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 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
761 vdev_t
*rvd
= spa
->spa_root_vdev
;
763 ASSERTV(uint64_t *newguid
= arg
);
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 (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
];
955 if (tqs
->stqs_taskq
== NULL
) {
956 ASSERT3U(tqs
->stqs_count
, ==, 0);
960 for (i
= 0; i
< tqs
->stqs_count
; i
++) {
961 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
962 taskq_destroy(tqs
->stqs_taskq
[i
]);
965 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
966 tqs
->stqs_taskq
= NULL
;
970 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
971 * Note that a type may have multiple discrete taskqs to avoid lock contention
972 * on the taskq itself. In that case we choose which taskq at random by using
973 * the low bits of gethrtime().
976 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
977 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
979 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
982 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
983 ASSERT3U(tqs
->stqs_count
, !=, 0);
985 if (tqs
->stqs_count
== 1) {
986 tq
= tqs
->stqs_taskq
[0];
988 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
991 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
995 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
998 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
999 task_func_t
*func
, void *arg
, uint_t flags
)
1001 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1005 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1006 ASSERT3U(tqs
->stqs_count
, !=, 0);
1008 if (tqs
->stqs_count
== 1) {
1009 tq
= tqs
->stqs_taskq
[0];
1011 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1014 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1016 taskq_wait_id(tq
, id
);
1020 spa_create_zio_taskqs(spa_t
*spa
)
1024 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1025 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1026 spa_taskqs_init(spa
, t
, q
);
1032 * Disabled until spa_thread() can be adapted for Linux.
1034 #undef HAVE_SPA_THREAD
1036 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1038 spa_thread(void *arg
)
1040 callb_cpr_t cprinfo
;
1043 user_t
*pu
= PTOU(curproc
);
1045 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1048 ASSERT(curproc
!= &p0
);
1049 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1050 "zpool-%s", spa
->spa_name
);
1051 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1053 /* bind this thread to the requested psrset */
1054 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1056 mutex_enter(&cpu_lock
);
1057 mutex_enter(&pidlock
);
1058 mutex_enter(&curproc
->p_lock
);
1060 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1061 0, NULL
, NULL
) == 0) {
1062 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1065 "Couldn't bind process for zfs pool \"%s\" to "
1066 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1069 mutex_exit(&curproc
->p_lock
);
1070 mutex_exit(&pidlock
);
1071 mutex_exit(&cpu_lock
);
1075 if (zio_taskq_sysdc
) {
1076 sysdc_thread_enter(curthread
, 100, 0);
1079 spa
->spa_proc
= curproc
;
1080 spa
->spa_did
= curthread
->t_did
;
1082 spa_create_zio_taskqs(spa
);
1084 mutex_enter(&spa
->spa_proc_lock
);
1085 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1087 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1088 cv_broadcast(&spa
->spa_proc_cv
);
1090 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1091 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1092 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1093 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1095 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1096 spa
->spa_proc_state
= SPA_PROC_GONE
;
1097 spa
->spa_proc
= &p0
;
1098 cv_broadcast(&spa
->spa_proc_cv
);
1099 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1101 mutex_enter(&curproc
->p_lock
);
1107 * Activate an uninitialized pool.
1110 spa_activate(spa_t
*spa
, int mode
)
1112 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1114 spa
->spa_state
= POOL_STATE_ACTIVE
;
1115 spa
->spa_mode
= mode
;
1117 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1118 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1120 /* Try to create a covering process */
1121 mutex_enter(&spa
->spa_proc_lock
);
1122 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1123 ASSERT(spa
->spa_proc
== &p0
);
1126 #ifdef HAVE_SPA_THREAD
1127 /* Only create a process if we're going to be around a while. */
1128 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1129 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1131 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1132 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1133 cv_wait(&spa
->spa_proc_cv
,
1134 &spa
->spa_proc_lock
);
1136 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1137 ASSERT(spa
->spa_proc
!= &p0
);
1138 ASSERT(spa
->spa_did
!= 0);
1142 "Couldn't create process for zfs pool \"%s\"\n",
1147 #endif /* HAVE_SPA_THREAD */
1148 mutex_exit(&spa
->spa_proc_lock
);
1150 /* If we didn't create a process, we need to create our taskqs. */
1151 if (spa
->spa_proc
== &p0
) {
1152 spa_create_zio_taskqs(spa
);
1155 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1156 offsetof(vdev_t
, vdev_config_dirty_node
));
1157 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1158 offsetof(objset_t
, os_evicting_node
));
1159 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1160 offsetof(vdev_t
, vdev_state_dirty_node
));
1162 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1163 offsetof(struct vdev
, vdev_txg_node
));
1165 avl_create(&spa
->spa_errlist_scrub
,
1166 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1167 offsetof(spa_error_entry_t
, se_avl
));
1168 avl_create(&spa
->spa_errlist_last
,
1169 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1170 offsetof(spa_error_entry_t
, se_avl
));
1172 spa_keystore_init(&spa
->spa_keystore
);
1175 * This taskq is used to perform zvol-minor-related tasks
1176 * asynchronously. This has several advantages, including easy
1177 * resolution of various deadlocks (zfsonlinux bug #3681).
1179 * The taskq must be single threaded to ensure tasks are always
1180 * processed in the order in which they were dispatched.
1182 * A taskq per pool allows one to keep the pools independent.
1183 * This way if one pool is suspended, it will not impact another.
1185 * The preferred location to dispatch a zvol minor task is a sync
1186 * task. In this context, there is easy access to the spa_t and minimal
1187 * error handling is required because the sync task must succeed.
1189 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1193 * The taskq to upgrade datasets in this pool. Currently used by
1194 * feature SPA_FEATURE_USEROBJ_ACCOUNTING.
1196 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", boot_ncpus
,
1197 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1201 * Opposite of spa_activate().
1204 spa_deactivate(spa_t
*spa
)
1208 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1209 ASSERT(spa
->spa_dsl_pool
== NULL
);
1210 ASSERT(spa
->spa_root_vdev
== NULL
);
1211 ASSERT(spa
->spa_async_zio_root
== NULL
);
1212 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1214 spa_evicting_os_wait(spa
);
1216 if (spa
->spa_zvol_taskq
) {
1217 taskq_destroy(spa
->spa_zvol_taskq
);
1218 spa
->spa_zvol_taskq
= NULL
;
1221 if (spa
->spa_upgrade_taskq
) {
1222 taskq_destroy(spa
->spa_upgrade_taskq
);
1223 spa
->spa_upgrade_taskq
= NULL
;
1226 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1228 list_destroy(&spa
->spa_config_dirty_list
);
1229 list_destroy(&spa
->spa_evicting_os_list
);
1230 list_destroy(&spa
->spa_state_dirty_list
);
1232 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1234 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1235 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1236 spa_taskqs_fini(spa
, t
, q
);
1240 metaslab_class_destroy(spa
->spa_normal_class
);
1241 spa
->spa_normal_class
= NULL
;
1243 metaslab_class_destroy(spa
->spa_log_class
);
1244 spa
->spa_log_class
= NULL
;
1247 * If this was part of an import or the open otherwise failed, we may
1248 * still have errors left in the queues. Empty them just in case.
1250 spa_errlog_drain(spa
);
1251 avl_destroy(&spa
->spa_errlist_scrub
);
1252 avl_destroy(&spa
->spa_errlist_last
);
1254 spa_keystore_fini(&spa
->spa_keystore
);
1256 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1258 mutex_enter(&spa
->spa_proc_lock
);
1259 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1260 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1261 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1262 cv_broadcast(&spa
->spa_proc_cv
);
1263 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1264 ASSERT(spa
->spa_proc
!= &p0
);
1265 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1267 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1268 spa
->spa_proc_state
= SPA_PROC_NONE
;
1270 ASSERT(spa
->spa_proc
== &p0
);
1271 mutex_exit(&spa
->spa_proc_lock
);
1274 * We want to make sure spa_thread() has actually exited the ZFS
1275 * module, so that the module can't be unloaded out from underneath
1278 if (spa
->spa_did
!= 0) {
1279 thread_join(spa
->spa_did
);
1285 * Verify a pool configuration, and construct the vdev tree appropriately. This
1286 * will create all the necessary vdevs in the appropriate layout, with each vdev
1287 * in the CLOSED state. This will prep the pool before open/creation/import.
1288 * All vdev validation is done by the vdev_alloc() routine.
1291 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1292 uint_t id
, int atype
)
1299 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1302 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1305 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1308 if (error
== ENOENT
)
1314 return (SET_ERROR(EINVAL
));
1317 for (c
= 0; c
< children
; c
++) {
1319 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1327 ASSERT(*vdp
!= NULL
);
1333 * Opposite of spa_load().
1336 spa_unload(spa_t
*spa
)
1340 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1345 spa_async_suspend(spa
);
1350 if (spa
->spa_sync_on
) {
1351 txg_sync_stop(spa
->spa_dsl_pool
);
1352 spa
->spa_sync_on
= B_FALSE
;
1356 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1357 * to call it earlier, before we wait for async i/o to complete.
1358 * This ensures that there is no async metaslab prefetching, by
1359 * calling taskq_wait(mg_taskq).
1361 if (spa
->spa_root_vdev
!= NULL
) {
1362 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1363 for (c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++)
1364 vdev_metaslab_fini(spa
->spa_root_vdev
->vdev_child
[c
]);
1365 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1368 if (spa
->spa_mmp
.mmp_thread
)
1369 mmp_thread_stop(spa
);
1372 * Wait for any outstanding async I/O to complete.
1374 if (spa
->spa_async_zio_root
!= NULL
) {
1375 for (i
= 0; i
< max_ncpus
; i
++)
1376 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1377 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1378 spa
->spa_async_zio_root
= NULL
;
1381 bpobj_close(&spa
->spa_deferred_bpobj
);
1383 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1388 if (spa
->spa_root_vdev
)
1389 vdev_free(spa
->spa_root_vdev
);
1390 ASSERT(spa
->spa_root_vdev
== NULL
);
1393 * Close the dsl pool.
1395 if (spa
->spa_dsl_pool
) {
1396 dsl_pool_close(spa
->spa_dsl_pool
);
1397 spa
->spa_dsl_pool
= NULL
;
1398 spa
->spa_meta_objset
= NULL
;
1404 * Drop and purge level 2 cache
1406 spa_l2cache_drop(spa
);
1408 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1409 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1410 if (spa
->spa_spares
.sav_vdevs
) {
1411 kmem_free(spa
->spa_spares
.sav_vdevs
,
1412 spa
->spa_spares
.sav_count
* sizeof (void *));
1413 spa
->spa_spares
.sav_vdevs
= NULL
;
1415 if (spa
->spa_spares
.sav_config
) {
1416 nvlist_free(spa
->spa_spares
.sav_config
);
1417 spa
->spa_spares
.sav_config
= NULL
;
1419 spa
->spa_spares
.sav_count
= 0;
1421 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1422 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1423 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1425 if (spa
->spa_l2cache
.sav_vdevs
) {
1426 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1427 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1428 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1430 if (spa
->spa_l2cache
.sav_config
) {
1431 nvlist_free(spa
->spa_l2cache
.sav_config
);
1432 spa
->spa_l2cache
.sav_config
= NULL
;
1434 spa
->spa_l2cache
.sav_count
= 0;
1436 spa
->spa_async_suspended
= 0;
1438 if (spa
->spa_comment
!= NULL
) {
1439 spa_strfree(spa
->spa_comment
);
1440 spa
->spa_comment
= NULL
;
1443 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1447 * Load (or re-load) the current list of vdevs describing the active spares for
1448 * this pool. When this is called, we have some form of basic information in
1449 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1450 * then re-generate a more complete list including status information.
1453 spa_load_spares(spa_t
*spa
)
1460 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1463 * First, close and free any existing spare vdevs.
1465 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1466 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1468 /* Undo the call to spa_activate() below */
1469 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1470 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1471 spa_spare_remove(tvd
);
1476 if (spa
->spa_spares
.sav_vdevs
)
1477 kmem_free(spa
->spa_spares
.sav_vdevs
,
1478 spa
->spa_spares
.sav_count
* sizeof (void *));
1480 if (spa
->spa_spares
.sav_config
== NULL
)
1483 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1484 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1486 spa
->spa_spares
.sav_count
= (int)nspares
;
1487 spa
->spa_spares
.sav_vdevs
= NULL
;
1493 * Construct the array of vdevs, opening them to get status in the
1494 * process. For each spare, there is potentially two different vdev_t
1495 * structures associated with it: one in the list of spares (used only
1496 * for basic validation purposes) and one in the active vdev
1497 * configuration (if it's spared in). During this phase we open and
1498 * validate each vdev on the spare list. If the vdev also exists in the
1499 * active configuration, then we also mark this vdev as an active spare.
1501 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1503 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1504 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1505 VDEV_ALLOC_SPARE
) == 0);
1508 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1510 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1511 B_FALSE
)) != NULL
) {
1512 if (!tvd
->vdev_isspare
)
1516 * We only mark the spare active if we were successfully
1517 * able to load the vdev. Otherwise, importing a pool
1518 * with a bad active spare would result in strange
1519 * behavior, because multiple pool would think the spare
1520 * is actively in use.
1522 * There is a vulnerability here to an equally bizarre
1523 * circumstance, where a dead active spare is later
1524 * brought back to life (onlined or otherwise). Given
1525 * the rarity of this scenario, and the extra complexity
1526 * it adds, we ignore the possibility.
1528 if (!vdev_is_dead(tvd
))
1529 spa_spare_activate(tvd
);
1533 vd
->vdev_aux
= &spa
->spa_spares
;
1535 if (vdev_open(vd
) != 0)
1538 if (vdev_validate_aux(vd
) == 0)
1543 * Recompute the stashed list of spares, with status information
1546 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1547 DATA_TYPE_NVLIST_ARRAY
) == 0);
1549 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1551 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1552 spares
[i
] = vdev_config_generate(spa
,
1553 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1554 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1555 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1556 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1557 nvlist_free(spares
[i
]);
1558 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1562 * Load (or re-load) the current list of vdevs describing the active l2cache for
1563 * this pool. When this is called, we have some form of basic information in
1564 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1565 * then re-generate a more complete list including status information.
1566 * Devices which are already active have their details maintained, and are
1570 spa_load_l2cache(spa_t
*spa
)
1574 int i
, j
, oldnvdevs
;
1576 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1577 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1579 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1581 oldvdevs
= sav
->sav_vdevs
;
1582 oldnvdevs
= sav
->sav_count
;
1583 sav
->sav_vdevs
= NULL
;
1586 if (sav
->sav_config
== NULL
) {
1592 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1593 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1594 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1597 * Process new nvlist of vdevs.
1599 for (i
= 0; i
< nl2cache
; i
++) {
1600 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1604 for (j
= 0; j
< oldnvdevs
; j
++) {
1606 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1608 * Retain previous vdev for add/remove ops.
1616 if (newvdevs
[i
] == NULL
) {
1620 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1621 VDEV_ALLOC_L2CACHE
) == 0);
1626 * Commit this vdev as an l2cache device,
1627 * even if it fails to open.
1629 spa_l2cache_add(vd
);
1634 spa_l2cache_activate(vd
);
1636 if (vdev_open(vd
) != 0)
1639 (void) vdev_validate_aux(vd
);
1641 if (!vdev_is_dead(vd
))
1642 l2arc_add_vdev(spa
, vd
);
1646 sav
->sav_vdevs
= newvdevs
;
1647 sav
->sav_count
= (int)nl2cache
;
1650 * Recompute the stashed list of l2cache devices, with status
1651 * information this time.
1653 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1654 DATA_TYPE_NVLIST_ARRAY
) == 0);
1656 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1657 for (i
= 0; i
< sav
->sav_count
; i
++)
1658 l2cache
[i
] = vdev_config_generate(spa
,
1659 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1660 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1661 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1665 * Purge vdevs that were dropped
1667 for (i
= 0; i
< oldnvdevs
; i
++) {
1672 ASSERT(vd
->vdev_isl2cache
);
1674 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1675 pool
!= 0ULL && l2arc_vdev_present(vd
))
1676 l2arc_remove_vdev(vd
);
1677 vdev_clear_stats(vd
);
1683 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1685 for (i
= 0; i
< sav
->sav_count
; i
++)
1686 nvlist_free(l2cache
[i
]);
1688 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1692 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1695 char *packed
= NULL
;
1700 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1704 nvsize
= *(uint64_t *)db
->db_data
;
1705 dmu_buf_rele(db
, FTAG
);
1707 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1708 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1711 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1712 vmem_free(packed
, nvsize
);
1718 * Checks to see if the given vdev could not be opened, in which case we post a
1719 * sysevent to notify the autoreplace code that the device has been removed.
1722 spa_check_removed(vdev_t
*vd
)
1726 for (c
= 0; c
< vd
->vdev_children
; c
++)
1727 spa_check_removed(vd
->vdev_child
[c
]);
1729 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1731 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1732 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
1737 spa_config_valid_zaps(vdev_t
*vd
, vdev_t
*mvd
)
1741 ASSERT3U(vd
->vdev_children
, ==, mvd
->vdev_children
);
1743 vd
->vdev_top_zap
= mvd
->vdev_top_zap
;
1744 vd
->vdev_leaf_zap
= mvd
->vdev_leaf_zap
;
1746 for (i
= 0; i
< vd
->vdev_children
; i
++) {
1747 spa_config_valid_zaps(vd
->vdev_child
[i
], mvd
->vdev_child
[i
]);
1752 * Validate the current config against the MOS config
1755 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1757 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1761 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1763 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1764 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1766 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1769 * If we're doing a normal import, then build up any additional
1770 * diagnostic information about missing devices in this config.
1771 * We'll pass this up to the user for further processing.
1773 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1774 nvlist_t
**child
, *nv
;
1777 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
1779 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1781 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1782 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1783 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1785 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1786 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1788 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1793 VERIFY(nvlist_add_nvlist_array(nv
,
1794 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1795 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1796 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1798 for (i
= 0; i
< idx
; i
++)
1799 nvlist_free(child
[i
]);
1802 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1806 * Compare the root vdev tree with the information we have
1807 * from the MOS config (mrvd). Check each top-level vdev
1808 * with the corresponding MOS config top-level (mtvd).
1810 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1811 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1812 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1815 * Resolve any "missing" vdevs in the current configuration.
1816 * If we find that the MOS config has more accurate information
1817 * about the top-level vdev then use that vdev instead.
1819 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1820 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1822 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1826 * Device specific actions.
1828 if (mtvd
->vdev_islog
) {
1829 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1832 * XXX - once we have 'readonly' pool
1833 * support we should be able to handle
1834 * missing data devices by transitioning
1835 * the pool to readonly.
1841 * Swap the missing vdev with the data we were
1842 * able to obtain from the MOS config.
1844 vdev_remove_child(rvd
, tvd
);
1845 vdev_remove_child(mrvd
, mtvd
);
1847 vdev_add_child(rvd
, mtvd
);
1848 vdev_add_child(mrvd
, tvd
);
1850 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1852 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1856 if (mtvd
->vdev_islog
) {
1858 * Load the slog device's state from the MOS
1859 * config since it's possible that the label
1860 * does not contain the most up-to-date
1863 vdev_load_log_state(tvd
, mtvd
);
1868 * Per-vdev ZAP info is stored exclusively in the MOS.
1870 spa_config_valid_zaps(tvd
, mtvd
);
1875 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1878 * Ensure we were able to validate the config.
1880 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1884 * Check for missing log devices
1887 spa_check_logs(spa_t
*spa
)
1889 boolean_t rv
= B_FALSE
;
1890 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1892 switch (spa
->spa_log_state
) {
1895 case SPA_LOG_MISSING
:
1896 /* need to recheck in case slog has been restored */
1897 case SPA_LOG_UNKNOWN
:
1898 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1899 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1901 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1908 spa_passivate_log(spa_t
*spa
)
1910 vdev_t
*rvd
= spa
->spa_root_vdev
;
1911 boolean_t slog_found
= B_FALSE
;
1914 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1916 if (!spa_has_slogs(spa
))
1919 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1920 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1921 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1923 if (tvd
->vdev_islog
) {
1924 metaslab_group_passivate(mg
);
1925 slog_found
= B_TRUE
;
1929 return (slog_found
);
1933 spa_activate_log(spa_t
*spa
)
1935 vdev_t
*rvd
= spa
->spa_root_vdev
;
1938 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1940 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1941 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1942 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1944 if (tvd
->vdev_islog
)
1945 metaslab_group_activate(mg
);
1950 spa_offline_log(spa_t
*spa
)
1954 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1955 NULL
, DS_FIND_CHILDREN
);
1958 * We successfully offlined the log device, sync out the
1959 * current txg so that the "stubby" block can be removed
1962 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1968 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1972 for (i
= 0; i
< sav
->sav_count
; i
++)
1973 spa_check_removed(sav
->sav_vdevs
[i
]);
1977 spa_claim_notify(zio_t
*zio
)
1979 spa_t
*spa
= zio
->io_spa
;
1984 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1985 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1986 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1987 mutex_exit(&spa
->spa_props_lock
);
1990 typedef struct spa_load_error
{
1991 uint64_t sle_meta_count
;
1992 uint64_t sle_data_count
;
1996 spa_load_verify_done(zio_t
*zio
)
1998 blkptr_t
*bp
= zio
->io_bp
;
1999 spa_load_error_t
*sle
= zio
->io_private
;
2000 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2001 int error
= zio
->io_error
;
2002 spa_t
*spa
= zio
->io_spa
;
2004 abd_free(zio
->io_abd
);
2006 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2007 type
!= DMU_OT_INTENT_LOG
)
2008 atomic_inc_64(&sle
->sle_meta_count
);
2010 atomic_inc_64(&sle
->sle_data_count
);
2013 mutex_enter(&spa
->spa_scrub_lock
);
2014 spa
->spa_scrub_inflight
--;
2015 cv_broadcast(&spa
->spa_scrub_io_cv
);
2016 mutex_exit(&spa
->spa_scrub_lock
);
2020 * Maximum number of concurrent scrub i/os to create while verifying
2021 * a pool while importing it.
2023 int spa_load_verify_maxinflight
= 10000;
2024 int spa_load_verify_metadata
= B_TRUE
;
2025 int spa_load_verify_data
= B_TRUE
;
2029 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2030 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2035 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
2038 * Note: normally this routine will not be called if
2039 * spa_load_verify_metadata is not set. However, it may be useful
2040 * to manually set the flag after the traversal has begun.
2042 if (!spa_load_verify_metadata
)
2044 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2048 size
= BP_GET_PSIZE(bp
);
2050 mutex_enter(&spa
->spa_scrub_lock
);
2051 while (spa
->spa_scrub_inflight
>= spa_load_verify_maxinflight
)
2052 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2053 spa
->spa_scrub_inflight
++;
2054 mutex_exit(&spa
->spa_scrub_lock
);
2056 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2057 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2058 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2059 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2065 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2067 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2068 return (SET_ERROR(ENAMETOOLONG
));
2074 spa_load_verify(spa_t
*spa
)
2077 spa_load_error_t sle
= { 0 };
2078 zpool_rewind_policy_t policy
;
2079 boolean_t verify_ok
= B_FALSE
;
2082 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
2084 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
2087 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2088 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2089 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2091 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2095 rio
= zio_root(spa
, NULL
, &sle
,
2096 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2098 if (spa_load_verify_metadata
) {
2099 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2100 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2101 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2104 (void) zio_wait(rio
);
2106 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2107 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2109 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
2110 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
2114 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2115 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2117 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2118 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2119 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2120 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2121 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2122 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2123 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2125 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2129 if (error
!= ENXIO
&& error
!= EIO
)
2130 error
= SET_ERROR(EIO
);
2134 return (verify_ok
? 0 : EIO
);
2138 * Find a value in the pool props object.
2141 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2143 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2144 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2148 * Find a value in the pool directory object.
2151 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2153 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2154 name
, sizeof (uint64_t), 1, val
));
2158 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2160 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2165 * Fix up config after a partly-completed split. This is done with the
2166 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2167 * pool have that entry in their config, but only the splitting one contains
2168 * a list of all the guids of the vdevs that are being split off.
2170 * This function determines what to do with that list: either rejoin
2171 * all the disks to the pool, or complete the splitting process. To attempt
2172 * the rejoin, each disk that is offlined is marked online again, and
2173 * we do a reopen() call. If the vdev label for every disk that was
2174 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2175 * then we call vdev_split() on each disk, and complete the split.
2177 * Otherwise we leave the config alone, with all the vdevs in place in
2178 * the original pool.
2181 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2188 boolean_t attempt_reopen
;
2190 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2193 /* check that the config is complete */
2194 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2195 &glist
, &gcount
) != 0)
2198 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2200 /* attempt to online all the vdevs & validate */
2201 attempt_reopen
= B_TRUE
;
2202 for (i
= 0; i
< gcount
; i
++) {
2203 if (glist
[i
] == 0) /* vdev is hole */
2206 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2207 if (vd
[i
] == NULL
) {
2209 * Don't bother attempting to reopen the disks;
2210 * just do the split.
2212 attempt_reopen
= B_FALSE
;
2214 /* attempt to re-online it */
2215 vd
[i
]->vdev_offline
= B_FALSE
;
2219 if (attempt_reopen
) {
2220 vdev_reopen(spa
->spa_root_vdev
);
2222 /* check each device to see what state it's in */
2223 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2224 if (vd
[i
] != NULL
&&
2225 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2232 * If every disk has been moved to the new pool, or if we never
2233 * even attempted to look at them, then we split them off for
2236 if (!attempt_reopen
|| gcount
== extracted
) {
2237 for (i
= 0; i
< gcount
; i
++)
2240 vdev_reopen(spa
->spa_root_vdev
);
2243 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2247 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2248 boolean_t mosconfig
)
2250 nvlist_t
*config
= spa
->spa_config
;
2251 char *ereport
= FM_EREPORT_ZFS_POOL
;
2257 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2258 return (SET_ERROR(EINVAL
));
2260 ASSERT(spa
->spa_comment
== NULL
);
2261 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2262 spa
->spa_comment
= spa_strdup(comment
);
2265 * Versioning wasn't explicitly added to the label until later, so if
2266 * it's not present treat it as the initial version.
2268 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2269 &spa
->spa_ubsync
.ub_version
) != 0)
2270 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2272 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2273 &spa
->spa_config_txg
);
2275 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2276 spa_guid_exists(pool_guid
, 0)) {
2277 error
= SET_ERROR(EEXIST
);
2279 spa
->spa_config_guid
= pool_guid
;
2281 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2283 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2287 nvlist_free(spa
->spa_load_info
);
2288 spa
->spa_load_info
= fnvlist_alloc();
2290 gethrestime(&spa
->spa_loaded_ts
);
2291 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2292 mosconfig
, &ereport
);
2296 * Don't count references from objsets that are already closed
2297 * and are making their way through the eviction process.
2299 spa_evicting_os_wait(spa
);
2300 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2302 if (error
!= EEXIST
) {
2303 spa
->spa_loaded_ts
.tv_sec
= 0;
2304 spa
->spa_loaded_ts
.tv_nsec
= 0;
2306 if (error
!= EBADF
) {
2307 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, NULL
, 0, 0);
2310 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2318 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2319 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2320 * spa's per-vdev ZAP list.
2323 vdev_count_verify_zaps(vdev_t
*vd
)
2325 spa_t
*spa
= vd
->vdev_spa
;
2329 if (vd
->vdev_top_zap
!= 0) {
2331 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2332 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2334 if (vd
->vdev_leaf_zap
!= 0) {
2336 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2337 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2340 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2341 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2349 * Determine whether the activity check is required.
2352 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2355 uint64_t hostid
= 0;
2356 uint64_t tryconfig_txg
= 0;
2357 uint64_t tryconfig_timestamp
= 0;
2360 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2361 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2362 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2364 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2365 &tryconfig_timestamp
);
2368 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2369 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_HOSTID
, &hostid
);
2372 * Disable the MMP activity check - This is used by zdb which
2373 * is intended to be used on potentially active pools.
2375 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
2379 * Skip the activity check when the MMP feature is disabled.
2381 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
2384 * If the tryconfig_* values are nonzero, they are the results of an
2385 * earlier tryimport. If they match the uberblock we just found, then
2386 * the pool has not changed and we return false so we do not test a
2389 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
2390 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
)
2394 * Allow the activity check to be skipped when importing the pool
2395 * on the same host which last imported it.
2397 if (hostid
== spa_get_hostid())
2401 * Skip the activity test when the pool was cleanly exported.
2403 if (state
!= POOL_STATE_ACTIVE
)
2410 * Perform the import activity check. If the user canceled the import or
2411 * we detected activity then fail.
2414 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2416 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2417 uint64_t txg
= ub
->ub_txg
;
2418 uint64_t timestamp
= ub
->ub_timestamp
;
2419 uint64_t import_delay
= NANOSEC
;
2420 hrtime_t import_expire
;
2421 nvlist_t
*mmp_label
= NULL
;
2422 vdev_t
*rvd
= spa
->spa_root_vdev
;
2427 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
2428 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2432 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2433 * during the earlier tryimport. If the txg recorded there is 0 then
2434 * the pool is known to be active on another host.
2436 * Otherwise, the pool might be in use on another node. Check for
2437 * changes in the uberblocks on disk if necessary.
2439 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2440 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
2441 ZPOOL_CONFIG_LOAD_INFO
);
2443 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
2444 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
2445 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2446 error
= SET_ERROR(EREMOTEIO
);
2452 * Preferentially use the zfs_multihost_interval from the node which
2453 * last imported the pool. This value is stored in an MMP uberblock as.
2455 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2457 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
)
2458 import_delay
= MAX(import_delay
, import_intervals
*
2459 ub
->ub_mmp_delay
* MAX(vdev_count_leaves(spa
), 1));
2461 /* Apply a floor using the local default values. */
2462 import_delay
= MAX(import_delay
, import_intervals
*
2463 MSEC2NSEC(MAX(zfs_multihost_interval
, MMP_MIN_INTERVAL
)));
2465 /* Add a small random factor in case of simultaneous imports (0-25%) */
2466 import_expire
= gethrtime() + import_delay
+
2467 (import_delay
* spa_get_random(250) / 1000);
2469 while (gethrtime() < import_expire
) {
2470 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2472 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
) {
2473 error
= SET_ERROR(EREMOTEIO
);
2478 nvlist_free(mmp_label
);
2482 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
2484 error
= SET_ERROR(EINTR
);
2492 mutex_destroy(&mtx
);
2496 * If the pool is determined to be active store the status in the
2497 * spa->spa_load_info nvlist. If the remote hostname or hostid are
2498 * available from configuration read from disk store them as well.
2499 * This allows 'zpool import' to generate a more useful message.
2501 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
2502 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2503 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
2505 if (error
== EREMOTEIO
) {
2506 char *hostname
= "<unknown>";
2507 uint64_t hostid
= 0;
2510 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
2511 hostname
= fnvlist_lookup_string(mmp_label
,
2512 ZPOOL_CONFIG_HOSTNAME
);
2513 fnvlist_add_string(spa
->spa_load_info
,
2514 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
2517 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
2518 hostid
= fnvlist_lookup_uint64(mmp_label
,
2519 ZPOOL_CONFIG_HOSTID
);
2520 fnvlist_add_uint64(spa
->spa_load_info
,
2521 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
2525 fnvlist_add_uint64(spa
->spa_load_info
,
2526 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
2527 fnvlist_add_uint64(spa
->spa_load_info
,
2528 ZPOOL_CONFIG_MMP_TXG
, 0);
2530 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
2534 nvlist_free(mmp_label
);
2540 * Load an existing storage pool, using the pool's builtin spa_config as a
2541 * source of configuration information.
2543 __attribute__((always_inline
))
2545 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2546 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2550 nvlist_t
*nvroot
= NULL
;
2553 uberblock_t
*ub
= &spa
->spa_uberblock
;
2554 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2555 int orig_mode
= spa
->spa_mode
;
2558 boolean_t missing_feat_write
= B_FALSE
;
2559 boolean_t activity_check
= B_FALSE
;
2560 nvlist_t
*mos_config
;
2563 * If this is an untrusted config, access the pool in read-only mode.
2564 * This prevents things like resilvering recently removed devices.
2567 spa
->spa_mode
= FREAD
;
2569 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2571 spa
->spa_load_state
= state
;
2573 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2574 return (SET_ERROR(EINVAL
));
2576 parse
= (type
== SPA_IMPORT_EXISTING
?
2577 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2580 * Create "The Godfather" zio to hold all async IOs
2582 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2584 for (i
= 0; i
< max_ncpus
; i
++) {
2585 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2586 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2587 ZIO_FLAG_GODFATHER
);
2591 * Parse the configuration into a vdev tree. We explicitly set the
2592 * value that will be returned by spa_version() since parsing the
2593 * configuration requires knowing the version number.
2595 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2596 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2597 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2602 ASSERT(spa
->spa_root_vdev
== rvd
);
2603 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2604 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2606 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2607 ASSERT(spa_guid(spa
) == pool_guid
);
2611 * Try to open all vdevs, loading each label in the process.
2613 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2614 error
= vdev_open(rvd
);
2615 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2620 * We need to validate the vdev labels against the configuration that
2621 * we have in hand, which is dependent on the setting of mosconfig. If
2622 * mosconfig is true then we're validating the vdev labels based on
2623 * that config. Otherwise, we're validating against the cached config
2624 * (zpool.cache) that was read when we loaded the zfs module, and then
2625 * later we will recursively call spa_load() and validate against
2628 * If we're assembling a new pool that's been split off from an
2629 * existing pool, the labels haven't yet been updated so we skip
2630 * validation for now.
2632 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2633 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2634 error
= vdev_validate(rvd
, mosconfig
);
2635 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2640 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2641 return (SET_ERROR(ENXIO
));
2645 * Find the best uberblock.
2647 vdev_uberblock_load(rvd
, ub
, &label
);
2650 * If we weren't able to find a single valid uberblock, return failure.
2652 if (ub
->ub_txg
== 0) {
2654 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2658 * For pools which have the multihost property on determine if the
2659 * pool is truly inactive and can be safely imported. Prevent
2660 * hosts which don't have a hostid set from importing the pool.
2662 activity_check
= spa_activity_check_required(spa
, ub
, config
);
2663 if (activity_check
) {
2664 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
2665 spa_get_hostid() == 0) {
2667 fnvlist_add_uint64(spa
->spa_load_info
,
2668 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
2669 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
2672 error
= spa_activity_check(spa
, ub
, config
);
2678 fnvlist_add_uint64(spa
->spa_load_info
,
2679 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
2680 fnvlist_add_uint64(spa
->spa_load_info
,
2681 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
2685 * If the pool has an unsupported version we can't open it.
2687 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2689 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2692 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2696 * If we weren't able to find what's necessary for reading the
2697 * MOS in the label, return failure.
2699 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2700 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2702 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2707 * Update our in-core representation with the definitive values
2710 nvlist_free(spa
->spa_label_features
);
2711 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2717 * Look through entries in the label nvlist's features_for_read. If
2718 * there is a feature listed there which we don't understand then we
2719 * cannot open a pool.
2721 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2722 nvlist_t
*unsup_feat
;
2725 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2728 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2730 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2731 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2732 VERIFY(nvlist_add_string(unsup_feat
,
2733 nvpair_name(nvp
), "") == 0);
2737 if (!nvlist_empty(unsup_feat
)) {
2738 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2739 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2740 nvlist_free(unsup_feat
);
2741 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2745 nvlist_free(unsup_feat
);
2749 * If the vdev guid sum doesn't match the uberblock, we have an
2750 * incomplete configuration. We first check to see if the pool
2751 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2752 * If it is, defer the vdev_guid_sum check till later so we
2753 * can handle missing vdevs.
2755 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2756 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2757 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2758 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2760 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2761 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2762 spa_try_repair(spa
, config
);
2763 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2764 nvlist_free(spa
->spa_config_splitting
);
2765 spa
->spa_config_splitting
= NULL
;
2769 * Initialize internal SPA structures.
2771 spa
->spa_state
= POOL_STATE_ACTIVE
;
2772 spa
->spa_ubsync
= spa
->spa_uberblock
;
2773 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2774 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2775 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2776 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2777 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2778 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2780 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2782 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2783 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2785 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2786 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2788 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2789 boolean_t missing_feat_read
= B_FALSE
;
2790 nvlist_t
*unsup_feat
, *enabled_feat
;
2793 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2794 &spa
->spa_feat_for_read_obj
) != 0) {
2795 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2798 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2799 &spa
->spa_feat_for_write_obj
) != 0) {
2800 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2803 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2804 &spa
->spa_feat_desc_obj
) != 0) {
2805 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2808 enabled_feat
= fnvlist_alloc();
2809 unsup_feat
= fnvlist_alloc();
2811 if (!spa_features_check(spa
, B_FALSE
,
2812 unsup_feat
, enabled_feat
))
2813 missing_feat_read
= B_TRUE
;
2815 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2816 if (!spa_features_check(spa
, B_TRUE
,
2817 unsup_feat
, enabled_feat
)) {
2818 missing_feat_write
= B_TRUE
;
2822 fnvlist_add_nvlist(spa
->spa_load_info
,
2823 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2825 if (!nvlist_empty(unsup_feat
)) {
2826 fnvlist_add_nvlist(spa
->spa_load_info
,
2827 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2830 fnvlist_free(enabled_feat
);
2831 fnvlist_free(unsup_feat
);
2833 if (!missing_feat_read
) {
2834 fnvlist_add_boolean(spa
->spa_load_info
,
2835 ZPOOL_CONFIG_CAN_RDONLY
);
2839 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2840 * twofold: to determine whether the pool is available for
2841 * import in read-write mode and (if it is not) whether the
2842 * pool is available for import in read-only mode. If the pool
2843 * is available for import in read-write mode, it is displayed
2844 * as available in userland; if it is not available for import
2845 * in read-only mode, it is displayed as unavailable in
2846 * userland. If the pool is available for import in read-only
2847 * mode but not read-write mode, it is displayed as unavailable
2848 * in userland with a special note that the pool is actually
2849 * available for open in read-only mode.
2851 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2852 * missing a feature for write, we must first determine whether
2853 * the pool can be opened read-only before returning to
2854 * userland in order to know whether to display the
2855 * abovementioned note.
2857 if (missing_feat_read
|| (missing_feat_write
&&
2858 spa_writeable(spa
))) {
2859 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2864 * Load refcounts for ZFS features from disk into an in-memory
2865 * cache during SPA initialization.
2867 for (i
= 0; i
< SPA_FEATURES
; i
++) {
2870 error
= feature_get_refcount_from_disk(spa
,
2871 &spa_feature_table
[i
], &refcount
);
2873 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2874 } else if (error
== ENOTSUP
) {
2875 spa
->spa_feat_refcount_cache
[i
] =
2876 SPA_FEATURE_DISABLED
;
2878 return (spa_vdev_err(rvd
,
2879 VDEV_AUX_CORRUPT_DATA
, EIO
));
2884 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2885 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2886 &spa
->spa_feat_enabled_txg_obj
) != 0)
2887 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2890 spa
->spa_is_initializing
= B_TRUE
;
2891 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2892 spa
->spa_is_initializing
= B_FALSE
;
2894 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2898 nvlist_t
*policy
= NULL
, *nvconfig
;
2900 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2901 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2903 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2904 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2906 unsigned long myhostid
= 0;
2908 VERIFY(nvlist_lookup_string(nvconfig
,
2909 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2911 myhostid
= spa_get_hostid();
2912 if (hostid
&& myhostid
&& hostid
!= myhostid
) {
2913 nvlist_free(nvconfig
);
2914 return (SET_ERROR(EBADF
));
2917 if (nvlist_lookup_nvlist(spa
->spa_config
,
2918 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2919 VERIFY(nvlist_add_nvlist(nvconfig
,
2920 ZPOOL_REWIND_POLICY
, policy
) == 0);
2922 spa_config_set(spa
, nvconfig
);
2924 spa_deactivate(spa
);
2925 spa_activate(spa
, orig_mode
);
2927 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2930 /* Grab the checksum salt from the MOS. */
2931 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2932 DMU_POOL_CHECKSUM_SALT
, 1,
2933 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
2934 spa
->spa_cksum_salt
.zcs_bytes
);
2935 if (error
== ENOENT
) {
2936 /* Generate a new salt for subsequent use */
2937 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
2938 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
2939 } else if (error
!= 0) {
2940 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2943 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2944 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2945 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2947 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2950 * Load the bit that tells us to use the new accounting function
2951 * (raid-z deflation). If we have an older pool, this will not
2954 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2955 if (error
!= 0 && error
!= ENOENT
)
2956 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2958 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2959 &spa
->spa_creation_version
);
2960 if (error
!= 0 && error
!= ENOENT
)
2961 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2964 * Load the persistent error log. If we have an older pool, this will
2967 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2968 if (error
!= 0 && error
!= ENOENT
)
2969 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2971 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2972 &spa
->spa_errlog_scrub
);
2973 if (error
!= 0 && error
!= ENOENT
)
2974 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2977 * Load the history object. If we have an older pool, this
2978 * will not be present.
2980 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2981 if (error
!= 0 && error
!= ENOENT
)
2982 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2985 * Load the per-vdev ZAP map. If we have an older pool, this will not
2986 * be present; in this case, defer its creation to a later time to
2987 * avoid dirtying the MOS this early / out of sync context. See
2988 * spa_sync_config_object.
2991 /* The sentinel is only available in the MOS config. */
2992 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
2993 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2995 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
2996 &spa
->spa_all_vdev_zaps
);
2998 if (error
== ENOENT
) {
2999 VERIFY(!nvlist_exists(mos_config
,
3000 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
3001 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
3002 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3003 } else if (error
!= 0) {
3004 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3005 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
3007 * An older version of ZFS overwrote the sentinel value, so
3008 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3009 * destruction to later; see spa_sync_config_object.
3011 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
3013 * We're assuming that no vdevs have had their ZAPs created
3014 * before this. Better be sure of it.
3016 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3018 nvlist_free(mos_config
);
3021 * If we're assembling the pool from the split-off vdevs of
3022 * an existing pool, we don't want to attach the spares & cache
3027 * Load any hot spares for this pool.
3029 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
3030 if (error
!= 0 && error
!= ENOENT
)
3031 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3032 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3033 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3034 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3035 &spa
->spa_spares
.sav_config
) != 0)
3036 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3038 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3039 spa_load_spares(spa
);
3040 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3041 } else if (error
== 0) {
3042 spa
->spa_spares
.sav_sync
= B_TRUE
;
3046 * Load any level 2 ARC devices for this pool.
3048 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3049 &spa
->spa_l2cache
.sav_object
);
3050 if (error
!= 0 && error
!= ENOENT
)
3051 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3052 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3053 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3054 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3055 &spa
->spa_l2cache
.sav_config
) != 0)
3056 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3058 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3059 spa_load_l2cache(spa
);
3060 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3061 } else if (error
== 0) {
3062 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3065 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3067 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
3068 if (error
&& error
!= ENOENT
)
3069 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3072 uint64_t autoreplace
= 0;
3074 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3075 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3076 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3077 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3078 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3079 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
3080 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3081 &spa
->spa_dedup_ditto
);
3083 spa
->spa_autoreplace
= (autoreplace
!= 0);
3087 * If the 'multihost' property is set, then never allow a pool to
3088 * be imported when the system hostid is zero. The exception to
3089 * this rule is zdb which is always allowed to access pools.
3091 if (spa_multihost(spa
) && spa_get_hostid() == 0 &&
3092 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
3093 fnvlist_add_uint64(spa
->spa_load_info
,
3094 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3095 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3099 * If the 'autoreplace' property is set, then post a resource notifying
3100 * the ZFS DE that it should not issue any faults for unopenable
3101 * devices. We also iterate over the vdevs, and post a sysevent for any
3102 * unopenable vdevs so that the normal autoreplace handler can take
3105 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
3106 spa_check_removed(spa
->spa_root_vdev
);
3108 * For the import case, this is done in spa_import(), because
3109 * at this point we're using the spare definitions from
3110 * the MOS config, not necessarily from the userland config.
3112 if (state
!= SPA_LOAD_IMPORT
) {
3113 spa_aux_check_removed(&spa
->spa_spares
);
3114 spa_aux_check_removed(&spa
->spa_l2cache
);
3119 * Load the vdev state for all toplevel vdevs.
3124 * Propagate the leaf DTLs we just loaded all the way up the tree.
3126 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3127 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3128 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3131 * Load the DDTs (dedup tables).
3133 error
= ddt_load(spa
);
3135 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3137 spa_update_dspace(spa
);
3140 * Validate the config, using the MOS config to fill in any
3141 * information which might be missing. If we fail to validate
3142 * the config then declare the pool unfit for use. If we're
3143 * assembling a pool from a split, the log is not transferred
3146 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3149 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
3150 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3152 if (!spa_config_valid(spa
, nvconfig
)) {
3153 nvlist_free(nvconfig
);
3154 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3157 nvlist_free(nvconfig
);
3160 * Now that we've validated the config, check the state of the
3161 * root vdev. If it can't be opened, it indicates one or
3162 * more toplevel vdevs are faulted.
3164 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
3165 return (SET_ERROR(ENXIO
));
3167 if (spa_writeable(spa
) && spa_check_logs(spa
)) {
3168 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3169 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
3173 if (missing_feat_write
) {
3174 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
3177 * At this point, we know that we can open the pool in
3178 * read-only mode but not read-write mode. We now have enough
3179 * information and can return to userland.
3181 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
3185 * We've successfully opened the pool, verify that we're ready
3186 * to start pushing transactions.
3188 if (state
!= SPA_LOAD_TRYIMPORT
) {
3189 if ((error
= spa_load_verify(spa
)))
3190 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3194 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
3195 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3197 int need_update
= B_FALSE
;
3198 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3201 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
3204 * Claim log blocks that haven't been committed yet.
3205 * This must all happen in a single txg.
3206 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3207 * invoked from zil_claim_log_block()'s i/o done callback.
3208 * Price of rollback is that we abandon the log.
3210 spa
->spa_claiming
= B_TRUE
;
3212 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3213 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3214 zil_claim
, tx
, DS_FIND_CHILDREN
);
3217 spa
->spa_claiming
= B_FALSE
;
3219 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3220 spa
->spa_sync_on
= B_TRUE
;
3221 txg_sync_start(spa
->spa_dsl_pool
);
3222 mmp_thread_start(spa
);
3225 * Wait for all claims to sync. We sync up to the highest
3226 * claimed log block birth time so that claimed log blocks
3227 * don't appear to be from the future. spa_claim_max_txg
3228 * will have been set for us by either zil_check_log_chain()
3229 * (invoked from spa_check_logs()) or zil_claim() above.
3231 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3234 * If the config cache is stale, or we have uninitialized
3235 * metaslabs (see spa_vdev_add()), then update the config.
3237 * If this is a verbatim import, trust the current
3238 * in-core spa_config and update the disk labels.
3240 if (config_cache_txg
!= spa
->spa_config_txg
||
3241 state
== SPA_LOAD_IMPORT
||
3242 state
== SPA_LOAD_RECOVER
||
3243 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3244 need_update
= B_TRUE
;
3246 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3247 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3248 need_update
= B_TRUE
;
3251 * Update the config cache asychronously in case we're the
3252 * root pool, in which case the config cache isn't writable yet.
3255 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3258 * Check all DTLs to see if anything needs resilvering.
3260 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3261 vdev_resilver_needed(rvd
, NULL
, NULL
))
3262 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3265 * Log the fact that we booted up (so that we can detect if
3266 * we rebooted in the middle of an operation).
3268 spa_history_log_version(spa
, "open", NULL
);
3271 * Delete any inconsistent datasets.
3273 (void) dmu_objset_find(spa_name(spa
),
3274 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3277 * Clean up any stale temporary dataset userrefs.
3279 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3286 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
3288 int mode
= spa
->spa_mode
;
3291 spa_deactivate(spa
);
3293 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3295 spa_activate(spa
, mode
);
3296 spa_async_suspend(spa
);
3298 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
3302 * If spa_load() fails this function will try loading prior txg's. If
3303 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3304 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3305 * function will not rewind the pool and will return the same error as
3309 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
3310 uint64_t max_request
, int rewind_flags
)
3312 nvlist_t
*loadinfo
= NULL
;
3313 nvlist_t
*config
= NULL
;
3314 int load_error
, rewind_error
;
3315 uint64_t safe_rewind_txg
;
3318 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3319 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3320 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3322 spa
->spa_load_max_txg
= max_request
;
3323 if (max_request
!= UINT64_MAX
)
3324 spa
->spa_extreme_rewind
= B_TRUE
;
3327 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
3329 if (load_error
== 0)
3332 if (spa
->spa_root_vdev
!= NULL
)
3333 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3335 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3336 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3338 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3339 nvlist_free(config
);
3340 return (load_error
);
3343 if (state
== SPA_LOAD_RECOVER
) {
3344 /* Price of rolling back is discarding txgs, including log */
3345 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3348 * If we aren't rolling back save the load info from our first
3349 * import attempt so that we can restore it after attempting
3352 loadinfo
= spa
->spa_load_info
;
3353 spa
->spa_load_info
= fnvlist_alloc();
3356 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3357 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
3358 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
3359 TXG_INITIAL
: safe_rewind_txg
;
3362 * Continue as long as we're finding errors, we're still within
3363 * the acceptable rewind range, and we're still finding uberblocks
3365 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
3366 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
3367 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
3368 spa
->spa_extreme_rewind
= B_TRUE
;
3369 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
3372 spa
->spa_extreme_rewind
= B_FALSE
;
3373 spa
->spa_load_max_txg
= UINT64_MAX
;
3375 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
3376 spa_config_set(spa
, config
);
3378 nvlist_free(config
);
3380 if (state
== SPA_LOAD_RECOVER
) {
3381 ASSERT3P(loadinfo
, ==, NULL
);
3382 return (rewind_error
);
3384 /* Store the rewind info as part of the initial load info */
3385 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
3386 spa
->spa_load_info
);
3388 /* Restore the initial load info */
3389 fnvlist_free(spa
->spa_load_info
);
3390 spa
->spa_load_info
= loadinfo
;
3392 return (load_error
);
3399 * The import case is identical to an open except that the configuration is sent
3400 * down from userland, instead of grabbed from the configuration cache. For the
3401 * case of an open, the pool configuration will exist in the
3402 * POOL_STATE_UNINITIALIZED state.
3404 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3405 * the same time open the pool, without having to keep around the spa_t in some
3409 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
3413 spa_load_state_t state
= SPA_LOAD_OPEN
;
3415 int locked
= B_FALSE
;
3416 int firstopen
= B_FALSE
;
3421 * As disgusting as this is, we need to support recursive calls to this
3422 * function because dsl_dir_open() is called during spa_load(), and ends
3423 * up calling spa_open() again. The real fix is to figure out how to
3424 * avoid dsl_dir_open() calling this in the first place.
3426 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
3427 mutex_enter(&spa_namespace_lock
);
3431 if ((spa
= spa_lookup(pool
)) == NULL
) {
3433 mutex_exit(&spa_namespace_lock
);
3434 return (SET_ERROR(ENOENT
));
3437 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3438 zpool_rewind_policy_t policy
;
3442 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3444 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3445 state
= SPA_LOAD_RECOVER
;
3447 spa_activate(spa
, spa_mode_global
);
3449 if (state
!= SPA_LOAD_RECOVER
)
3450 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3452 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3453 policy
.zrp_request
);
3455 if (error
== EBADF
) {
3457 * If vdev_validate() returns failure (indicated by
3458 * EBADF), it indicates that one of the vdevs indicates
3459 * that the pool has been exported or destroyed. If
3460 * this is the case, the config cache is out of sync and
3461 * we should remove the pool from the namespace.
3464 spa_deactivate(spa
);
3465 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3468 mutex_exit(&spa_namespace_lock
);
3469 return (SET_ERROR(ENOENT
));
3474 * We can't open the pool, but we still have useful
3475 * information: the state of each vdev after the
3476 * attempted vdev_open(). Return this to the user.
3478 if (config
!= NULL
&& spa
->spa_config
) {
3479 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3481 VERIFY(nvlist_add_nvlist(*config
,
3482 ZPOOL_CONFIG_LOAD_INFO
,
3483 spa
->spa_load_info
) == 0);
3486 spa_deactivate(spa
);
3487 spa
->spa_last_open_failed
= error
;
3489 mutex_exit(&spa_namespace_lock
);
3495 spa_open_ref(spa
, tag
);
3498 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3501 * If we've recovered the pool, pass back any information we
3502 * gathered while doing the load.
3504 if (state
== SPA_LOAD_RECOVER
) {
3505 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3506 spa
->spa_load_info
) == 0);
3510 spa
->spa_last_open_failed
= 0;
3511 spa
->spa_last_ubsync_txg
= 0;
3512 spa
->spa_load_txg
= 0;
3513 mutex_exit(&spa_namespace_lock
);
3517 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
3525 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3528 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3532 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3534 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3538 * Lookup the given spa_t, incrementing the inject count in the process,
3539 * preventing it from being exported or destroyed.
3542 spa_inject_addref(char *name
)
3546 mutex_enter(&spa_namespace_lock
);
3547 if ((spa
= spa_lookup(name
)) == NULL
) {
3548 mutex_exit(&spa_namespace_lock
);
3551 spa
->spa_inject_ref
++;
3552 mutex_exit(&spa_namespace_lock
);
3558 spa_inject_delref(spa_t
*spa
)
3560 mutex_enter(&spa_namespace_lock
);
3561 spa
->spa_inject_ref
--;
3562 mutex_exit(&spa_namespace_lock
);
3566 * Add spares device information to the nvlist.
3569 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3579 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3581 if (spa
->spa_spares
.sav_count
== 0)
3584 VERIFY(nvlist_lookup_nvlist(config
,
3585 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3586 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3587 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3589 VERIFY(nvlist_add_nvlist_array(nvroot
,
3590 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3591 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3592 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3595 * Go through and find any spares which have since been
3596 * repurposed as an active spare. If this is the case, update
3597 * their status appropriately.
3599 for (i
= 0; i
< nspares
; i
++) {
3600 VERIFY(nvlist_lookup_uint64(spares
[i
],
3601 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3602 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3604 VERIFY(nvlist_lookup_uint64_array(
3605 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3606 (uint64_t **)&vs
, &vsc
) == 0);
3607 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3608 vs
->vs_aux
= VDEV_AUX_SPARED
;
3615 * Add l2cache device information to the nvlist, including vdev stats.
3618 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3621 uint_t i
, j
, nl2cache
;
3628 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3630 if (spa
->spa_l2cache
.sav_count
== 0)
3633 VERIFY(nvlist_lookup_nvlist(config
,
3634 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3635 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3636 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3637 if (nl2cache
!= 0) {
3638 VERIFY(nvlist_add_nvlist_array(nvroot
,
3639 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3640 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3641 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3644 * Update level 2 cache device stats.
3647 for (i
= 0; i
< nl2cache
; i
++) {
3648 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3649 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3652 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3654 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3655 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3661 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3662 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3664 vdev_get_stats(vd
, vs
);
3665 vdev_config_generate_stats(vd
, l2cache
[i
]);
3672 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3677 if (spa
->spa_feat_for_read_obj
!= 0) {
3678 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3679 spa
->spa_feat_for_read_obj
);
3680 zap_cursor_retrieve(&zc
, &za
) == 0;
3681 zap_cursor_advance(&zc
)) {
3682 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3683 za
.za_num_integers
== 1);
3684 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3685 za
.za_first_integer
));
3687 zap_cursor_fini(&zc
);
3690 if (spa
->spa_feat_for_write_obj
!= 0) {
3691 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3692 spa
->spa_feat_for_write_obj
);
3693 zap_cursor_retrieve(&zc
, &za
) == 0;
3694 zap_cursor_advance(&zc
)) {
3695 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3696 za
.za_num_integers
== 1);
3697 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3698 za
.za_first_integer
));
3700 zap_cursor_fini(&zc
);
3705 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3709 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3710 zfeature_info_t feature
= spa_feature_table
[i
];
3713 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3716 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3721 * Store a list of pool features and their reference counts in the
3724 * The first time this is called on a spa, allocate a new nvlist, fetch
3725 * the pool features and reference counts from disk, then save the list
3726 * in the spa. In subsequent calls on the same spa use the saved nvlist
3727 * and refresh its values from the cached reference counts. This
3728 * ensures we don't block here on I/O on a suspended pool so 'zpool
3729 * clear' can resume the pool.
3732 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3736 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3738 mutex_enter(&spa
->spa_feat_stats_lock
);
3739 features
= spa
->spa_feat_stats
;
3741 if (features
!= NULL
) {
3742 spa_feature_stats_from_cache(spa
, features
);
3744 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3745 spa
->spa_feat_stats
= features
;
3746 spa_feature_stats_from_disk(spa
, features
);
3749 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3752 mutex_exit(&spa
->spa_feat_stats_lock
);
3756 spa_get_stats(const char *name
, nvlist_t
**config
,
3757 char *altroot
, size_t buflen
)
3763 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3767 * This still leaves a window of inconsistency where the spares
3768 * or l2cache devices could change and the config would be
3769 * self-inconsistent.
3771 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3773 if (*config
!= NULL
) {
3774 uint64_t loadtimes
[2];
3776 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3777 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3778 VERIFY(nvlist_add_uint64_array(*config
,
3779 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3781 VERIFY(nvlist_add_uint64(*config
,
3782 ZPOOL_CONFIG_ERRCOUNT
,
3783 spa_get_errlog_size(spa
)) == 0);
3785 if (spa_suspended(spa
))
3786 VERIFY(nvlist_add_uint64(*config
,
3787 ZPOOL_CONFIG_SUSPENDED
,
3788 spa
->spa_failmode
) == 0);
3790 spa_add_spares(spa
, *config
);
3791 spa_add_l2cache(spa
, *config
);
3792 spa_add_feature_stats(spa
, *config
);
3797 * We want to get the alternate root even for faulted pools, so we cheat
3798 * and call spa_lookup() directly.
3802 mutex_enter(&spa_namespace_lock
);
3803 spa
= spa_lookup(name
);
3805 spa_altroot(spa
, altroot
, buflen
);
3809 mutex_exit(&spa_namespace_lock
);
3811 spa_altroot(spa
, altroot
, buflen
);
3816 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3817 spa_close(spa
, FTAG
);
3824 * Validate that the auxiliary device array is well formed. We must have an
3825 * array of nvlists, each which describes a valid leaf vdev. If this is an
3826 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3827 * specified, as long as they are well-formed.
3830 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3831 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3832 vdev_labeltype_t label
)
3839 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3842 * It's acceptable to have no devs specified.
3844 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3848 return (SET_ERROR(EINVAL
));
3851 * Make sure the pool is formatted with a version that supports this
3854 if (spa_version(spa
) < version
)
3855 return (SET_ERROR(ENOTSUP
));
3858 * Set the pending device list so we correctly handle device in-use
3861 sav
->sav_pending
= dev
;
3862 sav
->sav_npending
= ndev
;
3864 for (i
= 0; i
< ndev
; i
++) {
3865 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3869 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3871 error
= SET_ERROR(EINVAL
);
3877 if ((error
= vdev_open(vd
)) == 0 &&
3878 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3879 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3880 vd
->vdev_guid
) == 0);
3886 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3893 sav
->sav_pending
= NULL
;
3894 sav
->sav_npending
= 0;
3899 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3903 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3905 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3906 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3907 VDEV_LABEL_SPARE
)) != 0) {
3911 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3912 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3913 VDEV_LABEL_L2CACHE
));
3917 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3922 if (sav
->sav_config
!= NULL
) {
3928 * Generate new dev list by concatenating with the
3931 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3932 &olddevs
, &oldndevs
) == 0);
3934 newdevs
= kmem_alloc(sizeof (void *) *
3935 (ndevs
+ oldndevs
), KM_SLEEP
);
3936 for (i
= 0; i
< oldndevs
; i
++)
3937 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3939 for (i
= 0; i
< ndevs
; i
++)
3940 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3943 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3944 DATA_TYPE_NVLIST_ARRAY
) == 0);
3946 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3947 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3948 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3949 nvlist_free(newdevs
[i
]);
3950 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3953 * Generate a new dev list.
3955 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3957 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3963 * Stop and drop level 2 ARC devices
3966 spa_l2cache_drop(spa_t
*spa
)
3970 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3972 for (i
= 0; i
< sav
->sav_count
; i
++) {
3975 vd
= sav
->sav_vdevs
[i
];
3978 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3979 pool
!= 0ULL && l2arc_vdev_present(vd
))
3980 l2arc_remove_vdev(vd
);
3985 * Verify encryption parameters for spa creation. If we are encrypting, we must
3986 * have the encryption feature flag enabled.
3989 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
3990 boolean_t has_encryption
)
3992 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
3993 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
3995 return (SET_ERROR(ENOTSUP
));
3997 return (dmu_objset_create_crypt_check(NULL
, dcp
));
4004 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
4005 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
4008 char *altroot
= NULL
;
4013 uint64_t txg
= TXG_INITIAL
;
4014 nvlist_t
**spares
, **l2cache
;
4015 uint_t nspares
, nl2cache
;
4016 uint64_t version
, obj
, root_dsobj
= 0;
4017 boolean_t has_features
;
4018 boolean_t has_encryption
;
4026 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
4027 poolname
= (char *)pool
;
4030 * If this pool already exists, return failure.
4032 mutex_enter(&spa_namespace_lock
);
4033 if (spa_lookup(poolname
) != NULL
) {
4034 mutex_exit(&spa_namespace_lock
);
4035 return (SET_ERROR(EEXIST
));
4039 * Allocate a new spa_t structure.
4041 nvl
= fnvlist_alloc();
4042 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
4043 (void) nvlist_lookup_string(props
,
4044 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4045 spa
= spa_add(poolname
, nvl
, altroot
);
4047 spa_activate(spa
, spa_mode_global
);
4049 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4050 spa_deactivate(spa
);
4052 mutex_exit(&spa_namespace_lock
);
4057 * Temporary pool names should never be written to disk.
4059 if (poolname
!= pool
)
4060 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
4062 has_features
= B_FALSE
;
4063 has_encryption
= B_FALSE
;
4064 for (elem
= nvlist_next_nvpair(props
, NULL
);
4065 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4066 if (zpool_prop_feature(nvpair_name(elem
))) {
4067 has_features
= B_TRUE
;
4069 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
4070 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
4071 if (feat
== SPA_FEATURE_ENCRYPTION
)
4072 has_encryption
= B_TRUE
;
4076 /* verify encryption params, if they were provided */
4078 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
4080 spa_deactivate(spa
);
4082 mutex_exit(&spa_namespace_lock
);
4087 if (has_features
|| nvlist_lookup_uint64(props
,
4088 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4089 version
= SPA_VERSION
;
4091 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4093 spa
->spa_first_txg
= txg
;
4094 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4095 spa
->spa_uberblock
.ub_version
= version
;
4096 spa
->spa_ubsync
= spa
->spa_uberblock
;
4097 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4100 * Create "The Godfather" zio to hold all async IOs
4102 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4104 for (i
= 0; i
< max_ncpus
; i
++) {
4105 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4106 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4107 ZIO_FLAG_GODFATHER
);
4111 * Create the root vdev.
4113 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4115 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4117 ASSERT(error
!= 0 || rvd
!= NULL
);
4118 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4120 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4121 error
= SET_ERROR(EINVAL
);
4124 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4125 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4126 VDEV_ALLOC_ADD
)) == 0) {
4127 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
4128 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4129 vdev_expand(rvd
->vdev_child
[c
], txg
);
4133 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4137 spa_deactivate(spa
);
4139 mutex_exit(&spa_namespace_lock
);
4144 * Get the list of spares, if specified.
4146 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4147 &spares
, &nspares
) == 0) {
4148 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4150 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4151 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4152 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4153 spa_load_spares(spa
);
4154 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4155 spa
->spa_spares
.sav_sync
= B_TRUE
;
4159 * Get the list of level 2 cache devices, if specified.
4161 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4162 &l2cache
, &nl2cache
) == 0) {
4163 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4164 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4165 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4166 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4167 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4168 spa_load_l2cache(spa
);
4169 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4170 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4173 spa
->spa_is_initializing
= B_TRUE
;
4174 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
4175 spa
->spa_is_initializing
= B_FALSE
;
4178 * Create DDTs (dedup tables).
4182 spa_update_dspace(spa
);
4184 tx
= dmu_tx_create_assigned(dp
, txg
);
4187 * Create the pool's history object.
4189 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
4190 spa_history_create_obj(spa
, tx
);
4192 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4193 spa_history_log_version(spa
, "create", tx
);
4196 * Create the pool config object.
4198 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4199 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4200 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4202 if (zap_add(spa
->spa_meta_objset
,
4203 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4204 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4205 cmn_err(CE_PANIC
, "failed to add pool config");
4208 if (zap_add(spa
->spa_meta_objset
,
4209 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4210 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4211 cmn_err(CE_PANIC
, "failed to add pool version");
4214 /* Newly created pools with the right version are always deflated. */
4215 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4216 spa
->spa_deflate
= TRUE
;
4217 if (zap_add(spa
->spa_meta_objset
,
4218 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4219 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4220 cmn_err(CE_PANIC
, "failed to add deflate");
4225 * Create the deferred-free bpobj. Turn off compression
4226 * because sync-to-convergence takes longer if the blocksize
4229 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4230 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4231 ZIO_COMPRESS_OFF
, tx
);
4232 if (zap_add(spa
->spa_meta_objset
,
4233 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4234 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4235 cmn_err(CE_PANIC
, "failed to add bpobj");
4237 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4238 spa
->spa_meta_objset
, obj
));
4241 * Generate some random noise for salted checksums to operate on.
4243 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4244 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4247 * Set pool properties.
4249 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4250 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4251 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4252 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4253 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
4255 if (props
!= NULL
) {
4256 spa_configfile_set(spa
, props
, B_FALSE
);
4257 spa_sync_props(props
, tx
);
4263 * If the root dataset is encrypted we will need to create key mappings
4264 * for the zio layer before we start to write any data to disk and hold
4265 * them until after the first txg has been synced. Waiting for the first
4266 * transaction to complete also ensures that our bean counters are
4267 * appropriately updated.
4269 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0) {
4270 root_dsobj
= dsl_dir_phys(dp
->dp_root_dir
)->dd_head_dataset_obj
;
4271 VERIFY0(spa_keystore_create_mapping_impl(spa
, root_dsobj
,
4272 dp
->dp_root_dir
, FTAG
));
4275 spa
->spa_sync_on
= B_TRUE
;
4277 mmp_thread_start(spa
);
4278 txg_wait_synced(dp
, txg
);
4280 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0)
4281 VERIFY0(spa_keystore_remove_mapping(spa
, root_dsobj
, FTAG
));
4283 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4286 * Don't count references from objsets that are already closed
4287 * and are making their way through the eviction process.
4289 spa_evicting_os_wait(spa
);
4290 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4291 spa
->spa_load_state
= SPA_LOAD_NONE
;
4293 mutex_exit(&spa_namespace_lock
);
4299 * Import a non-root pool into the system.
4302 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4305 char *altroot
= NULL
;
4306 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4307 zpool_rewind_policy_t policy
;
4308 uint64_t mode
= spa_mode_global
;
4309 uint64_t readonly
= B_FALSE
;
4312 nvlist_t
**spares
, **l2cache
;
4313 uint_t nspares
, nl2cache
;
4316 * If a pool with this name exists, return failure.
4318 mutex_enter(&spa_namespace_lock
);
4319 if (spa_lookup(pool
) != NULL
) {
4320 mutex_exit(&spa_namespace_lock
);
4321 return (SET_ERROR(EEXIST
));
4325 * Create and initialize the spa structure.
4327 (void) nvlist_lookup_string(props
,
4328 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4329 (void) nvlist_lookup_uint64(props
,
4330 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4333 spa
= spa_add(pool
, config
, altroot
);
4334 spa
->spa_import_flags
= flags
;
4337 * Verbatim import - Take a pool and insert it into the namespace
4338 * as if it had been loaded at boot.
4340 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4342 spa_configfile_set(spa
, props
, B_FALSE
);
4344 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4345 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4347 mutex_exit(&spa_namespace_lock
);
4351 spa_activate(spa
, mode
);
4354 * Don't start async tasks until we know everything is healthy.
4356 spa_async_suspend(spa
);
4358 zpool_get_rewind_policy(config
, &policy
);
4359 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4360 state
= SPA_LOAD_RECOVER
;
4363 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4364 * because the user-supplied config is actually the one to trust when
4367 if (state
!= SPA_LOAD_RECOVER
)
4368 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4370 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4371 policy
.zrp_request
);
4374 * Propagate anything learned while loading the pool and pass it
4375 * back to caller (i.e. rewind info, missing devices, etc).
4377 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4378 spa
->spa_load_info
) == 0);
4380 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4382 * Toss any existing sparelist, as it doesn't have any validity
4383 * anymore, and conflicts with spa_has_spare().
4385 if (spa
->spa_spares
.sav_config
) {
4386 nvlist_free(spa
->spa_spares
.sav_config
);
4387 spa
->spa_spares
.sav_config
= NULL
;
4388 spa_load_spares(spa
);
4390 if (spa
->spa_l2cache
.sav_config
) {
4391 nvlist_free(spa
->spa_l2cache
.sav_config
);
4392 spa
->spa_l2cache
.sav_config
= NULL
;
4393 spa_load_l2cache(spa
);
4396 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4398 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4401 spa_configfile_set(spa
, props
, B_FALSE
);
4403 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4404 (error
= spa_prop_set(spa
, props
)))) {
4406 spa_deactivate(spa
);
4408 mutex_exit(&spa_namespace_lock
);
4412 spa_async_resume(spa
);
4415 * Override any spares and level 2 cache devices as specified by
4416 * the user, as these may have correct device names/devids, etc.
4418 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4419 &spares
, &nspares
) == 0) {
4420 if (spa
->spa_spares
.sav_config
)
4421 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4422 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4424 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4425 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4426 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4427 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4428 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4429 spa_load_spares(spa
);
4430 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4431 spa
->spa_spares
.sav_sync
= B_TRUE
;
4433 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4434 &l2cache
, &nl2cache
) == 0) {
4435 if (spa
->spa_l2cache
.sav_config
)
4436 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4437 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4439 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4440 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4441 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4442 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4443 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4444 spa_load_l2cache(spa
);
4445 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4446 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4450 * Check for any removed devices.
4452 if (spa
->spa_autoreplace
) {
4453 spa_aux_check_removed(&spa
->spa_spares
);
4454 spa_aux_check_removed(&spa
->spa_l2cache
);
4457 if (spa_writeable(spa
)) {
4459 * Update the config cache to include the newly-imported pool.
4461 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4465 * It's possible that the pool was expanded while it was exported.
4466 * We kick off an async task to handle this for us.
4468 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4470 spa_history_log_version(spa
, "import", NULL
);
4472 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4474 zvol_create_minors(spa
, pool
, B_TRUE
);
4476 mutex_exit(&spa_namespace_lock
);
4482 spa_tryimport(nvlist_t
*tryconfig
)
4484 nvlist_t
*config
= NULL
;
4490 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4493 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4497 * Create and initialize the spa structure.
4499 mutex_enter(&spa_namespace_lock
);
4500 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4501 spa_activate(spa
, FREAD
);
4504 * Pass off the heavy lifting to spa_load().
4505 * Pass TRUE for mosconfig because the user-supplied config
4506 * is actually the one to trust when doing an import.
4508 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4511 * If 'tryconfig' was at least parsable, return the current config.
4513 if (spa
->spa_root_vdev
!= NULL
) {
4514 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4515 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4517 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4519 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4520 spa
->spa_uberblock
.ub_timestamp
) == 0);
4521 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4522 spa
->spa_load_info
) == 0);
4523 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4524 spa
->spa_errata
) == 0);
4527 * If the bootfs property exists on this pool then we
4528 * copy it out so that external consumers can tell which
4529 * pools are bootable.
4531 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4532 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4535 * We have to play games with the name since the
4536 * pool was opened as TRYIMPORT_NAME.
4538 if (dsl_dsobj_to_dsname(spa_name(spa
),
4539 spa
->spa_bootfs
, tmpname
) == 0) {
4543 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4545 cp
= strchr(tmpname
, '/');
4547 (void) strlcpy(dsname
, tmpname
,
4550 (void) snprintf(dsname
, MAXPATHLEN
,
4551 "%s/%s", poolname
, ++cp
);
4553 VERIFY(nvlist_add_string(config
,
4554 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4555 kmem_free(dsname
, MAXPATHLEN
);
4557 kmem_free(tmpname
, MAXPATHLEN
);
4561 * Add the list of hot spares and level 2 cache devices.
4563 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4564 spa_add_spares(spa
, config
);
4565 spa_add_l2cache(spa
, config
);
4566 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4570 spa_deactivate(spa
);
4572 mutex_exit(&spa_namespace_lock
);
4578 * Pool export/destroy
4580 * The act of destroying or exporting a pool is very simple. We make sure there
4581 * is no more pending I/O and any references to the pool are gone. Then, we
4582 * update the pool state and sync all the labels to disk, removing the
4583 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4584 * we don't sync the labels or remove the configuration cache.
4587 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4588 boolean_t force
, boolean_t hardforce
)
4595 if (!(spa_mode_global
& FWRITE
))
4596 return (SET_ERROR(EROFS
));
4598 mutex_enter(&spa_namespace_lock
);
4599 if ((spa
= spa_lookup(pool
)) == NULL
) {
4600 mutex_exit(&spa_namespace_lock
);
4601 return (SET_ERROR(ENOENT
));
4605 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4606 * reacquire the namespace lock, and see if we can export.
4608 spa_open_ref(spa
, FTAG
);
4609 mutex_exit(&spa_namespace_lock
);
4610 spa_async_suspend(spa
);
4611 if (spa
->spa_zvol_taskq
) {
4612 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
4613 taskq_wait(spa
->spa_zvol_taskq
);
4615 mutex_enter(&spa_namespace_lock
);
4616 spa_close(spa
, FTAG
);
4618 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4621 * The pool will be in core if it's openable, in which case we can
4622 * modify its state. Objsets may be open only because they're dirty,
4623 * so we have to force it to sync before checking spa_refcnt.
4625 if (spa
->spa_sync_on
) {
4626 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4627 spa_evicting_os_wait(spa
);
4631 * A pool cannot be exported or destroyed if there are active
4632 * references. If we are resetting a pool, allow references by
4633 * fault injection handlers.
4635 if (!spa_refcount_zero(spa
) ||
4636 (spa
->spa_inject_ref
!= 0 &&
4637 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4638 spa_async_resume(spa
);
4639 mutex_exit(&spa_namespace_lock
);
4640 return (SET_ERROR(EBUSY
));
4643 if (spa
->spa_sync_on
) {
4645 * A pool cannot be exported if it has an active shared spare.
4646 * This is to prevent other pools stealing the active spare
4647 * from an exported pool. At user's own will, such pool can
4648 * be forcedly exported.
4650 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4651 spa_has_active_shared_spare(spa
)) {
4652 spa_async_resume(spa
);
4653 mutex_exit(&spa_namespace_lock
);
4654 return (SET_ERROR(EXDEV
));
4658 * We want this to be reflected on every label,
4659 * so mark them all dirty. spa_unload() will do the
4660 * final sync that pushes these changes out.
4662 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4663 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4664 spa
->spa_state
= new_state
;
4665 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4667 vdev_config_dirty(spa
->spa_root_vdev
);
4668 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4673 if (new_state
== POOL_STATE_DESTROYED
)
4674 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
4675 else if (new_state
== POOL_STATE_EXPORTED
)
4676 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
4678 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4680 spa_deactivate(spa
);
4683 if (oldconfig
&& spa
->spa_config
)
4684 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4686 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4688 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4691 mutex_exit(&spa_namespace_lock
);
4697 * Destroy a storage pool.
4700 spa_destroy(char *pool
)
4702 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4707 * Export a storage pool.
4710 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4711 boolean_t hardforce
)
4713 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4718 * Similar to spa_export(), this unloads the spa_t without actually removing it
4719 * from the namespace in any way.
4722 spa_reset(char *pool
)
4724 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4729 * ==========================================================================
4730 * Device manipulation
4731 * ==========================================================================
4735 * Add a device to a storage pool.
4738 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4742 vdev_t
*rvd
= spa
->spa_root_vdev
;
4744 nvlist_t
**spares
, **l2cache
;
4745 uint_t nspares
, nl2cache
;
4748 ASSERT(spa_writeable(spa
));
4750 txg
= spa_vdev_enter(spa
);
4752 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4753 VDEV_ALLOC_ADD
)) != 0)
4754 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4756 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4758 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4762 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4766 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4767 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4769 if (vd
->vdev_children
!= 0 &&
4770 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4771 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4774 * We must validate the spares and l2cache devices after checking the
4775 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4777 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4778 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4781 * Transfer each new top-level vdev from vd to rvd.
4783 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4786 * Set the vdev id to the first hole, if one exists.
4788 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4789 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4790 vdev_free(rvd
->vdev_child
[id
]);
4794 tvd
= vd
->vdev_child
[c
];
4795 vdev_remove_child(vd
, tvd
);
4797 vdev_add_child(rvd
, tvd
);
4798 vdev_config_dirty(tvd
);
4802 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4803 ZPOOL_CONFIG_SPARES
);
4804 spa_load_spares(spa
);
4805 spa
->spa_spares
.sav_sync
= B_TRUE
;
4808 if (nl2cache
!= 0) {
4809 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4810 ZPOOL_CONFIG_L2CACHE
);
4811 spa_load_l2cache(spa
);
4812 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4816 * We have to be careful when adding new vdevs to an existing pool.
4817 * If other threads start allocating from these vdevs before we
4818 * sync the config cache, and we lose power, then upon reboot we may
4819 * fail to open the pool because there are DVAs that the config cache
4820 * can't translate. Therefore, we first add the vdevs without
4821 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4822 * and then let spa_config_update() initialize the new metaslabs.
4824 * spa_load() checks for added-but-not-initialized vdevs, so that
4825 * if we lose power at any point in this sequence, the remaining
4826 * steps will be completed the next time we load the pool.
4828 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4830 mutex_enter(&spa_namespace_lock
);
4831 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4832 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
4833 mutex_exit(&spa_namespace_lock
);
4839 * Attach a device to a mirror. The arguments are the path to any device
4840 * in the mirror, and the nvroot for the new device. If the path specifies
4841 * a device that is not mirrored, we automatically insert the mirror vdev.
4843 * If 'replacing' is specified, the new device is intended to replace the
4844 * existing device; in this case the two devices are made into their own
4845 * mirror using the 'replacing' vdev, which is functionally identical to
4846 * the mirror vdev (it actually reuses all the same ops) but has a few
4847 * extra rules: you can't attach to it after it's been created, and upon
4848 * completion of resilvering, the first disk (the one being replaced)
4849 * is automatically detached.
4852 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4854 uint64_t txg
, dtl_max_txg
;
4855 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4857 char *oldvdpath
, *newvdpath
;
4860 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4862 ASSERT(spa_writeable(spa
));
4864 txg
= spa_vdev_enter(spa
);
4866 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4869 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4871 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4872 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4874 pvd
= oldvd
->vdev_parent
;
4876 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4877 VDEV_ALLOC_ATTACH
)) != 0)
4878 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4880 if (newrootvd
->vdev_children
!= 1)
4881 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4883 newvd
= newrootvd
->vdev_child
[0];
4885 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4886 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4888 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4889 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4892 * Spares can't replace logs
4894 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4895 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4899 * For attach, the only allowable parent is a mirror or the root
4902 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4903 pvd
->vdev_ops
!= &vdev_root_ops
)
4904 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4906 pvops
= &vdev_mirror_ops
;
4909 * Active hot spares can only be replaced by inactive hot
4912 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4913 oldvd
->vdev_isspare
&&
4914 !spa_has_spare(spa
, newvd
->vdev_guid
))
4915 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4918 * If the source is a hot spare, and the parent isn't already a
4919 * spare, then we want to create a new hot spare. Otherwise, we
4920 * want to create a replacing vdev. The user is not allowed to
4921 * attach to a spared vdev child unless the 'isspare' state is
4922 * the same (spare replaces spare, non-spare replaces
4925 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4926 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4927 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4928 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4929 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4930 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4933 if (newvd
->vdev_isspare
)
4934 pvops
= &vdev_spare_ops
;
4936 pvops
= &vdev_replacing_ops
;
4940 * Make sure the new device is big enough.
4942 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4943 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4946 * The new device cannot have a higher alignment requirement
4947 * than the top-level vdev.
4949 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4950 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4953 * If this is an in-place replacement, update oldvd's path and devid
4954 * to make it distinguishable from newvd, and unopenable from now on.
4956 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4957 spa_strfree(oldvd
->vdev_path
);
4958 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4960 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4961 newvd
->vdev_path
, "old");
4962 if (oldvd
->vdev_devid
!= NULL
) {
4963 spa_strfree(oldvd
->vdev_devid
);
4964 oldvd
->vdev_devid
= NULL
;
4968 /* mark the device being resilvered */
4969 newvd
->vdev_resilver_txg
= txg
;
4972 * If the parent is not a mirror, or if we're replacing, insert the new
4973 * mirror/replacing/spare vdev above oldvd.
4975 if (pvd
->vdev_ops
!= pvops
)
4976 pvd
= vdev_add_parent(oldvd
, pvops
);
4978 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4979 ASSERT(pvd
->vdev_ops
== pvops
);
4980 ASSERT(oldvd
->vdev_parent
== pvd
);
4983 * Extract the new device from its root and add it to pvd.
4985 vdev_remove_child(newrootvd
, newvd
);
4986 newvd
->vdev_id
= pvd
->vdev_children
;
4987 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4988 vdev_add_child(pvd
, newvd
);
4991 * Reevaluate the parent vdev state.
4993 vdev_propagate_state(pvd
);
4995 tvd
= newvd
->vdev_top
;
4996 ASSERT(pvd
->vdev_top
== tvd
);
4997 ASSERT(tvd
->vdev_parent
== rvd
);
4999 vdev_config_dirty(tvd
);
5002 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5003 * for any dmu_sync-ed blocks. It will propagate upward when
5004 * spa_vdev_exit() calls vdev_dtl_reassess().
5006 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
5008 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
5009 dtl_max_txg
- TXG_INITIAL
);
5011 if (newvd
->vdev_isspare
) {
5012 spa_spare_activate(newvd
);
5013 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
5016 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
5017 newvdpath
= spa_strdup(newvd
->vdev_path
);
5018 newvd_isspare
= newvd
->vdev_isspare
;
5021 * Mark newvd's DTL dirty in this txg.
5023 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
5026 * Schedule the resilver to restart in the future. We do this to
5027 * ensure that dmu_sync-ed blocks have been stitched into the
5028 * respective datasets.
5030 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
5032 if (spa
->spa_bootfs
)
5033 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
5035 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
5040 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
5042 spa_history_log_internal(spa
, "vdev attach", NULL
,
5043 "%s vdev=%s %s vdev=%s",
5044 replacing
&& newvd_isspare
? "spare in" :
5045 replacing
? "replace" : "attach", newvdpath
,
5046 replacing
? "for" : "to", oldvdpath
);
5048 spa_strfree(oldvdpath
);
5049 spa_strfree(newvdpath
);
5055 * Detach a device from a mirror or replacing vdev.
5057 * If 'replace_done' is specified, only detach if the parent
5058 * is a replacing vdev.
5061 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5065 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5066 boolean_t unspare
= B_FALSE
;
5067 uint64_t unspare_guid
= 0;
5070 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5071 ASSERT(spa_writeable(spa
));
5073 txg
= spa_vdev_enter(spa
);
5075 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5078 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5080 if (!vd
->vdev_ops
->vdev_op_leaf
)
5081 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5083 pvd
= vd
->vdev_parent
;
5086 * If the parent/child relationship is not as expected, don't do it.
5087 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5088 * vdev that's replacing B with C. The user's intent in replacing
5089 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5090 * the replace by detaching C, the expected behavior is to end up
5091 * M(A,B). But suppose that right after deciding to detach C,
5092 * the replacement of B completes. We would have M(A,C), and then
5093 * ask to detach C, which would leave us with just A -- not what
5094 * the user wanted. To prevent this, we make sure that the
5095 * parent/child relationship hasn't changed -- in this example,
5096 * that C's parent is still the replacing vdev R.
5098 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5099 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5102 * Only 'replacing' or 'spare' vdevs can be replaced.
5104 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5105 pvd
->vdev_ops
!= &vdev_spare_ops
)
5106 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5108 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5109 spa_version(spa
) >= SPA_VERSION_SPARES
);
5112 * Only mirror, replacing, and spare vdevs support detach.
5114 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5115 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5116 pvd
->vdev_ops
!= &vdev_spare_ops
)
5117 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5120 * If this device has the only valid copy of some data,
5121 * we cannot safely detach it.
5123 if (vdev_dtl_required(vd
))
5124 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5126 ASSERT(pvd
->vdev_children
>= 2);
5129 * If we are detaching the second disk from a replacing vdev, then
5130 * check to see if we changed the original vdev's path to have "/old"
5131 * at the end in spa_vdev_attach(). If so, undo that change now.
5133 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5134 vd
->vdev_path
!= NULL
) {
5135 size_t len
= strlen(vd
->vdev_path
);
5137 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
5138 cvd
= pvd
->vdev_child
[c
];
5140 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5143 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5144 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5145 spa_strfree(cvd
->vdev_path
);
5146 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5153 * If we are detaching the original disk from a spare, then it implies
5154 * that the spare should become a real disk, and be removed from the
5155 * active spare list for the pool.
5157 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5159 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5163 * Erase the disk labels so the disk can be used for other things.
5164 * This must be done after all other error cases are handled,
5165 * but before we disembowel vd (so we can still do I/O to it).
5166 * But if we can't do it, don't treat the error as fatal --
5167 * it may be that the unwritability of the disk is the reason
5168 * it's being detached!
5170 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5173 * Remove vd from its parent and compact the parent's children.
5175 vdev_remove_child(pvd
, vd
);
5176 vdev_compact_children(pvd
);
5179 * Remember one of the remaining children so we can get tvd below.
5181 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5184 * If we need to remove the remaining child from the list of hot spares,
5185 * do it now, marking the vdev as no longer a spare in the process.
5186 * We must do this before vdev_remove_parent(), because that can
5187 * change the GUID if it creates a new toplevel GUID. For a similar
5188 * reason, we must remove the spare now, in the same txg as the detach;
5189 * otherwise someone could attach a new sibling, change the GUID, and
5190 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5193 ASSERT(cvd
->vdev_isspare
);
5194 spa_spare_remove(cvd
);
5195 unspare_guid
= cvd
->vdev_guid
;
5196 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5197 cvd
->vdev_unspare
= B_TRUE
;
5201 * If the parent mirror/replacing vdev only has one child,
5202 * the parent is no longer needed. Remove it from the tree.
5204 if (pvd
->vdev_children
== 1) {
5205 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5206 cvd
->vdev_unspare
= B_FALSE
;
5207 vdev_remove_parent(cvd
);
5212 * We don't set tvd until now because the parent we just removed
5213 * may have been the previous top-level vdev.
5215 tvd
= cvd
->vdev_top
;
5216 ASSERT(tvd
->vdev_parent
== rvd
);
5219 * Reevaluate the parent vdev state.
5221 vdev_propagate_state(cvd
);
5224 * If the 'autoexpand' property is set on the pool then automatically
5225 * try to expand the size of the pool. For example if the device we
5226 * just detached was smaller than the others, it may be possible to
5227 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5228 * first so that we can obtain the updated sizes of the leaf vdevs.
5230 if (spa
->spa_autoexpand
) {
5232 vdev_expand(tvd
, txg
);
5235 vdev_config_dirty(tvd
);
5238 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5239 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5240 * But first make sure we're not on any *other* txg's DTL list, to
5241 * prevent vd from being accessed after it's freed.
5243 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
5244 for (t
= 0; t
< TXG_SIZE
; t
++)
5245 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5246 vd
->vdev_detached
= B_TRUE
;
5247 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5249 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5251 /* hang on to the spa before we release the lock */
5252 spa_open_ref(spa
, FTAG
);
5254 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5256 spa_history_log_internal(spa
, "detach", NULL
,
5258 spa_strfree(vdpath
);
5261 * If this was the removal of the original device in a hot spare vdev,
5262 * then we want to go through and remove the device from the hot spare
5263 * list of every other pool.
5266 spa_t
*altspa
= NULL
;
5268 mutex_enter(&spa_namespace_lock
);
5269 while ((altspa
= spa_next(altspa
)) != NULL
) {
5270 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5274 spa_open_ref(altspa
, FTAG
);
5275 mutex_exit(&spa_namespace_lock
);
5276 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5277 mutex_enter(&spa_namespace_lock
);
5278 spa_close(altspa
, FTAG
);
5280 mutex_exit(&spa_namespace_lock
);
5282 /* search the rest of the vdevs for spares to remove */
5283 spa_vdev_resilver_done(spa
);
5286 /* all done with the spa; OK to release */
5287 mutex_enter(&spa_namespace_lock
);
5288 spa_close(spa
, FTAG
);
5289 mutex_exit(&spa_namespace_lock
);
5295 * Split a set of devices from their mirrors, and create a new pool from them.
5298 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5299 nvlist_t
*props
, boolean_t exp
)
5302 uint64_t txg
, *glist
;
5304 uint_t c
, children
, lastlog
;
5305 nvlist_t
**child
, *nvl
, *tmp
;
5307 char *altroot
= NULL
;
5308 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5309 boolean_t activate_slog
;
5311 ASSERT(spa_writeable(spa
));
5313 txg
= spa_vdev_enter(spa
);
5315 /* clear the log and flush everything up to now */
5316 activate_slog
= spa_passivate_log(spa
);
5317 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5318 error
= spa_offline_log(spa
);
5319 txg
= spa_vdev_config_enter(spa
);
5322 spa_activate_log(spa
);
5325 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5327 /* check new spa name before going any further */
5328 if (spa_lookup(newname
) != NULL
)
5329 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5332 * scan through all the children to ensure they're all mirrors
5334 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5335 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5337 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5339 /* first, check to ensure we've got the right child count */
5340 rvd
= spa
->spa_root_vdev
;
5342 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5343 vdev_t
*vd
= rvd
->vdev_child
[c
];
5345 /* don't count the holes & logs as children */
5346 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5354 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5355 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5357 /* next, ensure no spare or cache devices are part of the split */
5358 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5359 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5360 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5362 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5363 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5365 /* then, loop over each vdev and validate it */
5366 for (c
= 0; c
< children
; c
++) {
5367 uint64_t is_hole
= 0;
5369 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5373 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5374 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5377 error
= SET_ERROR(EINVAL
);
5382 /* which disk is going to be split? */
5383 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5385 error
= SET_ERROR(EINVAL
);
5389 /* look it up in the spa */
5390 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5391 if (vml
[c
] == NULL
) {
5392 error
= SET_ERROR(ENODEV
);
5396 /* make sure there's nothing stopping the split */
5397 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5398 vml
[c
]->vdev_islog
||
5399 vml
[c
]->vdev_ishole
||
5400 vml
[c
]->vdev_isspare
||
5401 vml
[c
]->vdev_isl2cache
||
5402 !vdev_writeable(vml
[c
]) ||
5403 vml
[c
]->vdev_children
!= 0 ||
5404 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5405 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5406 error
= SET_ERROR(EINVAL
);
5410 if (vdev_dtl_required(vml
[c
])) {
5411 error
= SET_ERROR(EBUSY
);
5415 /* we need certain info from the top level */
5416 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5417 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5418 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5419 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5420 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5421 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5422 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5423 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5425 /* transfer per-vdev ZAPs */
5426 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
5427 VERIFY0(nvlist_add_uint64(child
[c
],
5428 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
5430 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
5431 VERIFY0(nvlist_add_uint64(child
[c
],
5432 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
5433 vml
[c
]->vdev_parent
->vdev_top_zap
));
5437 kmem_free(vml
, children
* sizeof (vdev_t
*));
5438 kmem_free(glist
, children
* sizeof (uint64_t));
5439 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5442 /* stop writers from using the disks */
5443 for (c
= 0; c
< children
; c
++) {
5445 vml
[c
]->vdev_offline
= B_TRUE
;
5447 vdev_reopen(spa
->spa_root_vdev
);
5450 * Temporarily record the splitting vdevs in the spa config. This
5451 * will disappear once the config is regenerated.
5453 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5454 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5455 glist
, children
) == 0);
5456 kmem_free(glist
, children
* sizeof (uint64_t));
5458 mutex_enter(&spa
->spa_props_lock
);
5459 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5461 mutex_exit(&spa
->spa_props_lock
);
5462 spa
->spa_config_splitting
= nvl
;
5463 vdev_config_dirty(spa
->spa_root_vdev
);
5465 /* configure and create the new pool */
5466 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5467 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5468 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5469 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5470 spa_version(spa
)) == 0);
5471 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5472 spa
->spa_config_txg
) == 0);
5473 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5474 spa_generate_guid(NULL
)) == 0);
5475 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
5476 (void) nvlist_lookup_string(props
,
5477 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5479 /* add the new pool to the namespace */
5480 newspa
= spa_add(newname
, config
, altroot
);
5481 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5482 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5483 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5485 /* release the spa config lock, retaining the namespace lock */
5486 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5488 if (zio_injection_enabled
)
5489 zio_handle_panic_injection(spa
, FTAG
, 1);
5491 spa_activate(newspa
, spa_mode_global
);
5492 spa_async_suspend(newspa
);
5494 /* create the new pool from the disks of the original pool */
5495 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5499 /* if that worked, generate a real config for the new pool */
5500 if (newspa
->spa_root_vdev
!= NULL
) {
5501 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5502 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5503 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5504 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5505 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5510 if (props
!= NULL
) {
5511 spa_configfile_set(newspa
, props
, B_FALSE
);
5512 error
= spa_prop_set(newspa
, props
);
5517 /* flush everything */
5518 txg
= spa_vdev_config_enter(newspa
);
5519 vdev_config_dirty(newspa
->spa_root_vdev
);
5520 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5522 if (zio_injection_enabled
)
5523 zio_handle_panic_injection(spa
, FTAG
, 2);
5525 spa_async_resume(newspa
);
5527 /* finally, update the original pool's config */
5528 txg
= spa_vdev_config_enter(spa
);
5529 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5530 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5533 for (c
= 0; c
< children
; c
++) {
5534 if (vml
[c
] != NULL
) {
5537 spa_history_log_internal(spa
, "detach", tx
,
5538 "vdev=%s", vml
[c
]->vdev_path
);
5543 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5544 vdev_config_dirty(spa
->spa_root_vdev
);
5545 spa
->spa_config_splitting
= NULL
;
5549 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5551 if (zio_injection_enabled
)
5552 zio_handle_panic_injection(spa
, FTAG
, 3);
5554 /* split is complete; log a history record */
5555 spa_history_log_internal(newspa
, "split", NULL
,
5556 "from pool %s", spa_name(spa
));
5558 kmem_free(vml
, children
* sizeof (vdev_t
*));
5560 /* if we're not going to mount the filesystems in userland, export */
5562 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5569 spa_deactivate(newspa
);
5572 txg
= spa_vdev_config_enter(spa
);
5574 /* re-online all offlined disks */
5575 for (c
= 0; c
< children
; c
++) {
5577 vml
[c
]->vdev_offline
= B_FALSE
;
5579 vdev_reopen(spa
->spa_root_vdev
);
5581 nvlist_free(spa
->spa_config_splitting
);
5582 spa
->spa_config_splitting
= NULL
;
5583 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5585 kmem_free(vml
, children
* sizeof (vdev_t
*));
5590 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5594 for (i
= 0; i
< count
; i
++) {
5597 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5600 if (guid
== target_guid
)
5608 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5609 nvlist_t
*dev_to_remove
)
5611 nvlist_t
**newdev
= NULL
;
5615 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5617 for (i
= 0, j
= 0; i
< count
; i
++) {
5618 if (dev
[i
] == dev_to_remove
)
5620 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5623 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5624 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5626 for (i
= 0; i
< count
- 1; i
++)
5627 nvlist_free(newdev
[i
]);
5630 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5634 * Evacuate the device.
5637 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5642 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5643 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5644 ASSERT(vd
== vd
->vdev_top
);
5647 * Evacuate the device. We don't hold the config lock as writer
5648 * since we need to do I/O but we do keep the
5649 * spa_namespace_lock held. Once this completes the device
5650 * should no longer have any blocks allocated on it.
5652 if (vd
->vdev_islog
) {
5653 if (vd
->vdev_stat
.vs_alloc
!= 0)
5654 error
= spa_offline_log(spa
);
5656 error
= SET_ERROR(ENOTSUP
);
5663 * The evacuation succeeded. Remove any remaining MOS metadata
5664 * associated with this vdev, and wait for these changes to sync.
5666 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5667 txg
= spa_vdev_config_enter(spa
);
5668 vd
->vdev_removing
= B_TRUE
;
5669 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5670 vdev_config_dirty(vd
);
5671 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5677 * Complete the removal by cleaning up the namespace.
5680 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5682 vdev_t
*rvd
= spa
->spa_root_vdev
;
5683 uint64_t id
= vd
->vdev_id
;
5684 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5686 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5687 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5688 ASSERT(vd
== vd
->vdev_top
);
5691 * Only remove any devices which are empty.
5693 if (vd
->vdev_stat
.vs_alloc
!= 0)
5696 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5698 if (list_link_active(&vd
->vdev_state_dirty_node
))
5699 vdev_state_clean(vd
);
5700 if (list_link_active(&vd
->vdev_config_dirty_node
))
5701 vdev_config_clean(vd
);
5706 vdev_compact_children(rvd
);
5708 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5709 vdev_add_child(rvd
, vd
);
5711 vdev_config_dirty(rvd
);
5714 * Reassess the health of our root vdev.
5720 * Remove a device from the pool -
5722 * Removing a device from the vdev namespace requires several steps
5723 * and can take a significant amount of time. As a result we use
5724 * the spa_vdev_config_[enter/exit] functions which allow us to
5725 * grab and release the spa_config_lock while still holding the namespace
5726 * lock. During each step the configuration is synced out.
5728 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5732 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5735 sysevent_t
*ev
= NULL
;
5736 metaslab_group_t
*mg
;
5737 nvlist_t
**spares
, **l2cache
, *nv
;
5739 uint_t nspares
, nl2cache
;
5741 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5743 ASSERT(spa_writeable(spa
));
5746 txg
= spa_vdev_enter(spa
);
5748 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5750 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5751 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5752 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5753 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5755 * Only remove the hot spare if it's not currently in use
5758 if (vd
== NULL
|| unspare
) {
5760 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5761 ev
= spa_event_create(spa
, vd
, NULL
,
5762 ESC_ZFS_VDEV_REMOVE_AUX
);
5763 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5764 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5765 spa_load_spares(spa
);
5766 spa
->spa_spares
.sav_sync
= B_TRUE
;
5768 error
= SET_ERROR(EBUSY
);
5770 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5771 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5772 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5773 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5775 * Cache devices can always be removed.
5777 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5778 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_AUX
);
5779 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5780 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5781 spa_load_l2cache(spa
);
5782 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5783 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5785 ASSERT(vd
== vd
->vdev_top
);
5790 * Stop allocating from this vdev.
5792 metaslab_group_passivate(mg
);
5795 * Wait for the youngest allocations and frees to sync,
5796 * and then wait for the deferral of those frees to finish.
5798 spa_vdev_config_exit(spa
, NULL
,
5799 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5802 * Attempt to evacuate the vdev.
5804 error
= spa_vdev_remove_evacuate(spa
, vd
);
5806 txg
= spa_vdev_config_enter(spa
);
5809 * If we couldn't evacuate the vdev, unwind.
5812 metaslab_group_activate(mg
);
5813 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5817 * Clean up the vdev namespace.
5819 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_DEV
);
5820 spa_vdev_remove_from_namespace(spa
, vd
);
5822 } else if (vd
!= NULL
) {
5824 * Normal vdevs cannot be removed (yet).
5826 error
= SET_ERROR(ENOTSUP
);
5829 * There is no vdev of any kind with the specified guid.
5831 error
= SET_ERROR(ENOENT
);
5835 error
= spa_vdev_exit(spa
, NULL
, txg
, error
);
5844 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5845 * currently spared, so we can detach it.
5848 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5850 vdev_t
*newvd
, *oldvd
;
5853 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5854 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5860 * Check for a completed replacement. We always consider the first
5861 * vdev in the list to be the oldest vdev, and the last one to be
5862 * the newest (see spa_vdev_attach() for how that works). In
5863 * the case where the newest vdev is faulted, we will not automatically
5864 * remove it after a resilver completes. This is OK as it will require
5865 * user intervention to determine which disk the admin wishes to keep.
5867 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5868 ASSERT(vd
->vdev_children
> 1);
5870 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5871 oldvd
= vd
->vdev_child
[0];
5873 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5874 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5875 !vdev_dtl_required(oldvd
))
5880 * Check for a completed resilver with the 'unspare' flag set.
5882 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5883 vdev_t
*first
= vd
->vdev_child
[0];
5884 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5886 if (last
->vdev_unspare
) {
5889 } else if (first
->vdev_unspare
) {
5896 if (oldvd
!= NULL
&&
5897 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5898 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5899 !vdev_dtl_required(oldvd
))
5903 * If there are more than two spares attached to a disk,
5904 * and those spares are not required, then we want to
5905 * attempt to free them up now so that they can be used
5906 * by other pools. Once we're back down to a single
5907 * disk+spare, we stop removing them.
5909 if (vd
->vdev_children
> 2) {
5910 newvd
= vd
->vdev_child
[1];
5912 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5913 vdev_dtl_empty(last
, DTL_MISSING
) &&
5914 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5915 !vdev_dtl_required(newvd
))
5924 spa_vdev_resilver_done(spa_t
*spa
)
5926 vdev_t
*vd
, *pvd
, *ppvd
;
5927 uint64_t guid
, sguid
, pguid
, ppguid
;
5929 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5931 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5932 pvd
= vd
->vdev_parent
;
5933 ppvd
= pvd
->vdev_parent
;
5934 guid
= vd
->vdev_guid
;
5935 pguid
= pvd
->vdev_guid
;
5936 ppguid
= ppvd
->vdev_guid
;
5939 * If we have just finished replacing a hot spared device, then
5940 * we need to detach the parent's first child (the original hot
5943 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5944 ppvd
->vdev_children
== 2) {
5945 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5946 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5948 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5950 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5951 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5953 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5955 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5958 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5962 * Update the stored path or FRU for this vdev.
5965 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5969 boolean_t sync
= B_FALSE
;
5971 ASSERT(spa_writeable(spa
));
5973 spa_vdev_state_enter(spa
, SCL_ALL
);
5975 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5976 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5978 if (!vd
->vdev_ops
->vdev_op_leaf
)
5979 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5982 if (strcmp(value
, vd
->vdev_path
) != 0) {
5983 spa_strfree(vd
->vdev_path
);
5984 vd
->vdev_path
= spa_strdup(value
);
5988 if (vd
->vdev_fru
== NULL
) {
5989 vd
->vdev_fru
= spa_strdup(value
);
5991 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5992 spa_strfree(vd
->vdev_fru
);
5993 vd
->vdev_fru
= spa_strdup(value
);
5998 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
6002 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
6004 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
6008 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
6010 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
6014 * ==========================================================================
6016 * ==========================================================================
6019 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
6021 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6023 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6024 return (SET_ERROR(EBUSY
));
6026 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
6030 spa_scan_stop(spa_t
*spa
)
6032 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6033 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6034 return (SET_ERROR(EBUSY
));
6035 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
6039 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
6041 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6043 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
6044 return (SET_ERROR(ENOTSUP
));
6047 * If a resilver was requested, but there is no DTL on a
6048 * writeable leaf device, we have nothing to do.
6050 if (func
== POOL_SCAN_RESILVER
&&
6051 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
6052 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6056 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6060 * ==========================================================================
6061 * SPA async task processing
6062 * ==========================================================================
6066 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6070 if (vd
->vdev_remove_wanted
) {
6071 vd
->vdev_remove_wanted
= B_FALSE
;
6072 vd
->vdev_delayed_close
= B_FALSE
;
6073 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6076 * We want to clear the stats, but we don't want to do a full
6077 * vdev_clear() as that will cause us to throw away
6078 * degraded/faulted state as well as attempt to reopen the
6079 * device, all of which is a waste.
6081 vd
->vdev_stat
.vs_read_errors
= 0;
6082 vd
->vdev_stat
.vs_write_errors
= 0;
6083 vd
->vdev_stat
.vs_checksum_errors
= 0;
6085 vdev_state_dirty(vd
->vdev_top
);
6088 for (c
= 0; c
< vd
->vdev_children
; c
++)
6089 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6093 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6097 if (vd
->vdev_probe_wanted
) {
6098 vd
->vdev_probe_wanted
= B_FALSE
;
6099 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6102 for (c
= 0; c
< vd
->vdev_children
; c
++)
6103 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6107 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6111 if (!spa
->spa_autoexpand
)
6114 for (c
= 0; c
< vd
->vdev_children
; c
++) {
6115 vdev_t
*cvd
= vd
->vdev_child
[c
];
6116 spa_async_autoexpand(spa
, cvd
);
6119 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6122 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
6126 spa_async_thread(void *arg
)
6128 spa_t
*spa
= (spa_t
*)arg
;
6131 ASSERT(spa
->spa_sync_on
);
6133 mutex_enter(&spa
->spa_async_lock
);
6134 tasks
= spa
->spa_async_tasks
;
6135 spa
->spa_async_tasks
= 0;
6136 mutex_exit(&spa
->spa_async_lock
);
6139 * See if the config needs to be updated.
6141 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6142 uint64_t old_space
, new_space
;
6144 mutex_enter(&spa_namespace_lock
);
6145 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6146 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6147 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6148 mutex_exit(&spa_namespace_lock
);
6151 * If the pool grew as a result of the config update,
6152 * then log an internal history event.
6154 if (new_space
!= old_space
) {
6155 spa_history_log_internal(spa
, "vdev online", NULL
,
6156 "pool '%s' size: %llu(+%llu)",
6157 spa_name(spa
), new_space
, new_space
- old_space
);
6162 * See if any devices need to be marked REMOVED.
6164 if (tasks
& SPA_ASYNC_REMOVE
) {
6165 spa_vdev_state_enter(spa
, SCL_NONE
);
6166 spa_async_remove(spa
, spa
->spa_root_vdev
);
6167 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6168 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6169 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6170 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6171 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6174 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6175 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6176 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6177 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6181 * See if any devices need to be probed.
6183 if (tasks
& SPA_ASYNC_PROBE
) {
6184 spa_vdev_state_enter(spa
, SCL_NONE
);
6185 spa_async_probe(spa
, spa
->spa_root_vdev
);
6186 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6190 * If any devices are done replacing, detach them.
6192 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6193 spa_vdev_resilver_done(spa
);
6196 * Kick off a resilver.
6198 if (tasks
& SPA_ASYNC_RESILVER
)
6199 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6202 * Let the world know that we're done.
6204 mutex_enter(&spa
->spa_async_lock
);
6205 spa
->spa_async_thread
= NULL
;
6206 cv_broadcast(&spa
->spa_async_cv
);
6207 mutex_exit(&spa
->spa_async_lock
);
6212 spa_async_suspend(spa_t
*spa
)
6214 mutex_enter(&spa
->spa_async_lock
);
6215 spa
->spa_async_suspended
++;
6216 while (spa
->spa_async_thread
!= NULL
)
6217 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6218 mutex_exit(&spa
->spa_async_lock
);
6222 spa_async_resume(spa_t
*spa
)
6224 mutex_enter(&spa
->spa_async_lock
);
6225 ASSERT(spa
->spa_async_suspended
!= 0);
6226 spa
->spa_async_suspended
--;
6227 mutex_exit(&spa
->spa_async_lock
);
6231 spa_async_tasks_pending(spa_t
*spa
)
6233 uint_t non_config_tasks
;
6235 boolean_t config_task_suspended
;
6237 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6238 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6239 if (spa
->spa_ccw_fail_time
== 0) {
6240 config_task_suspended
= B_FALSE
;
6242 config_task_suspended
=
6243 (gethrtime() - spa
->spa_ccw_fail_time
) <
6244 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
6247 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6251 spa_async_dispatch(spa_t
*spa
)
6253 mutex_enter(&spa
->spa_async_lock
);
6254 if (spa_async_tasks_pending(spa
) &&
6255 !spa
->spa_async_suspended
&&
6256 spa
->spa_async_thread
== NULL
&&
6258 spa
->spa_async_thread
= thread_create(NULL
, 0,
6259 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6260 mutex_exit(&spa
->spa_async_lock
);
6264 spa_async_request(spa_t
*spa
, int task
)
6266 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6267 mutex_enter(&spa
->spa_async_lock
);
6268 spa
->spa_async_tasks
|= task
;
6269 mutex_exit(&spa
->spa_async_lock
);
6273 * ==========================================================================
6274 * SPA syncing routines
6275 * ==========================================================================
6279 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6282 bpobj_enqueue(bpo
, bp
, tx
);
6287 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6291 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6297 * Note: this simple function is not inlined to make it easier to dtrace the
6298 * amount of time spent syncing frees.
6301 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6303 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6304 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6305 VERIFY(zio_wait(zio
) == 0);
6309 * Note: this simple function is not inlined to make it easier to dtrace the
6310 * amount of time spent syncing deferred frees.
6313 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6315 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6316 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6317 spa_free_sync_cb
, zio
, tx
), ==, 0);
6318 VERIFY0(zio_wait(zio
));
6322 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6324 char *packed
= NULL
;
6329 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6332 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6333 * information. This avoids the dmu_buf_will_dirty() path and
6334 * saves us a pre-read to get data we don't actually care about.
6336 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6337 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6339 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6341 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6343 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6345 vmem_free(packed
, bufsize
);
6347 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6348 dmu_buf_will_dirty(db
, tx
);
6349 *(uint64_t *)db
->db_data
= nvsize
;
6350 dmu_buf_rele(db
, FTAG
);
6354 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6355 const char *config
, const char *entry
)
6365 * Update the MOS nvlist describing the list of available devices.
6366 * spa_validate_aux() will have already made sure this nvlist is
6367 * valid and the vdevs are labeled appropriately.
6369 if (sav
->sav_object
== 0) {
6370 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6371 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6372 sizeof (uint64_t), tx
);
6373 VERIFY(zap_update(spa
->spa_meta_objset
,
6374 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6375 &sav
->sav_object
, tx
) == 0);
6378 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6379 if (sav
->sav_count
== 0) {
6380 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6382 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6383 for (i
= 0; i
< sav
->sav_count
; i
++)
6384 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6385 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6386 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6387 sav
->sav_count
) == 0);
6388 for (i
= 0; i
< sav
->sav_count
; i
++)
6389 nvlist_free(list
[i
]);
6390 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6393 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6394 nvlist_free(nvroot
);
6396 sav
->sav_sync
= B_FALSE
;
6400 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6401 * The all-vdev ZAP must be empty.
6404 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6406 spa_t
*spa
= vd
->vdev_spa
;
6409 if (vd
->vdev_top_zap
!= 0) {
6410 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6411 vd
->vdev_top_zap
, tx
));
6413 if (vd
->vdev_leaf_zap
!= 0) {
6414 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6415 vd
->vdev_leaf_zap
, tx
));
6417 for (i
= 0; i
< vd
->vdev_children
; i
++) {
6418 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6423 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6428 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6429 * its config may not be dirty but we still need to build per-vdev ZAPs.
6430 * Similarly, if the pool is being assembled (e.g. after a split), we
6431 * need to rebuild the AVZ although the config may not be dirty.
6433 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6434 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6437 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6439 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6440 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6441 spa
->spa_all_vdev_zaps
!= 0);
6443 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6447 /* Make and build the new AVZ */
6448 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6449 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6450 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6452 /* Diff old AVZ with new one */
6453 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6454 spa
->spa_all_vdev_zaps
);
6455 zap_cursor_retrieve(&zc
, &za
) == 0;
6456 zap_cursor_advance(&zc
)) {
6457 uint64_t vdzap
= za
.za_first_integer
;
6458 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6461 * ZAP is listed in old AVZ but not in new one;
6464 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6469 zap_cursor_fini(&zc
);
6471 /* Destroy the old AVZ */
6472 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6473 spa
->spa_all_vdev_zaps
, tx
));
6475 /* Replace the old AVZ in the dir obj with the new one */
6476 VERIFY0(zap_update(spa
->spa_meta_objset
,
6477 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6478 sizeof (new_avz
), 1, &new_avz
, tx
));
6480 spa
->spa_all_vdev_zaps
= new_avz
;
6481 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6485 /* Walk through the AVZ and destroy all listed ZAPs */
6486 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6487 spa
->spa_all_vdev_zaps
);
6488 zap_cursor_retrieve(&zc
, &za
) == 0;
6489 zap_cursor_advance(&zc
)) {
6490 uint64_t zap
= za
.za_first_integer
;
6491 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6494 zap_cursor_fini(&zc
);
6496 /* Destroy and unlink the AVZ itself */
6497 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6498 spa
->spa_all_vdev_zaps
, tx
));
6499 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6500 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6501 spa
->spa_all_vdev_zaps
= 0;
6504 if (spa
->spa_all_vdev_zaps
== 0) {
6505 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6506 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6507 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6509 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6511 /* Create ZAPs for vdevs that don't have them. */
6512 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6514 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6515 dmu_tx_get_txg(tx
), B_FALSE
);
6518 * If we're upgrading the spa version then make sure that
6519 * the config object gets updated with the correct version.
6521 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6522 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6523 spa
->spa_uberblock
.ub_version
);
6525 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6527 nvlist_free(spa
->spa_config_syncing
);
6528 spa
->spa_config_syncing
= config
;
6530 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6534 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6536 uint64_t *versionp
= arg
;
6537 uint64_t version
= *versionp
;
6538 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6541 * Setting the version is special cased when first creating the pool.
6543 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6545 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6546 ASSERT(version
>= spa_version(spa
));
6548 spa
->spa_uberblock
.ub_version
= version
;
6549 vdev_config_dirty(spa
->spa_root_vdev
);
6550 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6554 * Set zpool properties.
6557 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6559 nvlist_t
*nvp
= arg
;
6560 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6561 objset_t
*mos
= spa
->spa_meta_objset
;
6562 nvpair_t
*elem
= NULL
;
6564 mutex_enter(&spa
->spa_props_lock
);
6566 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6568 char *strval
, *fname
;
6570 const char *propname
;
6571 zprop_type_t proptype
;
6574 prop
= zpool_name_to_prop(nvpair_name(elem
));
6575 switch ((int)prop
) {
6578 * We checked this earlier in spa_prop_validate().
6580 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6582 fname
= strchr(nvpair_name(elem
), '@') + 1;
6583 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6585 spa_feature_enable(spa
, fid
, tx
);
6586 spa_history_log_internal(spa
, "set", tx
,
6587 "%s=enabled", nvpair_name(elem
));
6590 case ZPOOL_PROP_VERSION
:
6591 intval
= fnvpair_value_uint64(elem
);
6593 * The version is synced separately before other
6594 * properties and should be correct by now.
6596 ASSERT3U(spa_version(spa
), >=, intval
);
6599 case ZPOOL_PROP_ALTROOT
:
6601 * 'altroot' is a non-persistent property. It should
6602 * have been set temporarily at creation or import time.
6604 ASSERT(spa
->spa_root
!= NULL
);
6607 case ZPOOL_PROP_READONLY
:
6608 case ZPOOL_PROP_CACHEFILE
:
6610 * 'readonly' and 'cachefile' are also non-persisitent
6614 case ZPOOL_PROP_COMMENT
:
6615 strval
= fnvpair_value_string(elem
);
6616 if (spa
->spa_comment
!= NULL
)
6617 spa_strfree(spa
->spa_comment
);
6618 spa
->spa_comment
= spa_strdup(strval
);
6620 * We need to dirty the configuration on all the vdevs
6621 * so that their labels get updated. It's unnecessary
6622 * to do this for pool creation since the vdev's
6623 * configuration has already been dirtied.
6625 if (tx
->tx_txg
!= TXG_INITIAL
)
6626 vdev_config_dirty(spa
->spa_root_vdev
);
6627 spa_history_log_internal(spa
, "set", tx
,
6628 "%s=%s", nvpair_name(elem
), strval
);
6632 * Set pool property values in the poolprops mos object.
6634 if (spa
->spa_pool_props_object
== 0) {
6635 spa
->spa_pool_props_object
=
6636 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6637 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6641 /* normalize the property name */
6642 propname
= zpool_prop_to_name(prop
);
6643 proptype
= zpool_prop_get_type(prop
);
6645 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6646 ASSERT(proptype
== PROP_TYPE_STRING
);
6647 strval
= fnvpair_value_string(elem
);
6648 VERIFY0(zap_update(mos
,
6649 spa
->spa_pool_props_object
, propname
,
6650 1, strlen(strval
) + 1, strval
, tx
));
6651 spa_history_log_internal(spa
, "set", tx
,
6652 "%s=%s", nvpair_name(elem
), strval
);
6653 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6654 intval
= fnvpair_value_uint64(elem
);
6656 if (proptype
== PROP_TYPE_INDEX
) {
6658 VERIFY0(zpool_prop_index_to_string(
6659 prop
, intval
, &unused
));
6661 VERIFY0(zap_update(mos
,
6662 spa
->spa_pool_props_object
, propname
,
6663 8, 1, &intval
, tx
));
6664 spa_history_log_internal(spa
, "set", tx
,
6665 "%s=%lld", nvpair_name(elem
), intval
);
6667 ASSERT(0); /* not allowed */
6671 case ZPOOL_PROP_DELEGATION
:
6672 spa
->spa_delegation
= intval
;
6674 case ZPOOL_PROP_BOOTFS
:
6675 spa
->spa_bootfs
= intval
;
6677 case ZPOOL_PROP_FAILUREMODE
:
6678 spa
->spa_failmode
= intval
;
6680 case ZPOOL_PROP_AUTOEXPAND
:
6681 spa
->spa_autoexpand
= intval
;
6682 if (tx
->tx_txg
!= TXG_INITIAL
)
6683 spa_async_request(spa
,
6684 SPA_ASYNC_AUTOEXPAND
);
6686 case ZPOOL_PROP_MULTIHOST
:
6687 spa
->spa_multihost
= intval
;
6689 case ZPOOL_PROP_DEDUPDITTO
:
6690 spa
->spa_dedup_ditto
= intval
;
6699 mutex_exit(&spa
->spa_props_lock
);
6703 * Perform one-time upgrade on-disk changes. spa_version() does not
6704 * reflect the new version this txg, so there must be no changes this
6705 * txg to anything that the upgrade code depends on after it executes.
6706 * Therefore this must be called after dsl_pool_sync() does the sync
6710 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6712 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6714 ASSERT(spa
->spa_sync_pass
== 1);
6716 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6718 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6719 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6720 dsl_pool_create_origin(dp
, tx
);
6722 /* Keeping the origin open increases spa_minref */
6723 spa
->spa_minref
+= 3;
6726 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6727 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6728 dsl_pool_upgrade_clones(dp
, tx
);
6731 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6732 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6733 dsl_pool_upgrade_dir_clones(dp
, tx
);
6735 /* Keeping the freedir open increases spa_minref */
6736 spa
->spa_minref
+= 3;
6739 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6740 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6741 spa_feature_create_zap_objects(spa
, tx
);
6745 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6746 * when possibility to use lz4 compression for metadata was added
6747 * Old pools that have this feature enabled must be upgraded to have
6748 * this feature active
6750 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6751 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6752 SPA_FEATURE_LZ4_COMPRESS
);
6753 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6754 SPA_FEATURE_LZ4_COMPRESS
);
6756 if (lz4_en
&& !lz4_ac
)
6757 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6761 * If we haven't written the salt, do so now. Note that the
6762 * feature may not be activated yet, but that's fine since
6763 * the presence of this ZAP entry is backwards compatible.
6765 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
6766 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
6767 VERIFY0(zap_add(spa
->spa_meta_objset
,
6768 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
6769 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
6770 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
6773 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6777 * Sync the specified transaction group. New blocks may be dirtied as
6778 * part of the process, so we iterate until it converges.
6781 spa_sync(spa_t
*spa
, uint64_t txg
)
6783 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6784 objset_t
*mos
= spa
->spa_meta_objset
;
6785 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6786 metaslab_class_t
*mc
;
6787 vdev_t
*rvd
= spa
->spa_root_vdev
;
6791 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
6792 zfs_vdev_queue_depth_pct
/ 100;
6793 uint64_t queue_depth_total
;
6796 VERIFY(spa_writeable(spa
));
6799 * Lock out configuration changes.
6801 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6803 spa
->spa_syncing_txg
= txg
;
6804 spa
->spa_sync_pass
= 0;
6806 mutex_enter(&spa
->spa_alloc_lock
);
6807 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
6808 mutex_exit(&spa
->spa_alloc_lock
);
6811 * If there are any pending vdev state changes, convert them
6812 * into config changes that go out with this transaction group.
6814 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6815 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6817 * We need the write lock here because, for aux vdevs,
6818 * calling vdev_config_dirty() modifies sav_config.
6819 * This is ugly and will become unnecessary when we
6820 * eliminate the aux vdev wart by integrating all vdevs
6821 * into the root vdev tree.
6823 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6824 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6825 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6826 vdev_state_clean(vd
);
6827 vdev_config_dirty(vd
);
6829 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6830 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6832 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6834 tx
= dmu_tx_create_assigned(dp
, txg
);
6836 spa
->spa_sync_starttime
= gethrtime();
6837 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6838 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
6839 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6840 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6843 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6844 * set spa_deflate if we have no raid-z vdevs.
6846 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6847 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6850 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6851 vd
= rvd
->vdev_child
[i
];
6852 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6855 if (i
== rvd
->vdev_children
) {
6856 spa
->spa_deflate
= TRUE
;
6857 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6858 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6859 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6864 * Set the top-level vdev's max queue depth. Evaluate each
6865 * top-level's async write queue depth in case it changed.
6866 * The max queue depth will not change in the middle of syncing
6869 queue_depth_total
= 0;
6870 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
6871 vdev_t
*tvd
= rvd
->vdev_child
[c
];
6872 metaslab_group_t
*mg
= tvd
->vdev_mg
;
6874 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
6875 !metaslab_group_initialized(mg
))
6879 * It is safe to do a lock-free check here because only async
6880 * allocations look at mg_max_alloc_queue_depth, and async
6881 * allocations all happen from spa_sync().
6883 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
6884 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
6885 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
6887 mc
= spa_normal_class(spa
);
6888 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
6889 mc
->mc_alloc_max_slots
= queue_depth_total
;
6890 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
6892 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
6893 max_queue_depth
* rvd
->vdev_children
);
6896 * Iterate to convergence.
6899 int pass
= ++spa
->spa_sync_pass
;
6901 spa_sync_config_object(spa
, tx
);
6902 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6903 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6904 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6905 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6906 spa_errlog_sync(spa
, txg
);
6907 dsl_pool_sync(dp
, txg
);
6909 if (pass
< zfs_sync_pass_deferred_free
) {
6910 spa_sync_frees(spa
, free_bpl
, tx
);
6913 * We can not defer frees in pass 1, because
6914 * we sync the deferred frees later in pass 1.
6916 ASSERT3U(pass
, >, 1);
6917 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6918 &spa
->spa_deferred_bpobj
, tx
);
6922 dsl_scan_sync(dp
, tx
);
6924 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6928 spa_sync_upgrades(spa
, tx
);
6930 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
6932 * Note: We need to check if the MOS is dirty
6933 * because we could have marked the MOS dirty
6934 * without updating the uberblock (e.g. if we
6935 * have sync tasks but no dirty user data). We
6936 * need to check the uberblock's rootbp because
6937 * it is updated if we have synced out dirty
6938 * data (though in this case the MOS will most
6939 * likely also be dirty due to second order
6940 * effects, we don't want to rely on that here).
6942 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
6943 !dmu_objset_is_dirty(mos
, txg
)) {
6945 * Nothing changed on the first pass,
6946 * therefore this TXG is a no-op. Avoid
6947 * syncing deferred frees, so that we
6948 * can keep this TXG as a no-op.
6950 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
6952 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6953 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
6956 spa_sync_deferred_frees(spa
, tx
);
6959 } while (dmu_objset_is_dirty(mos
, txg
));
6962 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
6964 * Make sure that the number of ZAPs for all the vdevs matches
6965 * the number of ZAPs in the per-vdev ZAP list. This only gets
6966 * called if the config is dirty; otherwise there may be
6967 * outstanding AVZ operations that weren't completed in
6968 * spa_sync_config_object.
6970 uint64_t all_vdev_zap_entry_count
;
6971 ASSERT0(zap_count(spa
->spa_meta_objset
,
6972 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
6973 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
6974 all_vdev_zap_entry_count
);
6979 * Rewrite the vdev configuration (which includes the uberblock)
6980 * to commit the transaction group.
6982 * If there are no dirty vdevs, we sync the uberblock to a few
6983 * random top-level vdevs that are known to be visible in the
6984 * config cache (see spa_vdev_add() for a complete description).
6985 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6989 * We hold SCL_STATE to prevent vdev open/close/etc.
6990 * while we're attempting to write the vdev labels.
6992 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6994 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6995 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6997 int children
= rvd
->vdev_children
;
6998 int c0
= spa_get_random(children
);
7000 for (c
= 0; c
< children
; c
++) {
7001 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
7002 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
7004 svd
[svdcount
++] = vd
;
7005 if (svdcount
== SPA_DVAS_PER_BP
)
7008 error
= vdev_config_sync(svd
, svdcount
, txg
);
7010 error
= vdev_config_sync(rvd
->vdev_child
,
7011 rvd
->vdev_children
, txg
);
7015 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
7017 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7021 zio_suspend(spa
, NULL
);
7022 zio_resume_wait(spa
);
7026 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
7027 spa
->spa_deadman_tqid
= 0;
7030 * Clear the dirty config list.
7032 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
7033 vdev_config_clean(vd
);
7036 * Now that the new config has synced transactionally,
7037 * let it become visible to the config cache.
7039 if (spa
->spa_config_syncing
!= NULL
) {
7040 spa_config_set(spa
, spa
->spa_config_syncing
);
7041 spa
->spa_config_txg
= txg
;
7042 spa
->spa_config_syncing
= NULL
;
7045 dsl_pool_sync_done(dp
, txg
);
7047 mutex_enter(&spa
->spa_alloc_lock
);
7048 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7049 mutex_exit(&spa
->spa_alloc_lock
);
7052 * Update usable space statistics.
7054 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
7055 vdev_sync_done(vd
, txg
);
7057 spa_update_dspace(spa
);
7060 * It had better be the case that we didn't dirty anything
7061 * since vdev_config_sync().
7063 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7064 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7065 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7067 spa
->spa_sync_pass
= 0;
7070 * Update the last synced uberblock here. We want to do this at
7071 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7072 * will be guaranteed that all the processing associated with
7073 * that txg has been completed.
7075 spa
->spa_ubsync
= spa
->spa_uberblock
;
7076 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7078 spa_handle_ignored_writes(spa
);
7081 * If any async tasks have been requested, kick them off.
7083 spa_async_dispatch(spa
);
7087 * Sync all pools. We don't want to hold the namespace lock across these
7088 * operations, so we take a reference on the spa_t and drop the lock during the
7092 spa_sync_allpools(void)
7095 mutex_enter(&spa_namespace_lock
);
7096 while ((spa
= spa_next(spa
)) != NULL
) {
7097 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7098 !spa_writeable(spa
) || spa_suspended(spa
))
7100 spa_open_ref(spa
, FTAG
);
7101 mutex_exit(&spa_namespace_lock
);
7102 txg_wait_synced(spa_get_dsl(spa
), 0);
7103 mutex_enter(&spa_namespace_lock
);
7104 spa_close(spa
, FTAG
);
7106 mutex_exit(&spa_namespace_lock
);
7110 * ==========================================================================
7111 * Miscellaneous routines
7112 * ==========================================================================
7116 * Remove all pools in the system.
7124 * Remove all cached state. All pools should be closed now,
7125 * so every spa in the AVL tree should be unreferenced.
7127 mutex_enter(&spa_namespace_lock
);
7128 while ((spa
= spa_next(NULL
)) != NULL
) {
7130 * Stop async tasks. The async thread may need to detach
7131 * a device that's been replaced, which requires grabbing
7132 * spa_namespace_lock, so we must drop it here.
7134 spa_open_ref(spa
, FTAG
);
7135 mutex_exit(&spa_namespace_lock
);
7136 spa_async_suspend(spa
);
7137 mutex_enter(&spa_namespace_lock
);
7138 spa_close(spa
, FTAG
);
7140 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7142 spa_deactivate(spa
);
7146 mutex_exit(&spa_namespace_lock
);
7150 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7155 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7159 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7160 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7161 if (vd
->vdev_guid
== guid
)
7165 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7166 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7167 if (vd
->vdev_guid
== guid
)
7176 spa_upgrade(spa_t
*spa
, uint64_t version
)
7178 ASSERT(spa_writeable(spa
));
7180 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7183 * This should only be called for a non-faulted pool, and since a
7184 * future version would result in an unopenable pool, this shouldn't be
7187 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7188 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7190 spa
->spa_uberblock
.ub_version
= version
;
7191 vdev_config_dirty(spa
->spa_root_vdev
);
7193 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7195 txg_wait_synced(spa_get_dsl(spa
), 0);
7199 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7203 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7205 for (i
= 0; i
< sav
->sav_count
; i
++)
7206 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7209 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7210 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7211 &spareguid
) == 0 && spareguid
== guid
)
7219 * Check if a pool has an active shared spare device.
7220 * Note: reference count of an active spare is 2, as a spare and as a replace
7223 spa_has_active_shared_spare(spa_t
*spa
)
7227 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7229 for (i
= 0; i
< sav
->sav_count
; i
++) {
7230 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7231 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7240 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7242 sysevent_t
*ev
= NULL
;
7246 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
7248 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
7249 ev
->resource
= resource
;
7256 spa_event_post(sysevent_t
*ev
)
7260 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
7261 kmem_free(ev
, sizeof (*ev
));
7267 * Post a zevent corresponding to the given sysevent. The 'name' must be one
7268 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
7269 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7270 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7271 * or zdb as real changes.
7274 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7276 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
7279 #if defined(_KERNEL) && defined(HAVE_SPL)
7280 /* state manipulation functions */
7281 EXPORT_SYMBOL(spa_open
);
7282 EXPORT_SYMBOL(spa_open_rewind
);
7283 EXPORT_SYMBOL(spa_get_stats
);
7284 EXPORT_SYMBOL(spa_create
);
7285 EXPORT_SYMBOL(spa_import
);
7286 EXPORT_SYMBOL(spa_tryimport
);
7287 EXPORT_SYMBOL(spa_destroy
);
7288 EXPORT_SYMBOL(spa_export
);
7289 EXPORT_SYMBOL(spa_reset
);
7290 EXPORT_SYMBOL(spa_async_request
);
7291 EXPORT_SYMBOL(spa_async_suspend
);
7292 EXPORT_SYMBOL(spa_async_resume
);
7293 EXPORT_SYMBOL(spa_inject_addref
);
7294 EXPORT_SYMBOL(spa_inject_delref
);
7295 EXPORT_SYMBOL(spa_scan_stat_init
);
7296 EXPORT_SYMBOL(spa_scan_get_stats
);
7298 /* device maniion */
7299 EXPORT_SYMBOL(spa_vdev_add
);
7300 EXPORT_SYMBOL(spa_vdev_attach
);
7301 EXPORT_SYMBOL(spa_vdev_detach
);
7302 EXPORT_SYMBOL(spa_vdev_remove
);
7303 EXPORT_SYMBOL(spa_vdev_setpath
);
7304 EXPORT_SYMBOL(spa_vdev_setfru
);
7305 EXPORT_SYMBOL(spa_vdev_split_mirror
);
7307 /* spare statech is global across all pools) */
7308 EXPORT_SYMBOL(spa_spare_add
);
7309 EXPORT_SYMBOL(spa_spare_remove
);
7310 EXPORT_SYMBOL(spa_spare_exists
);
7311 EXPORT_SYMBOL(spa_spare_activate
);
7313 /* L2ARC statech is global across all pools) */
7314 EXPORT_SYMBOL(spa_l2cache_add
);
7315 EXPORT_SYMBOL(spa_l2cache_remove
);
7316 EXPORT_SYMBOL(spa_l2cache_exists
);
7317 EXPORT_SYMBOL(spa_l2cache_activate
);
7318 EXPORT_SYMBOL(spa_l2cache_drop
);
7321 EXPORT_SYMBOL(spa_scan
);
7322 EXPORT_SYMBOL(spa_scan_stop
);
7325 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
7326 EXPORT_SYMBOL(spa_sync_allpools
);
7329 EXPORT_SYMBOL(spa_prop_set
);
7330 EXPORT_SYMBOL(spa_prop_get
);
7331 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
7333 /* asynchronous event notification */
7334 EXPORT_SYMBOL(spa_event_notify
);
7337 #if defined(_KERNEL) && defined(HAVE_SPL)
7338 module_param(spa_load_verify_maxinflight
, int, 0644);
7339 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
7340 "Max concurrent traversal I/Os while verifying pool during import -X");
7342 module_param(spa_load_verify_metadata
, int, 0644);
7343 MODULE_PARM_DESC(spa_load_verify_metadata
,
7344 "Set to traverse metadata on pool import");
7346 module_param(spa_load_verify_data
, int, 0644);
7347 MODULE_PARM_DESC(spa_load_verify_data
,
7348 "Set to traverse data on pool import");
7351 module_param(zio_taskq_batch_pct
, uint
, 0444);
7352 MODULE_PARM_DESC(zio_taskq_batch_pct
,
7353 "Percentage of CPUs to run an IO worker thread");