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
*config
)
2336 uint64_t hostid
= 0;
2337 uint64_t tryconfig_txg
= 0;
2338 uint64_t tryconfig_timestamp
= 0;
2341 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2342 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2343 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2345 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2346 &tryconfig_timestamp
);
2349 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2350 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_HOSTID
, &hostid
);
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.
2378 if (hostid
== spa_get_hostid())
2382 * Skip the activity test when the pool was cleanly exported.
2384 if (state
!= POOL_STATE_ACTIVE
)
2391 * Perform the import activity check. If the user canceled the import or
2392 * we detected activity then fail.
2395 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2397 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2398 uint64_t txg
= ub
->ub_txg
;
2399 uint64_t timestamp
= ub
->ub_timestamp
;
2400 uint64_t import_delay
= NANOSEC
;
2401 hrtime_t import_expire
;
2402 nvlist_t
*mmp_label
= NULL
;
2403 vdev_t
*rvd
= spa
->spa_root_vdev
;
2408 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
2409 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2413 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2414 * during the earlier tryimport. If the txg recorded there is 0 then
2415 * the pool is known to be active on another host.
2417 * Otherwise, the pool might be in use on another node. Check for
2418 * changes in the uberblocks on disk if necessary.
2420 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2421 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
2422 ZPOOL_CONFIG_LOAD_INFO
);
2424 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
2425 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
2426 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2427 error
= SET_ERROR(EREMOTEIO
);
2433 * Preferentially use the zfs_multihost_interval from the node which
2434 * last imported the pool. This value is stored in an MMP uberblock as.
2436 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2438 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
)
2439 import_delay
= MAX(import_delay
, import_intervals
*
2440 ub
->ub_mmp_delay
* MAX(vdev_count_leaves(spa
), 1));
2442 /* Apply a floor using the local default values. */
2443 import_delay
= MAX(import_delay
, import_intervals
*
2444 MSEC2NSEC(MAX(zfs_multihost_interval
, MMP_MIN_INTERVAL
)));
2446 /* Add a small random factor in case of simultaneous imports (0-25%) */
2447 import_expire
= gethrtime() + import_delay
+
2448 (import_delay
* spa_get_random(250) / 1000);
2450 while (gethrtime() < import_expire
) {
2451 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2453 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
) {
2454 error
= SET_ERROR(EREMOTEIO
);
2459 nvlist_free(mmp_label
);
2463 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
2465 error
= SET_ERROR(EINTR
);
2473 mutex_destroy(&mtx
);
2477 * If the pool is determined to be active store the status in the
2478 * spa->spa_load_info nvlist. If the remote hostname or hostid are
2479 * available from configuration read from disk store them as well.
2480 * This allows 'zpool import' to generate a more useful message.
2482 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
2483 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2484 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
2486 if (error
== EREMOTEIO
) {
2487 char *hostname
= "<unknown>";
2488 uint64_t hostid
= 0;
2491 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
2492 hostname
= fnvlist_lookup_string(mmp_label
,
2493 ZPOOL_CONFIG_HOSTNAME
);
2494 fnvlist_add_string(spa
->spa_load_info
,
2495 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
2498 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
2499 hostid
= fnvlist_lookup_uint64(mmp_label
,
2500 ZPOOL_CONFIG_HOSTID
);
2501 fnvlist_add_uint64(spa
->spa_load_info
,
2502 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
2506 fnvlist_add_uint64(spa
->spa_load_info
,
2507 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
2508 fnvlist_add_uint64(spa
->spa_load_info
,
2509 ZPOOL_CONFIG_MMP_TXG
, 0);
2511 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
2515 nvlist_free(mmp_label
);
2521 * Load an existing storage pool, using the pool's builtin spa_config as a
2522 * source of configuration information.
2524 __attribute__((always_inline
))
2526 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2527 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2531 nvlist_t
*nvroot
= NULL
;
2534 uberblock_t
*ub
= &spa
->spa_uberblock
;
2535 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2536 int orig_mode
= spa
->spa_mode
;
2539 boolean_t missing_feat_write
= B_FALSE
;
2540 boolean_t activity_check
= B_FALSE
;
2543 * If this is an untrusted config, access the pool in read-only mode.
2544 * This prevents things like resilvering recently removed devices.
2547 spa
->spa_mode
= FREAD
;
2549 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2551 spa
->spa_load_state
= state
;
2553 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2554 return (SET_ERROR(EINVAL
));
2556 parse
= (type
== SPA_IMPORT_EXISTING
?
2557 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2560 * Create "The Godfather" zio to hold all async IOs
2562 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2564 for (int i
= 0; i
< max_ncpus
; i
++) {
2565 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2566 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2567 ZIO_FLAG_GODFATHER
);
2571 * Parse the configuration into a vdev tree. We explicitly set the
2572 * value that will be returned by spa_version() since parsing the
2573 * configuration requires knowing the version number.
2575 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2576 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2577 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2582 ASSERT(spa
->spa_root_vdev
== rvd
);
2583 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2584 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2586 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2587 ASSERT(spa_guid(spa
) == pool_guid
);
2591 * Try to open all vdevs, loading each label in the process.
2593 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2594 error
= vdev_open(rvd
);
2595 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2600 * We need to validate the vdev labels against the configuration that
2601 * we have in hand, which is dependent on the setting of mosconfig. If
2602 * mosconfig is true then we're validating the vdev labels based on
2603 * that config. Otherwise, we're validating against the cached config
2604 * (zpool.cache) that was read when we loaded the zfs module, and then
2605 * later we will recursively call spa_load() and validate against
2608 * If we're assembling a new pool that's been split off from an
2609 * existing pool, the labels haven't yet been updated so we skip
2610 * validation for now.
2612 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2613 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2614 error
= vdev_validate(rvd
, mosconfig
);
2615 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2620 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2621 return (SET_ERROR(ENXIO
));
2625 * Find the best uberblock.
2627 vdev_uberblock_load(rvd
, ub
, &label
);
2630 * If we weren't able to find a single valid uberblock, return failure.
2632 if (ub
->ub_txg
== 0) {
2634 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2638 * For pools which have the multihost property on determine if the
2639 * pool is truly inactive and can be safely imported. Prevent
2640 * hosts which don't have a hostid set from importing the pool.
2642 activity_check
= spa_activity_check_required(spa
, ub
, config
);
2643 if (activity_check
) {
2644 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
2645 spa_get_hostid() == 0) {
2647 fnvlist_add_uint64(spa
->spa_load_info
,
2648 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
2649 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
2652 error
= spa_activity_check(spa
, ub
, config
);
2658 fnvlist_add_uint64(spa
->spa_load_info
,
2659 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
2660 fnvlist_add_uint64(spa
->spa_load_info
,
2661 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
2665 * If the pool has an unsupported version we can't open it.
2667 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2669 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2672 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2676 * If we weren't able to find what's necessary for reading the
2677 * MOS in the label, return failure.
2679 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2680 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2682 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2687 * Update our in-core representation with the definitive values
2690 nvlist_free(spa
->spa_label_features
);
2691 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2697 * Look through entries in the label nvlist's features_for_read. If
2698 * there is a feature listed there which we don't understand then we
2699 * cannot open a pool.
2701 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2702 nvlist_t
*unsup_feat
;
2704 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2707 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
2709 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2710 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2711 VERIFY(nvlist_add_string(unsup_feat
,
2712 nvpair_name(nvp
), "") == 0);
2716 if (!nvlist_empty(unsup_feat
)) {
2717 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2718 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2719 nvlist_free(unsup_feat
);
2720 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2724 nvlist_free(unsup_feat
);
2728 * If the vdev guid sum doesn't match the uberblock, we have an
2729 * incomplete configuration. We first check to see if the pool
2730 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2731 * If it is, defer the vdev_guid_sum check till later so we
2732 * can handle missing vdevs.
2734 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2735 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2736 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2737 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2739 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2740 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2741 spa_try_repair(spa
, config
);
2742 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2743 nvlist_free(spa
->spa_config_splitting
);
2744 spa
->spa_config_splitting
= NULL
;
2748 * Initialize internal SPA structures.
2750 spa
->spa_state
= POOL_STATE_ACTIVE
;
2751 spa
->spa_ubsync
= spa
->spa_uberblock
;
2752 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2753 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2754 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2755 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2756 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2757 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2759 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2761 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2762 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2764 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2765 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2767 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2768 boolean_t missing_feat_read
= B_FALSE
;
2769 nvlist_t
*unsup_feat
, *enabled_feat
;
2771 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2772 &spa
->spa_feat_for_read_obj
) != 0) {
2773 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2776 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2777 &spa
->spa_feat_for_write_obj
) != 0) {
2778 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2781 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2782 &spa
->spa_feat_desc_obj
) != 0) {
2783 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2786 enabled_feat
= fnvlist_alloc();
2787 unsup_feat
= fnvlist_alloc();
2789 if (!spa_features_check(spa
, B_FALSE
,
2790 unsup_feat
, enabled_feat
))
2791 missing_feat_read
= B_TRUE
;
2793 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2794 if (!spa_features_check(spa
, B_TRUE
,
2795 unsup_feat
, enabled_feat
)) {
2796 missing_feat_write
= B_TRUE
;
2800 fnvlist_add_nvlist(spa
->spa_load_info
,
2801 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2803 if (!nvlist_empty(unsup_feat
)) {
2804 fnvlist_add_nvlist(spa
->spa_load_info
,
2805 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2808 fnvlist_free(enabled_feat
);
2809 fnvlist_free(unsup_feat
);
2811 if (!missing_feat_read
) {
2812 fnvlist_add_boolean(spa
->spa_load_info
,
2813 ZPOOL_CONFIG_CAN_RDONLY
);
2817 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2818 * twofold: to determine whether the pool is available for
2819 * import in read-write mode and (if it is not) whether the
2820 * pool is available for import in read-only mode. If the pool
2821 * is available for import in read-write mode, it is displayed
2822 * as available in userland; if it is not available for import
2823 * in read-only mode, it is displayed as unavailable in
2824 * userland. If the pool is available for import in read-only
2825 * mode but not read-write mode, it is displayed as unavailable
2826 * in userland with a special note that the pool is actually
2827 * available for open in read-only mode.
2829 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2830 * missing a feature for write, we must first determine whether
2831 * the pool can be opened read-only before returning to
2832 * userland in order to know whether to display the
2833 * abovementioned note.
2835 if (missing_feat_read
|| (missing_feat_write
&&
2836 spa_writeable(spa
))) {
2837 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2842 * Load refcounts for ZFS features from disk into an in-memory
2843 * cache during SPA initialization.
2845 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
2848 error
= feature_get_refcount_from_disk(spa
,
2849 &spa_feature_table
[i
], &refcount
);
2851 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2852 } else if (error
== ENOTSUP
) {
2853 spa
->spa_feat_refcount_cache
[i
] =
2854 SPA_FEATURE_DISABLED
;
2856 return (spa_vdev_err(rvd
,
2857 VDEV_AUX_CORRUPT_DATA
, EIO
));
2862 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2863 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2864 &spa
->spa_feat_enabled_txg_obj
) != 0)
2865 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2868 spa
->spa_is_initializing
= B_TRUE
;
2869 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2870 spa
->spa_is_initializing
= B_FALSE
;
2872 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2876 nvlist_t
*policy
= NULL
, *nvconfig
;
2878 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2879 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2881 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2882 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2884 unsigned long myhostid
= 0;
2886 VERIFY(nvlist_lookup_string(nvconfig
,
2887 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2889 myhostid
= spa_get_hostid();
2890 if (hostid
&& myhostid
&& hostid
!= myhostid
) {
2891 nvlist_free(nvconfig
);
2892 return (SET_ERROR(EBADF
));
2895 if (nvlist_lookup_nvlist(spa
->spa_config
,
2896 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2897 VERIFY(nvlist_add_nvlist(nvconfig
,
2898 ZPOOL_REWIND_POLICY
, policy
) == 0);
2900 spa_config_set(spa
, nvconfig
);
2902 spa_deactivate(spa
);
2903 spa_activate(spa
, orig_mode
);
2905 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2908 /* Grab the checksum salt from the MOS. */
2909 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2910 DMU_POOL_CHECKSUM_SALT
, 1,
2911 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
2912 spa
->spa_cksum_salt
.zcs_bytes
);
2913 if (error
== ENOENT
) {
2914 /* Generate a new salt for subsequent use */
2915 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
2916 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
2917 } else if (error
!= 0) {
2918 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2921 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2922 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2923 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2925 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2928 * Load the bit that tells us to use the new accounting function
2929 * (raid-z deflation). If we have an older pool, this will not
2932 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2933 if (error
!= 0 && error
!= ENOENT
)
2934 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2936 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2937 &spa
->spa_creation_version
);
2938 if (error
!= 0 && error
!= ENOENT
)
2939 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2942 * Load the persistent error log. If we have an older pool, this will
2945 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2946 if (error
!= 0 && error
!= ENOENT
)
2947 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2949 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2950 &spa
->spa_errlog_scrub
);
2951 if (error
!= 0 && error
!= ENOENT
)
2952 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2955 * Load the history object. If we have an older pool, this
2956 * will not be present.
2958 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2959 if (error
!= 0 && error
!= ENOENT
)
2960 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2963 * Load the per-vdev ZAP map. If we have an older pool, this will not
2964 * be present; in this case, defer its creation to a later time to
2965 * avoid dirtying the MOS this early / out of sync context. See
2966 * spa_sync_config_object.
2969 /* The sentinel is only available in the MOS config. */
2970 nvlist_t
*mos_config
;
2971 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
2972 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2974 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
2975 &spa
->spa_all_vdev_zaps
);
2977 if (error
== ENOENT
) {
2978 VERIFY(!nvlist_exists(mos_config
,
2979 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
2980 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
2981 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
2982 } else if (error
!= 0) {
2983 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2984 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
2986 * An older version of ZFS overwrote the sentinel value, so
2987 * we have orphaned per-vdev ZAPs in the MOS. Defer their
2988 * destruction to later; see spa_sync_config_object.
2990 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
2992 * We're assuming that no vdevs have had their ZAPs created
2993 * before this. Better be sure of it.
2995 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
2997 nvlist_free(mos_config
);
3000 * If we're assembling the pool from the split-off vdevs of
3001 * an existing pool, we don't want to attach the spares & cache
3006 * Load any hot spares for this pool.
3008 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
3009 if (error
!= 0 && error
!= ENOENT
)
3010 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3011 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3012 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3013 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3014 &spa
->spa_spares
.sav_config
) != 0)
3015 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3017 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3018 spa_load_spares(spa
);
3019 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3020 } else if (error
== 0) {
3021 spa
->spa_spares
.sav_sync
= B_TRUE
;
3025 * Load any level 2 ARC devices for this pool.
3027 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3028 &spa
->spa_l2cache
.sav_object
);
3029 if (error
!= 0 && error
!= ENOENT
)
3030 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3031 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3032 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3033 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3034 &spa
->spa_l2cache
.sav_config
) != 0)
3035 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3037 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3038 spa_load_l2cache(spa
);
3039 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3040 } else if (error
== 0) {
3041 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3044 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3046 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
3047 if (error
&& error
!= ENOENT
)
3048 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3051 uint64_t autoreplace
= 0;
3053 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3054 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3055 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3056 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3057 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3058 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
3059 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3060 &spa
->spa_dedup_ditto
);
3062 spa
->spa_autoreplace
= (autoreplace
!= 0);
3066 * If the 'multihost' property is set, then never allow a pool to
3067 * be imported when the system hostid is zero. The exception to
3068 * this rule is zdb which is always allowed to access pools.
3070 if (spa_multihost(spa
) && spa_get_hostid() == 0 &&
3071 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
3072 fnvlist_add_uint64(spa
->spa_load_info
,
3073 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3074 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3078 * If the 'autoreplace' property is set, then post a resource notifying
3079 * the ZFS DE that it should not issue any faults for unopenable
3080 * devices. We also iterate over the vdevs, and post a sysevent for any
3081 * unopenable vdevs so that the normal autoreplace handler can take
3084 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
3085 spa_check_removed(spa
->spa_root_vdev
);
3087 * For the import case, this is done in spa_import(), because
3088 * at this point we're using the spare definitions from
3089 * the MOS config, not necessarily from the userland config.
3091 if (state
!= SPA_LOAD_IMPORT
) {
3092 spa_aux_check_removed(&spa
->spa_spares
);
3093 spa_aux_check_removed(&spa
->spa_l2cache
);
3098 * Load the vdev state for all toplevel vdevs.
3103 * Propagate the leaf DTLs we just loaded all the way up the tree.
3105 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3106 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3107 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3110 * Load the DDTs (dedup tables).
3112 error
= ddt_load(spa
);
3114 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3116 spa_update_dspace(spa
);
3119 * Validate the config, using the MOS config to fill in any
3120 * information which might be missing. If we fail to validate
3121 * the config then declare the pool unfit for use. If we're
3122 * assembling a pool from a split, the log is not transferred
3125 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3128 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
3129 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3131 if (!spa_config_valid(spa
, nvconfig
)) {
3132 nvlist_free(nvconfig
);
3133 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3136 nvlist_free(nvconfig
);
3139 * Now that we've validated the config, check the state of the
3140 * root vdev. If it can't be opened, it indicates one or
3141 * more toplevel vdevs are faulted.
3143 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
3144 return (SET_ERROR(ENXIO
));
3146 if (spa_writeable(spa
) && spa_check_logs(spa
)) {
3147 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3148 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
3152 if (missing_feat_write
) {
3153 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
3156 * At this point, we know that we can open the pool in
3157 * read-only mode but not read-write mode. We now have enough
3158 * information and can return to userland.
3160 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
3164 * We've successfully opened the pool, verify that we're ready
3165 * to start pushing transactions.
3167 if (state
!= SPA_LOAD_TRYIMPORT
) {
3168 if ((error
= spa_load_verify(spa
)))
3169 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3173 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
3174 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3176 int need_update
= B_FALSE
;
3177 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3179 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
3182 * Claim log blocks that haven't been committed yet.
3183 * This must all happen in a single txg.
3184 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3185 * invoked from zil_claim_log_block()'s i/o done callback.
3186 * Price of rollback is that we abandon the log.
3188 spa
->spa_claiming
= B_TRUE
;
3190 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3191 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3192 zil_claim
, tx
, DS_FIND_CHILDREN
);
3195 spa
->spa_claiming
= B_FALSE
;
3197 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3198 spa
->spa_sync_on
= B_TRUE
;
3199 txg_sync_start(spa
->spa_dsl_pool
);
3200 mmp_thread_start(spa
);
3203 * Wait for all claims to sync. We sync up to the highest
3204 * claimed log block birth time so that claimed log blocks
3205 * don't appear to be from the future. spa_claim_max_txg
3206 * will have been set for us by either zil_check_log_chain()
3207 * (invoked from spa_check_logs()) or zil_claim() above.
3209 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3212 * If the config cache is stale, or we have uninitialized
3213 * metaslabs (see spa_vdev_add()), then update the config.
3215 * If this is a verbatim import, trust the current
3216 * in-core spa_config and update the disk labels.
3218 if (config_cache_txg
!= spa
->spa_config_txg
||
3219 state
== SPA_LOAD_IMPORT
||
3220 state
== SPA_LOAD_RECOVER
||
3221 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3222 need_update
= B_TRUE
;
3224 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
3225 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3226 need_update
= B_TRUE
;
3229 * Update the config cache asychronously in case we're the
3230 * root pool, in which case the config cache isn't writable yet.
3233 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3236 * Check all DTLs to see if anything needs resilvering.
3238 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3239 vdev_resilver_needed(rvd
, NULL
, NULL
))
3240 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3243 * Log the fact that we booted up (so that we can detect if
3244 * we rebooted in the middle of an operation).
3246 spa_history_log_version(spa
, "open", NULL
);
3249 * Delete any inconsistent datasets.
3251 (void) dmu_objset_find(spa_name(spa
),
3252 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3255 * Clean up any stale temporary dataset userrefs.
3257 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3264 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
3266 int mode
= spa
->spa_mode
;
3269 spa_deactivate(spa
);
3271 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3273 spa_activate(spa
, mode
);
3274 spa_async_suspend(spa
);
3276 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
3280 * If spa_load() fails this function will try loading prior txg's. If
3281 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3282 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3283 * function will not rewind the pool and will return the same error as
3287 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
3288 uint64_t max_request
, int rewind_flags
)
3290 nvlist_t
*loadinfo
= NULL
;
3291 nvlist_t
*config
= NULL
;
3292 int load_error
, rewind_error
;
3293 uint64_t safe_rewind_txg
;
3296 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3297 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3298 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3300 spa
->spa_load_max_txg
= max_request
;
3301 if (max_request
!= UINT64_MAX
)
3302 spa
->spa_extreme_rewind
= B_TRUE
;
3305 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
3307 if (load_error
== 0)
3310 if (spa
->spa_root_vdev
!= NULL
)
3311 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3313 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3314 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3316 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3317 nvlist_free(config
);
3318 return (load_error
);
3321 if (state
== SPA_LOAD_RECOVER
) {
3322 /* Price of rolling back is discarding txgs, including log */
3323 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3326 * If we aren't rolling back save the load info from our first
3327 * import attempt so that we can restore it after attempting
3330 loadinfo
= spa
->spa_load_info
;
3331 spa
->spa_load_info
= fnvlist_alloc();
3334 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3335 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
3336 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
3337 TXG_INITIAL
: safe_rewind_txg
;
3340 * Continue as long as we're finding errors, we're still within
3341 * the acceptable rewind range, and we're still finding uberblocks
3343 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
3344 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
3345 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
3346 spa
->spa_extreme_rewind
= B_TRUE
;
3347 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
3350 spa
->spa_extreme_rewind
= B_FALSE
;
3351 spa
->spa_load_max_txg
= UINT64_MAX
;
3353 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
3354 spa_config_set(spa
, config
);
3356 nvlist_free(config
);
3358 if (state
== SPA_LOAD_RECOVER
) {
3359 ASSERT3P(loadinfo
, ==, NULL
);
3360 return (rewind_error
);
3362 /* Store the rewind info as part of the initial load info */
3363 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
3364 spa
->spa_load_info
);
3366 /* Restore the initial load info */
3367 fnvlist_free(spa
->spa_load_info
);
3368 spa
->spa_load_info
= loadinfo
;
3370 return (load_error
);
3377 * The import case is identical to an open except that the configuration is sent
3378 * down from userland, instead of grabbed from the configuration cache. For the
3379 * case of an open, the pool configuration will exist in the
3380 * POOL_STATE_UNINITIALIZED state.
3382 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3383 * the same time open the pool, without having to keep around the spa_t in some
3387 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
3391 spa_load_state_t state
= SPA_LOAD_OPEN
;
3393 int locked
= B_FALSE
;
3394 int firstopen
= B_FALSE
;
3399 * As disgusting as this is, we need to support recursive calls to this
3400 * function because dsl_dir_open() is called during spa_load(), and ends
3401 * up calling spa_open() again. The real fix is to figure out how to
3402 * avoid dsl_dir_open() calling this in the first place.
3404 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
3405 mutex_enter(&spa_namespace_lock
);
3409 if ((spa
= spa_lookup(pool
)) == NULL
) {
3411 mutex_exit(&spa_namespace_lock
);
3412 return (SET_ERROR(ENOENT
));
3415 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3416 zpool_rewind_policy_t policy
;
3420 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3422 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3423 state
= SPA_LOAD_RECOVER
;
3425 spa_activate(spa
, spa_mode_global
);
3427 if (state
!= SPA_LOAD_RECOVER
)
3428 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3430 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3431 policy
.zrp_request
);
3433 if (error
== EBADF
) {
3435 * If vdev_validate() returns failure (indicated by
3436 * EBADF), it indicates that one of the vdevs indicates
3437 * that the pool has been exported or destroyed. If
3438 * this is the case, the config cache is out of sync and
3439 * we should remove the pool from the namespace.
3442 spa_deactivate(spa
);
3443 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3446 mutex_exit(&spa_namespace_lock
);
3447 return (SET_ERROR(ENOENT
));
3452 * We can't open the pool, but we still have useful
3453 * information: the state of each vdev after the
3454 * attempted vdev_open(). Return this to the user.
3456 if (config
!= NULL
&& spa
->spa_config
) {
3457 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3459 VERIFY(nvlist_add_nvlist(*config
,
3460 ZPOOL_CONFIG_LOAD_INFO
,
3461 spa
->spa_load_info
) == 0);
3464 spa_deactivate(spa
);
3465 spa
->spa_last_open_failed
= error
;
3467 mutex_exit(&spa_namespace_lock
);
3473 spa_open_ref(spa
, tag
);
3476 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3479 * If we've recovered the pool, pass back any information we
3480 * gathered while doing the load.
3482 if (state
== SPA_LOAD_RECOVER
) {
3483 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3484 spa
->spa_load_info
) == 0);
3488 spa
->spa_last_open_failed
= 0;
3489 spa
->spa_last_ubsync_txg
= 0;
3490 spa
->spa_load_txg
= 0;
3491 mutex_exit(&spa_namespace_lock
);
3495 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
3503 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3506 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3510 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3512 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3516 * Lookup the given spa_t, incrementing the inject count in the process,
3517 * preventing it from being exported or destroyed.
3520 spa_inject_addref(char *name
)
3524 mutex_enter(&spa_namespace_lock
);
3525 if ((spa
= spa_lookup(name
)) == NULL
) {
3526 mutex_exit(&spa_namespace_lock
);
3529 spa
->spa_inject_ref
++;
3530 mutex_exit(&spa_namespace_lock
);
3536 spa_inject_delref(spa_t
*spa
)
3538 mutex_enter(&spa_namespace_lock
);
3539 spa
->spa_inject_ref
--;
3540 mutex_exit(&spa_namespace_lock
);
3544 * Add spares device information to the nvlist.
3547 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3557 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3559 if (spa
->spa_spares
.sav_count
== 0)
3562 VERIFY(nvlist_lookup_nvlist(config
,
3563 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3564 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3565 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3567 VERIFY(nvlist_add_nvlist_array(nvroot
,
3568 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3569 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3570 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3573 * Go through and find any spares which have since been
3574 * repurposed as an active spare. If this is the case, update
3575 * their status appropriately.
3577 for (i
= 0; i
< nspares
; i
++) {
3578 VERIFY(nvlist_lookup_uint64(spares
[i
],
3579 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3580 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3582 VERIFY(nvlist_lookup_uint64_array(
3583 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3584 (uint64_t **)&vs
, &vsc
) == 0);
3585 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3586 vs
->vs_aux
= VDEV_AUX_SPARED
;
3593 * Add l2cache device information to the nvlist, including vdev stats.
3596 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3599 uint_t i
, j
, nl2cache
;
3606 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3608 if (spa
->spa_l2cache
.sav_count
== 0)
3611 VERIFY(nvlist_lookup_nvlist(config
,
3612 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3613 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3614 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3615 if (nl2cache
!= 0) {
3616 VERIFY(nvlist_add_nvlist_array(nvroot
,
3617 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3618 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3619 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3622 * Update level 2 cache device stats.
3625 for (i
= 0; i
< nl2cache
; i
++) {
3626 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3627 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3630 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3632 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3633 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3639 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3640 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3642 vdev_get_stats(vd
, vs
);
3643 vdev_config_generate_stats(vd
, l2cache
[i
]);
3650 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3655 if (spa
->spa_feat_for_read_obj
!= 0) {
3656 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3657 spa
->spa_feat_for_read_obj
);
3658 zap_cursor_retrieve(&zc
, &za
) == 0;
3659 zap_cursor_advance(&zc
)) {
3660 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3661 za
.za_num_integers
== 1);
3662 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3663 za
.za_first_integer
));
3665 zap_cursor_fini(&zc
);
3668 if (spa
->spa_feat_for_write_obj
!= 0) {
3669 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3670 spa
->spa_feat_for_write_obj
);
3671 zap_cursor_retrieve(&zc
, &za
) == 0;
3672 zap_cursor_advance(&zc
)) {
3673 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3674 za
.za_num_integers
== 1);
3675 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3676 za
.za_first_integer
));
3678 zap_cursor_fini(&zc
);
3683 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3687 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3688 zfeature_info_t feature
= spa_feature_table
[i
];
3691 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3694 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3699 * Store a list of pool features and their reference counts in the
3702 * The first time this is called on a spa, allocate a new nvlist, fetch
3703 * the pool features and reference counts from disk, then save the list
3704 * in the spa. In subsequent calls on the same spa use the saved nvlist
3705 * and refresh its values from the cached reference counts. This
3706 * ensures we don't block here on I/O on a suspended pool so 'zpool
3707 * clear' can resume the pool.
3710 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3714 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3716 mutex_enter(&spa
->spa_feat_stats_lock
);
3717 features
= spa
->spa_feat_stats
;
3719 if (features
!= NULL
) {
3720 spa_feature_stats_from_cache(spa
, features
);
3722 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3723 spa
->spa_feat_stats
= features
;
3724 spa_feature_stats_from_disk(spa
, features
);
3727 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3730 mutex_exit(&spa
->spa_feat_stats_lock
);
3734 spa_get_stats(const char *name
, nvlist_t
**config
,
3735 char *altroot
, size_t buflen
)
3741 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3745 * This still leaves a window of inconsistency where the spares
3746 * or l2cache devices could change and the config would be
3747 * self-inconsistent.
3749 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3751 if (*config
!= NULL
) {
3752 uint64_t loadtimes
[2];
3754 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3755 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3756 VERIFY(nvlist_add_uint64_array(*config
,
3757 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3759 VERIFY(nvlist_add_uint64(*config
,
3760 ZPOOL_CONFIG_ERRCOUNT
,
3761 spa_get_errlog_size(spa
)) == 0);
3763 if (spa_suspended(spa
))
3764 VERIFY(nvlist_add_uint64(*config
,
3765 ZPOOL_CONFIG_SUSPENDED
,
3766 spa
->spa_failmode
) == 0);
3768 spa_add_spares(spa
, *config
);
3769 spa_add_l2cache(spa
, *config
);
3770 spa_add_feature_stats(spa
, *config
);
3775 * We want to get the alternate root even for faulted pools, so we cheat
3776 * and call spa_lookup() directly.
3780 mutex_enter(&spa_namespace_lock
);
3781 spa
= spa_lookup(name
);
3783 spa_altroot(spa
, altroot
, buflen
);
3787 mutex_exit(&spa_namespace_lock
);
3789 spa_altroot(spa
, altroot
, buflen
);
3794 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3795 spa_close(spa
, FTAG
);
3802 * Validate that the auxiliary device array is well formed. We must have an
3803 * array of nvlists, each which describes a valid leaf vdev. If this is an
3804 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3805 * specified, as long as they are well-formed.
3808 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3809 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3810 vdev_labeltype_t label
)
3817 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3820 * It's acceptable to have no devs specified.
3822 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3826 return (SET_ERROR(EINVAL
));
3829 * Make sure the pool is formatted with a version that supports this
3832 if (spa_version(spa
) < version
)
3833 return (SET_ERROR(ENOTSUP
));
3836 * Set the pending device list so we correctly handle device in-use
3839 sav
->sav_pending
= dev
;
3840 sav
->sav_npending
= ndev
;
3842 for (i
= 0; i
< ndev
; i
++) {
3843 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3847 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3849 error
= SET_ERROR(EINVAL
);
3855 if ((error
= vdev_open(vd
)) == 0 &&
3856 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3857 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3858 vd
->vdev_guid
) == 0);
3864 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3871 sav
->sav_pending
= NULL
;
3872 sav
->sav_npending
= 0;
3877 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3881 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3883 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3884 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3885 VDEV_LABEL_SPARE
)) != 0) {
3889 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3890 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3891 VDEV_LABEL_L2CACHE
));
3895 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3900 if (sav
->sav_config
!= NULL
) {
3906 * Generate new dev list by concatenating with the
3909 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3910 &olddevs
, &oldndevs
) == 0);
3912 newdevs
= kmem_alloc(sizeof (void *) *
3913 (ndevs
+ oldndevs
), KM_SLEEP
);
3914 for (i
= 0; i
< oldndevs
; i
++)
3915 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3917 for (i
= 0; i
< ndevs
; i
++)
3918 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3921 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3922 DATA_TYPE_NVLIST_ARRAY
) == 0);
3924 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3925 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3926 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3927 nvlist_free(newdevs
[i
]);
3928 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3931 * Generate a new dev list.
3933 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3935 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3941 * Stop and drop level 2 ARC devices
3944 spa_l2cache_drop(spa_t
*spa
)
3948 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3950 for (i
= 0; i
< sav
->sav_count
; i
++) {
3953 vd
= sav
->sav_vdevs
[i
];
3956 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3957 pool
!= 0ULL && l2arc_vdev_present(vd
))
3958 l2arc_remove_vdev(vd
);
3963 * Verify encryption parameters for spa creation. If we are encrypting, we must
3964 * have the encryption feature flag enabled.
3967 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
3968 boolean_t has_encryption
)
3970 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
3971 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
3973 return (SET_ERROR(ENOTSUP
));
3975 return (dmu_objset_create_crypt_check(NULL
, dcp
));
3982 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3983 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
3986 char *altroot
= NULL
;
3991 uint64_t txg
= TXG_INITIAL
;
3992 nvlist_t
**spares
, **l2cache
;
3993 uint_t nspares
, nl2cache
;
3994 uint64_t version
, obj
, root_dsobj
= 0;
3995 boolean_t has_features
;
3996 boolean_t has_encryption
;
4002 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
4003 poolname
= (char *)pool
;
4006 * If this pool already exists, return failure.
4008 mutex_enter(&spa_namespace_lock
);
4009 if (spa_lookup(poolname
) != NULL
) {
4010 mutex_exit(&spa_namespace_lock
);
4011 return (SET_ERROR(EEXIST
));
4015 * Allocate a new spa_t structure.
4017 nvl
= fnvlist_alloc();
4018 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
4019 (void) nvlist_lookup_string(props
,
4020 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4021 spa
= spa_add(poolname
, nvl
, altroot
);
4023 spa_activate(spa
, spa_mode_global
);
4025 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4026 spa_deactivate(spa
);
4028 mutex_exit(&spa_namespace_lock
);
4033 * Temporary pool names should never be written to disk.
4035 if (poolname
!= pool
)
4036 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
4038 has_features
= B_FALSE
;
4039 has_encryption
= B_FALSE
;
4040 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
4041 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4042 if (zpool_prop_feature(nvpair_name(elem
))) {
4043 has_features
= B_TRUE
;
4045 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
4046 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
4047 if (feat
== SPA_FEATURE_ENCRYPTION
)
4048 has_encryption
= B_TRUE
;
4052 /* verify encryption params, if they were provided */
4054 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
4056 spa_deactivate(spa
);
4058 mutex_exit(&spa_namespace_lock
);
4063 if (has_features
|| nvlist_lookup_uint64(props
,
4064 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4065 version
= SPA_VERSION
;
4067 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4069 spa
->spa_first_txg
= txg
;
4070 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4071 spa
->spa_uberblock
.ub_version
= version
;
4072 spa
->spa_ubsync
= spa
->spa_uberblock
;
4073 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4076 * Create "The Godfather" zio to hold all async IOs
4078 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4080 for (int i
= 0; i
< max_ncpus
; i
++) {
4081 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4082 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4083 ZIO_FLAG_GODFATHER
);
4087 * Create the root vdev.
4089 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4091 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4093 ASSERT(error
!= 0 || rvd
!= NULL
);
4094 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4096 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4097 error
= SET_ERROR(EINVAL
);
4100 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4101 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4102 VDEV_ALLOC_ADD
)) == 0) {
4103 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
4104 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4105 vdev_expand(rvd
->vdev_child
[c
], txg
);
4109 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4113 spa_deactivate(spa
);
4115 mutex_exit(&spa_namespace_lock
);
4120 * Get the list of spares, if specified.
4122 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4123 &spares
, &nspares
) == 0) {
4124 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4126 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4127 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4128 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4129 spa_load_spares(spa
);
4130 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4131 spa
->spa_spares
.sav_sync
= B_TRUE
;
4135 * Get the list of level 2 cache devices, if specified.
4137 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4138 &l2cache
, &nl2cache
) == 0) {
4139 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4140 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4141 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4142 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4143 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4144 spa_load_l2cache(spa
);
4145 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4146 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4149 spa
->spa_is_initializing
= B_TRUE
;
4150 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
4151 spa
->spa_is_initializing
= B_FALSE
;
4154 * Create DDTs (dedup tables).
4158 spa_update_dspace(spa
);
4160 tx
= dmu_tx_create_assigned(dp
, txg
);
4163 * Create the pool's history object.
4165 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
4166 spa_history_create_obj(spa
, tx
);
4168 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4169 spa_history_log_version(spa
, "create", tx
);
4172 * Create the pool config object.
4174 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4175 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4176 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4178 if (zap_add(spa
->spa_meta_objset
,
4179 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4180 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4181 cmn_err(CE_PANIC
, "failed to add pool config");
4184 if (zap_add(spa
->spa_meta_objset
,
4185 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4186 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4187 cmn_err(CE_PANIC
, "failed to add pool version");
4190 /* Newly created pools with the right version are always deflated. */
4191 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4192 spa
->spa_deflate
= TRUE
;
4193 if (zap_add(spa
->spa_meta_objset
,
4194 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4195 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4196 cmn_err(CE_PANIC
, "failed to add deflate");
4201 * Create the deferred-free bpobj. Turn off compression
4202 * because sync-to-convergence takes longer if the blocksize
4205 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4206 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4207 ZIO_COMPRESS_OFF
, tx
);
4208 if (zap_add(spa
->spa_meta_objset
,
4209 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4210 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4211 cmn_err(CE_PANIC
, "failed to add bpobj");
4213 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4214 spa
->spa_meta_objset
, obj
));
4217 * Generate some random noise for salted checksums to operate on.
4219 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4220 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4223 * Set pool properties.
4225 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4226 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4227 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4228 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4229 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
4231 if (props
!= NULL
) {
4232 spa_configfile_set(spa
, props
, B_FALSE
);
4233 spa_sync_props(props
, tx
);
4239 * If the root dataset is encrypted we will need to create key mappings
4240 * for the zio layer before we start to write any data to disk and hold
4241 * them until after the first txg has been synced. Waiting for the first
4242 * transaction to complete also ensures that our bean counters are
4243 * appropriately updated.
4245 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0) {
4246 root_dsobj
= dsl_dir_phys(dp
->dp_root_dir
)->dd_head_dataset_obj
;
4247 VERIFY0(spa_keystore_create_mapping_impl(spa
, root_dsobj
,
4248 dp
->dp_root_dir
, FTAG
));
4251 spa
->spa_sync_on
= B_TRUE
;
4253 mmp_thread_start(spa
);
4254 txg_wait_synced(dp
, txg
);
4256 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0)
4257 VERIFY0(spa_keystore_remove_mapping(spa
, root_dsobj
, FTAG
));
4259 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4262 * Don't count references from objsets that are already closed
4263 * and are making their way through the eviction process.
4265 spa_evicting_os_wait(spa
);
4266 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4267 spa
->spa_load_state
= SPA_LOAD_NONE
;
4269 mutex_exit(&spa_namespace_lock
);
4275 * Import a non-root pool into the system.
4278 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4281 char *altroot
= NULL
;
4282 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4283 zpool_rewind_policy_t policy
;
4284 uint64_t mode
= spa_mode_global
;
4285 uint64_t readonly
= B_FALSE
;
4288 nvlist_t
**spares
, **l2cache
;
4289 uint_t nspares
, nl2cache
;
4292 * If a pool with this name exists, return failure.
4294 mutex_enter(&spa_namespace_lock
);
4295 if (spa_lookup(pool
) != NULL
) {
4296 mutex_exit(&spa_namespace_lock
);
4297 return (SET_ERROR(EEXIST
));
4301 * Create and initialize the spa structure.
4303 (void) nvlist_lookup_string(props
,
4304 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4305 (void) nvlist_lookup_uint64(props
,
4306 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4309 spa
= spa_add(pool
, config
, altroot
);
4310 spa
->spa_import_flags
= flags
;
4313 * Verbatim import - Take a pool and insert it into the namespace
4314 * as if it had been loaded at boot.
4316 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4318 spa_configfile_set(spa
, props
, B_FALSE
);
4320 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4321 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4323 mutex_exit(&spa_namespace_lock
);
4327 spa_activate(spa
, mode
);
4330 * Don't start async tasks until we know everything is healthy.
4332 spa_async_suspend(spa
);
4334 zpool_get_rewind_policy(config
, &policy
);
4335 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4336 state
= SPA_LOAD_RECOVER
;
4339 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4340 * because the user-supplied config is actually the one to trust when
4343 if (state
!= SPA_LOAD_RECOVER
)
4344 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4346 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4347 policy
.zrp_request
);
4350 * Propagate anything learned while loading the pool and pass it
4351 * back to caller (i.e. rewind info, missing devices, etc).
4353 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4354 spa
->spa_load_info
) == 0);
4356 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4358 * Toss any existing sparelist, as it doesn't have any validity
4359 * anymore, and conflicts with spa_has_spare().
4361 if (spa
->spa_spares
.sav_config
) {
4362 nvlist_free(spa
->spa_spares
.sav_config
);
4363 spa
->spa_spares
.sav_config
= NULL
;
4364 spa_load_spares(spa
);
4366 if (spa
->spa_l2cache
.sav_config
) {
4367 nvlist_free(spa
->spa_l2cache
.sav_config
);
4368 spa
->spa_l2cache
.sav_config
= NULL
;
4369 spa_load_l2cache(spa
);
4372 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4374 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4377 spa_configfile_set(spa
, props
, B_FALSE
);
4379 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4380 (error
= spa_prop_set(spa
, props
)))) {
4382 spa_deactivate(spa
);
4384 mutex_exit(&spa_namespace_lock
);
4388 spa_async_resume(spa
);
4391 * Override any spares and level 2 cache devices as specified by
4392 * the user, as these may have correct device names/devids, etc.
4394 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4395 &spares
, &nspares
) == 0) {
4396 if (spa
->spa_spares
.sav_config
)
4397 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4398 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4400 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4401 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4402 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4403 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4404 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4405 spa_load_spares(spa
);
4406 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4407 spa
->spa_spares
.sav_sync
= B_TRUE
;
4409 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4410 &l2cache
, &nl2cache
) == 0) {
4411 if (spa
->spa_l2cache
.sav_config
)
4412 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4413 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4415 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4416 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4417 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4418 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4419 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4420 spa_load_l2cache(spa
);
4421 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4422 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4426 * Check for any removed devices.
4428 if (spa
->spa_autoreplace
) {
4429 spa_aux_check_removed(&spa
->spa_spares
);
4430 spa_aux_check_removed(&spa
->spa_l2cache
);
4433 if (spa_writeable(spa
)) {
4435 * Update the config cache to include the newly-imported pool.
4437 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4441 * It's possible that the pool was expanded while it was exported.
4442 * We kick off an async task to handle this for us.
4444 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4446 spa_history_log_version(spa
, "import", NULL
);
4448 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4450 zvol_create_minors(spa
, pool
, B_TRUE
);
4452 mutex_exit(&spa_namespace_lock
);
4458 spa_tryimport(nvlist_t
*tryconfig
)
4460 nvlist_t
*config
= NULL
;
4466 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4469 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4473 * Create and initialize the spa structure.
4475 mutex_enter(&spa_namespace_lock
);
4476 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4477 spa_activate(spa
, FREAD
);
4480 * Pass off the heavy lifting to spa_load().
4481 * Pass TRUE for mosconfig because the user-supplied config
4482 * is actually the one to trust when doing an import.
4484 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4487 * If 'tryconfig' was at least parsable, return the current config.
4489 if (spa
->spa_root_vdev
!= NULL
) {
4490 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4491 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4493 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4495 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4496 spa
->spa_uberblock
.ub_timestamp
) == 0);
4497 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4498 spa
->spa_load_info
) == 0);
4499 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4500 spa
->spa_errata
) == 0);
4503 * If the bootfs property exists on this pool then we
4504 * copy it out so that external consumers can tell which
4505 * pools are bootable.
4507 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4508 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4511 * We have to play games with the name since the
4512 * pool was opened as TRYIMPORT_NAME.
4514 if (dsl_dsobj_to_dsname(spa_name(spa
),
4515 spa
->spa_bootfs
, tmpname
) == 0) {
4519 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4521 cp
= strchr(tmpname
, '/');
4523 (void) strlcpy(dsname
, tmpname
,
4526 (void) snprintf(dsname
, MAXPATHLEN
,
4527 "%s/%s", poolname
, ++cp
);
4529 VERIFY(nvlist_add_string(config
,
4530 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4531 kmem_free(dsname
, MAXPATHLEN
);
4533 kmem_free(tmpname
, MAXPATHLEN
);
4537 * Add the list of hot spares and level 2 cache devices.
4539 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4540 spa_add_spares(spa
, config
);
4541 spa_add_l2cache(spa
, config
);
4542 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4546 spa_deactivate(spa
);
4548 mutex_exit(&spa_namespace_lock
);
4554 * Pool export/destroy
4556 * The act of destroying or exporting a pool is very simple. We make sure there
4557 * is no more pending I/O and any references to the pool are gone. Then, we
4558 * update the pool state and sync all the labels to disk, removing the
4559 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4560 * we don't sync the labels or remove the configuration cache.
4563 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4564 boolean_t force
, boolean_t hardforce
)
4571 if (!(spa_mode_global
& FWRITE
))
4572 return (SET_ERROR(EROFS
));
4574 mutex_enter(&spa_namespace_lock
);
4575 if ((spa
= spa_lookup(pool
)) == NULL
) {
4576 mutex_exit(&spa_namespace_lock
);
4577 return (SET_ERROR(ENOENT
));
4581 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4582 * reacquire the namespace lock, and see if we can export.
4584 spa_open_ref(spa
, FTAG
);
4585 mutex_exit(&spa_namespace_lock
);
4586 spa_async_suspend(spa
);
4587 if (spa
->spa_zvol_taskq
) {
4588 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
4589 taskq_wait(spa
->spa_zvol_taskq
);
4591 mutex_enter(&spa_namespace_lock
);
4592 spa_close(spa
, FTAG
);
4594 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4597 * The pool will be in core if it's openable, in which case we can
4598 * modify its state. Objsets may be open only because they're dirty,
4599 * so we have to force it to sync before checking spa_refcnt.
4601 if (spa
->spa_sync_on
) {
4602 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4603 spa_evicting_os_wait(spa
);
4607 * A pool cannot be exported or destroyed if there are active
4608 * references. If we are resetting a pool, allow references by
4609 * fault injection handlers.
4611 if (!spa_refcount_zero(spa
) ||
4612 (spa
->spa_inject_ref
!= 0 &&
4613 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4614 spa_async_resume(spa
);
4615 mutex_exit(&spa_namespace_lock
);
4616 return (SET_ERROR(EBUSY
));
4619 if (spa
->spa_sync_on
) {
4621 * A pool cannot be exported if it has an active shared spare.
4622 * This is to prevent other pools stealing the active spare
4623 * from an exported pool. At user's own will, such pool can
4624 * be forcedly exported.
4626 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4627 spa_has_active_shared_spare(spa
)) {
4628 spa_async_resume(spa
);
4629 mutex_exit(&spa_namespace_lock
);
4630 return (SET_ERROR(EXDEV
));
4634 * We want this to be reflected on every label,
4635 * so mark them all dirty. spa_unload() will do the
4636 * final sync that pushes these changes out.
4638 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4639 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4640 spa
->spa_state
= new_state
;
4641 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4643 vdev_config_dirty(spa
->spa_root_vdev
);
4644 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4649 if (new_state
== POOL_STATE_DESTROYED
)
4650 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
4651 else if (new_state
== POOL_STATE_EXPORTED
)
4652 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
4654 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4656 spa_deactivate(spa
);
4659 if (oldconfig
&& spa
->spa_config
)
4660 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4662 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4664 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4667 mutex_exit(&spa_namespace_lock
);
4673 * Destroy a storage pool.
4676 spa_destroy(char *pool
)
4678 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4683 * Export a storage pool.
4686 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4687 boolean_t hardforce
)
4689 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4694 * Similar to spa_export(), this unloads the spa_t without actually removing it
4695 * from the namespace in any way.
4698 spa_reset(char *pool
)
4700 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4705 * ==========================================================================
4706 * Device manipulation
4707 * ==========================================================================
4711 * Add a device to a storage pool.
4714 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4718 vdev_t
*rvd
= spa
->spa_root_vdev
;
4720 nvlist_t
**spares
, **l2cache
;
4721 uint_t nspares
, nl2cache
;
4723 ASSERT(spa_writeable(spa
));
4725 txg
= spa_vdev_enter(spa
);
4727 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4728 VDEV_ALLOC_ADD
)) != 0)
4729 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4731 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4733 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4737 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4741 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4742 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4744 if (vd
->vdev_children
!= 0 &&
4745 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4746 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4749 * We must validate the spares and l2cache devices after checking the
4750 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4752 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4753 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4756 * Transfer each new top-level vdev from vd to rvd.
4758 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4761 * Set the vdev id to the first hole, if one exists.
4763 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4764 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4765 vdev_free(rvd
->vdev_child
[id
]);
4769 tvd
= vd
->vdev_child
[c
];
4770 vdev_remove_child(vd
, tvd
);
4772 vdev_add_child(rvd
, tvd
);
4773 vdev_config_dirty(tvd
);
4777 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4778 ZPOOL_CONFIG_SPARES
);
4779 spa_load_spares(spa
);
4780 spa
->spa_spares
.sav_sync
= B_TRUE
;
4783 if (nl2cache
!= 0) {
4784 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4785 ZPOOL_CONFIG_L2CACHE
);
4786 spa_load_l2cache(spa
);
4787 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4791 * We have to be careful when adding new vdevs to an existing pool.
4792 * If other threads start allocating from these vdevs before we
4793 * sync the config cache, and we lose power, then upon reboot we may
4794 * fail to open the pool because there are DVAs that the config cache
4795 * can't translate. Therefore, we first add the vdevs without
4796 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4797 * and then let spa_config_update() initialize the new metaslabs.
4799 * spa_load() checks for added-but-not-initialized vdevs, so that
4800 * if we lose power at any point in this sequence, the remaining
4801 * steps will be completed the next time we load the pool.
4803 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4805 mutex_enter(&spa_namespace_lock
);
4806 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4807 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
4808 mutex_exit(&spa_namespace_lock
);
4814 * Attach a device to a mirror. The arguments are the path to any device
4815 * in the mirror, and the nvroot for the new device. If the path specifies
4816 * a device that is not mirrored, we automatically insert the mirror vdev.
4818 * If 'replacing' is specified, the new device is intended to replace the
4819 * existing device; in this case the two devices are made into their own
4820 * mirror using the 'replacing' vdev, which is functionally identical to
4821 * the mirror vdev (it actually reuses all the same ops) but has a few
4822 * extra rules: you can't attach to it after it's been created, and upon
4823 * completion of resilvering, the first disk (the one being replaced)
4824 * is automatically detached.
4827 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4829 uint64_t txg
, dtl_max_txg
;
4830 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4831 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4833 char *oldvdpath
, *newvdpath
;
4837 ASSERT(spa_writeable(spa
));
4839 txg
= spa_vdev_enter(spa
);
4841 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4844 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4846 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4847 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4849 pvd
= oldvd
->vdev_parent
;
4851 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4852 VDEV_ALLOC_ATTACH
)) != 0)
4853 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4855 if (newrootvd
->vdev_children
!= 1)
4856 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4858 newvd
= newrootvd
->vdev_child
[0];
4860 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4861 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4863 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4864 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4867 * Spares can't replace logs
4869 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4870 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4874 * For attach, the only allowable parent is a mirror or the root
4877 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4878 pvd
->vdev_ops
!= &vdev_root_ops
)
4879 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4881 pvops
= &vdev_mirror_ops
;
4884 * Active hot spares can only be replaced by inactive hot
4887 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4888 oldvd
->vdev_isspare
&&
4889 !spa_has_spare(spa
, newvd
->vdev_guid
))
4890 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4893 * If the source is a hot spare, and the parent isn't already a
4894 * spare, then we want to create a new hot spare. Otherwise, we
4895 * want to create a replacing vdev. The user is not allowed to
4896 * attach to a spared vdev child unless the 'isspare' state is
4897 * the same (spare replaces spare, non-spare replaces
4900 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4901 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4902 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4903 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4904 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4905 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4908 if (newvd
->vdev_isspare
)
4909 pvops
= &vdev_spare_ops
;
4911 pvops
= &vdev_replacing_ops
;
4915 * Make sure the new device is big enough.
4917 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4918 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4921 * The new device cannot have a higher alignment requirement
4922 * than the top-level vdev.
4924 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4925 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4928 * If this is an in-place replacement, update oldvd's path and devid
4929 * to make it distinguishable from newvd, and unopenable from now on.
4931 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4932 spa_strfree(oldvd
->vdev_path
);
4933 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4935 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4936 newvd
->vdev_path
, "old");
4937 if (oldvd
->vdev_devid
!= NULL
) {
4938 spa_strfree(oldvd
->vdev_devid
);
4939 oldvd
->vdev_devid
= NULL
;
4943 /* mark the device being resilvered */
4944 newvd
->vdev_resilver_txg
= txg
;
4947 * If the parent is not a mirror, or if we're replacing, insert the new
4948 * mirror/replacing/spare vdev above oldvd.
4950 if (pvd
->vdev_ops
!= pvops
)
4951 pvd
= vdev_add_parent(oldvd
, pvops
);
4953 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4954 ASSERT(pvd
->vdev_ops
== pvops
);
4955 ASSERT(oldvd
->vdev_parent
== pvd
);
4958 * Extract the new device from its root and add it to pvd.
4960 vdev_remove_child(newrootvd
, newvd
);
4961 newvd
->vdev_id
= pvd
->vdev_children
;
4962 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4963 vdev_add_child(pvd
, newvd
);
4966 * Reevaluate the parent vdev state.
4968 vdev_propagate_state(pvd
);
4970 tvd
= newvd
->vdev_top
;
4971 ASSERT(pvd
->vdev_top
== tvd
);
4972 ASSERT(tvd
->vdev_parent
== rvd
);
4974 vdev_config_dirty(tvd
);
4977 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4978 * for any dmu_sync-ed blocks. It will propagate upward when
4979 * spa_vdev_exit() calls vdev_dtl_reassess().
4981 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4983 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4984 dtl_max_txg
- TXG_INITIAL
);
4986 if (newvd
->vdev_isspare
) {
4987 spa_spare_activate(newvd
);
4988 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
4991 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4992 newvdpath
= spa_strdup(newvd
->vdev_path
);
4993 newvd_isspare
= newvd
->vdev_isspare
;
4996 * Mark newvd's DTL dirty in this txg.
4998 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
5001 * Schedule the resilver to restart in the future. We do this to
5002 * ensure that dmu_sync-ed blocks have been stitched into the
5003 * respective datasets.
5005 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
5007 if (spa
->spa_bootfs
)
5008 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
5010 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
5015 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
5017 spa_history_log_internal(spa
, "vdev attach", NULL
,
5018 "%s vdev=%s %s vdev=%s",
5019 replacing
&& newvd_isspare
? "spare in" :
5020 replacing
? "replace" : "attach", newvdpath
,
5021 replacing
? "for" : "to", oldvdpath
);
5023 spa_strfree(oldvdpath
);
5024 spa_strfree(newvdpath
);
5030 * Detach a device from a mirror or replacing vdev.
5032 * If 'replace_done' is specified, only detach if the parent
5033 * is a replacing vdev.
5036 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5040 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5041 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5042 boolean_t unspare
= B_FALSE
;
5043 uint64_t unspare_guid
= 0;
5046 ASSERT(spa_writeable(spa
));
5048 txg
= spa_vdev_enter(spa
);
5050 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5053 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5055 if (!vd
->vdev_ops
->vdev_op_leaf
)
5056 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5058 pvd
= vd
->vdev_parent
;
5061 * If the parent/child relationship is not as expected, don't do it.
5062 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5063 * vdev that's replacing B with C. The user's intent in replacing
5064 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5065 * the replace by detaching C, the expected behavior is to end up
5066 * M(A,B). But suppose that right after deciding to detach C,
5067 * the replacement of B completes. We would have M(A,C), and then
5068 * ask to detach C, which would leave us with just A -- not what
5069 * the user wanted. To prevent this, we make sure that the
5070 * parent/child relationship hasn't changed -- in this example,
5071 * that C's parent is still the replacing vdev R.
5073 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5074 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5077 * Only 'replacing' or 'spare' vdevs can be replaced.
5079 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5080 pvd
->vdev_ops
!= &vdev_spare_ops
)
5081 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5083 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5084 spa_version(spa
) >= SPA_VERSION_SPARES
);
5087 * Only mirror, replacing, and spare vdevs support detach.
5089 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5090 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5091 pvd
->vdev_ops
!= &vdev_spare_ops
)
5092 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5095 * If this device has the only valid copy of some data,
5096 * we cannot safely detach it.
5098 if (vdev_dtl_required(vd
))
5099 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5101 ASSERT(pvd
->vdev_children
>= 2);
5104 * If we are detaching the second disk from a replacing vdev, then
5105 * check to see if we changed the original vdev's path to have "/old"
5106 * at the end in spa_vdev_attach(). If so, undo that change now.
5108 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5109 vd
->vdev_path
!= NULL
) {
5110 size_t len
= strlen(vd
->vdev_path
);
5112 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
5113 cvd
= pvd
->vdev_child
[c
];
5115 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5118 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5119 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5120 spa_strfree(cvd
->vdev_path
);
5121 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5128 * If we are detaching the original disk from a spare, then it implies
5129 * that the spare should become a real disk, and be removed from the
5130 * active spare list for the pool.
5132 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5134 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5138 * Erase the disk labels so the disk can be used for other things.
5139 * This must be done after all other error cases are handled,
5140 * but before we disembowel vd (so we can still do I/O to it).
5141 * But if we can't do it, don't treat the error as fatal --
5142 * it may be that the unwritability of the disk is the reason
5143 * it's being detached!
5145 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5148 * Remove vd from its parent and compact the parent's children.
5150 vdev_remove_child(pvd
, vd
);
5151 vdev_compact_children(pvd
);
5154 * Remember one of the remaining children so we can get tvd below.
5156 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5159 * If we need to remove the remaining child from the list of hot spares,
5160 * do it now, marking the vdev as no longer a spare in the process.
5161 * We must do this before vdev_remove_parent(), because that can
5162 * change the GUID if it creates a new toplevel GUID. For a similar
5163 * reason, we must remove the spare now, in the same txg as the detach;
5164 * otherwise someone could attach a new sibling, change the GUID, and
5165 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5168 ASSERT(cvd
->vdev_isspare
);
5169 spa_spare_remove(cvd
);
5170 unspare_guid
= cvd
->vdev_guid
;
5171 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5172 cvd
->vdev_unspare
= B_TRUE
;
5176 * If the parent mirror/replacing vdev only has one child,
5177 * the parent is no longer needed. Remove it from the tree.
5179 if (pvd
->vdev_children
== 1) {
5180 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5181 cvd
->vdev_unspare
= B_FALSE
;
5182 vdev_remove_parent(cvd
);
5187 * We don't set tvd until now because the parent we just removed
5188 * may have been the previous top-level vdev.
5190 tvd
= cvd
->vdev_top
;
5191 ASSERT(tvd
->vdev_parent
== rvd
);
5194 * Reevaluate the parent vdev state.
5196 vdev_propagate_state(cvd
);
5199 * If the 'autoexpand' property is set on the pool then automatically
5200 * try to expand the size of the pool. For example if the device we
5201 * just detached was smaller than the others, it may be possible to
5202 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5203 * first so that we can obtain the updated sizes of the leaf vdevs.
5205 if (spa
->spa_autoexpand
) {
5207 vdev_expand(tvd
, txg
);
5210 vdev_config_dirty(tvd
);
5213 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5214 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5215 * But first make sure we're not on any *other* txg's DTL list, to
5216 * prevent vd from being accessed after it's freed.
5218 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
5219 for (int t
= 0; t
< TXG_SIZE
; t
++)
5220 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5221 vd
->vdev_detached
= B_TRUE
;
5222 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5224 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5226 /* hang on to the spa before we release the lock */
5227 spa_open_ref(spa
, FTAG
);
5229 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5231 spa_history_log_internal(spa
, "detach", NULL
,
5233 spa_strfree(vdpath
);
5236 * If this was the removal of the original device in a hot spare vdev,
5237 * then we want to go through and remove the device from the hot spare
5238 * list of every other pool.
5241 spa_t
*altspa
= NULL
;
5243 mutex_enter(&spa_namespace_lock
);
5244 while ((altspa
= spa_next(altspa
)) != NULL
) {
5245 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5249 spa_open_ref(altspa
, FTAG
);
5250 mutex_exit(&spa_namespace_lock
);
5251 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5252 mutex_enter(&spa_namespace_lock
);
5253 spa_close(altspa
, FTAG
);
5255 mutex_exit(&spa_namespace_lock
);
5257 /* search the rest of the vdevs for spares to remove */
5258 spa_vdev_resilver_done(spa
);
5261 /* all done with the spa; OK to release */
5262 mutex_enter(&spa_namespace_lock
);
5263 spa_close(spa
, FTAG
);
5264 mutex_exit(&spa_namespace_lock
);
5270 * Split a set of devices from their mirrors, and create a new pool from them.
5273 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5274 nvlist_t
*props
, boolean_t exp
)
5277 uint64_t txg
, *glist
;
5279 uint_t c
, children
, lastlog
;
5280 nvlist_t
**child
, *nvl
, *tmp
;
5282 char *altroot
= NULL
;
5283 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5284 boolean_t activate_slog
;
5286 ASSERT(spa_writeable(spa
));
5288 txg
= spa_vdev_enter(spa
);
5290 /* clear the log and flush everything up to now */
5291 activate_slog
= spa_passivate_log(spa
);
5292 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5293 error
= spa_offline_log(spa
);
5294 txg
= spa_vdev_config_enter(spa
);
5297 spa_activate_log(spa
);
5300 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5302 /* check new spa name before going any further */
5303 if (spa_lookup(newname
) != NULL
)
5304 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5307 * scan through all the children to ensure they're all mirrors
5309 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5310 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5312 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5314 /* first, check to ensure we've got the right child count */
5315 rvd
= spa
->spa_root_vdev
;
5317 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5318 vdev_t
*vd
= rvd
->vdev_child
[c
];
5320 /* don't count the holes & logs as children */
5321 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5329 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5330 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5332 /* next, ensure no spare or cache devices are part of the split */
5333 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5334 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5335 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5337 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5338 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5340 /* then, loop over each vdev and validate it */
5341 for (c
= 0; c
< children
; c
++) {
5342 uint64_t is_hole
= 0;
5344 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5348 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5349 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5352 error
= SET_ERROR(EINVAL
);
5357 /* which disk is going to be split? */
5358 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5360 error
= SET_ERROR(EINVAL
);
5364 /* look it up in the spa */
5365 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5366 if (vml
[c
] == NULL
) {
5367 error
= SET_ERROR(ENODEV
);
5371 /* make sure there's nothing stopping the split */
5372 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5373 vml
[c
]->vdev_islog
||
5374 vml
[c
]->vdev_ishole
||
5375 vml
[c
]->vdev_isspare
||
5376 vml
[c
]->vdev_isl2cache
||
5377 !vdev_writeable(vml
[c
]) ||
5378 vml
[c
]->vdev_children
!= 0 ||
5379 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5380 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5381 error
= SET_ERROR(EINVAL
);
5385 if (vdev_dtl_required(vml
[c
])) {
5386 error
= SET_ERROR(EBUSY
);
5390 /* we need certain info from the top level */
5391 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5392 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5393 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5394 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5395 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5396 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5397 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5398 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5400 /* transfer per-vdev ZAPs */
5401 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
5402 VERIFY0(nvlist_add_uint64(child
[c
],
5403 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
5405 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
5406 VERIFY0(nvlist_add_uint64(child
[c
],
5407 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
5408 vml
[c
]->vdev_parent
->vdev_top_zap
));
5412 kmem_free(vml
, children
* sizeof (vdev_t
*));
5413 kmem_free(glist
, children
* sizeof (uint64_t));
5414 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5417 /* stop writers from using the disks */
5418 for (c
= 0; c
< children
; c
++) {
5420 vml
[c
]->vdev_offline
= B_TRUE
;
5422 vdev_reopen(spa
->spa_root_vdev
);
5425 * Temporarily record the splitting vdevs in the spa config. This
5426 * will disappear once the config is regenerated.
5428 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5429 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5430 glist
, children
) == 0);
5431 kmem_free(glist
, children
* sizeof (uint64_t));
5433 mutex_enter(&spa
->spa_props_lock
);
5434 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5436 mutex_exit(&spa
->spa_props_lock
);
5437 spa
->spa_config_splitting
= nvl
;
5438 vdev_config_dirty(spa
->spa_root_vdev
);
5440 /* configure and create the new pool */
5441 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5442 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5443 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5444 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5445 spa_version(spa
)) == 0);
5446 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5447 spa
->spa_config_txg
) == 0);
5448 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5449 spa_generate_guid(NULL
)) == 0);
5450 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
5451 (void) nvlist_lookup_string(props
,
5452 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5454 /* add the new pool to the namespace */
5455 newspa
= spa_add(newname
, config
, altroot
);
5456 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5457 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5458 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5460 /* release the spa config lock, retaining the namespace lock */
5461 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5463 if (zio_injection_enabled
)
5464 zio_handle_panic_injection(spa
, FTAG
, 1);
5466 spa_activate(newspa
, spa_mode_global
);
5467 spa_async_suspend(newspa
);
5469 /* create the new pool from the disks of the original pool */
5470 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5474 /* if that worked, generate a real config for the new pool */
5475 if (newspa
->spa_root_vdev
!= NULL
) {
5476 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5477 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5478 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5479 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5480 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5485 if (props
!= NULL
) {
5486 spa_configfile_set(newspa
, props
, B_FALSE
);
5487 error
= spa_prop_set(newspa
, props
);
5492 /* flush everything */
5493 txg
= spa_vdev_config_enter(newspa
);
5494 vdev_config_dirty(newspa
->spa_root_vdev
);
5495 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5497 if (zio_injection_enabled
)
5498 zio_handle_panic_injection(spa
, FTAG
, 2);
5500 spa_async_resume(newspa
);
5502 /* finally, update the original pool's config */
5503 txg
= spa_vdev_config_enter(spa
);
5504 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5505 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5508 for (c
= 0; c
< children
; c
++) {
5509 if (vml
[c
] != NULL
) {
5512 spa_history_log_internal(spa
, "detach", tx
,
5513 "vdev=%s", vml
[c
]->vdev_path
);
5518 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5519 vdev_config_dirty(spa
->spa_root_vdev
);
5520 spa
->spa_config_splitting
= NULL
;
5524 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5526 if (zio_injection_enabled
)
5527 zio_handle_panic_injection(spa
, FTAG
, 3);
5529 /* split is complete; log a history record */
5530 spa_history_log_internal(newspa
, "split", NULL
,
5531 "from pool %s", spa_name(spa
));
5533 kmem_free(vml
, children
* sizeof (vdev_t
*));
5535 /* if we're not going to mount the filesystems in userland, export */
5537 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5544 spa_deactivate(newspa
);
5547 txg
= spa_vdev_config_enter(spa
);
5549 /* re-online all offlined disks */
5550 for (c
= 0; c
< children
; c
++) {
5552 vml
[c
]->vdev_offline
= B_FALSE
;
5554 vdev_reopen(spa
->spa_root_vdev
);
5556 nvlist_free(spa
->spa_config_splitting
);
5557 spa
->spa_config_splitting
= NULL
;
5558 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5560 kmem_free(vml
, children
* sizeof (vdev_t
*));
5565 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5567 for (int i
= 0; i
< count
; i
++) {
5570 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5573 if (guid
== target_guid
)
5581 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5582 nvlist_t
*dev_to_remove
)
5584 nvlist_t
**newdev
= NULL
;
5587 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5589 for (int i
= 0, j
= 0; i
< count
; i
++) {
5590 if (dev
[i
] == dev_to_remove
)
5592 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5595 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5596 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5598 for (int i
= 0; i
< count
- 1; i
++)
5599 nvlist_free(newdev
[i
]);
5602 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5606 * Evacuate the device.
5609 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5614 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5615 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5616 ASSERT(vd
== vd
->vdev_top
);
5619 * Evacuate the device. We don't hold the config lock as writer
5620 * since we need to do I/O but we do keep the
5621 * spa_namespace_lock held. Once this completes the device
5622 * should no longer have any blocks allocated on it.
5624 if (vd
->vdev_islog
) {
5625 if (vd
->vdev_stat
.vs_alloc
!= 0)
5626 error
= spa_offline_log(spa
);
5628 error
= SET_ERROR(ENOTSUP
);
5635 * The evacuation succeeded. Remove any remaining MOS metadata
5636 * associated with this vdev, and wait for these changes to sync.
5638 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5639 txg
= spa_vdev_config_enter(spa
);
5640 vd
->vdev_removing
= B_TRUE
;
5641 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5642 vdev_config_dirty(vd
);
5643 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5649 * Complete the removal by cleaning up the namespace.
5652 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5654 vdev_t
*rvd
= spa
->spa_root_vdev
;
5655 uint64_t id
= vd
->vdev_id
;
5656 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5658 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5659 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5660 ASSERT(vd
== vd
->vdev_top
);
5663 * Only remove any devices which are empty.
5665 if (vd
->vdev_stat
.vs_alloc
!= 0)
5668 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5670 if (list_link_active(&vd
->vdev_state_dirty_node
))
5671 vdev_state_clean(vd
);
5672 if (list_link_active(&vd
->vdev_config_dirty_node
))
5673 vdev_config_clean(vd
);
5678 vdev_compact_children(rvd
);
5680 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5681 vdev_add_child(rvd
, vd
);
5683 vdev_config_dirty(rvd
);
5686 * Reassess the health of our root vdev.
5692 * Remove a device from the pool -
5694 * Removing a device from the vdev namespace requires several steps
5695 * and can take a significant amount of time. As a result we use
5696 * the spa_vdev_config_[enter/exit] functions which allow us to
5697 * grab and release the spa_config_lock while still holding the namespace
5698 * lock. During each step the configuration is synced out.
5700 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5704 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5707 sysevent_t
*ev
= NULL
;
5708 metaslab_group_t
*mg
;
5709 nvlist_t
**spares
, **l2cache
, *nv
;
5711 uint_t nspares
, nl2cache
;
5713 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5715 ASSERT(spa_writeable(spa
));
5718 txg
= spa_vdev_enter(spa
);
5720 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5722 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5723 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5724 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5725 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5727 * Only remove the hot spare if it's not currently in use
5730 if (vd
== NULL
|| unspare
) {
5732 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5733 ev
= spa_event_create(spa
, vd
, NULL
,
5734 ESC_ZFS_VDEV_REMOVE_AUX
);
5735 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5736 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5737 spa_load_spares(spa
);
5738 spa
->spa_spares
.sav_sync
= B_TRUE
;
5740 error
= SET_ERROR(EBUSY
);
5742 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5743 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5744 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5745 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5747 * Cache devices can always be removed.
5749 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5750 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_AUX
);
5751 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5752 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5753 spa_load_l2cache(spa
);
5754 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5755 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5757 ASSERT(vd
== vd
->vdev_top
);
5762 * Stop allocating from this vdev.
5764 metaslab_group_passivate(mg
);
5767 * Wait for the youngest allocations and frees to sync,
5768 * and then wait for the deferral of those frees to finish.
5770 spa_vdev_config_exit(spa
, NULL
,
5771 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5774 * Attempt to evacuate the vdev.
5776 error
= spa_vdev_remove_evacuate(spa
, vd
);
5778 txg
= spa_vdev_config_enter(spa
);
5781 * If we couldn't evacuate the vdev, unwind.
5784 metaslab_group_activate(mg
);
5785 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5789 * Clean up the vdev namespace.
5791 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_DEV
);
5792 spa_vdev_remove_from_namespace(spa
, vd
);
5794 } else if (vd
!= NULL
) {
5796 * Normal vdevs cannot be removed (yet).
5798 error
= SET_ERROR(ENOTSUP
);
5801 * There is no vdev of any kind with the specified guid.
5803 error
= SET_ERROR(ENOENT
);
5807 error
= spa_vdev_exit(spa
, NULL
, txg
, error
);
5816 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5817 * currently spared, so we can detach it.
5820 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5822 vdev_t
*newvd
, *oldvd
;
5824 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5825 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5831 * Check for a completed replacement. We always consider the first
5832 * vdev in the list to be the oldest vdev, and the last one to be
5833 * the newest (see spa_vdev_attach() for how that works). In
5834 * the case where the newest vdev is faulted, we will not automatically
5835 * remove it after a resilver completes. This is OK as it will require
5836 * user intervention to determine which disk the admin wishes to keep.
5838 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5839 ASSERT(vd
->vdev_children
> 1);
5841 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5842 oldvd
= vd
->vdev_child
[0];
5844 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5845 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5846 !vdev_dtl_required(oldvd
))
5851 * Check for a completed resilver with the 'unspare' flag set.
5853 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5854 vdev_t
*first
= vd
->vdev_child
[0];
5855 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5857 if (last
->vdev_unspare
) {
5860 } else if (first
->vdev_unspare
) {
5867 if (oldvd
!= NULL
&&
5868 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5869 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5870 !vdev_dtl_required(oldvd
))
5874 * If there are more than two spares attached to a disk,
5875 * and those spares are not required, then we want to
5876 * attempt to free them up now so that they can be used
5877 * by other pools. Once we're back down to a single
5878 * disk+spare, we stop removing them.
5880 if (vd
->vdev_children
> 2) {
5881 newvd
= vd
->vdev_child
[1];
5883 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5884 vdev_dtl_empty(last
, DTL_MISSING
) &&
5885 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5886 !vdev_dtl_required(newvd
))
5895 spa_vdev_resilver_done(spa_t
*spa
)
5897 vdev_t
*vd
, *pvd
, *ppvd
;
5898 uint64_t guid
, sguid
, pguid
, ppguid
;
5900 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5902 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5903 pvd
= vd
->vdev_parent
;
5904 ppvd
= pvd
->vdev_parent
;
5905 guid
= vd
->vdev_guid
;
5906 pguid
= pvd
->vdev_guid
;
5907 ppguid
= ppvd
->vdev_guid
;
5910 * If we have just finished replacing a hot spared device, then
5911 * we need to detach the parent's first child (the original hot
5914 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5915 ppvd
->vdev_children
== 2) {
5916 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5917 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5919 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5921 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5922 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5924 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5926 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5929 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5933 * Update the stored path or FRU for this vdev.
5936 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5940 boolean_t sync
= B_FALSE
;
5942 ASSERT(spa_writeable(spa
));
5944 spa_vdev_state_enter(spa
, SCL_ALL
);
5946 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5947 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5949 if (!vd
->vdev_ops
->vdev_op_leaf
)
5950 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5953 if (strcmp(value
, vd
->vdev_path
) != 0) {
5954 spa_strfree(vd
->vdev_path
);
5955 vd
->vdev_path
= spa_strdup(value
);
5959 if (vd
->vdev_fru
== NULL
) {
5960 vd
->vdev_fru
= spa_strdup(value
);
5962 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5963 spa_strfree(vd
->vdev_fru
);
5964 vd
->vdev_fru
= spa_strdup(value
);
5969 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5973 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5975 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5979 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5981 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5985 * ==========================================================================
5987 * ==========================================================================
5990 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
5992 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5994 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5995 return (SET_ERROR(EBUSY
));
5997 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
6001 spa_scan_stop(spa_t
*spa
)
6003 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6004 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6005 return (SET_ERROR(EBUSY
));
6006 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
6010 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
6012 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6014 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
6015 return (SET_ERROR(ENOTSUP
));
6018 * If a resilver was requested, but there is no DTL on a
6019 * writeable leaf device, we have nothing to do.
6021 if (func
== POOL_SCAN_RESILVER
&&
6022 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
6023 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6027 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6031 * ==========================================================================
6032 * SPA async task processing
6033 * ==========================================================================
6037 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6039 if (vd
->vdev_remove_wanted
) {
6040 vd
->vdev_remove_wanted
= B_FALSE
;
6041 vd
->vdev_delayed_close
= B_FALSE
;
6042 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6045 * We want to clear the stats, but we don't want to do a full
6046 * vdev_clear() as that will cause us to throw away
6047 * degraded/faulted state as well as attempt to reopen the
6048 * device, all of which is a waste.
6050 vd
->vdev_stat
.vs_read_errors
= 0;
6051 vd
->vdev_stat
.vs_write_errors
= 0;
6052 vd
->vdev_stat
.vs_checksum_errors
= 0;
6054 vdev_state_dirty(vd
->vdev_top
);
6057 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6058 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6062 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6064 if (vd
->vdev_probe_wanted
) {
6065 vd
->vdev_probe_wanted
= B_FALSE
;
6066 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6069 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6070 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6074 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6076 if (!spa
->spa_autoexpand
)
6079 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6080 vdev_t
*cvd
= vd
->vdev_child
[c
];
6081 spa_async_autoexpand(spa
, cvd
);
6084 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6087 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
6091 spa_async_thread(void *arg
)
6093 spa_t
*spa
= (spa_t
*)arg
;
6096 ASSERT(spa
->spa_sync_on
);
6098 mutex_enter(&spa
->spa_async_lock
);
6099 tasks
= spa
->spa_async_tasks
;
6100 spa
->spa_async_tasks
= 0;
6101 mutex_exit(&spa
->spa_async_lock
);
6104 * See if the config needs to be updated.
6106 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6107 uint64_t old_space
, new_space
;
6109 mutex_enter(&spa_namespace_lock
);
6110 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6111 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6112 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6113 mutex_exit(&spa_namespace_lock
);
6116 * If the pool grew as a result of the config update,
6117 * then log an internal history event.
6119 if (new_space
!= old_space
) {
6120 spa_history_log_internal(spa
, "vdev online", NULL
,
6121 "pool '%s' size: %llu(+%llu)",
6122 spa_name(spa
), new_space
, new_space
- old_space
);
6127 * See if any devices need to be marked REMOVED.
6129 if (tasks
& SPA_ASYNC_REMOVE
) {
6130 spa_vdev_state_enter(spa
, SCL_NONE
);
6131 spa_async_remove(spa
, spa
->spa_root_vdev
);
6132 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6133 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6134 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6135 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6136 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6139 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6140 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6141 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6142 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6146 * See if any devices need to be probed.
6148 if (tasks
& SPA_ASYNC_PROBE
) {
6149 spa_vdev_state_enter(spa
, SCL_NONE
);
6150 spa_async_probe(spa
, spa
->spa_root_vdev
);
6151 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6155 * If any devices are done replacing, detach them.
6157 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6158 spa_vdev_resilver_done(spa
);
6161 * Kick off a resilver.
6163 if (tasks
& SPA_ASYNC_RESILVER
)
6164 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6167 * Let the world know that we're done.
6169 mutex_enter(&spa
->spa_async_lock
);
6170 spa
->spa_async_thread
= NULL
;
6171 cv_broadcast(&spa
->spa_async_cv
);
6172 mutex_exit(&spa
->spa_async_lock
);
6177 spa_async_suspend(spa_t
*spa
)
6179 mutex_enter(&spa
->spa_async_lock
);
6180 spa
->spa_async_suspended
++;
6181 while (spa
->spa_async_thread
!= NULL
)
6182 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6183 mutex_exit(&spa
->spa_async_lock
);
6187 spa_async_resume(spa_t
*spa
)
6189 mutex_enter(&spa
->spa_async_lock
);
6190 ASSERT(spa
->spa_async_suspended
!= 0);
6191 spa
->spa_async_suspended
--;
6192 mutex_exit(&spa
->spa_async_lock
);
6196 spa_async_tasks_pending(spa_t
*spa
)
6198 uint_t non_config_tasks
;
6200 boolean_t config_task_suspended
;
6202 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6203 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6204 if (spa
->spa_ccw_fail_time
== 0) {
6205 config_task_suspended
= B_FALSE
;
6207 config_task_suspended
=
6208 (gethrtime() - spa
->spa_ccw_fail_time
) <
6209 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
6212 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6216 spa_async_dispatch(spa_t
*spa
)
6218 mutex_enter(&spa
->spa_async_lock
);
6219 if (spa_async_tasks_pending(spa
) &&
6220 !spa
->spa_async_suspended
&&
6221 spa
->spa_async_thread
== NULL
&&
6223 spa
->spa_async_thread
= thread_create(NULL
, 0,
6224 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6225 mutex_exit(&spa
->spa_async_lock
);
6229 spa_async_request(spa_t
*spa
, int task
)
6231 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6232 mutex_enter(&spa
->spa_async_lock
);
6233 spa
->spa_async_tasks
|= task
;
6234 mutex_exit(&spa
->spa_async_lock
);
6238 * ==========================================================================
6239 * SPA syncing routines
6240 * ==========================================================================
6244 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6247 bpobj_enqueue(bpo
, bp
, tx
);
6252 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6256 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6262 * Note: this simple function is not inlined to make it easier to dtrace the
6263 * amount of time spent syncing frees.
6266 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6268 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6269 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6270 VERIFY(zio_wait(zio
) == 0);
6274 * Note: this simple function is not inlined to make it easier to dtrace the
6275 * amount of time spent syncing deferred frees.
6278 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6280 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6281 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6282 spa_free_sync_cb
, zio
, tx
), ==, 0);
6283 VERIFY0(zio_wait(zio
));
6287 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6289 char *packed
= NULL
;
6294 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6297 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6298 * information. This avoids the dmu_buf_will_dirty() path and
6299 * saves us a pre-read to get data we don't actually care about.
6301 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6302 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6304 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6306 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6308 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6310 vmem_free(packed
, bufsize
);
6312 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6313 dmu_buf_will_dirty(db
, tx
);
6314 *(uint64_t *)db
->db_data
= nvsize
;
6315 dmu_buf_rele(db
, FTAG
);
6319 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6320 const char *config
, const char *entry
)
6330 * Update the MOS nvlist describing the list of available devices.
6331 * spa_validate_aux() will have already made sure this nvlist is
6332 * valid and the vdevs are labeled appropriately.
6334 if (sav
->sav_object
== 0) {
6335 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6336 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6337 sizeof (uint64_t), tx
);
6338 VERIFY(zap_update(spa
->spa_meta_objset
,
6339 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6340 &sav
->sav_object
, tx
) == 0);
6343 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6344 if (sav
->sav_count
== 0) {
6345 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6347 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6348 for (i
= 0; i
< sav
->sav_count
; i
++)
6349 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6350 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6351 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6352 sav
->sav_count
) == 0);
6353 for (i
= 0; i
< sav
->sav_count
; i
++)
6354 nvlist_free(list
[i
]);
6355 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6358 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6359 nvlist_free(nvroot
);
6361 sav
->sav_sync
= B_FALSE
;
6365 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6366 * The all-vdev ZAP must be empty.
6369 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6371 spa_t
*spa
= vd
->vdev_spa
;
6373 if (vd
->vdev_top_zap
!= 0) {
6374 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6375 vd
->vdev_top_zap
, tx
));
6377 if (vd
->vdev_leaf_zap
!= 0) {
6378 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6379 vd
->vdev_leaf_zap
, tx
));
6381 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
6382 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6387 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6392 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6393 * its config may not be dirty but we still need to build per-vdev ZAPs.
6394 * Similarly, if the pool is being assembled (e.g. after a split), we
6395 * need to rebuild the AVZ although the config may not be dirty.
6397 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6398 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6401 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6403 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6404 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6405 spa
->spa_all_vdev_zaps
!= 0);
6407 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6408 /* Make and build the new AVZ */
6409 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6410 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6411 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6413 /* Diff old AVZ with new one */
6417 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6418 spa
->spa_all_vdev_zaps
);
6419 zap_cursor_retrieve(&zc
, &za
) == 0;
6420 zap_cursor_advance(&zc
)) {
6421 uint64_t vdzap
= za
.za_first_integer
;
6422 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6425 * ZAP is listed in old AVZ but not in new one;
6428 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6433 zap_cursor_fini(&zc
);
6435 /* Destroy the old AVZ */
6436 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6437 spa
->spa_all_vdev_zaps
, tx
));
6439 /* Replace the old AVZ in the dir obj with the new one */
6440 VERIFY0(zap_update(spa
->spa_meta_objset
,
6441 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6442 sizeof (new_avz
), 1, &new_avz
, tx
));
6444 spa
->spa_all_vdev_zaps
= new_avz
;
6445 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6449 /* Walk through the AVZ and destroy all listed ZAPs */
6450 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6451 spa
->spa_all_vdev_zaps
);
6452 zap_cursor_retrieve(&zc
, &za
) == 0;
6453 zap_cursor_advance(&zc
)) {
6454 uint64_t zap
= za
.za_first_integer
;
6455 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6458 zap_cursor_fini(&zc
);
6460 /* Destroy and unlink the AVZ itself */
6461 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6462 spa
->spa_all_vdev_zaps
, tx
));
6463 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6464 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6465 spa
->spa_all_vdev_zaps
= 0;
6468 if (spa
->spa_all_vdev_zaps
== 0) {
6469 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6470 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6471 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6473 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6475 /* Create ZAPs for vdevs that don't have them. */
6476 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6478 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6479 dmu_tx_get_txg(tx
), B_FALSE
);
6482 * If we're upgrading the spa version then make sure that
6483 * the config object gets updated with the correct version.
6485 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6486 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6487 spa
->spa_uberblock
.ub_version
);
6489 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6491 nvlist_free(spa
->spa_config_syncing
);
6492 spa
->spa_config_syncing
= config
;
6494 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6498 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6500 uint64_t *versionp
= arg
;
6501 uint64_t version
= *versionp
;
6502 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6505 * Setting the version is special cased when first creating the pool.
6507 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6509 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6510 ASSERT(version
>= spa_version(spa
));
6512 spa
->spa_uberblock
.ub_version
= version
;
6513 vdev_config_dirty(spa
->spa_root_vdev
);
6514 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6518 * Set zpool properties.
6521 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6523 nvlist_t
*nvp
= arg
;
6524 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6525 objset_t
*mos
= spa
->spa_meta_objset
;
6526 nvpair_t
*elem
= NULL
;
6528 mutex_enter(&spa
->spa_props_lock
);
6530 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6532 char *strval
, *fname
;
6534 const char *propname
;
6535 zprop_type_t proptype
;
6538 prop
= zpool_name_to_prop(nvpair_name(elem
));
6539 switch ((int)prop
) {
6542 * We checked this earlier in spa_prop_validate().
6544 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6546 fname
= strchr(nvpair_name(elem
), '@') + 1;
6547 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6549 spa_feature_enable(spa
, fid
, tx
);
6550 spa_history_log_internal(spa
, "set", tx
,
6551 "%s=enabled", nvpair_name(elem
));
6554 case ZPOOL_PROP_VERSION
:
6555 intval
= fnvpair_value_uint64(elem
);
6557 * The version is synced separately before other
6558 * properties and should be correct by now.
6560 ASSERT3U(spa_version(spa
), >=, intval
);
6563 case ZPOOL_PROP_ALTROOT
:
6565 * 'altroot' is a non-persistent property. It should
6566 * have been set temporarily at creation or import time.
6568 ASSERT(spa
->spa_root
!= NULL
);
6571 case ZPOOL_PROP_READONLY
:
6572 case ZPOOL_PROP_CACHEFILE
:
6574 * 'readonly' and 'cachefile' are also non-persisitent
6578 case ZPOOL_PROP_COMMENT
:
6579 strval
= fnvpair_value_string(elem
);
6580 if (spa
->spa_comment
!= NULL
)
6581 spa_strfree(spa
->spa_comment
);
6582 spa
->spa_comment
= spa_strdup(strval
);
6584 * We need to dirty the configuration on all the vdevs
6585 * so that their labels get updated. It's unnecessary
6586 * to do this for pool creation since the vdev's
6587 * configuration has already been dirtied.
6589 if (tx
->tx_txg
!= TXG_INITIAL
)
6590 vdev_config_dirty(spa
->spa_root_vdev
);
6591 spa_history_log_internal(spa
, "set", tx
,
6592 "%s=%s", nvpair_name(elem
), strval
);
6596 * Set pool property values in the poolprops mos object.
6598 if (spa
->spa_pool_props_object
== 0) {
6599 spa
->spa_pool_props_object
=
6600 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6601 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6605 /* normalize the property name */
6606 propname
= zpool_prop_to_name(prop
);
6607 proptype
= zpool_prop_get_type(prop
);
6609 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6610 ASSERT(proptype
== PROP_TYPE_STRING
);
6611 strval
= fnvpair_value_string(elem
);
6612 VERIFY0(zap_update(mos
,
6613 spa
->spa_pool_props_object
, propname
,
6614 1, strlen(strval
) + 1, strval
, tx
));
6615 spa_history_log_internal(spa
, "set", tx
,
6616 "%s=%s", nvpair_name(elem
), strval
);
6617 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6618 intval
= fnvpair_value_uint64(elem
);
6620 if (proptype
== PROP_TYPE_INDEX
) {
6622 VERIFY0(zpool_prop_index_to_string(
6623 prop
, intval
, &unused
));
6625 VERIFY0(zap_update(mos
,
6626 spa
->spa_pool_props_object
, propname
,
6627 8, 1, &intval
, tx
));
6628 spa_history_log_internal(spa
, "set", tx
,
6629 "%s=%lld", nvpair_name(elem
), intval
);
6631 ASSERT(0); /* not allowed */
6635 case ZPOOL_PROP_DELEGATION
:
6636 spa
->spa_delegation
= intval
;
6638 case ZPOOL_PROP_BOOTFS
:
6639 spa
->spa_bootfs
= intval
;
6641 case ZPOOL_PROP_FAILUREMODE
:
6642 spa
->spa_failmode
= intval
;
6644 case ZPOOL_PROP_AUTOEXPAND
:
6645 spa
->spa_autoexpand
= intval
;
6646 if (tx
->tx_txg
!= TXG_INITIAL
)
6647 spa_async_request(spa
,
6648 SPA_ASYNC_AUTOEXPAND
);
6650 case ZPOOL_PROP_MULTIHOST
:
6651 spa
->spa_multihost
= intval
;
6653 case ZPOOL_PROP_DEDUPDITTO
:
6654 spa
->spa_dedup_ditto
= intval
;
6663 mutex_exit(&spa
->spa_props_lock
);
6667 * Perform one-time upgrade on-disk changes. spa_version() does not
6668 * reflect the new version this txg, so there must be no changes this
6669 * txg to anything that the upgrade code depends on after it executes.
6670 * Therefore this must be called after dsl_pool_sync() does the sync
6674 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6676 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6678 ASSERT(spa
->spa_sync_pass
== 1);
6680 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6682 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6683 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6684 dsl_pool_create_origin(dp
, tx
);
6686 /* Keeping the origin open increases spa_minref */
6687 spa
->spa_minref
+= 3;
6690 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6691 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6692 dsl_pool_upgrade_clones(dp
, tx
);
6695 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6696 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6697 dsl_pool_upgrade_dir_clones(dp
, tx
);
6699 /* Keeping the freedir open increases spa_minref */
6700 spa
->spa_minref
+= 3;
6703 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6704 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6705 spa_feature_create_zap_objects(spa
, tx
);
6709 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6710 * when possibility to use lz4 compression for metadata was added
6711 * Old pools that have this feature enabled must be upgraded to have
6712 * this feature active
6714 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6715 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6716 SPA_FEATURE_LZ4_COMPRESS
);
6717 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6718 SPA_FEATURE_LZ4_COMPRESS
);
6720 if (lz4_en
&& !lz4_ac
)
6721 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6725 * If we haven't written the salt, do so now. Note that the
6726 * feature may not be activated yet, but that's fine since
6727 * the presence of this ZAP entry is backwards compatible.
6729 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
6730 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
6731 VERIFY0(zap_add(spa
->spa_meta_objset
,
6732 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
6733 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
6734 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
6737 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6741 * Sync the specified transaction group. New blocks may be dirtied as
6742 * part of the process, so we iterate until it converges.
6745 spa_sync(spa_t
*spa
, uint64_t txg
)
6747 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6748 objset_t
*mos
= spa
->spa_meta_objset
;
6749 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6750 vdev_t
*rvd
= spa
->spa_root_vdev
;
6754 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
6755 zfs_vdev_queue_depth_pct
/ 100;
6757 VERIFY(spa_writeable(spa
));
6760 * Lock out configuration changes.
6762 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6764 spa
->spa_syncing_txg
= txg
;
6765 spa
->spa_sync_pass
= 0;
6767 mutex_enter(&spa
->spa_alloc_lock
);
6768 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
6769 mutex_exit(&spa
->spa_alloc_lock
);
6772 * If there are any pending vdev state changes, convert them
6773 * into config changes that go out with this transaction group.
6775 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6776 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6778 * We need the write lock here because, for aux vdevs,
6779 * calling vdev_config_dirty() modifies sav_config.
6780 * This is ugly and will become unnecessary when we
6781 * eliminate the aux vdev wart by integrating all vdevs
6782 * into the root vdev tree.
6784 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6785 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6786 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6787 vdev_state_clean(vd
);
6788 vdev_config_dirty(vd
);
6790 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6791 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6793 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6795 tx
= dmu_tx_create_assigned(dp
, txg
);
6797 spa
->spa_sync_starttime
= gethrtime();
6798 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6799 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
6800 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6801 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6804 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6805 * set spa_deflate if we have no raid-z vdevs.
6807 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6808 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6811 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6812 vd
= rvd
->vdev_child
[i
];
6813 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6816 if (i
== rvd
->vdev_children
) {
6817 spa
->spa_deflate
= TRUE
;
6818 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6819 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6820 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6825 * Set the top-level vdev's max queue depth. Evaluate each
6826 * top-level's async write queue depth in case it changed.
6827 * The max queue depth will not change in the middle of syncing
6830 uint64_t queue_depth_total
= 0;
6831 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
6832 vdev_t
*tvd
= rvd
->vdev_child
[c
];
6833 metaslab_group_t
*mg
= tvd
->vdev_mg
;
6835 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
6836 !metaslab_group_initialized(mg
))
6840 * It is safe to do a lock-free check here because only async
6841 * allocations look at mg_max_alloc_queue_depth, and async
6842 * allocations all happen from spa_sync().
6844 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
6845 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
6846 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
6848 metaslab_class_t
*mc
= spa_normal_class(spa
);
6849 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
6850 mc
->mc_alloc_max_slots
= queue_depth_total
;
6851 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
6853 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
6854 max_queue_depth
* rvd
->vdev_children
);
6857 * Iterate to convergence.
6860 int pass
= ++spa
->spa_sync_pass
;
6862 spa_sync_config_object(spa
, tx
);
6863 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6864 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6865 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6866 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6867 spa_errlog_sync(spa
, txg
);
6868 dsl_pool_sync(dp
, txg
);
6870 if (pass
< zfs_sync_pass_deferred_free
) {
6871 spa_sync_frees(spa
, free_bpl
, tx
);
6874 * We can not defer frees in pass 1, because
6875 * we sync the deferred frees later in pass 1.
6877 ASSERT3U(pass
, >, 1);
6878 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6879 &spa
->spa_deferred_bpobj
, tx
);
6883 dsl_scan_sync(dp
, tx
);
6885 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6889 spa_sync_upgrades(spa
, tx
);
6891 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
6893 * Note: We need to check if the MOS is dirty
6894 * because we could have marked the MOS dirty
6895 * without updating the uberblock (e.g. if we
6896 * have sync tasks but no dirty user data). We
6897 * need to check the uberblock's rootbp because
6898 * it is updated if we have synced out dirty
6899 * data (though in this case the MOS will most
6900 * likely also be dirty due to second order
6901 * effects, we don't want to rely on that here).
6903 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
6904 !dmu_objset_is_dirty(mos
, txg
)) {
6906 * Nothing changed on the first pass,
6907 * therefore this TXG is a no-op. Avoid
6908 * syncing deferred frees, so that we
6909 * can keep this TXG as a no-op.
6911 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
6913 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6914 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
6917 spa_sync_deferred_frees(spa
, tx
);
6920 } while (dmu_objset_is_dirty(mos
, txg
));
6923 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
6925 * Make sure that the number of ZAPs for all the vdevs matches
6926 * the number of ZAPs in the per-vdev ZAP list. This only gets
6927 * called if the config is dirty; otherwise there may be
6928 * outstanding AVZ operations that weren't completed in
6929 * spa_sync_config_object.
6931 uint64_t all_vdev_zap_entry_count
;
6932 ASSERT0(zap_count(spa
->spa_meta_objset
,
6933 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
6934 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
6935 all_vdev_zap_entry_count
);
6940 * Rewrite the vdev configuration (which includes the uberblock)
6941 * to commit the transaction group.
6943 * If there are no dirty vdevs, we sync the uberblock to a few
6944 * random top-level vdevs that are known to be visible in the
6945 * config cache (see spa_vdev_add() for a complete description).
6946 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6950 * We hold SCL_STATE to prevent vdev open/close/etc.
6951 * while we're attempting to write the vdev labels.
6953 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6955 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6956 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6958 int children
= rvd
->vdev_children
;
6959 int c0
= spa_get_random(children
);
6961 for (int c
= 0; c
< children
; c
++) {
6962 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6963 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6965 svd
[svdcount
++] = vd
;
6966 if (svdcount
== SPA_DVAS_PER_BP
)
6969 error
= vdev_config_sync(svd
, svdcount
, txg
);
6971 error
= vdev_config_sync(rvd
->vdev_child
,
6972 rvd
->vdev_children
, txg
);
6976 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6978 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6982 zio_suspend(spa
, NULL
);
6983 zio_resume_wait(spa
);
6987 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6988 spa
->spa_deadman_tqid
= 0;
6991 * Clear the dirty config list.
6993 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6994 vdev_config_clean(vd
);
6997 * Now that the new config has synced transactionally,
6998 * let it become visible to the config cache.
7000 if (spa
->spa_config_syncing
!= NULL
) {
7001 spa_config_set(spa
, spa
->spa_config_syncing
);
7002 spa
->spa_config_txg
= txg
;
7003 spa
->spa_config_syncing
= NULL
;
7006 dsl_pool_sync_done(dp
, txg
);
7008 mutex_enter(&spa
->spa_alloc_lock
);
7009 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7010 mutex_exit(&spa
->spa_alloc_lock
);
7013 * Update usable space statistics.
7015 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
7016 vdev_sync_done(vd
, txg
);
7018 spa_update_dspace(spa
);
7021 * It had better be the case that we didn't dirty anything
7022 * since vdev_config_sync().
7024 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7025 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7026 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7028 spa
->spa_sync_pass
= 0;
7031 * Update the last synced uberblock here. We want to do this at
7032 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7033 * will be guaranteed that all the processing associated with
7034 * that txg has been completed.
7036 spa
->spa_ubsync
= spa
->spa_uberblock
;
7037 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7039 spa_handle_ignored_writes(spa
);
7042 * If any async tasks have been requested, kick them off.
7044 spa_async_dispatch(spa
);
7048 * Sync all pools. We don't want to hold the namespace lock across these
7049 * operations, so we take a reference on the spa_t and drop the lock during the
7053 spa_sync_allpools(void)
7056 mutex_enter(&spa_namespace_lock
);
7057 while ((spa
= spa_next(spa
)) != NULL
) {
7058 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7059 !spa_writeable(spa
) || spa_suspended(spa
))
7061 spa_open_ref(spa
, FTAG
);
7062 mutex_exit(&spa_namespace_lock
);
7063 txg_wait_synced(spa_get_dsl(spa
), 0);
7064 mutex_enter(&spa_namespace_lock
);
7065 spa_close(spa
, FTAG
);
7067 mutex_exit(&spa_namespace_lock
);
7071 * ==========================================================================
7072 * Miscellaneous routines
7073 * ==========================================================================
7077 * Remove all pools in the system.
7085 * Remove all cached state. All pools should be closed now,
7086 * so every spa in the AVL tree should be unreferenced.
7088 mutex_enter(&spa_namespace_lock
);
7089 while ((spa
= spa_next(NULL
)) != NULL
) {
7091 * Stop async tasks. The async thread may need to detach
7092 * a device that's been replaced, which requires grabbing
7093 * spa_namespace_lock, so we must drop it here.
7095 spa_open_ref(spa
, FTAG
);
7096 mutex_exit(&spa_namespace_lock
);
7097 spa_async_suspend(spa
);
7098 mutex_enter(&spa_namespace_lock
);
7099 spa_close(spa
, FTAG
);
7101 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7103 spa_deactivate(spa
);
7107 mutex_exit(&spa_namespace_lock
);
7111 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7116 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7120 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7121 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7122 if (vd
->vdev_guid
== guid
)
7126 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7127 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7128 if (vd
->vdev_guid
== guid
)
7137 spa_upgrade(spa_t
*spa
, uint64_t version
)
7139 ASSERT(spa_writeable(spa
));
7141 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7144 * This should only be called for a non-faulted pool, and since a
7145 * future version would result in an unopenable pool, this shouldn't be
7148 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7149 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7151 spa
->spa_uberblock
.ub_version
= version
;
7152 vdev_config_dirty(spa
->spa_root_vdev
);
7154 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7156 txg_wait_synced(spa_get_dsl(spa
), 0);
7160 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7164 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7166 for (i
= 0; i
< sav
->sav_count
; i
++)
7167 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7170 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7171 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7172 &spareguid
) == 0 && spareguid
== guid
)
7180 * Check if a pool has an active shared spare device.
7181 * Note: reference count of an active spare is 2, as a spare and as a replace
7184 spa_has_active_shared_spare(spa_t
*spa
)
7188 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7190 for (i
= 0; i
< sav
->sav_count
; i
++) {
7191 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7192 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7201 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7203 sysevent_t
*ev
= NULL
;
7207 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
7209 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
7210 ev
->resource
= resource
;
7217 spa_event_post(sysevent_t
*ev
)
7221 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
7222 kmem_free(ev
, sizeof (*ev
));
7228 * Post a zevent corresponding to the given sysevent. The 'name' must be one
7229 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
7230 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7231 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7232 * or zdb as real changes.
7235 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7237 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
7240 #if defined(_KERNEL) && defined(HAVE_SPL)
7241 /* state manipulation functions */
7242 EXPORT_SYMBOL(spa_open
);
7243 EXPORT_SYMBOL(spa_open_rewind
);
7244 EXPORT_SYMBOL(spa_get_stats
);
7245 EXPORT_SYMBOL(spa_create
);
7246 EXPORT_SYMBOL(spa_import
);
7247 EXPORT_SYMBOL(spa_tryimport
);
7248 EXPORT_SYMBOL(spa_destroy
);
7249 EXPORT_SYMBOL(spa_export
);
7250 EXPORT_SYMBOL(spa_reset
);
7251 EXPORT_SYMBOL(spa_async_request
);
7252 EXPORT_SYMBOL(spa_async_suspend
);
7253 EXPORT_SYMBOL(spa_async_resume
);
7254 EXPORT_SYMBOL(spa_inject_addref
);
7255 EXPORT_SYMBOL(spa_inject_delref
);
7256 EXPORT_SYMBOL(spa_scan_stat_init
);
7257 EXPORT_SYMBOL(spa_scan_get_stats
);
7259 /* device maniion */
7260 EXPORT_SYMBOL(spa_vdev_add
);
7261 EXPORT_SYMBOL(spa_vdev_attach
);
7262 EXPORT_SYMBOL(spa_vdev_detach
);
7263 EXPORT_SYMBOL(spa_vdev_remove
);
7264 EXPORT_SYMBOL(spa_vdev_setpath
);
7265 EXPORT_SYMBOL(spa_vdev_setfru
);
7266 EXPORT_SYMBOL(spa_vdev_split_mirror
);
7268 /* spare statech is global across all pools) */
7269 EXPORT_SYMBOL(spa_spare_add
);
7270 EXPORT_SYMBOL(spa_spare_remove
);
7271 EXPORT_SYMBOL(spa_spare_exists
);
7272 EXPORT_SYMBOL(spa_spare_activate
);
7274 /* L2ARC statech is global across all pools) */
7275 EXPORT_SYMBOL(spa_l2cache_add
);
7276 EXPORT_SYMBOL(spa_l2cache_remove
);
7277 EXPORT_SYMBOL(spa_l2cache_exists
);
7278 EXPORT_SYMBOL(spa_l2cache_activate
);
7279 EXPORT_SYMBOL(spa_l2cache_drop
);
7282 EXPORT_SYMBOL(spa_scan
);
7283 EXPORT_SYMBOL(spa_scan_stop
);
7286 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
7287 EXPORT_SYMBOL(spa_sync_allpools
);
7290 EXPORT_SYMBOL(spa_prop_set
);
7291 EXPORT_SYMBOL(spa_prop_get
);
7292 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
7294 /* asynchronous event notification */
7295 EXPORT_SYMBOL(spa_event_notify
);
7298 #if defined(_KERNEL) && defined(HAVE_SPL)
7299 module_param(spa_load_verify_maxinflight
, int, 0644);
7300 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
7301 "Max concurrent traversal I/Os while verifying pool during import -X");
7303 module_param(spa_load_verify_metadata
, int, 0644);
7304 MODULE_PARM_DESC(spa_load_verify_metadata
,
7305 "Set to traverse metadata on pool import");
7307 module_param(spa_load_verify_data
, int, 0644);
7308 MODULE_PARM_DESC(spa_load_verify_data
,
7309 "Set to traverse data on pool import");
7312 module_param(zio_taskq_batch_pct
, uint
, 0444);
7313 MODULE_PARM_DESC(zio_taskq_batch_pct
,
7314 "Percentage of CPUs to run an IO worker thread");