047 "zpool-%s", spa
->spa_name
);
1048 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1050 /* bind this thread to the requested psrset */
1051 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1053 mutex_enter(&cpu_lock
);
1054 mutex_enter(&pidlock
);
1055 mutex_enter(&curproc
->p_lock
);
1057 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1058 0, NULL
, NULL
) == 0) {
1059 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1062 "Couldn't bind process for zfs pool \"%s\" to "
1063 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1066 mutex_exit(&curproc
->p_lock
);
1067 mutex_exit(&pidlock
);
1068 mutex_exit(&cpu_lock
);
1072 if (zio_taskq_sysdc
) {
1073 sysdc_thread_enter(curthread
, 100, 0);
1076 spa
->spa_proc
= curproc
;
1077 spa
->spa_did
= curthread
->t_did
;
1079 spa_create_zio_taskqs(spa
);
1081 mutex_enter(&spa
->spa_proc_lock
);
1082 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1084 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1085 cv_broadcast(&spa
->spa_proc_cv
);
1087 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1088 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1089 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1090 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1092 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1093 spa
->spa_proc_state
= SPA_PROC_GONE
;
1094 spa
->spa_proc
= &p0
;
1095 cv_broadcast(&spa
->spa_proc_cv
);
1096 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1098 mutex_enter(&curproc
->p_lock
);
1104 * Activate an uninitialized pool.
1107 spa_activate(spa_t
*spa
, int mode
)
1109 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1111 spa
->spa_state
= POOL_STATE_ACTIVE
;
1112 spa
->spa_mode
= mode
;
1114 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1115 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1117 /* Try to create a covering process */
1118 mutex_enter(&spa
->spa_proc_lock
);
1119 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1120 ASSERT(spa
->spa_proc
== &p0
);
1123 #ifdef HAVE_SPA_THREAD
1124 /* Only create a process if we're going to be around a while. */
1125 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1126 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1128 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1129 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1130 cv_wait(&spa
->spa_proc_cv
,
1131 &spa
->spa_proc_lock
);
1133 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1134 ASSERT(spa
->spa_proc
!= &p0
);
1135 ASSERT(spa
->spa_did
!= 0);
1139 "Couldn't create process for zfs pool \"%s\"\n",
1144 #endif /* HAVE_SPA_THREAD */
1145 mutex_exit(&spa
->spa_proc_lock
);
1147 /* If we didn't create a process, we need to create our taskqs. */
1148 if (spa
->spa_proc
== &p0
) {
1149 spa_create_zio_taskqs(spa
);
1152 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1153 offsetof(vdev_t
, vdev_config_dirty_node
));
1154 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1155 offsetof(objset_t
, os_evicting_node
));
1156 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1157 offsetof(vdev_t
, vdev_state_dirty_node
));
1159 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1160 offsetof(struct vdev
, vdev_txg_node
));
1162 avl_create(&spa
->spa_errlist_scrub
,
1163 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1164 offsetof(spa_error_entry_t
, se_avl
));
1165 avl_create(&spa
->spa_errlist_last
,
1166 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1167 offsetof(spa_error_entry_t
, se_avl
));
1169 spa_keystore_init(&spa
->spa_keystore
);
1172 * This taskq is used to perform zvol-minor-related tasks
1173 * asynchronously. This has several advantages, including easy
1174 * resolution of various deadlocks (zfsonlinux bug #3681).
1176 * The taskq must be single threaded to ensure tasks are always
1177 * processed in the order in which they were dispatched.
1179 * A taskq per pool allows one to keep the pools independent.
1180 * This way if one pool is suspended, it will not impact another.
1182 * The preferred location to dispatch a zvol minor task is a sync
1183 * task. In this context, there is easy access to the spa_t and minimal
1184 * error handling is required because the sync task must succeed.
1186 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1190 * The taskq to upgrade datasets in this pool. Currently used by
1191 * feature SPA_FEATURE_USEROBJ_ACCOUNTING.
1193 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", boot_ncpus
,
1194 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1198 * Opposite of spa_activate().
1201 spa_deactivate(spa_t
*spa
)
1203 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1204 ASSERT(spa
->spa_dsl_pool
== NULL
);
1205 ASSERT(spa
->spa_root_vdev
== NULL
);
1206 ASSERT(spa
->spa_async_zio_root
== NULL
);
1207 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1209 spa_evicting_os_wait(spa
);
1211 if (spa
->spa_zvol_taskq
) {
1212 taskq_destroy(spa
->spa_zvol_taskq
);
1213 spa
->spa_zvol_taskq
= NULL
;
1216 if (spa
->spa_upgrade_taskq
) {
1217 taskq_destroy(spa
->spa_upgrade_taskq
);
1218 spa
->spa_upgrade_taskq
= NULL
;
1221 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1223 list_destroy(&spa
->spa_config_dirty_list
);
1224 list_destroy(&spa
->spa_evicting_os_list
);
1225 list_destroy(&spa
->spa_state_dirty_list
);
1227 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1229 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1230 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1231 spa_taskqs_fini(spa
, t
, q
);
1235 metaslab_class_destroy(spa
->spa_normal_class
);
1236 spa
->spa_normal_class
= NULL
;
1238 metaslab_class_destroy(spa
->spa_log_class
);
1239 spa
->spa_log_class
= NULL
;
1242 * If this was part of an import or the open otherwise failed, we may
1243 * still have errors left in the queues. Empty them just in case.
1245 spa_errlog_drain(spa
);
1246 avl_destroy(&spa
->spa_errlist_scrub
);
1247 avl_destroy(&spa
->spa_errlist_last
);
1249 spa_keystore_fini(&spa
->spa_keystore
);
1251 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1253 mutex_enter(&spa
->spa_proc_lock
);
1254 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1255 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1256 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1257 cv_broadcast(&spa
->spa_proc_cv
);
1258 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1259 ASSERT(spa
->spa_proc
!= &p0
);
1260 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1262 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1263 spa
->spa_proc_state
= SPA_PROC_NONE
;
1265 ASSERT(spa
->spa_proc
== &p0
);
1266 mutex_exit(&spa
->spa_proc_lock
);
1269 * We want to make sure spa_thread() has actually exited the ZFS
1270 * module, so that the module can't be unloaded out from underneath
1273 if (spa
->spa_did
!= 0) {
1274 thread_join(spa
->spa_did
);
1280 * Verify a pool configuration, and construct the vdev tree appropriately. This
1281 * will create all the necessary vdevs in the appropriate layout, with each vdev
1282 * in the CLOSED state. This will prep the pool before open/creation/import.
1283 * All vdev validation is done by the vdev_alloc() routine.
1286 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1287 uint_t id
, int atype
)
1293 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1296 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1299 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1302 if (error
== ENOENT
)
1308 return (SET_ERROR(EINVAL
));
1311 for (int c
= 0; c
< children
; c
++) {
1313 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1321 ASSERT(*vdp
!= NULL
);
1327 * Opposite of spa_load().
1330 spa_unload(spa_t
*spa
)
1334 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1339 spa_async_suspend(spa
);
1344 if (spa
->spa_sync_on
) {
1345 txg_sync_stop(spa
->spa_dsl_pool
);
1346 spa
->spa_sync_on
= B_FALSE
;
1350 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1351 * to call it earlier, before we wait for async i/o to complete.
1352 * This ensures that there is no async metaslab prefetching, by
1353 * calling taskq_wait(mg_taskq).
1355 if (spa
->spa_root_vdev
!= NULL
) {
1356 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1357 for (int c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++)
1358 vdev_metaslab_fini(spa
->spa_root_vdev
->vdev_child
[c
]);
1359 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1362 if (spa
->spa_mmp
.mmp_thread
)
1363 mmp_thread_stop(spa
);
1366 * Wait for any outstanding async I/O to complete.
1368 if (spa
->spa_async_zio_root
!= NULL
) {
1369 for (int i
= 0; i
< max_ncpus
; i
++)
1370 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1371 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1372 spa
->spa_async_zio_root
= NULL
;
1375 bpobj_close(&spa
->spa_deferred_bpobj
);
1377 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1382 if (spa
->spa_root_vdev
)
1383 vdev_free(spa
->spa_root_vdev
);
1384 ASSERT(spa
->spa_root_vdev
== NULL
);
1387 * Close the dsl pool.
1389 if (spa
->spa_dsl_pool
) {
1390 dsl_pool_close(spa
->spa_dsl_pool
);
1391 spa
->spa_dsl_pool
= NULL
;
1392 spa
->spa_meta_objset
= NULL
;
1398 * Drop and purge level 2 cache
1400 spa_l2cache_drop(spa
);
1402 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1403 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1404 if (spa
->spa_spares
.sav_vdevs
) {
1405 kmem_free(spa
->spa_spares
.sav_vdevs
,
1406 spa
->spa_spares
.sav_count
* sizeof (void *));
1407 spa
->spa_spares
.sav_vdevs
= NULL
;
1409 if (spa
->spa_spares
.sav_config
) {
1410 nvlist_free(spa
->spa_spares
.sav_config
);
1411 spa
->spa_spares
.sav_config
= NULL
;
1413 spa
->spa_spares
.sav_count
= 0;
1415 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1416 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1417 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1419 if (spa
->spa_l2cache
.sav_vdevs
) {
1420 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1421 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1422 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1424 if (spa
->spa_l2cache
.sav_config
) {
1425 nvlist_free(spa
->spa_l2cache
.sav_config
);
1426 spa
->spa_l2cache
.sav_config
= NULL
;
1428 spa
->spa_l2cache
.sav_count
= 0;
1430 spa
->spa_async_suspended
= 0;
1432 if (spa
->spa_comment
!= NULL
) {
1433 spa_strfree(spa
->spa_comment
);
1434 spa
->spa_comment
= NULL
;
1437 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1441 * Load (or re-load) the current list of vdevs describing the active spares for
1442 * this pool. When this is called, we have some form of basic information in
1443 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1444 * then re-generate a more complete list including status information.
1447 spa_load_spares(spa_t
*spa
)
1454 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1457 * First, close and free any existing spare vdevs.
1459 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1460 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1462 /* Undo the call to spa_activate() below */
1463 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1464 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1465 spa_spare_remove(tvd
);
1470 if (spa
->spa_spares
.sav_vdevs
)
1471 kmem_free(spa
->spa_spares
.sav_vdevs
,
1472 spa
->spa_spares
.sav_count
* sizeof (void *));
1474 if (spa
->spa_spares
.sav_config
== NULL
)
1477 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1478 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1480 spa
->spa_spares
.sav_count
= (int)nspares
;
1481 spa
->spa_spares
.sav_vdevs
= NULL
;
1487 * Construct the array of vdevs, opening them to get status in the
1488 * process. For each spare, there is potentially two different vdev_t
1489 * structures associated with it: one in the list of spares (used only
1490 * for basic validation purposes) and one in the active vdev
1491 * configuration (if it's spared in). During this phase we open and
1492 * validate each vdev on the spare list. If the vdev also exists in the
1493 * active configuration, then we also mark this vdev as an active spare.
1495 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1497 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1498 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1499 VDEV_ALLOC_SPARE
) == 0);
1502 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1504 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1505 B_FALSE
)) != NULL
) {
1506 if (!tvd
->vdev_isspare
)
1510 * We only mark the spare active if we were successfully
1511 * able to load the vdev. Otherwise, importing a pool
1512 * with a bad active spare would result in strange
1513 * behavior, because multiple pool would think the spare
1514 * is actively in use.
1516 * There is a vulnerability here to an equally bizarre
1517 * circumstance, where a dead active spare is later
1518 * brought back to life (onlined or otherwise). Given
1519 * the rarity of this scenario, and the extra complexity
1520 * it adds, we ignore the possibility.
1522 if (!vdev_is_dead(tvd
))
1523 spa_spare_activate(tvd
);
1527 vd
->vdev_aux
= &spa
->spa_spares
;
1529 if (vdev_open(vd
) != 0)
1532 if (vdev_validate_aux(vd
) == 0)
1537 * Recompute the stashed list of spares, with status information
1540 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1541 DATA_TYPE_NVLIST_ARRAY
) == 0);
1543 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1545 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1546 spares
[i
] = vdev_config_generate(spa
,
1547 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1548 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1549 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1550 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1551 nvlist_free(spares
[i
]);
1552 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1556 * Load (or re-load) the current list of vdevs describing the active l2cache for
1557 * this pool. When this is called, we have some form of basic information in
1558 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1559 * then re-generate a more complete list including status information.
1560 * Devices which are already active have their details maintained, and are
1564 spa_load_l2cache(spa_t
*spa
)
1568 int i
, j
, oldnvdevs
;
1570 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1571 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1573 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1575 oldvdevs
= sav
->sav_vdevs
;
1576 oldnvdevs
= sav
->sav_count
;
1577 sav
->sav_vdevs
= NULL
;
1580 if (sav
->sav_config
== NULL
) {
1586 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1587 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1588 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1591 * Process new nvlist of vdevs.
1593 for (i
= 0; i
< nl2cache
; i
++) {
1594 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1598 for (j
= 0; j
< oldnvdevs
; j
++) {
1600 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1602 * Retain previous vdev for add/remove ops.
1610 if (newvdevs
[i
] == NULL
) {
1614 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1615 VDEV_ALLOC_L2CACHE
) == 0);
1620 * Commit this vdev as an l2cache device,
1621 * even if it fails to open.
1623 spa_l2cache_add(vd
);
1628 spa_l2cache_activate(vd
);
1630 if (vdev_open(vd
) != 0)
1633 (void) vdev_validate_aux(vd
);
1635 if (!vdev_is_dead(vd
))
1636 l2arc_add_vdev(spa
, vd
);
1640 sav
->sav_vdevs
= newvdevs
;
1641 sav
->sav_count
= (int)nl2cache
;
1644 * Recompute the stashed list of l2cache devices, with status
1645 * information this time.
1647 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1648 DATA_TYPE_NVLIST_ARRAY
) == 0);
1650 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1651 for (i
= 0; i
< sav
->sav_count
; i
++)
1652 l2cache
[i
] = vdev_config_generate(spa
,
1653 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1654 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1655 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1659 * Purge vdevs that were dropped
1661 for (i
= 0; i
< oldnvdevs
; i
++) {
1666 ASSERT(vd
->vdev_isl2cache
);
1668 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1669 pool
!= 0ULL && l2arc_vdev_present(vd
))
1670 l2arc_remove_vdev(vd
);
1671 vdev_clear_stats(vd
);
1677 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1679 for (i
= 0; i
< sav
->sav_count
; i
++)
1680 nvlist_free(l2cache
[i
]);
1682 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1686 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1689 char *packed
= NULL
;
1694 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1698 nvsize
= *(uint64_t *)db
->db_data
;
1699 dmu_buf_rele(db
, FTAG
);
1701 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1702 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1705 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1706 vmem_free(packed
, nvsize
);
1712 * Checks to see if the given vdev could not be opened, in which case we post a
1713 * sysevent to notify the autoreplace code that the device has been removed.
1716 spa_check_removed(vdev_t
*vd
)
1718 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1719 spa_check_removed(vd
->vdev_child
[c
]);
1721 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1723 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1724 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
1729 spa_config_valid_zaps(vdev_t
*vd
, vdev_t
*mvd
)
1731 ASSERT3U(vd
->vdev_children
, ==, mvd
->vdev_children
);
1733 vd
->vdev_top_zap
= mvd
->vdev_top_zap
;
1734 vd
->vdev_leaf_zap
= mvd
->vdev_leaf_zap
;
1736 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
1737 spa_config_valid_zaps(vd
->vdev_child
[i
], mvd
->vdev_child
[i
]);
1742 * Validate the current config against the MOS config
1745 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1747 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1750 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1752 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1753 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1755 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1758 * If we're doing a normal import, then build up any additional
1759 * diagnostic information about missing devices in this config.
1760 * We'll pass this up to the user for further processing.
1762 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1763 nvlist_t
**child
, *nv
;
1766 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
1768 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1770 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1771 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1772 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1774 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1775 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1777 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1782 VERIFY(nvlist_add_nvlist_array(nv
,
1783 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1784 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1785 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1787 for (int i
= 0; i
< idx
; i
++)
1788 nvlist_free(child
[i
]);
1791 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1795 * Compare the root vdev tree with the information we have
1796 * from the MOS config (mrvd). Check each top-level vdev
1797 * with the corresponding MOS config top-level (mtvd).
1799 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1800 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1801 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1804 * Resolve any "missing" vdevs in the current configuration.
1805 * If we find that the MOS config has more accurate information
1806 * about the top-level vdev then use that vdev instead.
1808 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1809 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1811 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1815 * Device specific actions.
1817 if (mtvd
->vdev_islog
) {
1818 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1821 * XXX - once we have 'readonly' pool
1822 * support we should be able to handle
1823 * missing data devices by transitioning
1824 * the pool to readonly.
1830 * Swap the missing vdev with the data we were
1831 * able to obtain from the MOS config.
1833 vdev_remove_child(rvd
, tvd
);
1834 vdev_remove_child(mrvd
, mtvd
);
1836 vdev_add_child(rvd
, mtvd
);
1837 vdev_add_child(mrvd
, tvd
);
1839 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1841 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1845 if (mtvd
->vdev_islog
) {
1847 * Load the slog device's state from the MOS
1848 * config since it's possible that the label
1849 * does not contain the most up-to-date
1852 vdev_load_log_state(tvd
, mtvd
);
1857 * Per-vdev ZAP info is stored exclusively in the MOS.
1859 spa_config_valid_zaps(tvd
, mtvd
);
1864 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1867 * Ensure we were able to validate the config.
1869 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1873 * Check for missing log devices
1876 spa_check_logs(spa_t
*spa
)
1878 boolean_t rv
= B_FALSE
;
1879 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1881 switch (spa
->spa_log_state
) {
1884 case SPA_LOG_MISSING
:
1885 /* need to recheck in case slog has been restored */
1886 case SPA_LOG_UNKNOWN
:
1887 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1888 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1890 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1897 spa_passivate_log(spa_t
*spa
)
1899 vdev_t
*rvd
= spa
->spa_root_vdev
;
1900 boolean_t slog_found
= B_FALSE
;
1902 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1904 if (!spa_has_slogs(spa
))
1907 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1908 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1909 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1911 if (tvd
->vdev_islog
) {
1912 metaslab_group_passivate(mg
);
1913 slog_found
= B_TRUE
;
1917 return (slog_found
);
1921 spa_activate_log(spa_t
*spa
)
1923 vdev_t
*rvd
= spa
->spa_root_vdev
;
1925 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1927 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1928 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1929 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1931 if (tvd
->vdev_islog
)
1932 metaslab_group_activate(mg
);
1937 spa_offline_log(spa_t
*spa
)
1941 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1942 NULL
, DS_FIND_CHILDREN
);
1945 * We successfully offlined the log device, sync out the
1946 * current txg so that the "stubby" block can be removed
1949 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1955 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1957 for (int i
= 0; i
< sav
->sav_count
; i
++)
1958 spa_check_removed(sav
->sav_vdevs
[i
]);
1962 spa_claim_notify(zio_t
*zio
)
1964 spa_t
*spa
= zio
->io_spa
;
1969 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1970 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1971 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1972 mutex_exit(&spa
->spa_props_lock
);
1975 typedef struct spa_load_error
{
1976 uint64_t sle_meta_count
;
1977 uint64_t sle_data_count
;
1981 spa_load_verify_done(zio_t
*zio
)
1983 blkptr_t
*bp
= zio
->io_bp
;
1984 spa_load_error_t
*sle
= zio
->io_private
;
1985 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1986 int error
= zio
->io_error
;
1987 spa_t
*spa
= zio
->io_spa
;
1989 abd_free(zio
->io_abd
);
1991 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1992 type
!= DMU_OT_INTENT_LOG
)
1993 atomic_inc_64(&sle
->sle_meta_count
);
1995 atomic_inc_64(&sle
->sle_data_count
);
1998 mutex_enter(&spa
->spa_scrub_lock
);
1999 spa
->spa_load_verify_ios
--;
2000 cv_broadcast(&spa
->spa_scrub_io_cv
);
2001 mutex_exit(&spa
->spa_scrub_lock
);
2005 * Maximum number of concurrent scrub i/os to create while verifying
2006 * a pool while importing it.
2008 int spa_load_verify_maxinflight
= 10000;
2009 int spa_load_verify_metadata
= B_TRUE
;
2010 int spa_load_verify_data
= B_TRUE
;
2014 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2015 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2017 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
2020 * Note: normally this routine will not be called if
2021 * spa_load_verify_metadata is not set. However, it may be useful
2022 * to manually set the flag after the traversal has begun.
2024 if (!spa_load_verify_metadata
)
2026 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2030 size_t size
= BP_GET_PSIZE(bp
);
2032 mutex_enter(&spa
->spa_scrub_lock
);
2033 while (spa
->spa_load_verify_ios
>= spa_load_verify_maxinflight
)
2034 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2035 spa
->spa_load_verify_ios
++;
2036 mutex_exit(&spa
->spa_scrub_lock
);
2038 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2039 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2040 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2041 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2047 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2049 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2050 return (SET_ERROR(ENAMETOOLONG
));
2056 spa_load_verify(spa_t
*spa
)
2059 spa_load_error_t sle
= { 0 };
2060 zpool_rewind_policy_t policy
;
2061 boolean_t verify_ok
= B_FALSE
;
2064 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
2066 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
2069 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2070 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2071 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2073 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2077 rio
= zio_root(spa
, NULL
, &sle
,
2078 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2080 if (spa_load_verify_metadata
) {
2081 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2082 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2083 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2086 (void) zio_wait(rio
);
2088 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2089 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2091 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
2092 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
2096 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2097 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2099 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2100 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2101 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2102 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2103 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2104 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2105 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2107 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2111 if (error
!= ENXIO
&& error
!= EIO
)
2112 error
= SET_ERROR(EIO
);
2116 return (verify_ok
? 0 : EIO
);
2120 * Find a value in the pool props object.
2123 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2125 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2126 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2130 * Find a value in the pool directory object.
2133 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2135 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2136 name
, sizeof (uint64_t), 1, val
));
2140 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2142 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2147 * Fix up config after a partly-completed split. This is done with the
2148 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2149 * pool have that entry in their config, but only the splitting one contains
2150 * a list of all the guids of the vdevs that are being split off.
2152 * This function determines what to do with that list: either rejoin
2153 * all the disks to the pool, or complete the splitting process. To attempt
2154 * the rejoin, each disk that is offlined is marked online again, and
2155 * we do a reopen() call. If the vdev label for every disk that was
2156 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2157 * then we call vdev_split() on each disk, and complete the split.
2159 * Otherwise we leave the config alone, with all the vdevs in place in
2160 * the original pool.
2163 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2170 boolean_t attempt_reopen
;
2172 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2175 /* check that the config is complete */
2176 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2177 &glist
, &gcount
) != 0)
2180 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2182 /* attempt to online all the vdevs & validate */
2183 attempt_reopen
= B_TRUE
;
2184 for (i
= 0; i
< gcount
; i
++) {
2185 if (glist
[i
] == 0) /* vdev is hole */
2188 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2189 if (vd
[i
] == NULL
) {
2191 * Don't bother attempting to reopen the disks;
2192 * just do the split.
2194 attempt_reopen
= B_FALSE
;
2196 /* attempt to re-online it */
2197 vd
[i
]->vdev_offline
= B_FALSE
;
2201 if (attempt_reopen
) {
2202 vdev_reopen(spa
->spa_root_vdev
);
2204 /* check each device to see what state it's in */
2205 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2206 if (vd
[i
] != NULL
&&
2207 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2214 * If every disk has been moved to the new pool, or if we never
2215 * even attempted to look at them, then we split them off for
2218 if (!attempt_reopen
|| gcount
== extracted
) {
2219 for (i
= 0; i
< gcount
; i
++)
2222 vdev_reopen(spa
->spa_root_vdev
);
2225 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2229 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2230 boolean_t mosconfig
)
2232 nvlist_t
*config
= spa
->spa_config
;
2233 char *ereport
= FM_EREPORT_ZFS_POOL
;
2239 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2240 return (SET_ERROR(EINVAL
));
2242 ASSERT(spa
->spa_comment
== NULL
);
2243 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2244 spa
->spa_comment
= spa_strdup(comment
);
2247 * Versioning wasn't explicitly added to the label until later, so if
2248 * it's not present treat it as the initial version.
2250 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2251 &spa
->spa_ubsync
.ub_version
) != 0)
2252 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2254 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2255 &spa
->spa_config_txg
);
2257 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2258 spa_guid_exists(pool_guid
, 0)) {
2259 error
= SET_ERROR(EEXIST
);
2261 spa
->spa_config_guid
= pool_guid
;
2263 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2265 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2269 nvlist_free(spa
->spa_load_info
);
2270 spa
->spa_load_info
= fnvlist_alloc();
2272 gethrestime(&spa
->spa_loaded_ts
);
2273 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2274 mosconfig
, &ereport
);
2278 * Don't count references from objsets that are already closed
2279 * and are making their way through the eviction process.
2281 spa_evicting_os_wait(spa
);
2282 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2284 if (error
!= EEXIST
) {
2285 spa
->spa_loaded_ts
.tv_sec
= 0;
2286 spa
->spa_loaded_ts
.tv_nsec
= 0;
2288 if (error
!= EBADF
) {
2289 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, NULL
, 0, 0);
2292 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2300 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2301 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2302 * spa's per-vdev ZAP list.
2305 vdev_count_verify_zaps(vdev_t
*vd
)
2307 spa_t
*spa
= vd
->vdev_spa
;
2310 if (vd
->vdev_top_zap
!= 0) {
2312 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2313 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2315 if (vd
->vdev_leaf_zap
!= 0) {
2317 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2318 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2321 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2322 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2330 * Determine whether the activity check is required.
2333 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
2337 uint64_t hostid
= 0;
2338 uint64_t tryconfig_txg
= 0;
2339 uint64_t tryconfig_timestamp
= 0;
2342 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2343 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2344 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2346 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2347 &tryconfig_timestamp
);
2350 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2353 * Disable the MMP activity check - This is used by zdb which
2354 * is intended to be used on potentially active pools.
2356 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
2360 * Skip the activity check when the MMP feature is disabled.
2362 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
2365 * If the tryconfig_* values are nonzero, they are the results of an
2366 * earlier tryimport. If they match the uberblock we just found, then
2367 * the pool has not changed and we return false so we do not test a
2370 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
2371 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
)
2375 * Allow the activity check to be skipped when importing the pool
2376 * on the same host which last imported it. Since the hostid from
2377 * configuration may be stale use the one read from the label.
2379 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
2380 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
2382 if (hostid
== spa_get_hostid())
2386 * Skip the activity test when the pool was cleanly exported.
2388 if (state
!= POOL_STATE_ACTIVE
)
2395 * Perform the import activity check. If the user canceled the import or
2396 * we detected activity then fail.
2399 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2401 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2402 uint64_t txg
= ub
->ub_txg
;
2403 uint64_t timestamp
= ub
->ub_timestamp
;
2404 uint64_t import_delay
= NANOSEC
;
2405 hrtime_t import_expire
;
2406 nvlist_t
*mmp_label
= NULL
;
2407 vdev_t
*rvd
= spa
->spa_root_vdev
;
2412 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
2413 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2417 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2418 * during the earlier tryimport. If the txg recorded there is 0 then
2419 * the pool is known to be active on another host.
2421 * Otherwise, the pool might be in use on another node. Check for
2422 * changes in the uberblocks on disk if necessary.
2424 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2425 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
2426 ZPOOL_CONFIG_LOAD_INFO
);
2428 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
2429 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
2430 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2431 error
= SET_ERROR(EREMOTEIO
);
2437 * Preferentially use the zfs_multihost_interval from the node which
2438 * last imported the pool. This value is stored in an MMP uberblock as.
2440 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2442 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
)
2443 import_delay
= MAX(import_delay
, import_intervals
*
2444 ub
->ub_mmp_delay
* MAX(vdev_count_leaves(spa
), 1));
2446 /* Apply a floor using the local default values. */
2447 import_delay
= MAX(import_delay
, import_intervals
*
2448 MSEC2NSEC(MAX(zfs_multihost_interval
, MMP_MIN_INTERVAL
)));
2450 /* Add a small random factor in case of simultaneous imports (0-25%) */
2451 import_expire
= gethrtime() + import_delay
+
2452 (import_delay
* spa_get_random(250) / 1000);
2454 while (gethrtime() < import_expire
) {
2455 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2457 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
) {
2458 error
= SET_ERROR(EREMOTEIO
);
2463 nvlist_free(mmp_label
);
2467 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
2469 error
= SET_ERROR(EINTR
);
2477 mutex_destroy(&mtx
);
2481 * If the pool is determined to be active store the status in the
2482 * spa->spa_load_info nvlist. If the remote hostname or hostid are
2483 * available from configuration read from disk store them as well.
2484 * This allows 'zpool import' to generate a more useful message.
2486 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
2487 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2488 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
2490 if (error
== EREMOTEIO
) {
2491 char *hostname
= "<unknown>";
2492 uint64_t hostid
= 0;
2495 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
2496 hostname
= fnvlist_lookup_string(mmp_label
,
2497 ZPOOL_CONFIG_HOSTNAME
);
2498 fnvlist_add_string(spa
->spa_load_info
,
2499 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
2502 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
2503 hostid
= fnvlist_lookup_uint64(mmp_label
,
2504 ZPOOL_CONFIG_HOSTID
);
2505 fnvlist_add_uint64(spa
->spa_load_info
,
2506 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
2510 fnvlist_add_uint64(spa
->spa_load_info
,
2511 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
2512 fnvlist_add_uint64(spa
->spa_load_info
,
2513 ZPOOL_CONFIG_MMP_TXG
, 0);
2515 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
2519 nvlist_free(mmp_label
);
2525 * Load an existing storage pool, using the pool's builtin spa_config as a
2526 * source of configuration information.
2528 __attribute__((always_inline
))
2530 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2531 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2535 nvlist_t
*nvroot
= NULL
;
2538 uberblock_t
*ub
= &spa
->spa_uberblock
;
2539 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2540 int orig_mode
= spa
->spa_mode
;
2543 boolean_t missing_feat_write
= B_FALSE
;
2544 boolean_t activity_check
= B_FALSE
;
2547 * If this is an untrusted config, access the pool in read-only mode.
2548 * This prevents things like resilvering recently removed devices.
2551 spa
->spa_mode
= FREAD
;
2553 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2555 spa
->spa_load_state
= state
;
2557 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2558 return (SET_ERROR(EINVAL
));
2560 parse
= (type
== SPA_IMPORT_EXISTING
?
2561 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2564 * Create "The Godfather" zio to hold all async IOs
2566 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2568 for (int i
= 0; i
< max_ncpus
; i
++) {
2569 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2570 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2571 ZIO_FLAG_GODFATHER
);
2575 * Parse the configuration into a vdev tree. We explicitly set the
2576 * value that will be returned by spa_version() since parsing the
2577 * configuration requires knowing the version number.
2579 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2580 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2581 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2586 ASSERT(spa
->spa_root_vdev
== rvd
);
2587 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2588 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2590 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2591 ASSERT(spa_guid(spa
) == pool_guid
);
2595 * Try to open all vdevs, loading each label in the process.
2597 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2598 error
= vdev_open(rvd
);
2599 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2604 * We need to validate the vdev labels against the configuration that
2605 * we have in hand, which is dependent on the setting of mosconfig. If
2606 * mosconfig is true then we're validating the vdev labels based on
2607 * that config. Otherwise, we're validating against the cached config
2608 * (zpool.cache) that was read when we loaded the zfs module, and then
2609 * later we will recursively call spa_load() and validate against
2612 * If we're assembling a new pool that's been split off from an
2613 * existing pool, the labels haven't yet been updated so we skip
2614 * validation for now.
2616 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2617 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2618 error
= vdev_validate(rvd
, mosconfig
);
2619 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2624 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2625 return (SET_ERROR(ENXIO
));
2629 * Find the best uberblock.
2631 vdev_uberblock_load(rvd
, ub
, &label
);
2634 * If we weren't able to find a single valid uberblock, return failure.
2636 if (ub
->ub_txg
== 0) {
2638 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2642 * For pools which have the multihost property on determine if the
2643 * pool is truly inactive and can be safely imported. Prevent
2644 * hosts which don't have a hostid set from importing the pool.
2646 activity_check
= spa_activity_check_required(spa
, ub
, label
, config
);
2647 if (activity_check
) {
2648 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
2649 spa_get_hostid() == 0) {
2651 fnvlist_add_uint64(spa
->spa_load_info
,
2652 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
2653 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
2656 error
= spa_activity_check(spa
, ub
, config
);
2662 fnvlist_add_uint64(spa
->spa_load_info
,
2663 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
2664 fnvlist_add_uint64(spa
->spa_load_info
,
2665 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
2669 * If the pool has an unsupported version we can't open it.
2671 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2673 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2676 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2680 * If we weren't able to find what's necessary for reading the
2681 * MOS in the label, return failure.
2683 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2684 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2686 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2691 * Update our in-core representation with the definitive values
2694 nvlist_free(spa
->spa_label_features
);
2695 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2701 * Look through entries in the label nvlist's features_for_read. If
2702 * there is a feature listed there which we don't understand then we
2703 * cannot open a pool.
2705 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2706 nvlist_t
*unsup_feat
;
2708 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2711 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
2713 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2714 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2715 VERIFY(nvlist_add_string(unsup_feat
,
2716 nvpair_name(nvp
), "") == 0);
2720 if (!nvlist_empty(unsup_feat
)) {
2721 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2722 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2723 nvlist_free(unsup_feat
);
2724 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2728 nvlist_free(unsup_feat
);
2732 * If the vdev guid sum doesn't match the uberblock, we have an
2733 * incomplete configuration. We first check to see if the pool
2734 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2735 * If it is, defer the vdev_guid_sum check till later so we
2736 * can handle missing vdevs.
2738 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2739 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2740 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2741 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2743 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2744 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2745 spa_try_repair(spa
, config
);
2746 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2747 nvlist_free(spa
->spa_config_splitting
);
2748 spa
->spa_config_splitting
= NULL
;
2752 * Initialize internal SPA structures.
2754 spa
->spa_state
= POOL_STATE_ACTIVE
;
2755 spa
->spa_ubsync
= spa
->spa_uberblock
;
2756 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2757 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2758 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2759 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2760 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2761 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2763 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2765 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2766 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2768 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2769 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2771 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2772 boolean_t missing_feat_read
= B_FALSE
;
2773 nvlist_t
*unsup_feat
, *enabled_feat
;
2775 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2776 &spa
->spa_feat_for_read_obj
) != 0) {
2777 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2780 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2781 &spa
->spa_feat_for_write_obj
) != 0) {
2782 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2785 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2786 &spa
->spa_feat_desc_obj
) != 0) {
2787 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2790 enabled_feat
= fnvlist_alloc();
2791 unsup_feat
= fnvlist_alloc();
2793 if (!spa_features_check(spa
, B_FALSE
,
2794 unsup_feat
, enabled_feat
))
2795 missing_feat_read
= B_TRUE
;
2797 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2798 if (!spa_features_check(spa
, B_TRUE
,
2799 unsup_feat
, enabled_feat
)) {
2800 missing_feat_write
= B_TRUE
;
2804 fnvlist_add_nvlist(spa
->spa_load_info
,
2805 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2807 if (!nvlist_empty(unsup_feat
)) {
2808 fnvlist_add_nvlist(spa
->spa_load_info
,
2809 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2812 fnvlist_free(enabled_feat
);
2813 fnvlist_free(unsup_feat
);
2815 if (!missing_feat_read
) {
2816 fnvlist_add_boolean(spa
->spa_load_info
,
2817 ZPOOL_CONFIG_CAN_RDONLY
);
2821 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2822 * twofold: to determine whether the pool is available for
2823 * import in read-write mode and (if it is not) whether the
2824 * pool is available for import in read-only mode. If the pool
2825 * is available for import in read-write mode, it is displayed
2826 * as available in userland; if it is not available for import
2827 * in read-only mode, it is displayed as unavailable in
2828 * userland. If the pool is available for import in read-only
2829 * mode but not read-write mode, it is displayed as unavailable
2830 * in userland with a special note that the pool is actually
2831 * available for open in read-only mode.
2833 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2834 * missing a feature for write, we must first determine whether
2835 * the pool can be opened read-only before returning to
2836 * userland in order to know whether to display the
2837 * abovementioned note.
2839 if (missing_feat_read
|| (missing_feat_write
&&
2840 spa_writeable(spa
))) {
2841 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2846 * Load refcounts for ZFS features from disk into an in-memory
2847 * cache during SPA initialization.
2849 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
2852 error
= feature_get_refcount_from_disk(spa
,
2853 &spa_feature_table
[i
], &refcount
);
2855 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2856 } else if (error
== ENOTSUP
) {
2857 spa
->spa_feat_refcount_cache
[i
] =
2858 SPA_FEATURE_DISABLED
;
2860 return (spa_vdev_err(rvd
,
2861 VDEV_AUX_CORRUPT_DATA
, EIO
));
2866 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2867 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2868 &spa
->spa_feat_enabled_txg_obj
) != 0)
2869 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2872 spa
->spa_is_initializing
= B_TRUE
;
2873 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2874 spa
->spa_is_initializing
= B_FALSE
;
2876 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2880 nvlist_t
*policy
= NULL
, *nvconfig
;
2882 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2883 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2885 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2886 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2888 unsigned long myhostid
= 0;
2890 VERIFY(nvlist_lookup_string(nvconfig
,
2891 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2893 myhostid
= spa_get_hostid();
2894 if (hostid
&& myhostid
&& hostid
!= myhostid
) {
2895 nvlist_free(nvconfig
);
2896 return (SET_ERROR(EBADF
));
2899 if (nvlist_lookup_nvlist(spa
->spa_config
,
2900 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2901 VERIFY(nvlist_add_nvlist(nvconfig
,
2902 ZPOOL_REWIND_POLICY
, policy
) == 0);
2904 spa_config_set(spa
, nvconfig
);
2906 spa_deactivate(spa
);
2907 spa_activate(spa
, orig_mode
);
2909 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2912 /* Grab the checksum salt from the MOS. */
2913 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2914 DMU_POOL_CHECKSUM_SALT
, 1,
2915 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
2916 spa
->spa_cksum_salt
.zcs_bytes
);
2917 if (error
== ENOENT
) {
2918 /* Generate a new salt for subsequent use */
2919 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
2920 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
2921 } else if (error
!= 0) {
2922 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2925 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2926 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2927 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2929 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2932 * Load the bit that tells us to use the new accounting function
2933 * (raid-z deflation). If we have an older pool, this will not
2936 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2937 if (error
!= 0 && error
!= ENOENT
)
2938 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2940 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2941 &spa
->spa_creation_version
);
2942 if (error
!= 0 && error
!= ENOENT
)
2943 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2946 * Load the persistent error log. If we have an older pool, this will
2949 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2950 if (error
!= 0 && error
!= ENOENT
)
2951 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2953 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2954 &spa
->spa_errlog_scrub
);
2955 if (error
!= 0 && error
!= ENOENT
)
2956 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2959 * Load the history object. If we have an older pool, this
2960 * will not be present.
2962 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2963 if (error
!= 0 && error
!= ENOENT
)
2964 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2967 * Load the per-vdev ZAP map. If we have an older pool, this will not
2968 * be present; in this case, defer its creation to a later time to
2969 * avoid dirtying the MOS this early / out of sync context. See
2970 * spa_sync_config_object.
2973 /* The sentinel is only available in the MOS config. */
2974 nvlist_t
*mos_config
;
2975 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
2976 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2978 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
2979 &spa
->spa_all_vdev_zaps
);
2981 if (error
== ENOENT
) {
2982 VERIFY(!nvlist_exists(mos_config
,
2983 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
2984 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
2985 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
2986 } else if (error
!= 0) {
2987 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2988 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
2990 * An older version of ZFS overwrote the sentinel value, so
2991 * we have orphaned per-vdev ZAPs in the MOS. Defer their
2992 * destruction to later; see spa_sync_config_object.
2994 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
2996 * We're assuming that no vdevs have had their ZAPs created
2997 * before this. Better be sure of it.
2999 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3001 nvlist_free(mos_config
);
3004 * If we're assembling the pool from the split-off vdevs of
3005 * an existing pool, we don't want to attach the spares & cache
3010 * Load any hot spares for this pool.
3012 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
3013 if (error
!= 0 && error
!= ENOENT
)
3014 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3015 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3016 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3017 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3018 &spa
->spa_spares
.sav_config
) != 0)
3019 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3021 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3022 spa_load_spares(spa
);
3023 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3024 } else if (error
== 0) {
3025 spa
->spa_spares
.sav_sync
= B_TRUE
;
3029 * Load any level 2 ARC devices for this pool.
3031 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3032 &spa
->spa_l2cache
.sav_object
);
3033 if (error
!= 0 && error
!= ENOENT
)
3034 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3035 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3036 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3037 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3038 &spa
->spa_l2cache
.sav_config
) != 0)
3039 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3041 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3042 spa_load_l2cache(spa
);
3043 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3044 } else if (error
== 0) {
3045 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3048 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3050 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
3051 if (error
&& error
!= ENOENT
)
3052 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3055 uint64_t autoreplace
= 0;
3057 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3058 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3059 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3060 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3061 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3062 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
3063 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3064 &spa
->spa_dedup_ditto
);
3066 spa
->spa_autoreplace
= (autoreplace
!= 0);
3070 * If the 'multihost' property is set, then never allow a pool to
3071 * be imported when the system hostid is zero. The exception to
3072 * this rule is zdb which is always allowed to access pools.
3074 if (spa_multihost(spa
) && spa_get_hostid() == 0 &&
3075 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
3076 fnvlist_add_uint64(spa
->spa_load_info
,
3077 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3078 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3082 * If the 'autoreplace' property is set, then post a resource notifying
3083 * the ZFS DE that it should not issue any faults for unopenable
3084 * devices. We also iterate over the vdevs, and post a sysevent for any
3085 * unopenable vdevs so that the normal autoreplace handler can take
3088 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
3089 spa_check_removed(spa
->spa_root_vdev
);
3091 * For the import case, this is done in spa_import(), because
3092 * at this point we're using the spare definitions from
3093 * the MOS config, not necessarily from the userland config.
3095 if (state
!= SPA_LOAD_IMPORT
) {
3096 spa_aux_check_removed(&spa
->spa_spares
);
3097 spa_aux_check_removed(&spa
->spa_l2cache
);
3102 * Load the vdev state for all toplevel vdevs.
3107 * Propagate the leaf DTLs we just loaded all the way up the tree.
3109 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3110 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3111 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3114 * Load the DDTs (dedup tables).
3116 error
= ddt_load(spa
);
3118 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3120 spa_update_dspace(spa
);
3123 * Validate the config, using the MOS config to fill in any
3124 * information which might be missing. If we fail to validate
3125 * the config then declare the pool unfit for use. If we're
3126 * assembling a pool from a split, the log is not transferred
3129 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3132 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
3133 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3135 if (!spa_config_valid(spa
, nvconfig
)) {
3136 nvlist_free(nvconfig
);
3137 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3140 nvlist_free(nvconfig
);
3143 * Now that we've validated the config, check the state of the
3144 * root vdev. If it can't be opened, it indicates one or
3145 * more toplevel vdevs are faulted.
3147 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
3148 return (SET_ERROR(ENXIO
));
3150 if (spa_writeable(spa
) && spa_check_logs(spa
)) {
3151 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3152 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
3156 if (missing_feat_write
) {
3157 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
3160 * At this point, we know that we can open the pool in
3161 * read-only mode but not read-write mode. We now have enough
3162 * information and can return to userland.
3164 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
3168 * We've successfully opened the pool, verify that we're ready
3169 * to start pushing transactions.
3171 if (state
!= SPA_LOAD_TRYIMPORT
) {
3172 if ((error
= spa_load_verify(spa
)))
3173 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3177 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
3178 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3180 int need_update
= B_FALSE
;
3181 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3183 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
3186 * Claim log blocks that haven't been committed yet.
3187 * This must all happen in a single txg.
3188 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3189 * invoked from zil_claim_log_block()'s i/o done callback.
3190 * Price of rollback is that we abandon the log.
3192 spa
->spa_claiming
= B_TRUE
;
3194 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3195 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3196 zil_claim
, tx
, DS_FIND_CHILDREN
);
3199 spa
->spa_claiming
= B_FALSE
;
3201 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3202 spa
->spa_sync_on
= B_TRUE
;
3203 txg_sync_start(spa
->spa_dsl_pool
);
3204 mmp_thread_start(spa
);
3207 * Wait for all claims to sync. We sync up to the highest
3208 * claimed log block birth time so that claimed log blocks
3209 * don't appear to be from the future. spa_claim_max_txg
3210 * will have been set for us by either zil_check_log_chain()
3211 * (invoked from spa_check_logs()) or zil_claim() above.
3213 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3216 * If the config cache is stale, or we have uninitialized
3217 * metaslabs (see spa_vdev_add()), then update the config.
3219 * If this is a verbatim import, trust the current
3220 * in-core spa_config and update the disk labels.
3222 if (config_cache_txg
!= spa
->spa_config_txg
||
3223 state
== SPA_LOAD_IMPORT
||
3224 state
== SPA_LOAD_RECOVER
||
3225 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3226 need_update
= B_TRUE
;
3228 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
3229 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3230 need_update
= B_TRUE
;
3233 * Update the config cache asychronously in case we're the
3234 * root pool, in which case the config cache isn't writable yet.
3237 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3240 * Check all DTLs to see if anything needs resilvering.
3242 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3243 vdev_resilver_needed(rvd
, NULL
, NULL
))
3244 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3247 * Log the fact that we booted up (so that we can detect if
3248 * we rebooted in the middle of an operation).
3250 spa_history_log_version(spa
, "open", NULL
);
3253 * Delete any inconsistent datasets.
3255 (void) dmu_objset_find(spa_name(spa
),
3256 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3259 * Clean up any stale temporary dataset userrefs.
3261 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3268 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
3270 int mode
= spa
->spa_mode
;
3273 spa_deactivate(spa
);
3275 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3277 spa_activate(spa
, mode
);
3278 spa_async_suspend(spa
);
3280 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
3284 * If spa_load() fails this function will try loading prior txg's. If
3285 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3286 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3287 * function will not rewind the pool and will return the same error as
3291 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
3292 uint64_t max_request
, int rewind_flags
)
3294 nvlist_t
*loadinfo
= NULL
;
3295 nvlist_t
*config
= NULL
;
3296 int load_error
, rewind_error
;
3297 uint64_t safe_rewind_txg
;
3300 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3301 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3302 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3304 spa
->spa_load_max_txg
= max_request
;
3305 if (max_request
!= UINT64_MAX
)
3306 spa
->spa_extreme_rewind
= B_TRUE
;
3309 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
3311 if (load_error
== 0)
3314 if (spa
->spa_root_vdev
!= NULL
)
3315 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3317 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3318 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3320 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3321 nvlist_free(config
);
3322 return (load_error
);
3325 if (state
== SPA_LOAD_RECOVER
) {
3326 /* Price of rolling back is discarding txgs, including log */
3327 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3330 * If we aren't rolling back save the load info from our first
3331 * import attempt so that we can restore it after attempting
3334 loadinfo
= spa
->spa_load_info
;
3335 spa
->spa_load_info
= fnvlist_alloc();
3338 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3339 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
3340 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
3341 TXG_INITIAL
: safe_rewind_txg
;
3344 * Continue as long as we're finding errors, we're still within
3345 * the acceptable rewind range, and we're still finding uberblocks
3347 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
3348 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
3349 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
3350 spa
->spa_extreme_rewind
= B_TRUE
;
3351 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
3354 spa
->spa_extreme_rewind
= B_FALSE
;
3355 spa
->spa_load_max_txg
= UINT64_MAX
;
3357 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
3358 spa_config_set(spa
, config
);
3360 nvlist_free(config
);
3362 if (state
== SPA_LOAD_RECOVER
) {
3363 ASSERT3P(loadinfo
, ==, NULL
);
3364 return (rewind_error
);
3366 /* Store the rewind info as part of the initial load info */
3367 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
3368 spa
->spa_load_info
);
3370 /* Restore the initial load info */
3371 fnvlist_free(spa
->spa_load_info
);
3372 spa
->spa_load_info
= loadinfo
;
3374 return (load_error
);
3381 * The import case is identical to an open except that the configuration is sent
3382 * down from userland, instead of grabbed from the configuration cache. For the
3383 * case of an open, the pool configuration will exist in the
3384 * POOL_STATE_UNINITIALIZED state.
3386 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3387 * the same time open the pool, without having to keep around the spa_t in some
3391 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
3395 spa_load_state_t state
= SPA_LOAD_OPEN
;
3397 int locked
= B_FALSE
;
3398 int firstopen
= B_FALSE
;
3403 * As disgusting as this is, we need to support recursive calls to this
3404 * function because dsl_dir_open() is called during spa_load(), and ends
3405 * up calling spa_open() again. The real fix is to figure out how to
3406 * avoid dsl_dir_open() calling this in the first place.
3408 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
3409 mutex_enter(&spa_namespace_lock
);
3413 if ((spa
= spa_lookup(pool
)) == NULL
) {
3415 mutex_exit(&spa_namespace_lock
);
3416 return (SET_ERROR(ENOENT
));
3419 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3420 zpool_rewind_policy_t policy
;
3424 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3426 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3427 state
= SPA_LOAD_RECOVER
;
3429 spa_activate(spa
, spa_mode_global
);
3431 if (state
!= SPA_LOAD_RECOVER
)
3432 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3434 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3435 policy
.zrp_request
);
3437 if (error
== EBADF
) {
3439 * If vdev_validate() returns failure (indicated by
3440 * EBADF), it indicates that one of the vdevs indicates
3441 * that the pool has been exported or destroyed. If
3442 * this is the case, the config cache is out of sync and
3443 * we should remove the pool from the namespace.
3446 spa_deactivate(spa
);
3447 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3450 mutex_exit(&spa_namespace_lock
);
3451 return (SET_ERROR(ENOENT
));
3456 * We can't open the pool, but we still have useful
3457 * information: the state of each vdev after the
3458 * attempted vdev_open(). Return this to the user.
3460 if (config
!= NULL
&& spa
->spa_config
) {
3461 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3463 VERIFY(nvlist_add_nvlist(*config
,
3464 ZPOOL_CONFIG_LOAD_INFO
,
3465 spa
->spa_load_info
) == 0);
3468 spa_deactivate(spa
);
3469 spa
->spa_last_open_failed
= error
;
3471 mutex_exit(&spa_namespace_lock
);
3477 spa_open_ref(spa
, tag
);
3480 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3483 * If we've recovered the pool, pass back any information we
3484 * gathered while doing the load.
3486 if (state
== SPA_LOAD_RECOVER
) {
3487 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3488 spa
->spa_load_info
) == 0);
3492 spa
->spa_last_open_failed
= 0;
3493 spa
->spa_last_ubsync_txg
= 0;
3494 spa
->spa_load_txg
= 0;
3495 mutex_exit(&spa_namespace_lock
);
3499 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
3507 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3510 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3514 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3516 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3520 * Lookup the given spa_t, incrementing the inject count in the process,
3521 * preventing it from being exported or destroyed.
3524 spa_inject_addref(char *name
)
3528 mutex_enter(&spa_namespace_lock
);
3529 if ((spa
= spa_lookup(name
)) == NULL
) {
3530 mutex_exit(&spa_namespace_lock
);
3533 spa
->spa_inject_ref
++;
3534 mutex_exit(&spa_namespace_lock
);
3540 spa_inject_delref(spa_t
*spa
)
3542 mutex_enter(&spa_namespace_lock
);
3543 spa
->spa_inject_ref
--;
3544 mutex_exit(&spa_namespace_lock
);
3548 * Add spares device information to the nvlist.
3551 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3561 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3563 if (spa
->spa_spares
.sav_count
== 0)
3566 VERIFY(nvlist_lookup_nvlist(config
,
3567 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3568 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3569 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3571 VERIFY(nvlist_add_nvlist_array(nvroot
,
3572 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3573 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3574 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3577 * Go through and find any spares which have since been
3578 * repurposed as an active spare. If this is the case, update
3579 * their status appropriately.
3581 for (i
= 0; i
< nspares
; i
++) {
3582 VERIFY(nvlist_lookup_uint64(spares
[i
],
3583 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3584 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3586 VERIFY(nvlist_lookup_uint64_array(
3587 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3588 (uint64_t **)&vs
, &vsc
) == 0);
3589 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3590 vs
->vs_aux
= VDEV_AUX_SPARED
;
3597 * Add l2cache device information to the nvlist, including vdev stats.
3600 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3603 uint_t i
, j
, nl2cache
;
3610 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3612 if (spa
->spa_l2cache
.sav_count
== 0)
3615 VERIFY(nvlist_lookup_nvlist(config
,
3616 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3617 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3618 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3619 if (nl2cache
!= 0) {
3620 VERIFY(nvlist_add_nvlist_array(nvroot
,
3621 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3622 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3623 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3626 * Update level 2 cache device stats.
3629 for (i
= 0; i
< nl2cache
; i
++) {
3630 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3631 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3634 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3636 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3637 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3643 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3644 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3646 vdev_get_stats(vd
, vs
);
3647 vdev_config_generate_stats(vd
, l2cache
[i
]);
3654 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3659 if (spa
->spa_feat_for_read_obj
!= 0) {
3660 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3661 spa
->spa_feat_for_read_obj
);
3662 zap_cursor_retrieve(&zc
, &za
) == 0;
3663 zap_cursor_advance(&zc
)) {
3664 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3665 za
.za_num_integers
== 1);
3666 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3667 za
.za_first_integer
));
3669 zap_cursor_fini(&zc
);
3672 if (spa
->spa_feat_for_write_obj
!= 0) {
3673 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3674 spa
->spa_feat_for_write_obj
);
3675 zap_cursor_retrieve(&zc
, &za
) == 0;
3676 zap_cursor_advance(&zc
)) {
3677 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3678 za
.za_num_integers
== 1);
3679 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3680 za
.za_first_integer
));
3682 zap_cursor_fini(&zc
);
3687 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3691 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3692 zfeature_info_t feature
= spa_feature_table
[i
];
3695 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3698 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3703 * Store a list of pool features and their reference counts in the
3706 * The first time this is called on a spa, allocate a new nvlist, fetch
3707 * the pool features and reference counts from disk, then save the list
3708 * in the spa. In subsequent calls on the same spa use the saved nvlist
3709 * and refresh its values from the cached reference counts. This
3710 * ensures we don't block here on I/O on a suspended pool so 'zpool
3711 * clear' can resume the pool.
3714 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3718 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3720 mutex_enter(&spa
->spa_feat_stats_lock
);
3721 features
= spa
->spa_feat_stats
;
3723 if (features
!= NULL
) {
3724 spa_feature_stats_from_cache(spa
, features
);
3726 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3727 spa
->spa_feat_stats
= features
;
3728 spa_feature_stats_from_disk(spa
, features
);
3731 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3734 mutex_exit(&spa
->spa_feat_stats_lock
);
3738 spa_get_stats(const char *name
, nvlist_t
**config
,
3739 char *altroot
, size_t buflen
)
3745 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3749 * This still leaves a window of inconsistency where the spares
3750 * or l2cache devices could change and the config would be
3751 * self-inconsistent.
3753 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3755 if (*config
!= NULL
) {
3756 uint64_t loadtimes
[2];
3758 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3759 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3760 VERIFY(nvlist_add_uint64_array(*config
,
3761 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3763 VERIFY(nvlist_add_uint64(*config
,
3764 ZPOOL_CONFIG_ERRCOUNT
,
3765 spa_get_errlog_size(spa
)) == 0);
3767 if (spa_suspended(spa
))
3768 VERIFY(nvlist_add_uint64(*config
,
3769 ZPOOL_CONFIG_SUSPENDED
,
3770 spa
->spa_failmode
) == 0);
3772 spa_add_spares(spa
, *config
);
3773 spa_add_l2cache(spa
, *config
);
3774 spa_add_feature_stats(spa
, *config
);
3779 * We want to get the alternate root even for faulted pools, so we cheat
3780 * and call spa_lookup() directly.
3784 mutex_enter(&spa_namespace_lock
);
3785 spa
= spa_lookup(name
);
3787 spa_altroot(spa
, altroot
, buflen
);
3791 mutex_exit(&spa_namespace_lock
);
3793 spa_altroot(spa
, altroot
, buflen
);
3798 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3799 spa_close(spa
, FTAG
);
3806 * Validate that the auxiliary device array is well formed. We must have an
3807 * array of nvlists, each which describes a valid leaf vdev. If this is an
3808 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3809 * specified, as long as they are well-formed.
3812 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3813 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3814 vdev_labeltype_t label
)
3821 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3824 * It's acceptable to have no devs specified.
3826 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3830 return (SET_ERROR(EINVAL
));
3833 * Make sure the pool is formatted with a version that supports this
3836 if (spa_version(spa
) < version
)
3837 return (SET_ERROR(ENOTSUP
));
3840 * Set the pending device list so we correctly handle device in-use
3843 sav
->sav_pending
= dev
;
3844 sav
->sav_npending
= ndev
;
3846 for (i
= 0; i
< ndev
; i
++) {
3847 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3851 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3853 error
= SET_ERROR(EINVAL
);
3859 if ((error
= vdev_open(vd
)) == 0 &&
3860 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3861 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3862 vd
->vdev_guid
) == 0);
3868 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3875 sav
->sav_pending
= NULL
;
3876 sav
->sav_npending
= 0;
3881 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3885 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3887 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3888 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3889 VDEV_LABEL_SPARE
)) != 0) {
3893 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3894 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3895 VDEV_LABEL_L2CACHE
));
3899 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3904 if (sav
->sav_config
!= NULL
) {
3910 * Generate new dev list by concatenating with the
3913 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3914 &olddevs
, &oldndevs
) == 0);
3916 newdevs
= kmem_alloc(sizeof (void *) *
3917 (ndevs
+ oldndevs
), KM_SLEEP
);
3918 for (i
= 0; i
< oldndevs
; i
++)
3919 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3921 for (i
= 0; i
< ndevs
; i
++)
3922 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3925 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3926 DATA_TYPE_NVLIST_ARRAY
) == 0);
3928 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3929 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3930 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3931 nvlist_free(newdevs
[i
]);
3932 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3935 * Generate a new dev list.
3937 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3939 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3945 * Stop and drop level 2 ARC devices
3948 spa_l2cache_drop(spa_t
*spa
)
3952 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3954 for (i
= 0; i
< sav
->sav_count
; i
++) {
3957 vd
= sav
->sav_vdevs
[i
];
3960 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3961 pool
!= 0ULL && l2arc_vdev_present(vd
))
3962 l2arc_remove_vdev(vd
);
3967 * Verify encryption parameters for spa creation. If we are encrypting, we must
3968 * have the encryption feature flag enabled.
3971 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
3972 boolean_t has_encryption
)
3974 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
3975 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
3977 return (SET_ERROR(ENOTSUP
));
3979 return (dmu_objset_create_crypt_check(NULL
, dcp
));
3986 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3987 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
3990 char *altroot
= NULL
;
3995 uint64_t txg
= TXG_INITIAL
;
3996 nvlist_t
**spares
, **l2cache
;
3997 uint_t nspares
, nl2cache
;
3998 uint64_t version
, obj
, root_dsobj
= 0;
3999 boolean_t has_features
;
4000 boolean_t has_encryption
;
4006 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
4007 poolname
= (char *)pool
;
4010 * If this pool already exists, return failure.
4012 mutex_enter(&spa_namespace_lock
);
4013 if (spa_lookup(poolname
) != NULL
) {
4014 mutex_exit(&spa_namespace_lock
);
4015 return (SET_ERROR(EEXIST
));
4019 * Allocate a new spa_t structure.
4021 nvl
= fnvlist_alloc();
4022 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
4023 (void) nvlist_lookup_string(props
,
4024 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4025 spa
= spa_add(poolname
, nvl
, altroot
);
4027 spa_activate(spa
, spa_mode_global
);
4029 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4030 spa_deactivate(spa
);
4032 mutex_exit(&spa_namespace_lock
);
4037 * Temporary pool names should never be written to disk.
4039 if (poolname
!= pool
)
4040 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
4042 has_features
= B_FALSE
;
4043 has_encryption
= B_FALSE
;
4044 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
4045 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4046 if (zpool_prop_feature(nvpair_name(elem
))) {
4047 has_features
= B_TRUE
;
4049 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
4050 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
4051 if (feat
== SPA_FEATURE_ENCRYPTION
)
4052 has_encryption
= B_TRUE
;
4056 /* verify encryption params, if they were provided */
4058 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
4060 spa_deactivate(spa
);
4062 mutex_exit(&spa_namespace_lock
);
4067 if (has_features
|| nvlist_lookup_uint64(props
,
4068 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4069 version
= SPA_VERSION
;
4071 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4073 spa
->spa_first_txg
= txg
;
4074 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4075 spa
->spa_uberblock
.ub_version
= version
;
4076 spa
->spa_ubsync
= spa
->spa_uberblock
;
4077 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4080 * Create "The Godfather" zio to hold all async IOs
4082 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4084 for (int i
= 0; i
< max_ncpus
; i
++) {
4085 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4086 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4087 ZIO_FLAG_GODFATHER
);
4091 * Create the root vdev.
4093 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4095 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4097 ASSERT(error
!= 0 || rvd
!= NULL
);
4098 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4100 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4101 error
= SET_ERROR(EINVAL
);
4104 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4105 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4106 VDEV_ALLOC_ADD
)) == 0) {
4107 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
4108 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4109 vdev_expand(rvd
->vdev_child
[c
], txg
);
4113 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4117 spa_deactivate(spa
);
4119 mutex_exit(&spa_namespace_lock
);
4124 * Get the list of spares, if specified.
4126 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4127 &spares
, &nspares
) == 0) {
4128 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4130 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4131 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4132 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4133 spa_load_spares(spa
);
4134 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4135 spa
->spa_spares
.sav_sync
= B_TRUE
;
4139 * Get the list of level 2 cache devices, if specified.
4141 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4142 &l2cache
, &nl2cache
) == 0) {
4143 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4144 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4145 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4146 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4147 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4148 spa_load_l2cache(spa
);
4149 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4150 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4153 spa
->spa_is_initializing
= B_TRUE
;
4154 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
4155 spa
->spa_is_initializing
= B_FALSE
;
4158 * Create DDTs (dedup tables).
4162 spa_update_dspace(spa
);
4164 tx
= dmu_tx_create_assigned(dp
, txg
);
4167 * Create the pool's history object.
4169 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
4170 spa_history_create_obj(spa
, tx
);
4172 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4173 spa_history_log_version(spa
, "create", tx
);
4176 * Create the pool config object.
4178 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4179 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4180 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4182 if (zap_add(spa
->spa_meta_objset
,
4183 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4184 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4185 cmn_err(CE_PANIC
, "failed to add pool config");
4188 if (zap_add(spa
->spa_meta_objset
,
4189 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4190 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4191 cmn_err(CE_PANIC
, "failed to add pool version");
4194 /* Newly created pools with the right version are always deflated. */
4195 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4196 spa
->spa_deflate
= TRUE
;
4197 if (zap_add(spa
->spa_meta_objset
,
4198 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4199 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4200 cmn_err(CE_PANIC
, "failed to add deflate");
4205 * Create the deferred-free bpobj. Turn off compression
4206 * because sync-to-convergence takes longer if the blocksize
4209 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4210 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4211 ZIO_COMPRESS_OFF
, tx
);
4212 if (zap_add(spa
->spa_meta_objset
,
4213 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4214 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4215 cmn_err(CE_PANIC
, "failed to add bpobj");
4217 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4218 spa
->spa_meta_objset
, obj
));
4221 * Generate some random noise for salted checksums to operate on.
4223 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4224 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4227 * Set pool properties.
4229 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4230 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4231 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4232 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4233 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
4235 if (props
!= NULL
) {
4236 spa_configfile_set(spa
, props
, B_FALSE
);
4237 spa_sync_props(props
, tx
);
4243 * If the root dataset is encrypted we will need to create key mappings
4244 * for the zio layer before we start to write any data to disk and hold
4245 * them until after the first txg has been synced. Waiting for the first
4246 * transaction to complete also ensures that our bean counters are
4247 * appropriately updated.
4249 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0) {
4250 root_dsobj
= dsl_dir_phys(dp
->dp_root_dir
)->dd_head_dataset_obj
;
4251 VERIFY0(spa_keystore_create_mapping_impl(spa
, root_dsobj
,
4252 dp
->dp_root_dir
, FTAG
));
4255 spa
->spa_sync_on
= B_TRUE
;
4257 mmp_thread_start(spa
);
4258 txg_wait_synced(dp
, txg
);
4260 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0)
4261 VERIFY0(spa_keystore_remove_mapping(spa
, root_dsobj
, FTAG
));
4263 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4266 * Don't count references from objsets that are already closed
4267 * and are making their way through the eviction process.
4269 spa_evicting_os_wait(spa
);
4270 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4271 spa
->spa_load_state
= SPA_LOAD_NONE
;
4273 mutex_exit(&spa_namespace_lock
);
4279 * Import a non-root pool into the system.
4282 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4285 char *altroot
= NULL
;
4286 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4287 zpool_rewind_policy_t policy
;
4288 uint64_t mode
= spa_mode_global
;
4289 uint64_t readonly
= B_FALSE
;
4292 nvlist_t
**spares
, **l2cache
;
4293 uint_t nspares
, nl2cache
;
4296 * If a pool with this name exists, return failure.
4298 mutex_enter(&spa_namespace_lock
);
4299 if (spa_lookup(pool
) != NULL
) {
4300 mutex_exit(&spa_namespace_lock
);
4301 return (SET_ERROR(EEXIST
));
4305 * Create and initialize the spa structure.
4307 (void) nvlist_lookup_string(props
,
4308 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4309 (void) nvlist_lookup_uint64(props
,
4310 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4313 spa
= spa_add(pool
, config
, altroot
);
4314 spa
->spa_import_flags
= flags
;
4317 * Verbatim import - Take a pool and insert it into the namespace
4318 * as if it had been loaded at boot.
4320 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4322 spa_configfile_set(spa
, props
, B_FALSE
);
4324 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4325 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4327 mutex_exit(&spa_namespace_lock
);
4331 spa_activate(spa
, mode
);
4334 * Don't start async tasks until we know everything is healthy.
4336 spa_async_suspend(spa
);
4338 zpool_get_rewind_policy(config
, &policy
);
4339 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4340 state
= SPA_LOAD_RECOVER
;
4343 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4344 * because the user-supplied config is actually the one to trust when
4347 if (state
!= SPA_LOAD_RECOVER
)
4348 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4350 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4351 policy
.zrp_request
);
4354 * Propagate anything learned while loading the pool and pass it
4355 * back to caller (i.e. rewind info, missing devices, etc).
4357 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4358 spa
->spa_load_info
) == 0);
4360 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4362 * Toss any existing sparelist, as it doesn't have any validity
4363 * anymore, and conflicts with spa_has_spare().
4365 if (spa
->spa_spares
.sav_config
) {
4366 nvlist_free(spa
->spa_spares
.sav_config
);
4367 spa
->spa_spares
.sav_config
= NULL
;
4368 spa_load_spares(spa
);
4370 if (spa
->spa_l2cache
.sav_config
) {
4371 nvlist_free(spa
->spa_l2cache
.sav_config
);
4372 spa
->spa_l2cache
.sav_config
= NULL
;
4373 spa_load_l2cache(spa
);
4376 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4378 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4381 spa_configfile_set(spa
, props
, B_FALSE
);
4383 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4384 (error
= spa_prop_set(spa
, props
)))) {
4386 spa_deactivate(spa
);
4388 mutex_exit(&spa_namespace_lock
);
4392 spa_async_resume(spa
);
4395 * Override any spares and level 2 cache devices as specified by
4396 * the user, as these may have correct device names/devids, etc.
4398 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4399 &spares
, &nspares
) == 0) {
4400 if (spa
->spa_spares
.sav_config
)
4401 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4402 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4404 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4405 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4406 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4407 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4408 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4409 spa_load_spares(spa
);
4410 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4411 spa
->spa_spares
.sav_sync
= B_TRUE
;
4413 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4414 &l2cache
, &nl2cache
) == 0) {
4415 if (spa
->spa_l2cache
.sav_config
)
4416 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4417 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4419 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4420 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4421 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4422 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4423 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4424 spa_load_l2cache(spa
);
4425 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4426 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4430 * Check for any removed devices.
4432 if (spa
->spa_autoreplace
) {
4433 spa_aux_check_removed(&spa
->spa_spares
);
4434 spa_aux_check_removed(&spa
->spa_l2cache
);
4437 if (spa_writeable(spa
)) {
4439 * Update the config cache to include the newly-imported pool.
4441 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4445 * It's possible that the pool was expanded while it was exported.
4446 * We kick off an async task to handle this for us.
4448 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4450 spa_history_log_version(spa
, "import", NULL
);
4452 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4454 zvol_create_minors(spa
, pool
, B_TRUE
);
4456 mutex_exit(&spa_namespace_lock
);
4462 spa_tryimport(nvlist_t
*tryconfig
)
4464 nvlist_t
*config
= NULL
;
4470 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4473 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4477 * Create and initialize the spa structure.
4479 mutex_enter(&spa_namespace_lock
);
4480 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4481 spa_activate(spa
, FREAD
);
4484 * Pass off the heavy lifting to spa_load().
4485 * Pass TRUE for mosconfig because the user-supplied config
4486 * is actually the one to trust when doing an import.
4488 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4491 * If 'tryconfig' was at least parsable, return the current config.
4493 if (spa
->spa_root_vdev
!= NULL
) {
4494 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4495 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4497 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4499 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4500 spa
->spa_uberblock
.ub_timestamp
) == 0);
4501 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4502 spa
->spa_load_info
) == 0);
4503 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4504 spa
->spa_errata
) == 0);
4507 * If the bootfs property exists on this pool then we
4508 * copy it out so that external consumers can tell which
4509 * pools are bootable.
4511 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4512 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4515 * We have to play games with the name since the
4516 * pool was opened as TRYIMPORT_NAME.
4518 if (dsl_dsobj_to_dsname(spa_name(spa
),
4519 spa
->spa_bootfs
, tmpname
) == 0) {
4523 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4525 cp
= strchr(tmpname
, '/');
4527 (void) strlcpy(dsname
, tmpname
,
4530 (void) snprintf(dsname
, MAXPATHLEN
,
4531 "%s/%s", poolname
, ++cp
);
4533 VERIFY(nvlist_add_string(config
,
4534 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4535 kmem_free(dsname
, MAXPATHLEN
);
4537 kmem_free(tmpname
, MAXPATHLEN
);
4541 * Add the list of hot spares and level 2 cache devices.
4543 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4544 spa_add_spares(spa
, config
);
4545 spa_add_l2cache(spa
, config
);
4546 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4550 spa_deactivate(spa
);
4552 mutex_exit(&spa_namespace_lock
);
4558 * Pool export/destroy
4560 * The act of destroying or exporting a pool is very simple. We make sure there
4561 * is no more pending I/O and any references to the pool are gone. Then, we
4562 * update the pool state and sync all the labels to disk, removing the
4563 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4564 * we don't sync the labels or remove the configuration cache.
4567 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4568 boolean_t force
, boolean_t hardforce
)
4575 if (!(spa_mode_global
& FWRITE
))
4576 return (SET_ERROR(EROFS
));
4578 mutex_enter(&spa_namespace_lock
);
4579 if ((spa
= spa_lookup(pool
)) == NULL
) {
4580 mutex_exit(&spa_namespace_lock
);
4581 return (SET_ERROR(ENOENT
));
4585 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4586 * reacquire the namespace lock, and see if we can export.
4588 spa_open_ref(spa
, FTAG
);
4589 mutex_exit(&spa_namespace_lock
);
4590 spa_async_suspend(spa
);
4591 if (spa
->spa_zvol_taskq
) {
4592 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
4593 taskq_wait(spa
->spa_zvol_taskq
);
4595 mutex_enter(&spa_namespace_lock
);
4596 spa_close(spa
, FTAG
);
4598 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4601 * The pool will be in core if it's openable, in which case we can
4602 * modify its state. Objsets may be open only because they're dirty,
4603 * so we have to force it to sync before checking spa_refcnt.
4605 if (spa
->spa_sync_on
) {
4606 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4607 spa_evicting_os_wait(spa
);
4611 * A pool cannot be exported or destroyed if there are active
4612 * references. If we are resetting a pool, allow references by
4613 * fault injection handlers.
4615 if (!spa_refcount_zero(spa
) ||
4616 (spa
->spa_inject_ref
!= 0 &&
4617 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4618 spa_async_resume(spa
);
4619 mutex_exit(&spa_namespace_lock
);
4620 return (SET_ERROR(EBUSY
));
4623 if (spa
->spa_sync_on
) {
4625 * A pool cannot be exported if it has an active shared spare.
4626 * This is to prevent other pools stealing the active spare
4627 * from an exported pool. At user's own will, such pool can
4628 * be forcedly exported.
4630 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4631 spa_has_active_shared_spare(spa
)) {
4632 spa_async_resume(spa
);
4633 mutex_exit(&spa_namespace_lock
);
4634 return (SET_ERROR(EXDEV
));
4638 * We want this to be reflected on every label,
4639 * so mark them all dirty. spa_unload() will do the
4640 * final sync that pushes these changes out.
4642 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4643 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4644 spa
->spa_state
= new_state
;
4645 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4647 vdev_config_dirty(spa
->spa_root_vdev
);
4648 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4653 if (new_state
== POOL_STATE_DESTROYED
)
4654 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
4655 else if (new_state
== POOL_STATE_EXPORTED
)
4656 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
4658 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4660 spa_deactivate(spa
);
4663 if (oldconfig
&& spa
->spa_config
)
4664 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4666 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4668 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4671 mutex_exit(&spa_namespace_lock
);
4677 * Destroy a storage pool.
4680 spa_destroy(char *pool
)
4682 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4687 * Export a storage pool.
4690 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4691 boolean_t hardforce
)
4693 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4698 * Similar to spa_export(), this unloads the spa_t without actually removing it
4699 * from the namespace in any way.
4702 spa_reset(char *pool
)
4704 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4709 * ==========================================================================
4710 * Device manipulation
4711 * ==========================================================================
4715 * Add a device to a storage pool.
4718 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4722 vdev_t
*rvd
= spa
->spa_root_vdev
;
4724 nvlist_t
**spares
, **l2cache
;
4725 uint_t nspares
, nl2cache
;
4727 ASSERT(spa_writeable(spa
));
4729 txg
= spa_vdev_enter(spa
);
4731 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4732 VDEV_ALLOC_ADD
)) != 0)
4733 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4735 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4737 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4741 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4745 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4746 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4748 if (vd
->vdev_children
!= 0 &&
4749 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4750 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4753 * We must validate the spares and l2cache devices after checking the
4754 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4756 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4757 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4760 * Transfer each new top-level vdev from vd to rvd.
4762 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4765 * Set the vdev id to the first hole, if one exists.
4767 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4768 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4769 vdev_free(rvd
->vdev_child
[id
]);
4773 tvd
= vd
->vdev_child
[c
];
4774 vdev_remove_child(vd
, tvd
);
4776 vdev_add_child(rvd
, tvd
);
4777 vdev_config_dirty(tvd
);
4781 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4782 ZPOOL_CONFIG_SPARES
);
4783 spa_load_spares(spa
);
4784 spa
->spa_spares
.sav_sync
= B_TRUE
;
4787 if (nl2cache
!= 0) {
4788 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4789 ZPOOL_CONFIG_L2CACHE
);
4790 spa_load_l2cache(spa
);
4791 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4795 * We have to be careful when adding new vdevs to an existing pool.
4796 * If other threads start allocating from these vdevs before we
4797 * sync the config cache, and we lose power, then upon reboot we may
4798 * fail to open the pool because there are DVAs that the config cache
4799 * can't translate. Therefore, we first add the vdevs without
4800 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4801 * and then let spa_config_update() initialize the new metaslabs.
4803 * spa_load() checks for added-but-not-initialized vdevs, so that
4804 * if we lose power at any point in this sequence, the remaining
4805 * steps will be completed the next time we load the pool.
4807 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4809 mutex_enter(&spa_namespace_lock
);
4810 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4811 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
4812 mutex_exit(&spa_namespace_lock
);
4818 * Attach a device to a mirror. The arguments are the path to any device
4819 * in the mirror, and the nvroot for the new device. If the path specifies
4820 * a device that is not mirrored, we automatically insert the mirror vdev.
4822 * If 'replacing' is specified, the new device is intended to replace the
4823 * existing device; in this case the two devices are made into their own
4824 * mirror using the 'replacing' vdev, which is functionally identical to
4825 * the mirror vdev (it actually reuses all the same ops) but has a few
4826 * extra rules: you can't attach to it after it's been created, and upon
4827 * completion of resilvering, the first disk (the one being replaced)
4828 * is automatically detached.
4831 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4833 uint64_t txg
, dtl_max_txg
;
4834 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4835 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4837 char *oldvdpath
, *newvdpath
;
4841 ASSERT(spa_writeable(spa
));
4843 txg
= spa_vdev_enter(spa
);
4845 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4848 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4850 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4851 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4853 pvd
= oldvd
->vdev_parent
;
4855 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4856 VDEV_ALLOC_ATTACH
)) != 0)
4857 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4859 if (newrootvd
->vdev_children
!= 1)
4860 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4862 newvd
= newrootvd
->vdev_child
[0];
4864 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4865 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4867 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4868 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4871 * Spares can't replace logs
4873 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4874 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4878 * For attach, the only allowable parent is a mirror or the root
4881 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4882 pvd
->vdev_ops
!= &vdev_root_ops
)
4883 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4885 pvops
= &vdev_mirror_ops
;
4888 * Active hot spares can only be replaced by inactive hot
4891 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4892 oldvd
->vdev_isspare
&&
4893 !spa_has_spare(spa
, newvd
->vdev_guid
))
4894 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4897 * If the source is a hot spare, and the parent isn't already a
4898 * spare, then we want to create a new hot spare. Otherwise, we
4899 * want to create a replacing vdev. The user is not allowed to
4900 * attach to a spared vdev child unless the 'isspare' state is
4901 * the same (spare replaces spare, non-spare replaces
4904 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4905 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4906 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4907 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4908 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4909 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4912 if (newvd
->vdev_isspare
)
4913 pvops
= &vdev_spare_ops
;
4915 pvops
= &vdev_replacing_ops
;
4919 * Make sure the new device is big enough.
4921 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4922 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4925 * The new device cannot have a higher alignment requirement
4926 * than the top-level vdev.
4928 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4929 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4932 * If this is an in-place replacement, update oldvd's path and devid
4933 * to make it distinguishable from newvd, and unopenable from now on.
4935 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4936 spa_strfree(oldvd
->vdev_path
);
4937 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4939 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4940 newvd
->vdev_path
, "old");
4941 if (oldvd
->vdev_devid
!= NULL
) {
4942 spa_strfree(oldvd
->vdev_devid
);
4943 oldvd
->vdev_devid
= NULL
;
4947 /* mark the device being resilvered */
4948 newvd
->vdev_resilver_txg
= txg
;
4951 * If the parent is not a mirror, or if we're replacing, insert the new
4952 * mirror/replacing/spare vdev above oldvd.
4954 if (pvd
->vdev_ops
!= pvops
)
4955 pvd
= vdev_add_parent(oldvd
, pvops
);
4957 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4958 ASSERT(pvd
->vdev_ops
== pvops
);
4959 ASSERT(oldvd
->vdev_parent
== pvd
);
4962 * Extract the new device from its root and add it to pvd.
4964 vdev_remove_child(newrootvd
, newvd
);
4965 newvd
->vdev_id
= pvd
->vdev_children
;
4966 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4967 vdev_add_child(pvd
, newvd
);
4970 * Reevaluate the parent vdev state.
4972 vdev_propagate_state(pvd
);
4974 tvd
= newvd
->vdev_top
;
4975 ASSERT(pvd
->vdev_top
== tvd
);
4976 ASSERT(tvd
->vdev_parent
== rvd
);
4978 vdev_config_dirty(tvd
);
4981 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4982 * for any dmu_sync-ed blocks. It will propagate upward when
4983 * spa_vdev_exit() calls vdev_dtl_reassess().
4985 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4987 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4988 dtl_max_txg
- TXG_INITIAL
);
4990 if (newvd
->vdev_isspare
) {
4991 spa_spare_activate(newvd
);
4992 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
4995 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4996 newvdpath
= spa_strdup(newvd
->vdev_path
);
4997 newvd_isspare
= newvd
->vdev_isspare
;
5000 * Mark newvd's DTL dirty in this txg.
5002 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
5005 * Schedule the resilver to restart in the future. We do this to
5006 * ensure that dmu_sync-ed blocks have been stitched into the
5007 * respective datasets.
5009 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
5011 if (spa
->spa_bootfs
)
5012 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
5014 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
5019 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
5021 spa_history_log_internal(spa
, "vdev attach", NULL
,
5022 "%s vdev=%s %s vdev=%s",
5023 replacing
&& newvd_isspare
? "spare in" :
5024 replacing
? "replace" : "attach", newvdpath
,
5025 replacing
? "for" : "to", oldvdpath
);
5027 spa_strfree(oldvdpath
);
5028 spa_strfree(newvdpath
);
5034 * Detach a device from a mirror or replacing vdev.
5036 * If 'replace_done' is specified, only detach if the parent
5037 * is a replacing vdev.
5040 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5044 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5045 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5046 boolean_t unspare
= B_FALSE
;
5047 uint64_t unspare_guid
= 0;
5050 ASSERT(spa_writeable(spa
));
5052 txg
= spa_vdev_enter(spa
);
5054 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5057 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5059 if (!vd
->vdev_ops
->vdev_op_leaf
)
5060 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5062 pvd
= vd
->vdev_parent
;
5065 * If the parent/child relationship is not as expected, don't do it.
5066 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5067 * vdev that's replacing B with C. The user's intent in replacing
5068 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5069 * the replace by detaching C, the expected behavior is to end up
5070 * M(A,B). But suppose that right after deciding to detach C,
5071 * the replacement of B completes. We would have M(A,C), and then
5072 * ask to detach C, which would leave us with just A -- not what
5073 * the user wanted. To prevent this, we make sure that the
5074 * parent/child relationship hasn't changed -- in this example,
5075 * that C's parent is still the replacing vdev R.
5077 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5078 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5081 * Only 'replacing' or 'spare' vdevs can be replaced.
5083 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5084 pvd
->vdev_ops
!= &vdev_spare_ops
)
5085 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5087 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5088 spa_version(spa
) >= SPA_VERSION_SPARES
);
5091 * Only mirror, replacing, and spare vdevs support detach.
5093 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5094 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5095 pvd
->vdev_ops
!= &vdev_spare_ops
)
5096 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5099 * If this device has the only valid copy of some data,
5100 * we cannot safely detach it.
5102 if (vdev_dtl_required(vd
))
5103 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5105 ASSERT(pvd
->vdev_children
>= 2);
5108 * If we are detaching the second disk from a replacing vdev, then
5109 * check to see if we changed the original vdev's path to have "/old"
5110 * at the end in spa_vdev_attach(). If so, undo that change now.
5112 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5113 vd
->vdev_path
!= NULL
) {
5114 size_t len
= strlen(vd
->vdev_path
);
5116 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
5117 cvd
= pvd
->vdev_child
[c
];
5119 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5122 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5123 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5124 spa_strfree(cvd
->vdev_path
);
5125 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5132 * If we are detaching the original disk from a spare, then it implies
5133 * that the spare should become a real disk, and be removed from the
5134 * active spare list for the pool.
5136 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5138 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5142 * Erase the disk labels so the disk can be used for other things.
5143 * This must be done after all other error cases are handled,
5144 * but before we disembowel vd (so we can still do I/O to it).
5145 * But if we can't do it, don't treat the error as fatal --
5146 * it may be that the unwritability of the disk is the reason
5147 * it's being detached!
5149 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5152 * Remove vd from its parent and compact the parent's children.
5154 vdev_remove_child(pvd
, vd
);
5155 vdev_compact_children(pvd
);
5158 * Remember one of the remaining children so we can get tvd below.
5160 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5163 * If we need to remove the remaining child from the list of hot spares,
5164 * do it now, marking the vdev as no longer a spare in the process.
5165 * We must do this before vdev_remove_parent(), because that can
5166 * change the GUID if it creates a new toplevel GUID. For a similar
5167 * reason, we must remove the spare now, in the same txg as the detach;
5168 * otherwise someone could attach a new sibling, change the GUID, and
5169 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5172 ASSERT(cvd
->vdev_isspare
);
5173 spa_spare_remove(cvd
);
5174 unspare_guid
= cvd
->vdev_guid
;
5175 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5176 cvd
->vdev_unspare
= B_TRUE
;
5180 * If the parent mirror/replacing vdev only has one child,
5181 * the parent is no longer needed. Remove it from the tree.
5183 if (pvd
->vdev_children
== 1) {
5184 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5185 cvd
->vdev_unspare
= B_FALSE
;
5186 vdev_remove_parent(cvd
);
5191 * We don't set tvd until now because the parent we just removed
5192 * may have been the previous top-level vdev.
5194 tvd
= cvd
->vdev_top
;
5195 ASSERT(tvd
->vdev_parent
== rvd
);
5198 * Reevaluate the parent vdev state.
5200 vdev_propagate_state(cvd
);
5203 * If the 'autoexpand' property is set on the pool then automatically
5204 * try to expand the size of the pool. For example if the device we
5205 * just detached was smaller than the others, it may be possible to
5206 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5207 * first so that we can obtain the updated sizes of the leaf vdevs.
5209 if (spa
->spa_autoexpand
) {
5211 vdev_expand(tvd
, txg
);
5214 vdev_config_dirty(tvd
);
5217 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5218 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5219 * But first make sure we're not on any *other* txg's DTL list, to
5220 * prevent vd from being accessed after it's freed.
5222 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
5223 for (int t
= 0; t
< TXG_SIZE
; t
++)
5224 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5225 vd
->vdev_detached
= B_TRUE
;
5226 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5228 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5230 /* hang on to the spa before we release the lock */
5231 spa_open_ref(spa
, FTAG
);
5233 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5235 spa_history_log_internal(spa
, "detach", NULL
,
5237 spa_strfree(vdpath
);
5240 * If this was the removal of the original device in a hot spare vdev,
5241 * then we want to go through and remove the device from the hot spare
5242 * list of every other pool.
5245 spa_t
*altspa
= NULL
;
5247 mutex_enter(&spa_namespace_lock
);
5248 while ((altspa
= spa_next(altspa
)) != NULL
) {
5249 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5253 spa_open_ref(altspa
, FTAG
);
5254 mutex_exit(&spa_namespace_lock
);
5255 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5256 mutex_enter(&spa_namespace_lock
);
5257 spa_close(altspa
, FTAG
);
5259 mutex_exit(&spa_namespace_lock
);
5261 /* search the rest of the vdevs for spares to remove */
5262 spa_vdev_resilver_done(spa
);
5265 /* all done with the spa; OK to release */
5266 mutex_enter(&spa_namespace_lock
);
5267 spa_close(spa
, FTAG
);
5268 mutex_exit(&spa_namespace_lock
);
5274 * Split a set of devices from their mirrors, and create a new pool from them.
5277 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5278 nvlist_t
*props
, boolean_t exp
)
5281 uint64_t txg
, *glist
;
5283 uint_t c
, children
, lastlog
;
5284 nvlist_t
**child
, *nvl
, *tmp
;
5286 char *altroot
= NULL
;
5287 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5288 boolean_t activate_slog
;
5290 ASSERT(spa_writeable(spa
));
5292 txg
= spa_vdev_enter(spa
);
5294 /* clear the log and flush everything up to now */
5295 activate_slog
= spa_passivate_log(spa
);
5296 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5297 error
= spa_offline_log(spa
);
5298 txg
= spa_vdev_config_enter(spa
);
5301 spa_activate_log(spa
);
5304 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5306 /* check new spa name before going any further */
5307 if (spa_lookup(newname
) != NULL
)
5308 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5311 * scan through all the children to ensure they're all mirrors
5313 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5314 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5316 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5318 /* first, check to ensure we've got the right child count */
5319 rvd
= spa
->spa_root_vdev
;
5321 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5322 vdev_t
*vd
= rvd
->vdev_child
[c
];
5324 /* don't count the holes & logs as children */
5325 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5333 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5334 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5336 /* next, ensure no spare or cache devices are part of the split */
5337 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5338 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5339 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5341 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5342 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5344 /* then, loop over each vdev and validate it */
5345 for (c
= 0; c
< children
; c
++) {
5346 uint64_t is_hole
= 0;
5348 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5352 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5353 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5356 error
= SET_ERROR(EINVAL
);
5361 /* which disk is going to be split? */
5362 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5364 error
= SET_ERROR(EINVAL
);
5368 /* look it up in the spa */
5369 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5370 if (vml
[c
] == NULL
) {
5371 error
= SET_ERROR(ENODEV
);
5375 /* make sure there's nothing stopping the split */
5376 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5377 vml
[c
]->vdev_islog
||
5378 vml
[c
]->vdev_ishole
||
5379 vml
[c
]->vdev_isspare
||
5380 vml
[c
]->vdev_isl2cache
||
5381 !vdev_writeable(vml
[c
]) ||
5382 vml
[c
]->vdev_children
!= 0 ||
5383 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5384 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5385 error
= SET_ERROR(EINVAL
);
5389 if (vdev_dtl_required(vml
[c
])) {
5390 error
= SET_ERROR(EBUSY
);
5394 /* we need certain info from the top level */
5395 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5396 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5397 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5398 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5399 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5400 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5401 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5402 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5404 /* transfer per-vdev ZAPs */
5405 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
5406 VERIFY0(nvlist_add_uint64(child
[c
],
5407 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
5409 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
5410 VERIFY0(nvlist_add_uint64(child
[c
],
5411 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
5412 vml
[c
]->vdev_parent
->vdev_top_zap
));
5416 kmem_free(vml
, children
* sizeof (vdev_t
*));
5417 kmem_free(glist
, children
* sizeof (uint64_t));
5418 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5421 /* stop writers from using the disks */
5422 for (c
= 0; c
< children
; c
++) {
5424 vml
[c
]->vdev_offline
= B_TRUE
;
5426 vdev_reopen(spa
->spa_root_vdev
);
5429 * Temporarily record the splitting vdevs in the spa config. This
5430 * will disappear once the config is regenerated.
5432 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5433 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5434 glist
, children
) == 0);
5435 kmem_free(glist
, children
* sizeof (uint64_t));
5437 mutex_enter(&spa
->spa_props_lock
);
5438 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5440 mutex_exit(&spa
->spa_props_lock
);
5441 spa
->spa_config_splitting
= nvl
;
5442 vdev_config_dirty(spa
->spa_root_vdev
);
5444 /* configure and create the new pool */
5445 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5446 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5447 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5448 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5449 spa_version(spa
)) == 0);
5450 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5451 spa
->spa_config_txg
) == 0);
5452 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5453 spa_generate_guid(NULL
)) == 0);
5454 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
5455 (void) nvlist_lookup_string(props
,
5456 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5458 /* add the new pool to the namespace */
5459 newspa
= spa_add(newname
, config
, altroot
);
5460 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5461 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5462 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5464 /* release the spa config lock, retaining the namespace lock */
5465 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5467 if (zio_injection_enabled
)
5468 zio_handle_panic_injection(spa
, FTAG
, 1);
5470 spa_activate(newspa
, spa_mode_global
);
5471 spa_async_suspend(newspa
);
5473 /* create the new pool from the disks of the original pool */
5474 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5478 /* if that worked, generate a real config for the new pool */
5479 if (newspa
->spa_root_vdev
!= NULL
) {
5480 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5481 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5482 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5483 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5484 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5489 if (props
!= NULL
) {
5490 spa_configfile_set(newspa
, props
, B_FALSE
);
5491 error
= spa_prop_set(newspa
, props
);
5496 /* flush everything */
5497 txg
= spa_vdev_config_enter(newspa
);
5498 vdev_config_dirty(newspa
->spa_root_vdev
);
5499 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5501 if (zio_injection_enabled
)
5502 zio_handle_panic_injection(spa
, FTAG
, 2);
5504 spa_async_resume(newspa
);
5506 /* finally, update the original pool's config */
5507 txg
= spa_vdev_config_enter(spa
);
5508 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5509 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5512 for (c
= 0; c
< children
; c
++) {
5513 if (vml
[c
] != NULL
) {
5516 spa_history_log_internal(spa
, "detach", tx
,
5517 "vdev=%s", vml
[c
]->vdev_path
);
5522 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5523 vdev_config_dirty(spa
->spa_root_vdev
);
5524 spa
->spa_config_splitting
= NULL
;
5528 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5530 if (zio_injection_enabled
)
5531 zio_handle_panic_injection(spa
, FTAG
, 3);
5533 /* split is complete; log a history record */
5534 spa_history_log_internal(newspa
, "split", NULL
,
5535 "from pool %s", spa_name(spa
));
5537 kmem_free(vml
, children
* sizeof (vdev_t
*));
5539 /* if we're not going to mount the filesystems in userland, export */
5541 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5548 spa_deactivate(newspa
);
5551 txg
= spa_vdev_config_enter(spa
);
5553 /* re-online all offlined disks */
5554 for (c
= 0; c
< children
; c
++) {
5556 vml
[c
]->vdev_offline
= B_FALSE
;
5558 vdev_reopen(spa
->spa_root_vdev
);
5560 nvlist_free(spa
->spa_config_splitting
);
5561 spa
->spa_config_splitting
= NULL
;
5562 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5564 kmem_free(vml
, children
* sizeof (vdev_t
*));
5569 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5571 for (int i
= 0; i
< count
; i
++) {
5574 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5577 if (guid
== target_guid
)
5585 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5586 nvlist_t
*dev_to_remove
)
5588 nvlist_t
**newdev
= NULL
;
5591 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5593 for (int i
= 0, j
= 0; i
< count
; i
++) {
5594 if (dev
[i
] == dev_to_remove
)
5596 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5599 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5600 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5602 for (int i
= 0; i
< count
- 1; i
++)
5603 nvlist_free(newdev
[i
]);
5606 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5610 * Evacuate the device.
5613 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5618 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5619 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5620 ASSERT(vd
== vd
->vdev_top
);
5623 * Evacuate the device. We don't hold the config lock as writer
5624 * since we need to do I/O but we do keep the
5625 * spa_namespace_lock held. Once this completes the device
5626 * should no longer have any blocks allocated on it.
5628 if (vd
->vdev_islog
) {
5629 if (vd
->vdev_stat
.vs_alloc
!= 0)
5630 error
= spa_offline_log(spa
);
5632 error
= SET_ERROR(ENOTSUP
);
5639 * The evacuation succeeded. Remove any remaining MOS metadata
5640 * associated with this vdev, and wait for these changes to sync.
5642 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5643 txg
= spa_vdev_config_enter(spa
);
5644 vd
->vdev_removing
= B_TRUE
;
5645 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5646 vdev_config_dirty(vd
);
5647 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5653 * Complete the removal by cleaning up the namespace.
5656 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5658 vdev_t
*rvd
= spa
->spa_root_vdev
;
5659 uint64_t id
= vd
->vdev_id
;
5660 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5662 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5663 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5664 ASSERT(vd
== vd
->vdev_top
);
5667 * Only remove any devices which are empty.
5669 if (vd
->vdev_stat
.vs_alloc
!= 0)
5672 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5674 if (list_link_active(&vd
->vdev_state_dirty_node
))
5675 vdev_state_clean(vd
);
5676 if (list_link_active(&vd
->vdev_config_dirty_node
))
5677 vdev_config_clean(vd
);
5682 vdev_compact_children(rvd
);
5684 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5685 vdev_add_child(rvd
, vd
);
5687 vdev_config_dirty(rvd
);
5690 * Reassess the health of our root vdev.
5696 * Remove a device from the pool -
5698 * Removing a device from the vdev namespace requires several steps
5699 * and can take a significant amount of time. As a result we use
5700 * the spa_vdev_config_[enter/exit] functions which allow us to
5701 * grab and release the spa_config_lock while still holding the namespace
5702 * lock. During each step the configuration is synced out.
5704 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5708 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5711 sysevent_t
*ev
= NULL
;
5712 metaslab_group_t
*mg
;
5713 nvlist_t
**spares
, **l2cache
, *nv
;
5715 uint_t nspares
, nl2cache
;
5717 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5719 ASSERT(spa_writeable(spa
));
5722 txg
= spa_vdev_enter(spa
);
5724 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5726 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5727 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5728 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5729 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5731 * Only remove the hot spare if it's not currently in use
5734 if (vd
== NULL
|| unspare
) {
5736 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5737 ev
= spa_event_create(spa
, vd
, NULL
,
5738 ESC_ZFS_VDEV_REMOVE_AUX
);
5739 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5740 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5741 spa_load_spares(spa
);
5742 spa
->spa_spares
.sav_sync
= B_TRUE
;
5744 error
= SET_ERROR(EBUSY
);
5746 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5747 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5748 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5749 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5751 * Cache devices can always be removed.
5753 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5754 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_AUX
);
5755 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5756 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5757 spa_load_l2cache(spa
);
5758 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5759 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5761 ASSERT(vd
== vd
->vdev_top
);
5766 * Stop allocating from this vdev.
5768 metaslab_group_passivate(mg
);
5771 * Wait for the youngest allocations and frees to sync,
5772 * and then wait for the deferral of those frees to finish.
5774 spa_vdev_config_exit(spa
, NULL
,
5775 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5778 * Attempt to evacuate the vdev.
5780 error
= spa_vdev_remove_evacuate(spa
, vd
);
5782 txg
= spa_vdev_config_enter(spa
);
5785 * If we couldn't evacuate the vdev, unwind.
5788 metaslab_group_activate(mg
);
5789 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5793 * Clean up the vdev namespace.
5795 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_DEV
);
5796 spa_vdev_remove_from_namespace(spa
, vd
);
5798 } else if (vd
!= NULL
) {
5800 * Normal vdevs cannot be removed (yet).
5802 error
= SET_ERROR(ENOTSUP
);
5805 * There is no vdev of any kind with the specified guid.
5807 error
= SET_ERROR(ENOENT
);
5811 error
= spa_vdev_exit(spa
, NULL
, txg
, error
);
5820 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5821 * currently spared, so we can detach it.
5824 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5826 vdev_t
*newvd
, *oldvd
;
5828 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5829 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5835 * Check for a completed replacement. We always consider the first
5836 * vdev in the list to be the oldest vdev, and the last one to be
5837 * the newest (see spa_vdev_attach() for how that works). In
5838 * the case where the newest vdev is faulted, we will not automatically
5839 * remove it after a resilver completes. This is OK as it will require
5840 * user intervention to determine which disk the admin wishes to keep.
5842 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5843 ASSERT(vd
->vdev_children
> 1);
5845 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5846 oldvd
= vd
->vdev_child
[0];
5848 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5849 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5850 !vdev_dtl_required(oldvd
))
5855 * Check for a completed resilver with the 'unspare' flag set.
5857 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5858 vdev_t
*first
= vd
->vdev_child
[0];
5859 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5861 if (last
->vdev_unspare
) {
5864 } else if (first
->vdev_unspare
) {
5871 if (oldvd
!= NULL
&&
5872 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5873 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5874 !vdev_dtl_required(oldvd
))
5878 * If there are more than two spares attached to a disk,
5879 * and those spares are not required, then we want to
5880 * attempt to free them up now so that they can be used
5881 * by other pools. Once we're back down to a single
5882 * disk+spare, we stop removing them.
5884 if (vd
->vdev_children
> 2) {
5885 newvd
= vd
->vdev_child
[1];
5887 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5888 vdev_dtl_empty(last
, DTL_MISSING
) &&
5889 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5890 !vdev_dtl_required(newvd
))
5899 spa_vdev_resilver_done(spa_t
*spa
)
5901 vdev_t
*vd
, *pvd
, *ppvd
;
5902 uint64_t guid
, sguid
, pguid
, ppguid
;
5904 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5906 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5907 pvd
= vd
->vdev_parent
;
5908 ppvd
= pvd
->vdev_parent
;
5909 guid
= vd
->vdev_guid
;
5910 pguid
= pvd
->vdev_guid
;
5911 ppguid
= ppvd
->vdev_guid
;
5914 * If we have just finished replacing a hot spared device, then
5915 * we need to detach the parent's first child (the original hot
5918 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5919 ppvd
->vdev_children
== 2) {
5920 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5921 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5923 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5925 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5926 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5928 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5930 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5933 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5937 * Update the stored path or FRU for this vdev.
5940 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5944 boolean_t sync
= B_FALSE
;
5946 ASSERT(spa_writeable(spa
));
5948 spa_vdev_state_enter(spa
, SCL_ALL
);
5950 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5951 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5953 if (!vd
->vdev_ops
->vdev_op_leaf
)
5954 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5957 if (strcmp(value
, vd
->vdev_path
) != 0) {
5958 spa_strfree(vd
->vdev_path
);
5959 vd
->vdev_path
= spa_strdup(value
);
5963 if (vd
->vdev_fru
== NULL
) {
5964 vd
->vdev_fru
= spa_strdup(value
);
5966 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5967 spa_strfree(vd
->vdev_fru
);
5968 vd
->vdev_fru
= spa_strdup(value
);
5973 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5977 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5979 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5983 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5985 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5989 * ==========================================================================
5991 * ==========================================================================
5994 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
5996 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5998 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5999 return (SET_ERROR(EBUSY
));
6001 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
6005 spa_scan_stop(spa_t
*spa
)
6007 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6008 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6009 return (SET_ERROR(EBUSY
));
6010 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
6014 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
6016 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6018 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
6019 return (SET_ERROR(ENOTSUP
));
6022 * If a resilver was requested, but there is no DTL on a
6023 * writeable leaf device, we have nothing to do.
6025 if (func
== POOL_SCAN_RESILVER
&&
6026 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
6027 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6031 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6035 * ==========================================================================
6036 * SPA async task processing
6037 * ==========================================================================
6041 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6043 if (vd
->vdev_remove_wanted
) {
6044 vd
->vdev_remove_wanted
= B_FALSE
;
6045 vd
->vdev_delayed_close
= B_FALSE
;
6046 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6049 * We want to clear the stats, but we don't want to do a full
6050 * vdev_clear() as that will cause us to throw away
6051 * degraded/faulted state as well as attempt to reopen the
6052 * device, all of which is a waste.
6054 vd
->vdev_stat
.vs_read_errors
= 0;
6055 vd
->vdev_stat
.vs_write_errors
= 0;
6056 vd
->vdev_stat
.vs_checksum_errors
= 0;
6058 vdev_state_dirty(vd
->vdev_top
);
6061 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6062 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6066 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6068 if (vd
->vdev_probe_wanted
) {
6069 vd
->vdev_probe_wanted
= B_FALSE
;
6070 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6073 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6074 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6078 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6080 if (!spa
->spa_autoexpand
)
6083 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6084 vdev_t
*cvd
= vd
->vdev_child
[c
];
6085 spa_async_autoexpand(spa
, cvd
);
6088 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6091 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
6095 spa_async_thread(void *arg
)
6097 spa_t
*spa
= (spa_t
*)arg
;
6100 ASSERT(spa
->spa_sync_on
);
6102 mutex_enter(&spa
->spa_async_lock
);
6103 tasks
= spa
->spa_async_tasks
;
6104 spa
->spa_async_tasks
= 0;
6105 mutex_exit(&spa
->spa_async_lock
);
6108 * See if the config needs to be updated.
6110 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6111 uint64_t old_space
, new_space
;
6113 mutex_enter(&spa_namespace_lock
);
6114 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6115 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6116 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6117 mutex_exit(&spa_namespace_lock
);
6120 * If the pool grew as a result of the config update,
6121 * then log an internal history event.
6123 if (new_space
!= old_space
) {
6124 spa_history_log_internal(spa
, "vdev online", NULL
,
6125 "pool '%s' size: %llu(+%llu)",
6126 spa_name(spa
), new_space
, new_space
- old_space
);
6131 * See if any devices need to be marked REMOVED.
6133 if (tasks
& SPA_ASYNC_REMOVE
) {
6134 spa_vdev_state_enter(spa
, SCL_NONE
);
6135 spa_async_remove(spa
, spa
->spa_root_vdev
);
6136 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6137 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6138 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6139 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6140 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6143 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6144 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6145 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6146 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6150 * See if any devices need to be probed.
6152 if (tasks
& SPA_ASYNC_PROBE
) {
6153 spa_vdev_state_enter(spa
, SCL_NONE
);
6154 spa_async_probe(spa
, spa
->spa_root_vdev
);
6155 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6159 * If any devices are done replacing, detach them.
6161 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6162 spa_vdev_resilver_done(spa
);
6165 * Kick off a resilver.
6167 if (tasks
& SPA_ASYNC_RESILVER
)
6168 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6171 * Let the world know that we're done.
6173 mutex_enter(&spa
->spa_async_lock
);
6174 spa
->spa_async_thread
= NULL
;
6175 cv_broadcast(&spa
->spa_async_cv
);
6176 mutex_exit(&spa
->spa_async_lock
);
6181 spa_async_suspend(spa_t
*spa
)
6183 mutex_enter(&spa
->spa_async_lock
);
6184 spa
->spa_async_suspended
++;
6185 while (spa
->spa_async_thread
!= NULL
)
6186 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6187 mutex_exit(&spa
->spa_async_lock
);
6191 spa_async_resume(spa_t
*spa
)
6193 mutex_enter(&spa
->spa_async_lock
);
6194 ASSERT(spa
->spa_async_suspended
!= 0);
6195 spa
->spa_async_suspended
--;
6196 mutex_exit(&spa
->spa_async_lock
);
6200 spa_async_tasks_pending(spa_t
*spa
)
6202 uint_t non_config_tasks
;
6204 boolean_t config_task_suspended
;
6206 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6207 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6208 if (spa
->spa_ccw_fail_time
== 0) {
6209 config_task_suspended
= B_FALSE
;
6211 config_task_suspended
=
6212 (gethrtime() - spa
->spa_ccw_fail_time
) <
6213 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
6216 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6220 spa_async_dispatch(spa_t
*spa
)
6222 mutex_enter(&spa
->spa_async_lock
);
6223 if (spa_async_tasks_pending(spa
) &&
6224 !spa
->spa_async_suspended
&&
6225 spa
->spa_async_thread
== NULL
&&
6227 spa
->spa_async_thread
= thread_create(NULL
, 0,
6228 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6229 mutex_exit(&spa
->spa_async_lock
);
6233 spa_async_request(spa_t
*spa
, int task
)
6235 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6236 mutex_enter(&spa
->spa_async_lock
);
6237 spa
->spa_async_tasks
|= task
;
6238 mutex_exit(&spa
->spa_async_lock
);
6242 * ==========================================================================
6243 * SPA syncing routines
6244 * ==========================================================================
6248 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6251 bpobj_enqueue(bpo
, bp
, tx
);
6256 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6260 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6266 * Note: this simple function is not inlined to make it easier to dtrace the
6267 * amount of time spent syncing frees.
6270 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6272 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6273 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6274 VERIFY(zio_wait(zio
) == 0);
6278 * Note: this simple function is not inlined to make it easier to dtrace the
6279 * amount of time spent syncing deferred frees.
6282 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6284 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6285 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6286 spa_free_sync_cb
, zio
, tx
), ==, 0);
6287 VERIFY0(zio_wait(zio
));
6291 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6293 char *packed
= NULL
;
6298 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6301 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6302 * information. This avoids the dmu_buf_will_dirty() path and
6303 * saves us a pre-read to get data we don't actually care about.
6305 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6306 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6308 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6310 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6312 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6314 vmem_free(packed
, bufsize
);
6316 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6317 dmu_buf_will_dirty(db
, tx
);
6318 *(uint64_t *)db
->db_data
= nvsize
;
6319 dmu_buf_rele(db
, FTAG
);
6323 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6324 const char *config
, const char *entry
)
6334 * Update the MOS nvlist describing the list of available devices.
6335 * spa_validate_aux() will have already made sure this nvlist is
6336 * valid and the vdevs are labeled appropriately.
6338 if (sav
->sav_object
== 0) {
6339 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6340 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6341 sizeof (uint64_t), tx
);
6342 VERIFY(zap_update(spa
->spa_meta_objset
,
6343 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6344 &sav
->sav_object
, tx
) == 0);
6347 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6348 if (sav
->sav_count
== 0) {
6349 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6351 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6352 for (i
= 0; i
< sav
->sav_count
; i
++)
6353 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6354 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6355 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6356 sav
->sav_count
) == 0);
6357 for (i
= 0; i
< sav
->sav_count
; i
++)
6358 nvlist_free(list
[i
]);
6359 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6362 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6363 nvlist_free(nvroot
);
6365 sav
->sav_sync
= B_FALSE
;
6369 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6370 * The all-vdev ZAP must be empty.
6373 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6375 spa_t
*spa
= vd
->vdev_spa
;
6377 if (vd
->vdev_top_zap
!= 0) {
6378 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6379 vd
->vdev_top_zap
, tx
));
6381 if (vd
->vdev_leaf_zap
!= 0) {
6382 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6383 vd
->vdev_leaf_zap
, tx
));
6385 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
6386 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6391 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6396 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6397 * its config may not be dirty but we still need to build per-vdev ZAPs.
6398 * Similarly, if the pool is being assembled (e.g. after a split), we
6399 * need to rebuild the AVZ although the config may not be dirty.
6401 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6402 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6405 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6407 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6408 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6409 spa
->spa_all_vdev_zaps
!= 0);
6411 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6412 /* Make and build the new AVZ */
6413 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6414 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6415 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6417 /* Diff old AVZ with new one */
6421 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6422 spa
->spa_all_vdev_zaps
);
6423 zap_cursor_retrieve(&zc
, &za
) == 0;
6424 zap_cursor_advance(&zc
)) {
6425 uint64_t vdzap
= za
.za_first_integer
;
6426 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6429 * ZAP is listed in old AVZ but not in new one;
6432 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6437 zap_cursor_fini(&zc
);
6439 /* Destroy the old AVZ */
6440 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6441 spa
->spa_all_vdev_zaps
, tx
));
6443 /* Replace the old AVZ in the dir obj with the new one */
6444 VERIFY0(zap_update(spa
->spa_meta_objset
,
6445 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6446 sizeof (new_avz
), 1, &new_avz
, tx
));
6448 spa
->spa_all_vdev_zaps
= new_avz
;
6449 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6453 /* Walk through the AVZ and destroy all listed ZAPs */
6454 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6455 spa
->spa_all_vdev_zaps
);
6456 zap_cursor_retrieve(&zc
, &za
) == 0;
6457 zap_cursor_advance(&zc
)) {
6458 uint64_t zap
= za
.za_first_integer
;
6459 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6462 zap_cursor_fini(&zc
);
6464 /* Destroy and unlink the AVZ itself */
6465 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6466 spa
->spa_all_vdev_zaps
, tx
));
6467 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6468 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6469 spa
->spa_all_vdev_zaps
= 0;
6472 if (spa
->spa_all_vdev_zaps
== 0) {
6473 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6474 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6475 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6477 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6479 /* Create ZAPs for vdevs that don't have them. */
6480 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6482 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6483 dmu_tx_get_txg(tx
), B_FALSE
);
6486 * If we're upgrading the spa version then make sure that
6487 * the config object gets updated with the correct version.
6489 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6490 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6491 spa
->spa_uberblock
.ub_version
);
6493 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6495 nvlist_free(spa
->spa_config_syncing
);
6496 spa
->spa_config_syncing
= config
;
6498 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6502 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6504 uint64_t *versionp
= arg
;
6505 uint64_t version
= *versionp
;
6506 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6509 * Setting the version is special cased when first creating the pool.
6511 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6513 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6514 ASSERT(version
>= spa_version(spa
));
6516 spa
->spa_uberblock
.ub_version
= version
;
6517 vdev_config_dirty(spa
->spa_root_vdev
);
6518 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6522 * Set zpool properties.
6525 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6527 nvlist_t
*nvp
= arg
;
6528 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6529 objset_t
*mos
= spa
->spa_meta_objset
;
6530 nvpair_t
*elem
= NULL
;
6532 mutex_enter(&spa
->spa_props_lock
);
6534 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6536 char *strval
, *fname
;
6538 const char *propname
;
6539 zprop_type_t proptype
;
6542 prop
= zpool_name_to_prop(nvpair_name(elem
));
6543 switch ((int)prop
) {
6546 * We checked this earlier in spa_prop_validate().
6548 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6550 fname
= strchr(nvpair_name(elem
), '@') + 1;
6551 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6553 spa_feature_enable(spa
, fid
, tx
);
6554 spa_history_log_internal(spa
, "set", tx
,
6555 "%s=enabled", nvpair_name(elem
));
6558 case ZPOOL_PROP_VERSION
:
6559 intval
= fnvpair_value_uint64(elem
);
6561 * The version is synced separately before other
6562 * properties and should be correct by now.
6564 ASSERT3U(spa_version(spa
), >=, intval
);
6567 case ZPOOL_PROP_ALTROOT
:
6569 * 'altroot' is a non-persistent property. It should
6570 * have been set temporarily at creation or import time.
6572 ASSERT(spa
->spa_root
!= NULL
);
6575 case ZPOOL_PROP_READONLY
:
6576 case ZPOOL_PROP_CACHEFILE
:
6578 * 'readonly' and 'cachefile' are also non-persisitent
6582 case ZPOOL_PROP_COMMENT
:
6583 strval
= fnvpair_value_string(elem
);
6584 if (spa
->spa_comment
!= NULL
)
6585 spa_strfree(spa
->spa_comment
);
6586 spa
->spa_comment
= spa_strdup(strval
);
6588 * We need to dirty the configuration on all the vdevs
6589 * so that their labels get updated. It's unnecessary
6590 * to do this for pool creation since the vdev's
6591 * configuration has already been dirtied.
6593 if (tx
->tx_txg
!= TXG_INITIAL
)
6594 vdev_config_dirty(spa
->spa_root_vdev
);
6595 spa_history_log_internal(spa
, "set", tx
,
6596 "%s=%s", nvpair_name(elem
), strval
);
6600 * Set pool property values in the poolprops mos object.
6602 if (spa
->spa_pool_props_object
== 0) {
6603 spa
->spa_pool_props_object
=
6604 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6605 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6609 /* normalize the property name */
6610 propname
= zpool_prop_to_name(prop
);
6611 proptype
= zpool_prop_get_type(prop
);
6613 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6614 ASSERT(proptype
== PROP_TYPE_STRING
);
6615 strval
= fnvpair_value_string(elem
);
6616 VERIFY0(zap_update(mos
,
6617 spa
->spa_pool_props_object
, propname
,
6618 1, strlen(strval
) + 1, strval
, tx
));
6619 spa_history_log_internal(spa
, "set", tx
,
6620 "%s=%s", nvpair_name(elem
), strval
);
6621 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6622 intval
= fnvpair_value_uint64(elem
);
6624 if (proptype
== PROP_TYPE_INDEX
) {
6626 VERIFY0(zpool_prop_index_to_string(
6627 prop
, intval
, &unused
));
6629 VERIFY0(zap_update(mos
,
6630 spa
->spa_pool_props_object
, propname
,
6631 8, 1, &intval
, tx
));
6632 spa_history_log_internal(spa
, "set", tx
,
6633 "%s=%lld", nvpair_name(elem
), intval
);
6635 ASSERT(0); /* not allowed */
6639 case ZPOOL_PROP_DELEGATION
:
6640 spa
->spa_delegation
= intval
;
6642 case ZPOOL_PROP_BOOTFS
:
6643 spa
->spa_bootfs
= intval
;
6645 case ZPOOL_PROP_FAILUREMODE
:
6646 spa
->spa_failmode
= intval
;
6648 case ZPOOL_PROP_AUTOEXPAND
:
6649 spa
->spa_autoexpand
= intval
;
6650 if (tx
->tx_txg
!= TXG_INITIAL
)
6651 spa_async_request(spa
,
6652 SPA_ASYNC_AUTOEXPAND
);
6654 case ZPOOL_PROP_MULTIHOST
:
6655 spa
->spa_multihost
= intval
;
6657 case ZPOOL_PROP_DEDUPDITTO
:
6658 spa
->spa_dedup_ditto
= intval
;
6667 mutex_exit(&spa
->spa_props_lock
);
6671 * Perform one-time upgrade on-disk changes. spa_version() does not
6672 * reflect the new version this txg, so there must be no changes this
6673 * txg to anything that the upgrade code depends on after it executes.
6674 * Therefore this must be called after dsl_pool_sync() does the sync
6678 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6680 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6682 ASSERT(spa
->spa_sync_pass
== 1);
6684 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6686 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6687 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6688 dsl_pool_create_origin(dp
, tx
);
6690 /* Keeping the origin open increases spa_minref */
6691 spa
->spa_minref
+= 3;
6694 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6695 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6696 dsl_pool_upgrade_clones(dp
, tx
);
6699 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6700 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6701 dsl_pool_upgrade_dir_clones(dp
, tx
);
6703 /* Keeping the freedir open increases spa_minref */
6704 spa
->spa_minref
+= 3;
6707 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6708 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6709 spa_feature_create_zap_objects(spa
, tx
);
6713 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6714 * when possibility to use lz4 compression for metadata was added
6715 * Old pools that have this feature enabled must be upgraded to have
6716 * this feature active
6718 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6719 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6720 SPA_FEATURE_LZ4_COMPRESS
);
6721 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6722 SPA_FEATURE_LZ4_COMPRESS
);
6724 if (lz4_en
&& !lz4_ac
)
6725 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6729 * If we haven't written the salt, do so now. Note that the
6730 * feature may not be activated yet, but that's fine since
6731 * the presence of this ZAP entry is backwards compatible.
6733 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
6734 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
6735 VERIFY0(zap_add(spa
->spa_meta_objset
,
6736 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
6737 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
6738 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
6741 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6745 * Sync the specified transaction group. New blocks may be dirtied as
6746 * part of the process, so we iterate until it converges.
6749 spa_sync(spa_t
*spa
, uint64_t txg
)
6751 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6752 objset_t
*mos
= spa
->spa_meta_objset
;
6753 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6754 vdev_t
*rvd
= spa
->spa_root_vdev
;
6758 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
6759 zfs_vdev_queue_depth_pct
/ 100;
6761 VERIFY(spa_writeable(spa
));
6764 * Lock out configuration changes.
6766 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6768 spa
->spa_syncing_txg
= txg
;
6769 spa
->spa_sync_pass
= 0;
6771 mutex_enter(&spa
->spa_alloc_lock
);
6772 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
6773 mutex_exit(&spa
->spa_alloc_lock
);
6776 * If there are any pending vdev state changes, convert them
6777 * into config changes that go out with this transaction group.
6779 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6780 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6782 * We need the write lock here because, for aux vdevs,
6783 * calling vdev_config_dirty() modifies sav_config.
6784 * This is ugly and will become unnecessary when we
6785 * eliminate the aux vdev wart by integrating all vdevs
6786 * into the root vdev tree.
6788 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6789 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6790 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6791 vdev_state_clean(vd
);
6792 vdev_config_dirty(vd
);
6794 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6795 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6797 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6799 tx
= dmu_tx_create_assigned(dp
, txg
);
6801 spa
->spa_sync_starttime
= gethrtime();
6802 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6803 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
6804 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6805 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6808 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6809 * set spa_deflate if we have no raid-z vdevs.
6811 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6812 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6815 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6816 vd
= rvd
->vdev_child
[i
];
6817 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6820 if (i
== rvd
->vdev_children
) {
6821 spa
->spa_deflate
= TRUE
;
6822 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6823 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6824 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6829 * Set the top-level vdev's max queue depth. Evaluate each
6830 * top-level's async write queue depth in case it changed.
6831 * The max queue depth will not change in the middle of syncing
6834 uint64_t queue_depth_total
= 0;
6835 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
6836 vdev_t
*tvd
= rvd
->vdev_child
[c
];
6837 metaslab_group_t
*mg
= tvd
->vdev_mg
;
6839 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
6840 !metaslab_group_initialized(mg
))
6844 * It is safe to do a lock-free check here because only async
6845 * allocations look at mg_max_alloc_queue_depth, and async
6846 * allocations all happen from spa_sync().
6848 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
6849 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
6850 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
6852 metaslab_class_t
*mc
= spa_normal_class(spa
);
6853 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
6854 mc
->mc_alloc_max_slots
= queue_depth_total
;
6855 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
6857 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
6858 max_queue_depth
* rvd
->vdev_children
);
6861 * Iterate to convergence.
6864 int pass
= ++spa
->spa_sync_pass
;
6866 spa_sync_config_object(spa
, tx
);
6867 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6868 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6869 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6870 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6871 spa_errlog_sync(spa
, txg
);
6872 dsl_pool_sync(dp
, txg
);
6874 if (pass
< zfs_sync_pass_deferred_free
) {
6875 spa_sync_frees(spa
, free_bpl
, tx
);
6878 * We can not defer frees in pass 1, because
6879 * we sync the deferred frees later in pass 1.
6881 ASSERT3U(pass
, >, 1);
6882 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6883 &spa
->spa_deferred_bpobj
, tx
);
6887 dsl_scan_sync(dp
, tx
);
6889 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6893 spa_sync_upgrades(spa
, tx
);
6895 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
6897 * Note: We need to check if the MOS is dirty
6898 * because we could have marked the MOS dirty
6899 * without updating the uberblock (e.g. if we
6900 * have sync tasks but no dirty user data). We
6901 * need to check the uberblock's rootbp because
6902 * it is updated if we have synced out dirty
6903 * data (though in this case the MOS will most
6904 * likely also be dirty due to second order
6905 * effects, we don't want to rely on that here).
6907 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
6908 !dmu_objset_is_dirty(mos
, txg
)) {
6910 * Nothing changed on the first pass,
6911 * therefore this TXG is a no-op. Avoid
6912 * syncing deferred frees, so that we
6913 * can keep this TXG as a no-op.
6915 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
6917 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6918 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
6921 spa_sync_deferred_frees(spa
, tx
);
6924 } while (dmu_objset_is_dirty(mos
, txg
));
6927 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
6929 * Make sure that the number of ZAPs for all the vdevs matches
6930 * the number of ZAPs in the per-vdev ZAP list. This only gets
6931 * called if the config is dirty; otherwise there may be
6932 * outstanding AVZ operations that weren't completed in
6933 * spa_sync_config_object.
6935 uint64_t all_vdev_zap_entry_count
;
6936 ASSERT0(zap_count(spa
->spa_meta_objset
,
6937 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
6938 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
6939 all_vdev_zap_entry_count
);
6944 * Rewrite the vdev configuration (which includes the uberblock)
6945 * to commit the transaction group.
6947 * If there are no dirty vdevs, we sync the uberblock to a few
6948 * random top-level vdevs that are known to be visible in the
6949 * config cache (see spa_vdev_add() for a complete description).
6950 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6954 * We hold SCL_STATE to prevent vdev open/close/etc.
6955 * while we're attempting to write the vdev labels.
6957 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6959 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6960 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6962 int children
= rvd
->vdev_children
;
6963 int c0
= spa_get_random(children
);
6965 for (int c
= 0; c
< children
; c
++) {
6966 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6967 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6969 svd
[svdcount
++] = vd
;
6970 if (svdcount
== SPA_DVAS_PER_BP
)
6973 error
= vdev_config_sync(svd
, svdcount
, txg
);
6975 error
= vdev_config_sync(rvd
->vdev_child
,
6976 rvd
->vdev_children
, txg
);
6980 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6982 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6986 zio_suspend(spa
, NULL
);
6987 zio_resume_wait(spa
);
6991 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6992 spa
->spa_deadman_tqid
= 0;
6995 * Clear the dirty config list.
6997 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6998 vdev_config_clean(vd
);
7001 * Now that the new config has synced transactionally,
7002 * let it become visible to the config cache.
7004 if (spa
->spa_config_syncing
!= NULL
) {
7005 spa_config_set(spa
, spa
->spa_config_syncing
);
7006 spa
->spa_config_txg
= txg
;
7007 spa
->spa_config_syncing
= NULL
;
7010 dsl_pool_sync_done(dp
, txg
);
7012 mutex_enter(&spa
->spa_alloc_lock
);
7013 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7014 mutex_exit(&spa
->spa_alloc_lock
);
7017 * Update usable space statistics.
7019 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
7020 vdev_sync_done(vd
, txg
);
7022 spa_update_dspace(spa
);
7025 * It had better be the case that we didn't dirty anything
7026 * since vdev_config_sync().
7028 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7029 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7030 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7032 spa
->spa_sync_pass
= 0;
7035 * Update the last synced uberblock here. We want to do this at
7036 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7037 * will be guaranteed that all the processing associated with
7038 * that txg has been completed.
7040 spa
->spa_ubsync
= spa
->spa_uberblock
;
7041 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7043 spa_handle_ignored_writes(spa
);
7046 * If any async tasks have been requested, kick them off.
7048 spa_async_dispatch(spa
);
7052 * Sync all pools. We don't want to hold the namespace lock across these
7053 * operations, so we take a reference on the spa_t and drop the lock during the
7057 spa_sync_allpools(void)
7060 mutex_enter(&spa_namespace_lock
);
7061 while ((spa
= spa_next(spa
)) != NULL
) {
7062 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7063 !spa_writeable(spa
) || spa_suspended(spa
))
7065 spa_open_ref(spa
, FTAG
);
7066 mutex_exit(&spa_namespace_lock
);
7067 txg_wait_synced(spa_get_dsl(spa
), 0);
7068 mutex_enter(&spa_namespace_lock
);
7069 spa_close(spa
, FTAG
);
7071 mutex_exit(&spa_namespace_lock
);
7075 * ==========================================================================
7076 * Miscellaneous routines
7077 * ==========================================================================
7081 * Remove all pools in the system.
7089 * Remove all cached state. All pools should be closed now,
7090 * so every spa in the AVL tree should be unreferenced.
7092 mutex_enter(&spa_namespace_lock
);
7093 while ((spa
= spa_next(NULL
)) != NULL
) {
7095 * Stop async tasks. The async thread may need to detach
7096 * a device that's been replaced, which requires grabbing
7097 * spa_namespace_lock, so we must drop it here.
7099 spa_open_ref(spa
, FTAG
);
7100 mutex_exit(&spa_namespace_lock
);
7101 spa_async_suspend(spa
);
7102 mutex_enter(&spa_namespace_lock
);
7103 spa_close(spa
, FTAG
);
7105 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7107 spa_deactivate(spa
);
7111 mutex_exit(&spa_namespace_lock
);
7115 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7120 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7124 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7125 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7126 if (vd
->vdev_guid
== guid
)
7130 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7131 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7132 if (vd
->vdev_guid
== guid
)
7141 spa_upgrade(spa_t
*spa
, uint64_t version
)
7143 ASSERT(spa_writeable(spa
));
7145 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7148 * This should only be called for a non-faulted pool, and since a
7149 * future version would result in an unopenable pool, this shouldn't be
7152 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7153 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7155 spa
->spa_uberblock
.ub_version
= version
;
7156 vdev_config_dirty(spa
->spa_root_vdev
);
7158 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7160 txg_wait_synced(spa_get_dsl(spa
), 0);
7164 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7168 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7170 for (i
= 0; i
< sav
->sav_count
; i
++)
7171 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7174 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7175 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7176 &spareguid
) == 0 && spareguid
== guid
)
7184 * Check if a pool has an active shared spare device.
7185 * Note: reference count of an active spare is 2, as a spare and as a replace
7188 spa_has_active_shared_spare(spa_t
*spa
)
7192 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7194 for (i
= 0; i
< sav
->sav_count
; i
++) {
7195 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7196 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7205 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7207 sysevent_t
*ev
= NULL
;
7211 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
7213 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
7214 ev
->resource
= resource
;
7221 spa_event_post(sysevent_t
*ev
)
7225 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
7226 kmem_free(ev
, sizeof (*ev
));
7232 * Post a zevent corresponding to the given sysevent. The 'name' must be one
7233 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
7234 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7235 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7236 * or zdb as real changes.
7239 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7241 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
7244 #if defined(_KERNEL) && defined(HAVE_SPL)
7245 /* state manipulation functions */
7246 EXPORT_SYMBOL(spa_open
);
7247 EXPORT_SYMBOL(spa_open_rewind
);
7248 EXPORT_SYMBOL(spa_get_stats
);
7249 EXPORT_SYMBOL(spa_create
);
7250 EXPORT_SYMBOL(spa_import
);
7251 EXPORT_SYMBOL(spa_tryimport
);
7252 EXPORT_SYMBOL(spa_destroy
);
7253 EXPORT_SYMBOL(spa_export
);
7254 EXPORT_SYMBOL(spa_reset
);
7255 EXPORT_SYMBOL(spa_async_request
);
7256 EXPORT_SYMBOL(spa_async_suspend
);
7257 EXPORT_SYMBOL(spa_async_resume
);
7258 EXPORT_SYMBOL(spa_inject_addref
);
7259 EXPORT_SYMBOL(spa_inject_delref
);
7260 EXPORT_SYMBOL(spa_scan_stat_init
);
7261 EXPORT_SYMBOL(spa_scan_get_stats
);
7263 /* device maniion */
7264 EXPORT_SYMBOL(spa_vdev_add
);
7265 EXPORT_SYMBOL(spa_vdev_attach
);
7266 EXPORT_SYMBOL(spa_vdev_detach
);
7267 EXPORT_SYMBOL(spa_vdev_remove
);
7268 EXPORT_SYMBOL(spa_vdev_setpath
);
7269 EXPORT_SYMBOL(spa_vdev_setfru
);
7270 EXPORT_SYMBOL(spa_vdev_split_mirror
);
7272 /* spare statech is global across all pools) */
7273 EXPORT_SYMBOL(spa_spare_add
);
7274 EXPORT_SYMBOL(spa_spare_remove
);
7275 EXPORT_SYMBOL(spa_spare_exists
);
7276 EXPORT_SYMBOL(spa_spare_activate
);
7278 /* L2ARC statech is global across all pools) */
7279 EXPORT_SYMBOL(spa_l2cache_add
);
7280 EXPORT_SYMBOL(spa_l2cache_remove
);
7281 EXPORT_SYMBOL(spa_l2cache_exists
);
7282 EXPORT_SYMBOL(spa_l2cache_activate
);
7283 EXPORT_SYMBOL(spa_l2cache_drop
);
7286 EXPORT_SYMBOL(spa_scan
);
7287 EXPORT_SYMBOL(spa_scan_stop
);
7290 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
7291 EXPORT_SYMBOL(spa_sync_allpools
);
7294 EXPORT_SYMBOL(spa_prop_set
);
7295 EXPORT_SYMBOL(spa_prop_get
);
7296 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
7298 /* asynchronous event notification */
7299 EXPORT_SYMBOL(spa_event_notify
);
7302 #if defined(_KERNEL) && defined(HAVE_SPL)
7303 module_param(spa_load_verify_maxinflight
, int, 0644);
7304 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
7305 "Max concurrent traversal I/Os while verifying pool during import -X");
7307 module_param(spa_load_verify_metadata
, int, 0644);
7308 MODULE_PARM_DESC(spa_load_verify_metadata
,
7309 "Set to traverse metadata on pool import");
7311 module_param(spa_load_verify_data
, int, 0644);
7312 MODULE_PARM_DESC(spa_load_verify_data
,
7313 "Set to traverse data on pool import");
7316 module_param(zio_taskq_batch_pct
, uint
, 0444);
7317 MODULE_PARM_DESC(zio_taskq_batch_pct
,
7318 "Percentage of CPUs to run an IO worker thread");