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 * Taskq dedicated to prefetcher threads: this is used to prevent the
1187 * pool traverse code from monopolizing the global (and limited)
1188 * system_taskq by inappropriately scheduling long running tasks on it.
1190 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", boot_ncpus
,
1191 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1194 * The taskq to upgrade datasets in this pool. Currently used by
1195 * feature SPA_FEATURE_USEROBJ_ACCOUNTING.
1197 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", boot_ncpus
,
1198 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1202 * Opposite of spa_activate().
1205 spa_deactivate(spa_t
*spa
)
1209 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1210 ASSERT(spa
->spa_dsl_pool
== NULL
);
1211 ASSERT(spa
->spa_root_vdev
== NULL
);
1212 ASSERT(spa
->spa_async_zio_root
== NULL
);
1213 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1215 spa_evicting_os_wait(spa
);
1217 if (spa
->spa_zvol_taskq
) {
1218 taskq_destroy(spa
->spa_zvol_taskq
);
1219 spa
->spa_zvol_taskq
= NULL
;
1222 if (spa
->spa_prefetch_taskq
) {
1223 taskq_destroy(spa
->spa_prefetch_taskq
);
1224 spa
->spa_prefetch_taskq
= NULL
;
1227 if (spa
->spa_upgrade_taskq
) {
1228 taskq_destroy(spa
->spa_upgrade_taskq
);
1229 spa
->spa_upgrade_taskq
= NULL
;
1232 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1234 list_destroy(&spa
->spa_config_dirty_list
);
1235 list_destroy(&spa
->spa_evicting_os_list
);
1236 list_destroy(&spa
->spa_state_dirty_list
);
1238 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1240 for (t
= 0; t
< ZIO_TYPES
; t
++) {
1241 for (q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1242 spa_taskqs_fini(spa
, t
, q
);
1246 metaslab_class_destroy(spa
->spa_normal_class
);
1247 spa
->spa_normal_class
= NULL
;
1249 metaslab_class_destroy(spa
->spa_log_class
);
1250 spa
->spa_log_class
= NULL
;
1253 * If this was part of an import or the open otherwise failed, we may
1254 * still have errors left in the queues. Empty them just in case.
1256 spa_errlog_drain(spa
);
1258 avl_destroy(&spa
->spa_errlist_scrub
);
1259 avl_destroy(&spa
->spa_errlist_last
);
1261 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1263 mutex_enter(&spa
->spa_proc_lock
);
1264 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1265 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1266 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1267 cv_broadcast(&spa
->spa_proc_cv
);
1268 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1269 ASSERT(spa
->spa_proc
!= &p0
);
1270 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1272 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1273 spa
->spa_proc_state
= SPA_PROC_NONE
;
1275 ASSERT(spa
->spa_proc
== &p0
);
1276 mutex_exit(&spa
->spa_proc_lock
);
1279 * We want to make sure spa_thread() has actually exited the ZFS
1280 * module, so that the module can't be unloaded out from underneath
1283 if (spa
->spa_did
!= 0) {
1284 thread_join(spa
->spa_did
);
1290 * Verify a pool configuration, and construct the vdev tree appropriately. This
1291 * will create all the necessary vdevs in the appropriate layout, with each vdev
1292 * in the CLOSED state. This will prep the pool before open/creation/import.
1293 * All vdev validation is done by the vdev_alloc() routine.
1296 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1297 uint_t id
, int atype
)
1304 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1307 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1310 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1313 if (error
== ENOENT
)
1319 return (SET_ERROR(EINVAL
));
1322 for (c
= 0; c
< children
; c
++) {
1324 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1332 ASSERT(*vdp
!= NULL
);
1338 * Opposite of spa_load().
1341 spa_unload(spa_t
*spa
)
1345 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1350 spa_async_suspend(spa
);
1355 if (spa
->spa_sync_on
) {
1356 txg_sync_stop(spa
->spa_dsl_pool
);
1357 spa
->spa_sync_on
= B_FALSE
;
1361 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1362 * to call it earlier, before we wait for async i/o to complete.
1363 * This ensures that there is no async metaslab prefetching, by
1364 * calling taskq_wait(mg_taskq).
1366 if (spa
->spa_root_vdev
!= NULL
) {
1367 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1368 for (c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++)
1369 vdev_metaslab_fini(spa
->spa_root_vdev
->vdev_child
[c
]);
1370 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1373 if (spa
->spa_mmp
.mmp_thread
)
1374 mmp_thread_stop(spa
);
1377 * Wait for any outstanding async I/O to complete.
1379 if (spa
->spa_async_zio_root
!= NULL
) {
1380 for (i
= 0; i
< max_ncpus
; i
++)
1381 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1382 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1383 spa
->spa_async_zio_root
= NULL
;
1386 bpobj_close(&spa
->spa_deferred_bpobj
);
1388 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1393 if (spa
->spa_root_vdev
)
1394 vdev_free(spa
->spa_root_vdev
);
1395 ASSERT(spa
->spa_root_vdev
== NULL
);
1398 * Close the dsl pool.
1400 if (spa
->spa_dsl_pool
) {
1401 dsl_pool_close(spa
->spa_dsl_pool
);
1402 spa
->spa_dsl_pool
= NULL
;
1403 spa
->spa_meta_objset
= NULL
;
1409 * Drop and purge level 2 cache
1411 spa_l2cache_drop(spa
);
1413 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1414 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1415 if (spa
->spa_spares
.sav_vdevs
) {
1416 kmem_free(spa
->spa_spares
.sav_vdevs
,
1417 spa
->spa_spares
.sav_count
* sizeof (void *));
1418 spa
->spa_spares
.sav_vdevs
= NULL
;
1420 if (spa
->spa_spares
.sav_config
) {
1421 nvlist_free(spa
->spa_spares
.sav_config
);
1422 spa
->spa_spares
.sav_config
= NULL
;
1424 spa
->spa_spares
.sav_count
= 0;
1426 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1427 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1428 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1430 if (spa
->spa_l2cache
.sav_vdevs
) {
1431 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1432 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1433 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1435 if (spa
->spa_l2cache
.sav_config
) {
1436 nvlist_free(spa
->spa_l2cache
.sav_config
);
1437 spa
->spa_l2cache
.sav_config
= NULL
;
1439 spa
->spa_l2cache
.sav_count
= 0;
1441 spa
->spa_async_suspended
= 0;
1443 if (spa
->spa_comment
!= NULL
) {
1444 spa_strfree(spa
->spa_comment
);
1445 spa
->spa_comment
= NULL
;
1448 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1452 * Load (or re-load) the current list of vdevs describing the active spares for
1453 * this pool. When this is called, we have some form of basic information in
1454 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1455 * then re-generate a more complete list including status information.
1458 spa_load_spares(spa_t
*spa
)
1465 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1468 * First, close and free any existing spare vdevs.
1470 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1471 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1473 /* Undo the call to spa_activate() below */
1474 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1475 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1476 spa_spare_remove(tvd
);
1481 if (spa
->spa_spares
.sav_vdevs
)
1482 kmem_free(spa
->spa_spares
.sav_vdevs
,
1483 spa
->spa_spares
.sav_count
* sizeof (void *));
1485 if (spa
->spa_spares
.sav_config
== NULL
)
1488 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1489 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1491 spa
->spa_spares
.sav_count
= (int)nspares
;
1492 spa
->spa_spares
.sav_vdevs
= NULL
;
1498 * Construct the array of vdevs, opening them to get status in the
1499 * process. For each spare, there is potentially two different vdev_t
1500 * structures associated with it: one in the list of spares (used only
1501 * for basic validation purposes) and one in the active vdev
1502 * configuration (if it's spared in). During this phase we open and
1503 * validate each vdev on the spare list. If the vdev also exists in the
1504 * active configuration, then we also mark this vdev as an active spare.
1506 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1508 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1509 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1510 VDEV_ALLOC_SPARE
) == 0);
1513 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1515 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1516 B_FALSE
)) != NULL
) {
1517 if (!tvd
->vdev_isspare
)
1521 * We only mark the spare active if we were successfully
1522 * able to load the vdev. Otherwise, importing a pool
1523 * with a bad active spare would result in strange
1524 * behavior, because multiple pool would think the spare
1525 * is actively in use.
1527 * There is a vulnerability here to an equally bizarre
1528 * circumstance, where a dead active spare is later
1529 * brought back to life (onlined or otherwise). Given
1530 * the rarity of this scenario, and the extra complexity
1531 * it adds, we ignore the possibility.
1533 if (!vdev_is_dead(tvd
))
1534 spa_spare_activate(tvd
);
1538 vd
->vdev_aux
= &spa
->spa_spares
;
1540 if (vdev_open(vd
) != 0)
1543 if (vdev_validate_aux(vd
) == 0)
1548 * Recompute the stashed list of spares, with status information
1551 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1552 DATA_TYPE_NVLIST_ARRAY
) == 0);
1554 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1556 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1557 spares
[i
] = vdev_config_generate(spa
,
1558 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1559 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1560 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1561 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1562 nvlist_free(spares
[i
]);
1563 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1567 * Load (or re-load) the current list of vdevs describing the active l2cache for
1568 * this pool. When this is called, we have some form of basic information in
1569 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1570 * then re-generate a more complete list including status information.
1571 * Devices which are already active have their details maintained, and are
1575 spa_load_l2cache(spa_t
*spa
)
1579 int i
, j
, oldnvdevs
;
1581 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1582 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1584 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1586 oldvdevs
= sav
->sav_vdevs
;
1587 oldnvdevs
= sav
->sav_count
;
1588 sav
->sav_vdevs
= NULL
;
1591 if (sav
->sav_config
== NULL
) {
1597 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1598 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1599 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1602 * Process new nvlist of vdevs.
1604 for (i
= 0; i
< nl2cache
; i
++) {
1605 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1609 for (j
= 0; j
< oldnvdevs
; j
++) {
1611 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1613 * Retain previous vdev for add/remove ops.
1621 if (newvdevs
[i
] == NULL
) {
1625 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1626 VDEV_ALLOC_L2CACHE
) == 0);
1631 * Commit this vdev as an l2cache device,
1632 * even if it fails to open.
1634 spa_l2cache_add(vd
);
1639 spa_l2cache_activate(vd
);
1641 if (vdev_open(vd
) != 0)
1644 (void) vdev_validate_aux(vd
);
1646 if (!vdev_is_dead(vd
))
1647 l2arc_add_vdev(spa
, vd
);
1651 sav
->sav_vdevs
= newvdevs
;
1652 sav
->sav_count
= (int)nl2cache
;
1655 * Recompute the stashed list of l2cache devices, with status
1656 * information this time.
1658 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1659 DATA_TYPE_NVLIST_ARRAY
) == 0);
1661 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1662 for (i
= 0; i
< sav
->sav_count
; i
++)
1663 l2cache
[i
] = vdev_config_generate(spa
,
1664 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1665 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1666 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1670 * Purge vdevs that were dropped
1672 for (i
= 0; i
< oldnvdevs
; i
++) {
1677 ASSERT(vd
->vdev_isl2cache
);
1679 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1680 pool
!= 0ULL && l2arc_vdev_present(vd
))
1681 l2arc_remove_vdev(vd
);
1682 vdev_clear_stats(vd
);
1688 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1690 for (i
= 0; i
< sav
->sav_count
; i
++)
1691 nvlist_free(l2cache
[i
]);
1693 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1697 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1700 char *packed
= NULL
;
1705 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1709 nvsize
= *(uint64_t *)db
->db_data
;
1710 dmu_buf_rele(db
, FTAG
);
1712 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1713 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1716 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1717 vmem_free(packed
, nvsize
);
1723 * Checks to see if the given vdev could not be opened, in which case we post a
1724 * sysevent to notify the autoreplace code that the device has been removed.
1727 spa_check_removed(vdev_t
*vd
)
1731 for (c
= 0; c
< vd
->vdev_children
; c
++)
1732 spa_check_removed(vd
->vdev_child
[c
]);
1734 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1736 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1737 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
1742 spa_config_valid_zaps(vdev_t
*vd
, vdev_t
*mvd
)
1746 ASSERT3U(vd
->vdev_children
, ==, mvd
->vdev_children
);
1748 vd
->vdev_top_zap
= mvd
->vdev_top_zap
;
1749 vd
->vdev_leaf_zap
= mvd
->vdev_leaf_zap
;
1751 for (i
= 0; i
< vd
->vdev_children
; i
++) {
1752 spa_config_valid_zaps(vd
->vdev_child
[i
], mvd
->vdev_child
[i
]);
1757 * Validate the current config against the MOS config
1760 spa_config_valid(spa_t
*spa
, nvlist_t
*config
)
1762 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
1766 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1768 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1769 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1771 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1774 * If we're doing a normal import, then build up any additional
1775 * diagnostic information about missing devices in this config.
1776 * We'll pass this up to the user for further processing.
1778 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1779 nvlist_t
**child
, *nv
;
1782 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
1784 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1786 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1787 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1788 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1790 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1791 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1793 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1798 VERIFY(nvlist_add_nvlist_array(nv
,
1799 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1800 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1801 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1803 for (i
= 0; i
< idx
; i
++)
1804 nvlist_free(child
[i
]);
1807 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1811 * Compare the root vdev tree with the information we have
1812 * from the MOS config (mrvd). Check each top-level vdev
1813 * with the corresponding MOS config top-level (mtvd).
1815 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1816 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1817 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1820 * Resolve any "missing" vdevs in the current configuration.
1821 * If we find that the MOS config has more accurate information
1822 * about the top-level vdev then use that vdev instead.
1824 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1825 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1827 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1831 * Device specific actions.
1833 if (mtvd
->vdev_islog
) {
1834 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1837 * XXX - once we have 'readonly' pool
1838 * support we should be able to handle
1839 * missing data devices by transitioning
1840 * the pool to readonly.
1846 * Swap the missing vdev with the data we were
1847 * able to obtain from the MOS config.
1849 vdev_remove_child(rvd
, tvd
);
1850 vdev_remove_child(mrvd
, mtvd
);
1852 vdev_add_child(rvd
, mtvd
);
1853 vdev_add_child(mrvd
, tvd
);
1855 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1857 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1861 if (mtvd
->vdev_islog
) {
1863 * Load the slog device's state from the MOS
1864 * config since it's possible that the label
1865 * does not contain the most up-to-date
1868 vdev_load_log_state(tvd
, mtvd
);
1873 * Per-vdev ZAP info is stored exclusively in the MOS.
1875 spa_config_valid_zaps(tvd
, mtvd
);
1880 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1883 * Ensure we were able to validate the config.
1885 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1889 * Check for missing log devices
1892 spa_check_logs(spa_t
*spa
)
1894 boolean_t rv
= B_FALSE
;
1895 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1897 switch (spa
->spa_log_state
) {
1900 case SPA_LOG_MISSING
:
1901 /* need to recheck in case slog has been restored */
1902 case SPA_LOG_UNKNOWN
:
1903 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1904 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1906 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1913 spa_passivate_log(spa_t
*spa
)
1915 vdev_t
*rvd
= spa
->spa_root_vdev
;
1916 boolean_t slog_found
= B_FALSE
;
1919 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1921 if (!spa_has_slogs(spa
))
1924 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1925 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1926 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1928 if (tvd
->vdev_islog
) {
1929 metaslab_group_passivate(mg
);
1930 slog_found
= B_TRUE
;
1934 return (slog_found
);
1938 spa_activate_log(spa_t
*spa
)
1940 vdev_t
*rvd
= spa
->spa_root_vdev
;
1943 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1945 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
1946 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1947 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1949 if (tvd
->vdev_islog
)
1950 metaslab_group_activate(mg
);
1955 spa_offline_log(spa_t
*spa
)
1959 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1960 NULL
, DS_FIND_CHILDREN
);
1963 * We successfully offlined the log device, sync out the
1964 * current txg so that the "stubby" block can be removed
1967 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1973 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1977 for (i
= 0; i
< sav
->sav_count
; i
++)
1978 spa_check_removed(sav
->sav_vdevs
[i
]);
1982 spa_claim_notify(zio_t
*zio
)
1984 spa_t
*spa
= zio
->io_spa
;
1989 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1990 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1991 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1992 mutex_exit(&spa
->spa_props_lock
);
1995 typedef struct spa_load_error
{
1996 uint64_t sle_meta_count
;
1997 uint64_t sle_data_count
;
2001 spa_load_verify_done(zio_t
*zio
)
2003 blkptr_t
*bp
= zio
->io_bp
;
2004 spa_load_error_t
*sle
= zio
->io_private
;
2005 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2006 int error
= zio
->io_error
;
2007 spa_t
*spa
= zio
->io_spa
;
2009 abd_free(zio
->io_abd
);
2011 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2012 type
!= DMU_OT_INTENT_LOG
)
2013 atomic_inc_64(&sle
->sle_meta_count
);
2015 atomic_inc_64(&sle
->sle_data_count
);
2018 mutex_enter(&spa
->spa_scrub_lock
);
2019 spa
->spa_scrub_inflight
--;
2020 cv_broadcast(&spa
->spa_scrub_io_cv
);
2021 mutex_exit(&spa
->spa_scrub_lock
);
2025 * Maximum number of concurrent scrub i/os to create while verifying
2026 * a pool while importing it.
2028 int spa_load_verify_maxinflight
= 10000;
2029 int spa_load_verify_metadata
= B_TRUE
;
2030 int spa_load_verify_data
= B_TRUE
;
2034 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2035 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2040 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
2043 * Note: normally this routine will not be called if
2044 * spa_load_verify_metadata is not set. However, it may be useful
2045 * to manually set the flag after the traversal has begun.
2047 if (!spa_load_verify_metadata
)
2049 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2053 size
= BP_GET_PSIZE(bp
);
2055 mutex_enter(&spa
->spa_scrub_lock
);
2056 while (spa
->spa_scrub_inflight
>= spa_load_verify_maxinflight
)
2057 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2058 spa
->spa_scrub_inflight
++;
2059 mutex_exit(&spa
->spa_scrub_lock
);
2061 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2062 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2063 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2064 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2070 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2072 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2073 return (SET_ERROR(ENAMETOOLONG
));
2079 spa_load_verify(spa_t
*spa
)
2082 spa_load_error_t sle
= { 0 };
2083 zpool_rewind_policy_t policy
;
2084 boolean_t verify_ok
= B_FALSE
;
2087 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
2089 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
2092 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2093 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2094 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2096 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2100 rio
= zio_root(spa
, NULL
, &sle
,
2101 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2103 if (spa_load_verify_metadata
) {
2104 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2105 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
,
2106 spa_load_verify_cb
, rio
);
2109 (void) zio_wait(rio
);
2111 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2112 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2114 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
2115 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
2119 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2120 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2122 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2123 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2124 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2125 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2126 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2127 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2128 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2130 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2134 if (error
!= ENXIO
&& error
!= EIO
)
2135 error
= SET_ERROR(EIO
);
2139 return (verify_ok
? 0 : EIO
);
2143 * Find a value in the pool props object.
2146 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2148 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2149 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2153 * Find a value in the pool directory object.
2156 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2158 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2159 name
, sizeof (uint64_t), 1, val
));
2163 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2165 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2170 * Fix up config after a partly-completed split. This is done with the
2171 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2172 * pool have that entry in their config, but only the splitting one contains
2173 * a list of all the guids of the vdevs that are being split off.
2175 * This function determines what to do with that list: either rejoin
2176 * all the disks to the pool, or complete the splitting process. To attempt
2177 * the rejoin, each disk that is offlined is marked online again, and
2178 * we do a reopen() call. If the vdev label for every disk that was
2179 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2180 * then we call vdev_split() on each disk, and complete the split.
2182 * Otherwise we leave the config alone, with all the vdevs in place in
2183 * the original pool.
2186 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2193 boolean_t attempt_reopen
;
2195 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2198 /* check that the config is complete */
2199 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2200 &glist
, &gcount
) != 0)
2203 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2205 /* attempt to online all the vdevs & validate */
2206 attempt_reopen
= B_TRUE
;
2207 for (i
= 0; i
< gcount
; i
++) {
2208 if (glist
[i
] == 0) /* vdev is hole */
2211 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2212 if (vd
[i
] == NULL
) {
2214 * Don't bother attempting to reopen the disks;
2215 * just do the split.
2217 attempt_reopen
= B_FALSE
;
2219 /* attempt to re-online it */
2220 vd
[i
]->vdev_offline
= B_FALSE
;
2224 if (attempt_reopen
) {
2225 vdev_reopen(spa
->spa_root_vdev
);
2227 /* check each device to see what state it's in */
2228 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2229 if (vd
[i
] != NULL
&&
2230 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2237 * If every disk has been moved to the new pool, or if we never
2238 * even attempted to look at them, then we split them off for
2241 if (!attempt_reopen
|| gcount
== extracted
) {
2242 for (i
= 0; i
< gcount
; i
++)
2245 vdev_reopen(spa
->spa_root_vdev
);
2248 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2252 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2253 boolean_t mosconfig
)
2255 nvlist_t
*config
= spa
->spa_config
;
2256 char *ereport
= FM_EREPORT_ZFS_POOL
;
2262 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2263 return (SET_ERROR(EINVAL
));
2265 ASSERT(spa
->spa_comment
== NULL
);
2266 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2267 spa
->spa_comment
= spa_strdup(comment
);
2270 * Versioning wasn't explicitly added to the label until later, so if
2271 * it's not present treat it as the initial version.
2273 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2274 &spa
->spa_ubsync
.ub_version
) != 0)
2275 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2277 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2278 &spa
->spa_config_txg
);
2280 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2281 spa_guid_exists(pool_guid
, 0)) {
2282 error
= SET_ERROR(EEXIST
);
2284 spa
->spa_config_guid
= pool_guid
;
2286 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2288 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2292 nvlist_free(spa
->spa_load_info
);
2293 spa
->spa_load_info
= fnvlist_alloc();
2295 gethrestime(&spa
->spa_loaded_ts
);
2296 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2297 mosconfig
, &ereport
);
2301 * Don't count references from objsets that are already closed
2302 * and are making their way through the eviction process.
2304 spa_evicting_os_wait(spa
);
2305 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2307 if (error
!= EEXIST
) {
2308 spa
->spa_loaded_ts
.tv_sec
= 0;
2309 spa
->spa_loaded_ts
.tv_nsec
= 0;
2311 if (error
!= EBADF
) {
2312 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2315 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2323 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2324 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2325 * spa's per-vdev ZAP list.
2328 vdev_count_verify_zaps(vdev_t
*vd
)
2330 spa_t
*spa
= vd
->vdev_spa
;
2334 if (vd
->vdev_top_zap
!= 0) {
2336 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2337 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2339 if (vd
->vdev_leaf_zap
!= 0) {
2341 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2342 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2345 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2346 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2354 * Determine whether the activity check is required.
2357 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
2361 uint64_t hostid
= 0;
2362 uint64_t tryconfig_txg
= 0;
2363 uint64_t tryconfig_timestamp
= 0;
2366 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2367 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2368 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2370 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2371 &tryconfig_timestamp
);
2374 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2377 * Disable the MMP activity check - This is used by zdb which
2378 * is intended to be used on potentially active pools.
2380 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
2384 * Skip the activity check when the MMP feature is disabled.
2386 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
2389 * If the tryconfig_* values are nonzero, they are the results of an
2390 * earlier tryimport. If they match the uberblock we just found, then
2391 * the pool has not changed and we return false so we do not test a
2394 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
2395 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
)
2399 * Allow the activity check to be skipped when importing the pool
2400 * on the same host which last imported it. Since the hostid from
2401 * configuration may be stale use the one read from the label.
2403 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
2404 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
2406 if (hostid
== spa_get_hostid())
2410 * Skip the activity test when the pool was cleanly exported.
2412 if (state
!= POOL_STATE_ACTIVE
)
2419 * Perform the import activity check. If the user canceled the import or
2420 * we detected activity then fail.
2423 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2425 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2426 uint64_t txg
= ub
->ub_txg
;
2427 uint64_t timestamp
= ub
->ub_timestamp
;
2428 uint64_t import_delay
= NANOSEC
;
2429 hrtime_t import_expire
;
2430 nvlist_t
*mmp_label
= NULL
;
2431 vdev_t
*rvd
= spa
->spa_root_vdev
;
2436 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
2437 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2441 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2442 * during the earlier tryimport. If the txg recorded there is 0 then
2443 * the pool is known to be active on another host.
2445 * Otherwise, the pool might be in use on another node. Check for
2446 * changes in the uberblocks on disk if necessary.
2448 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2449 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
2450 ZPOOL_CONFIG_LOAD_INFO
);
2452 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
2453 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
2454 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2455 error
= SET_ERROR(EREMOTEIO
);
2461 * Preferentially use the zfs_multihost_interval from the node which
2462 * last imported the pool. This value is stored in an MMP uberblock as.
2464 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2466 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
)
2467 import_delay
= MAX(import_delay
, import_intervals
*
2468 ub
->ub_mmp_delay
* MAX(vdev_count_leaves(spa
), 1));
2470 /* Apply a floor using the local default values. */
2471 import_delay
= MAX(import_delay
, import_intervals
*
2472 MSEC2NSEC(MAX(zfs_multihost_interval
, MMP_MIN_INTERVAL
)));
2474 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu import_intervals=%u "
2475 "leaves=%u", import_delay
, ub
->ub_mmp_delay
, import_intervals
,
2476 vdev_count_leaves(spa
));
2478 /* Add a small random factor in case of simultaneous imports (0-25%) */
2479 import_expire
= gethrtime() + import_delay
+
2480 (import_delay
* spa_get_random(250) / 1000);
2482 while (gethrtime() < import_expire
) {
2483 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2485 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
) {
2486 error
= SET_ERROR(EREMOTEIO
);
2491 nvlist_free(mmp_label
);
2495 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
2497 error
= SET_ERROR(EINTR
);
2505 mutex_destroy(&mtx
);
2509 * If the pool is determined to be active store the status in the
2510 * spa->spa_load_info nvlist. If the remote hostname or hostid are
2511 * available from configuration read from disk store them as well.
2512 * This allows 'zpool import' to generate a more useful message.
2514 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
2515 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2516 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
2518 if (error
== EREMOTEIO
) {
2519 char *hostname
= "<unknown>";
2520 uint64_t hostid
= 0;
2523 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
2524 hostname
= fnvlist_lookup_string(mmp_label
,
2525 ZPOOL_CONFIG_HOSTNAME
);
2526 fnvlist_add_string(spa
->spa_load_info
,
2527 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
2530 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
2531 hostid
= fnvlist_lookup_uint64(mmp_label
,
2532 ZPOOL_CONFIG_HOSTID
);
2533 fnvlist_add_uint64(spa
->spa_load_info
,
2534 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
2538 fnvlist_add_uint64(spa
->spa_load_info
,
2539 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
2540 fnvlist_add_uint64(spa
->spa_load_info
,
2541 ZPOOL_CONFIG_MMP_TXG
, 0);
2543 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
2547 nvlist_free(mmp_label
);
2553 * Load an existing storage pool, using the pool's builtin spa_config as a
2554 * source of configuration information.
2556 __attribute__((always_inline
))
2558 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2559 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2563 nvlist_t
*nvroot
= NULL
;
2566 uberblock_t
*ub
= &spa
->spa_uberblock
;
2567 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2568 int orig_mode
= spa
->spa_mode
;
2571 boolean_t missing_feat_write
= B_FALSE
;
2572 boolean_t activity_check
= B_FALSE
;
2573 nvlist_t
*mos_config
;
2576 * If this is an untrusted config, access the pool in read-only mode.
2577 * This prevents things like resilvering recently removed devices.
2580 spa
->spa_mode
= FREAD
;
2582 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2584 spa
->spa_load_state
= state
;
2586 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2587 return (SET_ERROR(EINVAL
));
2589 parse
= (type
== SPA_IMPORT_EXISTING
?
2590 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2593 * Create "The Godfather" zio to hold all async IOs
2595 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2597 for (i
= 0; i
< max_ncpus
; i
++) {
2598 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2599 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2600 ZIO_FLAG_GODFATHER
);
2604 * Parse the configuration into a vdev tree. We explicitly set the
2605 * value that will be returned by spa_version() since parsing the
2606 * configuration requires knowing the version number.
2608 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2609 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2610 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2615 ASSERT(spa
->spa_root_vdev
== rvd
);
2616 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2617 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2619 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2620 ASSERT(spa_guid(spa
) == pool_guid
);
2624 * Try to open all vdevs, loading each label in the process.
2626 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2627 error
= vdev_open(rvd
);
2628 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2633 * We need to validate the vdev labels against the configuration that
2634 * we have in hand, which is dependent on the setting of mosconfig. If
2635 * mosconfig is true then we're validating the vdev labels based on
2636 * that config. Otherwise, we're validating against the cached config
2637 * (zpool.cache) that was read when we loaded the zfs module, and then
2638 * later we will recursively call spa_load() and validate against
2641 * If we're assembling a new pool that's been split off from an
2642 * existing pool, the labels haven't yet been updated so we skip
2643 * validation for now.
2645 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2646 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2647 error
= vdev_validate(rvd
, mosconfig
);
2648 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2653 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2654 return (SET_ERROR(ENXIO
));
2658 * Find the best uberblock.
2660 vdev_uberblock_load(rvd
, ub
, &label
);
2663 * If we weren't able to find a single valid uberblock, return failure.
2665 if (ub
->ub_txg
== 0) {
2667 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2671 * For pools which have the multihost property on determine if the
2672 * pool is truly inactive and can be safely imported. Prevent
2673 * hosts which don't have a hostid set from importing the pool.
2675 activity_check
= spa_activity_check_required(spa
, ub
, label
, config
);
2676 if (activity_check
) {
2677 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
2678 spa_get_hostid() == 0) {
2680 fnvlist_add_uint64(spa
->spa_load_info
,
2681 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
2682 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
2685 error
= spa_activity_check(spa
, ub
, config
);
2691 fnvlist_add_uint64(spa
->spa_load_info
,
2692 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
2693 fnvlist_add_uint64(spa
->spa_load_info
,
2694 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
2698 * If the pool has an unsupported version we can't open it.
2700 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2702 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2705 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2709 * If we weren't able to find what's necessary for reading the
2710 * MOS in the label, return failure.
2712 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2713 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2715 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2720 * Update our in-core representation with the definitive values
2723 nvlist_free(spa
->spa_label_features
);
2724 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2730 * Look through entries in the label nvlist's features_for_read. If
2731 * there is a feature listed there which we don't understand then we
2732 * cannot open a pool.
2734 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2735 nvlist_t
*unsup_feat
;
2738 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2741 for (nvp
= nvlist_next_nvpair(spa
->spa_label_features
, NULL
);
2743 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2744 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2745 VERIFY(nvlist_add_string(unsup_feat
,
2746 nvpair_name(nvp
), "") == 0);
2750 if (!nvlist_empty(unsup_feat
)) {
2751 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2752 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2753 nvlist_free(unsup_feat
);
2754 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2758 nvlist_free(unsup_feat
);
2762 * If the vdev guid sum doesn't match the uberblock, we have an
2763 * incomplete configuration. We first check to see if the pool
2764 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2765 * If it is, defer the vdev_guid_sum check till later so we
2766 * can handle missing vdevs.
2768 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2769 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2770 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2771 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2773 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2774 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2775 spa_try_repair(spa
, config
);
2776 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2777 nvlist_free(spa
->spa_config_splitting
);
2778 spa
->spa_config_splitting
= NULL
;
2782 * Initialize internal SPA structures.
2784 spa
->spa_state
= POOL_STATE_ACTIVE
;
2785 spa
->spa_ubsync
= spa
->spa_uberblock
;
2786 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2787 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2788 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2789 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2790 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2791 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2793 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2795 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2796 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2798 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2799 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2801 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2802 boolean_t missing_feat_read
= B_FALSE
;
2803 nvlist_t
*unsup_feat
, *enabled_feat
;
2806 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2807 &spa
->spa_feat_for_read_obj
) != 0) {
2808 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2811 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2812 &spa
->spa_feat_for_write_obj
) != 0) {
2813 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2816 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2817 &spa
->spa_feat_desc_obj
) != 0) {
2818 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2821 enabled_feat
= fnvlist_alloc();
2822 unsup_feat
= fnvlist_alloc();
2824 if (!spa_features_check(spa
, B_FALSE
,
2825 unsup_feat
, enabled_feat
))
2826 missing_feat_read
= B_TRUE
;
2828 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2829 if (!spa_features_check(spa
, B_TRUE
,
2830 unsup_feat
, enabled_feat
)) {
2831 missing_feat_write
= B_TRUE
;
2835 fnvlist_add_nvlist(spa
->spa_load_info
,
2836 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2838 if (!nvlist_empty(unsup_feat
)) {
2839 fnvlist_add_nvlist(spa
->spa_load_info
,
2840 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2843 fnvlist_free(enabled_feat
);
2844 fnvlist_free(unsup_feat
);
2846 if (!missing_feat_read
) {
2847 fnvlist_add_boolean(spa
->spa_load_info
,
2848 ZPOOL_CONFIG_CAN_RDONLY
);
2852 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2853 * twofold: to determine whether the pool is available for
2854 * import in read-write mode and (if it is not) whether the
2855 * pool is available for import in read-only mode. If the pool
2856 * is available for import in read-write mode, it is displayed
2857 * as available in userland; if it is not available for import
2858 * in read-only mode, it is displayed as unavailable in
2859 * userland. If the pool is available for import in read-only
2860 * mode but not read-write mode, it is displayed as unavailable
2861 * in userland with a special note that the pool is actually
2862 * available for open in read-only mode.
2864 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2865 * missing a feature for write, we must first determine whether
2866 * the pool can be opened read-only before returning to
2867 * userland in order to know whether to display the
2868 * abovementioned note.
2870 if (missing_feat_read
|| (missing_feat_write
&&
2871 spa_writeable(spa
))) {
2872 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2877 * Load refcounts for ZFS features from disk into an in-memory
2878 * cache during SPA initialization.
2880 for (i
= 0; i
< SPA_FEATURES
; i
++) {
2883 error
= feature_get_refcount_from_disk(spa
,
2884 &spa_feature_table
[i
], &refcount
);
2886 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2887 } else if (error
== ENOTSUP
) {
2888 spa
->spa_feat_refcount_cache
[i
] =
2889 SPA_FEATURE_DISABLED
;
2891 return (spa_vdev_err(rvd
,
2892 VDEV_AUX_CORRUPT_DATA
, EIO
));
2897 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2898 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2899 &spa
->spa_feat_enabled_txg_obj
) != 0)
2900 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2903 spa
->spa_is_initializing
= B_TRUE
;
2904 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2905 spa
->spa_is_initializing
= B_FALSE
;
2907 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2911 nvlist_t
*policy
= NULL
, *nvconfig
;
2913 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2914 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2916 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2917 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2919 unsigned long myhostid
= 0;
2921 VERIFY(nvlist_lookup_string(nvconfig
,
2922 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2924 myhostid
= spa_get_hostid();
2925 if (hostid
&& myhostid
&& hostid
!= myhostid
) {
2926 nvlist_free(nvconfig
);
2927 return (SET_ERROR(EBADF
));
2930 if (nvlist_lookup_nvlist(spa
->spa_config
,
2931 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2932 VERIFY(nvlist_add_nvlist(nvconfig
,
2933 ZPOOL_REWIND_POLICY
, policy
) == 0);
2935 spa_config_set(spa
, nvconfig
);
2937 spa_deactivate(spa
);
2938 spa_activate(spa
, orig_mode
);
2940 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2943 /* Grab the checksum salt from the MOS. */
2944 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2945 DMU_POOL_CHECKSUM_SALT
, 1,
2946 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
2947 spa
->spa_cksum_salt
.zcs_bytes
);
2948 if (error
== ENOENT
) {
2949 /* Generate a new salt for subsequent use */
2950 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
2951 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
2952 } else if (error
!= 0) {
2953 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2956 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2957 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2958 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2960 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2963 * Load the bit that tells us to use the new accounting function
2964 * (raid-z deflation). If we have an older pool, this will not
2967 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2968 if (error
!= 0 && error
!= ENOENT
)
2969 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2971 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2972 &spa
->spa_creation_version
);
2973 if (error
!= 0 && error
!= ENOENT
)
2974 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2977 * Load the persistent error log. If we have an older pool, this will
2980 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2981 if (error
!= 0 && error
!= ENOENT
)
2982 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2984 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2985 &spa
->spa_errlog_scrub
);
2986 if (error
!= 0 && error
!= ENOENT
)
2987 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2990 * Load the history object. If we have an older pool, this
2991 * will not be present.
2993 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2994 if (error
!= 0 && error
!= ENOENT
)
2995 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2998 * Load the per-vdev ZAP map. If we have an older pool, this will not
2999 * be present; in this case, defer its creation to a later time to
3000 * avoid dirtying the MOS this early / out of sync context. See
3001 * spa_sync_config_object.
3004 /* The sentinel is only available in the MOS config. */
3005 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
3006 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3008 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
3009 &spa
->spa_all_vdev_zaps
);
3011 if (error
== ENOENT
) {
3012 VERIFY(!nvlist_exists(mos_config
,
3013 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
3014 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
3015 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3016 } else if (error
!= 0) {
3017 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3018 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
3020 * An older version of ZFS overwrote the sentinel value, so
3021 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3022 * destruction to later; see spa_sync_config_object.
3024 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
3026 * We're assuming that no vdevs have had their ZAPs created
3027 * before this. Better be sure of it.
3029 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3031 nvlist_free(mos_config
);
3034 * If we're assembling the pool from the split-off vdevs of
3035 * an existing pool, we don't want to attach the spares & cache
3040 * Load any hot spares for this pool.
3042 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
3043 if (error
!= 0 && error
!= ENOENT
)
3044 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3045 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3046 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3047 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3048 &spa
->spa_spares
.sav_config
) != 0)
3049 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3051 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3052 spa_load_spares(spa
);
3053 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3054 } else if (error
== 0) {
3055 spa
->spa_spares
.sav_sync
= B_TRUE
;
3059 * Load any level 2 ARC devices for this pool.
3061 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3062 &spa
->spa_l2cache
.sav_object
);
3063 if (error
!= 0 && error
!= ENOENT
)
3064 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3065 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3066 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3067 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3068 &spa
->spa_l2cache
.sav_config
) != 0)
3069 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3071 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3072 spa_load_l2cache(spa
);
3073 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3074 } else if (error
== 0) {
3075 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3078 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3080 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
3081 if (error
&& error
!= ENOENT
)
3082 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3085 uint64_t autoreplace
= 0;
3087 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3088 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3089 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3090 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3091 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3092 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
3093 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3094 &spa
->spa_dedup_ditto
);
3096 spa
->spa_autoreplace
= (autoreplace
!= 0);
3100 * If the 'multihost' property is set, then never allow a pool to
3101 * be imported when the system hostid is zero. The exception to
3102 * this rule is zdb which is always allowed to access pools.
3104 if (spa_multihost(spa
) && spa_get_hostid() == 0 &&
3105 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
3106 fnvlist_add_uint64(spa
->spa_load_info
,
3107 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3108 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3112 * If the 'autoreplace' property is set, then post a resource notifying
3113 * the ZFS DE that it should not issue any faults for unopenable
3114 * devices. We also iterate over the vdevs, and post a sysevent for any
3115 * unopenable vdevs so that the normal autoreplace handler can take
3118 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
3119 spa_check_removed(spa
->spa_root_vdev
);
3121 * For the import case, this is done in spa_import(), because
3122 * at this point we're using the spare definitions from
3123 * the MOS config, not necessarily from the userland config.
3125 if (state
!= SPA_LOAD_IMPORT
) {
3126 spa_aux_check_removed(&spa
->spa_spares
);
3127 spa_aux_check_removed(&spa
->spa_l2cache
);
3132 * Load the vdev state for all toplevel vdevs.
3137 * Propagate the leaf DTLs we just loaded all the way up the tree.
3139 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3140 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3141 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3144 * Load the DDTs (dedup tables).
3146 error
= ddt_load(spa
);
3148 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3150 spa_update_dspace(spa
);
3153 * Validate the config, using the MOS config to fill in any
3154 * information which might be missing. If we fail to validate
3155 * the config then declare the pool unfit for use. If we're
3156 * assembling a pool from a split, the log is not transferred
3159 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3162 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
3163 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3165 if (!spa_config_valid(spa
, nvconfig
)) {
3166 nvlist_free(nvconfig
);
3167 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3170 nvlist_free(nvconfig
);
3173 * Now that we've validated the config, check the state of the
3174 * root vdev. If it can't be opened, it indicates one or
3175 * more toplevel vdevs are faulted.
3177 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
3178 return (SET_ERROR(ENXIO
));
3180 if (spa_writeable(spa
) && spa_check_logs(spa
)) {
3181 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3182 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
3186 if (missing_feat_write
) {
3187 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
3190 * At this point, we know that we can open the pool in
3191 * read-only mode but not read-write mode. We now have enough
3192 * information and can return to userland.
3194 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
3198 * We've successfully opened the pool, verify that we're ready
3199 * to start pushing transactions.
3201 if (state
!= SPA_LOAD_TRYIMPORT
) {
3202 if ((error
= spa_load_verify(spa
)))
3203 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3207 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
3208 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3210 int need_update
= B_FALSE
;
3211 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3214 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
3217 * Claim log blocks that haven't been committed yet.
3218 * This must all happen in a single txg.
3219 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3220 * invoked from zil_claim_log_block()'s i/o done callback.
3221 * Price of rollback is that we abandon the log.
3223 spa
->spa_claiming
= B_TRUE
;
3225 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3226 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3227 zil_claim
, tx
, DS_FIND_CHILDREN
);
3230 spa
->spa_claiming
= B_FALSE
;
3232 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3233 spa
->spa_sync_on
= B_TRUE
;
3234 txg_sync_start(spa
->spa_dsl_pool
);
3235 mmp_thread_start(spa
);
3238 * Wait for all claims to sync. We sync up to the highest
3239 * claimed log block birth time so that claimed log blocks
3240 * don't appear to be from the future. spa_claim_max_txg
3241 * will have been set for us by either zil_check_log_chain()
3242 * (invoked from spa_check_logs()) or zil_claim() above.
3244 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3247 * If the config cache is stale, or we have uninitialized
3248 * metaslabs (see spa_vdev_add()), then update the config.
3250 * If this is a verbatim import, trust the current
3251 * in-core spa_config and update the disk labels.
3253 if (config_cache_txg
!= spa
->spa_config_txg
||
3254 state
== SPA_LOAD_IMPORT
||
3255 state
== SPA_LOAD_RECOVER
||
3256 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3257 need_update
= B_TRUE
;
3259 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3260 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3261 need_update
= B_TRUE
;
3264 * Update the config cache asychronously in case we're the
3265 * root pool, in which case the config cache isn't writable yet.
3268 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3271 * Check all DTLs to see if anything needs resilvering.
3273 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3274 vdev_resilver_needed(rvd
, NULL
, NULL
))
3275 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3278 * Log the fact that we booted up (so that we can detect if
3279 * we rebooted in the middle of an operation).
3281 spa_history_log_version(spa
, "open", NULL
);
3284 * Delete any inconsistent datasets.
3286 (void) dmu_objset_find(spa_name(spa
),
3287 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3290 * Clean up any stale temporary dataset userrefs.
3292 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3299 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
3301 int mode
= spa
->spa_mode
;
3304 spa_deactivate(spa
);
3306 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3308 spa_activate(spa
, mode
);
3309 spa_async_suspend(spa
);
3311 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
3315 * If spa_load() fails this function will try loading prior txg's. If
3316 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3317 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3318 * function will not rewind the pool and will return the same error as
3322 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
3323 uint64_t max_request
, int rewind_flags
)
3325 nvlist_t
*loadinfo
= NULL
;
3326 nvlist_t
*config
= NULL
;
3327 int load_error
, rewind_error
;
3328 uint64_t safe_rewind_txg
;
3331 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3332 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3333 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3335 spa
->spa_load_max_txg
= max_request
;
3336 if (max_request
!= UINT64_MAX
)
3337 spa
->spa_extreme_rewind
= B_TRUE
;
3340 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
3342 if (load_error
== 0)
3345 if (spa
->spa_root_vdev
!= NULL
)
3346 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3348 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3349 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3351 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3352 nvlist_free(config
);
3353 return (load_error
);
3356 if (state
== SPA_LOAD_RECOVER
) {
3357 /* Price of rolling back is discarding txgs, including log */
3358 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3361 * If we aren't rolling back save the load info from our first
3362 * import attempt so that we can restore it after attempting
3365 loadinfo
= spa
->spa_load_info
;
3366 spa
->spa_load_info
= fnvlist_alloc();
3369 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3370 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
3371 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
3372 TXG_INITIAL
: safe_rewind_txg
;
3375 * Continue as long as we're finding errors, we're still within
3376 * the acceptable rewind range, and we're still finding uberblocks
3378 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
3379 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
3380 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
3381 spa
->spa_extreme_rewind
= B_TRUE
;
3382 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
3385 spa
->spa_extreme_rewind
= B_FALSE
;
3386 spa
->spa_load_max_txg
= UINT64_MAX
;
3388 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
3389 spa_config_set(spa
, config
);
3391 nvlist_free(config
);
3393 if (state
== SPA_LOAD_RECOVER
) {
3394 ASSERT3P(loadinfo
, ==, NULL
);
3395 return (rewind_error
);
3397 /* Store the rewind info as part of the initial load info */
3398 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
3399 spa
->spa_load_info
);
3401 /* Restore the initial load info */
3402 fnvlist_free(spa
->spa_load_info
);
3403 spa
->spa_load_info
= loadinfo
;
3405 return (load_error
);
3412 * The import case is identical to an open except that the configuration is sent
3413 * down from userland, instead of grabbed from the configuration cache. For the
3414 * case of an open, the pool configuration will exist in the
3415 * POOL_STATE_UNINITIALIZED state.
3417 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3418 * the same time open the pool, without having to keep around the spa_t in some
3422 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
3426 spa_load_state_t state
= SPA_LOAD_OPEN
;
3428 int locked
= B_FALSE
;
3429 int firstopen
= B_FALSE
;
3434 * As disgusting as this is, we need to support recursive calls to this
3435 * function because dsl_dir_open() is called during spa_load(), and ends
3436 * up calling spa_open() again. The real fix is to figure out how to
3437 * avoid dsl_dir_open() calling this in the first place.
3439 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
3440 mutex_enter(&spa_namespace_lock
);
3444 if ((spa
= spa_lookup(pool
)) == NULL
) {
3446 mutex_exit(&spa_namespace_lock
);
3447 return (SET_ERROR(ENOENT
));
3450 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3451 zpool_rewind_policy_t policy
;
3455 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3457 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3458 state
= SPA_LOAD_RECOVER
;
3460 spa_activate(spa
, spa_mode_global
);
3462 if (state
!= SPA_LOAD_RECOVER
)
3463 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3465 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3466 policy
.zrp_request
);
3468 if (error
== EBADF
) {
3470 * If vdev_validate() returns failure (indicated by
3471 * EBADF), it indicates that one of the vdevs indicates
3472 * that the pool has been exported or destroyed. If
3473 * this is the case, the config cache is out of sync and
3474 * we should remove the pool from the namespace.
3477 spa_deactivate(spa
);
3478 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3481 mutex_exit(&spa_namespace_lock
);
3482 return (SET_ERROR(ENOENT
));
3487 * We can't open the pool, but we still have useful
3488 * information: the state of each vdev after the
3489 * attempted vdev_open(). Return this to the user.
3491 if (config
!= NULL
&& spa
->spa_config
) {
3492 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3494 VERIFY(nvlist_add_nvlist(*config
,
3495 ZPOOL_CONFIG_LOAD_INFO
,
3496 spa
->spa_load_info
) == 0);
3499 spa_deactivate(spa
);
3500 spa
->spa_last_open_failed
= error
;
3502 mutex_exit(&spa_namespace_lock
);
3508 spa_open_ref(spa
, tag
);
3511 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3514 * If we've recovered the pool, pass back any information we
3515 * gathered while doing the load.
3517 if (state
== SPA_LOAD_RECOVER
) {
3518 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3519 spa
->spa_load_info
) == 0);
3523 spa
->spa_last_open_failed
= 0;
3524 spa
->spa_last_ubsync_txg
= 0;
3525 spa
->spa_load_txg
= 0;
3526 mutex_exit(&spa_namespace_lock
);
3530 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
3538 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3541 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3545 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3547 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3551 * Lookup the given spa_t, incrementing the inject count in the process,
3552 * preventing it from being exported or destroyed.
3555 spa_inject_addref(char *name
)
3559 mutex_enter(&spa_namespace_lock
);
3560 if ((spa
= spa_lookup(name
)) == NULL
) {
3561 mutex_exit(&spa_namespace_lock
);
3564 spa
->spa_inject_ref
++;
3565 mutex_exit(&spa_namespace_lock
);
3571 spa_inject_delref(spa_t
*spa
)
3573 mutex_enter(&spa_namespace_lock
);
3574 spa
->spa_inject_ref
--;
3575 mutex_exit(&spa_namespace_lock
);
3579 * Add spares device information to the nvlist.
3582 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3592 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3594 if (spa
->spa_spares
.sav_count
== 0)
3597 VERIFY(nvlist_lookup_nvlist(config
,
3598 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3599 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3600 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3602 VERIFY(nvlist_add_nvlist_array(nvroot
,
3603 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3604 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3605 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3608 * Go through and find any spares which have since been
3609 * repurposed as an active spare. If this is the case, update
3610 * their status appropriately.
3612 for (i
= 0; i
< nspares
; i
++) {
3613 VERIFY(nvlist_lookup_uint64(spares
[i
],
3614 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3615 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3617 VERIFY(nvlist_lookup_uint64_array(
3618 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3619 (uint64_t **)&vs
, &vsc
) == 0);
3620 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3621 vs
->vs_aux
= VDEV_AUX_SPARED
;
3628 * Add l2cache device information to the nvlist, including vdev stats.
3631 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3634 uint_t i
, j
, nl2cache
;
3641 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3643 if (spa
->spa_l2cache
.sav_count
== 0)
3646 VERIFY(nvlist_lookup_nvlist(config
,
3647 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3648 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3649 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3650 if (nl2cache
!= 0) {
3651 VERIFY(nvlist_add_nvlist_array(nvroot
,
3652 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3653 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3654 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3657 * Update level 2 cache device stats.
3660 for (i
= 0; i
< nl2cache
; i
++) {
3661 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3662 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3665 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3667 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3668 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3674 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3675 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3677 vdev_get_stats(vd
, vs
);
3678 vdev_config_generate_stats(vd
, l2cache
[i
]);
3685 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3690 if (spa
->spa_feat_for_read_obj
!= 0) {
3691 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3692 spa
->spa_feat_for_read_obj
);
3693 zap_cursor_retrieve(&zc
, &za
) == 0;
3694 zap_cursor_advance(&zc
)) {
3695 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3696 za
.za_num_integers
== 1);
3697 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3698 za
.za_first_integer
));
3700 zap_cursor_fini(&zc
);
3703 if (spa
->spa_feat_for_write_obj
!= 0) {
3704 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3705 spa
->spa_feat_for_write_obj
);
3706 zap_cursor_retrieve(&zc
, &za
) == 0;
3707 zap_cursor_advance(&zc
)) {
3708 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3709 za
.za_num_integers
== 1);
3710 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3711 za
.za_first_integer
));
3713 zap_cursor_fini(&zc
);
3718 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3722 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3723 zfeature_info_t feature
= spa_feature_table
[i
];
3726 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3729 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3734 * Store a list of pool features and their reference counts in the
3737 * The first time this is called on a spa, allocate a new nvlist, fetch
3738 * the pool features and reference counts from disk, then save the list
3739 * in the spa. In subsequent calls on the same spa use the saved nvlist
3740 * and refresh its values from the cached reference counts. This
3741 * ensures we don't block here on I/O on a suspended pool so 'zpool
3742 * clear' can resume the pool.
3745 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3749 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3751 mutex_enter(&spa
->spa_feat_stats_lock
);
3752 features
= spa
->spa_feat_stats
;
3754 if (features
!= NULL
) {
3755 spa_feature_stats_from_cache(spa
, features
);
3757 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3758 spa
->spa_feat_stats
= features
;
3759 spa_feature_stats_from_disk(spa
, features
);
3762 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3765 mutex_exit(&spa
->spa_feat_stats_lock
);
3769 spa_get_stats(const char *name
, nvlist_t
**config
,
3770 char *altroot
, size_t buflen
)
3776 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3780 * This still leaves a window of inconsistency where the spares
3781 * or l2cache devices could change and the config would be
3782 * self-inconsistent.
3784 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3786 if (*config
!= NULL
) {
3787 uint64_t loadtimes
[2];
3789 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3790 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3791 VERIFY(nvlist_add_uint64_array(*config
,
3792 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3794 VERIFY(nvlist_add_uint64(*config
,
3795 ZPOOL_CONFIG_ERRCOUNT
,
3796 spa_get_errlog_size(spa
)) == 0);
3798 if (spa_suspended(spa
)) {
3799 VERIFY(nvlist_add_uint64(*config
,
3800 ZPOOL_CONFIG_SUSPENDED
,
3801 spa
->spa_failmode
) == 0);
3802 VERIFY(nvlist_add_uint64(*config
,
3803 ZPOOL_CONFIG_SUSPENDED_REASON
,
3804 spa
->spa_suspended
) == 0);
3807 spa_add_spares(spa
, *config
);
3808 spa_add_l2cache(spa
, *config
);
3809 spa_add_feature_stats(spa
, *config
);
3814 * We want to get the alternate root even for faulted pools, so we cheat
3815 * and call spa_lookup() directly.
3819 mutex_enter(&spa_namespace_lock
);
3820 spa
= spa_lookup(name
);
3822 spa_altroot(spa
, altroot
, buflen
);
3826 mutex_exit(&spa_namespace_lock
);
3828 spa_altroot(spa
, altroot
, buflen
);
3833 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3834 spa_close(spa
, FTAG
);
3841 * Validate that the auxiliary device array is well formed. We must have an
3842 * array of nvlists, each which describes a valid leaf vdev. If this is an
3843 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3844 * specified, as long as they are well-formed.
3847 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3848 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3849 vdev_labeltype_t label
)
3856 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3859 * It's acceptable to have no devs specified.
3861 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3865 return (SET_ERROR(EINVAL
));
3868 * Make sure the pool is formatted with a version that supports this
3871 if (spa_version(spa
) < version
)
3872 return (SET_ERROR(ENOTSUP
));
3875 * Set the pending device list so we correctly handle device in-use
3878 sav
->sav_pending
= dev
;
3879 sav
->sav_npending
= ndev
;
3881 for (i
= 0; i
< ndev
; i
++) {
3882 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3886 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3888 error
= SET_ERROR(EINVAL
);
3894 if ((error
= vdev_open(vd
)) == 0 &&
3895 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3896 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3897 vd
->vdev_guid
) == 0);
3903 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3910 sav
->sav_pending
= NULL
;
3911 sav
->sav_npending
= 0;
3916 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3920 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3922 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3923 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3924 VDEV_LABEL_SPARE
)) != 0) {
3928 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3929 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3930 VDEV_LABEL_L2CACHE
));
3934 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3939 if (sav
->sav_config
!= NULL
) {
3945 * Generate new dev list by concatenating with the
3948 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3949 &olddevs
, &oldndevs
) == 0);
3951 newdevs
= kmem_alloc(sizeof (void *) *
3952 (ndevs
+ oldndevs
), KM_SLEEP
);
3953 for (i
= 0; i
< oldndevs
; i
++)
3954 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3956 for (i
= 0; i
< ndevs
; i
++)
3957 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3960 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3961 DATA_TYPE_NVLIST_ARRAY
) == 0);
3963 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3964 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3965 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3966 nvlist_free(newdevs
[i
]);
3967 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3970 * Generate a new dev list.
3972 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3974 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3980 * Stop and drop level 2 ARC devices
3983 spa_l2cache_drop(spa_t
*spa
)
3987 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3989 for (i
= 0; i
< sav
->sav_count
; i
++) {
3992 vd
= sav
->sav_vdevs
[i
];
3995 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3996 pool
!= 0ULL && l2arc_vdev_present(vd
))
3997 l2arc_remove_vdev(vd
);
4005 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
4009 char *altroot
= NULL
;
4014 uint64_t txg
= TXG_INITIAL
;
4015 nvlist_t
**spares
, **l2cache
;
4016 uint_t nspares
, nl2cache
;
4017 uint64_t version
, obj
;
4018 boolean_t has_features
;
4024 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
4025 poolname
= (char *)pool
;
4028 * If this pool already exists, return failure.
4030 mutex_enter(&spa_namespace_lock
);
4031 if (spa_lookup(poolname
) != NULL
) {
4032 mutex_exit(&spa_namespace_lock
);
4033 return (SET_ERROR(EEXIST
));
4037 * Allocate a new spa_t structure.
4039 nvl
= fnvlist_alloc();
4040 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
4041 (void) nvlist_lookup_string(props
,
4042 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4043 spa
= spa_add(poolname
, nvl
, altroot
);
4045 spa_activate(spa
, spa_mode_global
);
4047 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4048 spa_deactivate(spa
);
4050 mutex_exit(&spa_namespace_lock
);
4055 * Temporary pool names should never be written to disk.
4057 if (poolname
!= pool
)
4058 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
4060 has_features
= B_FALSE
;
4061 for (elem
= nvlist_next_nvpair(props
, NULL
);
4062 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4063 if (zpool_prop_feature(nvpair_name(elem
)))
4064 has_features
= B_TRUE
;
4067 if (has_features
|| nvlist_lookup_uint64(props
,
4068 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4069 version
= SPA_VERSION
;
4071 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4073 spa
->spa_first_txg
= txg
;
4074 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4075 spa
->spa_uberblock
.ub_version
= version
;
4076 spa
->spa_ubsync
= spa
->spa_uberblock
;
4077 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4080 * Create "The Godfather" zio to hold all async IOs
4082 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4084 for (i
= 0; i
< max_ncpus
; i
++) {
4085 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4086 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4087 ZIO_FLAG_GODFATHER
);
4091 * Create the root vdev.
4093 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4095 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4097 ASSERT(error
!= 0 || rvd
!= NULL
);
4098 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4100 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4101 error
= SET_ERROR(EINVAL
);
4104 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4105 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4106 VDEV_ALLOC_ADD
)) == 0) {
4107 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
4108 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4109 vdev_expand(rvd
->vdev_child
[c
], txg
);
4113 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4117 spa_deactivate(spa
);
4119 mutex_exit(&spa_namespace_lock
);
4124 * Get the list of spares, if specified.
4126 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4127 &spares
, &nspares
) == 0) {
4128 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4130 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4131 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4132 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4133 spa_load_spares(spa
);
4134 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4135 spa
->spa_spares
.sav_sync
= B_TRUE
;
4139 * Get the list of level 2 cache devices, if specified.
4141 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4142 &l2cache
, &nl2cache
) == 0) {
4143 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4144 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4145 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4146 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4147 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4148 spa_load_l2cache(spa
);
4149 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4150 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4153 spa
->spa_is_initializing
= B_TRUE
;
4154 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
4155 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
4156 spa
->spa_is_initializing
= B_FALSE
;
4159 * Create DDTs (dedup tables).
4163 spa_update_dspace(spa
);
4165 tx
= dmu_tx_create_assigned(dp
, txg
);
4168 * Create the pool's history object.
4170 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
4171 spa_history_create_obj(spa
, tx
);
4173 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4174 spa_history_log_version(spa
, "create", tx
);
4177 * Create the pool config object.
4179 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4180 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4181 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4183 if (zap_add(spa
->spa_meta_objset
,
4184 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4185 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4186 cmn_err(CE_PANIC
, "failed to add pool config");
4189 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
4190 spa_feature_create_zap_objects(spa
, tx
);
4192 if (zap_add(spa
->spa_meta_objset
,
4193 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4194 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4195 cmn_err(CE_PANIC
, "failed to add pool version");
4198 /* Newly created pools with the right version are always deflated. */
4199 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4200 spa
->spa_deflate
= TRUE
;
4201 if (zap_add(spa
->spa_meta_objset
,
4202 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4203 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4204 cmn_err(CE_PANIC
, "failed to add deflate");
4209 * Create the deferred-free bpobj. Turn off compression
4210 * because sync-to-convergence takes longer if the blocksize
4213 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4214 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4215 ZIO_COMPRESS_OFF
, tx
);
4216 if (zap_add(spa
->spa_meta_objset
,
4217 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4218 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4219 cmn_err(CE_PANIC
, "failed to add bpobj");
4221 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4222 spa
->spa_meta_objset
, obj
));
4225 * Generate some random noise for salted checksums to operate on.
4227 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4228 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4231 * Set pool properties.
4233 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4234 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4235 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4236 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4237 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
4239 if (props
!= NULL
) {
4240 spa_configfile_set(spa
, props
, B_FALSE
);
4241 spa_sync_props(props
, tx
);
4246 spa
->spa_sync_on
= B_TRUE
;
4247 txg_sync_start(spa
->spa_dsl_pool
);
4248 mmp_thread_start(spa
);
4251 * We explicitly wait for the first transaction to complete so that our
4252 * bean counters are appropriately updated.
4254 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
4256 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4259 * Don't count references from objsets that are already closed
4260 * and are making their way through the eviction process.
4262 spa_evicting_os_wait(spa
);
4263 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4264 spa
->spa_load_state
= SPA_LOAD_NONE
;
4266 mutex_exit(&spa_namespace_lock
);
4272 * Import a non-root pool into the system.
4275 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4278 char *altroot
= NULL
;
4279 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4280 zpool_rewind_policy_t policy
;
4281 uint64_t mode
= spa_mode_global
;
4282 uint64_t readonly
= B_FALSE
;
4285 nvlist_t
**spares
, **l2cache
;
4286 uint_t nspares
, nl2cache
;
4289 * If a pool with this name exists, return failure.
4291 mutex_enter(&spa_namespace_lock
);
4292 if (spa_lookup(pool
) != NULL
) {
4293 mutex_exit(&spa_namespace_lock
);
4294 return (SET_ERROR(EEXIST
));
4298 * Create and initialize the spa structure.
4300 (void) nvlist_lookup_string(props
,
4301 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4302 (void) nvlist_lookup_uint64(props
,
4303 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4306 spa
= spa_add(pool
, config
, altroot
);
4307 spa
->spa_import_flags
= flags
;
4310 * Verbatim import - Take a pool and insert it into the namespace
4311 * as if it had been loaded at boot.
4313 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4315 spa_configfile_set(spa
, props
, B_FALSE
);
4317 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4318 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4320 mutex_exit(&spa_namespace_lock
);
4324 spa_activate(spa
, mode
);
4327 * Don't start async tasks until we know everything is healthy.
4329 spa_async_suspend(spa
);
4331 zpool_get_rewind_policy(config
, &policy
);
4332 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4333 state
= SPA_LOAD_RECOVER
;
4336 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4337 * because the user-supplied config is actually the one to trust when
4340 if (state
!= SPA_LOAD_RECOVER
)
4341 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4343 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4344 policy
.zrp_request
);
4347 * Propagate anything learned while loading the pool and pass it
4348 * back to caller (i.e. rewind info, missing devices, etc).
4350 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4351 spa
->spa_load_info
) == 0);
4353 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4355 * Toss any existing sparelist, as it doesn't have any validity
4356 * anymore, and conflicts with spa_has_spare().
4358 if (spa
->spa_spares
.sav_config
) {
4359 nvlist_free(spa
->spa_spares
.sav_config
);
4360 spa
->spa_spares
.sav_config
= NULL
;
4361 spa_load_spares(spa
);
4363 if (spa
->spa_l2cache
.sav_config
) {
4364 nvlist_free(spa
->spa_l2cache
.sav_config
);
4365 spa
->spa_l2cache
.sav_config
= NULL
;
4366 spa_load_l2cache(spa
);
4369 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4371 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4374 spa_configfile_set(spa
, props
, B_FALSE
);
4376 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4377 (error
= spa_prop_set(spa
, props
)))) {
4379 spa_deactivate(spa
);
4381 mutex_exit(&spa_namespace_lock
);
4385 spa_async_resume(spa
);
4388 * Override any spares and level 2 cache devices as specified by
4389 * the user, as these may have correct device names/devids, etc.
4391 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4392 &spares
, &nspares
) == 0) {
4393 if (spa
->spa_spares
.sav_config
)
4394 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4395 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4397 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4398 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4399 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4400 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4401 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4402 spa_load_spares(spa
);
4403 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4404 spa
->spa_spares
.sav_sync
= B_TRUE
;
4406 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4407 &l2cache
, &nl2cache
) == 0) {
4408 if (spa
->spa_l2cache
.sav_config
)
4409 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4410 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4412 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4413 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4414 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4415 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4416 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4417 spa_load_l2cache(spa
);
4418 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4419 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4423 * Check for any removed devices.
4425 if (spa
->spa_autoreplace
) {
4426 spa_aux_check_removed(&spa
->spa_spares
);
4427 spa_aux_check_removed(&spa
->spa_l2cache
);
4430 if (spa_writeable(spa
)) {
4432 * Update the config cache to include the newly-imported pool.
4434 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4438 * It's possible that the pool was expanded while it was exported.
4439 * We kick off an async task to handle this for us.
4441 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4443 spa_history_log_version(spa
, "import", NULL
);
4445 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4447 zvol_create_minors(spa
, pool
, B_TRUE
);
4449 mutex_exit(&spa_namespace_lock
);
4455 spa_tryimport(nvlist_t
*tryconfig
)
4457 nvlist_t
*config
= NULL
;
4463 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4466 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4470 * Create and initialize the spa structure.
4472 mutex_enter(&spa_namespace_lock
);
4473 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4474 spa_activate(spa
, FREAD
);
4477 * Pass off the heavy lifting to spa_load().
4478 * Pass TRUE for mosconfig because the user-supplied config
4479 * is actually the one to trust when doing an import.
4481 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4484 * If 'tryconfig' was at least parsable, return the current config.
4486 if (spa
->spa_root_vdev
!= NULL
) {
4487 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4488 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4490 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4492 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4493 spa
->spa_uberblock
.ub_timestamp
) == 0);
4494 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4495 spa
->spa_load_info
) == 0);
4496 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4497 spa
->spa_errata
) == 0);
4500 * If the bootfs property exists on this pool then we
4501 * copy it out so that external consumers can tell which
4502 * pools are bootable.
4504 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4505 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4508 * We have to play games with the name since the
4509 * pool was opened as TRYIMPORT_NAME.
4511 if (dsl_dsobj_to_dsname(spa_name(spa
),
4512 spa
->spa_bootfs
, tmpname
) == 0) {
4516 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4518 cp
= strchr(tmpname
, '/');
4520 (void) strlcpy(dsname
, tmpname
,
4523 (void) snprintf(dsname
, MAXPATHLEN
,
4524 "%s/%s", poolname
, ++cp
);
4526 VERIFY(nvlist_add_string(config
,
4527 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4528 kmem_free(dsname
, MAXPATHLEN
);
4530 kmem_free(tmpname
, MAXPATHLEN
);
4534 * Add the list of hot spares and level 2 cache devices.
4536 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4537 spa_add_spares(spa
, config
);
4538 spa_add_l2cache(spa
, config
);
4539 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4543 spa_deactivate(spa
);
4545 mutex_exit(&spa_namespace_lock
);
4551 * Pool export/destroy
4553 * The act of destroying or exporting a pool is very simple. We make sure there
4554 * is no more pending I/O and any references to the pool are gone. Then, we
4555 * update the pool state and sync all the labels to disk, removing the
4556 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4557 * we don't sync the labels or remove the configuration cache.
4560 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4561 boolean_t force
, boolean_t hardforce
)
4568 if (!(spa_mode_global
& FWRITE
))
4569 return (SET_ERROR(EROFS
));
4571 mutex_enter(&spa_namespace_lock
);
4572 if ((spa
= spa_lookup(pool
)) == NULL
) {
4573 mutex_exit(&spa_namespace_lock
);
4574 return (SET_ERROR(ENOENT
));
4578 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4579 * reacquire the namespace lock, and see if we can export.
4581 spa_open_ref(spa
, FTAG
);
4582 mutex_exit(&spa_namespace_lock
);
4583 spa_async_suspend(spa
);
4584 if (spa
->spa_zvol_taskq
) {
4585 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
4586 taskq_wait(spa
->spa_zvol_taskq
);
4588 mutex_enter(&spa_namespace_lock
);
4589 spa_close(spa
, FTAG
);
4591 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4594 * The pool will be in core if it's openable, in which case we can
4595 * modify its state. Objsets may be open only because they're dirty,
4596 * so we have to force it to sync before checking spa_refcnt.
4598 if (spa
->spa_sync_on
) {
4599 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4600 spa_evicting_os_wait(spa
);
4604 * A pool cannot be exported or destroyed if there are active
4605 * references. If we are resetting a pool, allow references by
4606 * fault injection handlers.
4608 if (!spa_refcount_zero(spa
) ||
4609 (spa
->spa_inject_ref
!= 0 &&
4610 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4611 spa_async_resume(spa
);
4612 mutex_exit(&spa_namespace_lock
);
4613 return (SET_ERROR(EBUSY
));
4616 if (spa
->spa_sync_on
) {
4618 * A pool cannot be exported if it has an active shared spare.
4619 * This is to prevent other pools stealing the active spare
4620 * from an exported pool. At user's own will, such pool can
4621 * be forcedly exported.
4623 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4624 spa_has_active_shared_spare(spa
)) {
4625 spa_async_resume(spa
);
4626 mutex_exit(&spa_namespace_lock
);
4627 return (SET_ERROR(EXDEV
));
4631 * We want this to be reflected on every label,
4632 * so mark them all dirty. spa_unload() will do the
4633 * final sync that pushes these changes out.
4635 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4636 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4637 spa
->spa_state
= new_state
;
4638 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4640 vdev_config_dirty(spa
->spa_root_vdev
);
4641 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4646 if (new_state
== POOL_STATE_DESTROYED
)
4647 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
4648 else if (new_state
== POOL_STATE_EXPORTED
)
4649 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
4651 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4653 spa_deactivate(spa
);
4656 if (oldconfig
&& spa
->spa_config
)
4657 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4659 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4661 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4664 mutex_exit(&spa_namespace_lock
);
4670 * Destroy a storage pool.
4673 spa_destroy(char *pool
)
4675 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4680 * Export a storage pool.
4683 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4684 boolean_t hardforce
)
4686 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4691 * Similar to spa_export(), this unloads the spa_t without actually removing it
4692 * from the namespace in any way.
4695 spa_reset(char *pool
)
4697 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4702 * ==========================================================================
4703 * Device manipulation
4704 * ==========================================================================
4708 * Add a device to a storage pool.
4711 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4715 vdev_t
*rvd
= spa
->spa_root_vdev
;
4717 nvlist_t
**spares
, **l2cache
;
4718 uint_t nspares
, nl2cache
;
4721 ASSERT(spa_writeable(spa
));
4723 txg
= spa_vdev_enter(spa
);
4725 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4726 VDEV_ALLOC_ADD
)) != 0)
4727 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4729 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4731 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4735 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4739 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4740 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4742 if (vd
->vdev_children
!= 0 &&
4743 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4744 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4747 * We must validate the spares and l2cache devices after checking the
4748 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4750 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4751 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4754 * Transfer each new top-level vdev from vd to rvd.
4756 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4759 * Set the vdev id to the first hole, if one exists.
4761 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4762 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4763 vdev_free(rvd
->vdev_child
[id
]);
4767 tvd
= vd
->vdev_child
[c
];
4768 vdev_remove_child(vd
, tvd
);
4770 vdev_add_child(rvd
, tvd
);
4771 vdev_config_dirty(tvd
);
4775 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4776 ZPOOL_CONFIG_SPARES
);
4777 spa_load_spares(spa
);
4778 spa
->spa_spares
.sav_sync
= B_TRUE
;
4781 if (nl2cache
!= 0) {
4782 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4783 ZPOOL_CONFIG_L2CACHE
);
4784 spa_load_l2cache(spa
);
4785 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4789 * We have to be careful when adding new vdevs to an existing pool.
4790 * If other threads start allocating from these vdevs before we
4791 * sync the config cache, and we lose power, then upon reboot we may
4792 * fail to open the pool because there are DVAs that the config cache
4793 * can't translate. Therefore, we first add the vdevs without
4794 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4795 * and then let spa_config_update() initialize the new metaslabs.
4797 * spa_load() checks for added-but-not-initialized vdevs, so that
4798 * if we lose power at any point in this sequence, the remaining
4799 * steps will be completed the next time we load the pool.
4801 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4803 mutex_enter(&spa_namespace_lock
);
4804 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4805 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
4806 mutex_exit(&spa_namespace_lock
);
4812 * Attach a device to a mirror. The arguments are the path to any device
4813 * in the mirror, and the nvroot for the new device. If the path specifies
4814 * a device that is not mirrored, we automatically insert the mirror vdev.
4816 * If 'replacing' is specified, the new device is intended to replace the
4817 * existing device; in this case the two devices are made into their own
4818 * mirror using the 'replacing' vdev, which is functionally identical to
4819 * the mirror vdev (it actually reuses all the same ops) but has a few
4820 * extra rules: you can't attach to it after it's been created, and upon
4821 * completion of resilvering, the first disk (the one being replaced)
4822 * is automatically detached.
4825 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4827 uint64_t txg
, dtl_max_txg
;
4828 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4830 char *oldvdpath
, *newvdpath
;
4833 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4835 ASSERT(spa_writeable(spa
));
4837 txg
= spa_vdev_enter(spa
);
4839 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4842 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4844 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4845 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4847 pvd
= oldvd
->vdev_parent
;
4849 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4850 VDEV_ALLOC_ATTACH
)) != 0)
4851 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4853 if (newrootvd
->vdev_children
!= 1)
4854 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4856 newvd
= newrootvd
->vdev_child
[0];
4858 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4859 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4861 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4862 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4865 * Spares can't replace logs
4867 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4868 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4872 * For attach, the only allowable parent is a mirror or the root
4875 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4876 pvd
->vdev_ops
!= &vdev_root_ops
)
4877 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4879 pvops
= &vdev_mirror_ops
;
4882 * Active hot spares can only be replaced by inactive hot
4885 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4886 oldvd
->vdev_isspare
&&
4887 !spa_has_spare(spa
, newvd
->vdev_guid
))
4888 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4891 * If the source is a hot spare, and the parent isn't already a
4892 * spare, then we want to create a new hot spare. Otherwise, we
4893 * want to create a replacing vdev. The user is not allowed to
4894 * attach to a spared vdev child unless the 'isspare' state is
4895 * the same (spare replaces spare, non-spare replaces
4898 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4899 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4900 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4901 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4902 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4903 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4906 if (newvd
->vdev_isspare
)
4907 pvops
= &vdev_spare_ops
;
4909 pvops
= &vdev_replacing_ops
;
4913 * Make sure the new device is big enough.
4915 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4916 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4919 * The new device cannot have a higher alignment requirement
4920 * than the top-level vdev.
4922 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4923 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4926 * If this is an in-place replacement, update oldvd's path and devid
4927 * to make it distinguishable from newvd, and unopenable from now on.
4929 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4930 spa_strfree(oldvd
->vdev_path
);
4931 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4933 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4934 newvd
->vdev_path
, "old");
4935 if (oldvd
->vdev_devid
!= NULL
) {
4936 spa_strfree(oldvd
->vdev_devid
);
4937 oldvd
->vdev_devid
= NULL
;
4941 /* mark the device being resilvered */
4942 newvd
->vdev_resilver_txg
= txg
;
4945 * If the parent is not a mirror, or if we're replacing, insert the new
4946 * mirror/replacing/spare vdev above oldvd.
4948 if (pvd
->vdev_ops
!= pvops
)
4949 pvd
= vdev_add_parent(oldvd
, pvops
);
4951 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4952 ASSERT(pvd
->vdev_ops
== pvops
);
4953 ASSERT(oldvd
->vdev_parent
== pvd
);
4956 * Extract the new device from its root and add it to pvd.
4958 vdev_remove_child(newrootvd
, newvd
);
4959 newvd
->vdev_id
= pvd
->vdev_children
;
4960 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4961 vdev_add_child(pvd
, newvd
);
4964 * Reevaluate the parent vdev state.
4966 vdev_propagate_state(pvd
);
4968 tvd
= newvd
->vdev_top
;
4969 ASSERT(pvd
->vdev_top
== tvd
);
4970 ASSERT(tvd
->vdev_parent
== rvd
);
4972 vdev_config_dirty(tvd
);
4975 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4976 * for any dmu_sync-ed blocks. It will propagate upward when
4977 * spa_vdev_exit() calls vdev_dtl_reassess().
4979 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
4981 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
4982 dtl_max_txg
- TXG_INITIAL
);
4984 if (newvd
->vdev_isspare
) {
4985 spa_spare_activate(newvd
);
4986 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
4989 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
4990 newvdpath
= spa_strdup(newvd
->vdev_path
);
4991 newvd_isspare
= newvd
->vdev_isspare
;
4994 * Mark newvd's DTL dirty in this txg.
4996 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
4999 * Schedule the resilver to restart in the future. We do this to
5000 * ensure that dmu_sync-ed blocks have been stitched into the
5001 * respective datasets.
5003 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
5005 if (spa
->spa_bootfs
)
5006 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
5008 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
5013 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
5015 spa_history_log_internal(spa
, "vdev attach", NULL
,
5016 "%s vdev=%s %s vdev=%s",
5017 replacing
&& newvd_isspare
? "spare in" :
5018 replacing
? "replace" : "attach", newvdpath
,
5019 replacing
? "for" : "to", oldvdpath
);
5021 spa_strfree(oldvdpath
);
5022 spa_strfree(newvdpath
);
5028 * Detach a device from a mirror or replacing vdev.
5030 * If 'replace_done' is specified, only detach if the parent
5031 * is a replacing vdev.
5034 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5038 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5039 boolean_t unspare
= B_FALSE
;
5040 uint64_t unspare_guid
= 0;
5043 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5044 ASSERT(spa_writeable(spa
));
5046 txg
= spa_vdev_enter(spa
);
5048 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5051 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5053 if (!vd
->vdev_ops
->vdev_op_leaf
)
5054 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5056 pvd
= vd
->vdev_parent
;
5059 * If the parent/child relationship is not as expected, don't do it.
5060 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5061 * vdev that's replacing B with C. The user's intent in replacing
5062 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5063 * the replace by detaching C, the expected behavior is to end up
5064 * M(A,B). But suppose that right after deciding to detach C,
5065 * the replacement of B completes. We would have M(A,C), and then
5066 * ask to detach C, which would leave us with just A -- not what
5067 * the user wanted. To prevent this, we make sure that the
5068 * parent/child relationship hasn't changed -- in this example,
5069 * that C's parent is still the replacing vdev R.
5071 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5072 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5075 * Only 'replacing' or 'spare' vdevs can be replaced.
5077 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5078 pvd
->vdev_ops
!= &vdev_spare_ops
)
5079 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5081 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5082 spa_version(spa
) >= SPA_VERSION_SPARES
);
5085 * Only mirror, replacing, and spare vdevs support detach.
5087 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5088 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5089 pvd
->vdev_ops
!= &vdev_spare_ops
)
5090 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5093 * If this device has the only valid copy of some data,
5094 * we cannot safely detach it.
5096 if (vdev_dtl_required(vd
))
5097 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5099 ASSERT(pvd
->vdev_children
>= 2);
5102 * If we are detaching the second disk from a replacing vdev, then
5103 * check to see if we changed the original vdev's path to have "/old"
5104 * at the end in spa_vdev_attach(). If so, undo that change now.
5106 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5107 vd
->vdev_path
!= NULL
) {
5108 size_t len
= strlen(vd
->vdev_path
);
5110 for (c
= 0; c
< pvd
->vdev_children
; c
++) {
5111 cvd
= pvd
->vdev_child
[c
];
5113 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5116 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5117 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5118 spa_strfree(cvd
->vdev_path
);
5119 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5126 * If we are detaching the original disk from a spare, then it implies
5127 * that the spare should become a real disk, and be removed from the
5128 * active spare list for the pool.
5130 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5132 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5136 * Erase the disk labels so the disk can be used for other things.
5137 * This must be done after all other error cases are handled,
5138 * but before we disembowel vd (so we can still do I/O to it).
5139 * But if we can't do it, don't treat the error as fatal --
5140 * it may be that the unwritability of the disk is the reason
5141 * it's being detached!
5143 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5146 * Remove vd from its parent and compact the parent's children.
5148 vdev_remove_child(pvd
, vd
);
5149 vdev_compact_children(pvd
);
5152 * Remember one of the remaining children so we can get tvd below.
5154 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5157 * If we need to remove the remaining child from the list of hot spares,
5158 * do it now, marking the vdev as no longer a spare in the process.
5159 * We must do this before vdev_remove_parent(), because that can
5160 * change the GUID if it creates a new toplevel GUID. For a similar
5161 * reason, we must remove the spare now, in the same txg as the detach;
5162 * otherwise someone could attach a new sibling, change the GUID, and
5163 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5166 ASSERT(cvd
->vdev_isspare
);
5167 spa_spare_remove(cvd
);
5168 unspare_guid
= cvd
->vdev_guid
;
5169 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5170 cvd
->vdev_unspare
= B_TRUE
;
5174 * If the parent mirror/replacing vdev only has one child,
5175 * the parent is no longer needed. Remove it from the tree.
5177 if (pvd
->vdev_children
== 1) {
5178 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5179 cvd
->vdev_unspare
= B_FALSE
;
5180 vdev_remove_parent(cvd
);
5185 * We don't set tvd until now because the parent we just removed
5186 * may have been the previous top-level vdev.
5188 tvd
= cvd
->vdev_top
;
5189 ASSERT(tvd
->vdev_parent
== rvd
);
5192 * Reevaluate the parent vdev state.
5194 vdev_propagate_state(cvd
);
5197 * If the 'autoexpand' property is set on the pool then automatically
5198 * try to expand the size of the pool. For example if the device we
5199 * just detached was smaller than the others, it may be possible to
5200 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5201 * first so that we can obtain the updated sizes of the leaf vdevs.
5203 if (spa
->spa_autoexpand
) {
5205 vdev_expand(tvd
, txg
);
5208 vdev_config_dirty(tvd
);
5211 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5212 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5213 * But first make sure we're not on any *other* txg's DTL list, to
5214 * prevent vd from being accessed after it's freed.
5216 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
5217 for (t
= 0; t
< TXG_SIZE
; t
++)
5218 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5219 vd
->vdev_detached
= B_TRUE
;
5220 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5222 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5224 /* hang on to the spa before we release the lock */
5225 spa_open_ref(spa
, FTAG
);
5227 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5229 spa_history_log_internal(spa
, "detach", NULL
,
5231 spa_strfree(vdpath
);
5234 * If this was the removal of the original device in a hot spare vdev,
5235 * then we want to go through and remove the device from the hot spare
5236 * list of every other pool.
5239 spa_t
*altspa
= NULL
;
5241 mutex_enter(&spa_namespace_lock
);
5242 while ((altspa
= spa_next(altspa
)) != NULL
) {
5243 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5247 spa_open_ref(altspa
, FTAG
);
5248 mutex_exit(&spa_namespace_lock
);
5249 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5250 mutex_enter(&spa_namespace_lock
);
5251 spa_close(altspa
, FTAG
);
5253 mutex_exit(&spa_namespace_lock
);
5255 /* search the rest of the vdevs for spares to remove */
5256 spa_vdev_resilver_done(spa
);
5259 /* all done with the spa; OK to release */
5260 mutex_enter(&spa_namespace_lock
);
5261 spa_close(spa
, FTAG
);
5262 mutex_exit(&spa_namespace_lock
);
5268 * Split a set of devices from their mirrors, and create a new pool from them.
5271 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5272 nvlist_t
*props
, boolean_t exp
)
5275 uint64_t txg
, *glist
;
5277 uint_t c
, children
, lastlog
;
5278 nvlist_t
**child
, *nvl
, *tmp
;
5280 char *altroot
= NULL
;
5281 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5282 boolean_t activate_slog
;
5284 ASSERT(spa_writeable(spa
));
5286 txg
= spa_vdev_enter(spa
);
5288 /* clear the log and flush everything up to now */
5289 activate_slog
= spa_passivate_log(spa
);
5290 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5291 error
= spa_offline_log(spa
);
5292 txg
= spa_vdev_config_enter(spa
);
5295 spa_activate_log(spa
);
5298 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5300 /* check new spa name before going any further */
5301 if (spa_lookup(newname
) != NULL
)
5302 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5305 * scan through all the children to ensure they're all mirrors
5307 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5308 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5310 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5312 /* first, check to ensure we've got the right child count */
5313 rvd
= spa
->spa_root_vdev
;
5315 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5316 vdev_t
*vd
= rvd
->vdev_child
[c
];
5318 /* don't count the holes & logs as children */
5319 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5327 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5328 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5330 /* next, ensure no spare or cache devices are part of the split */
5331 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5332 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5333 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5335 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5336 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5338 /* then, loop over each vdev and validate it */
5339 for (c
= 0; c
< children
; c
++) {
5340 uint64_t is_hole
= 0;
5342 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5346 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5347 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5350 error
= SET_ERROR(EINVAL
);
5355 /* which disk is going to be split? */
5356 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5358 error
= SET_ERROR(EINVAL
);
5362 /* look it up in the spa */
5363 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5364 if (vml
[c
] == NULL
) {
5365 error
= SET_ERROR(ENODEV
);
5369 /* make sure there's nothing stopping the split */
5370 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5371 vml
[c
]->vdev_islog
||
5372 vml
[c
]->vdev_ishole
||
5373 vml
[c
]->vdev_isspare
||
5374 vml
[c
]->vdev_isl2cache
||
5375 !vdev_writeable(vml
[c
]) ||
5376 vml
[c
]->vdev_children
!= 0 ||
5377 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5378 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5379 error
= SET_ERROR(EINVAL
);
5383 if (vdev_dtl_required(vml
[c
])) {
5384 error
= SET_ERROR(EBUSY
);
5388 /* we need certain info from the top level */
5389 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5390 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5391 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5392 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5393 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5394 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5395 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5396 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5398 /* transfer per-vdev ZAPs */
5399 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
5400 VERIFY0(nvlist_add_uint64(child
[c
],
5401 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
5403 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
5404 VERIFY0(nvlist_add_uint64(child
[c
],
5405 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
5406 vml
[c
]->vdev_parent
->vdev_top_zap
));
5410 kmem_free(vml
, children
* sizeof (vdev_t
*));
5411 kmem_free(glist
, children
* sizeof (uint64_t));
5412 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5415 /* stop writers from using the disks */
5416 for (c
= 0; c
< children
; c
++) {
5418 vml
[c
]->vdev_offline
= B_TRUE
;
5420 vdev_reopen(spa
->spa_root_vdev
);
5423 * Temporarily record the splitting vdevs in the spa config. This
5424 * will disappear once the config is regenerated.
5426 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5427 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5428 glist
, children
) == 0);
5429 kmem_free(glist
, children
* sizeof (uint64_t));
5431 mutex_enter(&spa
->spa_props_lock
);
5432 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5434 mutex_exit(&spa
->spa_props_lock
);
5435 spa
->spa_config_splitting
= nvl
;
5436 vdev_config_dirty(spa
->spa_root_vdev
);
5438 /* configure and create the new pool */
5439 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5440 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5441 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5442 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5443 spa_version(spa
)) == 0);
5444 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5445 spa
->spa_config_txg
) == 0);
5446 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5447 spa_generate_guid(NULL
)) == 0);
5448 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
5449 (void) nvlist_lookup_string(props
,
5450 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5452 /* add the new pool to the namespace */
5453 newspa
= spa_add(newname
, config
, altroot
);
5454 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5455 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5456 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5458 /* release the spa config lock, retaining the namespace lock */
5459 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5461 if (zio_injection_enabled
)
5462 zio_handle_panic_injection(spa
, FTAG
, 1);
5464 spa_activate(newspa
, spa_mode_global
);
5465 spa_async_suspend(newspa
);
5467 /* create the new pool from the disks of the original pool */
5468 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5472 /* if that worked, generate a real config for the new pool */
5473 if (newspa
->spa_root_vdev
!= NULL
) {
5474 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5475 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5476 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5477 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5478 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5483 if (props
!= NULL
) {
5484 spa_configfile_set(newspa
, props
, B_FALSE
);
5485 error
= spa_prop_set(newspa
, props
);
5490 /* flush everything */
5491 txg
= spa_vdev_config_enter(newspa
);
5492 vdev_config_dirty(newspa
->spa_root_vdev
);
5493 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5495 if (zio_injection_enabled
)
5496 zio_handle_panic_injection(spa
, FTAG
, 2);
5498 spa_async_resume(newspa
);
5500 /* finally, update the original pool's config */
5501 txg
= spa_vdev_config_enter(spa
);
5502 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5503 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5506 for (c
= 0; c
< children
; c
++) {
5507 if (vml
[c
] != NULL
) {
5510 spa_history_log_internal(spa
, "detach", tx
,
5511 "vdev=%s", vml
[c
]->vdev_path
);
5516 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5517 vdev_config_dirty(spa
->spa_root_vdev
);
5518 spa
->spa_config_splitting
= NULL
;
5522 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5524 if (zio_injection_enabled
)
5525 zio_handle_panic_injection(spa
, FTAG
, 3);
5527 /* split is complete; log a history record */
5528 spa_history_log_internal(newspa
, "split", NULL
,
5529 "from pool %s", spa_name(spa
));
5531 kmem_free(vml
, children
* sizeof (vdev_t
*));
5533 /* if we're not going to mount the filesystems in userland, export */
5535 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5542 spa_deactivate(newspa
);
5545 txg
= spa_vdev_config_enter(spa
);
5547 /* re-online all offlined disks */
5548 for (c
= 0; c
< children
; c
++) {
5550 vml
[c
]->vdev_offline
= B_FALSE
;
5552 vdev_reopen(spa
->spa_root_vdev
);
5554 nvlist_free(spa
->spa_config_splitting
);
5555 spa
->spa_config_splitting
= NULL
;
5556 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5558 kmem_free(vml
, children
* sizeof (vdev_t
*));
5563 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5567 for (i
= 0; i
< count
; i
++) {
5570 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5573 if (guid
== target_guid
)
5581 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5582 nvlist_t
*dev_to_remove
)
5584 nvlist_t
**newdev
= NULL
;
5588 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5590 for (i
= 0, j
= 0; i
< count
; i
++) {
5591 if (dev
[i
] == dev_to_remove
)
5593 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5596 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5597 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5599 for (i
= 0; i
< count
- 1; i
++)
5600 nvlist_free(newdev
[i
]);
5603 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5607 * Evacuate the device.
5610 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5615 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5616 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5617 ASSERT(vd
== vd
->vdev_top
);
5620 * Evacuate the device. We don't hold the config lock as writer
5621 * since we need to do I/O but we do keep the
5622 * spa_namespace_lock held. Once this completes the device
5623 * should no longer have any blocks allocated on it.
5625 if (vd
->vdev_islog
) {
5626 if (vd
->vdev_stat
.vs_alloc
!= 0)
5627 error
= spa_offline_log(spa
);
5629 error
= SET_ERROR(ENOTSUP
);
5636 * The evacuation succeeded. Remove any remaining MOS metadata
5637 * associated with this vdev, and wait for these changes to sync.
5639 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5640 txg
= spa_vdev_config_enter(spa
);
5641 vd
->vdev_removing
= B_TRUE
;
5642 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5643 vdev_config_dirty(vd
);
5644 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5650 * Complete the removal by cleaning up the namespace.
5653 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5655 vdev_t
*rvd
= spa
->spa_root_vdev
;
5656 uint64_t id
= vd
->vdev_id
;
5657 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5659 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5660 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5661 ASSERT(vd
== vd
->vdev_top
);
5664 * Only remove any devices which are empty.
5666 if (vd
->vdev_stat
.vs_alloc
!= 0)
5669 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5671 if (list_link_active(&vd
->vdev_state_dirty_node
))
5672 vdev_state_clean(vd
);
5673 if (list_link_active(&vd
->vdev_config_dirty_node
))
5674 vdev_config_clean(vd
);
5679 vdev_compact_children(rvd
);
5681 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5682 vdev_add_child(rvd
, vd
);
5684 vdev_config_dirty(rvd
);
5687 * Reassess the health of our root vdev.
5693 * Remove a device from the pool -
5695 * Removing a device from the vdev namespace requires several steps
5696 * and can take a significant amount of time. As a result we use
5697 * the spa_vdev_config_[enter/exit] functions which allow us to
5698 * grab and release the spa_config_lock while still holding the namespace
5699 * lock. During each step the configuration is synced out.
5701 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5705 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5708 sysevent_t
*ev
= NULL
;
5709 metaslab_group_t
*mg
;
5710 nvlist_t
**spares
, **l2cache
, *nv
;
5712 uint_t nspares
, nl2cache
;
5714 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5716 ASSERT(spa_writeable(spa
));
5719 txg
= spa_vdev_enter(spa
);
5721 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5723 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5724 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5725 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5726 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5728 * Only remove the hot spare if it's not currently in use
5731 if (vd
== NULL
|| unspare
) {
5733 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5734 ev
= spa_event_create(spa
, vd
, NULL
,
5735 ESC_ZFS_VDEV_REMOVE_AUX
);
5736 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5737 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5738 spa_load_spares(spa
);
5739 spa
->spa_spares
.sav_sync
= B_TRUE
;
5741 error
= SET_ERROR(EBUSY
);
5743 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5744 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5745 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5746 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5748 * Cache devices can always be removed.
5750 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5751 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_AUX
);
5752 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5753 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5754 spa_load_l2cache(spa
);
5755 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5756 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5758 ASSERT(vd
== vd
->vdev_top
);
5763 * Stop allocating from this vdev.
5765 metaslab_group_passivate(mg
);
5768 * Wait for the youngest allocations and frees to sync,
5769 * and then wait for the deferral of those frees to finish.
5771 spa_vdev_config_exit(spa
, NULL
,
5772 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5775 * Attempt to evacuate the vdev.
5777 error
= spa_vdev_remove_evacuate(spa
, vd
);
5779 txg
= spa_vdev_config_enter(spa
);
5782 * If we couldn't evacuate the vdev, unwind.
5785 metaslab_group_activate(mg
);
5786 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5790 * Clean up the vdev namespace.
5792 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_DEV
);
5793 spa_vdev_remove_from_namespace(spa
, vd
);
5795 } else if (vd
!= NULL
) {
5797 * Normal vdevs cannot be removed (yet).
5799 error
= SET_ERROR(ENOTSUP
);
5802 * There is no vdev of any kind with the specified guid.
5804 error
= SET_ERROR(ENOENT
);
5808 error
= spa_vdev_exit(spa
, NULL
, txg
, error
);
5817 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5818 * currently spared, so we can detach it.
5821 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5823 vdev_t
*newvd
, *oldvd
;
5826 for (c
= 0; c
< vd
->vdev_children
; c
++) {
5827 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5833 * Check for a completed replacement. We always consider the first
5834 * vdev in the list to be the oldest vdev, and the last one to be
5835 * the newest (see spa_vdev_attach() for how that works). In
5836 * the case where the newest vdev is faulted, we will not automatically
5837 * remove it after a resilver completes. This is OK as it will require
5838 * user intervention to determine which disk the admin wishes to keep.
5840 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5841 ASSERT(vd
->vdev_children
> 1);
5843 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5844 oldvd
= vd
->vdev_child
[0];
5846 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5847 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5848 !vdev_dtl_required(oldvd
))
5853 * Check for a completed resilver with the 'unspare' flag set.
5855 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5856 vdev_t
*first
= vd
->vdev_child
[0];
5857 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5859 if (last
->vdev_unspare
) {
5862 } else if (first
->vdev_unspare
) {
5869 if (oldvd
!= NULL
&&
5870 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5871 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5872 !vdev_dtl_required(oldvd
))
5876 * If there are more than two spares attached to a disk,
5877 * and those spares are not required, then we want to
5878 * attempt to free them up now so that they can be used
5879 * by other pools. Once we're back down to a single
5880 * disk+spare, we stop removing them.
5882 if (vd
->vdev_children
> 2) {
5883 newvd
= vd
->vdev_child
[1];
5885 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5886 vdev_dtl_empty(last
, DTL_MISSING
) &&
5887 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5888 !vdev_dtl_required(newvd
))
5897 spa_vdev_resilver_done(spa_t
*spa
)
5899 vdev_t
*vd
, *pvd
, *ppvd
;
5900 uint64_t guid
, sguid
, pguid
, ppguid
;
5902 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5904 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5905 pvd
= vd
->vdev_parent
;
5906 ppvd
= pvd
->vdev_parent
;
5907 guid
= vd
->vdev_guid
;
5908 pguid
= pvd
->vdev_guid
;
5909 ppguid
= ppvd
->vdev_guid
;
5912 * If we have just finished replacing a hot spared device, then
5913 * we need to detach the parent's first child (the original hot
5916 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5917 ppvd
->vdev_children
== 2) {
5918 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5919 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5921 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5923 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5924 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5926 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5928 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5931 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5935 * Update the stored path or FRU for this vdev.
5938 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5942 boolean_t sync
= B_FALSE
;
5944 ASSERT(spa_writeable(spa
));
5946 spa_vdev_state_enter(spa
, SCL_ALL
);
5948 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5949 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5951 if (!vd
->vdev_ops
->vdev_op_leaf
)
5952 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5955 if (strcmp(value
, vd
->vdev_path
) != 0) {
5956 spa_strfree(vd
->vdev_path
);
5957 vd
->vdev_path
= spa_strdup(value
);
5961 if (vd
->vdev_fru
== NULL
) {
5962 vd
->vdev_fru
= spa_strdup(value
);
5964 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5965 spa_strfree(vd
->vdev_fru
);
5966 vd
->vdev_fru
= spa_strdup(value
);
5971 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
5975 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
5977 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
5981 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
5983 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
5987 * ==========================================================================
5989 * ==========================================================================
5992 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
5994 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5996 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
5997 return (SET_ERROR(EBUSY
));
5999 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
6003 spa_scan_stop(spa_t
*spa
)
6005 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6006 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6007 return (SET_ERROR(EBUSY
));
6008 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
6012 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
6014 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6016 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
6017 return (SET_ERROR(ENOTSUP
));
6020 * If a resilver was requested, but there is no DTL on a
6021 * writeable leaf device, we have nothing to do.
6023 if (func
== POOL_SCAN_RESILVER
&&
6024 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
6025 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6029 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6033 * ==========================================================================
6034 * SPA async task processing
6035 * ==========================================================================
6039 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6043 if (vd
->vdev_remove_wanted
) {
6044 vd
->vdev_remove_wanted
= B_FALSE
;
6045 vd
->vdev_delayed_close
= B_FALSE
;
6046 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6049 * We want to clear the stats, but we don't want to do a full
6050 * vdev_clear() as that will cause us to throw away
6051 * degraded/faulted state as well as attempt to reopen the
6052 * device, all of which is a waste.
6054 vd
->vdev_stat
.vs_read_errors
= 0;
6055 vd
->vdev_stat
.vs_write_errors
= 0;
6056 vd
->vdev_stat
.vs_checksum_errors
= 0;
6058 vdev_state_dirty(vd
->vdev_top
);
6061 for (c
= 0; c
< vd
->vdev_children
; c
++)
6062 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6066 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6070 if (vd
->vdev_probe_wanted
) {
6071 vd
->vdev_probe_wanted
= B_FALSE
;
6072 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6075 for (c
= 0; c
< vd
->vdev_children
; c
++)
6076 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6080 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6084 if (!spa
->spa_autoexpand
)
6087 for (c
= 0; c
< vd
->vdev_children
; c
++) {
6088 vdev_t
*cvd
= vd
->vdev_child
[c
];
6089 spa_async_autoexpand(spa
, cvd
);
6092 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6095 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
6099 spa_async_thread(spa_t
*spa
)
6103 ASSERT(spa
->spa_sync_on
);
6105 mutex_enter(&spa
->spa_async_lock
);
6106 tasks
= spa
->spa_async_tasks
;
6107 spa
->spa_async_tasks
= 0;
6108 mutex_exit(&spa
->spa_async_lock
);
6111 * See if the config needs to be updated.
6113 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6114 uint64_t old_space
, new_space
;
6116 mutex_enter(&spa_namespace_lock
);
6117 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6118 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6119 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6120 mutex_exit(&spa_namespace_lock
);
6123 * If the pool grew as a result of the config update,
6124 * then log an internal history event.
6126 if (new_space
!= old_space
) {
6127 spa_history_log_internal(spa
, "vdev online", NULL
,
6128 "pool '%s' size: %llu(+%llu)",
6129 spa_name(spa
), new_space
, new_space
- old_space
);
6134 * See if any devices need to be marked REMOVED.
6136 if (tasks
& SPA_ASYNC_REMOVE
) {
6137 spa_vdev_state_enter(spa
, SCL_NONE
);
6138 spa_async_remove(spa
, spa
->spa_root_vdev
);
6139 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6140 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6141 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6142 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6143 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6146 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6147 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6148 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6149 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6153 * See if any devices need to be probed.
6155 if (tasks
& SPA_ASYNC_PROBE
) {
6156 spa_vdev_state_enter(spa
, SCL_NONE
);
6157 spa_async_probe(spa
, spa
->spa_root_vdev
);
6158 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6162 * If any devices are done replacing, detach them.
6164 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6165 spa_vdev_resilver_done(spa
);
6168 * Kick off a resilver.
6170 if (tasks
& SPA_ASYNC_RESILVER
)
6171 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6174 * Let the world know that we're done.
6176 mutex_enter(&spa
->spa_async_lock
);
6177 spa
->spa_async_thread
= NULL
;
6178 cv_broadcast(&spa
->spa_async_cv
);
6179 mutex_exit(&spa
->spa_async_lock
);
6184 spa_async_suspend(spa_t
*spa
)
6186 mutex_enter(&spa
->spa_async_lock
);
6187 spa
->spa_async_suspended
++;
6188 while (spa
->spa_async_thread
!= NULL
)
6189 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6190 mutex_exit(&spa
->spa_async_lock
);
6194 spa_async_resume(spa_t
*spa
)
6196 mutex_enter(&spa
->spa_async_lock
);
6197 ASSERT(spa
->spa_async_suspended
!= 0);
6198 spa
->spa_async_suspended
--;
6199 mutex_exit(&spa
->spa_async_lock
);
6203 spa_async_tasks_pending(spa_t
*spa
)
6205 uint_t non_config_tasks
;
6207 boolean_t config_task_suspended
;
6209 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6210 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6211 if (spa
->spa_ccw_fail_time
== 0) {
6212 config_task_suspended
= B_FALSE
;
6214 config_task_suspended
=
6215 (gethrtime() - spa
->spa_ccw_fail_time
) <
6216 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
6219 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6223 spa_async_dispatch(spa_t
*spa
)
6225 mutex_enter(&spa
->spa_async_lock
);
6226 if (spa_async_tasks_pending(spa
) &&
6227 !spa
->spa_async_suspended
&&
6228 spa
->spa_async_thread
== NULL
&&
6230 spa
->spa_async_thread
= thread_create(NULL
, 0,
6231 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6232 mutex_exit(&spa
->spa_async_lock
);
6236 spa_async_request(spa_t
*spa
, int task
)
6238 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6239 mutex_enter(&spa
->spa_async_lock
);
6240 spa
->spa_async_tasks
|= task
;
6241 mutex_exit(&spa
->spa_async_lock
);
6245 * ==========================================================================
6246 * SPA syncing routines
6247 * ==========================================================================
6251 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6254 bpobj_enqueue(bpo
, bp
, tx
);
6259 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6263 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6269 * Note: this simple function is not inlined to make it easier to dtrace the
6270 * amount of time spent syncing frees.
6273 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6275 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6276 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6277 VERIFY(zio_wait(zio
) == 0);
6281 * Note: this simple function is not inlined to make it easier to dtrace the
6282 * amount of time spent syncing deferred frees.
6285 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6287 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6288 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6289 spa_free_sync_cb
, zio
, tx
), ==, 0);
6290 VERIFY0(zio_wait(zio
));
6294 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6296 char *packed
= NULL
;
6301 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6304 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6305 * information. This avoids the dmu_buf_will_dirty() path and
6306 * saves us a pre-read to get data we don't actually care about.
6308 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6309 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6311 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6313 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6315 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6317 vmem_free(packed
, bufsize
);
6319 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6320 dmu_buf_will_dirty(db
, tx
);
6321 *(uint64_t *)db
->db_data
= nvsize
;
6322 dmu_buf_rele(db
, FTAG
);
6326 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6327 const char *config
, const char *entry
)
6337 * Update the MOS nvlist describing the list of available devices.
6338 * spa_validate_aux() will have already made sure this nvlist is
6339 * valid and the vdevs are labeled appropriately.
6341 if (sav
->sav_object
== 0) {
6342 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6343 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6344 sizeof (uint64_t), tx
);
6345 VERIFY(zap_update(spa
->spa_meta_objset
,
6346 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6347 &sav
->sav_object
, tx
) == 0);
6350 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6351 if (sav
->sav_count
== 0) {
6352 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6354 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6355 for (i
= 0; i
< sav
->sav_count
; i
++)
6356 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6357 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6358 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6359 sav
->sav_count
) == 0);
6360 for (i
= 0; i
< sav
->sav_count
; i
++)
6361 nvlist_free(list
[i
]);
6362 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6365 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6366 nvlist_free(nvroot
);
6368 sav
->sav_sync
= B_FALSE
;
6372 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6373 * The all-vdev ZAP must be empty.
6376 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6378 spa_t
*spa
= vd
->vdev_spa
;
6381 if (vd
->vdev_top_zap
!= 0) {
6382 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6383 vd
->vdev_top_zap
, tx
));
6385 if (vd
->vdev_leaf_zap
!= 0) {
6386 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6387 vd
->vdev_leaf_zap
, tx
));
6389 for (i
= 0; i
< vd
->vdev_children
; i
++) {
6390 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6395 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6400 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6401 * its config may not be dirty but we still need to build per-vdev ZAPs.
6402 * Similarly, if the pool is being assembled (e.g. after a split), we
6403 * need to rebuild the AVZ although the config may not be dirty.
6405 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6406 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6409 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6411 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6412 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6413 spa
->spa_all_vdev_zaps
!= 0);
6415 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6419 /* Make and build the new AVZ */
6420 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6421 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6422 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6424 /* Diff old AVZ with new one */
6425 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6426 spa
->spa_all_vdev_zaps
);
6427 zap_cursor_retrieve(&zc
, &za
) == 0;
6428 zap_cursor_advance(&zc
)) {
6429 uint64_t vdzap
= za
.za_first_integer
;
6430 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6433 * ZAP is listed in old AVZ but not in new one;
6436 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6441 zap_cursor_fini(&zc
);
6443 /* Destroy the old AVZ */
6444 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6445 spa
->spa_all_vdev_zaps
, tx
));
6447 /* Replace the old AVZ in the dir obj with the new one */
6448 VERIFY0(zap_update(spa
->spa_meta_objset
,
6449 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6450 sizeof (new_avz
), 1, &new_avz
, tx
));
6452 spa
->spa_all_vdev_zaps
= new_avz
;
6453 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6457 /* Walk through the AVZ and destroy all listed ZAPs */
6458 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6459 spa
->spa_all_vdev_zaps
);
6460 zap_cursor_retrieve(&zc
, &za
) == 0;
6461 zap_cursor_advance(&zc
)) {
6462 uint64_t zap
= za
.za_first_integer
;
6463 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6466 zap_cursor_fini(&zc
);
6468 /* Destroy and unlink the AVZ itself */
6469 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6470 spa
->spa_all_vdev_zaps
, tx
));
6471 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6472 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6473 spa
->spa_all_vdev_zaps
= 0;
6476 if (spa
->spa_all_vdev_zaps
== 0) {
6477 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6478 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6479 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6481 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6483 /* Create ZAPs for vdevs that don't have them. */
6484 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6486 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6487 dmu_tx_get_txg(tx
), B_FALSE
);
6490 * If we're upgrading the spa version then make sure that
6491 * the config object gets updated with the correct version.
6493 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6494 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6495 spa
->spa_uberblock
.ub_version
);
6497 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6499 nvlist_free(spa
->spa_config_syncing
);
6500 spa
->spa_config_syncing
= config
;
6502 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6506 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6508 uint64_t *versionp
= arg
;
6509 uint64_t version
= *versionp
;
6510 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6513 * Setting the version is special cased when first creating the pool.
6515 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6517 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6518 ASSERT(version
>= spa_version(spa
));
6520 spa
->spa_uberblock
.ub_version
= version
;
6521 vdev_config_dirty(spa
->spa_root_vdev
);
6522 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6526 * Set zpool properties.
6529 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6531 nvlist_t
*nvp
= arg
;
6532 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6533 objset_t
*mos
= spa
->spa_meta_objset
;
6534 nvpair_t
*elem
= NULL
;
6536 mutex_enter(&spa
->spa_props_lock
);
6538 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6540 char *strval
, *fname
;
6542 const char *propname
;
6543 zprop_type_t proptype
;
6546 prop
= zpool_name_to_prop(nvpair_name(elem
));
6547 switch ((int)prop
) {
6550 * We checked this earlier in spa_prop_validate().
6552 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6554 fname
= strchr(nvpair_name(elem
), '@') + 1;
6555 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6557 spa_feature_enable(spa
, fid
, tx
);
6558 spa_history_log_internal(spa
, "set", tx
,
6559 "%s=enabled", nvpair_name(elem
));
6562 case ZPOOL_PROP_VERSION
:
6563 intval
= fnvpair_value_uint64(elem
);
6565 * The version is synced separately before other
6566 * properties and should be correct by now.
6568 ASSERT3U(spa_version(spa
), >=, intval
);
6571 case ZPOOL_PROP_ALTROOT
:
6573 * 'altroot' is a non-persistent property. It should
6574 * have been set temporarily at creation or import time.
6576 ASSERT(spa
->spa_root
!= NULL
);
6579 case ZPOOL_PROP_READONLY
:
6580 case ZPOOL_PROP_CACHEFILE
:
6582 * 'readonly' and 'cachefile' are also non-persisitent
6586 case ZPOOL_PROP_COMMENT
:
6587 strval
= fnvpair_value_string(elem
);
6588 if (spa
->spa_comment
!= NULL
)
6589 spa_strfree(spa
->spa_comment
);
6590 spa
->spa_comment
= spa_strdup(strval
);
6592 * We need to dirty the configuration on all the vdevs
6593 * so that their labels get updated. It's unnecessary
6594 * to do this for pool creation since the vdev's
6595 * configuration has already been dirtied.
6597 if (tx
->tx_txg
!= TXG_INITIAL
)
6598 vdev_config_dirty(spa
->spa_root_vdev
);
6599 spa_history_log_internal(spa
, "set", tx
,
6600 "%s=%s", nvpair_name(elem
), strval
);
6604 * Set pool property values in the poolprops mos object.
6606 if (spa
->spa_pool_props_object
== 0) {
6607 spa
->spa_pool_props_object
=
6608 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6609 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6613 /* normalize the property name */
6614 propname
= zpool_prop_to_name(prop
);
6615 proptype
= zpool_prop_get_type(prop
);
6617 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6618 ASSERT(proptype
== PROP_TYPE_STRING
);
6619 strval
= fnvpair_value_string(elem
);
6620 VERIFY0(zap_update(mos
,
6621 spa
->spa_pool_props_object
, propname
,
6622 1, strlen(strval
) + 1, strval
, tx
));
6623 spa_history_log_internal(spa
, "set", tx
,
6624 "%s=%s", nvpair_name(elem
), strval
);
6625 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6626 intval
= fnvpair_value_uint64(elem
);
6628 if (proptype
== PROP_TYPE_INDEX
) {
6630 VERIFY0(zpool_prop_index_to_string(
6631 prop
, intval
, &unused
));
6633 VERIFY0(zap_update(mos
,
6634 spa
->spa_pool_props_object
, propname
,
6635 8, 1, &intval
, tx
));
6636 spa_history_log_internal(spa
, "set", tx
,
6637 "%s=%lld", nvpair_name(elem
), intval
);
6639 ASSERT(0); /* not allowed */
6643 case ZPOOL_PROP_DELEGATION
:
6644 spa
->spa_delegation
= intval
;
6646 case ZPOOL_PROP_BOOTFS
:
6647 spa
->spa_bootfs
= intval
;
6649 case ZPOOL_PROP_FAILUREMODE
:
6650 spa
->spa_failmode
= intval
;
6652 case ZPOOL_PROP_AUTOEXPAND
:
6653 spa
->spa_autoexpand
= intval
;
6654 if (tx
->tx_txg
!= TXG_INITIAL
)
6655 spa_async_request(spa
,
6656 SPA_ASYNC_AUTOEXPAND
);
6658 case ZPOOL_PROP_MULTIHOST
:
6659 spa
->spa_multihost
= intval
;
6661 case ZPOOL_PROP_DEDUPDITTO
:
6662 spa
->spa_dedup_ditto
= intval
;
6671 mutex_exit(&spa
->spa_props_lock
);
6675 * Perform one-time upgrade on-disk changes. spa_version() does not
6676 * reflect the new version this txg, so there must be no changes this
6677 * txg to anything that the upgrade code depends on after it executes.
6678 * Therefore this must be called after dsl_pool_sync() does the sync
6682 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6684 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6686 ASSERT(spa
->spa_sync_pass
== 1);
6688 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6690 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6691 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6692 dsl_pool_create_origin(dp
, tx
);
6694 /* Keeping the origin open increases spa_minref */
6695 spa
->spa_minref
+= 3;
6698 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6699 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6700 dsl_pool_upgrade_clones(dp
, tx
);
6703 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6704 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6705 dsl_pool_upgrade_dir_clones(dp
, tx
);
6707 /* Keeping the freedir open increases spa_minref */
6708 spa
->spa_minref
+= 3;
6711 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6712 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6713 spa_feature_create_zap_objects(spa
, tx
);
6717 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6718 * when possibility to use lz4 compression for metadata was added
6719 * Old pools that have this feature enabled must be upgraded to have
6720 * this feature active
6722 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6723 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6724 SPA_FEATURE_LZ4_COMPRESS
);
6725 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6726 SPA_FEATURE_LZ4_COMPRESS
);
6728 if (lz4_en
&& !lz4_ac
)
6729 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6733 * If we haven't written the salt, do so now. Note that the
6734 * feature may not be activated yet, but that's fine since
6735 * the presence of this ZAP entry is backwards compatible.
6737 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
6738 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
6739 VERIFY0(zap_add(spa
->spa_meta_objset
,
6740 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
6741 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
6742 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
6745 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6749 * Sync the specified transaction group. New blocks may be dirtied as
6750 * part of the process, so we iterate until it converges.
6753 spa_sync(spa_t
*spa
, uint64_t txg
)
6755 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6756 objset_t
*mos
= spa
->spa_meta_objset
;
6757 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6758 metaslab_class_t
*mc
;
6759 vdev_t
*rvd
= spa
->spa_root_vdev
;
6763 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
6764 zfs_vdev_queue_depth_pct
/ 100;
6765 uint64_t queue_depth_total
;
6768 VERIFY(spa_writeable(spa
));
6771 * Lock out configuration changes.
6773 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6775 spa
->spa_syncing_txg
= txg
;
6776 spa
->spa_sync_pass
= 0;
6778 mutex_enter(&spa
->spa_alloc_lock
);
6779 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
6780 mutex_exit(&spa
->spa_alloc_lock
);
6783 * If there are any pending vdev state changes, convert them
6784 * into config changes that go out with this transaction group.
6786 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6787 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6789 * We need the write lock here because, for aux vdevs,
6790 * calling vdev_config_dirty() modifies sav_config.
6791 * This is ugly and will become unnecessary when we
6792 * eliminate the aux vdev wart by integrating all vdevs
6793 * into the root vdev tree.
6795 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6796 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6797 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6798 vdev_state_clean(vd
);
6799 vdev_config_dirty(vd
);
6801 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6802 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6804 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6806 tx
= dmu_tx_create_assigned(dp
, txg
);
6808 spa
->spa_sync_starttime
= gethrtime();
6809 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6810 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
6811 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6812 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6815 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6816 * set spa_deflate if we have no raid-z vdevs.
6818 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6819 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6822 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6823 vd
= rvd
->vdev_child
[i
];
6824 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6827 if (i
== rvd
->vdev_children
) {
6828 spa
->spa_deflate
= TRUE
;
6829 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6830 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6831 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6836 * Set the top-level vdev's max queue depth. Evaluate each
6837 * top-level's async write queue depth in case it changed.
6838 * The max queue depth will not change in the middle of syncing
6841 queue_depth_total
= 0;
6842 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
6843 vdev_t
*tvd
= rvd
->vdev_child
[c
];
6844 metaslab_group_t
*mg
= tvd
->vdev_mg
;
6846 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
6847 !metaslab_group_initialized(mg
))
6851 * It is safe to do a lock-free check here because only async
6852 * allocations look at mg_max_alloc_queue_depth, and async
6853 * allocations all happen from spa_sync().
6855 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
6856 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
6857 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
6859 mc
= spa_normal_class(spa
);
6860 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
6861 mc
->mc_alloc_max_slots
= queue_depth_total
;
6862 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
6864 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
6865 max_queue_depth
* rvd
->vdev_children
);
6868 * Iterate to convergence.
6871 int pass
= ++spa
->spa_sync_pass
;
6873 spa_sync_config_object(spa
, tx
);
6874 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6875 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6876 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6877 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6878 spa_errlog_sync(spa
, txg
);
6879 dsl_pool_sync(dp
, txg
);
6881 if (pass
< zfs_sync_pass_deferred_free
) {
6882 spa_sync_frees(spa
, free_bpl
, tx
);
6885 * We can not defer frees in pass 1, because
6886 * we sync the deferred frees later in pass 1.
6888 ASSERT3U(pass
, >, 1);
6889 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6890 &spa
->spa_deferred_bpobj
, tx
);
6894 dsl_scan_sync(dp
, tx
);
6896 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6900 spa_sync_upgrades(spa
, tx
);
6902 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
6904 * Note: We need to check if the MOS is dirty
6905 * because we could have marked the MOS dirty
6906 * without updating the uberblock (e.g. if we
6907 * have sync tasks but no dirty user data). We
6908 * need to check the uberblock's rootbp because
6909 * it is updated if we have synced out dirty
6910 * data (though in this case the MOS will most
6911 * likely also be dirty due to second order
6912 * effects, we don't want to rely on that here).
6914 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
6915 !dmu_objset_is_dirty(mos
, txg
)) {
6917 * Nothing changed on the first pass,
6918 * therefore this TXG is a no-op. Avoid
6919 * syncing deferred frees, so that we
6920 * can keep this TXG as a no-op.
6922 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
6924 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6925 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
6928 spa_sync_deferred_frees(spa
, tx
);
6931 } while (dmu_objset_is_dirty(mos
, txg
));
6934 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
6936 * Make sure that the number of ZAPs for all the vdevs matches
6937 * the number of ZAPs in the per-vdev ZAP list. This only gets
6938 * called if the config is dirty; otherwise there may be
6939 * outstanding AVZ operations that weren't completed in
6940 * spa_sync_config_object.
6942 uint64_t all_vdev_zap_entry_count
;
6943 ASSERT0(zap_count(spa
->spa_meta_objset
,
6944 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
6945 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
6946 all_vdev_zap_entry_count
);
6951 * Rewrite the vdev configuration (which includes the uberblock)
6952 * to commit the transaction group.
6954 * If there are no dirty vdevs, we sync the uberblock to a few
6955 * random top-level vdevs that are known to be visible in the
6956 * config cache (see spa_vdev_add() for a complete description).
6957 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6961 * We hold SCL_STATE to prevent vdev open/close/etc.
6962 * while we're attempting to write the vdev labels.
6964 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6966 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6967 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6969 int children
= rvd
->vdev_children
;
6970 int c0
= spa_get_random(children
);
6972 for (c
= 0; c
< children
; c
++) {
6973 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6974 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6976 svd
[svdcount
++] = vd
;
6977 if (svdcount
== SPA_DVAS_PER_BP
)
6980 error
= vdev_config_sync(svd
, svdcount
, txg
);
6982 error
= vdev_config_sync(rvd
->vdev_child
,
6983 rvd
->vdev_children
, txg
);
6987 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
6989 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6993 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
6994 zio_resume_wait(spa
);
6998 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6999 spa
->spa_deadman_tqid
= 0;
7002 * Clear the dirty config list.
7004 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
7005 vdev_config_clean(vd
);
7008 * Now that the new config has synced transactionally,
7009 * let it become visible to the config cache.
7011 if (spa
->spa_config_syncing
!= NULL
) {
7012 spa_config_set(spa
, spa
->spa_config_syncing
);
7013 spa
->spa_config_txg
= txg
;
7014 spa
->spa_config_syncing
= NULL
;
7017 dsl_pool_sync_done(dp
, txg
);
7019 mutex_enter(&spa
->spa_alloc_lock
);
7020 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7021 mutex_exit(&spa
->spa_alloc_lock
);
7024 * Update usable space statistics.
7026 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
7027 vdev_sync_done(vd
, txg
);
7029 spa_update_dspace(spa
);
7032 * It had better be the case that we didn't dirty anything
7033 * since vdev_config_sync().
7035 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7036 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7037 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7039 spa
->spa_sync_pass
= 0;
7042 * Update the last synced uberblock here. We want to do this at
7043 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7044 * will be guaranteed that all the processing associated with
7045 * that txg has been completed.
7047 spa
->spa_ubsync
= spa
->spa_uberblock
;
7048 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7050 spa_handle_ignored_writes(spa
);
7053 * If any async tasks have been requested, kick them off.
7055 spa_async_dispatch(spa
);
7059 * Sync all pools. We don't want to hold the namespace lock across these
7060 * operations, so we take a reference on the spa_t and drop the lock during the
7064 spa_sync_allpools(void)
7067 mutex_enter(&spa_namespace_lock
);
7068 while ((spa
= spa_next(spa
)) != NULL
) {
7069 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7070 !spa_writeable(spa
) || spa_suspended(spa
))
7072 spa_open_ref(spa
, FTAG
);
7073 mutex_exit(&spa_namespace_lock
);
7074 txg_wait_synced(spa_get_dsl(spa
), 0);
7075 mutex_enter(&spa_namespace_lock
);
7076 spa_close(spa
, FTAG
);
7078 mutex_exit(&spa_namespace_lock
);
7082 * ==========================================================================
7083 * Miscellaneous routines
7084 * ==========================================================================
7088 * Remove all pools in the system.
7096 * Remove all cached state. All pools should be closed now,
7097 * so every spa in the AVL tree should be unreferenced.
7099 mutex_enter(&spa_namespace_lock
);
7100 while ((spa
= spa_next(NULL
)) != NULL
) {
7102 * Stop async tasks. The async thread may need to detach
7103 * a device that's been replaced, which requires grabbing
7104 * spa_namespace_lock, so we must drop it here.
7106 spa_open_ref(spa
, FTAG
);
7107 mutex_exit(&spa_namespace_lock
);
7108 spa_async_suspend(spa
);
7109 mutex_enter(&spa_namespace_lock
);
7110 spa_close(spa
, FTAG
);
7112 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7114 spa_deactivate(spa
);
7118 mutex_exit(&spa_namespace_lock
);
7122 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7127 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7131 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7132 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7133 if (vd
->vdev_guid
== guid
)
7137 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7138 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7139 if (vd
->vdev_guid
== guid
)
7148 spa_upgrade(spa_t
*spa
, uint64_t version
)
7150 ASSERT(spa_writeable(spa
));
7152 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7155 * This should only be called for a non-faulted pool, and since a
7156 * future version would result in an unopenable pool, this shouldn't be
7159 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7160 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7162 spa
->spa_uberblock
.ub_version
= version
;
7163 vdev_config_dirty(spa
->spa_root_vdev
);
7165 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7167 txg_wait_synced(spa_get_dsl(spa
), 0);
7171 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7175 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7177 for (i
= 0; i
< sav
->sav_count
; i
++)
7178 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7181 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7182 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7183 &spareguid
) == 0 && spareguid
== guid
)
7191 * Check if a pool has an active shared spare device.
7192 * Note: reference count of an active spare is 2, as a spare and as a replace
7195 spa_has_active_shared_spare(spa_t
*spa
)
7199 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7201 for (i
= 0; i
< sav
->sav_count
; i
++) {
7202 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7203 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7212 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7214 sysevent_t
*ev
= NULL
;
7218 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
7220 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
7221 ev
->resource
= resource
;
7228 spa_event_post(sysevent_t
*ev
)
7232 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
7233 kmem_free(ev
, sizeof (*ev
));
7239 * Post a zevent corresponding to the given sysevent. The 'name' must be one
7240 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
7241 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7242 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7243 * or zdb as real changes.
7246 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7248 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
7251 #if defined(_KERNEL) && defined(HAVE_SPL)
7252 /* state manipulation functions */
7253 EXPORT_SYMBOL(spa_open
);
7254 EXPORT_SYMBOL(spa_open_rewind
);
7255 EXPORT_SYMBOL(spa_get_stats
);
7256 EXPORT_SYMBOL(spa_create
);
7257 EXPORT_SYMBOL(spa_import
);
7258 EXPORT_SYMBOL(spa_tryimport
);
7259 EXPORT_SYMBOL(spa_destroy
);
7260 EXPORT_SYMBOL(spa_export
);
7261 EXPORT_SYMBOL(spa_reset
);
7262 EXPORT_SYMBOL(spa_async_request
);
7263 EXPORT_SYMBOL(spa_async_suspend
);
7264 EXPORT_SYMBOL(spa_async_resume
);
7265 EXPORT_SYMBOL(spa_inject_addref
);
7266 EXPORT_SYMBOL(spa_inject_delref
);
7267 EXPORT_SYMBOL(spa_scan_stat_init
);
7268 EXPORT_SYMBOL(spa_scan_get_stats
);
7270 /* device maniion */
7271 EXPORT_SYMBOL(spa_vdev_add
);
7272 EXPORT_SYMBOL(spa_vdev_attach
);
7273 EXPORT_SYMBOL(spa_vdev_detach
);
7274 EXPORT_SYMBOL(spa_vdev_remove
);
7275 EXPORT_SYMBOL(spa_vdev_setpath
);
7276 EXPORT_SYMBOL(spa_vdev_setfru
);
7277 EXPORT_SYMBOL(spa_vdev_split_mirror
);
7279 /* spare statech is global across all pools) */
7280 EXPORT_SYMBOL(spa_spare_add
);
7281 EXPORT_SYMBOL(spa_spare_remove
);
7282 EXPORT_SYMBOL(spa_spare_exists
);
7283 EXPORT_SYMBOL(spa_spare_activate
);
7285 /* L2ARC statech is global across all pools) */
7286 EXPORT_SYMBOL(spa_l2cache_add
);
7287 EXPORT_SYMBOL(spa_l2cache_remove
);
7288 EXPORT_SYMBOL(spa_l2cache_exists
);
7289 EXPORT_SYMBOL(spa_l2cache_activate
);
7290 EXPORT_SYMBOL(spa_l2cache_drop
);
7293 EXPORT_SYMBOL(spa_scan
);
7294 EXPORT_SYMBOL(spa_scan_stop
);
7297 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
7298 EXPORT_SYMBOL(spa_sync_allpools
);
7301 EXPORT_SYMBOL(spa_prop_set
);
7302 EXPORT_SYMBOL(spa_prop_get
);
7303 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
7305 /* asynchronous event notification */
7306 EXPORT_SYMBOL(spa_event_notify
);
7309 #if defined(_KERNEL) && defined(HAVE_SPL)
7310 module_param(spa_load_verify_maxinflight
, int, 0644);
7311 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
7312 "Max concurrent traversal I/Os while verifying pool during import -X");
7314 module_param(spa_load_verify_metadata
, int, 0644);
7315 MODULE_PARM_DESC(spa_load_verify_metadata
,
7316 "Set to traverse metadata on pool import");
7318 module_param(spa_load_verify_data
, int, 0644);
7319 MODULE_PARM_DESC(spa_load_verify_data
,
7320 "Set to traverse data on pool import");
7323 module_param(zio_taskq_batch_pct
, uint
, 0444);
7324 MODULE_PARM_DESC(zio_taskq_batch_pct
,
7325 "Percentage of CPUs to run an IO worker thread");