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
)) == ZPOOL_PROP_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
);
444 case ZPOOL_PROP_INVAL
:
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
== ZPOOL_PROP_INVAL
) {
704 if (prop
== ZPOOL_PROP_VERSION
) {
705 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
707 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
708 ver
= SPA_VERSION_FEATURES
;
712 /* Save time if the version is already set. */
713 if (ver
== spa_version(spa
))
717 * In addition to the pool directory object, we might
718 * create the pool properties object, the features for
719 * read object, the features for write object, or the
720 * feature descriptions object.
722 error
= dsl_sync_task(spa
->spa_name
, NULL
,
723 spa_sync_version
, &ver
,
724 6, ZFS_SPACE_CHECK_RESERVED
);
735 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
736 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
743 * If the bootfs property value is dsobj, clear it.
746 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
748 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
749 VERIFY(zap_remove(spa
->spa_meta_objset
,
750 spa
->spa_pool_props_object
,
751 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
758 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
760 ASSERTV(uint64_t *newguid
= arg
);
761 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
762 vdev_t
*rvd
= spa
->spa_root_vdev
;
765 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
766 vdev_state
= rvd
->vdev_state
;
767 spa_config_exit(spa
, SCL_STATE
, FTAG
);
769 if (vdev_state
!= VDEV_STATE_HEALTHY
)
770 return (SET_ERROR(ENXIO
));
772 ASSERT3U(spa_guid(spa
), !=, *newguid
);
778 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
780 uint64_t *newguid
= arg
;
781 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
783 vdev_t
*rvd
= spa
->spa_root_vdev
;
785 oldguid
= spa_guid(spa
);
787 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
788 rvd
->vdev_guid
= *newguid
;
789 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
790 vdev_config_dirty(rvd
);
791 spa_config_exit(spa
, SCL_STATE
, FTAG
);
793 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
798 * Change the GUID for the pool. This is done so that we can later
799 * re-import a pool built from a clone of our own vdevs. We will modify
800 * the root vdev's guid, our own pool guid, and then mark all of our
801 * vdevs dirty. Note that we must make sure that all our vdevs are
802 * online when we do this, or else any vdevs that weren't present
803 * would be orphaned from our pool. We are also going to issue a
804 * sysevent to update any watchers.
807 spa_change_guid(spa_t
*spa
)
812 mutex_enter(&spa
->spa_vdev_top_lock
);
813 mutex_enter(&spa_namespace_lock
);
814 guid
= spa_generate_guid(NULL
);
816 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
817 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
820 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
821 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
824 mutex_exit(&spa_namespace_lock
);
825 mutex_exit(&spa
->spa_vdev_top_lock
);
831 * ==========================================================================
832 * SPA state manipulation (open/create/destroy/import/export)
833 * ==========================================================================
837 spa_error_entry_compare(const void *a
, const void *b
)
839 const spa_error_entry_t
*sa
= (const spa_error_entry_t
*)a
;
840 const spa_error_entry_t
*sb
= (const spa_error_entry_t
*)b
;
843 ret
= memcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
844 sizeof (zbookmark_phys_t
));
846 return (AVL_ISIGN(ret
));
850 * Utility function which retrieves copies of the current logs and
851 * re-initializes them in the process.
854 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
856 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
858 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
859 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
861 avl_create(&spa
->spa_errlist_scrub
,
862 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
863 offsetof(spa_error_entry_t
, se_avl
));
864 avl_create(&spa
->spa_errlist_last
,
865 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
866 offsetof(spa_error_entry_t
, se_avl
));
870 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
872 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
873 enum zti_modes mode
= ztip
->zti_mode
;
874 uint_t value
= ztip
->zti_value
;
875 uint_t count
= ztip
->zti_count
;
876 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
879 boolean_t batch
= B_FALSE
;
881 if (mode
== ZTI_MODE_NULL
) {
883 tqs
->stqs_taskq
= NULL
;
887 ASSERT3U(count
, >, 0);
889 tqs
->stqs_count
= count
;
890 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
894 ASSERT3U(value
, >=, 1);
895 value
= MAX(value
, 1);
896 flags
|= TASKQ_DYNAMIC
;
901 flags
|= TASKQ_THREADS_CPU_PCT
;
902 value
= MIN(zio_taskq_batch_pct
, 100);
906 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
908 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
912 for (uint_t i
= 0; i
< count
; i
++) {
916 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
917 zio_type_name
[t
], zio_taskq_types
[q
], i
);
919 (void) snprintf(name
, sizeof (name
), "%s_%s",
920 zio_type_name
[t
], zio_taskq_types
[q
]);
923 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
925 flags
|= TASKQ_DC_BATCH
;
927 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
928 spa
->spa_proc
, zio_taskq_basedc
, flags
);
930 pri_t pri
= maxclsyspri
;
932 * The write issue taskq can be extremely CPU
933 * intensive. Run it at slightly less important
934 * priority than the other taskqs. Under Linux this
935 * means incrementing the priority value on platforms
936 * like illumos it should be decremented.
938 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
941 tq
= taskq_create_proc(name
, value
, pri
, 50,
942 INT_MAX
, spa
->spa_proc
, flags
);
945 tqs
->stqs_taskq
[i
] = tq
;
950 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
952 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
954 if (tqs
->stqs_taskq
== NULL
) {
955 ASSERT3U(tqs
->stqs_count
, ==, 0);
959 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
960 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
961 taskq_destroy(tqs
->stqs_taskq
[i
]);
964 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
965 tqs
->stqs_taskq
= NULL
;
969 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
970 * Note that a type may have multiple discrete taskqs to avoid lock contention
971 * on the taskq itself. In that case we choose which taskq at random by using
972 * the low bits of gethrtime().
975 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
976 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
978 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
981 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
982 ASSERT3U(tqs
->stqs_count
, !=, 0);
984 if (tqs
->stqs_count
== 1) {
985 tq
= tqs
->stqs_taskq
[0];
987 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
990 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
994 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
997 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
998 task_func_t
*func
, void *arg
, uint_t flags
)
1000 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1004 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1005 ASSERT3U(tqs
->stqs_count
, !=, 0);
1007 if (tqs
->stqs_count
== 1) {
1008 tq
= tqs
->stqs_taskq
[0];
1010 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1013 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1015 taskq_wait_id(tq
, id
);
1019 spa_create_zio_taskqs(spa_t
*spa
)
1021 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1022 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1023 spa_taskqs_init(spa
, t
, q
);
1029 * Disabled until spa_thread() can be adapted for Linux.
1031 #undef HAVE_SPA_THREAD
1033 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1035 spa_thread(void *arg
)
1037 callb_cpr_t cprinfo
;
1040 user_t
*pu
= PTOU(curproc
);
1042 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1045 ASSERT(curproc
!= &p0
);
1046 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1047 "zpool-%s", spa
->spa_name
);
1048 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1050 /* bind this thread to the requested psrset */
1051 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1053 mutex_enter(&cpu_lock
);
1054 mutex_enter(&pidlock
);
1055 mutex_enter(&curproc
->p_lock
);
1057 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1058 0, NULL
, NULL
) == 0) {
1059 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1062 "Couldn't bind process for zfs pool \"%s\" to "
1063 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1066 mutex_exit(&curproc
->p_lock
);
1067 mutex_exit(&pidlock
);
1068 mutex_exit(&cpu_lock
);
1072 if (zio_taskq_sysdc
) {
1073 sysdc_thread_enter(curthread
, 100, 0);
1076 spa
->spa_proc
= curproc
;
1077 spa
->spa_did
= curthread
->t_did
;
1079 spa_create_zio_taskqs(spa
);
1081 mutex_enter(&spa
->spa_proc_lock
);
1082 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1084 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1085 cv_broadcast(&spa
->spa_proc_cv
);
1087 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1088 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1089 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1090 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1092 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1093 spa
->spa_proc_state
= SPA_PROC_GONE
;
1094 spa
->spa_proc
= &p0
;
1095 cv_broadcast(&spa
->spa_proc_cv
);
1096 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1098 mutex_enter(&curproc
->p_lock
);
1104 * Activate an uninitialized pool.
1107 spa_activate(spa_t
*spa
, int mode
)
1109 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1111 spa
->spa_state
= POOL_STATE_ACTIVE
;
1112 spa
->spa_mode
= mode
;
1114 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1115 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1117 /* Try to create a covering process */
1118 mutex_enter(&spa
->spa_proc_lock
);
1119 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1120 ASSERT(spa
->spa_proc
== &p0
);
1123 #ifdef HAVE_SPA_THREAD
1124 /* Only create a process if we're going to be around a while. */
1125 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1126 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1128 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1129 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1130 cv_wait(&spa
->spa_proc_cv
,
1131 &spa
->spa_proc_lock
);
1133 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1134 ASSERT(spa
->spa_proc
!= &p0
);
1135 ASSERT(spa
->spa_did
!= 0);
1139 "Couldn't create process for zfs pool \"%s\"\n",
1144 #endif /* HAVE_SPA_THREAD */
1145 mutex_exit(&spa
->spa_proc_lock
);
1147 /* If we didn't create a process, we need to create our taskqs. */
1148 if (spa
->spa_proc
== &p0
) {
1149 spa_create_zio_taskqs(spa
);
1152 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1153 offsetof(vdev_t
, vdev_config_dirty_node
));
1154 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1155 offsetof(objset_t
, os_evicting_node
));
1156 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1157 offsetof(vdev_t
, vdev_state_dirty_node
));
1159 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1160 offsetof(struct vdev
, vdev_txg_node
));
1162 avl_create(&spa
->spa_errlist_scrub
,
1163 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1164 offsetof(spa_error_entry_t
, se_avl
));
1165 avl_create(&spa
->spa_errlist_last
,
1166 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1167 offsetof(spa_error_entry_t
, se_avl
));
1169 spa_keystore_init(&spa
->spa_keystore
);
1172 * This taskq is used to perform zvol-minor-related tasks
1173 * asynchronously. This has several advantages, including easy
1174 * resolution of various deadlocks (zfsonlinux bug #3681).
1176 * The taskq must be single threaded to ensure tasks are always
1177 * processed in the order in which they were dispatched.
1179 * A taskq per pool allows one to keep the pools independent.
1180 * This way if one pool is suspended, it will not impact another.
1182 * The preferred location to dispatch a zvol minor task is a sync
1183 * task. In this context, there is easy access to the spa_t and minimal
1184 * error handling is required because the sync task must succeed.
1186 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1190 * Taskq dedicated to prefetcher threads: this is used to prevent the
1191 * pool traverse code from monopolizing the global (and limited)
1192 * system_taskq by inappropriately scheduling long running tasks on it.
1194 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", boot_ncpus
,
1195 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1198 * The taskq to upgrade datasets in this pool. Currently used by
1199 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1201 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", boot_ncpus
,
1202 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1206 * Opposite of spa_activate().
1209 spa_deactivate(spa_t
*spa
)
1211 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1212 ASSERT(spa
->spa_dsl_pool
== NULL
);
1213 ASSERT(spa
->spa_root_vdev
== NULL
);
1214 ASSERT(spa
->spa_async_zio_root
== NULL
);
1215 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1217 spa_evicting_os_wait(spa
);
1219 if (spa
->spa_zvol_taskq
) {
1220 taskq_destroy(spa
->spa_zvol_taskq
);
1221 spa
->spa_zvol_taskq
= NULL
;
1224 if (spa
->spa_prefetch_taskq
) {
1225 taskq_destroy(spa
->spa_prefetch_taskq
);
1226 spa
->spa_prefetch_taskq
= NULL
;
1229 if (spa
->spa_upgrade_taskq
) {
1230 taskq_destroy(spa
->spa_upgrade_taskq
);
1231 spa
->spa_upgrade_taskq
= NULL
;
1234 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1236 list_destroy(&spa
->spa_config_dirty_list
);
1237 list_destroy(&spa
->spa_evicting_os_list
);
1238 list_destroy(&spa
->spa_state_dirty_list
);
1240 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1242 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1243 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1244 spa_taskqs_fini(spa
, t
, q
);
1248 metaslab_class_destroy(spa
->spa_normal_class
);
1249 spa
->spa_normal_class
= NULL
;
1251 metaslab_class_destroy(spa
->spa_log_class
);
1252 spa
->spa_log_class
= NULL
;
1255 * If this was part of an import or the open otherwise failed, we may
1256 * still have errors left in the queues. Empty them just in case.
1258 spa_errlog_drain(spa
);
1259 avl_destroy(&spa
->spa_errlist_scrub
);
1260 avl_destroy(&spa
->spa_errlist_last
);
1262 spa_keystore_fini(&spa
->spa_keystore
);
1264 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1266 mutex_enter(&spa
->spa_proc_lock
);
1267 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1268 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1269 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1270 cv_broadcast(&spa
->spa_proc_cv
);
1271 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1272 ASSERT(spa
->spa_proc
!= &p0
);
1273 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1275 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1276 spa
->spa_proc_state
= SPA_PROC_NONE
;
1278 ASSERT(spa
->spa_proc
== &p0
);
1279 mutex_exit(&spa
->spa_proc_lock
);
1282 * We want to make sure spa_thread() has actually exited the ZFS
1283 * module, so that the module can't be unloaded out from underneath
1286 if (spa
->spa_did
!= 0) {
1287 thread_join(spa
->spa_did
);
1293 * Verify a pool configuration, and construct the vdev tree appropriately. This
1294 * will create all the necessary vdevs in the appropriate layout, with each vdev
1295 * in the CLOSED state. This will prep the pool before open/creation/import.
1296 * All vdev validation is done by the vdev_alloc() routine.
1299 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1300 uint_t id
, int atype
)
1306 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1309 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1312 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1315 if (error
== ENOENT
)
1321 return (SET_ERROR(EINVAL
));
1324 for (int c
= 0; c
< children
; c
++) {
1326 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1334 ASSERT(*vdp
!= NULL
);
1340 * Opposite of spa_load().
1343 spa_unload(spa_t
*spa
)
1347 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1352 spa_async_suspend(spa
);
1357 if (spa
->spa_sync_on
) {
1358 txg_sync_stop(spa
->spa_dsl_pool
);
1359 spa
->spa_sync_on
= B_FALSE
;
1363 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1364 * to call it earlier, before we wait for async i/o to complete.
1365 * This ensures that there is no async metaslab prefetching, by
1366 * calling taskq_wait(mg_taskq).
1368 if (spa
->spa_root_vdev
!= NULL
) {
1369 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1370 for (int c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++)
1371 vdev_metaslab_fini(spa
->spa_root_vdev
->vdev_child
[c
]);
1372 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1375 if (spa
->spa_mmp
.mmp_thread
)
1376 mmp_thread_stop(spa
);
1379 * Wait for any outstanding async I/O to complete.
1381 if (spa
->spa_async_zio_root
!= NULL
) {
1382 for (int i
= 0; i
< max_ncpus
; i
++)
1383 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1384 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1385 spa
->spa_async_zio_root
= NULL
;
1388 bpobj_close(&spa
->spa_deferred_bpobj
);
1390 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1395 if (spa
->spa_root_vdev
)
1396 vdev_free(spa
->spa_root_vdev
);
1397 ASSERT(spa
->spa_root_vdev
== NULL
);
1400 * Close the dsl pool.
1402 if (spa
->spa_dsl_pool
) {
1403 dsl_pool_close(spa
->spa_dsl_pool
);
1404 spa
->spa_dsl_pool
= NULL
;
1405 spa
->spa_meta_objset
= NULL
;
1411 * Drop and purge level 2 cache
1413 spa_l2cache_drop(spa
);
1415 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1416 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1417 if (spa
->spa_spares
.sav_vdevs
) {
1418 kmem_free(spa
->spa_spares
.sav_vdevs
,
1419 spa
->spa_spares
.sav_count
* sizeof (void *));
1420 spa
->spa_spares
.sav_vdevs
= NULL
;
1422 if (spa
->spa_spares
.sav_config
) {
1423 nvlist_free(spa
->spa_spares
.sav_config
);
1424 spa
->spa_spares
.sav_config
= NULL
;
1426 spa
->spa_spares
.sav_count
= 0;
1428 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1429 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1430 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1432 if (spa
->spa_l2cache
.sav_vdevs
) {
1433 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1434 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1435 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1437 if (spa
->spa_l2cache
.sav_config
) {
1438 nvlist_free(spa
->spa_l2cache
.sav_config
);
1439 spa
->spa_l2cache
.sav_config
= NULL
;
1441 spa
->spa_l2cache
.sav_count
= 0;
1443 spa
->spa_async_suspended
= 0;
1445 if (spa
->spa_comment
!= NULL
) {
1446 spa_strfree(spa
->spa_comment
);
1447 spa
->spa_comment
= NULL
;
1450 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1454 * Load (or re-load) the current list of vdevs describing the active spares for
1455 * this pool. When this is called, we have some form of basic information in
1456 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1457 * then re-generate a more complete list including status information.
1460 spa_load_spares(spa_t
*spa
)
1467 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1470 * First, close and free any existing spare vdevs.
1472 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1473 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1475 /* Undo the call to spa_activate() below */
1476 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1477 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1478 spa_spare_remove(tvd
);
1483 if (spa
->spa_spares
.sav_vdevs
)
1484 kmem_free(spa
->spa_spares
.sav_vdevs
,
1485 spa
->spa_spares
.sav_count
* sizeof (void *));
1487 if (spa
->spa_spares
.sav_config
== NULL
)
1490 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1491 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1493 spa
->spa_spares
.sav_count
= (int)nspares
;
1494 spa
->spa_spares
.sav_vdevs
= NULL
;
1500 * Construct the array of vdevs, opening them to get status in the
1501 * process. For each spare, there is potentially two different vdev_t
1502 * structures associated with it: one in the list of spares (used only
1503 * for basic validation purposes) and one in the active vdev
1504 * configuration (if it's spared in). During this phase we open and
1505 * validate each vdev on the spare list. If the vdev also exists in the
1506 * active configuration, then we also mark this vdev as an active spare.
1508 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1510 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1511 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1512 VDEV_ALLOC_SPARE
) == 0);
1515 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1517 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1518 B_FALSE
)) != NULL
) {
1519 if (!tvd
->vdev_isspare
)
1523 * We only mark the spare active if we were successfully
1524 * able to load the vdev. Otherwise, importing a pool
1525 * with a bad active spare would result in strange
1526 * behavior, because multiple pool would think the spare
1527 * is actively in use.
1529 * There is a vulnerability here to an equally bizarre
1530 * circumstance, where a dead active spare is later
1531 * brought back to life (onlined or otherwise). Given
1532 * the rarity of this scenario, and the extra complexity
1533 * it adds, we ignore the possibility.
1535 if (!vdev_is_dead(tvd
))
1536 spa_spare_activate(tvd
);
1540 vd
->vdev_aux
= &spa
->spa_spares
;
1542 if (vdev_open(vd
) != 0)
1545 if (vdev_validate_aux(vd
) == 0)
1550 * Recompute the stashed list of spares, with status information
1553 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1554 DATA_TYPE_NVLIST_ARRAY
) == 0);
1556 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1558 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1559 spares
[i
] = vdev_config_generate(spa
,
1560 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1561 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1562 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1563 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1564 nvlist_free(spares
[i
]);
1565 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1569 * Load (or re-load) the current list of vdevs describing the active l2cache for
1570 * this pool. When this is called, we have some form of basic information in
1571 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1572 * then re-generate a more complete list including status information.
1573 * Devices which are already active have their details maintained, and are
1577 spa_load_l2cache(spa_t
*spa
)
1579 nvlist_t
**l2cache
= NULL
;
1581 int i
, j
, oldnvdevs
;
1583 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1584 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1586 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1588 oldvdevs
= sav
->sav_vdevs
;
1589 oldnvdevs
= sav
->sav_count
;
1590 sav
->sav_vdevs
= NULL
;
1593 if (sav
->sav_config
== NULL
) {
1599 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1600 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1601 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1604 * Process new nvlist of vdevs.
1606 for (i
= 0; i
< nl2cache
; i
++) {
1607 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1611 for (j
= 0; j
< oldnvdevs
; j
++) {
1613 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1615 * Retain previous vdev for add/remove ops.
1623 if (newvdevs
[i
] == NULL
) {
1627 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1628 VDEV_ALLOC_L2CACHE
) == 0);
1633 * Commit this vdev as an l2cache device,
1634 * even if it fails to open.
1636 spa_l2cache_add(vd
);
1641 spa_l2cache_activate(vd
);
1643 if (vdev_open(vd
) != 0)
1646 (void) vdev_validate_aux(vd
);
1648 if (!vdev_is_dead(vd
))
1649 l2arc_add_vdev(spa
, vd
);
1653 sav
->sav_vdevs
= newvdevs
;
1654 sav
->sav_count
= (int)nl2cache
;
1657 * Recompute the stashed list of l2cache devices, with status
1658 * information this time.
1660 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1661 DATA_TYPE_NVLIST_ARRAY
) == 0);
1663 if (sav
->sav_count
> 0)
1664 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *),
1666 for (i
= 0; i
< sav
->sav_count
; i
++)
1667 l2cache
[i
] = vdev_config_generate(spa
,
1668 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1669 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1670 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1674 * Purge vdevs that were dropped
1676 for (i
= 0; i
< oldnvdevs
; i
++) {
1681 ASSERT(vd
->vdev_isl2cache
);
1683 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1684 pool
!= 0ULL && l2arc_vdev_present(vd
))
1685 l2arc_remove_vdev(vd
);
1686 vdev_clear_stats(vd
);
1692 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1694 for (i
= 0; i
< sav
->sav_count
; i
++)
1695 nvlist_free(l2cache
[i
]);
1697 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1701 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1704 char *packed
= NULL
;
1709 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1713 nvsize
= *(uint64_t *)db
->db_data
;
1714 dmu_buf_rele(db
, FTAG
);
1716 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1717 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1720 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1721 vmem_free(packed
, nvsize
);
1727 * Checks to see if the given vdev could not be opened, in which case we post a
1728 * sysevent to notify the autoreplace code that the device has been removed.
1731 spa_check_removed(vdev_t
*vd
)
1733 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1734 spa_check_removed(vd
->vdev_child
[c
]);
1736 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1738 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1739 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
1744 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 (uint64_t 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
;
1765 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nv
) == 0);
1767 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1768 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
1770 ASSERT3U(rvd
->vdev_children
, ==, mrvd
->vdev_children
);
1773 * If we're doing a normal import, then build up any additional
1774 * diagnostic information about missing devices in this config.
1775 * We'll pass this up to the user for further processing.
1777 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1778 nvlist_t
**child
, *nv
;
1781 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
1783 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1785 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1786 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1787 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1789 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1790 mtvd
->vdev_ops
!= &vdev_missing_ops
&&
1792 child
[idx
++] = vdev_config_generate(spa
, mtvd
,
1797 VERIFY(nvlist_add_nvlist_array(nv
,
1798 ZPOOL_CONFIG_CHILDREN
, child
, idx
) == 0);
1799 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
1800 ZPOOL_CONFIG_MISSING_DEVICES
, nv
) == 0);
1802 for (int i
= 0; i
< idx
; i
++)
1803 nvlist_free(child
[i
]);
1806 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1810 * Compare the root vdev tree with the information we have
1811 * from the MOS config (mrvd). Check each top-level vdev
1812 * with the corresponding MOS config top-level (mtvd).
1814 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1815 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1816 vdev_t
*mtvd
= mrvd
->vdev_child
[c
];
1819 * Resolve any "missing" vdevs in the current configuration.
1820 * If we find that the MOS config has more accurate information
1821 * about the top-level vdev then use that vdev instead.
1823 if (tvd
->vdev_ops
== &vdev_missing_ops
&&
1824 mtvd
->vdev_ops
!= &vdev_missing_ops
) {
1826 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
))
1830 * Device specific actions.
1832 if (mtvd
->vdev_islog
) {
1833 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1836 * XXX - once we have 'readonly' pool
1837 * support we should be able to handle
1838 * missing data devices by transitioning
1839 * the pool to readonly.
1845 * Swap the missing vdev with the data we were
1846 * able to obtain from the MOS config.
1848 vdev_remove_child(rvd
, tvd
);
1849 vdev_remove_child(mrvd
, mtvd
);
1851 vdev_add_child(rvd
, mtvd
);
1852 vdev_add_child(mrvd
, tvd
);
1854 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1856 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1860 if (mtvd
->vdev_islog
) {
1862 * Load the slog device's state from the MOS
1863 * config since it's possible that the label
1864 * does not contain the most up-to-date
1867 vdev_load_log_state(tvd
, mtvd
);
1872 * Per-vdev ZAP info is stored exclusively in the MOS.
1874 spa_config_valid_zaps(tvd
, mtvd
);
1879 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1882 * Ensure we were able to validate the config.
1884 return (rvd
->vdev_guid_sum
== spa
->spa_uberblock
.ub_guid_sum
);
1888 * Check for missing log devices
1891 spa_check_logs(spa_t
*spa
)
1893 boolean_t rv
= B_FALSE
;
1894 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1896 switch (spa
->spa_log_state
) {
1899 case SPA_LOG_MISSING
:
1900 /* need to recheck in case slog has been restored */
1901 case SPA_LOG_UNKNOWN
:
1902 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1903 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1905 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1912 spa_passivate_log(spa_t
*spa
)
1914 vdev_t
*rvd
= spa
->spa_root_vdev
;
1915 boolean_t slog_found
= B_FALSE
;
1917 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1919 if (!spa_has_slogs(spa
))
1922 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1923 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1924 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1926 if (tvd
->vdev_islog
) {
1927 metaslab_group_passivate(mg
);
1928 slog_found
= B_TRUE
;
1932 return (slog_found
);
1936 spa_activate_log(spa_t
*spa
)
1938 vdev_t
*rvd
= spa
->spa_root_vdev
;
1940 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1942 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1943 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1944 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1946 if (tvd
->vdev_islog
)
1947 metaslab_group_activate(mg
);
1952 spa_offline_log(spa_t
*spa
)
1956 error
= dmu_objset_find(spa_name(spa
), zil_vdev_offline
,
1957 NULL
, DS_FIND_CHILDREN
);
1960 * We successfully offlined the log device, sync out the
1961 * current txg so that the "stubby" block can be removed
1964 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1970 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1972 for (int i
= 0; i
< sav
->sav_count
; i
++)
1973 spa_check_removed(sav
->sav_vdevs
[i
]);
1977 spa_claim_notify(zio_t
*zio
)
1979 spa_t
*spa
= zio
->io_spa
;
1984 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1985 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1986 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1987 mutex_exit(&spa
->spa_props_lock
);
1990 typedef struct spa_load_error
{
1991 uint64_t sle_meta_count
;
1992 uint64_t sle_data_count
;
1996 spa_load_verify_done(zio_t
*zio
)
1998 blkptr_t
*bp
= zio
->io_bp
;
1999 spa_load_error_t
*sle
= zio
->io_private
;
2000 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2001 int error
= zio
->io_error
;
2002 spa_t
*spa
= zio
->io_spa
;
2004 abd_free(zio
->io_abd
);
2006 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2007 type
!= DMU_OT_INTENT_LOG
)
2008 atomic_inc_64(&sle
->sle_meta_count
);
2010 atomic_inc_64(&sle
->sle_data_count
);
2013 mutex_enter(&spa
->spa_scrub_lock
);
2014 spa
->spa_load_verify_ios
--;
2015 cv_broadcast(&spa
->spa_scrub_io_cv
);
2016 mutex_exit(&spa
->spa_scrub_lock
);
2020 * Maximum number of concurrent scrub i/os to create while verifying
2021 * a pool while importing it.
2023 int spa_load_verify_maxinflight
= 10000;
2024 int spa_load_verify_metadata
= B_TRUE
;
2025 int spa_load_verify_data
= B_TRUE
;
2029 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2030 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2032 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
2035 * Note: normally this routine will not be called if
2036 * spa_load_verify_metadata is not set. However, it may be useful
2037 * to manually set the flag after the traversal has begun.
2039 if (!spa_load_verify_metadata
)
2041 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2045 size_t size
= BP_GET_PSIZE(bp
);
2047 mutex_enter(&spa
->spa_scrub_lock
);
2048 while (spa
->spa_load_verify_ios
>= spa_load_verify_maxinflight
)
2049 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2050 spa
->spa_load_verify_ios
++;
2051 mutex_exit(&spa
->spa_scrub_lock
);
2053 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2054 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2055 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2056 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2062 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2064 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2065 return (SET_ERROR(ENAMETOOLONG
));
2071 spa_load_verify(spa_t
*spa
)
2074 spa_load_error_t sle
= { 0 };
2075 zpool_rewind_policy_t policy
;
2076 boolean_t verify_ok
= B_FALSE
;
2079 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
2081 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
2084 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2085 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2086 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2088 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2092 rio
= zio_root(spa
, NULL
, &sle
,
2093 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2095 if (spa_load_verify_metadata
) {
2096 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2097 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2098 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2101 (void) zio_wait(rio
);
2103 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2104 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2106 if (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
2107 sle
.sle_data_count
<= policy
.zrp_maxdata
) {
2111 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2112 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2114 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2115 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2116 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2117 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2118 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2119 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2120 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2122 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2126 if (error
!= ENXIO
&& error
!= EIO
)
2127 error
= SET_ERROR(EIO
);
2131 return (verify_ok
? 0 : EIO
);
2135 * Find a value in the pool props object.
2138 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2140 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2141 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2145 * Find a value in the pool directory object.
2148 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
)
2150 return (zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2151 name
, sizeof (uint64_t), 1, val
));
2155 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2157 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2162 * Fix up config after a partly-completed split. This is done with the
2163 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2164 * pool have that entry in their config, but only the splitting one contains
2165 * a list of all the guids of the vdevs that are being split off.
2167 * This function determines what to do with that list: either rejoin
2168 * all the disks to the pool, or complete the splitting process. To attempt
2169 * the rejoin, each disk that is offlined is marked online again, and
2170 * we do a reopen() call. If the vdev label for every disk that was
2171 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2172 * then we call vdev_split() on each disk, and complete the split.
2174 * Otherwise we leave the config alone, with all the vdevs in place in
2175 * the original pool.
2178 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2185 boolean_t attempt_reopen
;
2187 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2190 /* check that the config is complete */
2191 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2192 &glist
, &gcount
) != 0)
2195 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2197 /* attempt to online all the vdevs & validate */
2198 attempt_reopen
= B_TRUE
;
2199 for (i
= 0; i
< gcount
; i
++) {
2200 if (glist
[i
] == 0) /* vdev is hole */
2203 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2204 if (vd
[i
] == NULL
) {
2206 * Don't bother attempting to reopen the disks;
2207 * just do the split.
2209 attempt_reopen
= B_FALSE
;
2211 /* attempt to re-online it */
2212 vd
[i
]->vdev_offline
= B_FALSE
;
2216 if (attempt_reopen
) {
2217 vdev_reopen(spa
->spa_root_vdev
);
2219 /* check each device to see what state it's in */
2220 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2221 if (vd
[i
] != NULL
&&
2222 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2229 * If every disk has been moved to the new pool, or if we never
2230 * even attempted to look at them, then we split them off for
2233 if (!attempt_reopen
|| gcount
== extracted
) {
2234 for (i
= 0; i
< gcount
; i
++)
2237 vdev_reopen(spa
->spa_root_vdev
);
2240 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2244 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
,
2245 boolean_t mosconfig
)
2247 nvlist_t
*config
= spa
->spa_config
;
2248 char *ereport
= FM_EREPORT_ZFS_POOL
;
2254 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
))
2255 return (SET_ERROR(EINVAL
));
2257 ASSERT(spa
->spa_comment
== NULL
);
2258 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2259 spa
->spa_comment
= spa_strdup(comment
);
2262 * Versioning wasn't explicitly added to the label until later, so if
2263 * it's not present treat it as the initial version.
2265 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2266 &spa
->spa_ubsync
.ub_version
) != 0)
2267 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2269 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2270 &spa
->spa_config_txg
);
2272 if ((state
== SPA_LOAD_IMPORT
|| state
== SPA_LOAD_TRYIMPORT
) &&
2273 spa_guid_exists(pool_guid
, 0)) {
2274 error
= SET_ERROR(EEXIST
);
2276 spa
->spa_config_guid
= pool_guid
;
2278 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
,
2280 VERIFY(nvlist_dup(nvl
, &spa
->spa_config_splitting
,
2284 nvlist_free(spa
->spa_load_info
);
2285 spa
->spa_load_info
= fnvlist_alloc();
2287 gethrestime(&spa
->spa_loaded_ts
);
2288 error
= spa_load_impl(spa
, pool_guid
, config
, state
, type
,
2289 mosconfig
, &ereport
);
2293 * Don't count references from objsets that are already closed
2294 * and are making their way through the eviction process.
2296 spa_evicting_os_wait(spa
);
2297 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2299 if (error
!= EEXIST
) {
2300 spa
->spa_loaded_ts
.tv_sec
= 0;
2301 spa
->spa_loaded_ts
.tv_nsec
= 0;
2303 if (error
!= EBADF
) {
2304 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, NULL
, 0, 0);
2307 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2315 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2316 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2317 * spa's per-vdev ZAP list.
2320 vdev_count_verify_zaps(vdev_t
*vd
)
2322 spa_t
*spa
= vd
->vdev_spa
;
2325 if (vd
->vdev_top_zap
!= 0) {
2327 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2328 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2330 if (vd
->vdev_leaf_zap
!= 0) {
2332 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2333 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2336 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2337 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2345 * Determine whether the activity check is required.
2348 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
2352 uint64_t hostid
= 0;
2353 uint64_t tryconfig_txg
= 0;
2354 uint64_t tryconfig_timestamp
= 0;
2357 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2358 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2359 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2361 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2362 &tryconfig_timestamp
);
2365 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2368 * Disable the MMP activity check - This is used by zdb which
2369 * is intended to be used on potentially active pools.
2371 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
2375 * Skip the activity check when the MMP feature is disabled.
2377 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
2380 * If the tryconfig_* values are nonzero, they are the results of an
2381 * earlier tryimport. If they match the uberblock we just found, then
2382 * the pool has not changed and we return false so we do not test a
2385 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
2386 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
)
2390 * Allow the activity check to be skipped when importing the pool
2391 * on the same host which last imported it. Since the hostid from
2392 * configuration may be stale use the one read from the label.
2394 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
2395 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
2397 if (hostid
== spa_get_hostid())
2401 * Skip the activity test when the pool was cleanly exported.
2403 if (state
!= POOL_STATE_ACTIVE
)
2410 * Perform the import activity check. If the user canceled the import or
2411 * we detected activity then fail.
2414 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2416 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2417 uint64_t txg
= ub
->ub_txg
;
2418 uint64_t timestamp
= ub
->ub_timestamp
;
2419 uint64_t import_delay
= NANOSEC
;
2420 hrtime_t import_expire
;
2421 nvlist_t
*mmp_label
= NULL
;
2422 vdev_t
*rvd
= spa
->spa_root_vdev
;
2427 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
2428 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2432 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2433 * during the earlier tryimport. If the txg recorded there is 0 then
2434 * the pool is known to be active on another host.
2436 * Otherwise, the pool might be in use on another node. Check for
2437 * changes in the uberblocks on disk if necessary.
2439 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2440 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
2441 ZPOOL_CONFIG_LOAD_INFO
);
2443 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
2444 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
2445 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2446 error
= SET_ERROR(EREMOTEIO
);
2452 * Preferentially use the zfs_multihost_interval from the node which
2453 * last imported the pool. This value is stored in an MMP uberblock as.
2455 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2457 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
)
2458 import_delay
= MAX(import_delay
, import_intervals
*
2459 ub
->ub_mmp_delay
* MAX(vdev_count_leaves(spa
), 1));
2461 /* Apply a floor using the local default values. */
2462 import_delay
= MAX(import_delay
, import_intervals
*
2463 MSEC2NSEC(MAX(zfs_multihost_interval
, MMP_MIN_INTERVAL
)));
2465 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu import_intervals=%u "
2466 "leaves=%u", import_delay
, ub
->ub_mmp_delay
, import_intervals
,
2467 vdev_count_leaves(spa
));
2469 /* Add a small random factor in case of simultaneous imports (0-25%) */
2470 import_expire
= gethrtime() + import_delay
+
2471 (import_delay
* spa_get_random(250) / 1000);
2473 while (gethrtime() < import_expire
) {
2474 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2476 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
) {
2477 error
= SET_ERROR(EREMOTEIO
);
2482 nvlist_free(mmp_label
);
2486 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
2488 error
= SET_ERROR(EINTR
);
2496 mutex_destroy(&mtx
);
2500 * If the pool is determined to be active store the status in the
2501 * spa->spa_load_info nvlist. If the remote hostname or hostid are
2502 * available from configuration read from disk store them as well.
2503 * This allows 'zpool import' to generate a more useful message.
2505 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
2506 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2507 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
2509 if (error
== EREMOTEIO
) {
2510 char *hostname
= "<unknown>";
2511 uint64_t hostid
= 0;
2514 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
2515 hostname
= fnvlist_lookup_string(mmp_label
,
2516 ZPOOL_CONFIG_HOSTNAME
);
2517 fnvlist_add_string(spa
->spa_load_info
,
2518 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
2521 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
2522 hostid
= fnvlist_lookup_uint64(mmp_label
,
2523 ZPOOL_CONFIG_HOSTID
);
2524 fnvlist_add_uint64(spa
->spa_load_info
,
2525 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
2529 fnvlist_add_uint64(spa
->spa_load_info
,
2530 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
2531 fnvlist_add_uint64(spa
->spa_load_info
,
2532 ZPOOL_CONFIG_MMP_TXG
, 0);
2534 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
2538 nvlist_free(mmp_label
);
2544 * Load an existing storage pool, using the pool's builtin spa_config as a
2545 * source of configuration information.
2547 __attribute__((always_inline
))
2549 spa_load_impl(spa_t
*spa
, uint64_t pool_guid
, nvlist_t
*config
,
2550 spa_load_state_t state
, spa_import_type_t type
, boolean_t mosconfig
,
2554 nvlist_t
*nvroot
= NULL
;
2557 uberblock_t
*ub
= &spa
->spa_uberblock
;
2558 uint64_t children
, config_cache_txg
= spa
->spa_config_txg
;
2559 int orig_mode
= spa
->spa_mode
;
2562 boolean_t missing_feat_write
= B_FALSE
;
2563 boolean_t activity_check
= B_FALSE
;
2566 * If this is an untrusted config, access the pool in read-only mode.
2567 * This prevents things like resilvering recently removed devices.
2570 spa
->spa_mode
= FREAD
;
2572 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2574 spa
->spa_load_state
= state
;
2576 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvroot
))
2577 return (SET_ERROR(EINVAL
));
2579 parse
= (type
== SPA_IMPORT_EXISTING
?
2580 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2583 * Create "The Godfather" zio to hold all async IOs
2585 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2587 for (int i
= 0; i
< max_ncpus
; i
++) {
2588 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2589 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2590 ZIO_FLAG_GODFATHER
);
2594 * Parse the configuration into a vdev tree. We explicitly set the
2595 * value that will be returned by spa_version() since parsing the
2596 * configuration requires knowing the version number.
2598 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2599 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, parse
);
2600 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2605 ASSERT(spa
->spa_root_vdev
== rvd
);
2606 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2607 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2609 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2610 ASSERT(spa_guid(spa
) == pool_guid
);
2614 * Try to open all vdevs, loading each label in the process.
2616 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2617 error
= vdev_open(rvd
);
2618 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2623 * We need to validate the vdev labels against the configuration that
2624 * we have in hand, which is dependent on the setting of mosconfig. If
2625 * mosconfig is true then we're validating the vdev labels based on
2626 * that config. Otherwise, we're validating against the cached config
2627 * (zpool.cache) that was read when we loaded the zfs module, and then
2628 * later we will recursively call spa_load() and validate against
2631 * If we're assembling a new pool that's been split off from an
2632 * existing pool, the labels haven't yet been updated so we skip
2633 * validation for now.
2635 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2636 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2637 error
= vdev_validate(rvd
, mosconfig
);
2638 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2643 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
2644 return (SET_ERROR(ENXIO
));
2648 * Find the best uberblock.
2650 vdev_uberblock_load(rvd
, ub
, &label
);
2653 * If we weren't able to find a single valid uberblock, return failure.
2655 if (ub
->ub_txg
== 0) {
2657 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2661 * For pools which have the multihost property on determine if the
2662 * pool is truly inactive and can be safely imported. Prevent
2663 * hosts which don't have a hostid set from importing the pool.
2665 activity_check
= spa_activity_check_required(spa
, ub
, label
, config
);
2666 if (activity_check
) {
2667 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
2668 spa_get_hostid() == 0) {
2670 fnvlist_add_uint64(spa
->spa_load_info
,
2671 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
2672 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
2675 error
= spa_activity_check(spa
, ub
, config
);
2681 fnvlist_add_uint64(spa
->spa_load_info
,
2682 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
2683 fnvlist_add_uint64(spa
->spa_load_info
,
2684 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
2688 * If the pool has an unsupported version we can't open it.
2690 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2692 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2695 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2699 * If we weren't able to find what's necessary for reading the
2700 * MOS in the label, return failure.
2702 if (label
== NULL
|| nvlist_lookup_nvlist(label
,
2703 ZPOOL_CONFIG_FEATURES_FOR_READ
, &features
) != 0) {
2705 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2710 * Update our in-core representation with the definitive values
2713 nvlist_free(spa
->spa_label_features
);
2714 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2720 * Look through entries in the label nvlist's features_for_read. If
2721 * there is a feature listed there which we don't understand then we
2722 * cannot open a pool.
2724 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2725 nvlist_t
*unsup_feat
;
2727 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2730 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
2732 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2733 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2734 VERIFY(nvlist_add_string(unsup_feat
,
2735 nvpair_name(nvp
), "") == 0);
2739 if (!nvlist_empty(unsup_feat
)) {
2740 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2741 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2742 nvlist_free(unsup_feat
);
2743 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2747 nvlist_free(unsup_feat
);
2751 * If the vdev guid sum doesn't match the uberblock, we have an
2752 * incomplete configuration. We first check to see if the pool
2753 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2754 * If it is, defer the vdev_guid_sum check till later so we
2755 * can handle missing vdevs.
2757 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
2758 &children
) != 0 && mosconfig
&& type
!= SPA_IMPORT_ASSEMBLE
&&
2759 rvd
->vdev_guid_sum
!= ub
->ub_guid_sum
)
2760 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2762 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2763 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2764 spa_try_repair(spa
, config
);
2765 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2766 nvlist_free(spa
->spa_config_splitting
);
2767 spa
->spa_config_splitting
= NULL
;
2771 * Initialize internal SPA structures.
2773 spa
->spa_state
= POOL_STATE_ACTIVE
;
2774 spa
->spa_ubsync
= spa
->spa_uberblock
;
2775 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2776 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2777 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2778 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2779 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2780 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2782 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2784 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2785 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2787 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
) != 0)
2788 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2790 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2791 boolean_t missing_feat_read
= B_FALSE
;
2792 nvlist_t
*unsup_feat
, *enabled_feat
;
2794 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2795 &spa
->spa_feat_for_read_obj
) != 0) {
2796 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2799 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2800 &spa
->spa_feat_for_write_obj
) != 0) {
2801 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2804 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2805 &spa
->spa_feat_desc_obj
) != 0) {
2806 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2809 enabled_feat
= fnvlist_alloc();
2810 unsup_feat
= fnvlist_alloc();
2812 if (!spa_features_check(spa
, B_FALSE
,
2813 unsup_feat
, enabled_feat
))
2814 missing_feat_read
= B_TRUE
;
2816 if (spa_writeable(spa
) || state
== SPA_LOAD_TRYIMPORT
) {
2817 if (!spa_features_check(spa
, B_TRUE
,
2818 unsup_feat
, enabled_feat
)) {
2819 missing_feat_write
= B_TRUE
;
2823 fnvlist_add_nvlist(spa
->spa_load_info
,
2824 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2826 if (!nvlist_empty(unsup_feat
)) {
2827 fnvlist_add_nvlist(spa
->spa_load_info
,
2828 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2831 fnvlist_free(enabled_feat
);
2832 fnvlist_free(unsup_feat
);
2834 if (!missing_feat_read
) {
2835 fnvlist_add_boolean(spa
->spa_load_info
,
2836 ZPOOL_CONFIG_CAN_RDONLY
);
2840 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2841 * twofold: to determine whether the pool is available for
2842 * import in read-write mode and (if it is not) whether the
2843 * pool is available for import in read-only mode. If the pool
2844 * is available for import in read-write mode, it is displayed
2845 * as available in userland; if it is not available for import
2846 * in read-only mode, it is displayed as unavailable in
2847 * userland. If the pool is available for import in read-only
2848 * mode but not read-write mode, it is displayed as unavailable
2849 * in userland with a special note that the pool is actually
2850 * available for open in read-only mode.
2852 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2853 * missing a feature for write, we must first determine whether
2854 * the pool can be opened read-only before returning to
2855 * userland in order to know whether to display the
2856 * abovementioned note.
2858 if (missing_feat_read
|| (missing_feat_write
&&
2859 spa_writeable(spa
))) {
2860 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2865 * Load refcounts for ZFS features from disk into an in-memory
2866 * cache during SPA initialization.
2868 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
2871 error
= feature_get_refcount_from_disk(spa
,
2872 &spa_feature_table
[i
], &refcount
);
2874 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2875 } else if (error
== ENOTSUP
) {
2876 spa
->spa_feat_refcount_cache
[i
] =
2877 SPA_FEATURE_DISABLED
;
2879 return (spa_vdev_err(rvd
,
2880 VDEV_AUX_CORRUPT_DATA
, EIO
));
2885 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2886 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2887 &spa
->spa_feat_enabled_txg_obj
) != 0)
2888 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2891 spa
->spa_is_initializing
= B_TRUE
;
2892 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2893 spa
->spa_is_initializing
= B_FALSE
;
2895 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2899 nvlist_t
*policy
= NULL
, *nvconfig
;
2901 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
2902 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2904 if (!spa_is_root(spa
) && nvlist_lookup_uint64(nvconfig
,
2905 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2907 unsigned long myhostid
= 0;
2909 VERIFY(nvlist_lookup_string(nvconfig
,
2910 ZPOOL_CONFIG_HOSTNAME
, &hostname
) == 0);
2912 myhostid
= spa_get_hostid();
2913 if (hostid
&& myhostid
&& hostid
!= myhostid
) {
2914 nvlist_free(nvconfig
);
2915 return (SET_ERROR(EBADF
));
2918 if (nvlist_lookup_nvlist(spa
->spa_config
,
2919 ZPOOL_REWIND_POLICY
, &policy
) == 0)
2920 VERIFY(nvlist_add_nvlist(nvconfig
,
2921 ZPOOL_REWIND_POLICY
, policy
) == 0);
2923 spa_config_set(spa
, nvconfig
);
2925 spa_deactivate(spa
);
2926 spa_activate(spa
, orig_mode
);
2928 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, B_TRUE
));
2931 /* Grab the checksum salt from the MOS. */
2932 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2933 DMU_POOL_CHECKSUM_SALT
, 1,
2934 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
2935 spa
->spa_cksum_salt
.zcs_bytes
);
2936 if (error
== ENOENT
) {
2937 /* Generate a new salt for subsequent use */
2938 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
2939 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
2940 } else if (error
!= 0) {
2941 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2944 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
) != 0)
2945 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2946 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2948 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2951 * Load the bit that tells us to use the new accounting function
2952 * (raid-z deflation). If we have an older pool, this will not
2955 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
);
2956 if (error
!= 0 && error
!= ENOENT
)
2957 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2959 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2960 &spa
->spa_creation_version
);
2961 if (error
!= 0 && error
!= ENOENT
)
2962 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2965 * Load the persistent error log. If we have an older pool, this will
2968 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
);
2969 if (error
!= 0 && error
!= ENOENT
)
2970 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2972 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2973 &spa
->spa_errlog_scrub
);
2974 if (error
!= 0 && error
!= ENOENT
)
2975 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2978 * Load the history object. If we have an older pool, this
2979 * will not be present.
2981 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
);
2982 if (error
!= 0 && error
!= ENOENT
)
2983 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2986 * Load the per-vdev ZAP map. If we have an older pool, this will not
2987 * be present; in this case, defer its creation to a later time to
2988 * avoid dirtying the MOS this early / out of sync context. See
2989 * spa_sync_config_object.
2992 /* The sentinel is only available in the MOS config. */
2993 nvlist_t
*mos_config
;
2994 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0)
2995 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2997 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
2998 &spa
->spa_all_vdev_zaps
);
3000 if (error
== ENOENT
) {
3001 VERIFY(!nvlist_exists(mos_config
,
3002 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
3003 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
3004 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3005 } else if (error
!= 0) {
3006 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3007 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
3009 * An older version of ZFS overwrote the sentinel value, so
3010 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3011 * destruction to later; see spa_sync_config_object.
3013 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
3015 * We're assuming that no vdevs have had their ZAPs created
3016 * before this. Better be sure of it.
3018 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3020 nvlist_free(mos_config
);
3023 * If we're assembling the pool from the split-off vdevs of
3024 * an existing pool, we don't want to attach the spares & cache
3029 * Load any hot spares for this pool.
3031 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
);
3032 if (error
!= 0 && error
!= ENOENT
)
3033 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3034 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3035 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3036 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3037 &spa
->spa_spares
.sav_config
) != 0)
3038 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3040 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3041 spa_load_spares(spa
);
3042 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3043 } else if (error
== 0) {
3044 spa
->spa_spares
.sav_sync
= B_TRUE
;
3048 * Load any level 2 ARC devices for this pool.
3050 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3051 &spa
->spa_l2cache
.sav_object
);
3052 if (error
!= 0 && error
!= ENOENT
)
3053 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3054 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3055 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3056 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3057 &spa
->spa_l2cache
.sav_config
) != 0)
3058 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3060 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3061 spa_load_l2cache(spa
);
3062 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3063 } else if (error
== 0) {
3064 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3067 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3069 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
);
3070 if (error
&& error
!= ENOENT
)
3071 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3074 uint64_t autoreplace
= 0;
3076 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3077 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3078 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3079 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3080 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3081 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
3082 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3083 &spa
->spa_dedup_ditto
);
3085 spa
->spa_autoreplace
= (autoreplace
!= 0);
3089 * If the 'multihost' property is set, then never allow a pool to
3090 * be imported when the system hostid is zero. The exception to
3091 * this rule is zdb which is always allowed to access pools.
3093 if (spa_multihost(spa
) && spa_get_hostid() == 0 &&
3094 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
3095 fnvlist_add_uint64(spa
->spa_load_info
,
3096 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3097 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3101 * If the 'autoreplace' property is set, then post a resource notifying
3102 * the ZFS DE that it should not issue any faults for unopenable
3103 * devices. We also iterate over the vdevs, and post a sysevent for any
3104 * unopenable vdevs so that the normal autoreplace handler can take
3107 if (spa
->spa_autoreplace
&& state
!= SPA_LOAD_TRYIMPORT
) {
3108 spa_check_removed(spa
->spa_root_vdev
);
3110 * For the import case, this is done in spa_import(), because
3111 * at this point we're using the spare definitions from
3112 * the MOS config, not necessarily from the userland config.
3114 if (state
!= SPA_LOAD_IMPORT
) {
3115 spa_aux_check_removed(&spa
->spa_spares
);
3116 spa_aux_check_removed(&spa
->spa_l2cache
);
3121 * Load the vdev state for all toplevel vdevs.
3126 * Propagate the leaf DTLs we just loaded all the way up the tree.
3128 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3129 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3130 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3133 * Load the DDTs (dedup tables).
3135 error
= ddt_load(spa
);
3137 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3139 spa_update_dspace(spa
);
3142 * Validate the config, using the MOS config to fill in any
3143 * information which might be missing. If we fail to validate
3144 * the config then declare the pool unfit for use. If we're
3145 * assembling a pool from a split, the log is not transferred
3148 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3151 if (load_nvlist(spa
, spa
->spa_config_object
, &nvconfig
) != 0)
3152 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3154 if (!spa_config_valid(spa
, nvconfig
)) {
3155 nvlist_free(nvconfig
);
3156 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3159 nvlist_free(nvconfig
);
3162 * Now that we've validated the config, check the state of the
3163 * root vdev. If it can't be opened, it indicates one or
3164 * more toplevel vdevs are faulted.
3166 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
)
3167 return (SET_ERROR(ENXIO
));
3169 if (spa_writeable(spa
) && spa_check_logs(spa
)) {
3170 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3171 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
, ENXIO
));
3175 if (missing_feat_write
) {
3176 ASSERT(state
== SPA_LOAD_TRYIMPORT
);
3179 * At this point, we know that we can open the pool in
3180 * read-only mode but not read-write mode. We now have enough
3181 * information and can return to userland.
3183 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
, ENOTSUP
));
3187 * We've successfully opened the pool, verify that we're ready
3188 * to start pushing transactions.
3190 if (state
!= SPA_LOAD_TRYIMPORT
) {
3191 if ((error
= spa_load_verify(spa
)))
3192 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3196 if (spa_writeable(spa
) && (state
== SPA_LOAD_RECOVER
||
3197 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3199 int need_update
= B_FALSE
;
3200 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3202 ASSERT(state
!= SPA_LOAD_TRYIMPORT
);
3205 * Claim log blocks that haven't been committed yet.
3206 * This must all happen in a single txg.
3207 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3208 * invoked from zil_claim_log_block()'s i/o done callback.
3209 * Price of rollback is that we abandon the log.
3211 spa
->spa_claiming
= B_TRUE
;
3213 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3214 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3215 zil_claim
, tx
, DS_FIND_CHILDREN
);
3218 spa
->spa_claiming
= B_FALSE
;
3220 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3221 spa
->spa_sync_on
= B_TRUE
;
3222 txg_sync_start(spa
->spa_dsl_pool
);
3223 mmp_thread_start(spa
);
3226 * Wait for all claims to sync. We sync up to the highest
3227 * claimed log block birth time so that claimed log blocks
3228 * don't appear to be from the future. spa_claim_max_txg
3229 * will have been set for us by either zil_check_log_chain()
3230 * (invoked from spa_check_logs()) or zil_claim() above.
3232 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3235 * If the config cache is stale, or we have uninitialized
3236 * metaslabs (see spa_vdev_add()), then update the config.
3238 * If this is a verbatim import, trust the current
3239 * in-core spa_config and update the disk labels.
3241 if (config_cache_txg
!= spa
->spa_config_txg
||
3242 state
== SPA_LOAD_IMPORT
||
3243 state
== SPA_LOAD_RECOVER
||
3244 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3245 need_update
= B_TRUE
;
3247 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
3248 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3249 need_update
= B_TRUE
;
3252 * Update the config cache asychronously in case we're the
3253 * root pool, in which case the config cache isn't writable yet.
3256 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3259 * Check all DTLs to see if anything needs resilvering.
3261 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3262 vdev_resilver_needed(rvd
, NULL
, NULL
))
3263 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3266 * Log the fact that we booted up (so that we can detect if
3267 * we rebooted in the middle of an operation).
3269 spa_history_log_version(spa
, "open", NULL
);
3272 * Delete any inconsistent datasets.
3274 (void) dmu_objset_find(spa_name(spa
),
3275 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3278 * Clean up any stale temporary dataset userrefs.
3280 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3287 spa_load_retry(spa_t
*spa
, spa_load_state_t state
, int mosconfig
)
3289 int mode
= spa
->spa_mode
;
3292 spa_deactivate(spa
);
3294 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3296 spa_activate(spa
, mode
);
3297 spa_async_suspend(spa
);
3299 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
, mosconfig
));
3303 * If spa_load() fails this function will try loading prior txg's. If
3304 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3305 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3306 * function will not rewind the pool and will return the same error as
3310 spa_load_best(spa_t
*spa
, spa_load_state_t state
, int mosconfig
,
3311 uint64_t max_request
, int rewind_flags
)
3313 nvlist_t
*loadinfo
= NULL
;
3314 nvlist_t
*config
= NULL
;
3315 int load_error
, rewind_error
;
3316 uint64_t safe_rewind_txg
;
3319 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3320 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3321 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3323 spa
->spa_load_max_txg
= max_request
;
3324 if (max_request
!= UINT64_MAX
)
3325 spa
->spa_extreme_rewind
= B_TRUE
;
3328 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
,
3330 if (load_error
== 0)
3333 if (spa
->spa_root_vdev
!= NULL
)
3334 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3336 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3337 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3339 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3340 nvlist_free(config
);
3341 return (load_error
);
3344 if (state
== SPA_LOAD_RECOVER
) {
3345 /* Price of rolling back is discarding txgs, including log */
3346 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3349 * If we aren't rolling back save the load info from our first
3350 * import attempt so that we can restore it after attempting
3353 loadinfo
= spa
->spa_load_info
;
3354 spa
->spa_load_info
= fnvlist_alloc();
3357 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3358 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
3359 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
3360 TXG_INITIAL
: safe_rewind_txg
;
3363 * Continue as long as we're finding errors, we're still within
3364 * the acceptable rewind range, and we're still finding uberblocks
3366 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
3367 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
3368 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
3369 spa
->spa_extreme_rewind
= B_TRUE
;
3370 rewind_error
= spa_load_retry(spa
, state
, mosconfig
);
3373 spa
->spa_extreme_rewind
= B_FALSE
;
3374 spa
->spa_load_max_txg
= UINT64_MAX
;
3376 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
3377 spa_config_set(spa
, config
);
3379 nvlist_free(config
);
3381 if (state
== SPA_LOAD_RECOVER
) {
3382 ASSERT3P(loadinfo
, ==, NULL
);
3383 return (rewind_error
);
3385 /* Store the rewind info as part of the initial load info */
3386 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
3387 spa
->spa_load_info
);
3389 /* Restore the initial load info */
3390 fnvlist_free(spa
->spa_load_info
);
3391 spa
->spa_load_info
= loadinfo
;
3393 return (load_error
);
3400 * The import case is identical to an open except that the configuration is sent
3401 * down from userland, instead of grabbed from the configuration cache. For the
3402 * case of an open, the pool configuration will exist in the
3403 * POOL_STATE_UNINITIALIZED state.
3405 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3406 * the same time open the pool, without having to keep around the spa_t in some
3410 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
3414 spa_load_state_t state
= SPA_LOAD_OPEN
;
3416 int locked
= B_FALSE
;
3417 int firstopen
= B_FALSE
;
3422 * As disgusting as this is, we need to support recursive calls to this
3423 * function because dsl_dir_open() is called during spa_load(), and ends
3424 * up calling spa_open() again. The real fix is to figure out how to
3425 * avoid dsl_dir_open() calling this in the first place.
3427 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
3428 mutex_enter(&spa_namespace_lock
);
3432 if ((spa
= spa_lookup(pool
)) == NULL
) {
3434 mutex_exit(&spa_namespace_lock
);
3435 return (SET_ERROR(ENOENT
));
3438 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3439 zpool_rewind_policy_t policy
;
3443 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3445 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3446 state
= SPA_LOAD_RECOVER
;
3448 spa_activate(spa
, spa_mode_global
);
3450 if (state
!= SPA_LOAD_RECOVER
)
3451 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3453 error
= spa_load_best(spa
, state
, B_FALSE
, policy
.zrp_txg
,
3454 policy
.zrp_request
);
3456 if (error
== EBADF
) {
3458 * If vdev_validate() returns failure (indicated by
3459 * EBADF), it indicates that one of the vdevs indicates
3460 * that the pool has been exported or destroyed. If
3461 * this is the case, the config cache is out of sync and
3462 * we should remove the pool from the namespace.
3465 spa_deactivate(spa
);
3466 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
3469 mutex_exit(&spa_namespace_lock
);
3470 return (SET_ERROR(ENOENT
));
3475 * We can't open the pool, but we still have useful
3476 * information: the state of each vdev after the
3477 * attempted vdev_open(). Return this to the user.
3479 if (config
!= NULL
&& spa
->spa_config
) {
3480 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3482 VERIFY(nvlist_add_nvlist(*config
,
3483 ZPOOL_CONFIG_LOAD_INFO
,
3484 spa
->spa_load_info
) == 0);
3487 spa_deactivate(spa
);
3488 spa
->spa_last_open_failed
= error
;
3490 mutex_exit(&spa_namespace_lock
);
3496 spa_open_ref(spa
, tag
);
3499 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3502 * If we've recovered the pool, pass back any information we
3503 * gathered while doing the load.
3505 if (state
== SPA_LOAD_RECOVER
) {
3506 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3507 spa
->spa_load_info
) == 0);
3511 spa
->spa_last_open_failed
= 0;
3512 spa
->spa_last_ubsync_txg
= 0;
3513 spa
->spa_load_txg
= 0;
3514 mutex_exit(&spa_namespace_lock
);
3518 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
3526 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3529 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3533 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3535 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3539 * Lookup the given spa_t, incrementing the inject count in the process,
3540 * preventing it from being exported or destroyed.
3543 spa_inject_addref(char *name
)
3547 mutex_enter(&spa_namespace_lock
);
3548 if ((spa
= spa_lookup(name
)) == NULL
) {
3549 mutex_exit(&spa_namespace_lock
);
3552 spa
->spa_inject_ref
++;
3553 mutex_exit(&spa_namespace_lock
);
3559 spa_inject_delref(spa_t
*spa
)
3561 mutex_enter(&spa_namespace_lock
);
3562 spa
->spa_inject_ref
--;
3563 mutex_exit(&spa_namespace_lock
);
3567 * Add spares device information to the nvlist.
3570 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3580 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3582 if (spa
->spa_spares
.sav_count
== 0)
3585 VERIFY(nvlist_lookup_nvlist(config
,
3586 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3587 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3588 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3590 VERIFY(nvlist_add_nvlist_array(nvroot
,
3591 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3592 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3593 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3596 * Go through and find any spares which have since been
3597 * repurposed as an active spare. If this is the case, update
3598 * their status appropriately.
3600 for (i
= 0; i
< nspares
; i
++) {
3601 VERIFY(nvlist_lookup_uint64(spares
[i
],
3602 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3603 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3605 VERIFY(nvlist_lookup_uint64_array(
3606 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3607 (uint64_t **)&vs
, &vsc
) == 0);
3608 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3609 vs
->vs_aux
= VDEV_AUX_SPARED
;
3616 * Add l2cache device information to the nvlist, including vdev stats.
3619 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3622 uint_t i
, j
, nl2cache
;
3629 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3631 if (spa
->spa_l2cache
.sav_count
== 0)
3634 VERIFY(nvlist_lookup_nvlist(config
,
3635 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3636 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3637 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3638 if (nl2cache
!= 0) {
3639 VERIFY(nvlist_add_nvlist_array(nvroot
,
3640 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3641 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3642 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3645 * Update level 2 cache device stats.
3648 for (i
= 0; i
< nl2cache
; i
++) {
3649 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3650 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3653 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3655 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3656 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3662 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3663 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3665 vdev_get_stats(vd
, vs
);
3666 vdev_config_generate_stats(vd
, l2cache
[i
]);
3673 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
3678 if (spa
->spa_feat_for_read_obj
!= 0) {
3679 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3680 spa
->spa_feat_for_read_obj
);
3681 zap_cursor_retrieve(&zc
, &za
) == 0;
3682 zap_cursor_advance(&zc
)) {
3683 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3684 za
.za_num_integers
== 1);
3685 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3686 za
.za_first_integer
));
3688 zap_cursor_fini(&zc
);
3691 if (spa
->spa_feat_for_write_obj
!= 0) {
3692 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3693 spa
->spa_feat_for_write_obj
);
3694 zap_cursor_retrieve(&zc
, &za
) == 0;
3695 zap_cursor_advance(&zc
)) {
3696 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3697 za
.za_num_integers
== 1);
3698 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
3699 za
.za_first_integer
));
3701 zap_cursor_fini(&zc
);
3706 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
3710 for (i
= 0; i
< SPA_FEATURES
; i
++) {
3711 zfeature_info_t feature
= spa_feature_table
[i
];
3714 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
3717 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
3722 * Store a list of pool features and their reference counts in the
3725 * The first time this is called on a spa, allocate a new nvlist, fetch
3726 * the pool features and reference counts from disk, then save the list
3727 * in the spa. In subsequent calls on the same spa use the saved nvlist
3728 * and refresh its values from the cached reference counts. This
3729 * ensures we don't block here on I/O on a suspended pool so 'zpool
3730 * clear' can resume the pool.
3733 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3737 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3739 mutex_enter(&spa
->spa_feat_stats_lock
);
3740 features
= spa
->spa_feat_stats
;
3742 if (features
!= NULL
) {
3743 spa_feature_stats_from_cache(spa
, features
);
3745 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
3746 spa
->spa_feat_stats
= features
;
3747 spa_feature_stats_from_disk(spa
, features
);
3750 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3753 mutex_exit(&spa
->spa_feat_stats_lock
);
3757 spa_get_stats(const char *name
, nvlist_t
**config
,
3758 char *altroot
, size_t buflen
)
3764 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
3768 * This still leaves a window of inconsistency where the spares
3769 * or l2cache devices could change and the config would be
3770 * self-inconsistent.
3772 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3774 if (*config
!= NULL
) {
3775 uint64_t loadtimes
[2];
3777 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
3778 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
3779 VERIFY(nvlist_add_uint64_array(*config
,
3780 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
3782 VERIFY(nvlist_add_uint64(*config
,
3783 ZPOOL_CONFIG_ERRCOUNT
,
3784 spa_get_errlog_size(spa
)) == 0);
3786 if (spa_suspended(spa
)) {
3787 VERIFY(nvlist_add_uint64(*config
,
3788 ZPOOL_CONFIG_SUSPENDED
,
3789 spa
->spa_failmode
) == 0);
3790 VERIFY(nvlist_add_uint64(*config
,
3791 ZPOOL_CONFIG_SUSPENDED_REASON
,
3792 spa
->spa_suspended
) == 0);
3795 spa_add_spares(spa
, *config
);
3796 spa_add_l2cache(spa
, *config
);
3797 spa_add_feature_stats(spa
, *config
);
3802 * We want to get the alternate root even for faulted pools, so we cheat
3803 * and call spa_lookup() directly.
3807 mutex_enter(&spa_namespace_lock
);
3808 spa
= spa_lookup(name
);
3810 spa_altroot(spa
, altroot
, buflen
);
3814 mutex_exit(&spa_namespace_lock
);
3816 spa_altroot(spa
, altroot
, buflen
);
3821 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3822 spa_close(spa
, FTAG
);
3829 * Validate that the auxiliary device array is well formed. We must have an
3830 * array of nvlists, each which describes a valid leaf vdev. If this is an
3831 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3832 * specified, as long as they are well-formed.
3835 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
3836 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
3837 vdev_labeltype_t label
)
3844 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3847 * It's acceptable to have no devs specified.
3849 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
3853 return (SET_ERROR(EINVAL
));
3856 * Make sure the pool is formatted with a version that supports this
3859 if (spa_version(spa
) < version
)
3860 return (SET_ERROR(ENOTSUP
));
3863 * Set the pending device list so we correctly handle device in-use
3866 sav
->sav_pending
= dev
;
3867 sav
->sav_npending
= ndev
;
3869 for (i
= 0; i
< ndev
; i
++) {
3870 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
3874 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3876 error
= SET_ERROR(EINVAL
);
3882 if ((error
= vdev_open(vd
)) == 0 &&
3883 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
3884 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
3885 vd
->vdev_guid
) == 0);
3891 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
3898 sav
->sav_pending
= NULL
;
3899 sav
->sav_npending
= 0;
3904 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
3908 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3910 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3911 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
3912 VDEV_LABEL_SPARE
)) != 0) {
3916 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
3917 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
3918 VDEV_LABEL_L2CACHE
));
3922 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
3927 if (sav
->sav_config
!= NULL
) {
3933 * Generate new dev list by concatenating with the
3936 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
3937 &olddevs
, &oldndevs
) == 0);
3939 newdevs
= kmem_alloc(sizeof (void *) *
3940 (ndevs
+ oldndevs
), KM_SLEEP
);
3941 for (i
= 0; i
< oldndevs
; i
++)
3942 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
3944 for (i
= 0; i
< ndevs
; i
++)
3945 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
3948 VERIFY(nvlist_remove(sav
->sav_config
, config
,
3949 DATA_TYPE_NVLIST_ARRAY
) == 0);
3951 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
3952 config
, newdevs
, ndevs
+ oldndevs
) == 0);
3953 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
3954 nvlist_free(newdevs
[i
]);
3955 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
3958 * Generate a new dev list.
3960 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
3962 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
3968 * Stop and drop level 2 ARC devices
3971 spa_l2cache_drop(spa_t
*spa
)
3975 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
3977 for (i
= 0; i
< sav
->sav_count
; i
++) {
3980 vd
= sav
->sav_vdevs
[i
];
3983 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
3984 pool
!= 0ULL && l2arc_vdev_present(vd
))
3985 l2arc_remove_vdev(vd
);
3990 * Verify encryption parameters for spa creation. If we are encrypting, we must
3991 * have the encryption feature flag enabled.
3994 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
3995 boolean_t has_encryption
)
3997 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
3998 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
4000 return (SET_ERROR(ENOTSUP
));
4002 return (dmu_objset_create_crypt_check(NULL
, dcp
));
4009 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
4010 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
4013 char *altroot
= NULL
;
4018 uint64_t txg
= TXG_INITIAL
;
4019 nvlist_t
**spares
, **l2cache
;
4020 uint_t nspares
, nl2cache
;
4021 uint64_t version
, obj
, root_dsobj
= 0;
4022 boolean_t has_features
;
4023 boolean_t has_encryption
;
4029 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
4030 poolname
= (char *)pool
;
4033 * If this pool already exists, return failure.
4035 mutex_enter(&spa_namespace_lock
);
4036 if (spa_lookup(poolname
) != NULL
) {
4037 mutex_exit(&spa_namespace_lock
);
4038 return (SET_ERROR(EEXIST
));
4042 * Allocate a new spa_t structure.
4044 nvl
= fnvlist_alloc();
4045 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
4046 (void) nvlist_lookup_string(props
,
4047 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4048 spa
= spa_add(poolname
, nvl
, altroot
);
4050 spa_activate(spa
, spa_mode_global
);
4052 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4053 spa_deactivate(spa
);
4055 mutex_exit(&spa_namespace_lock
);
4060 * Temporary pool names should never be written to disk.
4062 if (poolname
!= pool
)
4063 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
4065 has_features
= B_FALSE
;
4066 has_encryption
= B_FALSE
;
4067 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
4068 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4069 if (zpool_prop_feature(nvpair_name(elem
))) {
4070 has_features
= B_TRUE
;
4072 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
4073 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
4074 if (feat
== SPA_FEATURE_ENCRYPTION
)
4075 has_encryption
= B_TRUE
;
4079 /* verify encryption params, if they were provided */
4081 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
4083 spa_deactivate(spa
);
4085 mutex_exit(&spa_namespace_lock
);
4090 if (has_features
|| nvlist_lookup_uint64(props
,
4091 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4092 version
= SPA_VERSION
;
4094 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4096 spa
->spa_first_txg
= txg
;
4097 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4098 spa
->spa_uberblock
.ub_version
= version
;
4099 spa
->spa_ubsync
= spa
->spa_uberblock
;
4100 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4103 * Create "The Godfather" zio to hold all async IOs
4105 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4107 for (int i
= 0; i
< max_ncpus
; i
++) {
4108 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4109 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4110 ZIO_FLAG_GODFATHER
);
4114 * Create the root vdev.
4116 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4118 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4120 ASSERT(error
!= 0 || rvd
!= NULL
);
4121 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4123 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4124 error
= SET_ERROR(EINVAL
);
4127 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4128 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4129 VDEV_ALLOC_ADD
)) == 0) {
4130 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
4131 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4132 vdev_expand(rvd
->vdev_child
[c
], txg
);
4136 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4140 spa_deactivate(spa
);
4142 mutex_exit(&spa_namespace_lock
);
4147 * Get the list of spares, if specified.
4149 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4150 &spares
, &nspares
) == 0) {
4151 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4153 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4154 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4155 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4156 spa_load_spares(spa
);
4157 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4158 spa
->spa_spares
.sav_sync
= B_TRUE
;
4162 * Get the list of level 2 cache devices, if specified.
4164 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4165 &l2cache
, &nl2cache
) == 0) {
4166 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4167 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4168 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4169 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4170 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4171 spa_load_l2cache(spa
);
4172 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4173 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4176 spa
->spa_is_initializing
= B_TRUE
;
4177 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
4178 spa
->spa_is_initializing
= B_FALSE
;
4181 * Create DDTs (dedup tables).
4185 spa_update_dspace(spa
);
4187 tx
= dmu_tx_create_assigned(dp
, txg
);
4190 * Create the pool's history object.
4192 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
4193 spa_history_create_obj(spa
, tx
);
4195 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4196 spa_history_log_version(spa
, "create", tx
);
4199 * Create the pool config object.
4201 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4202 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4203 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4205 if (zap_add(spa
->spa_meta_objset
,
4206 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4207 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4208 cmn_err(CE_PANIC
, "failed to add pool config");
4211 if (zap_add(spa
->spa_meta_objset
,
4212 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4213 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4214 cmn_err(CE_PANIC
, "failed to add pool version");
4217 /* Newly created pools with the right version are always deflated. */
4218 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4219 spa
->spa_deflate
= TRUE
;
4220 if (zap_add(spa
->spa_meta_objset
,
4221 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4222 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4223 cmn_err(CE_PANIC
, "failed to add deflate");
4228 * Create the deferred-free bpobj. Turn off compression
4229 * because sync-to-convergence takes longer if the blocksize
4232 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4233 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4234 ZIO_COMPRESS_OFF
, tx
);
4235 if (zap_add(spa
->spa_meta_objset
,
4236 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4237 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4238 cmn_err(CE_PANIC
, "failed to add bpobj");
4240 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4241 spa
->spa_meta_objset
, obj
));
4244 * Generate some random noise for salted checksums to operate on.
4246 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4247 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4250 * Set pool properties.
4252 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4253 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4254 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4255 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4256 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
4258 if (props
!= NULL
) {
4259 spa_configfile_set(spa
, props
, B_FALSE
);
4260 spa_sync_props(props
, tx
);
4266 * If the root dataset is encrypted we will need to create key mappings
4267 * for the zio layer before we start to write any data to disk and hold
4268 * them until after the first txg has been synced. Waiting for the first
4269 * transaction to complete also ensures that our bean counters are
4270 * appropriately updated.
4272 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0) {
4273 root_dsobj
= dsl_dir_phys(dp
->dp_root_dir
)->dd_head_dataset_obj
;
4274 VERIFY0(spa_keystore_create_mapping_impl(spa
, root_dsobj
,
4275 dp
->dp_root_dir
, FTAG
));
4278 spa
->spa_sync_on
= B_TRUE
;
4280 mmp_thread_start(spa
);
4281 txg_wait_synced(dp
, txg
);
4283 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0)
4284 VERIFY0(spa_keystore_remove_mapping(spa
, root_dsobj
, FTAG
));
4286 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4289 * Don't count references from objsets that are already closed
4290 * and are making their way through the eviction process.
4292 spa_evicting_os_wait(spa
);
4293 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4294 spa
->spa_load_state
= SPA_LOAD_NONE
;
4296 mutex_exit(&spa_namespace_lock
);
4302 * Import a non-root pool into the system.
4305 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4308 char *altroot
= NULL
;
4309 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4310 zpool_rewind_policy_t policy
;
4311 uint64_t mode
= spa_mode_global
;
4312 uint64_t readonly
= B_FALSE
;
4315 nvlist_t
**spares
, **l2cache
;
4316 uint_t nspares
, nl2cache
;
4319 * If a pool with this name exists, return failure.
4321 mutex_enter(&spa_namespace_lock
);
4322 if (spa_lookup(pool
) != NULL
) {
4323 mutex_exit(&spa_namespace_lock
);
4324 return (SET_ERROR(EEXIST
));
4328 * Create and initialize the spa structure.
4330 (void) nvlist_lookup_string(props
,
4331 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4332 (void) nvlist_lookup_uint64(props
,
4333 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4336 spa
= spa_add(pool
, config
, altroot
);
4337 spa
->spa_import_flags
= flags
;
4340 * Verbatim import - Take a pool and insert it into the namespace
4341 * as if it had been loaded at boot.
4343 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4345 spa_configfile_set(spa
, props
, B_FALSE
);
4347 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
4348 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4350 mutex_exit(&spa_namespace_lock
);
4354 spa_activate(spa
, mode
);
4357 * Don't start async tasks until we know everything is healthy.
4359 spa_async_suspend(spa
);
4361 zpool_get_rewind_policy(config
, &policy
);
4362 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4363 state
= SPA_LOAD_RECOVER
;
4366 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4367 * because the user-supplied config is actually the one to trust when
4370 if (state
!= SPA_LOAD_RECOVER
)
4371 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4373 error
= spa_load_best(spa
, state
, B_TRUE
, policy
.zrp_txg
,
4374 policy
.zrp_request
);
4377 * Propagate anything learned while loading the pool and pass it
4378 * back to caller (i.e. rewind info, missing devices, etc).
4380 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4381 spa
->spa_load_info
) == 0);
4383 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4385 * Toss any existing sparelist, as it doesn't have any validity
4386 * anymore, and conflicts with spa_has_spare().
4388 if (spa
->spa_spares
.sav_config
) {
4389 nvlist_free(spa
->spa_spares
.sav_config
);
4390 spa
->spa_spares
.sav_config
= NULL
;
4391 spa_load_spares(spa
);
4393 if (spa
->spa_l2cache
.sav_config
) {
4394 nvlist_free(spa
->spa_l2cache
.sav_config
);
4395 spa
->spa_l2cache
.sav_config
= NULL
;
4396 spa_load_l2cache(spa
);
4399 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4401 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4404 spa_configfile_set(spa
, props
, B_FALSE
);
4406 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4407 (error
= spa_prop_set(spa
, props
)))) {
4409 spa_deactivate(spa
);
4411 mutex_exit(&spa_namespace_lock
);
4415 spa_async_resume(spa
);
4418 * Override any spares and level 2 cache devices as specified by
4419 * the user, as these may have correct device names/devids, etc.
4421 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4422 &spares
, &nspares
) == 0) {
4423 if (spa
->spa_spares
.sav_config
)
4424 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4425 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4427 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4428 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4429 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4430 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4431 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4432 spa_load_spares(spa
);
4433 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4434 spa
->spa_spares
.sav_sync
= B_TRUE
;
4436 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4437 &l2cache
, &nl2cache
) == 0) {
4438 if (spa
->spa_l2cache
.sav_config
)
4439 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4440 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4442 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4443 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4444 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4445 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4446 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4447 spa_load_l2cache(spa
);
4448 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4449 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4453 * Check for any removed devices.
4455 if (spa
->spa_autoreplace
) {
4456 spa_aux_check_removed(&spa
->spa_spares
);
4457 spa_aux_check_removed(&spa
->spa_l2cache
);
4460 if (spa_writeable(spa
)) {
4462 * Update the config cache to include the newly-imported pool.
4464 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4468 * It's possible that the pool was expanded while it was exported.
4469 * We kick off an async task to handle this for us.
4471 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4473 spa_history_log_version(spa
, "import", NULL
);
4475 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4477 zvol_create_minors(spa
, pool
, B_TRUE
);
4479 mutex_exit(&spa_namespace_lock
);
4485 spa_tryimport(nvlist_t
*tryconfig
)
4487 nvlist_t
*config
= NULL
;
4493 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4496 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4500 * Create and initialize the spa structure.
4502 mutex_enter(&spa_namespace_lock
);
4503 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4504 spa_activate(spa
, FREAD
);
4507 * Pass off the heavy lifting to spa_load().
4508 * Pass TRUE for mosconfig because the user-supplied config
4509 * is actually the one to trust when doing an import.
4511 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
, B_TRUE
);
4514 * If 'tryconfig' was at least parsable, return the current config.
4516 if (spa
->spa_root_vdev
!= NULL
) {
4517 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4518 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4520 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4522 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4523 spa
->spa_uberblock
.ub_timestamp
) == 0);
4524 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4525 spa
->spa_load_info
) == 0);
4526 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
4527 spa
->spa_errata
) == 0);
4530 * If the bootfs property exists on this pool then we
4531 * copy it out so that external consumers can tell which
4532 * pools are bootable.
4534 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4535 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4538 * We have to play games with the name since the
4539 * pool was opened as TRYIMPORT_NAME.
4541 if (dsl_dsobj_to_dsname(spa_name(spa
),
4542 spa
->spa_bootfs
, tmpname
) == 0) {
4546 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4548 cp
= strchr(tmpname
, '/');
4550 (void) strlcpy(dsname
, tmpname
,
4553 (void) snprintf(dsname
, MAXPATHLEN
,
4554 "%s/%s", poolname
, ++cp
);
4556 VERIFY(nvlist_add_string(config
,
4557 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4558 kmem_free(dsname
, MAXPATHLEN
);
4560 kmem_free(tmpname
, MAXPATHLEN
);
4564 * Add the list of hot spares and level 2 cache devices.
4566 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4567 spa_add_spares(spa
, config
);
4568 spa_add_l2cache(spa
, config
);
4569 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4573 spa_deactivate(spa
);
4575 mutex_exit(&spa_namespace_lock
);
4581 * Pool export/destroy
4583 * The act of destroying or exporting a pool is very simple. We make sure there
4584 * is no more pending I/O and any references to the pool are gone. Then, we
4585 * update the pool state and sync all the labels to disk, removing the
4586 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4587 * we don't sync the labels or remove the configuration cache.
4590 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
4591 boolean_t force
, boolean_t hardforce
)
4598 if (!(spa_mode_global
& FWRITE
))
4599 return (SET_ERROR(EROFS
));
4601 mutex_enter(&spa_namespace_lock
);
4602 if ((spa
= spa_lookup(pool
)) == NULL
) {
4603 mutex_exit(&spa_namespace_lock
);
4604 return (SET_ERROR(ENOENT
));
4608 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4609 * reacquire the namespace lock, and see if we can export.
4611 spa_open_ref(spa
, FTAG
);
4612 mutex_exit(&spa_namespace_lock
);
4613 spa_async_suspend(spa
);
4614 if (spa
->spa_zvol_taskq
) {
4615 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
4616 taskq_wait(spa
->spa_zvol_taskq
);
4618 mutex_enter(&spa_namespace_lock
);
4619 spa_close(spa
, FTAG
);
4621 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
4624 * The pool will be in core if it's openable, in which case we can
4625 * modify its state. Objsets may be open only because they're dirty,
4626 * so we have to force it to sync before checking spa_refcnt.
4628 if (spa
->spa_sync_on
) {
4629 txg_wait_synced(spa
->spa_dsl_pool
, 0);
4630 spa_evicting_os_wait(spa
);
4634 * A pool cannot be exported or destroyed if there are active
4635 * references. If we are resetting a pool, allow references by
4636 * fault injection handlers.
4638 if (!spa_refcount_zero(spa
) ||
4639 (spa
->spa_inject_ref
!= 0 &&
4640 new_state
!= POOL_STATE_UNINITIALIZED
)) {
4641 spa_async_resume(spa
);
4642 mutex_exit(&spa_namespace_lock
);
4643 return (SET_ERROR(EBUSY
));
4646 if (spa
->spa_sync_on
) {
4648 * A pool cannot be exported if it has an active shared spare.
4649 * This is to prevent other pools stealing the active spare
4650 * from an exported pool. At user's own will, such pool can
4651 * be forcedly exported.
4653 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
4654 spa_has_active_shared_spare(spa
)) {
4655 spa_async_resume(spa
);
4656 mutex_exit(&spa_namespace_lock
);
4657 return (SET_ERROR(EXDEV
));
4661 * We want this to be reflected on every label,
4662 * so mark them all dirty. spa_unload() will do the
4663 * final sync that pushes these changes out.
4665 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
4666 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4667 spa
->spa_state
= new_state
;
4668 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
4670 vdev_config_dirty(spa
->spa_root_vdev
);
4671 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4676 if (new_state
== POOL_STATE_DESTROYED
)
4677 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
4678 else if (new_state
== POOL_STATE_EXPORTED
)
4679 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
4681 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
4683 spa_deactivate(spa
);
4686 if (oldconfig
&& spa
->spa_config
)
4687 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
4689 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
4691 spa_config_sync(spa
, B_TRUE
, B_TRUE
);
4694 mutex_exit(&spa_namespace_lock
);
4700 * Destroy a storage pool.
4703 spa_destroy(char *pool
)
4705 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
4710 * Export a storage pool.
4713 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
4714 boolean_t hardforce
)
4716 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
4721 * Similar to spa_export(), this unloads the spa_t without actually removing it
4722 * from the namespace in any way.
4725 spa_reset(char *pool
)
4727 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
4732 * ==========================================================================
4733 * Device manipulation
4734 * ==========================================================================
4738 * Add a device to a storage pool.
4741 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
4745 vdev_t
*rvd
= spa
->spa_root_vdev
;
4747 nvlist_t
**spares
, **l2cache
;
4748 uint_t nspares
, nl2cache
;
4750 ASSERT(spa_writeable(spa
));
4752 txg
= spa_vdev_enter(spa
);
4754 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
4755 VDEV_ALLOC_ADD
)) != 0)
4756 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
4758 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
4760 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
4764 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
4768 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
4769 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
4771 if (vd
->vdev_children
!= 0 &&
4772 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
4773 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4776 * We must validate the spares and l2cache devices after checking the
4777 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4779 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
4780 return (spa_vdev_exit(spa
, vd
, txg
, error
));
4783 * Transfer each new top-level vdev from vd to rvd.
4785 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4788 * Set the vdev id to the first hole, if one exists.
4790 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
4791 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
4792 vdev_free(rvd
->vdev_child
[id
]);
4796 tvd
= vd
->vdev_child
[c
];
4797 vdev_remove_child(vd
, tvd
);
4799 vdev_add_child(rvd
, tvd
);
4800 vdev_config_dirty(tvd
);
4804 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
4805 ZPOOL_CONFIG_SPARES
);
4806 spa_load_spares(spa
);
4807 spa
->spa_spares
.sav_sync
= B_TRUE
;
4810 if (nl2cache
!= 0) {
4811 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
4812 ZPOOL_CONFIG_L2CACHE
);
4813 spa_load_l2cache(spa
);
4814 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4818 * We have to be careful when adding new vdevs to an existing pool.
4819 * If other threads start allocating from these vdevs before we
4820 * sync the config cache, and we lose power, then upon reboot we may
4821 * fail to open the pool because there are DVAs that the config cache
4822 * can't translate. Therefore, we first add the vdevs without
4823 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4824 * and then let spa_config_update() initialize the new metaslabs.
4826 * spa_load() checks for added-but-not-initialized vdevs, so that
4827 * if we lose power at any point in this sequence, the remaining
4828 * steps will be completed the next time we load the pool.
4830 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
4832 mutex_enter(&spa_namespace_lock
);
4833 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4834 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
4835 mutex_exit(&spa_namespace_lock
);
4841 * Attach a device to a mirror. The arguments are the path to any device
4842 * in the mirror, and the nvroot for the new device. If the path specifies
4843 * a device that is not mirrored, we automatically insert the mirror vdev.
4845 * If 'replacing' is specified, the new device is intended to replace the
4846 * existing device; in this case the two devices are made into their own
4847 * mirror using the 'replacing' vdev, which is functionally identical to
4848 * the mirror vdev (it actually reuses all the same ops) but has a few
4849 * extra rules: you can't attach to it after it's been created, and upon
4850 * completion of resilvering, the first disk (the one being replaced)
4851 * is automatically detached.
4854 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
4856 uint64_t txg
, dtl_max_txg
;
4857 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
4858 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
4860 char *oldvdpath
, *newvdpath
;
4864 ASSERT(spa_writeable(spa
));
4866 txg
= spa_vdev_enter(spa
);
4868 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
4871 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
4873 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
4874 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
4876 pvd
= oldvd
->vdev_parent
;
4878 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
4879 VDEV_ALLOC_ATTACH
)) != 0)
4880 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
4882 if (newrootvd
->vdev_children
!= 1)
4883 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4885 newvd
= newrootvd
->vdev_child
[0];
4887 if (!newvd
->vdev_ops
->vdev_op_leaf
)
4888 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
4890 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
4891 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
4894 * Spares can't replace logs
4896 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
4897 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4901 * For attach, the only allowable parent is a mirror or the root
4904 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
4905 pvd
->vdev_ops
!= &vdev_root_ops
)
4906 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4908 pvops
= &vdev_mirror_ops
;
4911 * Active hot spares can only be replaced by inactive hot
4914 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4915 oldvd
->vdev_isspare
&&
4916 !spa_has_spare(spa
, newvd
->vdev_guid
))
4917 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4920 * If the source is a hot spare, and the parent isn't already a
4921 * spare, then we want to create a new hot spare. Otherwise, we
4922 * want to create a replacing vdev. The user is not allowed to
4923 * attach to a spared vdev child unless the 'isspare' state is
4924 * the same (spare replaces spare, non-spare replaces
4927 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
4928 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
4929 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4930 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
4931 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
4932 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
4935 if (newvd
->vdev_isspare
)
4936 pvops
= &vdev_spare_ops
;
4938 pvops
= &vdev_replacing_ops
;
4942 * Make sure the new device is big enough.
4944 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
4945 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
4948 * The new device cannot have a higher alignment requirement
4949 * than the top-level vdev.
4951 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
4952 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
4955 * If this is an in-place replacement, update oldvd's path and devid
4956 * to make it distinguishable from newvd, and unopenable from now on.
4958 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
4959 spa_strfree(oldvd
->vdev_path
);
4960 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
4962 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
4963 newvd
->vdev_path
, "old");
4964 if (oldvd
->vdev_devid
!= NULL
) {
4965 spa_strfree(oldvd
->vdev_devid
);
4966 oldvd
->vdev_devid
= NULL
;
4970 /* mark the device being resilvered */
4971 newvd
->vdev_resilver_txg
= txg
;
4974 * If the parent is not a mirror, or if we're replacing, insert the new
4975 * mirror/replacing/spare vdev above oldvd.
4977 if (pvd
->vdev_ops
!= pvops
)
4978 pvd
= vdev_add_parent(oldvd
, pvops
);
4980 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
4981 ASSERT(pvd
->vdev_ops
== pvops
);
4982 ASSERT(oldvd
->vdev_parent
== pvd
);
4985 * Extract the new device from its root and add it to pvd.
4987 vdev_remove_child(newrootvd
, newvd
);
4988 newvd
->vdev_id
= pvd
->vdev_children
;
4989 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
4990 vdev_add_child(pvd
, newvd
);
4993 * Reevaluate the parent vdev state.
4995 vdev_propagate_state(pvd
);
4997 tvd
= newvd
->vdev_top
;
4998 ASSERT(pvd
->vdev_top
== tvd
);
4999 ASSERT(tvd
->vdev_parent
== rvd
);
5001 vdev_config_dirty(tvd
);
5004 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5005 * for any dmu_sync-ed blocks. It will propagate upward when
5006 * spa_vdev_exit() calls vdev_dtl_reassess().
5008 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
5010 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
5011 dtl_max_txg
- TXG_INITIAL
);
5013 if (newvd
->vdev_isspare
) {
5014 spa_spare_activate(newvd
);
5015 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
5018 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
5019 newvdpath
= spa_strdup(newvd
->vdev_path
);
5020 newvd_isspare
= newvd
->vdev_isspare
;
5023 * Mark newvd's DTL dirty in this txg.
5025 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
5028 * Schedule the resilver to restart in the future. We do this to
5029 * ensure that dmu_sync-ed blocks have been stitched into the
5030 * respective datasets.
5032 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
5034 if (spa
->spa_bootfs
)
5035 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
5037 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
5042 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
5044 spa_history_log_internal(spa
, "vdev attach", NULL
,
5045 "%s vdev=%s %s vdev=%s",
5046 replacing
&& newvd_isspare
? "spare in" :
5047 replacing
? "replace" : "attach", newvdpath
,
5048 replacing
? "for" : "to", oldvdpath
);
5050 spa_strfree(oldvdpath
);
5051 spa_strfree(newvdpath
);
5057 * Detach a device from a mirror or replacing vdev.
5059 * If 'replace_done' is specified, only detach if the parent
5060 * is a replacing vdev.
5063 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5067 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5068 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5069 boolean_t unspare
= B_FALSE
;
5070 uint64_t unspare_guid
= 0;
5073 ASSERT(spa_writeable(spa
));
5075 txg
= spa_vdev_enter(spa
);
5077 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5080 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5082 if (!vd
->vdev_ops
->vdev_op_leaf
)
5083 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5085 pvd
= vd
->vdev_parent
;
5088 * If the parent/child relationship is not as expected, don't do it.
5089 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5090 * vdev that's replacing B with C. The user's intent in replacing
5091 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5092 * the replace by detaching C, the expected behavior is to end up
5093 * M(A,B). But suppose that right after deciding to detach C,
5094 * the replacement of B completes. We would have M(A,C), and then
5095 * ask to detach C, which would leave us with just A -- not what
5096 * the user wanted. To prevent this, we make sure that the
5097 * parent/child relationship hasn't changed -- in this example,
5098 * that C's parent is still the replacing vdev R.
5100 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5101 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5104 * Only 'replacing' or 'spare' vdevs can be replaced.
5106 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5107 pvd
->vdev_ops
!= &vdev_spare_ops
)
5108 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5110 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5111 spa_version(spa
) >= SPA_VERSION_SPARES
);
5114 * Only mirror, replacing, and spare vdevs support detach.
5116 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5117 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5118 pvd
->vdev_ops
!= &vdev_spare_ops
)
5119 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5122 * If this device has the only valid copy of some data,
5123 * we cannot safely detach it.
5125 if (vdev_dtl_required(vd
))
5126 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5128 ASSERT(pvd
->vdev_children
>= 2);
5131 * If we are detaching the second disk from a replacing vdev, then
5132 * check to see if we changed the original vdev's path to have "/old"
5133 * at the end in spa_vdev_attach(). If so, undo that change now.
5135 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5136 vd
->vdev_path
!= NULL
) {
5137 size_t len
= strlen(vd
->vdev_path
);
5139 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
5140 cvd
= pvd
->vdev_child
[c
];
5142 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5145 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5146 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5147 spa_strfree(cvd
->vdev_path
);
5148 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5155 * If we are detaching the original disk from a spare, then it implies
5156 * that the spare should become a real disk, and be removed from the
5157 * active spare list for the pool.
5159 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5161 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5165 * Erase the disk labels so the disk can be used for other things.
5166 * This must be done after all other error cases are handled,
5167 * but before we disembowel vd (so we can still do I/O to it).
5168 * But if we can't do it, don't treat the error as fatal --
5169 * it may be that the unwritability of the disk is the reason
5170 * it's being detached!
5172 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5175 * Remove vd from its parent and compact the parent's children.
5177 vdev_remove_child(pvd
, vd
);
5178 vdev_compact_children(pvd
);
5181 * Remember one of the remaining children so we can get tvd below.
5183 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5186 * If we need to remove the remaining child from the list of hot spares,
5187 * do it now, marking the vdev as no longer a spare in the process.
5188 * We must do this before vdev_remove_parent(), because that can
5189 * change the GUID if it creates a new toplevel GUID. For a similar
5190 * reason, we must remove the spare now, in the same txg as the detach;
5191 * otherwise someone could attach a new sibling, change the GUID, and
5192 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5195 ASSERT(cvd
->vdev_isspare
);
5196 spa_spare_remove(cvd
);
5197 unspare_guid
= cvd
->vdev_guid
;
5198 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5199 cvd
->vdev_unspare
= B_TRUE
;
5203 * If the parent mirror/replacing vdev only has one child,
5204 * the parent is no longer needed. Remove it from the tree.
5206 if (pvd
->vdev_children
== 1) {
5207 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5208 cvd
->vdev_unspare
= B_FALSE
;
5209 vdev_remove_parent(cvd
);
5214 * We don't set tvd until now because the parent we just removed
5215 * may have been the previous top-level vdev.
5217 tvd
= cvd
->vdev_top
;
5218 ASSERT(tvd
->vdev_parent
== rvd
);
5221 * Reevaluate the parent vdev state.
5223 vdev_propagate_state(cvd
);
5226 * If the 'autoexpand' property is set on the pool then automatically
5227 * try to expand the size of the pool. For example if the device we
5228 * just detached was smaller than the others, it may be possible to
5229 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5230 * first so that we can obtain the updated sizes of the leaf vdevs.
5232 if (spa
->spa_autoexpand
) {
5234 vdev_expand(tvd
, txg
);
5237 vdev_config_dirty(tvd
);
5240 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5241 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5242 * But first make sure we're not on any *other* txg's DTL list, to
5243 * prevent vd from being accessed after it's freed.
5245 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
5246 for (int t
= 0; t
< TXG_SIZE
; t
++)
5247 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5248 vd
->vdev_detached
= B_TRUE
;
5249 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5251 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5253 /* hang on to the spa before we release the lock */
5254 spa_open_ref(spa
, FTAG
);
5256 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5258 spa_history_log_internal(spa
, "detach", NULL
,
5260 spa_strfree(vdpath
);
5263 * If this was the removal of the original device in a hot spare vdev,
5264 * then we want to go through and remove the device from the hot spare
5265 * list of every other pool.
5268 spa_t
*altspa
= NULL
;
5270 mutex_enter(&spa_namespace_lock
);
5271 while ((altspa
= spa_next(altspa
)) != NULL
) {
5272 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5276 spa_open_ref(altspa
, FTAG
);
5277 mutex_exit(&spa_namespace_lock
);
5278 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5279 mutex_enter(&spa_namespace_lock
);
5280 spa_close(altspa
, FTAG
);
5282 mutex_exit(&spa_namespace_lock
);
5284 /* search the rest of the vdevs for spares to remove */
5285 spa_vdev_resilver_done(spa
);
5288 /* all done with the spa; OK to release */
5289 mutex_enter(&spa_namespace_lock
);
5290 spa_close(spa
, FTAG
);
5291 mutex_exit(&spa_namespace_lock
);
5297 * Split a set of devices from their mirrors, and create a new pool from them.
5300 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5301 nvlist_t
*props
, boolean_t exp
)
5304 uint64_t txg
, *glist
;
5306 uint_t c
, children
, lastlog
;
5307 nvlist_t
**child
, *nvl
, *tmp
;
5309 char *altroot
= NULL
;
5310 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5311 boolean_t activate_slog
;
5313 ASSERT(spa_writeable(spa
));
5315 txg
= spa_vdev_enter(spa
);
5317 /* clear the log and flush everything up to now */
5318 activate_slog
= spa_passivate_log(spa
);
5319 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5320 error
= spa_offline_log(spa
);
5321 txg
= spa_vdev_config_enter(spa
);
5324 spa_activate_log(spa
);
5327 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5329 /* check new spa name before going any further */
5330 if (spa_lookup(newname
) != NULL
)
5331 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5334 * scan through all the children to ensure they're all mirrors
5336 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5337 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5339 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5341 /* first, check to ensure we've got the right child count */
5342 rvd
= spa
->spa_root_vdev
;
5344 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5345 vdev_t
*vd
= rvd
->vdev_child
[c
];
5347 /* don't count the holes & logs as children */
5348 if (vd
->vdev_islog
|| vd
->vdev_ishole
) {
5356 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5357 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5359 /* next, ensure no spare or cache devices are part of the split */
5360 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5361 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5362 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5364 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5365 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5367 /* then, loop over each vdev and validate it */
5368 for (c
= 0; c
< children
; c
++) {
5369 uint64_t is_hole
= 0;
5371 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5375 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5376 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5379 error
= SET_ERROR(EINVAL
);
5384 /* which disk is going to be split? */
5385 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5387 error
= SET_ERROR(EINVAL
);
5391 /* look it up in the spa */
5392 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5393 if (vml
[c
] == NULL
) {
5394 error
= SET_ERROR(ENODEV
);
5398 /* make sure there's nothing stopping the split */
5399 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5400 vml
[c
]->vdev_islog
||
5401 vml
[c
]->vdev_ishole
||
5402 vml
[c
]->vdev_isspare
||
5403 vml
[c
]->vdev_isl2cache
||
5404 !vdev_writeable(vml
[c
]) ||
5405 vml
[c
]->vdev_children
!= 0 ||
5406 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5407 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5408 error
= SET_ERROR(EINVAL
);
5412 if (vdev_dtl_required(vml
[c
])) {
5413 error
= SET_ERROR(EBUSY
);
5417 /* we need certain info from the top level */
5418 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5419 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5420 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5421 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5422 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5423 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5424 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5425 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5427 /* transfer per-vdev ZAPs */
5428 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
5429 VERIFY0(nvlist_add_uint64(child
[c
],
5430 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
5432 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
5433 VERIFY0(nvlist_add_uint64(child
[c
],
5434 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
5435 vml
[c
]->vdev_parent
->vdev_top_zap
));
5439 kmem_free(vml
, children
* sizeof (vdev_t
*));
5440 kmem_free(glist
, children
* sizeof (uint64_t));
5441 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5444 /* stop writers from using the disks */
5445 for (c
= 0; c
< children
; c
++) {
5447 vml
[c
]->vdev_offline
= B_TRUE
;
5449 vdev_reopen(spa
->spa_root_vdev
);
5452 * Temporarily record the splitting vdevs in the spa config. This
5453 * will disappear once the config is regenerated.
5455 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5456 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5457 glist
, children
) == 0);
5458 kmem_free(glist
, children
* sizeof (uint64_t));
5460 mutex_enter(&spa
->spa_props_lock
);
5461 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5463 mutex_exit(&spa
->spa_props_lock
);
5464 spa
->spa_config_splitting
= nvl
;
5465 vdev_config_dirty(spa
->spa_root_vdev
);
5467 /* configure and create the new pool */
5468 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5469 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5470 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5471 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5472 spa_version(spa
)) == 0);
5473 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5474 spa
->spa_config_txg
) == 0);
5475 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5476 spa_generate_guid(NULL
)) == 0);
5477 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
5478 (void) nvlist_lookup_string(props
,
5479 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5481 /* add the new pool to the namespace */
5482 newspa
= spa_add(newname
, config
, altroot
);
5483 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5484 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5485 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5487 /* release the spa config lock, retaining the namespace lock */
5488 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5490 if (zio_injection_enabled
)
5491 zio_handle_panic_injection(spa
, FTAG
, 1);
5493 spa_activate(newspa
, spa_mode_global
);
5494 spa_async_suspend(newspa
);
5496 /* create the new pool from the disks of the original pool */
5497 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
, B_TRUE
);
5501 /* if that worked, generate a real config for the new pool */
5502 if (newspa
->spa_root_vdev
!= NULL
) {
5503 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5504 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5505 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5506 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5507 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5512 if (props
!= NULL
) {
5513 spa_configfile_set(newspa
, props
, B_FALSE
);
5514 error
= spa_prop_set(newspa
, props
);
5519 /* flush everything */
5520 txg
= spa_vdev_config_enter(newspa
);
5521 vdev_config_dirty(newspa
->spa_root_vdev
);
5522 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5524 if (zio_injection_enabled
)
5525 zio_handle_panic_injection(spa
, FTAG
, 2);
5527 spa_async_resume(newspa
);
5529 /* finally, update the original pool's config */
5530 txg
= spa_vdev_config_enter(spa
);
5531 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5532 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5535 for (c
= 0; c
< children
; c
++) {
5536 if (vml
[c
] != NULL
) {
5539 spa_history_log_internal(spa
, "detach", tx
,
5540 "vdev=%s", vml
[c
]->vdev_path
);
5545 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5546 vdev_config_dirty(spa
->spa_root_vdev
);
5547 spa
->spa_config_splitting
= NULL
;
5551 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5553 if (zio_injection_enabled
)
5554 zio_handle_panic_injection(spa
, FTAG
, 3);
5556 /* split is complete; log a history record */
5557 spa_history_log_internal(newspa
, "split", NULL
,
5558 "from pool %s", spa_name(spa
));
5560 kmem_free(vml
, children
* sizeof (vdev_t
*));
5562 /* if we're not going to mount the filesystems in userland, export */
5564 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
5571 spa_deactivate(newspa
);
5574 txg
= spa_vdev_config_enter(spa
);
5576 /* re-online all offlined disks */
5577 for (c
= 0; c
< children
; c
++) {
5579 vml
[c
]->vdev_offline
= B_FALSE
;
5581 vdev_reopen(spa
->spa_root_vdev
);
5583 nvlist_free(spa
->spa_config_splitting
);
5584 spa
->spa_config_splitting
= NULL
;
5585 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
5587 kmem_free(vml
, children
* sizeof (vdev_t
*));
5592 spa_nvlist_lookup_by_guid(nvlist_t
**nvpp
, int count
, uint64_t target_guid
)
5594 for (int i
= 0; i
< count
; i
++) {
5597 VERIFY(nvlist_lookup_uint64(nvpp
[i
], ZPOOL_CONFIG_GUID
,
5600 if (guid
== target_guid
)
5608 spa_vdev_remove_aux(nvlist_t
*config
, char *name
, nvlist_t
**dev
, int count
,
5609 nvlist_t
*dev_to_remove
)
5611 nvlist_t
**newdev
= NULL
;
5614 newdev
= kmem_alloc((count
- 1) * sizeof (void *), KM_SLEEP
);
5616 for (int i
= 0, j
= 0; i
< count
; i
++) {
5617 if (dev
[i
] == dev_to_remove
)
5619 VERIFY(nvlist_dup(dev
[i
], &newdev
[j
++], KM_SLEEP
) == 0);
5622 VERIFY(nvlist_remove(config
, name
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5623 VERIFY(nvlist_add_nvlist_array(config
, name
, newdev
, count
- 1) == 0);
5625 for (int i
= 0; i
< count
- 1; i
++)
5626 nvlist_free(newdev
[i
]);
5629 kmem_free(newdev
, (count
- 1) * sizeof (void *));
5633 * Evacuate the device.
5636 spa_vdev_remove_evacuate(spa_t
*spa
, vdev_t
*vd
)
5641 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5642 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
5643 ASSERT(vd
== vd
->vdev_top
);
5646 * Evacuate the device. We don't hold the config lock as writer
5647 * since we need to do I/O but we do keep the
5648 * spa_namespace_lock held. Once this completes the device
5649 * should no longer have any blocks allocated on it.
5651 if (vd
->vdev_islog
) {
5652 if (vd
->vdev_stat
.vs_alloc
!= 0)
5653 error
= spa_offline_log(spa
);
5655 error
= SET_ERROR(ENOTSUP
);
5662 * The evacuation succeeded. Remove any remaining MOS metadata
5663 * associated with this vdev, and wait for these changes to sync.
5665 ASSERT0(vd
->vdev_stat
.vs_alloc
);
5666 txg
= spa_vdev_config_enter(spa
);
5667 vd
->vdev_removing
= B_TRUE
;
5668 vdev_dirty_leaves(vd
, VDD_DTL
, txg
);
5669 vdev_config_dirty(vd
);
5670 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5676 * Complete the removal by cleaning up the namespace.
5679 spa_vdev_remove_from_namespace(spa_t
*spa
, vdev_t
*vd
)
5681 vdev_t
*rvd
= spa
->spa_root_vdev
;
5682 uint64_t id
= vd
->vdev_id
;
5683 boolean_t last_vdev
= (id
== (rvd
->vdev_children
- 1));
5685 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
5686 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5687 ASSERT(vd
== vd
->vdev_top
);
5690 * Only remove any devices which are empty.
5692 if (vd
->vdev_stat
.vs_alloc
!= 0)
5695 (void) vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5697 if (list_link_active(&vd
->vdev_state_dirty_node
))
5698 vdev_state_clean(vd
);
5699 if (list_link_active(&vd
->vdev_config_dirty_node
))
5700 vdev_config_clean(vd
);
5705 vdev_compact_children(rvd
);
5707 vd
= vdev_alloc_common(spa
, id
, 0, &vdev_hole_ops
);
5708 vdev_add_child(rvd
, vd
);
5710 vdev_config_dirty(rvd
);
5713 * Reassess the health of our root vdev.
5719 * Remove a device from the pool -
5721 * Removing a device from the vdev namespace requires several steps
5722 * and can take a significant amount of time. As a result we use
5723 * the spa_vdev_config_[enter/exit] functions which allow us to
5724 * grab and release the spa_config_lock while still holding the namespace
5725 * lock. During each step the configuration is synced out.
5727 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5731 spa_vdev_remove(spa_t
*spa
, uint64_t guid
, boolean_t unspare
)
5734 sysevent_t
*ev
= NULL
;
5735 metaslab_group_t
*mg
;
5736 nvlist_t
**spares
, **l2cache
, *nv
;
5738 uint_t nspares
, nl2cache
;
5740 boolean_t locked
= MUTEX_HELD(&spa_namespace_lock
);
5742 ASSERT(spa_writeable(spa
));
5745 txg
= spa_vdev_enter(spa
);
5747 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5749 if (spa
->spa_spares
.sav_vdevs
!= NULL
&&
5750 nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
5751 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0 &&
5752 (nv
= spa_nvlist_lookup_by_guid(spares
, nspares
, guid
)) != NULL
) {
5754 * Only remove the hot spare if it's not currently in use
5757 if (vd
== NULL
|| unspare
) {
5759 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5760 ev
= spa_event_create(spa
, vd
, NULL
,
5761 ESC_ZFS_VDEV_REMOVE_AUX
);
5762 spa_vdev_remove_aux(spa
->spa_spares
.sav_config
,
5763 ZPOOL_CONFIG_SPARES
, spares
, nspares
, nv
);
5764 spa_load_spares(spa
);
5765 spa
->spa_spares
.sav_sync
= B_TRUE
;
5767 error
= SET_ERROR(EBUSY
);
5769 } else if (spa
->spa_l2cache
.sav_vdevs
!= NULL
&&
5770 nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
5771 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0 &&
5772 (nv
= spa_nvlist_lookup_by_guid(l2cache
, nl2cache
, guid
)) != NULL
) {
5774 * Cache devices can always be removed.
5776 vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
);
5777 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_AUX
);
5778 spa_vdev_remove_aux(spa
->spa_l2cache
.sav_config
,
5779 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
, nv
);
5780 spa_load_l2cache(spa
);
5781 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5782 } else if (vd
!= NULL
&& vd
->vdev_islog
) {
5784 ASSERT(vd
== vd
->vdev_top
);
5789 * Stop allocating from this vdev.
5791 metaslab_group_passivate(mg
);
5794 * Wait for the youngest allocations and frees to sync,
5795 * and then wait for the deferral of those frees to finish.
5797 spa_vdev_config_exit(spa
, NULL
,
5798 txg
+ TXG_CONCURRENT_STATES
+ TXG_DEFER_SIZE
, 0, FTAG
);
5801 * Attempt to evacuate the vdev.
5803 error
= spa_vdev_remove_evacuate(spa
, vd
);
5805 txg
= spa_vdev_config_enter(spa
);
5808 * If we couldn't evacuate the vdev, unwind.
5811 metaslab_group_activate(mg
);
5812 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5816 * Clean up the vdev namespace.
5818 ev
= spa_event_create(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE_DEV
);
5819 spa_vdev_remove_from_namespace(spa
, vd
);
5821 } else if (vd
!= NULL
) {
5823 * Normal vdevs cannot be removed (yet).
5825 error
= SET_ERROR(ENOTSUP
);
5828 * There is no vdev of any kind with the specified guid.
5830 error
= SET_ERROR(ENOENT
);
5834 error
= spa_vdev_exit(spa
, NULL
, txg
, error
);
5843 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5844 * currently spared, so we can detach it.
5847 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
5849 vdev_t
*newvd
, *oldvd
;
5851 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5852 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
5858 * Check for a completed replacement. We always consider the first
5859 * vdev in the list to be the oldest vdev, and the last one to be
5860 * the newest (see spa_vdev_attach() for how that works). In
5861 * the case where the newest vdev is faulted, we will not automatically
5862 * remove it after a resilver completes. This is OK as it will require
5863 * user intervention to determine which disk the admin wishes to keep.
5865 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
5866 ASSERT(vd
->vdev_children
> 1);
5868 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
5869 oldvd
= vd
->vdev_child
[0];
5871 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5872 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5873 !vdev_dtl_required(oldvd
))
5878 * Check for a completed resilver with the 'unspare' flag set.
5880 if (vd
->vdev_ops
== &vdev_spare_ops
) {
5881 vdev_t
*first
= vd
->vdev_child
[0];
5882 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
5884 if (last
->vdev_unspare
) {
5887 } else if (first
->vdev_unspare
) {
5894 if (oldvd
!= NULL
&&
5895 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
5896 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
5897 !vdev_dtl_required(oldvd
))
5901 * If there are more than two spares attached to a disk,
5902 * and those spares are not required, then we want to
5903 * attempt to free them up now so that they can be used
5904 * by other pools. Once we're back down to a single
5905 * disk+spare, we stop removing them.
5907 if (vd
->vdev_children
> 2) {
5908 newvd
= vd
->vdev_child
[1];
5910 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
5911 vdev_dtl_empty(last
, DTL_MISSING
) &&
5912 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
5913 !vdev_dtl_required(newvd
))
5922 spa_vdev_resilver_done(spa_t
*spa
)
5924 vdev_t
*vd
, *pvd
, *ppvd
;
5925 uint64_t guid
, sguid
, pguid
, ppguid
;
5927 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5929 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
5930 pvd
= vd
->vdev_parent
;
5931 ppvd
= pvd
->vdev_parent
;
5932 guid
= vd
->vdev_guid
;
5933 pguid
= pvd
->vdev_guid
;
5934 ppguid
= ppvd
->vdev_guid
;
5937 * If we have just finished replacing a hot spared device, then
5938 * we need to detach the parent's first child (the original hot
5941 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
5942 ppvd
->vdev_children
== 2) {
5943 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
5944 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
5946 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
5948 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5949 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
5951 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
5953 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5956 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5960 * Update the stored path or FRU for this vdev.
5963 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
5967 boolean_t sync
= B_FALSE
;
5969 ASSERT(spa_writeable(spa
));
5971 spa_vdev_state_enter(spa
, SCL_ALL
);
5973 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
5974 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
5976 if (!vd
->vdev_ops
->vdev_op_leaf
)
5977 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
5980 if (strcmp(value
, vd
->vdev_path
) != 0) {
5981 spa_strfree(vd
->vdev_path
);
5982 vd
->vdev_path
= spa_strdup(value
);
5986 if (vd
->vdev_fru
== NULL
) {
5987 vd
->vdev_fru
= spa_strdup(value
);
5989 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
5990 spa_strfree(vd
->vdev_fru
);
5991 vd
->vdev_fru
= spa_strdup(value
);
5996 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
6000 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
6002 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
6006 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
6008 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
6012 * ==========================================================================
6014 * ==========================================================================
6017 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
6019 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6021 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6022 return (SET_ERROR(EBUSY
));
6024 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
6028 spa_scan_stop(spa_t
*spa
)
6030 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6031 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6032 return (SET_ERROR(EBUSY
));
6033 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
6037 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
6039 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6041 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
6042 return (SET_ERROR(ENOTSUP
));
6045 * If a resilver was requested, but there is no DTL on a
6046 * writeable leaf device, we have nothing to do.
6048 if (func
== POOL_SCAN_RESILVER
&&
6049 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
6050 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6054 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6058 * ==========================================================================
6059 * SPA async task processing
6060 * ==========================================================================
6064 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6066 if (vd
->vdev_remove_wanted
) {
6067 vd
->vdev_remove_wanted
= B_FALSE
;
6068 vd
->vdev_delayed_close
= B_FALSE
;
6069 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6072 * We want to clear the stats, but we don't want to do a full
6073 * vdev_clear() as that will cause us to throw away
6074 * degraded/faulted state as well as attempt to reopen the
6075 * device, all of which is a waste.
6077 vd
->vdev_stat
.vs_read_errors
= 0;
6078 vd
->vdev_stat
.vs_write_errors
= 0;
6079 vd
->vdev_stat
.vs_checksum_errors
= 0;
6081 vdev_state_dirty(vd
->vdev_top
);
6084 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6085 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6089 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6091 if (vd
->vdev_probe_wanted
) {
6092 vd
->vdev_probe_wanted
= B_FALSE
;
6093 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6096 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6097 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6101 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6103 if (!spa
->spa_autoexpand
)
6106 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6107 vdev_t
*cvd
= vd
->vdev_child
[c
];
6108 spa_async_autoexpand(spa
, cvd
);
6111 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6114 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
6118 spa_async_thread(void *arg
)
6120 spa_t
*spa
= (spa_t
*)arg
;
6123 ASSERT(spa
->spa_sync_on
);
6125 mutex_enter(&spa
->spa_async_lock
);
6126 tasks
= spa
->spa_async_tasks
;
6127 spa
->spa_async_tasks
= 0;
6128 mutex_exit(&spa
->spa_async_lock
);
6131 * See if the config needs to be updated.
6133 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6134 uint64_t old_space
, new_space
;
6136 mutex_enter(&spa_namespace_lock
);
6137 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6138 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6139 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6140 mutex_exit(&spa_namespace_lock
);
6143 * If the pool grew as a result of the config update,
6144 * then log an internal history event.
6146 if (new_space
!= old_space
) {
6147 spa_history_log_internal(spa
, "vdev online", NULL
,
6148 "pool '%s' size: %llu(+%llu)",
6149 spa_name(spa
), new_space
, new_space
- old_space
);
6154 * See if any devices need to be marked REMOVED.
6156 if (tasks
& SPA_ASYNC_REMOVE
) {
6157 spa_vdev_state_enter(spa
, SCL_NONE
);
6158 spa_async_remove(spa
, spa
->spa_root_vdev
);
6159 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6160 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6161 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6162 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6163 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6166 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6167 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6168 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6169 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6173 * See if any devices need to be probed.
6175 if (tasks
& SPA_ASYNC_PROBE
) {
6176 spa_vdev_state_enter(spa
, SCL_NONE
);
6177 spa_async_probe(spa
, spa
->spa_root_vdev
);
6178 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6182 * If any devices are done replacing, detach them.
6184 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6185 spa_vdev_resilver_done(spa
);
6188 * Kick off a resilver.
6190 if (tasks
& SPA_ASYNC_RESILVER
)
6191 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6194 * Let the world know that we're done.
6196 mutex_enter(&spa
->spa_async_lock
);
6197 spa
->spa_async_thread
= NULL
;
6198 cv_broadcast(&spa
->spa_async_cv
);
6199 mutex_exit(&spa
->spa_async_lock
);
6204 spa_async_suspend(spa_t
*spa
)
6206 mutex_enter(&spa
->spa_async_lock
);
6207 spa
->spa_async_suspended
++;
6208 while (spa
->spa_async_thread
!= NULL
)
6209 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6210 mutex_exit(&spa
->spa_async_lock
);
6214 spa_async_resume(spa_t
*spa
)
6216 mutex_enter(&spa
->spa_async_lock
);
6217 ASSERT(spa
->spa_async_suspended
!= 0);
6218 spa
->spa_async_suspended
--;
6219 mutex_exit(&spa
->spa_async_lock
);
6223 spa_async_tasks_pending(spa_t
*spa
)
6225 uint_t non_config_tasks
;
6227 boolean_t config_task_suspended
;
6229 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6230 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6231 if (spa
->spa_ccw_fail_time
== 0) {
6232 config_task_suspended
= B_FALSE
;
6234 config_task_suspended
=
6235 (gethrtime() - spa
->spa_ccw_fail_time
) <
6236 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
6239 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6243 spa_async_dispatch(spa_t
*spa
)
6245 mutex_enter(&spa
->spa_async_lock
);
6246 if (spa_async_tasks_pending(spa
) &&
6247 !spa
->spa_async_suspended
&&
6248 spa
->spa_async_thread
== NULL
&&
6250 spa
->spa_async_thread
= thread_create(NULL
, 0,
6251 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6252 mutex_exit(&spa
->spa_async_lock
);
6256 spa_async_request(spa_t
*spa
, int task
)
6258 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6259 mutex_enter(&spa
->spa_async_lock
);
6260 spa
->spa_async_tasks
|= task
;
6261 mutex_exit(&spa
->spa_async_lock
);
6265 * ==========================================================================
6266 * SPA syncing routines
6267 * ==========================================================================
6271 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6274 bpobj_enqueue(bpo
, bp
, tx
);
6279 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6283 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6289 * Note: this simple function is not inlined to make it easier to dtrace the
6290 * amount of time spent syncing frees.
6293 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6295 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6296 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6297 VERIFY(zio_wait(zio
) == 0);
6301 * Note: this simple function is not inlined to make it easier to dtrace the
6302 * amount of time spent syncing deferred frees.
6305 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6307 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6308 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6309 spa_free_sync_cb
, zio
, tx
), ==, 0);
6310 VERIFY0(zio_wait(zio
));
6314 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6316 char *packed
= NULL
;
6321 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6324 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6325 * information. This avoids the dmu_buf_will_dirty() path and
6326 * saves us a pre-read to get data we don't actually care about.
6328 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6329 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6331 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6333 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6335 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6337 vmem_free(packed
, bufsize
);
6339 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6340 dmu_buf_will_dirty(db
, tx
);
6341 *(uint64_t *)db
->db_data
= nvsize
;
6342 dmu_buf_rele(db
, FTAG
);
6346 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6347 const char *config
, const char *entry
)
6357 * Update the MOS nvlist describing the list of available devices.
6358 * spa_validate_aux() will have already made sure this nvlist is
6359 * valid and the vdevs are labeled appropriately.
6361 if (sav
->sav_object
== 0) {
6362 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6363 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6364 sizeof (uint64_t), tx
);
6365 VERIFY(zap_update(spa
->spa_meta_objset
,
6366 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6367 &sav
->sav_object
, tx
) == 0);
6370 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6371 if (sav
->sav_count
== 0) {
6372 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6374 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6375 for (i
= 0; i
< sav
->sav_count
; i
++)
6376 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6377 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6378 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6379 sav
->sav_count
) == 0);
6380 for (i
= 0; i
< sav
->sav_count
; i
++)
6381 nvlist_free(list
[i
]);
6382 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6385 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6386 nvlist_free(nvroot
);
6388 sav
->sav_sync
= B_FALSE
;
6392 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6393 * The all-vdev ZAP must be empty.
6396 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6398 spa_t
*spa
= vd
->vdev_spa
;
6400 if (vd
->vdev_top_zap
!= 0) {
6401 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6402 vd
->vdev_top_zap
, tx
));
6404 if (vd
->vdev_leaf_zap
!= 0) {
6405 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6406 vd
->vdev_leaf_zap
, tx
));
6408 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
6409 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6414 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6419 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6420 * its config may not be dirty but we still need to build per-vdev ZAPs.
6421 * Similarly, if the pool is being assembled (e.g. after a split), we
6422 * need to rebuild the AVZ although the config may not be dirty.
6424 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6425 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6428 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6430 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6431 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6432 spa
->spa_all_vdev_zaps
!= 0);
6434 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6435 /* Make and build the new AVZ */
6436 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6437 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6438 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6440 /* Diff old AVZ with new one */
6444 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6445 spa
->spa_all_vdev_zaps
);
6446 zap_cursor_retrieve(&zc
, &za
) == 0;
6447 zap_cursor_advance(&zc
)) {
6448 uint64_t vdzap
= za
.za_first_integer
;
6449 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6452 * ZAP is listed in old AVZ but not in new one;
6455 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6460 zap_cursor_fini(&zc
);
6462 /* Destroy the old AVZ */
6463 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6464 spa
->spa_all_vdev_zaps
, tx
));
6466 /* Replace the old AVZ in the dir obj with the new one */
6467 VERIFY0(zap_update(spa
->spa_meta_objset
,
6468 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6469 sizeof (new_avz
), 1, &new_avz
, tx
));
6471 spa
->spa_all_vdev_zaps
= new_avz
;
6472 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6476 /* Walk through the AVZ and destroy all listed ZAPs */
6477 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6478 spa
->spa_all_vdev_zaps
);
6479 zap_cursor_retrieve(&zc
, &za
) == 0;
6480 zap_cursor_advance(&zc
)) {
6481 uint64_t zap
= za
.za_first_integer
;
6482 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6485 zap_cursor_fini(&zc
);
6487 /* Destroy and unlink the AVZ itself */
6488 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6489 spa
->spa_all_vdev_zaps
, tx
));
6490 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6491 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6492 spa
->spa_all_vdev_zaps
= 0;
6495 if (spa
->spa_all_vdev_zaps
== 0) {
6496 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6497 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6498 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6500 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6502 /* Create ZAPs for vdevs that don't have them. */
6503 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6505 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6506 dmu_tx_get_txg(tx
), B_FALSE
);
6509 * If we're upgrading the spa version then make sure that
6510 * the config object gets updated with the correct version.
6512 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6513 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6514 spa
->spa_uberblock
.ub_version
);
6516 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6518 nvlist_free(spa
->spa_config_syncing
);
6519 spa
->spa_config_syncing
= config
;
6521 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6525 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6527 uint64_t *versionp
= arg
;
6528 uint64_t version
= *versionp
;
6529 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6532 * Setting the version is special cased when first creating the pool.
6534 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6536 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6537 ASSERT(version
>= spa_version(spa
));
6539 spa
->spa_uberblock
.ub_version
= version
;
6540 vdev_config_dirty(spa
->spa_root_vdev
);
6541 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6545 * Set zpool properties.
6548 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6550 nvlist_t
*nvp
= arg
;
6551 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6552 objset_t
*mos
= spa
->spa_meta_objset
;
6553 nvpair_t
*elem
= NULL
;
6555 mutex_enter(&spa
->spa_props_lock
);
6557 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6559 char *strval
, *fname
;
6561 const char *propname
;
6562 zprop_type_t proptype
;
6565 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
6566 case ZPOOL_PROP_INVAL
:
6568 * We checked this earlier in spa_prop_validate().
6570 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6572 fname
= strchr(nvpair_name(elem
), '@') + 1;
6573 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6575 spa_feature_enable(spa
, fid
, tx
);
6576 spa_history_log_internal(spa
, "set", tx
,
6577 "%s=enabled", nvpair_name(elem
));
6580 case ZPOOL_PROP_VERSION
:
6581 intval
= fnvpair_value_uint64(elem
);
6583 * The version is synced separately before other
6584 * properties and should be correct by now.
6586 ASSERT3U(spa_version(spa
), >=, intval
);
6589 case ZPOOL_PROP_ALTROOT
:
6591 * 'altroot' is a non-persistent property. It should
6592 * have been set temporarily at creation or import time.
6594 ASSERT(spa
->spa_root
!= NULL
);
6597 case ZPOOL_PROP_READONLY
:
6598 case ZPOOL_PROP_CACHEFILE
:
6600 * 'readonly' and 'cachefile' are also non-persisitent
6604 case ZPOOL_PROP_COMMENT
:
6605 strval
= fnvpair_value_string(elem
);
6606 if (spa
->spa_comment
!= NULL
)
6607 spa_strfree(spa
->spa_comment
);
6608 spa
->spa_comment
= spa_strdup(strval
);
6610 * We need to dirty the configuration on all the vdevs
6611 * so that their labels get updated. It's unnecessary
6612 * to do this for pool creation since the vdev's
6613 * configuration has already been dirtied.
6615 if (tx
->tx_txg
!= TXG_INITIAL
)
6616 vdev_config_dirty(spa
->spa_root_vdev
);
6617 spa_history_log_internal(spa
, "set", tx
,
6618 "%s=%s", nvpair_name(elem
), strval
);
6622 * Set pool property values in the poolprops mos object.
6624 if (spa
->spa_pool_props_object
== 0) {
6625 spa
->spa_pool_props_object
=
6626 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6627 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6631 /* normalize the property name */
6632 propname
= zpool_prop_to_name(prop
);
6633 proptype
= zpool_prop_get_type(prop
);
6635 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6636 ASSERT(proptype
== PROP_TYPE_STRING
);
6637 strval
= fnvpair_value_string(elem
);
6638 VERIFY0(zap_update(mos
,
6639 spa
->spa_pool_props_object
, propname
,
6640 1, strlen(strval
) + 1, strval
, tx
));
6641 spa_history_log_internal(spa
, "set", tx
,
6642 "%s=%s", nvpair_name(elem
), strval
);
6643 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6644 intval
= fnvpair_value_uint64(elem
);
6646 if (proptype
== PROP_TYPE_INDEX
) {
6648 VERIFY0(zpool_prop_index_to_string(
6649 prop
, intval
, &unused
));
6651 VERIFY0(zap_update(mos
,
6652 spa
->spa_pool_props_object
, propname
,
6653 8, 1, &intval
, tx
));
6654 spa_history_log_internal(spa
, "set", tx
,
6655 "%s=%lld", nvpair_name(elem
), intval
);
6657 ASSERT(0); /* not allowed */
6661 case ZPOOL_PROP_DELEGATION
:
6662 spa
->spa_delegation
= intval
;
6664 case ZPOOL_PROP_BOOTFS
:
6665 spa
->spa_bootfs
= intval
;
6667 case ZPOOL_PROP_FAILUREMODE
:
6668 spa
->spa_failmode
= intval
;
6670 case ZPOOL_PROP_AUTOEXPAND
:
6671 spa
->spa_autoexpand
= intval
;
6672 if (tx
->tx_txg
!= TXG_INITIAL
)
6673 spa_async_request(spa
,
6674 SPA_ASYNC_AUTOEXPAND
);
6676 case ZPOOL_PROP_MULTIHOST
:
6677 spa
->spa_multihost
= intval
;
6679 case ZPOOL_PROP_DEDUPDITTO
:
6680 spa
->spa_dedup_ditto
= intval
;
6689 mutex_exit(&spa
->spa_props_lock
);
6693 * Perform one-time upgrade on-disk changes. spa_version() does not
6694 * reflect the new version this txg, so there must be no changes this
6695 * txg to anything that the upgrade code depends on after it executes.
6696 * Therefore this must be called after dsl_pool_sync() does the sync
6700 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6702 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6704 ASSERT(spa
->spa_sync_pass
== 1);
6706 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6708 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6709 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6710 dsl_pool_create_origin(dp
, tx
);
6712 /* Keeping the origin open increases spa_minref */
6713 spa
->spa_minref
+= 3;
6716 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6717 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6718 dsl_pool_upgrade_clones(dp
, tx
);
6721 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6722 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6723 dsl_pool_upgrade_dir_clones(dp
, tx
);
6725 /* Keeping the freedir open increases spa_minref */
6726 spa
->spa_minref
+= 3;
6729 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6730 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6731 spa_feature_create_zap_objects(spa
, tx
);
6735 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6736 * when possibility to use lz4 compression for metadata was added
6737 * Old pools that have this feature enabled must be upgraded to have
6738 * this feature active
6740 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6741 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6742 SPA_FEATURE_LZ4_COMPRESS
);
6743 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6744 SPA_FEATURE_LZ4_COMPRESS
);
6746 if (lz4_en
&& !lz4_ac
)
6747 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6751 * If we haven't written the salt, do so now. Note that the
6752 * feature may not be activated yet, but that's fine since
6753 * the presence of this ZAP entry is backwards compatible.
6755 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
6756 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
6757 VERIFY0(zap_add(spa
->spa_meta_objset
,
6758 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
6759 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
6760 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
6763 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6767 * Sync the specified transaction group. New blocks may be dirtied as
6768 * part of the process, so we iterate until it converges.
6771 spa_sync(spa_t
*spa
, uint64_t txg
)
6773 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6774 objset_t
*mos
= spa
->spa_meta_objset
;
6775 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
6776 vdev_t
*rvd
= spa
->spa_root_vdev
;
6780 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
6781 zfs_vdev_queue_depth_pct
/ 100;
6783 VERIFY(spa_writeable(spa
));
6786 * Lock out configuration changes.
6788 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6790 spa
->spa_syncing_txg
= txg
;
6791 spa
->spa_sync_pass
= 0;
6793 mutex_enter(&spa
->spa_alloc_lock
);
6794 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
6795 mutex_exit(&spa
->spa_alloc_lock
);
6798 * If there are any pending vdev state changes, convert them
6799 * into config changes that go out with this transaction group.
6801 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6802 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
6804 * We need the write lock here because, for aux vdevs,
6805 * calling vdev_config_dirty() modifies sav_config.
6806 * This is ugly and will become unnecessary when we
6807 * eliminate the aux vdev wart by integrating all vdevs
6808 * into the root vdev tree.
6810 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6811 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
6812 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
6813 vdev_state_clean(vd
);
6814 vdev_config_dirty(vd
);
6816 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
6817 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
6819 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6821 tx
= dmu_tx_create_assigned(dp
, txg
);
6823 spa
->spa_sync_starttime
= gethrtime();
6824 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
6825 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
6826 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
6827 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
6830 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6831 * set spa_deflate if we have no raid-z vdevs.
6833 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
6834 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
6837 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
6838 vd
= rvd
->vdev_child
[i
];
6839 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
6842 if (i
== rvd
->vdev_children
) {
6843 spa
->spa_deflate
= TRUE
;
6844 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
6845 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
6846 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
6851 * Set the top-level vdev's max queue depth. Evaluate each
6852 * top-level's async write queue depth in case it changed.
6853 * The max queue depth will not change in the middle of syncing
6856 uint64_t queue_depth_total
= 0;
6857 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
6858 vdev_t
*tvd
= rvd
->vdev_child
[c
];
6859 metaslab_group_t
*mg
= tvd
->vdev_mg
;
6861 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
6862 !metaslab_group_initialized(mg
))
6866 * It is safe to do a lock-free check here because only async
6867 * allocations look at mg_max_alloc_queue_depth, and async
6868 * allocations all happen from spa_sync().
6870 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
6871 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
6872 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
6874 metaslab_class_t
*mc
= spa_normal_class(spa
);
6875 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
6876 mc
->mc_alloc_max_slots
= queue_depth_total
;
6877 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
6879 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
6880 max_queue_depth
* rvd
->vdev_children
);
6883 * Iterate to convergence.
6886 int pass
= ++spa
->spa_sync_pass
;
6888 spa_sync_config_object(spa
, tx
);
6889 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
6890 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
6891 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
6892 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
6893 spa_errlog_sync(spa
, txg
);
6894 dsl_pool_sync(dp
, txg
);
6896 if (pass
< zfs_sync_pass_deferred_free
) {
6897 spa_sync_frees(spa
, free_bpl
, tx
);
6900 * We can not defer frees in pass 1, because
6901 * we sync the deferred frees later in pass 1.
6903 ASSERT3U(pass
, >, 1);
6904 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
6905 &spa
->spa_deferred_bpobj
, tx
);
6909 dsl_scan_sync(dp
, tx
);
6911 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
)))
6915 spa_sync_upgrades(spa
, tx
);
6917 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
6919 * Note: We need to check if the MOS is dirty
6920 * because we could have marked the MOS dirty
6921 * without updating the uberblock (e.g. if we
6922 * have sync tasks but no dirty user data). We
6923 * need to check the uberblock's rootbp because
6924 * it is updated if we have synced out dirty
6925 * data (though in this case the MOS will most
6926 * likely also be dirty due to second order
6927 * effects, we don't want to rely on that here).
6929 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
6930 !dmu_objset_is_dirty(mos
, txg
)) {
6932 * Nothing changed on the first pass,
6933 * therefore this TXG is a no-op. Avoid
6934 * syncing deferred frees, so that we
6935 * can keep this TXG as a no-op.
6937 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
6939 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
6940 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
6943 spa_sync_deferred_frees(spa
, tx
);
6946 } while (dmu_objset_is_dirty(mos
, txg
));
6949 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
6951 * Make sure that the number of ZAPs for all the vdevs matches
6952 * the number of ZAPs in the per-vdev ZAP list. This only gets
6953 * called if the config is dirty; otherwise there may be
6954 * outstanding AVZ operations that weren't completed in
6955 * spa_sync_config_object.
6957 uint64_t all_vdev_zap_entry_count
;
6958 ASSERT0(zap_count(spa
->spa_meta_objset
,
6959 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
6960 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
6961 all_vdev_zap_entry_count
);
6966 * Rewrite the vdev configuration (which includes the uberblock)
6967 * to commit the transaction group.
6969 * If there are no dirty vdevs, we sync the uberblock to a few
6970 * random top-level vdevs that are known to be visible in the
6971 * config cache (see spa_vdev_add() for a complete description).
6972 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6976 * We hold SCL_STATE to prevent vdev open/close/etc.
6977 * while we're attempting to write the vdev labels.
6979 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6981 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
6982 vdev_t
*svd
[SPA_DVAS_PER_BP
];
6984 int children
= rvd
->vdev_children
;
6985 int c0
= spa_get_random(children
);
6987 for (int c
= 0; c
< children
; c
++) {
6988 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
6989 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
)
6991 svd
[svdcount
++] = vd
;
6992 if (svdcount
== SPA_DVAS_PER_BP
)
6995 error
= vdev_config_sync(svd
, svdcount
, txg
);
6997 error
= vdev_config_sync(rvd
->vdev_child
,
6998 rvd
->vdev_children
, txg
);
7002 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
7004 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7008 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
7009 zio_resume_wait(spa
);
7013 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
7014 spa
->spa_deadman_tqid
= 0;
7017 * Clear the dirty config list.
7019 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
7020 vdev_config_clean(vd
);
7023 * Now that the new config has synced transactionally,
7024 * let it become visible to the config cache.
7026 if (spa
->spa_config_syncing
!= NULL
) {
7027 spa_config_set(spa
, spa
->spa_config_syncing
);
7028 spa
->spa_config_txg
= txg
;
7029 spa
->spa_config_syncing
= NULL
;
7032 dsl_pool_sync_done(dp
, txg
);
7034 mutex_enter(&spa
->spa_alloc_lock
);
7035 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7036 mutex_exit(&spa
->spa_alloc_lock
);
7039 * Update usable space statistics.
7041 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
7042 vdev_sync_done(vd
, txg
);
7044 spa_update_dspace(spa
);
7047 * It had better be the case that we didn't dirty anything
7048 * since vdev_config_sync().
7050 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7051 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7052 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7054 spa
->spa_sync_pass
= 0;
7057 * Update the last synced uberblock here. We want to do this at
7058 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7059 * will be guaranteed that all the processing associated with
7060 * that txg has been completed.
7062 spa
->spa_ubsync
= spa
->spa_uberblock
;
7063 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7065 spa_handle_ignored_writes(spa
);
7068 * If any async tasks have been requested, kick them off.
7070 spa_async_dispatch(spa
);
7074 * Sync all pools. We don't want to hold the namespace lock across these
7075 * operations, so we take a reference on the spa_t and drop the lock during the
7079 spa_sync_allpools(void)
7082 mutex_enter(&spa_namespace_lock
);
7083 while ((spa
= spa_next(spa
)) != NULL
) {
7084 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7085 !spa_writeable(spa
) || spa_suspended(spa
))
7087 spa_open_ref(spa
, FTAG
);
7088 mutex_exit(&spa_namespace_lock
);
7089 txg_wait_synced(spa_get_dsl(spa
), 0);
7090 mutex_enter(&spa_namespace_lock
);
7091 spa_close(spa
, FTAG
);
7093 mutex_exit(&spa_namespace_lock
);
7097 * ==========================================================================
7098 * Miscellaneous routines
7099 * ==========================================================================
7103 * Remove all pools in the system.
7111 * Remove all cached state. All pools should be closed now,
7112 * so every spa in the AVL tree should be unreferenced.
7114 mutex_enter(&spa_namespace_lock
);
7115 while ((spa
= spa_next(NULL
)) != NULL
) {
7117 * Stop async tasks. The async thread may need to detach
7118 * a device that's been replaced, which requires grabbing
7119 * spa_namespace_lock, so we must drop it here.
7121 spa_open_ref(spa
, FTAG
);
7122 mutex_exit(&spa_namespace_lock
);
7123 spa_async_suspend(spa
);
7124 mutex_enter(&spa_namespace_lock
);
7125 spa_close(spa
, FTAG
);
7127 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7129 spa_deactivate(spa
);
7133 mutex_exit(&spa_namespace_lock
);
7137 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7142 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7146 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7147 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7148 if (vd
->vdev_guid
== guid
)
7152 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7153 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7154 if (vd
->vdev_guid
== guid
)
7163 spa_upgrade(spa_t
*spa
, uint64_t version
)
7165 ASSERT(spa_writeable(spa
));
7167 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7170 * This should only be called for a non-faulted pool, and since a
7171 * future version would result in an unopenable pool, this shouldn't be
7174 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7175 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7177 spa
->spa_uberblock
.ub_version
= version
;
7178 vdev_config_dirty(spa
->spa_root_vdev
);
7180 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7182 txg_wait_synced(spa_get_dsl(spa
), 0);
7186 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7190 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7192 for (i
= 0; i
< sav
->sav_count
; i
++)
7193 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7196 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7197 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7198 &spareguid
) == 0 && spareguid
== guid
)
7206 * Check if a pool has an active shared spare device.
7207 * Note: reference count of an active spare is 2, as a spare and as a replace
7210 spa_has_active_shared_spare(spa_t
*spa
)
7214 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7216 for (i
= 0; i
< sav
->sav_count
; i
++) {
7217 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7218 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7227 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7229 sysevent_t
*ev
= NULL
;
7233 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
7235 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
7236 ev
->resource
= resource
;
7243 spa_event_post(sysevent_t
*ev
)
7247 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
7248 kmem_free(ev
, sizeof (*ev
));
7254 * Post a zevent corresponding to the given sysevent. The 'name' must be one
7255 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
7256 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7257 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7258 * or zdb as real changes.
7261 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7263 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
7266 #if defined(_KERNEL) && defined(HAVE_SPL)
7267 /* state manipulation functions */
7268 EXPORT_SYMBOL(spa_open
);
7269 EXPORT_SYMBOL(spa_open_rewind
);
7270 EXPORT_SYMBOL(spa_get_stats
);
7271 EXPORT_SYMBOL(spa_create
);
7272 EXPORT_SYMBOL(spa_import
);
7273 EXPORT_SYMBOL(spa_tryimport
);
7274 EXPORT_SYMBOL(spa_destroy
);
7275 EXPORT_SYMBOL(spa_export
);
7276 EXPORT_SYMBOL(spa_reset
);
7277 EXPORT_SYMBOL(spa_async_request
);
7278 EXPORT_SYMBOL(spa_async_suspend
);
7279 EXPORT_SYMBOL(spa_async_resume
);
7280 EXPORT_SYMBOL(spa_inject_addref
);
7281 EXPORT_SYMBOL(spa_inject_delref
);
7282 EXPORT_SYMBOL(spa_scan_stat_init
);
7283 EXPORT_SYMBOL(spa_scan_get_stats
);
7285 /* device maniion */
7286 EXPORT_SYMBOL(spa_vdev_add
);
7287 EXPORT_SYMBOL(spa_vdev_attach
);
7288 EXPORT_SYMBOL(spa_vdev_detach
);
7289 EXPORT_SYMBOL(spa_vdev_remove
);
7290 EXPORT_SYMBOL(spa_vdev_setpath
);
7291 EXPORT_SYMBOL(spa_vdev_setfru
);
7292 EXPORT_SYMBOL(spa_vdev_split_mirror
);
7294 /* spare statech is global across all pools) */
7295 EXPORT_SYMBOL(spa_spare_add
);
7296 EXPORT_SYMBOL(spa_spare_remove
);
7297 EXPORT_SYMBOL(spa_spare_exists
);
7298 EXPORT_SYMBOL(spa_spare_activate
);
7300 /* L2ARC statech is global across all pools) */
7301 EXPORT_SYMBOL(spa_l2cache_add
);
7302 EXPORT_SYMBOL(spa_l2cache_remove
);
7303 EXPORT_SYMBOL(spa_l2cache_exists
);
7304 EXPORT_SYMBOL(spa_l2cache_activate
);
7305 EXPORT_SYMBOL(spa_l2cache_drop
);
7308 EXPORT_SYMBOL(spa_scan
);
7309 EXPORT_SYMBOL(spa_scan_stop
);
7312 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
7313 EXPORT_SYMBOL(spa_sync_allpools
);
7316 EXPORT_SYMBOL(spa_prop_set
);
7317 EXPORT_SYMBOL(spa_prop_get
);
7318 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
7320 /* asynchronous event notification */
7321 EXPORT_SYMBOL(spa_event_notify
);
7324 #if defined(_KERNEL) && defined(HAVE_SPL)
7325 module_param(spa_load_verify_maxinflight
, int, 0644);
7326 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
7327 "Max concurrent traversal I/Os while verifying pool during import -X");
7329 module_param(spa_load_verify_metadata
, int, 0644);
7330 MODULE_PARM_DESC(spa_load_verify_metadata
,
7331 "Set to traverse metadata on pool import");
7333 module_param(spa_load_verify_data
, int, 0644);
7334 MODULE_PARM_DESC(spa_load_verify_data
,
7335 "Set to traverse data on pool import");
7338 module_param(zio_taskq_batch_pct
, uint
, 0444);
7339 MODULE_PARM_DESC(zio_taskq_batch_pct
,
7340 "Percentage of CPUs to run an IO worker thread");