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
);
1031 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1033 spa_thread(void *arg
)
1035 callb_cpr_t cprinfo
;
1038 user_t
*pu
= PTOU(curproc
);
1040 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1043 ASSERT(curproc
!= &p0
);
1044 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1045 "zpool-%s", spa
->spa_name
);
1046 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1048 /* bind this thread to the requested psrset */
1049 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1051 mutex_enter(&cpu_lock
);
1052 mutex_enter(&pidlock
);
1053 mutex_enter(&curproc
->p_lock
);
1055 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1056 0, NULL
, NULL
) == 0) {
1057 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1060 "Couldn't bind process for zfs pool \"%s\" to "
1061 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1064 mutex_exit(&curproc
->p_lock
);
1065 mutex_exit(&pidlock
);
1066 mutex_exit(&cpu_lock
);
1070 if (zio_taskq_sysdc
) {
1071 sysdc_thread_enter(curthread
, 100, 0);
1074 spa
->spa_proc
= curproc
;
1075 spa
->spa_did
= curthread
->t_did
;
1077 spa_create_zio_taskqs(spa
);
1079 mutex_enter(&spa
->spa_proc_lock
);
1080 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1082 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1083 cv_broadcast(&spa
->spa_proc_cv
);
1085 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1086 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1087 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1088 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1090 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1091 spa
->spa_proc_state
= SPA_PROC_GONE
;
1092 spa
->spa_proc
= &p0
;
1093 cv_broadcast(&spa
->spa_proc_cv
);
1094 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1096 mutex_enter(&curproc
->p_lock
);
1102 * Activate an uninitialized pool.
1105 spa_activate(spa_t
*spa
, int mode
)
1107 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1109 spa
->spa_state
= POOL_STATE_ACTIVE
;
1110 spa
->spa_mode
= mode
;
1112 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1113 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1115 /* Try to create a covering process */
1116 mutex_enter(&spa
->spa_proc_lock
);
1117 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1118 ASSERT(spa
->spa_proc
== &p0
);
1121 #ifdef HAVE_SPA_THREAD
1122 /* Only create a process if we're going to be around a while. */
1123 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1124 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1126 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1127 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1128 cv_wait(&spa
->spa_proc_cv
,
1129 &spa
->spa_proc_lock
);
1131 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1132 ASSERT(spa
->spa_proc
!= &p0
);
1133 ASSERT(spa
->spa_did
!= 0);
1137 "Couldn't create process for zfs pool \"%s\"\n",
1142 #endif /* HAVE_SPA_THREAD */
1143 mutex_exit(&spa
->spa_proc_lock
);
1145 /* If we didn't create a process, we need to create our taskqs. */
1146 if (spa
->spa_proc
== &p0
) {
1147 spa_create_zio_taskqs(spa
);
1150 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1151 offsetof(vdev_t
, vdev_config_dirty_node
));
1152 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1153 offsetof(objset_t
, os_evicting_node
));
1154 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1155 offsetof(vdev_t
, vdev_state_dirty_node
));
1157 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1158 offsetof(struct vdev
, vdev_txg_node
));
1160 avl_create(&spa
->spa_errlist_scrub
,
1161 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1162 offsetof(spa_error_entry_t
, se_avl
));
1163 avl_create(&spa
->spa_errlist_last
,
1164 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1165 offsetof(spa_error_entry_t
, se_avl
));
1168 * This taskq is used to perform zvol-minor-related tasks
1169 * asynchronously. This has several advantages, including easy
1170 * resolution of various deadlocks (zfsonlinux bug #3681).
1172 * The taskq must be single threaded to ensure tasks are always
1173 * processed in the order in which they were dispatched.
1175 * A taskq per pool allows one to keep the pools independent.
1176 * This way if one pool is suspended, it will not impact another.
1178 * The preferred location to dispatch a zvol minor task is a sync
1179 * task. In this context, there is easy access to the spa_t and minimal
1180 * error handling is required because the sync task must succeed.
1182 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1186 * The taskq to upgrade datasets in this pool. Currently used by
1187 * feature SPA_FEATURE_USEROBJ_ACCOUNTING.
1189 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", boot_ncpus
,
1190 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1194 * Opposite of spa_activate().
1197 spa_deactivate(spa_t
*spa
)
1201 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1202 ASSERT(spa
->spa_dsl_pool
== NULL
);
1203 ASSERT(spa
->spa_root_vdev
== NULL
);
1204 ASSERT(spa
->spa_async_zio_root
== NULL
);
1205 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1207 spa_evicting_os_wait(spa
);
1209 if (spa
->spa_zvol_taskq
) {
1210 taskq_destroy(spa
->spa_zvol_taskq
);
1211 spa
->spa_zvol_taskq
= NULL
;
1214 if (spa
->spa_upgrade_taskq
) {
1215 taskq_destroy(spa
->spa_upgrade_taskq
);
1216 spa
->spa_upgrade_taskq
= NULL
;
1219 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1221 list_destroy(&spa
->spa_config_dirty_list
);
1222 list_destroy(&spa
->spa_evicting_os_list
);
1223 list_destroy(&spa
->spa_state_dirty_list
);
1225 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1227 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1228 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1229 spa_taskqs_fini(spa
, t
, q
);
1233 metaslab_class_destroy(spa
->spa_normal_class
);
1234 spa
->spa_normal_class
= NULL
;
1236 metaslab_class_destroy(spa
->spa_log_class
);
1237 spa
->spa_log_class
= NULL
;
1240 * If this was part of an import or the open otherwise failed, we may
1241 * still have errors left in the queues. Empty them just in case.
1243 spa_errlog_drain(spa
);
1245 avl_destroy(&spa
->spa_errlist_scrub
);
1246 avl_destroy(&spa
->spa_errlist_last
);
1248 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1250 mutex_enter(&spa
->spa_proc_lock
);
1251 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1252 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1253 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1254 cv_broadcast(&spa
->spa_proc_cv
);
1255 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1256 ASSERT(spa
->spa_proc
!= &p0
);
1257 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1259 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1260 spa
->spa_proc_state
= SPA_PROC_NONE
;
1262 ASSERT(spa
->spa_proc
== &p0
);
1263 mutex_exit(&spa
->spa_proc_lock
);
1266 * We want to make sure spa_thread() has actually exited the ZFS
1267 * module, so that the module can't be unloaded out from underneath
1270 if (spa
->spa_did
!= 0) {
1271 thread_join(spa
->spa_did
);
1277 * Verify a pool configuration, and construct the vdev tree appropriately. This
1278 * will create all the necessary vdevs in the appropriate layout, with each vdev
1279 * in the CLOSED state. This will prep the pool before open/creation/import.
1280 * All vdev validation is done by the vdev_alloc() routine.
1283 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1284 uint_t id
, int atype
)
1291 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1294 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1297 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1300 if (error
== ENOENT
)
1306 return (SET_ERROR(EINVAL
));
1309 for (c
= 0; c
< children
; c
++) {
1311 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1319 ASSERT(*vdp
!= NULL
);
1325 * Opposite of spa_load().
1328 spa_unload(spa_t
*spa
)
1332 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1337 spa_async_suspend(spa
);
1342 if (spa
->spa_sync_on
) {
1343 txg_sync_stop(spa
->spa_dsl_pool
);
1344 spa
->spa_sync_on
= B_FALSE
;
1348 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1349 * to call it earlier, before we wait for async i/o to complete.
1350 * This ensures that there is no async metaslab prefetching, by
1351 * calling taskq_wait(mg_taskq).
1353 if (spa
->spa_root_vdev
!= NULL
) {
1354 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1355 for (c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++)
1356 vdev_metaslab_fini(spa
->spa_root_vdev
->vdev_child
[c
]);
1357 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1360 if (spa
->spa_mmp
.mmp_thread
)
1361 mmp_thread_stop(spa
);
1364 * Wait for any outstanding async I/O to complete.
1366 if (spa
->spa_async_zio_root
!= NULL
) {
1367 for (i
= 0; i
< max_ncpus
; i
++)
1368 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1369 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1370 spa
->spa_async_zio_root
= NULL
;
1373 bpobj_close(&spa
->spa_deferred_bpobj
);
1375 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1380 if (spa
->spa_root_vdev
)
1381 vdev_free(spa
->spa_root_vdev
);
1382 ASSERT(spa
->spa_root_vdev
== NULL
);
1385 * Close the dsl pool.
1387 if (spa
->spa_dsl_pool
) {
1388 dsl_pool_close(spa
->spa_dsl_pool
);
1389 spa
->spa_dsl_pool
= NULL
;
1390 spa
->spa_meta_objset
= NULL
;
1396 * Drop and purge level 2 cache
1398 spa_l2cache_drop(spa
);
1400 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1401 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1402 if (spa
->spa_spares
.sav_vdevs
) {
1403 kmem_free(spa
->spa_spares
.sav_vdevs
,
1404 spa
->spa_spares
.sav_count
* sizeof (void *));
1405 spa
->spa_spares
.sav_vdevs
= NULL
;
1407 if (spa
->spa_spares
.sav_config
) {
1408 nvlist_free(spa
->spa_spares
.sav_config
);
1409 spa
->spa_spares
.sav_config
= NULL
;
1411 spa
->spa_spares
.sav_count
= 0;
1413 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1414 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1415 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1417 if (spa
->spa_l2cache
.sav_vdevs
) {
1418 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1419 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1420 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1422 if (spa
->spa_l2cache
.sav_config
) {
1423 nvlist_free(spa
->spa_l2cache
.sav_config
);
1424 spa
->spa_l2cache
.sav_config
= NULL
;
1426 spa
->spa_l2cache
.sav_count
= 0;
1428 spa
->spa_async_suspended
= 0;
1430 if (spa
->spa_comment
!= NULL
) {
1431 spa_strfree(spa
->spa_comment
);
1432 spa
->spa_comment
= NULL
;
1435 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1439 * Load (or re-load) the current list of vdevs describing the active spares for
1440 * this pool. When this is called, we have some form of basic information in
1441 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1442 * then re-generate a more complete list including status information.
1445 spa_load_spares(spa_t
*spa
)
1452 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1455 * First, close and free any existing spare vdevs.
1457 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1458 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1460 /* Undo the call to spa_activate() below */
1461 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1462 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1463 spa_spare_remove(tvd
);
1468 if (spa
->spa_spares
.sav_vdevs
)
1469 kmem_free(spa
->spa_spares
.sav_vdevs
,
1470 spa
->spa_spares
.sav_count
* sizeof (void *));
1472 if (spa
->spa_spares
.sav_config
== NULL
)
1475 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1476 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1478 spa
->spa_spares
.sav_count
= (int)nspares
;
1479 spa
->spa_spares
.sav_vdevs
= NULL
;
1485 * Construct the array of vdevs, opening them to get status in the
1486 * process. For each spare, there is potentially two different vdev_t
1487 * structures associated with it: one in the list of spares (used only
1488 * for basic validation purposes) and one in the active vdev
1489 * configuration (if it's spared in). During this phase we open and
1490 * validate each vdev on the spare list. If the vdev also exists in the
1491 * active configuration, then we also mark this vdev as an active spare.
1493 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1495 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1496 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1497 VDEV_ALLOC_SPARE
) == 0);
1500 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1502 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1503 B_FALSE
)) != NULL
) {
1504 if (!tvd
->vdev_isspare
)
1508 * We only mark the spare active if we were successfully
1509 * able to load the vdev. Otherwise, importing a pool
1510 * with a bad active spare would result in strange
1511 * behavior, because multiple pool would think the spare
1512 * is actively in use.
1514 * There is a vulnerability here to an equally bizarre
1515 * circumstance, where a dead active spare is later
1516 * brought back to life (onlined or otherwise). Given
1517 * the rarity of this scenario, and the extra complexity
1518 * it adds, we ignore the possibility.
1520 if (!vdev_is_dead(tvd
))
1521 spa_spare_activate(tvd
);
1525 vd
->vdev_aux
= &spa
->spa_spares
;
1527 if (vdev_open(vd
) != 0)
1530 if (vdev_validate_aux(vd
) == 0)
1535 * Recompute the stashed list of spares, with status information
1538 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1539 DATA_TYPE_NVLIST_ARRAY
) == 0);
1541 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1543 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1544 spares
[i
] = vdev_config_generate(spa
,
1545 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1546 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1547 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1548 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1549 nvlist_free(spares
[i
]);
1550 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1554 * Load (or re-load) the current list of vdevs describing the active l2cache for
1555 * this pool. When this is called, we have some form of basic information in
1556 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1557 * then re-generate a more complete list including status information.
1558 * Devices which are already active have their details maintained, and are
1562 spa_load_l2cache(spa_t
*spa
)
1566 int i
, j
, oldnvdevs
;
1568 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1569 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1571 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1573 oldvdevs
= sav
->sav_vdevs
;
1574 oldnvdevs
= sav
->sav_count
;
1575 sav
->sav_vdevs
= NULL
;
1578 if (sav
->sav_config
== NULL
) {
1584 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1585 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1586 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1589 * Process new nvlist of vdevs.
1591 for (i
= 0; i
< nl2cache
; i
++) {
1592 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1596 for (j
= 0; j
< oldnvdevs
; j
++) {
1598 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1600 * Retain previous vdev for add/remove ops.
1608 if (newvdevs
[i
] == NULL
) {
1612 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1613 VDEV_ALLOC_L2CACHE
) == 0);
1618 * Commit this vdev as an l2cache device,
1619 * even if it fails to open.
1621 spa_l2cache_add(vd
);
1626 spa_l2cache_activate(vd
);
1628 if (vdev_open(vd
) != 0)
1631 (void) vdev_validate_aux(vd
);
1633 if (!vdev_is_dead(vd
))
1634 l2arc_add_vdev(spa
, vd
);
1638 sav
->sav_vdevs
= newvdevs
;
1639 sav
->sav_count
= (int)nl2cache
;
1642 * Recompute the stashed list of l2cache devices, with status
1643 * information this time.
1645 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1646 DATA_TYPE_NVLIST_ARRAY
) == 0);
1648 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1649 for (i
= 0; i
< sav
->sav_count
; i
++)
1650 l2cache
[i
] = vdev_config_generate(spa
,
1651 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1652 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1653 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1657 * Purge vdevs that were dropped
1659 for (i
= 0; i
< oldnvdevs
; i
++) {
1664 ASSERT(vd
->vdev_isl2cache
);
1666 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1667 pool
!= 0ULL && l2arc_vdev_present(vd
))
1668 l2arc_remove_vdev(vd
);
1669 vdev_clear_stats(vd
);
1675 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1677 for (i
= 0; i
< sav
->sav_count
; i
++)
1678 nvlist_free(l2cache
[i
]);
1680 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1684 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1687 char *packed
= NULL
;
1692 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1696 nvsize
= *(uint64_t *)db
->db_data
;
1697 dmu_buf_rele(db
, FTAG
);
1699 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1700 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1703 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1704 vmem_free(packed
, nvsize
);
1710 * Checks to see if the given vdev could not be opened, in which case we post a
1711 * sysevent to notify the autoreplace code that the device has been removed.
1714 spa_check_removed(vdev_t
*vd
)
1718 for (c
= 0; c
< vd
->vdev_children
; c
++)
1719 spa_check_removed(vd
->vdev_child
[c
]);
1721 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1723 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1724 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
1729 spa_config_valid_zaps(vdev_t
*vd
, vdev_t
*mvd
)
1733 ASSERT3U(vd
->vdev_children
, ==, mvd
->vdev_children
);
1735 vd
->vdev_top_zap
= mvd
->vdev_top_zap
;
1736 vd
->vdev_leaf_zap
= mvd
->vdev_leaf_zap
;
1738 for (i
= 0; i
< vd
->vdev_children
; i
++) {
1739 spa_config_valid_zaps(vd
->vdev_child
[i
], mvd
->vdev_child
[i
]);
1744 * Validate the current config against the MOS config
1747 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1749 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1753 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1755 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1756 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1758 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1761 * If we're doing a normal import, then build up any additional
1762 * diagnostic information about missing devices in this config.
1763 * We'll pass this up to the user for further processing.
1765 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1766 nvlist_t
**child
, *nv
;
1769 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
1771 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1773 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1774 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1775 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1777 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1778 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1780 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1785 VERIFY(nvlist_add_nvlist_array(nv
,
1786 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1787 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1788 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1790 for (i
= 0; i
< idx
; i
++)
1791 nvlist_free(child
[i
]);
1794 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1798 * Compare the root vdev tree with the information we have
1799 * from the MOS config (mrvd). Check each top-level vdev
1800 * with the corresponding MOS config top-level (mtvd).
1802 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1803 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1804 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1807 * Resolve any "missing" vdevs in the current configuration.
1808 * If we find that the MOS config has more accurate information
1809 * about the top-level vdev then use that vdev instead.
1811 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1812 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1814 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1818 * Device specific actions.
1820 if (mtvd
->vdev_islog
) {
1821 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1824 * XXX - once we have 'readonly' pool
1825 * support we should be able to handle
1826 * missing data devices by transitioning
1827 * the pool to readonly.
1833 * Swap the missing vdev with the data we were
1834 * able to obtain from the MOS config.
1836 vdev_remove_child(rvd
, tvd
);
1837 vdev_remove_child(mrvd
, mtvd
);
1839 vdev_add_child(rvd
, mtvd
);
1840 vdev_add_child(mrvd
, tvd
);
1842 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1844 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1848 if (mtvd
->vdev_islog
) {
1850 * Load the slog device's state from the MOS
1851 * config since it's possible that the label
1852 * does not contain the most up-to-date
1855 vdev_load_log_state(tvd
, mtvd
);
1860 * Per-vdev ZAP info is stored exclusively in the MOS.
1862 spa_config_valid_zaps(tvd
, mtvd
);
1867 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1870 * Ensure we were able to validate the config.
1872 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1876 * Check for missing log devices
1879 spa_check_logs(spa_t
*spa
)
1881 boolean_t rv
= B_FALSE
;
1882 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1884 switch (spa
->spa_log_state
) {
1887 case SPA_LOG_MISSING
:
1888 /* need to recheck in case slog has been restored */
1889 case SPA_LOG_UNKNOWN
:
1890 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1891 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1893 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1900 spa_passivate_log(spa_t
*spa
)
1902 vdev_t
*rvd
= spa
->spa_root_vdev
;
1903 boolean_t slog_found
= B_FALSE
;
1906 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1908 if (!spa_has_slogs(spa
))
1911 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1912 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1913 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1915 if (tvd
->vdev_islog
) {
1916 metaslab_group_passivate(mg
);
1917 slog_found
= B_TRUE
;
1921 return (slog_found
);
1925 spa_activate_log(spa_t
*spa
)
1927 vdev_t
*rvd
= spa
->spa_root_vdev
;
1930 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1932 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1933 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1934 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1936 if (tvd
->vdev_islog
)
1937 metaslab_group_activate(mg
);
1942 spa_offline_log(spa_t
*spa
)
1946 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1947 NULL
, DS_FIND_CHILDREN
);
1950 * We successfully offlined the log device, sync out the
1951 * current txg so that the "stubby" block can be removed
1954 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1960 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1964 for (i
= 0; i
< sav
->sav_count
; i
++)
1965 spa_check_removed(sav
->sav_vdevs
[i
]);
1969 spa_claim_notify(zio_t
*zio
)
1971 spa_t
*spa
= zio
->io_spa
;
1976 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1977 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1978 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1979 mutex_exit(&spa
->spa_props_lock
);
1982 typedef struct spa_load_error
{
1983 uint64_t sle_meta_count
;
1984 uint64_t sle_data_count
;
1988 spa_load_verify_done(zio_t
*zio
)
1990 blkptr_t
*bp
= zio
->io_bp
;
1991 spa_load_error_t
*sle
= zio
->io_private
;
1992 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1993 int error
= zio
->io_error
;
1994 spa_t
*spa
= zio
->io_spa
;
1996 abd_free(zio
->io_abd
);
1998 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1999 type
!= DMU_OT_INTENT_LOG
)
2000 atomic_inc_64(&sle
->sle_meta_count
);
2002 atomic_inc_64(&sle
->sle_data_count
);
2005 mutex_enter(&spa
->spa_scrub_lock
);
2006 spa
->spa_scrub_inflight
--;
2007 cv_broadcast(&spa
->spa_scrub_io_cv
);
2008 mutex_exit(&spa
->spa_scrub_lock
);
2012 * Maximum number of concurrent scrub i/os to create while verifying
2013 * a pool while importing it.
2015 int spa_load_verify_maxinflight
= 10000;
2016 int spa_load_verify_metadata
= B_TRUE
;
2017 int spa_load_verify_data
= B_TRUE
;
2021 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2022 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2027 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
2030 * Note: normally this routine will not be called if
2031 * spa_load_verify_metadata is not set. However, it may be useful
2032 * to manually set the flag after the traversal has begun.
2034 if (!spa_load_verify_metadata
)
2036 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2040 size
= BP_GET_PSIZE(bp
);
2042 mutex_enter(&spa
->spa_scrub_lock
);
2043 while (spa
->spa_scrub_inflight
>= spa_load_verify_maxinflight
)
2044 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2045 spa
->spa_scrub_inflight
++;
2046 mutex_exit(&spa
->spa_scrub_lock
);
2048 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2049 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2050 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2051 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2057 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2059 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2060 return (SET_ERROR(ENAMETOOLONG
));
2066 spa_load_verify(spa_t
*spa
)
2069 spa_load_error_t sle
= { 0 };
2070 zpool_rewind_policy_t policy
;
2071 boolean_t verify_ok
= B_FALSE
;
2074 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
2076 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
2079 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2080 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2081 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2083 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2087 rio
= zio_root(spa
, NULL
, &sle
,
2088 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2090 if (spa_load_verify_metadata
) {
2091 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2092 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
,
2093 spa_load_verify_cb
, rio
);
2096 (void) zio_wait(rio
);
2098 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2099 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2101 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
2102 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
2106 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2107 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2109 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2110 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2111 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2112 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2113 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2114 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2115 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2117 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2121 if (error
!= ENXIO
&& error
!= EIO
)
2122 error
= SET_ERROR(EIO
);
2126 return (verify_ok
? 0 : EIO
);
2130 * Find a value in the pool props object.
2133 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2135 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2136 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2140 * Find a value in the pool directory object.
2143 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2145 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2146 name
, sizeof (uint64_t), 1, val
));
2150 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2152 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2157 * Fix up config after a partly-completed split. This is done with the
2158 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2159 * pool have that entry in their config, but only the splitting one contains
2160 * a list of all the guids of the vdevs that are being split off.
2162 * This function determines what to do with that list: either rejoin
2163 * all the disks to the pool, or complete the splitting process. To attempt
2164 * the rejoin, each disk that is offlined is marked online again, and
2165 * we do a reopen() call. If the vdev label for every disk that was
2166 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2167 * then we call vdev_split() on each disk, and complete the split.
2169 * Otherwise we leave the config alone, with all the vdevs in place in
2170 * the original pool.
2173 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2180 boolean_t attempt_reopen
;
2182 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2185 /* check that the config is complete */
2186 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2187 &glist
, &gcount
) != 0)
2190 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2192 /* attempt to online all the vdevs & validate */
2193 attempt_reopen
= B_TRUE
;
2194 for (i
= 0; i
< gcount
; i
++) {
2195 if (glist
[i
] == 0) /* vdev is hole */
2198 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2199 if (vd
[i
] == NULL
) {
2201 * Don't bother attempting to reopen the disks;
2202 * just do the split.
2204 attempt_reopen
= B_FALSE
;
2206 /* attempt to re-online it */
2207 vd
[i
]->vdev_offline
= B_FALSE
;
2211 if (attempt_reopen
) {
2212 vdev_reopen(spa
->spa_root_vdev
);
2214 /* check each device to see what state it's in */
2215 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2216 if (vd
[i
] != NULL
&&
2217 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2224 * If every disk has been moved to the new pool, or if we never
2225 * even attempted to look at them, then we split them off for
2228 if (!attempt_reopen
|| gcount
== extracted
) {
2229 for (i
= 0; i
< gcount
; i
++)
2232 vdev_reopen(spa
->spa_root_vdev
);
2235 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2239 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2240 boolean_t mosconfig
)
2242 nvlist_t
*config
= spa
->spa_config
;
2243 char *ereport
= FM_EREPORT_ZFS_POOL
;
2249 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2250 return (SET_ERROR(EINVAL
));
2252 ASSERT(spa
->spa_comment
== NULL
);
2253 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2254 spa
->spa_comment
= spa_strdup(comment
);
2257 * Versioning wasn't explicitly added to the label until later, so if
2258 * it's not present treat it as the initial version.
2260 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2261 &spa
->spa_ubsync
.ub_version
) != 0)
2262 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2264 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2265 &spa
->spa_config_txg
);
2267 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2268 spa_guid_exists(pool_guid
, 0)) {
2269 error
= SET_ERROR(EEXIST
);
2271 spa
->spa_config_guid
= pool_guid
;
2273 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2275 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2279 nvlist_free(spa
->spa_load_info
);
2280 spa
->spa_load_info
= fnvlist_alloc();
2282 gethrestime(&spa
->spa_loaded_ts
);
2283 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2284 mosconfig
, &ereport
);
2288 * Don't count references from objsets that are already closed
2289 * and are making their way through the eviction process.
2291 spa_evicting_os_wait(spa
);
2292 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2294 if (error
!= EEXIST
) {
2295 spa
->spa_loaded_ts
.tv_sec
= 0;
2296 spa
->spa_loaded_ts
.tv_nsec
= 0;
2298 if (error
!= EBADF
) {
2299 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2302 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2310 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2311 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2312 * spa's per-vdev ZAP list.
2315 vdev_count_verify_zaps(vdev_t
*vd
)
2317 spa_t
*spa
= vd
->vdev_spa
;
2321 if (vd
->vdev_top_zap
!= 0) {
2323 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2324 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2326 if (vd
->vdev_leaf_zap
!= 0) {
2328 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2329 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2332 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2333 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2341 * Determine whether the activity check is required.
2344 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2347 uint64_t hostid
= 0;
2348 uint64_t tryconfig_txg
= 0;
2349 uint64_t tryconfig_timestamp
= 0;
2352 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2353 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2354 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2356 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2357 &tryconfig_timestamp
);
2360 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2361 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_HOSTID
, &hostid
);
2364 * Disable the MMP activity check - This is used by zdb which
2365 * is intended to be used on potentially active pools.
2367 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
2371 * Skip the activity check when the MMP feature is disabled.
2373 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
2376 * If the tryconfig_* values are nonzero, they are the results of an
2377 * earlier tryimport. If they match the uberblock we just found, then
2378 * the pool has not changed and we return false so we do not test a
2381 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
2382 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
)
2386 * Allow the activity check to be skipped when importing the pool
2387 * on the same host which last imported it.
2389 if (hostid
== spa_get_hostid())
2393 * Skip the activity test when the pool was cleanly exported.
2395 if (state
!= POOL_STATE_ACTIVE
)
2402 * Perform the import activity check. If the user canceled the import or
2403 * we detected activity then fail.
2406 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2408 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2409 uint64_t txg
= ub
->ub_txg
;
2410 uint64_t timestamp
= ub
->ub_timestamp
;
2411 uint64_t import_delay
= NANOSEC
;
2412 hrtime_t import_expire
;
2413 nvlist_t
*mmp_label
= NULL
;
2414 vdev_t
*rvd
= spa
->spa_root_vdev
;
2419 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
2420 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2424 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2425 * during the earlier tryimport. If the txg recorded there is 0 then
2426 * the pool is known to be active on another host.
2428 * Otherwise, the pool might be in use on another node. Check for
2429 * changes in the uberblocks on disk if necessary.
2431 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2432 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
2433 ZPOOL_CONFIG_LOAD_INFO
);
2435 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
2436 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
2437 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2438 error
= SET_ERROR(EREMOTEIO
);
2444 * Preferentially use the zfs_multihost_interval from the node which
2445 * last imported the pool. This value is stored in an MMP uberblock as.
2447 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2449 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
)
2450 import_delay
= MAX(import_delay
, import_intervals
*
2451 ub
->ub_mmp_delay
* MAX(vdev_count_leaves(spa
), 1));
2453 /* Apply a floor using the local default values. */
2454 import_delay
= MAX(import_delay
, import_intervals
*
2455 MSEC2NSEC(MAX(zfs_multihost_interval
, MMP_MIN_INTERVAL
)));
2457 /* Add a small random factor in case of simultaneous imports (0-25%) */
2458 import_expire
= gethrtime() + import_delay
+
2459 (import_delay
* spa_get_random(250) / 1000);
2461 while (gethrtime() < import_expire
) {
2462 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2464 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
) {
2465 error
= SET_ERROR(EREMOTEIO
);
2470 nvlist_free(mmp_label
);
2474 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
2476 error
= SET_ERROR(EINTR
);
2484 mutex_destroy(&mtx
);
2488 * If the pool is determined to be active store the status in the
2489 * spa->spa_load_info nvlist. If the remote hostname or hostid are
2490 * available from configuration read from disk store them as well.
2491 * This allows 'zpool import' to generate a more useful message.
2493 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
2494 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2495 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
2497 if (error
== EREMOTEIO
) {
2498 char *hostname
= "<unknown>";
2499 uint64_t hostid
= 0;
2502 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
2503 hostname
= fnvlist_lookup_string(mmp_label
,
2504 ZPOOL_CONFIG_HOSTNAME
);
2505 fnvlist_add_string(spa
->spa_load_info
,
2506 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
2509 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
2510 hostid
= fnvlist_lookup_uint64(mmp_label
,
2511 ZPOOL_CONFIG_HOSTID
);
2512 fnvlist_add_uint64(spa
->spa_load_info
,
2513 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
2517 fnvlist_add_uint64(spa
->spa_load_info
,
2518 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
2519 fnvlist_add_uint64(spa
->spa_load_info
,
2520 ZPOOL_CONFIG_MMP_TXG
, 0);
2522 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
2526 nvlist_free(mmp_label
);
2532 * Load an existing storage pool, using the pool's builtin spa_config as a
2533 * source of configuration information.
2535 __attribute__((always_inline
))
2537 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2538 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2542 nvlist_t
*nvroot
= NULL
;
2545 uberblock_t
*ub
= &spa
->spa_uberblock
;
2546 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2547 int orig_mode
= spa
->spa_mode
;
2550 boolean_t missing_feat_write
= B_FALSE
;
2551 boolean_t activity_check
= B_FALSE
;
2552 nvlist_t
*mos_config
;
2555 * If this is an untrusted config, access the pool in read-only mode.
2556 * This prevents things like resilvering recently removed devices.
2559 spa
->spa_mode
= FREAD
;
2561 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2563 spa
->spa_load_state
= state
;
2565 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2566 return (SET_ERROR(EINVAL
));
2568 parse
= (type
== SPA_IMPORT_EXISTING
?
2569 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2572 * Create "The Godfather" zio to hold all async IOs
2574 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2576 for (i
= 0; i
< max_ncpus
; i
++) {
2577 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2578 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2579 ZIO_FLAG_GODFATHER
);
2583 * Parse the configuration into a vdev tree. We explicitly set the
2584 * value that will be returned by spa_version() since parsing the
2585 * configuration requires knowing the version number.
2587 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2588 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2589 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2594 ASSERT(spa
->spa_root_vdev
== rvd
);
2595 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2596 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2598 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2599 ASSERT(spa_guid(spa
) == pool_guid
);
2603 * Try to open all vdevs, loading each label in the process.
2605 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2606 error
= vdev_open(rvd
);
2607 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2612 * We need to validate the vdev labels against the configuration that
2613 * we have in hand, which is dependent on the setting of mosconfig. If
2614 * mosconfig is true then we're validating the vdev labels based on
2615 * that config. Otherwise, we're validating against the cached config
2616 * (zpool.cache) that was read when we loaded the zfs module, and then
2617 * later we will recursively call spa_load() and validate against
2620 * If we're assembling a new pool that's been split off from an
2621 * existing pool, the labels haven't yet been updated so we skip
2622 * validation for now.
2624 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2625 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2626 error
= vdev_validate(rvd
, mosconfig
);
2627 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2632 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2633 return (SET_ERROR(ENXIO
));
2637 * Find the best uberblock.
2639 vdev_uberblock_load(rvd
, ub
, &label
);
2642 * If we weren't able to find a single valid uberblock, return failure.
2644 if (ub
->ub_txg
== 0) {
2646 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2650 * For pools which have the multihost property on determine if the
2651 * pool is truly inactive and can be safely imported. Prevent
2652 * hosts which don't have a hostid set from importing the pool.
2654 activity_check
= spa_activity_check_required(spa
, ub
, config
);
2655 if (activity_check
) {
2656 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
2657 spa_get_hostid() == 0) {
2659 fnvlist_add_uint64(spa
->spa_load_info
,
2660 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
2661 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
2664 error
= spa_activity_check(spa
, ub
, config
);
2670 fnvlist_add_uint64(spa
->spa_load_info
,
2671 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
2672 fnvlist_add_uint64(spa
->spa_load_info
,
2673 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
2677 * If the pool has an unsupported version we can't open it.
2679 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2681 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2684 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2688 * If we weren't able to find what's necessary for reading the
2689 * MOS in the label, return failure.
2691 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2692 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2694 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2699 * Update our in-core representation with the definitive values
2702 nvlist_free(spa
->spa_label_features
);
2703 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2709 * Look through entries in the label nvlist's features_for_read. If
2710 * there is a feature listed there which we don't understand then we
2711 * cannot open a pool.
2713 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2714 nvlist_t
*unsup_feat
;
2717 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2720 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2722 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2723 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2724 VERIFY(nvlist_add_string(unsup_feat
,
2725 nvpair_name(nvp
), "") == 0);
2729 if (!nvlist_empty(unsup_feat
)) {
2730 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2731 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2732 nvlist_free(unsup_feat
);
2733 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2737 nvlist_free(unsup_feat
);
2741 * If the vdev guid sum doesn't match the uberblock, we have an
2742 * incomplete configuration. We first check to see if the pool
2743 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2744 * If it is, defer the vdev_guid_sum check till later so we
2745 * can handle missing vdevs.
2747 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2748 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2749 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2750 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2752 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2753 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2754 spa_try_repair(spa
, config
);
2755 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2756 nvlist_free(spa
->spa_config_splitting
);
2757 spa
->spa_config_splitting
= NULL
;
2761 * Initialize internal SPA structures.
2763 spa
->spa_state
= POOL_STATE_ACTIVE
;
2764 spa
->spa_ubsync
= spa
->spa_uberblock
;
2765 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2766 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2767 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2768 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2769 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2770 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2772 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2774 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2775 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2777 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2778 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2780 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2781 boolean_t missing_feat_read
= B_FALSE
;
2782 nvlist_t
*unsup_feat
, *enabled_feat
;
2785 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2786 &spa
->spa_feat_for_read_obj
) != 0) {
2787 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2790 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2791 &spa
->spa_feat_for_write_obj
) != 0) {
2792 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2795 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2796 &spa
->spa_feat_desc_obj
) != 0) {
2797 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2800 enabled_feat
= fnvlist_alloc();
2801 unsup_feat
= fnvlist_alloc();
2803 if (!spa_features_check(spa
, B_FALSE
,
2804 unsup_feat
, enabled_feat
))
2805 missing_feat_read
= B_TRUE
;
2807 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2808 if (!spa_features_check(spa
, B_TRUE
,
2809 unsup_feat
, enabled_feat
)) {
2810 missing_feat_write
= B_TRUE
;
2814 fnvlist_add_nvlist(spa
->spa_load_info
,
2815 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2817 if (!nvlist_empty(unsup_feat
)) {
2818 fnvlist_add_nvlist(spa
->spa_load_info
,
2819 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2822 fnvlist_free(enabled_feat
);
2823 fnvlist_free(unsup_feat
);
2825 if (!missing_feat_read
) {
2826 fnvlist_add_boolean(spa
->spa_load_info
,
2827 ZPOOL_CONFIG_CAN_RDONLY
);
2831 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2832 * twofold: to determine whether the pool is available for
2833 * import in read-write mode and (if it is not) whether the
2834 * pool is available for import in read-only mode. If the pool
2835 * is available for import in read-write mode, it is displayed
2836 * as available in userland; if it is not available for import
2837 * in read-only mode, it is displayed as unavailable in
2838 * userland. If the pool is available for import in read-only
2839 * mode but not read-write mode, it is displayed as unavailable
2840 * in userland with a special note that the pool is actually
2841 * available for open in read-only mode.
2843 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2844 * missing a feature for write, we must first determine whether
2845 * the pool can be opened read-only before returning to
2846 * userland in order to know whether to display the
2847 * abovementioned note.
2849 if (missing_feat_read
|| (missing_feat_write
&&
2850 spa_writeable(spa
))) {
2851 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2856 * Load refcounts for ZFS features from disk into an in-memory
2857 * cache during SPA initialization.
2859 for (i
= 0; i
< SPA_FEATURES
; i
++) {
2862 error
= feature_get_refcount_from_disk(spa
,
2863 &spa_feature_table
[i
], &refcount
);
2865 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2866 } else if (error
== ENOTSUP
) {
2867 spa
->spa_feat_refcount_cache
[i
] =
2868 SPA_FEATURE_DISABLED
;
2870 return (spa_vdev_err(rvd
,
2871 VDEV_AUX_CORRUPT_DATA
, EIO
));
2876 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2877 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2878 &spa
->spa_feat_enabled_txg_obj
) != 0)
2879 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2882 spa
->spa_is_initializing
= B_TRUE
;
2883 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2884 spa
->spa_is_initializing
= B_FALSE
;
2886 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2890 nvlist_t
*policy
= NULL
, *nvconfig
;
2892 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2893 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2895 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2896 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2898 unsigned long myhostid
= 0;
2900 VERIFY(nvlist_lookup_string(nvconfig
,
2901 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2903 myhostid
= spa_get_hostid();
2904 if (hostid
&& myhostid
&& hostid
!= myhostid
) {
2905 nvlist_free(nvconfig
);
2906 return (SET_ERROR(EBADF
));
2909 if (nvlist_lookup_nvlist(spa
->spa_config
,
2910 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2911 VERIFY(nvlist_add_nvlist(nvconfig
,
2912 ZPOOL_REWIND_POLICY
, policy
) == 0);
2914 spa_config_set(spa
, nvconfig
);
2916 spa_deactivate(spa
);
2917 spa_activate(spa
, orig_mode
);
2919 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2922 /* Grab the checksum salt from the MOS. */
2923 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2924 DMU_POOL_CHECKSUM_SALT
, 1,
2925 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
2926 spa
->spa_cksum_salt
.zcs_bytes
);
2927 if (error
== ENOENT
) {
2928 /* Generate a new salt for subsequent use */
2929 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
2930 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
2931 } else if (error
!= 0) {
2932 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2935 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2936 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2937 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2939 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2942 * Load the bit that tells us to use the new accounting function
2943 * (raid-z deflation). If we have an older pool, this will not
2946 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2947 if (error
!= 0 && error
!= ENOENT
)
2948 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2950 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2951 &spa
->spa_creation_version
);
2952 if (error
!= 0 && error
!= ENOENT
)
2953 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2956 * Load the persistent error log. If we have an older pool, this will
2959 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2960 if (error
!= 0 && error
!= ENOENT
)
2961 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2963 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2964 &spa
->spa_errlog_scrub
);
2965 if (error
!= 0 && error
!= ENOENT
)
2966 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2969 * Load the history object. If we have an older pool, this
2970 * will not be present.
2972 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2973 if (error
!= 0 && error
!= ENOENT
)
2974 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2977 * Load the per-vdev ZAP map. If we have an older pool, this will not
2978 * be present; in this case, defer its creation to a later time to
2979 * avoid dirtying the MOS this early / out of sync context. See
2980 * spa_sync_config_object.
2983 /* The sentinel is only available in the MOS config. */
2984 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
2985 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2987 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
2988 &spa
->spa_all_vdev_zaps
);
2990 if (error
== ENOENT
) {
2991 VERIFY(!nvlist_exists(mos_config
,
2992 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
2993 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
2994 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
2995 } else if (error
!= 0) {
2996 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2997 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
2999 * An older version of ZFS overwrote the sentinel value, so
3000 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3001 * destruction to later; see spa_sync_config_object.
3003 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
3005 * We're assuming that no vdevs have had their ZAPs created
3006 * before this. Better be sure of it.
3008 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3010 nvlist_free(mos_config
);
3013 * If we're assembling the pool from the split-off vdevs of
3014 * an existing pool, we don't want to attach the spares & cache
3019 * Load any hot spares for this pool.
3021 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
3022 if (error
!= 0 && error
!= ENOENT
)
3023 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3024 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3025 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3026 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3027 &spa
->spa_spares
.sav_config
) != 0)
3028 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3030 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3031 spa_load_spares(spa
);
3032 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3033 } else if (error
== 0) {
3034 spa
->spa_spares
.sav_sync
= B_TRUE
;
3038 * Load any level 2 ARC devices for this pool.
3040 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3041 &spa
->spa_l2cache
.sav_object
);
3042 if (error
!= 0 && error
!= ENOENT
)
3043 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3044 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3045 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3046 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3047 &spa
->spa_l2cache
.sav_config
) != 0)
3048 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3050 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3051 spa_load_l2cache(spa
);
3052 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3053 } else if (error
== 0) {
3054 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3057 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3059 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
3060 if (error
&& error
!= ENOENT
)
3061 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3064 uint64_t autoreplace
= 0;
3066 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3067 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3068 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3069 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3070 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3071 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
3072 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3073 &spa
->spa_dedup_ditto
);
3075 spa
->spa_autoreplace
= (autoreplace
!= 0);
3079 * If the 'multihost' property is set, then never allow a pool to
3080 * be imported when the system hostid is zero. The exception to
3081 * this rule is zdb which is always allowed to access pools.
3083 if (spa_multihost(spa
) && spa_get_hostid() == 0 &&
3084 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
3085 fnvlist_add_uint64(spa
->spa_load_info
,
3086 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3087 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3091 * If the 'autoreplace' property is set, then post a resource notifying
3092 * the ZFS DE that it should not issue any faults for unopenable
3093 * devices. We also iterate over the vdevs, and post a sysevent for any
3094 * unopenable vdevs so that the normal autoreplace handler can take
3097 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
3098 spa_check_removed(spa
->spa_root_vdev
);
3100 * For the import case, this is done in spa_import(), because
3101 * at this point we're using the spare definitions from
3102 * the MOS config, not necessarily from the userland config.
3104 if (state
!= SPA_LOAD_IMPORT
) {
3105 spa_aux_check_removed(&spa
->spa_spares
);
3106 spa_aux_check_removed(&spa
->spa_l2cache
);
3111 * Load the vdev state for all toplevel vdevs.
3116 * Propagate the leaf DTLs we just loaded all the way up the tree.
3118 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3119 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3120 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3123 * Load the DDTs (dedup tables).
3125 error
= ddt_load(spa
);
3127 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3129 spa_update_dspace(spa
);
3132 * Validate the config, using the MOS config to fill in any
3133 * information which might be missing. If we fail to validate
3134 * the config then declare the pool unfit for use. If we're
3135 * assembling a pool from a split, the log is not transferred
3138 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3141 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
3142 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3144 if (!spa_config_valid(spa
, nvconfig
)) {
3145 nvlist_free(nvconfig
);
3146 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3149 nvlist_free(nvconfig
);
3152 * Now that we've validated the config, check the state of the
3153 * root vdev. If it can't be opened, it indicates one or
3154 * more toplevel vdevs are faulted.
3156 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
3157 return (SET_ERROR(ENXIO
));
3159 if (spa_writeable(spa
) && spa_check_logs(spa
)) {
3160 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3161 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
3165 if (missing_feat_write
) {
3166 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
3169 * At this point, we know that we can open the pool in
3170 * read-only mode but not read-write mode. We now have enough
3171 * information and can return to userland.
3173 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
3177 * We've successfully opened the pool, verify that we're ready
3178 * to start pushing transactions.
3180 if (state
!= SPA_LOAD_TRYIMPORT
) {
3181 if ((error
= spa_load_verify(spa
)))
3182 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3186 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
3187 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3189 int need_update
= B_FALSE
;
3190 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3193 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
3196 * Claim log blocks that haven't been committed yet.
3197 * This must all happen in a single txg.
3198 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3199 * invoked from zil_claim_log_block()'s i/o done callback.
3200 * Price of rollback is that we abandon the log.
3202 spa
->spa_claiming
= B_TRUE
;
3204 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3205 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3206 zil_claim
, tx
, DS_FIND_CHILDREN
);
3209 spa
->spa_claiming
= B_FALSE
;
3211 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3212 spa
->spa_sync_on
= B_TRUE
;
3213 txg_sync_start(spa
->spa_dsl_pool
);
3214 mmp_thread_start(spa
);
3217 * Wait for all claims to sync. We sync up to the highest
3218 * claimed log block birth time so that claimed log blocks
3219 * don't appear to be from the future. spa_claim_max_txg
3220 * will have been set for us by either zil_check_log_chain()
3221 * (invoked from spa_check_logs()) or zil_claim() above.
3223 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3226 * If the config cache is stale, or we have uninitialized
3227 * metaslabs (see spa_vdev_add()), then update the config.
3229 * If this is a verbatim import, trust the current
3230 * in-core spa_config and update the disk labels.
3232 if (config_cache_txg
!= spa
->spa_config_txg
||
3233 state
== SPA_LOAD_IMPORT
||
3234 state
== SPA_LOAD_RECOVER
||
3235 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3236 need_update
= B_TRUE
;
3238 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3239 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3240 need_update
= B_TRUE
;
3243 * Update the config cache asychronously in case we're the
3244 * root pool, in which case the config cache isn't writable yet.
3247 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3250 * Check all DTLs to see if anything needs resilvering.
3252 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3253 vdev_resilver_needed(rvd
, NULL
, NULL
))
3254 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3257 * Log the fact that we booted up (so that we can detect if
3258 * we rebooted in the middle of an operation).
3260 spa_history_log_version(spa
, "open");
3263 * Delete any inconsistent datasets.
3265 (void) dmu_objset_find(spa_name(spa
),
3266 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3269 * Clean up any stale temporary dataset userrefs.
3271 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3278 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
3280 int mode
= spa
->spa_mode
;
3283 spa_deactivate(spa
);
3285 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3287 spa_activate(spa
, mode
);
3288 spa_async_suspend(spa
);
3290 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
3294 * If spa_load() fails this function will try loading prior txg's. If
3295 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3296 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3297 * function will not rewind the pool and will return the same error as
3301 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
3302 uint64_t max_request
, int rewind_flags
)
3304 nvlist_t
*loadinfo
= NULL
;
3305 nvlist_t
*config
= NULL
;
3306 int load_error
, rewind_error
;
3307 uint64_t safe_rewind_txg
;
3310 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3311 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3312 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3314 spa
->spa_load_max_txg
= max_request
;
3315 if (max_request
!= UINT64_MAX
)
3316 spa
->spa_extreme_rewind
= B_TRUE
;
3319 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
3321 if (load_error
== 0)
3324 if (spa
->spa_root_vdev
!= NULL
)
3325 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3327 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3328 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3330 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3331 nvlist_free(config
);
3332 return (load_error
);
3335 if (state
== SPA_LOAD_RECOVER
) {
3336 /* Price of rolling back is discarding txgs, including log */
3337 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3340 * If we aren't rolling back save the load info from our first
3341 * import attempt so that we can restore it after attempting
3344 loadinfo
= spa
->spa_load_info
;
3345 spa
->spa_load_info
= fnvlist_alloc();
3348 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3349 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
3350 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
3351 TXG_INITIAL
: safe_rewind_txg
;
3354 * Continue as long as we're finding errors, we're still within
3355 * the acceptable rewind range, and we're still finding uberblocks
3357 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
3358 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
3359 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
3360 spa
->spa_extreme_rewind
= B_TRUE
;
3361 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
3364 spa
->spa_extreme_rewind
= B_FALSE
;
3365 spa
->spa_load_max_txg
= UINT64_MAX
;
3367 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
3368 spa_config_set(spa
, config
);
3370 nvlist_free(config
);
3372 if (state
== SPA_LOAD_RECOVER
) {
3373 ASSERT3P(loadinfo
, ==, NULL
);
3374 return (rewind_error
);
3376 /* Store the rewind info as part of the initial load info */
3377 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
3378 spa
->spa_load_info
);
3380 /* Restore the initial load info */
3381 fnvlist_free(spa
->spa_load_info
);
3382 spa
->spa_load_info
= loadinfo
;
3384 return (load_error
);
3391 * The import case is identical to an open except that the configuration is sent
3392 * down from userland, instead of grabbed from the configuration cache. For the
3393 * case of an open, the pool configuration will exist in the
3394 * POOL_STATE_UNINITIALIZED state.
3396 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3397 * the same time open the pool, without having to keep around the spa_t in some
3401 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
3405 spa_load_state_t state
= SPA_LOAD_OPEN
;
3407 int locked
= B_FALSE
;
3408 int firstopen
= B_FALSE
;
3413 * As disgusting as this is, we need to support recursive calls to this
3414 * function because dsl_dir_open() is called during spa_load(), and ends
3415 * up calling spa_open() again. The real fix is to figure out how to
3416 * avoid dsl_dir_open() calling this in the first place.
3418 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
3419 mutex_enter(&spa_namespace_lock
);
3423 if ((spa
= spa_lookup(pool
)) == NULL
) {
3425 mutex_exit(&spa_namespace_lock
);
3426 return (SET_ERROR(ENOENT
));
3429 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3430 zpool_rewind_policy_t policy
;
3434 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3436 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3437 state
= SPA_LOAD_RECOVER
;
3439 spa_activate(spa
, spa_mode_global
);
3441 if (state
!= SPA_LOAD_RECOVER
)
3442 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3444 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3445 policy
.zrp_request
);
3447 if (error
== EBADF
) {
3449 * If vdev_validate() returns failure (indicated by
3450 * EBADF), it indicates that one of the vdevs indicates
3451 * that the pool has been exported or destroyed. If
3452 * this is the case, the config cache is out of sync and
3453 * we should remove the pool from the namespace.
3456 spa_deactivate(spa
);
3457 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3460 mutex_exit(&spa_namespace_lock
);
3461 return (SET_ERROR(ENOENT
));
3466 * We can't open the pool, but we still have useful
3467 * information: the state of each vdev after the
3468 * attempted vdev_open(). Return this to the user.
3470 if (config
!= NULL
&& spa
->spa_config
) {
3471 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3473 VERIFY(nvlist_add_nvlist(*config
,
3474 ZPOOL_CONFIG_LOAD_INFO
,
3475 spa
->spa_load_info
) == 0);
3478 spa_deactivate(spa
);
3479 spa
->spa_last_open_failed
= error
;
3481 mutex_exit(&spa_namespace_lock
);
3487 spa_open_ref(spa
, tag
);
3490 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3493 * If we've recovered the pool, pass back any information we
3494 * gathered while doing the load.
3496 if (state
== SPA_LOAD_RECOVER
) {
3497 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3498 spa
->spa_load_info
) == 0);
3502 spa
->spa_last_open_failed
= 0;
3503 spa
->spa_last_ubsync_txg
= 0;
3504 spa
->spa_load_txg
= 0;
3505 mutex_exit(&spa_namespace_lock
);
3509 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
3517 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3520 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3524 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3526 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3530 * Lookup the given spa_t, incrementing the inject count in the process,
3531 * preventing it from being exported or destroyed.
3534 spa_inject_addref(char *name
)
3538 mutex_enter(&spa_namespace_lock
);
3539 if ((spa
= spa_lookup(name
)) == NULL
) {
3540 mutex_exit(&spa_namespace_lock
);
3543 spa
->spa_inject_ref
++;
3544 mutex_exit(&spa_namespace_lock
);
3550 spa_inject_delref(spa_t
*spa
)
3552 mutex_enter(&spa_namespace_lock
);
3553 spa
->spa_inject_ref
--;
3554 mutex_exit(&spa_namespace_lock
);
3558 * Add spares device information to the nvlist.
3561 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3571 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3573 if (spa
->spa_spares
.sav_count
== 0)
3576 VERIFY(nvlist_lookup_nvlist(config
,
3577 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3578 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3579 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3581 VERIFY(nvlist_add_nvlist_array(nvroot
,
3582 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3583 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3584 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3587 * Go through and find any spares which have since been
3588 * repurposed as an active spare. If this is the case, update
3589 * their status appropriately.
3591 for (i
= 0; i
< nspares
; i
++) {
3592 VERIFY(nvlist_lookup_uint64(spares
[i
],
3593 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3594 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3596 VERIFY(nvlist_lookup_uint64_array(
3597 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3598 (uint64_t **)&vs
, &vsc
) == 0);
3599 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3600 vs
->vs_aux
= VDEV_AUX_SPARED
;
3607 * Add l2cache device information to the nvlist, including vdev stats.
3610 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3613 uint_t i
, j
, nl2cache
;
3620 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3622 if (spa
->spa_l2cache
.sav_count
== 0)
3625 VERIFY(nvlist_lookup_nvlist(config
,
3626 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3627 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3628 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3629 if (nl2cache
!= 0) {
3630 VERIFY(nvlist_add_nvlist_array(nvroot
,
3631 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3632 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3633 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3636 * Update level 2 cache device stats.
3639 for (i
= 0; i
< nl2cache
; i
++) {
3640 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3641 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3644 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3646 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3647 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3653 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3654 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3656 vdev_get_stats(vd
, vs
);
3657 vdev_config_generate_stats(vd
, l2cache
[i
]);
3664 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3669 if (spa
->spa_feat_for_read_obj
!= 0) {
3670 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3671 spa
->spa_feat_for_read_obj
);
3672 zap_cursor_retrieve(&zc
, &za
) == 0;
3673 zap_cursor_advance(&zc
)) {
3674 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3675 za
.za_num_integers
== 1);
3676 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3677 za
.za_first_integer
));
3679 zap_cursor_fini(&zc
);
3682 if (spa
->spa_feat_for_write_obj
!= 0) {
3683 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3684 spa
->spa_feat_for_write_obj
);
3685 zap_cursor_retrieve(&zc
, &za
) == 0;
3686 zap_cursor_advance(&zc
)) {
3687 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3688 za
.za_num_integers
== 1);
3689 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3690 za
.za_first_integer
));
3692 zap_cursor_fini(&zc
);
3697 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3701 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3702 zfeature_info_t feature
= spa_feature_table
[i
];
3705 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3708 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3713 * Store a list of pool features and their reference counts in the
3716 * The first time this is called on a spa, allocate a new nvlist, fetch
3717 * the pool features and reference counts from disk, then save the list
3718 * in the spa. In subsequent calls on the same spa use the saved nvlist
3719 * and refresh its values from the cached reference counts. This
3720 * ensures we don't block here on I/O on a suspended pool so 'zpool
3721 * clear' can resume the pool.
3724 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3728 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3730 mutex_enter(&spa
->spa_feat_stats_lock
);
3731 features
= spa
->spa_feat_stats
;
3733 if (features
!= NULL
) {
3734 spa_feature_stats_from_cache(spa
, features
);
3736 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3737 spa
->spa_feat_stats
= features
;
3738 spa_feature_stats_from_disk(spa
, features
);
3741 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3744 mutex_exit(&spa
->spa_feat_stats_lock
);
3748 spa_get_stats(const char *name
, nvlist_t
**config
,
3749 char *altroot
, size_t buflen
)
3755 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3759 * This still leaves a window of inconsistency where the spares
3760 * or l2cache devices could change and the config would be
3761 * self-inconsistent.
3763 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3765 if (*config
!= NULL
) {
3766 uint64_t loadtimes
[2];
3768 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3769 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3770 VERIFY(nvlist_add_uint64_array(*config
,
3771 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3773 VERIFY(nvlist_add_uint64(*config
,
3774 ZPOOL_CONFIG_ERRCOUNT
,
3775 spa_get_errlog_size(spa
)) == 0);
3777 if (spa_suspended(spa
))
3778 VERIFY(nvlist_add_uint64(*config
,
3779 ZPOOL_CONFIG_SUSPENDED
,
3780 spa
->spa_failmode
) == 0);
3782 spa_add_spares(spa
, *config
);
3783 spa_add_l2cache(spa
, *config
);
3784 spa_add_feature_stats(spa
, *config
);
3789 * We want to get the alternate root even for faulted pools, so we cheat
3790 * and call spa_lookup() directly.
3794 mutex_enter(&spa_namespace_lock
);
3795 spa
= spa_lookup(name
);
3797 spa_altroot(spa
, altroot
, buflen
);
3801 mutex_exit(&spa_namespace_lock
);
3803 spa_altroot(spa
, altroot
, buflen
);
3808 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3809 spa_close(spa
, FTAG
);
3816 * Validate that the auxiliary device array is well formed. We must have an
3817 * array of nvlists, each which describes a valid leaf vdev. If this is an
3818 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3819 * specified, as long as they are well-formed.
3822 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3823 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3824 vdev_labeltype_t label
)
3831 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3834 * It's acceptable to have no devs specified.
3836 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3840 return (SET_ERROR(EINVAL
));
3843 * Make sure the pool is formatted with a version that supports this
3846 if (spa_version(spa
) < version
)
3847 return (SET_ERROR(ENOTSUP
));
3850 * Set the pending device list so we correctly handle device in-use
3853 sav
->sav_pending
= dev
;
3854 sav
->sav_npending
= ndev
;
3856 for (i
= 0; i
< ndev
; i
++) {
3857 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3861 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3863 error
= SET_ERROR(EINVAL
);
3869 if ((error
= vdev_open(vd
)) == 0 &&
3870 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3871 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3872 vd
->vdev_guid
) == 0);
3878 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3885 sav
->sav_pending
= NULL
;
3886 sav
->sav_npending
= 0;
3891 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3895 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3897 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3898 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3899 VDEV_LABEL_SPARE
)) != 0) {
3903 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3904 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3905 VDEV_LABEL_L2CACHE
));
3909 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3914 if (sav
->sav_config
!= NULL
) {
3920 * Generate new dev list by concatenating with the
3923 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3924 &olddevs
, &oldndevs
) == 0);
3926 newdevs
= kmem_alloc(sizeof (void *) *
3927 (ndevs
+ oldndevs
), KM_SLEEP
);
3928 for (i
= 0; i
< oldndevs
; i
++)
3929 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3931 for (i
= 0; i
< ndevs
; i
++)
3932 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3935 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3936 DATA_TYPE_NVLIST_ARRAY
) == 0);
3938 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3939 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3940 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3941 nvlist_free(newdevs
[i
]);
3942 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3945 * Generate a new dev list.
3947 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3949 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3955 * Stop and drop level 2 ARC devices
3958 spa_l2cache_drop(spa_t
*spa
)
3962 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3964 for (i
= 0; i
< sav
->sav_count
; i
++) {
3967 vd
= sav
->sav_vdevs
[i
];
3970 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3971 pool
!= 0ULL && l2arc_vdev_present(vd
))
3972 l2arc_remove_vdev(vd
);
3980 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
3984 char *altroot
= NULL
;
3989 uint64_t txg
= TXG_INITIAL
;
3990 nvlist_t
**spares
, **l2cache
;
3991 uint_t nspares
, nl2cache
;
3992 uint64_t version
, obj
;
3993 boolean_t has_features
;
3999 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
4000 poolname
= (char *)pool
;
4003 * If this pool already exists, return failure.
4005 mutex_enter(&spa_namespace_lock
);
4006 if (spa_lookup(poolname
) != NULL
) {
4007 mutex_exit(&spa_namespace_lock
);
4008 return (SET_ERROR(EEXIST
));
4012 * Allocate a new spa_t structure.
4014 nvl
= fnvlist_alloc();
4015 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
4016 (void) nvlist_lookup_string(props
,
4017 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4018 spa
= spa_add(poolname
, nvl
, altroot
);
4020 spa_activate(spa
, spa_mode_global
);
4022 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4023 spa_deactivate(spa
);
4025 mutex_exit(&spa_namespace_lock
);
4030 * Temporary pool names should never be written to disk.
4032 if (poolname
!= pool
)
4033 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
4035 has_features
= B_FALSE
;
4036 for (elem
= nvlist_next_nvpair(props
, NULL
);
4037 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4038 if (zpool_prop_feature(nvpair_name(elem
)))
4039 has_features
= B_TRUE
;
4042 if (has_features
|| nvlist_lookup_uint64(props
,
4043 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4044 version
= SPA_VERSION
;
4046 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4048 spa
->spa_first_txg
= txg
;
4049 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4050 spa
->spa_uberblock
.ub_version
= version
;
4051 spa
->spa_ubsync
= spa
->spa_uberblock
;
4052 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4055 * Create "The Godfather" zio to hold all async IOs
4057 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4059 for (i
= 0; i
< max_ncpus
; i
++) {
4060 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4061 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4062 ZIO_FLAG_GODFATHER
);
4066 * Create the root vdev.
4068 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4070 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4072 ASSERT(error
!= 0 || rvd
!= NULL
);
4073 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4075 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4076 error
= SET_ERROR(EINVAL
);
4079 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4080 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4081 VDEV_ALLOC_ADD
)) == 0) {
4082 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
4083 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4084 vdev_expand(rvd
->vdev_child
[c
], txg
);
4088 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4092 spa_deactivate(spa
);
4094 mutex_exit(&spa_namespace_lock
);
4099 * Get the list of spares, if specified.
4101 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4102 &spares
, &nspares
) == 0) {
4103 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4105 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4106 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4107 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4108 spa_load_spares(spa
);
4109 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4110 spa
->spa_spares
.sav_sync
= B_TRUE
;
4114 * Get the list of level 2 cache devices, if specified.
4116 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4117 &l2cache
, &nl2cache
) == 0) {
4118 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4119 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4120 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4121 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4122 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4123 spa_load_l2cache(spa
);
4124 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4125 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4128 spa
->spa_is_initializing
= B_TRUE
;
4129 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
4130 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
4131 spa
->spa_is_initializing
= B_FALSE
;
4134 * Create DDTs (dedup tables).
4138 spa_update_dspace(spa
);
4140 tx
= dmu_tx_create_assigned(dp
, txg
);
4143 * Create the pool config object.
4145 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4146 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4147 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4149 if (zap_add(spa
->spa_meta_objset
,
4150 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4151 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4152 cmn_err(CE_PANIC
, "failed to add pool config");
4155 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
4156 spa_feature_create_zap_objects(spa
, tx
);
4158 if (zap_add(spa
->spa_meta_objset
,
4159 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4160 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4161 cmn_err(CE_PANIC
, "failed to add pool version");
4164 /* Newly created pools with the right version are always deflated. */
4165 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4166 spa
->spa_deflate
= TRUE
;
4167 if (zap_add(spa
->spa_meta_objset
,
4168 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4169 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4170 cmn_err(CE_PANIC
, "failed to add deflate");
4175 * Create the deferred-free bpobj. Turn off compression
4176 * because sync-to-convergence takes longer if the blocksize
4179 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4180 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4181 ZIO_COMPRESS_OFF
, tx
);
4182 if (zap_add(spa
->spa_meta_objset
,
4183 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4184 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4185 cmn_err(CE_PANIC
, "failed to add bpobj");
4187 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4188 spa
->spa_meta_objset
, obj
));
4191 * Create the pool's history object.
4193 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
4194 spa_history_create_obj(spa
, tx
);
4197 * Generate some random noise for salted checksums to operate on.
4199 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4200 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4203 * Set pool properties.
4205 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4206 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4207 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4208 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4209 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
4211 if (props
!= NULL
) {
4212 spa_configfile_set(spa
, props
, B_FALSE
);
4213 spa_sync_props(props
, tx
);
4218 spa
->spa_sync_on
= B_TRUE
;
4219 txg_sync_start(spa
->spa_dsl_pool
);
4220 mmp_thread_start(spa
);
4223 * We explicitly wait for the first transaction to complete so that our
4224 * bean counters are appropriately updated.
4226 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
4228 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4229 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4231 spa_history_log_version(spa
, "create");
4234 * Don't count references from objsets that are already closed
4235 * and are making their way through the eviction process.
4237 spa_evicting_os_wait(spa
);
4238 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4239 spa
->spa_load_state
= SPA_LOAD_NONE
;
4241 mutex_exit(&spa_namespace_lock
);
4247 * Import a non-root pool into the system.
4250 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4253 char *altroot
= NULL
;
4254 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4255 zpool_rewind_policy_t policy
;
4256 uint64_t mode
= spa_mode_global
;
4257 uint64_t readonly
= B_FALSE
;
4260 nvlist_t
**spares
, **l2cache
;
4261 uint_t nspares
, nl2cache
;
4264 * If a pool with this name exists, return failure.
4266 mutex_enter(&spa_namespace_lock
);
4267 if (spa_lookup(pool
) != NULL
) {
4268 mutex_exit(&spa_namespace_lock
);
4269 return (SET_ERROR(EEXIST
));
4273 * Create and initialize the spa structure.
4275 (void) nvlist_lookup_string(props
,
4276 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4277 (void) nvlist_lookup_uint64(props
,
4278 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4281 spa
= spa_add(pool
, config
, altroot
);
4282 spa
->spa_import_flags
= flags
;
4285 * Verbatim import - Take a pool and insert it into the namespace
4286 * as if it had been loaded at boot.
4288 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4290 spa_configfile_set(spa
, props
, B_FALSE
);
4292 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4293 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4295 mutex_exit(&spa_namespace_lock
);
4299 spa_activate(spa
, mode
);
4302 * Don't start async tasks until we know everything is healthy.
4304 spa_async_suspend(spa
);
4306 zpool_get_rewind_policy(config
, &policy
);
4307 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4308 state
= SPA_LOAD_RECOVER
;
4311 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4312 * because the user-supplied config is actually the one to trust when
4315 if (state
!= SPA_LOAD_RECOVER
)
4316 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4318 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4319 policy
.zrp_request
);
4322 * Propagate anything learned while loading the pool and pass it
4323 * back to caller (i.e. rewind info, missing devices, etc).
4325 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4326 spa
->spa_load_info
) == 0);
4328 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4330 * Toss any existing sparelist, as it doesn't have any validity
4331 * anymore, and conflicts with spa_has_spare().
4333 if (spa
->spa_spares
.sav_config
) {
4334 nvlist_free(spa
->spa_spares
.sav_config
);
4335 spa
->spa_spares
.sav_config
= NULL
;
4336 spa_load_spares(spa
);
4338 if (spa
->spa_l2cache
.sav_config
) {
4339 nvlist_free(spa
->spa_l2cache
.sav_config
);
4340 spa
->spa_l2cache
.sav_config
= NULL
;
4341 spa_load_l2cache(spa
);
4344 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4346 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4349 spa_configfile_set(spa
, props
, B_FALSE
);
4351 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4352 (error
= spa_prop_set(spa
, props
)))) {
4354 spa_deactivate(spa
);
4356 mutex_exit(&spa_namespace_lock
);
4360 spa_async_resume(spa
);
4363 * Override any spares and level 2 cache devices as specified by
4364 * the user, as these may have correct device names/devids, etc.
4366 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4367 &spares
, &nspares
) == 0) {
4368 if (spa
->spa_spares
.sav_config
)
4369 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4370 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4372 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4373 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4374 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4375 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4376 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4377 spa_load_spares(spa
);
4378 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4379 spa
->spa_spares
.sav_sync
= B_TRUE
;
4381 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4382 &l2cache
, &nl2cache
) == 0) {
4383 if (spa
->spa_l2cache
.sav_config
)
4384 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4385 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4387 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4388 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4389 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4390 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4391 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4392 spa_load_l2cache(spa
);
4393 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4394 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4398 * Check for any removed devices.
4400 if (spa
->spa_autoreplace
) {
4401 spa_aux_check_removed(&spa
->spa_spares
);
4402 spa_aux_check_removed(&spa
->spa_l2cache
);
4405 if (spa_writeable(spa
)) {
4407 * Update the config cache to include the newly-imported pool.
4409 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4413 * It's possible that the pool was expanded while it was exported.
4414 * We kick off an async task to handle this for us.
4416 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4418 spa_history_log_version(spa
, "import");
4420 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4422 zvol_create_minors(spa
, pool
, B_TRUE
);
4424 mutex_exit(&spa_namespace_lock
);
4430 spa_tryimport(nvlist_t
*tryconfig
)
4432 nvlist_t
*config
= NULL
;
4438 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4441 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4445 * Create and initialize the spa structure.
4447 mutex_enter(&spa_namespace_lock
);
4448 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4449 spa_activate(spa
, FREAD
);
4452 * Pass off the heavy lifting to spa_load().
4453 * Pass TRUE for mosconfig because the user-supplied config
4454 * is actually the one to trust when doing an import.
4456 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4459 * If 'tryconfig' was at least parsable, return the current config.
4461 if (spa
->spa_root_vdev
!= NULL
) {
4462 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4463 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4465 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4467 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4468 spa
->spa_uberblock
.ub_timestamp
) == 0);
4469 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4470 spa
->spa_load_info
) == 0);
4471 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4472 spa
->spa_errata
) == 0);
4475 * If the bootfs property exists on this pool then we
4476 * copy it out so that external consumers can tell which
4477 * pools are bootable.
4479 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4480 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4483 * We have to play games with the name since the
4484 * pool was opened as TRYIMPORT_NAME.
4486 if (dsl_dsobj_to_dsname(spa_name(spa
),
4487 spa
->spa_bootfs
, tmpname
) == 0) {
4491 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4493 cp
= strchr(tmpname
, '/');
4495 (void) strlcpy(dsname
, tmpname
,
4498 (void) snprintf(dsname
, MAXPATHLEN
,
4499 "%s/%s", poolname
, ++cp
);
4501 VERIFY(nvlist_add_string(config
,
4502 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4503 kmem_free(dsname
, MAXPATHLEN
);
4505 kmem_free(tmpname
, MAXPATHLEN
);
4509 * Add the list of hot spares and level 2 cache devices.
4511 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4512 spa_add_spares(spa
, config
);
4513 spa_add_l2cache(spa
, config
);
4514 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4518 spa_deactivate(spa
);
4520 mutex_exit(&spa_namespace_lock
);
4526 * Pool export/destroy
4528 * The act of destroying or exporting a pool is very simple. We make sure there
4529 * is no more pending I/O and any references to the pool are gone. Then, we
4530 * update the pool state and sync all the labels to disk, removing the
4531 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4532 * we don't sync the labels or remove the configuration cache.
4535 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4536 boolean_t force
, boolean_t hardforce
)
4543 if (!(spa_mode_global
& FWRITE
))
4544 return (SET_ERROR(EROFS
));
4546 mutex_enter(&spa_namespace_lock
);
4547 if ((spa
= spa_lookup(pool
)) == NULL
) {
4548 mutex_exit(&spa_namespace_lock
);
4549 return (SET_ERROR(ENOENT
));
4553 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4554 * reacquire the namespace lock, and see if we can export.
4556 spa_open_ref(spa
, FTAG
);
4557 mutex_exit(&spa_namespace_lock
);
4558 spa_async_suspend(spa
);
4559 if (spa
->spa_zvol_taskq
) {
4560 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
4561 taskq_wait(spa
->spa_zvol_taskq
);
4563 mutex_enter(&spa_namespace_lock
);
4564 spa_close(spa
, FTAG
);
4566 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4569 * The pool will be in core if it's openable, in which case we can
4570 * modify its state. Objsets may be open only because they're dirty,
4571 * so we have to force it to sync before checking spa_refcnt.
4573 if (spa
->spa_sync_on
) {
4574 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4575 spa_evicting_os_wait(spa
);
4579 * A pool cannot be exported or destroyed if there are active
4580 * references. If we are resetting a pool, allow references by
4581 * fault injection handlers.
4583 if (!spa_refcount_zero(spa
) ||
4584 (spa
->spa_inject_ref
!= 0 &&
4585 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4586 spa_async_resume(spa
);
4587 mutex_exit(&spa_namespace_lock
);
4588 return (SET_ERROR(EBUSY
));
4591 if (spa
->spa_sync_on
) {
4593 * A pool cannot be exported if it has an active shared spare.
4594 * This is to prevent other pools stealing the active spare
4595 * from an exported pool. At user's own will, such pool can
4596 * be forcedly exported.
4598 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4599 spa_has_active_shared_spare(spa
)) {
4600 spa_async_resume(spa
);
4601 mutex_exit(&spa_namespace_lock
);
4602 return (SET_ERROR(EXDEV
));
4606 * We want this to be reflected on every label,
4607 * so mark them all dirty. spa_unload() will do the
4608 * final sync that pushes these changes out.
4610 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4611 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4612 spa
->spa_state
= new_state
;
4613 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4615 vdev_config_dirty(spa
->spa_root_vdev
);
4616 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4621 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
4623 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4625 spa_deactivate(spa
);
4628 if (oldconfig
&& spa
->spa_config
)
4629 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4631 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4633 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4636 mutex_exit(&spa_namespace_lock
);
4642 * Destroy a storage pool.
4645 spa_destroy(char *pool
)
4647 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4652 * Export a storage pool.
4655 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4656 boolean_t hardforce
)
4658 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4663 * Similar to spa_export(), this unloads the spa_t without actually removing it
4664 * from the namespace in any way.
4667 spa_reset(char *pool
)
4669 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4674 * ==========================================================================
4675 * Device manipulation
4676 * ==========================================================================
4680 * Add a device to a storage pool.
4683 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4687 vdev_t
*rvd
= spa
->spa_root_vdev
;
4689 nvlist_t
**spares
, **l2cache
;
4690 uint_t nspares
, nl2cache
;
4693 ASSERT(spa_writeable(spa
));
4695 txg
= spa_vdev_enter(spa
);
4697 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4698 VDEV_ALLOC_ADD
)) != 0)
4699 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4701 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4703 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4707 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4711 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4712 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4714 if (vd
->vdev_children
!= 0 &&
4715 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4716 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4719 * We must validate the spares and l2cache devices after checking the
4720 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4722 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4723 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4726 * Transfer each new top-level vdev from vd to rvd.
4728 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4731 * Set the vdev id to the first hole, if one exists.
4733 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4734 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4735 vdev_free(rvd
->vdev_child
[id
]);
4739 tvd
= vd
->vdev_child
[c
];
4740 vdev_remove_child(vd
, tvd
);
4742 vdev_add_child(rvd
, tvd
);
4743 vdev_config_dirty(tvd
);
4747 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4748 ZPOOL_CONFIG_SPARES
);
4749 spa_load_spares(spa
);
4750 spa
->spa_spares
.sav_sync
= B_TRUE
;
4753 if (nl2cache
!= 0) {
4754 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4755 ZPOOL_CONFIG_L2CACHE
);
4756 spa_load_l2cache(spa
);
4757 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4761 * We have to be careful when adding new vdevs to an existing pool.
4762 * If other threads start allocating from these vdevs before we
4763 * sync the config cache, and we lose power, then upon reboot we may
4764 * fail to open the pool because there are DVAs that the config cache
4765 * can't translate. Therefore, we first add the vdevs without
4766 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4767 * and then let spa_config_update() initialize the new metaslabs.
4769 * spa_load() checks for added-but-not-initialized vdevs, so that
4770 * if we lose power at any point in this sequence, the remaining
4771 * steps will be completed the next time we load the pool.
4773 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4775 mutex_enter(&spa_namespace_lock
);
4776 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4777 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
4778 mutex_exit(&spa_namespace_lock
);
4784 * Attach a device to a mirror. The arguments are the path to any device
4785 * in the mirror, and the nvroot for the new device. If the path specifies
4786 * a device that is not mirrored, we automatically insert the mirror vdev.
4788 * If 'replacing' is specified, the new device is intended to replace the
4789 * existing device; in this case the two devices are made into their own
4790 * mirror using the 'replacing' vdev, which is functionally identical to
4791 * the mirror vdev (it actually reuses all the same ops) but has a few
4792 * extra rules: you can't attach to it after it's been created, and upon
4793 * completion of resilvering, the first disk (the one being replaced)
4794 * is automatically detached.
4797 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4799 uint64_t txg
, dtl_max_txg
;
4800 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4802 char *oldvdpath
, *newvdpath
;
4805 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4807 ASSERT(spa_writeable(spa
));
4809 txg
= spa_vdev_enter(spa
);
4811 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4814 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4816 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4817 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4819 pvd
= oldvd
->vdev_parent
;
4821 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4822 VDEV_ALLOC_ATTACH
)) != 0)
4823 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4825 if (newrootvd
->vdev_children
!= 1)
4826 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4828 newvd
= newrootvd
->vdev_child
[0];
4830 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4831 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4833 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4834 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4837 * Spares can't replace logs
4839 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4840 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4844 * For attach, the only allowable parent is a mirror or the root
4847 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4848 pvd
->vdev_ops
!= &vdev_root_ops
)
4849 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4851 pvops
= &vdev_mirror_ops
;
4854 * Active hot spares can only be replaced by inactive hot
4857 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4858 oldvd
->vdev_isspare
&&
4859 !spa_has_spare(spa
, newvd
->vdev_guid
))
4860 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4863 * If the source is a hot spare, and the parent isn't already a
4864 * spare, then we want to create a new hot spare. Otherwise, we
4865 * want to create a replacing vdev. The user is not allowed to
4866 * attach to a spared vdev child unless the 'isspare' state is
4867 * the same (spare replaces spare, non-spare replaces
4870 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4871 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4872 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4873 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4874 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4875 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4878 if (newvd
->vdev_isspare
)
4879 pvops
= &vdev_spare_ops
;
4881 pvops
= &vdev_replacing_ops
;
4885 * Make sure the new device is big enough.
4887 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4888 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4891 * The new device cannot have a higher alignment requirement
4892 * than the top-level vdev.
4894 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4895 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4898 * If this is an in-place replacement, update oldvd's path and devid
4899 * to make it distinguishable from newvd, and unopenable from now on.
4901 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4902 spa_strfree(oldvd
->vdev_path
);
4903 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4905 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4906 newvd
->vdev_path
, "old");
4907 if (oldvd
->vdev_devid
!= NULL
) {
4908 spa_strfree(oldvd
->vdev_devid
);
4909 oldvd
->vdev_devid
= NULL
;
4913 /* mark the device being resilvered */
4914 newvd
->vdev_resilver_txg
= txg
;
4917 * If the parent is not a mirror, or if we're replacing, insert the new
4918 * mirror/replacing/spare vdev above oldvd.
4920 if (pvd
->vdev_ops
!= pvops
)
4921 pvd
= vdev_add_parent(oldvd
, pvops
);
4923 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4924 ASSERT(pvd
->vdev_ops
== pvops
);
4925 ASSERT(oldvd
->vdev_parent
== pvd
);
4928 * Extract the new device from its root and add it to pvd.
4930 vdev_remove_child(newrootvd
, newvd
);
4931 newvd
->vdev_id
= pvd
->vdev_children
;
4932 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4933 vdev_add_child(pvd
, newvd
);
4936 * Reevaluate the parent vdev state.
4938 vdev_propagate_state(pvd
);
4940 tvd
= newvd
->vdev_top
;
4941 ASSERT(pvd
->vdev_top
== tvd
);
4942 ASSERT(tvd
->vdev_parent
== rvd
);
4944 vdev_config_dirty(tvd
);
4947 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4948 * for any dmu_sync-ed blocks. It will propagate upward when
4949 * spa_vdev_exit() calls vdev_dtl_reassess().
4951 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4953 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4954 dtl_max_txg
- TXG_INITIAL
);
4956 if (newvd
->vdev_isspare
) {
4957 spa_spare_activate(newvd
);
4958 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
4961 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4962 newvdpath
= spa_strdup(newvd
->vdev_path
);
4963 newvd_isspare
= newvd
->vdev_isspare
;
4966 * Mark newvd's DTL dirty in this txg.
4968 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4971 * Schedule the resilver to restart in the future. We do this to
4972 * ensure that dmu_sync-ed blocks have been stitched into the
4973 * respective datasets.
4975 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
4977 if (spa
->spa_bootfs
)
4978 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
4980 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
4985 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
4987 spa_history_log_internal(spa
, "vdev attach", NULL
,
4988 "%s vdev=%s %s vdev=%s",
4989 replacing
&& newvd_isspare
? "spare in" :
4990 replacing
? "replace" : "attach", newvdpath
,
4991 replacing
? "for" : "to", oldvdpath
);
4993 spa_strfree(oldvdpath
);
4994 spa_strfree(newvdpath
);
5000 * Detach a device from a mirror or replacing vdev.
5002 * If 'replace_done' is specified, only detach if the parent
5003 * is a replacing vdev.
5006 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5010 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5011 boolean_t unspare
= B_FALSE
;
5012 uint64_t unspare_guid
= 0;
5015 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5016 ASSERT(spa_writeable(spa
));
5018 txg
= spa_vdev_enter(spa
);
5020 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5023 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5025 if (!vd
->vdev_ops
->vdev_op_leaf
)
5026 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5028 pvd
= vd
->vdev_parent
;
5031 * If the parent/child relationship is not as expected, don't do it.
5032 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5033 * vdev that's replacing B with C. The user's intent in replacing
5034 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5035 * the replace by detaching C, the expected behavior is to end up
5036 * M(A,B). But suppose that right after deciding to detach C,
5037 * the replacement of B completes. We would have M(A,C), and then
5038 * ask to detach C, which would leave us with just A -- not what
5039 * the user wanted. To prevent this, we make sure that the
5040 * parent/child relationship hasn't changed -- in this example,
5041 * that C's parent is still the replacing vdev R.
5043 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5044 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5047 * Only 'replacing' or 'spare' vdevs can be replaced.
5049 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5050 pvd
->vdev_ops
!= &vdev_spare_ops
)
5051 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5053 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5054 spa_version(spa
) >= SPA_VERSION_SPARES
);
5057 * Only mirror, replacing, and spare vdevs support detach.
5059 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5060 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5061 pvd
->vdev_ops
!= &vdev_spare_ops
)
5062 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5065 * If this device has the only valid copy of some data,
5066 * we cannot safely detach it.
5068 if (vdev_dtl_required(vd
))
5069 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5071 ASSERT(pvd
->vdev_children
>= 2);
5074 * If we are detaching the second disk from a replacing vdev, then
5075 * check to see if we changed the original vdev's path to have "/old"
5076 * at the end in spa_vdev_attach(). If so, undo that change now.
5078 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5079 vd
->vdev_path
!= NULL
) {
5080 size_t len
= strlen(vd
->vdev_path
);
5082 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
5083 cvd
= pvd
->vdev_child
[c
];
5085 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5088 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5089 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5090 spa_strfree(cvd
->vdev_path
);
5091 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5098 * If we are detaching the original disk from a spare, then it implies
5099 * that the spare should become a real disk, and be removed from the
5100 * active spare list for the pool.
5102 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5104 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5108 * Erase the disk labels so the disk can be used for other things.
5109 * This must be done after all other error cases are handled,
5110 * but before we disembowel vd (so we can still do I/O to it).
5111 * But if we can't do it, don't treat the error as fatal --
5112 * it may be that the unwritability of the disk is the reason
5113 * it's being detached!
5115 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5118 * Remove vd from its parent and compact the parent's children.
5120 vdev_remove_child(pvd
, vd
);
5121 vdev_compact_children(pvd
);
5124 * Remember one of the remaining children so we can get tvd below.
5126 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5129 * If we need to remove the remaining child from the list of hot spares,
5130 * do it now, marking the vdev as no longer a spare in the process.
5131 * We must do this before vdev_remove_parent(), because that can
5132 * change the GUID if it creates a new toplevel GUID. For a similar
5133 * reason, we must remove the spare now, in the same txg as the detach;
5134 * otherwise someone could attach a new sibling, change the GUID, and
5135 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5138 ASSERT(cvd
->vdev_isspare
);
5139 spa_spare_remove(cvd
);
5140 unspare_guid
= cvd
->vdev_guid
;
5141 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5142 cvd
->vdev_unspare
= B_TRUE
;
5146 * If the parent mirror/replacing vdev only has one child,
5147 * the parent is no longer needed. Remove it from the tree.
5149 if (pvd
->vdev_children
== 1) {
5150 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5151 cvd
->vdev_unspare
= B_FALSE
;
5152 vdev_remove_parent(cvd
);
5157 * We don't set tvd until now because the parent we just removed
5158 * may have been the previous top-level vdev.
5160 tvd
= cvd
->vdev_top
;
5161 ASSERT(tvd
->vdev_parent
== rvd
);
5164 * Reevaluate the parent vdev state.
5166 vdev_propagate_state(cvd
);
5169 * If the 'autoexpand' property is set on the pool then automatically
5170 * try to expand the size of the pool. For example if the device we
5171 * just detached was smaller than the others, it may be possible to
5172 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5173 * first so that we can obtain the updated sizes of the leaf vdevs.
5175 if (spa
->spa_autoexpand
) {
5177 vdev_expand(tvd
, txg
);
5180 vdev_config_dirty(tvd
);
5183 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5184 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5185 * But first make sure we're not on any *other* txg's DTL list, to
5186 * prevent vd from being accessed after it's freed.
5188 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
5189 for (t
= 0; t
< TXG_SIZE
; t
++)
5190 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5191 vd
->vdev_detached
= B_TRUE
;
5192 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5194 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5196 /* hang on to the spa before we release the lock */
5197 spa_open_ref(spa
, FTAG
);
5199 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5201 spa_history_log_internal(spa
, "detach", NULL
,
5203 spa_strfree(vdpath
);
5206 * If this was the removal of the original device in a hot spare vdev,
5207 * then we want to go through and remove the device from the hot spare
5208 * list of every other pool.
5211 spa_t
*altspa
= NULL
;
5213 mutex_enter(&spa_namespace_lock
);
5214 while ((altspa
= spa_next(altspa
)) != NULL
) {
5215 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5219 spa_open_ref(altspa
, FTAG
);
5220 mutex_exit(&spa_namespace_lock
);
5221 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5222 mutex_enter(&spa_namespace_lock
);
5223 spa_close(altspa
, FTAG
);
5225 mutex_exit(&spa_namespace_lock
);
5227 /* search the rest of the vdevs for spares to remove */
5228 spa_vdev_resilver_done(spa
);
5231 /* all done with the spa; OK to release */
5232 mutex_enter(&spa_namespace_lock
);
5233 spa_close(spa
, FTAG
);
5234 mutex_exit(&spa_namespace_lock
);
5240 * Split a set of devices from their mirrors, and create a new pool from them.
5243 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5244 nvlist_t
*props
, boolean_t exp
)
5247 uint64_t txg
, *glist
;
5249 uint_t c
, children
, lastlog
;
5250 nvlist_t
**child
, *nvl
, *tmp
;
5252 char *altroot
= NULL
;
5253 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5254 boolean_t activate_slog
;
5256 ASSERT(spa_writeable(spa
));
5258 txg
= spa_vdev_enter(spa
);
5260 /* clear the log and flush everything up to now */
5261 activate_slog
= spa_passivate_log(spa
);
5262 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5263 error
= spa_offline_log(spa
);
5264 txg
= spa_vdev_config_enter(spa
);
5267 spa_activate_log(spa
);
5270 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5272 /* check new spa name before going any further */
5273 if (spa_lookup(newname
) != NULL
)
5274 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5277 * scan through all the children to ensure they're all mirrors
5279 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5280 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5282 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5284 /* first, check to ensure we've got the right child count */
5285 rvd
= spa
->spa_root_vdev
;
5287 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5288 vdev_t
*vd
= rvd
->vdev_child
[c
];
5290 /* don't count the holes & logs as children */
5291 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5299 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5300 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5302 /* next, ensure no spare or cache devices are part of the split */
5303 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5304 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5305 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5307 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5308 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5310 /* then, loop over each vdev and validate it */
5311 for (c
= 0; c
< children
; c
++) {
5312 uint64_t is_hole
= 0;
5314 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5318 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5319 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5322 error
= SET_ERROR(EINVAL
);
5327 /* which disk is going to be split? */
5328 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5330 error
= SET_ERROR(EINVAL
);
5334 /* look it up in the spa */
5335 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5336 if (vml
[c
] == NULL
) {
5337 error
= SET_ERROR(ENODEV
);
5341 /* make sure there's nothing stopping the split */
5342 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5343 vml
[c
]->vdev_islog
||
5344 vml
[c
]->vdev_ishole
||
5345 vml
[c
]->vdev_isspare
||
5346 vml
[c
]->vdev_isl2cache
||
5347 !vdev_writeable(vml
[c
]) ||
5348 vml
[c
]->vdev_children
!= 0 ||
5349 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5350 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5351 error
= SET_ERROR(EINVAL
);
5355 if (vdev_dtl_required(vml
[c
])) {
5356 error
= SET_ERROR(EBUSY
);
5360 /* we need certain info from the top level */
5361 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5362 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5363 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5364 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5365 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5366 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5367 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5368 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5370 /* transfer per-vdev ZAPs */
5371 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
5372 VERIFY0(nvlist_add_uint64(child
[c
],
5373 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
5375 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
5376 VERIFY0(nvlist_add_uint64(child
[c
],
5377 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
5378 vml
[c
]->vdev_parent
->vdev_top_zap
));
5382 kmem_free(vml
, children
* sizeof (vdev_t
*));
5383 kmem_free(glist
, children
* sizeof (uint64_t));
5384 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5387 /* stop writers from using the disks */
5388 for (c
= 0; c
< children
; c
++) {
5390 vml
[c
]->vdev_offline
= B_TRUE
;
5392 vdev_reopen(spa
->spa_root_vdev
);
5395 * Temporarily record the splitting vdevs in the spa config. This
5396 * will disappear once the config is regenerated.
5398 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5399 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5400 glist
, children
) == 0);
5401 kmem_free(glist
, children
* sizeof (uint64_t));
5403 mutex_enter(&spa
->spa_props_lock
);
5404 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5406 mutex_exit(&spa
->spa_props_lock
);
5407 spa
->spa_config_splitting
= nvl
;
5408 vdev_config_dirty(spa
->spa_root_vdev
);
5410 /* configure and create the new pool */
5411 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5412 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5413 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5414 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5415 spa_version(spa
)) == 0);
5416 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5417 spa
->spa_config_txg
) == 0);
5418 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5419 spa_generate_guid(NULL
)) == 0);
5420 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
5421 (void) nvlist_lookup_string(props
,
5422 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5424 /* add the new pool to the namespace */
5425 newspa
= spa_add(newname
, config
, altroot
);
5426 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5427 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5428 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5430 /* release the spa config lock, retaining the namespace lock */
5431 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5433 if (zio_injection_enabled
)
5434 zio_handle_panic_injection(spa
, FTAG
, 1);
5436 spa_activate(newspa
, spa_mode_global
);
5437 spa_async_suspend(newspa
);
5439 /* create the new pool from the disks of the original pool */
5440 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5444 /* if that worked, generate a real config for the new pool */
5445 if (newspa
->spa_root_vdev
!= NULL
) {
5446 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5447 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5448 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5449 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5450 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5455 if (props
!= NULL
) {
5456 spa_configfile_set(newspa
, props
, B_FALSE
);
5457 error
= spa_prop_set(newspa
, props
);
5462 /* flush everything */
5463 txg
= spa_vdev_config_enter(newspa
);
5464 vdev_config_dirty(newspa
->spa_root_vdev
);
5465 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5467 if (zio_injection_enabled
)
5468 zio_handle_panic_injection(spa
, FTAG
, 2);
5470 spa_async_resume(newspa
);
5472 /* finally, update the original pool's config */
5473 txg
= spa_vdev_config_enter(spa
);
5474 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5475 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5478 for (c
= 0; c
< children
; c
++) {
5479 if (vml
[c
] != NULL
) {
5482 spa_history_log_internal(spa
, "detach", tx
,
5483 "vdev=%s", vml
[c
]->vdev_path
);
5488 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5489 vdev_config_dirty(spa
->spa_root_vdev
);
5490 spa
->spa_config_splitting
= NULL
;
5494 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5496 if (zio_injection_enabled
)
5497 zio_handle_panic_injection(spa
, FTAG
, 3);
5499 /* split is complete; log a history record */
5500 spa_history_log_internal(newspa
, "split", NULL
,
5501 "from pool %s", spa_name(spa
));
5503 kmem_free(vml
, children
* sizeof (vdev_t
*));
5505 /* if we're not going to mount the filesystems in userland, export */
5507 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5514 spa_deactivate(newspa
);
5517 txg
= spa_vdev_config_enter(spa
);
5519 /* re-online all offlined disks */
5520 for (c
= 0; c
< children
; c
++) {
5522 vml
[c
]->vdev_offline
= B_FALSE
;
5524 vdev_reopen(spa
->spa_root_vdev
);
5526 nvlist_free(spa
->spa_config_splitting
);
5527 spa
->spa_config_splitting
= NULL
;
5528 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5530 kmem_free(vml
, children
* sizeof (vdev_t
*));
5535 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5539 for (i
= 0; i
< count
; i
++) {
5542 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5545 if (guid
== target_guid
)
5553 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5554 nvlist_t
*dev_to_remove
)
5556 nvlist_t
**newdev
= NULL
;
5560 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5562 for (i
= 0, j
= 0; i
< count
; i
++) {
5563 if (dev
[i
] == dev_to_remove
)
5565 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5568 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5569 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5571 for (i
= 0; i
< count
- 1; i
++)
5572 nvlist_free(newdev
[i
]);
5575 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5579 * Evacuate the device.
5582 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5587 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5588 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5589 ASSERT(vd
== vd
->vdev_top
);
5592 * Evacuate the device. We don't hold the config lock as writer
5593 * since we need to do I/O but we do keep the
5594 * spa_namespace_lock held. Once this completes the device
5595 * should no longer have any blocks allocated on it.
5597 if (vd
->vdev_islog
) {
5598 if (vd
->vdev_stat
.vs_alloc
!= 0)
5599 error
= spa_offline_log(spa
);
5601 error
= SET_ERROR(ENOTSUP
);
5608 * The evacuation succeeded. Remove any remaining MOS metadata
5609 * associated with this vdev, and wait for these changes to sync.
5611 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5612 txg
= spa_vdev_config_enter(spa
);
5613 vd
->vdev_removing
= B_TRUE
;
5614 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5615 vdev_config_dirty(vd
);
5616 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5622 * Complete the removal by cleaning up the namespace.
5625 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5627 vdev_t
*rvd
= spa
->spa_root_vdev
;
5628 uint64_t id
= vd
->vdev_id
;
5629 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5631 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5632 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5633 ASSERT(vd
== vd
->vdev_top
);
5636 * Only remove any devices which are empty.
5638 if (vd
->vdev_stat
.vs_alloc
!= 0)
5641 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5643 if (list_link_active(&vd
->vdev_state_dirty_node
))
5644 vdev_state_clean(vd
);
5645 if (list_link_active(&vd
->vdev_config_dirty_node
))
5646 vdev_config_clean(vd
);
5651 vdev_compact_children(rvd
);
5653 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5654 vdev_add_child(rvd
, vd
);
5656 vdev_config_dirty(rvd
);
5659 * Reassess the health of our root vdev.
5665 * Remove a device from the pool -
5667 * Removing a device from the vdev namespace requires several steps
5668 * and can take a significant amount of time. As a result we use
5669 * the spa_vdev_config_[enter/exit] functions which allow us to
5670 * grab and release the spa_config_lock while still holding the namespace
5671 * lock. During each step the configuration is synced out.
5673 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5677 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5680 sysevent_t
*ev
= NULL
;
5681 metaslab_group_t
*mg
;
5682 nvlist_t
**spares
, **l2cache
, *nv
;
5684 uint_t nspares
, nl2cache
;
5686 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5688 ASSERT(spa_writeable(spa
));
5691 txg
= spa_vdev_enter(spa
);
5693 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5695 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5696 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5697 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5698 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5700 * Only remove the hot spare if it's not currently in use
5703 if (vd
== NULL
|| unspare
) {
5705 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5706 ev
= spa_event_create(spa
, vd
, NULL
,
5707 ESC_ZFS_VDEV_REMOVE_AUX
);
5708 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5709 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5710 spa_load_spares(spa
);
5711 spa
->spa_spares
.sav_sync
= B_TRUE
;
5713 error
= SET_ERROR(EBUSY
);
5715 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5716 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5717 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5718 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5720 * Cache devices can always be removed.
5722 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5723 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_AUX
);
5724 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5725 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5726 spa_load_l2cache(spa
);
5727 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5728 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5730 ASSERT(vd
== vd
->vdev_top
);
5735 * Stop allocating from this vdev.
5737 metaslab_group_passivate(mg
);
5740 * Wait for the youngest allocations and frees to sync,
5741 * and then wait for the deferral of those frees to finish.
5743 spa_vdev_config_exit(spa
, NULL
,
5744 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5747 * Attempt to evacuate the vdev.
5749 error
= spa_vdev_remove_evacuate(spa
, vd
);
5751 txg
= spa_vdev_config_enter(spa
);
5754 * If we couldn't evacuate the vdev, unwind.
5757 metaslab_group_activate(mg
);
5758 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5762 * Clean up the vdev namespace.
5764 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_DEV
);
5765 spa_vdev_remove_from_namespace(spa
, vd
);
5767 } else if (vd
!= NULL
) {
5769 * Normal vdevs cannot be removed (yet).
5771 error
= SET_ERROR(ENOTSUP
);
5774 * There is no vdev of any kind with the specified guid.
5776 error
= SET_ERROR(ENOENT
);
5780 error
= spa_vdev_exit(spa
, NULL
, txg
, error
);
5789 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5790 * currently spared, so we can detach it.
5793 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5795 vdev_t
*newvd
, *oldvd
;
5798 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5799 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5805 * Check for a completed replacement. We always consider the first
5806 * vdev in the list to be the oldest vdev, and the last one to be
5807 * the newest (see spa_vdev_attach() for how that works). In
5808 * the case where the newest vdev is faulted, we will not automatically
5809 * remove it after a resilver completes. This is OK as it will require
5810 * user intervention to determine which disk the admin wishes to keep.
5812 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5813 ASSERT(vd
->vdev_children
> 1);
5815 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5816 oldvd
= vd
->vdev_child
[0];
5818 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5819 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5820 !vdev_dtl_required(oldvd
))
5825 * Check for a completed resilver with the 'unspare' flag set.
5827 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5828 vdev_t
*first
= vd
->vdev_child
[0];
5829 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5831 if (last
->vdev_unspare
) {
5834 } else if (first
->vdev_unspare
) {
5841 if (oldvd
!= NULL
&&
5842 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5843 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5844 !vdev_dtl_required(oldvd
))
5848 * If there are more than two spares attached to a disk,
5849 * and those spares are not required, then we want to
5850 * attempt to free them up now so that they can be used
5851 * by other pools. Once we're back down to a single
5852 * disk+spare, we stop removing them.
5854 if (vd
->vdev_children
> 2) {
5855 newvd
= vd
->vdev_child
[1];
5857 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5858 vdev_dtl_empty(last
, DTL_MISSING
) &&
5859 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5860 !vdev_dtl_required(newvd
))
5869 spa_vdev_resilver_done(spa_t
*spa
)
5871 vdev_t
*vd
, *pvd
, *ppvd
;
5872 uint64_t guid
, sguid
, pguid
, ppguid
;
5874 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5876 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5877 pvd
= vd
->vdev_parent
;
5878 ppvd
= pvd
->vdev_parent
;
5879 guid
= vd
->vdev_guid
;
5880 pguid
= pvd
->vdev_guid
;
5881 ppguid
= ppvd
->vdev_guid
;
5884 * If we have just finished replacing a hot spared device, then
5885 * we need to detach the parent's first child (the original hot
5888 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5889 ppvd
->vdev_children
== 2) {
5890 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5891 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5893 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5895 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5896 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5898 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5900 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5903 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5907 * Update the stored path or FRU for this vdev.
5910 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5914 boolean_t sync
= B_FALSE
;
5916 ASSERT(spa_writeable(spa
));
5918 spa_vdev_state_enter(spa
, SCL_ALL
);
5920 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5921 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5923 if (!vd
->vdev_ops
->vdev_op_leaf
)
5924 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5927 if (strcmp(value
, vd
->vdev_path
) != 0) {
5928 spa_strfree(vd
->vdev_path
);
5929 vd
->vdev_path
= spa_strdup(value
);
5933 if (vd
->vdev_fru
== NULL
) {
5934 vd
->vdev_fru
= spa_strdup(value
);
5936 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5937 spa_strfree(vd
->vdev_fru
);
5938 vd
->vdev_fru
= spa_strdup(value
);
5943 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5947 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5949 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5953 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5955 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5959 * ==========================================================================
5961 * ==========================================================================
5964 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
5966 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5968 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5969 return (SET_ERROR(EBUSY
));
5971 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
5975 spa_scan_stop(spa_t
*spa
)
5977 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5978 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5979 return (SET_ERROR(EBUSY
));
5980 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
5984 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
5986 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5988 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
5989 return (SET_ERROR(ENOTSUP
));
5992 * If a resilver was requested, but there is no DTL on a
5993 * writeable leaf device, we have nothing to do.
5995 if (func
== POOL_SCAN_RESILVER
&&
5996 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
5997 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6001 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6005 * ==========================================================================
6006 * SPA async task processing
6007 * ==========================================================================
6011 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6015 if (vd
->vdev_remove_wanted
) {
6016 vd
->vdev_remove_wanted
= B_FALSE
;
6017 vd
->vdev_delayed_close
= B_FALSE
;
6018 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6021 * We want to clear the stats, but we don't want to do a full
6022 * vdev_clear() as that will cause us to throw away
6023 * degraded/faulted state as well as attempt to reopen the
6024 * device, all of which is a waste.
6026 vd
->vdev_stat
.vs_read_errors
= 0;
6027 vd
->vdev_stat
.vs_write_errors
= 0;
6028 vd
->vdev_stat
.vs_checksum_errors
= 0;
6030 vdev_state_dirty(vd
->vdev_top
);
6033 for (c
= 0; c
< vd
->vdev_children
; c
++)
6034 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6038 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6042 if (vd
->vdev_probe_wanted
) {
6043 vd
->vdev_probe_wanted
= B_FALSE
;
6044 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6047 for (c
= 0; c
< vd
->vdev_children
; c
++)
6048 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6052 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6056 if (!spa
->spa_autoexpand
)
6059 for (c
= 0; c
< vd
->vdev_children
; c
++) {
6060 vdev_t
*cvd
= vd
->vdev_child
[c
];
6061 spa_async_autoexpand(spa
, cvd
);
6064 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6067 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
6071 spa_async_thread(spa_t
*spa
)
6075 ASSERT(spa
->spa_sync_on
);
6077 mutex_enter(&spa
->spa_async_lock
);
6078 tasks
= spa
->spa_async_tasks
;
6079 spa
->spa_async_tasks
= 0;
6080 mutex_exit(&spa
->spa_async_lock
);
6083 * See if the config needs to be updated.
6085 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6086 uint64_t old_space
, new_space
;
6088 mutex_enter(&spa_namespace_lock
);
6089 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6090 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6091 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6092 mutex_exit(&spa_namespace_lock
);
6095 * If the pool grew as a result of the config update,
6096 * then log an internal history event.
6098 if (new_space
!= old_space
) {
6099 spa_history_log_internal(spa
, "vdev online", NULL
,
6100 "pool '%s' size: %llu(+%llu)",
6101 spa_name(spa
), new_space
, new_space
- old_space
);
6106 * See if any devices need to be marked REMOVED.
6108 if (tasks
& SPA_ASYNC_REMOVE
) {
6109 spa_vdev_state_enter(spa
, SCL_NONE
);
6110 spa_async_remove(spa
, spa
->spa_root_vdev
);
6111 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6112 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6113 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6114 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6115 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6118 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6119 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6120 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6121 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6125 * See if any devices need to be probed.
6127 if (tasks
& SPA_ASYNC_PROBE
) {
6128 spa_vdev_state_enter(spa
, SCL_NONE
);
6129 spa_async_probe(spa
, spa
->spa_root_vdev
);
6130 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6134 * If any devices are done replacing, detach them.
6136 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6137 spa_vdev_resilver_done(spa
);
6140 * Kick off a resilver.
6142 if (tasks
& SPA_ASYNC_RESILVER
)
6143 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6146 * Let the world know that we're done.
6148 mutex_enter(&spa
->spa_async_lock
);
6149 spa
->spa_async_thread
= NULL
;
6150 cv_broadcast(&spa
->spa_async_cv
);
6151 mutex_exit(&spa
->spa_async_lock
);
6156 spa_async_suspend(spa_t
*spa
)
6158 mutex_enter(&spa
->spa_async_lock
);
6159 spa
->spa_async_suspended
++;
6160 while (spa
->spa_async_thread
!= NULL
)
6161 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6162 mutex_exit(&spa
->spa_async_lock
);
6166 spa_async_resume(spa_t
*spa
)
6168 mutex_enter(&spa
->spa_async_lock
);
6169 ASSERT(spa
->spa_async_suspended
!= 0);
6170 spa
->spa_async_suspended
--;
6171 mutex_exit(&spa
->spa_async_lock
);
6175 spa_async_tasks_pending(spa_t
*spa
)
6177 uint_t non_config_tasks
;
6179 boolean_t config_task_suspended
;
6181 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6182 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6183 if (spa
->spa_ccw_fail_time
== 0) {
6184 config_task_suspended
= B_FALSE
;
6186 config_task_suspended
=
6187 (gethrtime() - spa
->spa_ccw_fail_time
) <
6188 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
6191 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6195 spa_async_dispatch(spa_t
*spa
)
6197 mutex_enter(&spa
->spa_async_lock
);
6198 if (spa_async_tasks_pending(spa
) &&
6199 !spa
->spa_async_suspended
&&
6200 spa
->spa_async_thread
== NULL
&&
6202 spa
->spa_async_thread
= thread_create(NULL
, 0,
6203 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6204 mutex_exit(&spa
->spa_async_lock
);
6208 spa_async_request(spa_t
*spa
, int task
)
6210 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6211 mutex_enter(&spa
->spa_async_lock
);
6212 spa
->spa_async_tasks
|= task
;
6213 mutex_exit(&spa
->spa_async_lock
);
6217 * ==========================================================================
6218 * SPA syncing routines
6219 * ==========================================================================
6223 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6226 bpobj_enqueue(bpo
, bp
, tx
);
6231 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6235 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6241 * Note: this simple function is not inlined to make it easier to dtrace the
6242 * amount of time spent syncing frees.
6245 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6247 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6248 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6249 VERIFY(zio_wait(zio
) == 0);
6253 * Note: this simple function is not inlined to make it easier to dtrace the
6254 * amount of time spent syncing deferred frees.
6257 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6259 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6260 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6261 spa_free_sync_cb
, zio
, tx
), ==, 0);
6262 VERIFY0(zio_wait(zio
));
6266 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6268 char *packed
= NULL
;
6273 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6276 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6277 * information. This avoids the dmu_buf_will_dirty() path and
6278 * saves us a pre-read to get data we don't actually care about.
6280 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6281 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6283 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6285 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6287 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6289 vmem_free(packed
, bufsize
);
6291 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6292 dmu_buf_will_dirty(db
, tx
);
6293 *(uint64_t *)db
->db_data
= nvsize
;
6294 dmu_buf_rele(db
, FTAG
);
6298 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6299 const char *config
, const char *entry
)
6309 * Update the MOS nvlist describing the list of available devices.
6310 * spa_validate_aux() will have already made sure this nvlist is
6311 * valid and the vdevs are labeled appropriately.
6313 if (sav
->sav_object
== 0) {
6314 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6315 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6316 sizeof (uint64_t), tx
);
6317 VERIFY(zap_update(spa
->spa_meta_objset
,
6318 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6319 &sav
->sav_object
, tx
) == 0);
6322 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6323 if (sav
->sav_count
== 0) {
6324 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6326 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6327 for (i
= 0; i
< sav
->sav_count
; i
++)
6328 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6329 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6330 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6331 sav
->sav_count
) == 0);
6332 for (i
= 0; i
< sav
->sav_count
; i
++)
6333 nvlist_free(list
[i
]);
6334 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6337 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6338 nvlist_free(nvroot
);
6340 sav
->sav_sync
= B_FALSE
;
6344 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6345 * The all-vdev ZAP must be empty.
6348 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6350 spa_t
*spa
= vd
->vdev_spa
;
6353 if (vd
->vdev_top_zap
!= 0) {
6354 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6355 vd
->vdev_top_zap
, tx
));
6357 if (vd
->vdev_leaf_zap
!= 0) {
6358 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6359 vd
->vdev_leaf_zap
, tx
));
6361 for (i
= 0; i
< vd
->vdev_children
; i
++) {
6362 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6367 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6372 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6373 * its config may not be dirty but we still need to build per-vdev ZAPs.
6374 * Similarly, if the pool is being assembled (e.g. after a split), we
6375 * need to rebuild the AVZ although the config may not be dirty.
6377 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6378 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6381 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6383 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6384 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6385 spa
->spa_all_vdev_zaps
!= 0);
6387 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6391 /* Make and build the new AVZ */
6392 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6393 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6394 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6396 /* Diff old AVZ with new one */
6397 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6398 spa
->spa_all_vdev_zaps
);
6399 zap_cursor_retrieve(&zc
, &za
) == 0;
6400 zap_cursor_advance(&zc
)) {
6401 uint64_t vdzap
= za
.za_first_integer
;
6402 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6405 * ZAP is listed in old AVZ but not in new one;
6408 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6413 zap_cursor_fini(&zc
);
6415 /* Destroy the old AVZ */
6416 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6417 spa
->spa_all_vdev_zaps
, tx
));
6419 /* Replace the old AVZ in the dir obj with the new one */
6420 VERIFY0(zap_update(spa
->spa_meta_objset
,
6421 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6422 sizeof (new_avz
), 1, &new_avz
, tx
));
6424 spa
->spa_all_vdev_zaps
= new_avz
;
6425 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6429 /* Walk through the AVZ and destroy all listed ZAPs */
6430 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6431 spa
->spa_all_vdev_zaps
);
6432 zap_cursor_retrieve(&zc
, &za
) == 0;
6433 zap_cursor_advance(&zc
)) {
6434 uint64_t zap
= za
.za_first_integer
;
6435 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6438 zap_cursor_fini(&zc
);
6440 /* Destroy and unlink the AVZ itself */
6441 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6442 spa
->spa_all_vdev_zaps
, tx
));
6443 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6444 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6445 spa
->spa_all_vdev_zaps
= 0;
6448 if (spa
->spa_all_vdev_zaps
== 0) {
6449 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6450 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6451 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6453 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6455 /* Create ZAPs for vdevs that don't have them. */
6456 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6458 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6459 dmu_tx_get_txg(tx
), B_FALSE
);
6462 * If we're upgrading the spa version then make sure that
6463 * the config object gets updated with the correct version.
6465 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6466 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6467 spa
->spa_uberblock
.ub_version
);
6469 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6471 nvlist_free(spa
->spa_config_syncing
);
6472 spa
->spa_config_syncing
= config
;
6474 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6478 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6480 uint64_t *versionp
= arg
;
6481 uint64_t version
= *versionp
;
6482 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6485 * Setting the version is special cased when first creating the pool.
6487 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6489 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6490 ASSERT(version
>= spa_version(spa
));
6492 spa
->spa_uberblock
.ub_version
= version
;
6493 vdev_config_dirty(spa
->spa_root_vdev
);
6494 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6498 * Set zpool properties.
6501 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6503 nvlist_t
*nvp
= arg
;
6504 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6505 objset_t
*mos
= spa
->spa_meta_objset
;
6506 nvpair_t
*elem
= NULL
;
6508 mutex_enter(&spa
->spa_props_lock
);
6510 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6512 char *strval
, *fname
;
6514 const char *propname
;
6515 zprop_type_t proptype
;
6518 prop
= zpool_name_to_prop(nvpair_name(elem
));
6519 switch ((int)prop
) {
6522 * We checked this earlier in spa_prop_validate().
6524 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6526 fname
= strchr(nvpair_name(elem
), '@') + 1;
6527 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6529 spa_feature_enable(spa
, fid
, tx
);
6530 spa_history_log_internal(spa
, "set", tx
,
6531 "%s=enabled", nvpair_name(elem
));
6534 case ZPOOL_PROP_VERSION
:
6535 intval
= fnvpair_value_uint64(elem
);
6537 * The version is synced separately before other
6538 * properties and should be correct by now.
6540 ASSERT3U(spa_version(spa
), >=, intval
);
6543 case ZPOOL_PROP_ALTROOT
:
6545 * 'altroot' is a non-persistent property. It should
6546 * have been set temporarily at creation or import time.
6548 ASSERT(spa
->spa_root
!= NULL
);
6551 case ZPOOL_PROP_READONLY
:
6552 case ZPOOL_PROP_CACHEFILE
:
6554 * 'readonly' and 'cachefile' are also non-persisitent
6558 case ZPOOL_PROP_COMMENT
:
6559 strval
= fnvpair_value_string(elem
);
6560 if (spa
->spa_comment
!= NULL
)
6561 spa_strfree(spa
->spa_comment
);
6562 spa
->spa_comment
= spa_strdup(strval
);
6564 * We need to dirty the configuration on all the vdevs
6565 * so that their labels get updated. It's unnecessary
6566 * to do this for pool creation since the vdev's
6567 * configuration has already been dirtied.
6569 if (tx
->tx_txg
!= TXG_INITIAL
)
6570 vdev_config_dirty(spa
->spa_root_vdev
);
6571 spa_history_log_internal(spa
, "set", tx
,
6572 "%s=%s", nvpair_name(elem
), strval
);
6576 * Set pool property values in the poolprops mos object.
6578 if (spa
->spa_pool_props_object
== 0) {
6579 spa
->spa_pool_props_object
=
6580 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6581 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6585 /* normalize the property name */
6586 propname
= zpool_prop_to_name(prop
);
6587 proptype
= zpool_prop_get_type(prop
);
6589 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6590 ASSERT(proptype
== PROP_TYPE_STRING
);
6591 strval
= fnvpair_value_string(elem
);
6592 VERIFY0(zap_update(mos
,
6593 spa
->spa_pool_props_object
, propname
,
6594 1, strlen(strval
) + 1, strval
, tx
));
6595 spa_history_log_internal(spa
, "set", tx
,
6596 "%s=%s", nvpair_name(elem
), strval
);
6597 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6598 intval
= fnvpair_value_uint64(elem
);
6600 if (proptype
== PROP_TYPE_INDEX
) {
6602 VERIFY0(zpool_prop_index_to_string(
6603 prop
, intval
, &unused
));
6605 VERIFY0(zap_update(mos
,
6606 spa
->spa_pool_props_object
, propname
,
6607 8, 1, &intval
, tx
));
6608 spa_history_log_internal(spa
, "set", tx
,
6609 "%s=%lld", nvpair_name(elem
), intval
);
6611 ASSERT(0); /* not allowed */
6615 case ZPOOL_PROP_DELEGATION
:
6616 spa
->spa_delegation
= intval
;
6618 case ZPOOL_PROP_BOOTFS
:
6619 spa
->spa_bootfs
= intval
;
6621 case ZPOOL_PROP_FAILUREMODE
:
6622 spa
->spa_failmode
= intval
;
6624 case ZPOOL_PROP_AUTOEXPAND
:
6625 spa
->spa_autoexpand
= intval
;
6626 if (tx
->tx_txg
!= TXG_INITIAL
)
6627 spa_async_request(spa
,
6628 SPA_ASYNC_AUTOEXPAND
);
6630 case ZPOOL_PROP_MULTIHOST
:
6631 spa
->spa_multihost
= intval
;
6633 case ZPOOL_PROP_DEDUPDITTO
:
6634 spa
->spa_dedup_ditto
= intval
;
6643 mutex_exit(&spa
->spa_props_lock
);
6647 * Perform one-time upgrade on-disk changes. spa_version() does not
6648 * reflect the new version this txg, so there must be no changes this
6649 * txg to anything that the upgrade code depends on after it executes.
6650 * Therefore this must be called after dsl_pool_sync() does the sync
6654 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6656 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6658 ASSERT(spa
->spa_sync_pass
== 1);
6660 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6662 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6663 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6664 dsl_pool_create_origin(dp
, tx
);
6666 /* Keeping the origin open increases spa_minref */
6667 spa
->spa_minref
+= 3;
6670 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6671 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6672 dsl_pool_upgrade_clones(dp
, tx
);
6675 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6676 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6677 dsl_pool_upgrade_dir_clones(dp
, tx
);
6679 /* Keeping the freedir open increases spa_minref */
6680 spa
->spa_minref
+= 3;
6683 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6684 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6685 spa_feature_create_zap_objects(spa
, tx
);
6689 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6690 * when possibility to use lz4 compression for metadata was added
6691 * Old pools that have this feature enabled must be upgraded to have
6692 * this feature active
6694 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6695 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6696 SPA_FEATURE_LZ4_COMPRESS
);
6697 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6698 SPA_FEATURE_LZ4_COMPRESS
);
6700 if (lz4_en
&& !lz4_ac
)
6701 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6705 * If we haven't written the salt, do so now. Note that the
6706 * feature may not be activated yet, but that's fine since
6707 * the presence of this ZAP entry is backwards compatible.
6709 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
6710 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
6711 VERIFY0(zap_add(spa
->spa_meta_objset
,
6712 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
6713 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
6714 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
6717 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6721 * Sync the specified transaction group. New blocks may be dirtied as
6722 * part of the process, so we iterate until it converges.
6725 spa_sync(spa_t
*spa
, uint64_t txg
)
6727 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6728 objset_t
*mos
= spa
->spa_meta_objset
;
6729 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6730 metaslab_class_t
*mc
;
6731 vdev_t
*rvd
= spa
->spa_root_vdev
;
6735 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
6736 zfs_vdev_queue_depth_pct
/ 100;
6737 uint64_t queue_depth_total
;
6740 VERIFY(spa_writeable(spa
));
6743 * Lock out configuration changes.
6745 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6747 spa
->spa_syncing_txg
= txg
;
6748 spa
->spa_sync_pass
= 0;
6750 mutex_enter(&spa
->spa_alloc_lock
);
6751 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
6752 mutex_exit(&spa
->spa_alloc_lock
);
6755 * If there are any pending vdev state changes, convert them
6756 * into config changes that go out with this transaction group.
6758 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6759 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6761 * We need the write lock here because, for aux vdevs,
6762 * calling vdev_config_dirty() modifies sav_config.
6763 * This is ugly and will become unnecessary when we
6764 * eliminate the aux vdev wart by integrating all vdevs
6765 * into the root vdev tree.
6767 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6768 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6769 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6770 vdev_state_clean(vd
);
6771 vdev_config_dirty(vd
);
6773 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6774 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6776 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6778 tx
= dmu_tx_create_assigned(dp
, txg
);
6780 spa
->spa_sync_starttime
= gethrtime();
6781 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6782 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
6783 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6784 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6787 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6788 * set spa_deflate if we have no raid-z vdevs.
6790 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6791 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6794 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6795 vd
= rvd
->vdev_child
[i
];
6796 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6799 if (i
== rvd
->vdev_children
) {
6800 spa
->spa_deflate
= TRUE
;
6801 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6802 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6803 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6808 * Set the top-level vdev's max queue depth. Evaluate each
6809 * top-level's async write queue depth in case it changed.
6810 * The max queue depth will not change in the middle of syncing
6813 queue_depth_total
= 0;
6814 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
6815 vdev_t
*tvd
= rvd
->vdev_child
[c
];
6816 metaslab_group_t
*mg
= tvd
->vdev_mg
;
6818 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
6819 !metaslab_group_initialized(mg
))
6823 * It is safe to do a lock-free check here because only async
6824 * allocations look at mg_max_alloc_queue_depth, and async
6825 * allocations all happen from spa_sync().
6827 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
6828 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
6829 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
6831 mc
= spa_normal_class(spa
);
6832 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
6833 mc
->mc_alloc_max_slots
= queue_depth_total
;
6834 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
6836 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
6837 max_queue_depth
* rvd
->vdev_children
);
6840 * Iterate to convergence.
6843 int pass
= ++spa
->spa_sync_pass
;
6845 spa_sync_config_object(spa
, tx
);
6846 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6847 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6848 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6849 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6850 spa_errlog_sync(spa
, txg
);
6851 dsl_pool_sync(dp
, txg
);
6853 if (pass
< zfs_sync_pass_deferred_free
) {
6854 spa_sync_frees(spa
, free_bpl
, tx
);
6857 * We can not defer frees in pass 1, because
6858 * we sync the deferred frees later in pass 1.
6860 ASSERT3U(pass
, >, 1);
6861 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6862 &spa
->spa_deferred_bpobj
, tx
);
6866 dsl_scan_sync(dp
, tx
);
6868 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6872 spa_sync_upgrades(spa
, tx
);
6874 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
6876 * Note: We need to check if the MOS is dirty
6877 * because we could have marked the MOS dirty
6878 * without updating the uberblock (e.g. if we
6879 * have sync tasks but no dirty user data). We
6880 * need to check the uberblock's rootbp because
6881 * it is updated if we have synced out dirty
6882 * data (though in this case the MOS will most
6883 * likely also be dirty due to second order
6884 * effects, we don't want to rely on that here).
6886 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
6887 !dmu_objset_is_dirty(mos
, txg
)) {
6889 * Nothing changed on the first pass,
6890 * therefore this TXG is a no-op. Avoid
6891 * syncing deferred frees, so that we
6892 * can keep this TXG as a no-op.
6894 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
6896 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6897 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
6900 spa_sync_deferred_frees(spa
, tx
);
6903 } while (dmu_objset_is_dirty(mos
, txg
));
6906 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
6908 * Make sure that the number of ZAPs for all the vdevs matches
6909 * the number of ZAPs in the per-vdev ZAP list. This only gets
6910 * called if the config is dirty; otherwise there may be
6911 * outstanding AVZ operations that weren't completed in
6912 * spa_sync_config_object.
6914 uint64_t all_vdev_zap_entry_count
;
6915 ASSERT0(zap_count(spa
->spa_meta_objset
,
6916 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
6917 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
6918 all_vdev_zap_entry_count
);
6923 * Rewrite the vdev configuration (which includes the uberblock)
6924 * to commit the transaction group.
6926 * If there are no dirty vdevs, we sync the uberblock to a few
6927 * random top-level vdevs that are known to be visible in the
6928 * config cache (see spa_vdev_add() for a complete description).
6929 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6933 * We hold SCL_STATE to prevent vdev open/close/etc.
6934 * while we're attempting to write the vdev labels.
6936 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6938 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6939 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6941 int children
= rvd
->vdev_children
;
6942 int c0
= spa_get_random(children
);
6944 for (c
= 0; c
< children
; c
++) {
6945 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6946 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6948 svd
[svdcount
++] = vd
;
6949 if (svdcount
== SPA_DVAS_PER_BP
)
6952 error
= vdev_config_sync(svd
, svdcount
, txg
);
6954 error
= vdev_config_sync(rvd
->vdev_child
,
6955 rvd
->vdev_children
, txg
);
6959 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6961 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6965 zio_suspend(spa
, NULL
);
6966 zio_resume_wait(spa
);
6970 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6971 spa
->spa_deadman_tqid
= 0;
6974 * Clear the dirty config list.
6976 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
6977 vdev_config_clean(vd
);
6980 * Now that the new config has synced transactionally,
6981 * let it become visible to the config cache.
6983 if (spa
->spa_config_syncing
!= NULL
) {
6984 spa_config_set(spa
, spa
->spa_config_syncing
);
6985 spa
->spa_config_txg
= txg
;
6986 spa
->spa_config_syncing
= NULL
;
6989 dsl_pool_sync_done(dp
, txg
);
6991 mutex_enter(&spa
->spa_alloc_lock
);
6992 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
6993 mutex_exit(&spa
->spa_alloc_lock
);
6996 * Update usable space statistics.
6998 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
6999 vdev_sync_done(vd
, txg
);
7001 spa_update_dspace(spa
);
7004 * It had better be the case that we didn't dirty anything
7005 * since vdev_config_sync().
7007 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7008 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7009 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7011 spa
->spa_sync_pass
= 0;
7014 * Update the last synced uberblock here. We want to do this at
7015 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7016 * will be guaranteed that all the processing associated with
7017 * that txg has been completed.
7019 spa
->spa_ubsync
= spa
->spa_uberblock
;
7020 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7022 spa_handle_ignored_writes(spa
);
7025 * If any async tasks have been requested, kick them off.
7027 spa_async_dispatch(spa
);
7031 * Sync all pools. We don't want to hold the namespace lock across these
7032 * operations, so we take a reference on the spa_t and drop the lock during the
7036 spa_sync_allpools(void)
7039 mutex_enter(&spa_namespace_lock
);
7040 while ((spa
= spa_next(spa
)) != NULL
) {
7041 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7042 !spa_writeable(spa
) || spa_suspended(spa
))
7044 spa_open_ref(spa
, FTAG
);
7045 mutex_exit(&spa_namespace_lock
);
7046 txg_wait_synced(spa_get_dsl(spa
), 0);
7047 mutex_enter(&spa_namespace_lock
);
7048 spa_close(spa
, FTAG
);
7050 mutex_exit(&spa_namespace_lock
);
7054 * ==========================================================================
7055 * Miscellaneous routines
7056 * ==========================================================================
7060 * Remove all pools in the system.
7068 * Remove all cached state. All pools should be closed now,
7069 * so every spa in the AVL tree should be unreferenced.
7071 mutex_enter(&spa_namespace_lock
);
7072 while ((spa
= spa_next(NULL
)) != NULL
) {
7074 * Stop async tasks. The async thread may need to detach
7075 * a device that's been replaced, which requires grabbing
7076 * spa_namespace_lock, so we must drop it here.
7078 spa_open_ref(spa
, FTAG
);
7079 mutex_exit(&spa_namespace_lock
);
7080 spa_async_suspend(spa
);
7081 mutex_enter(&spa_namespace_lock
);
7082 spa_close(spa
, FTAG
);
7084 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7086 spa_deactivate(spa
);
7090 mutex_exit(&spa_namespace_lock
);
7094 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7099 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7103 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7104 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7105 if (vd
->vdev_guid
== guid
)
7109 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7110 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7111 if (vd
->vdev_guid
== guid
)
7120 spa_upgrade(spa_t
*spa
, uint64_t version
)
7122 ASSERT(spa_writeable(spa
));
7124 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7127 * This should only be called for a non-faulted pool, and since a
7128 * future version would result in an unopenable pool, this shouldn't be
7131 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7132 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7134 spa
->spa_uberblock
.ub_version
= version
;
7135 vdev_config_dirty(spa
->spa_root_vdev
);
7137 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7139 txg_wait_synced(spa_get_dsl(spa
), 0);
7143 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7147 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7149 for (i
= 0; i
< sav
->sav_count
; i
++)
7150 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7153 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7154 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7155 &spareguid
) == 0 && spareguid
== guid
)
7163 * Check if a pool has an active shared spare device.
7164 * Note: reference count of an active spare is 2, as a spare and as a replace
7167 spa_has_active_shared_spare(spa_t
*spa
)
7171 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7173 for (i
= 0; i
< sav
->sav_count
; i
++) {
7174 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7175 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7184 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7186 sysevent_t
*ev
= NULL
;
7190 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
7192 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
7193 ev
->resource
= resource
;
7200 spa_event_post(sysevent_t
*ev
)
7204 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
7205 kmem_free(ev
, sizeof (*ev
));
7211 * Post a zevent corresponding to the given sysevent. The 'name' must be one
7212 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
7213 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7214 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7215 * or zdb as real changes.
7218 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7220 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
7223 #if defined(_KERNEL) && defined(HAVE_SPL)
7224 /* state manipulation functions */
7225 EXPORT_SYMBOL(spa_open
);
7226 EXPORT_SYMBOL(spa_open_rewind
);
7227 EXPORT_SYMBOL(spa_get_stats
);
7228 EXPORT_SYMBOL(spa_create
);
7229 EXPORT_SYMBOL(spa_import
);
7230 EXPORT_SYMBOL(spa_tryimport
);
7231 EXPORT_SYMBOL(spa_destroy
);
7232 EXPORT_SYMBOL(spa_export
);
7233 EXPORT_SYMBOL(spa_reset
);
7234 EXPORT_SYMBOL(spa_async_request
);
7235 EXPORT_SYMBOL(spa_async_suspend
);
7236 EXPORT_SYMBOL(spa_async_resume
);
7237 EXPORT_SYMBOL(spa_inject_addref
);
7238 EXPORT_SYMBOL(spa_inject_delref
);
7239 EXPORT_SYMBOL(spa_scan_stat_init
);
7240 EXPORT_SYMBOL(spa_scan_get_stats
);
7242 /* device maniion */
7243 EXPORT_SYMBOL(spa_vdev_add
);
7244 EXPORT_SYMBOL(spa_vdev_attach
);
7245 EXPORT_SYMBOL(spa_vdev_detach
);
7246 EXPORT_SYMBOL(spa_vdev_remove
);
7247 EXPORT_SYMBOL(spa_vdev_setpath
);
7248 EXPORT_SYMBOL(spa_vdev_setfru
);
7249 EXPORT_SYMBOL(spa_vdev_split_mirror
);
7251 /* spare statech is global across all pools) */
7252 EXPORT_SYMBOL(spa_spare_add
);
7253 EXPORT_SYMBOL(spa_spare_remove
);
7254 EXPORT_SYMBOL(spa_spare_exists
);
7255 EXPORT_SYMBOL(spa_spare_activate
);
7257 /* L2ARC statech is global across all pools) */
7258 EXPORT_SYMBOL(spa_l2cache_add
);
7259 EXPORT_SYMBOL(spa_l2cache_remove
);
7260 EXPORT_SYMBOL(spa_l2cache_exists
);
7261 EXPORT_SYMBOL(spa_l2cache_activate
);
7262 EXPORT_SYMBOL(spa_l2cache_drop
);
7265 EXPORT_SYMBOL(spa_scan
);
7266 EXPORT_SYMBOL(spa_scan_stop
);
7269 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
7270 EXPORT_SYMBOL(spa_sync_allpools
);
7273 EXPORT_SYMBOL(spa_prop_set
);
7274 EXPORT_SYMBOL(spa_prop_get
);
7275 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
7277 /* asynchronous event notification */
7278 EXPORT_SYMBOL(spa_event_notify
);
7281 #if defined(_KERNEL) && defined(HAVE_SPL)
7282 module_param(spa_load_verify_maxinflight
, int, 0644);
7283 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
7284 "Max concurrent traversal I/Os while verifying pool during import -X");
7286 module_param(spa_load_verify_metadata
, int, 0644);
7287 MODULE_PARM_DESC(spa_load_verify_metadata
,
7288 "Set to traverse metadata on pool import");
7290 module_param(spa_load_verify_data
, int, 0644);
7291 MODULE_PARM_DESC(spa_load_verify_data
,
7292 "Set to traverse data on pool import");
7295 module_param(zio_taskq_batch_pct
, uint
, 0444);
7296 MODULE_PARM_DESC(zio_taskq_batch_pct
,
7297 "Percentage of CPUs to run an IO worker thread");