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_removal.h>
56 #include <sys/vdev_indirect_mapping.h>
57 #include <sys/vdev_indirect_births.h>
58 #include <sys/vdev_disk.h>
59 #include <sys/metaslab.h>
60 #include <sys/metaslab_impl.h>
62 #include <sys/uberblock_impl.h>
65 #include <sys/bpobj.h>
66 #include <sys/dmu_traverse.h>
67 #include <sys/dmu_objset.h>
68 #include <sys/unique.h>
69 #include <sys/dsl_pool.h>
70 #include <sys/dsl_dataset.h>
71 #include <sys/dsl_dir.h>
72 #include <sys/dsl_prop.h>
73 #include <sys/dsl_synctask.h>
74 #include <sys/fs/zfs.h>
76 #include <sys/callb.h>
77 #include <sys/systeminfo.h>
78 #include <sys/spa_boot.h>
79 #include <sys/zfs_ioctl.h>
80 #include <sys/dsl_scan.h>
81 #include <sys/zfeature.h>
82 #include <sys/dsl_destroy.h>
86 #include <sys/fm/protocol.h>
87 #include <sys/fm/util.h>
88 #include <sys/callb.h>
93 #include "zfs_comutil.h"
96 * The interval, in seconds, at which failed configuration cache file writes
99 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 void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
154 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
155 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
156 static int spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
,
157 boolean_t reloading
);
158 static void spa_vdev_resilver_done(spa_t
*spa
);
160 uint_t zio_taskq_batch_pct
= 75; /* 1 thread per cpu in pset */
161 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
162 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
164 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
167 * Report any spa_load_verify errors found, but do not fail spa_load.
168 * This is used by zdb to analyze non-idle pools.
170 boolean_t spa_load_verify_dryrun
= B_FALSE
;
173 * This (illegal) pool name is used when temporarily importing a spa_t in order
174 * to get the vdev stats associated with the imported devices.
176 #define TRYIMPORT_NAME "$import"
179 * For debugging purposes: print out vdev tree during pool import.
181 int spa_load_print_vdev_tree
= B_FALSE
;
184 * A non-zero value for zfs_max_missing_tvds means that we allow importing
185 * pools with missing top-level vdevs. This is strictly intended for advanced
186 * pool recovery cases since missing data is almost inevitable. Pools with
187 * missing devices can only be imported read-only for safety reasons, and their
188 * fail-mode will be automatically set to "continue".
190 * With 1 missing vdev we should be able to import the pool and mount all
191 * datasets. User data that was not modified after the missing device has been
192 * added should be recoverable. This means that snapshots created prior to the
193 * addition of that device should be completely intact.
195 * With 2 missing vdevs, some datasets may fail to mount since there are
196 * dataset statistics that are stored as regular metadata. Some data might be
197 * recoverable if those vdevs were added recently.
199 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
200 * may be missing entirely. Chances of data recovery are very low. Note that
201 * there are also risks of performing an inadvertent rewind as we might be
202 * missing all the vdevs with the latest uberblocks.
204 unsigned long zfs_max_missing_tvds
= 0;
207 * The parameters below are similar to zfs_max_missing_tvds but are only
208 * intended for a preliminary open of the pool with an untrusted config which
209 * might be incomplete or out-dated.
211 * We are more tolerant for pools opened from a cachefile since we could have
212 * an out-dated cachefile where a device removal was not registered.
213 * We could have set the limit arbitrarily high but in the case where devices
214 * are really missing we would want to return the proper error codes; we chose
215 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
216 * and we get a chance to retrieve the trusted config.
218 uint64_t zfs_max_missing_tvds_cachefile
= SPA_DVAS_PER_BP
- 1;
220 * In the case where config was assembled by scanning device paths (/dev/dsks
221 * by default) we are less tolerant since all the existing devices should have
222 * been detected and we want spa_load to return the right error codes.
224 uint64_t zfs_max_missing_tvds_scan
= 0;
227 * ==========================================================================
228 * SPA properties routines
229 * ==========================================================================
233 * Add a (source=src, propname=propval) list to an nvlist.
236 spa_prop_add_list(nvlist_t
*nvl
, zpool_prop_t prop
, char *strval
,
237 uint64_t intval
, zprop_source_t src
)
239 const char *propname
= zpool_prop_to_name(prop
);
242 VERIFY(nvlist_alloc(&propval
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
243 VERIFY(nvlist_add_uint64(propval
, ZPROP_SOURCE
, src
) == 0);
246 VERIFY(nvlist_add_string(propval
, ZPROP_VALUE
, strval
) == 0);
248 VERIFY(nvlist_add_uint64(propval
, ZPROP_VALUE
, intval
) == 0);
250 VERIFY(nvlist_add_nvlist(nvl
, propname
, propval
) == 0);
251 nvlist_free(propval
);
255 * Get property values from the spa configuration.
258 spa_prop_get_config(spa_t
*spa
, nvlist_t
**nvp
)
260 vdev_t
*rvd
= spa
->spa_root_vdev
;
261 dsl_pool_t
*pool
= spa
->spa_dsl_pool
;
262 uint64_t size
, alloc
, cap
, version
;
263 const zprop_source_t src
= ZPROP_SRC_NONE
;
264 spa_config_dirent_t
*dp
;
265 metaslab_class_t
*mc
= spa_normal_class(spa
);
267 ASSERT(MUTEX_HELD(&spa
->spa_props_lock
));
270 alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
271 size
= metaslab_class_get_space(spa_normal_class(spa
));
272 spa_prop_add_list(*nvp
, ZPOOL_PROP_NAME
, spa_name(spa
), 0, src
);
273 spa_prop_add_list(*nvp
, ZPOOL_PROP_SIZE
, NULL
, size
, src
);
274 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALLOCATED
, NULL
, alloc
, src
);
275 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREE
, NULL
,
278 spa_prop_add_list(*nvp
, ZPOOL_PROP_FRAGMENTATION
, NULL
,
279 metaslab_class_fragmentation(mc
), src
);
280 spa_prop_add_list(*nvp
, ZPOOL_PROP_EXPANDSZ
, NULL
,
281 metaslab_class_expandable_space(mc
), src
);
282 spa_prop_add_list(*nvp
, ZPOOL_PROP_READONLY
, NULL
,
283 (spa_mode(spa
) == FREAD
), src
);
285 cap
= (size
== 0) ? 0 : (alloc
* 100 / size
);
286 spa_prop_add_list(*nvp
, ZPOOL_PROP_CAPACITY
, NULL
, cap
, src
);
288 spa_prop_add_list(*nvp
, ZPOOL_PROP_DEDUPRATIO
, NULL
,
289 ddt_get_pool_dedup_ratio(spa
), src
);
291 spa_prop_add_list(*nvp
, ZPOOL_PROP_HEALTH
, NULL
,
292 rvd
->vdev_state
, src
);
294 version
= spa_version(spa
);
295 if (version
== zpool_prop_default_numeric(ZPOOL_PROP_VERSION
)) {
296 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
297 version
, ZPROP_SRC_DEFAULT
);
299 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
,
300 version
, ZPROP_SRC_LOCAL
);
306 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
307 * when opening pools before this version freedir will be NULL.
309 if (pool
->dp_free_dir
!= NULL
) {
310 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
311 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
314 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
318 if (pool
->dp_leak_dir
!= NULL
) {
319 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
320 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
323 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
328 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
330 if (spa
->spa_comment
!= NULL
) {
331 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
335 if (spa
->spa_root
!= NULL
)
336 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
339 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
340 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
341 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
343 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
344 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
347 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
)) {
348 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
349 DNODE_MAX_SIZE
, ZPROP_SRC_NONE
);
351 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXDNODESIZE
, NULL
,
352 DNODE_MIN_SIZE
, ZPROP_SRC_NONE
);
355 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
356 if (dp
->scd_path
== NULL
) {
357 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
358 "none", 0, ZPROP_SRC_LOCAL
);
359 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
360 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
361 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
367 * Get zpool property values.
370 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
372 objset_t
*mos
= spa
->spa_meta_objset
;
377 err
= nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
);
381 mutex_enter(&spa
->spa_props_lock
);
384 * Get properties from the spa config.
386 spa_prop_get_config(spa
, nvp
);
388 /* If no pool property object, no more prop to get. */
389 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
390 mutex_exit(&spa
->spa_props_lock
);
395 * Get properties from the MOS pool property object.
397 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
398 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
399 zap_cursor_advance(&zc
)) {
402 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
405 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPOOL_PROP_INVAL
)
408 switch (za
.za_integer_length
) {
410 /* integer property */
411 if (za
.za_first_integer
!=
412 zpool_prop_default_numeric(prop
))
413 src
= ZPROP_SRC_LOCAL
;
415 if (prop
== ZPOOL_PROP_BOOTFS
) {
417 dsl_dataset_t
*ds
= NULL
;
419 dp
= spa_get_dsl(spa
);
420 dsl_pool_config_enter(dp
, FTAG
);
421 if ((err
= dsl_dataset_hold_obj(dp
,
422 za
.za_first_integer
, FTAG
, &ds
))) {
423 dsl_pool_config_exit(dp
, FTAG
);
427 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
429 dsl_dataset_name(ds
, strval
);
430 dsl_dataset_rele(ds
, FTAG
);
431 dsl_pool_config_exit(dp
, FTAG
);
434 intval
= za
.za_first_integer
;
437 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
440 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
445 /* string property */
446 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
447 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
448 za
.za_name
, 1, za
.za_num_integers
, strval
);
450 kmem_free(strval
, za
.za_num_integers
);
453 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
454 kmem_free(strval
, za
.za_num_integers
);
461 zap_cursor_fini(&zc
);
462 mutex_exit(&spa
->spa_props_lock
);
464 if (err
&& err
!= ENOENT
) {
474 * Validate the given pool properties nvlist and modify the list
475 * for the property values to be set.
478 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
481 int error
= 0, reset_bootfs
= 0;
483 boolean_t has_feature
= B_FALSE
;
486 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
488 char *strval
, *slash
, *check
, *fname
;
489 const char *propname
= nvpair_name(elem
);
490 zpool_prop_t prop
= zpool_name_to_prop(propname
);
493 case ZPOOL_PROP_INVAL
:
494 if (!zpool_prop_feature(propname
)) {
495 error
= SET_ERROR(EINVAL
);
500 * Sanitize the input.
502 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
503 error
= SET_ERROR(EINVAL
);
507 if (nvpair_value_uint64(elem
, &intval
) != 0) {
508 error
= SET_ERROR(EINVAL
);
513 error
= SET_ERROR(EINVAL
);
517 fname
= strchr(propname
, '@') + 1;
518 if (zfeature_lookup_name(fname
, NULL
) != 0) {
519 error
= SET_ERROR(EINVAL
);
523 has_feature
= B_TRUE
;
526 case ZPOOL_PROP_VERSION
:
527 error
= nvpair_value_uint64(elem
, &intval
);
529 (intval
< spa_version(spa
) ||
530 intval
> SPA_VERSION_BEFORE_FEATURES
||
532 error
= SET_ERROR(EINVAL
);
535 case ZPOOL_PROP_DELEGATION
:
536 case ZPOOL_PROP_AUTOREPLACE
:
537 case ZPOOL_PROP_LISTSNAPS
:
538 case ZPOOL_PROP_AUTOEXPAND
:
539 error
= nvpair_value_uint64(elem
, &intval
);
540 if (!error
&& intval
> 1)
541 error
= SET_ERROR(EINVAL
);
544 case ZPOOL_PROP_MULTIHOST
:
545 error
= nvpair_value_uint64(elem
, &intval
);
546 if (!error
&& intval
> 1)
547 error
= SET_ERROR(EINVAL
);
549 if (!error
&& !spa_get_hostid())
550 error
= SET_ERROR(ENOTSUP
);
554 case ZPOOL_PROP_BOOTFS
:
556 * If the pool version is less than SPA_VERSION_BOOTFS,
557 * or the pool is still being created (version == 0),
558 * the bootfs property cannot be set.
560 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
561 error
= SET_ERROR(ENOTSUP
);
566 * Make sure the vdev config is bootable
568 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
569 error
= SET_ERROR(ENOTSUP
);
575 error
= nvpair_value_string(elem
, &strval
);
581 if (strval
== NULL
|| strval
[0] == '\0') {
582 objnum
= zpool_prop_default_numeric(
587 error
= dmu_objset_hold(strval
, FTAG
, &os
);
592 * Must be ZPL, and its property settings
593 * must be supported by GRUB (compression
594 * is not gzip, and large blocks or large
595 * dnodes are not used).
598 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
599 error
= SET_ERROR(ENOTSUP
);
601 dsl_prop_get_int_ds(dmu_objset_ds(os
),
602 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
604 !BOOTFS_COMPRESS_VALID(propval
)) {
605 error
= SET_ERROR(ENOTSUP
);
607 dsl_prop_get_int_ds(dmu_objset_ds(os
),
608 zfs_prop_to_name(ZFS_PROP_DNODESIZE
),
610 propval
!= ZFS_DNSIZE_LEGACY
) {
611 error
= SET_ERROR(ENOTSUP
);
613 objnum
= dmu_objset_id(os
);
615 dmu_objset_rele(os
, FTAG
);
619 case ZPOOL_PROP_FAILUREMODE
:
620 error
= nvpair_value_uint64(elem
, &intval
);
621 if (!error
&& intval
> ZIO_FAILURE_MODE_PANIC
)
622 error
= SET_ERROR(EINVAL
);
625 * This is a special case which only occurs when
626 * the pool has completely failed. This allows
627 * the user to change the in-core failmode property
628 * without syncing it out to disk (I/Os might
629 * currently be blocked). We do this by returning
630 * EIO to the caller (spa_prop_set) to trick it
631 * into thinking we encountered a property validation
634 if (!error
&& spa_suspended(spa
)) {
635 spa
->spa_failmode
= intval
;
636 error
= SET_ERROR(EIO
);
640 case ZPOOL_PROP_CACHEFILE
:
641 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
644 if (strval
[0] == '\0')
647 if (strcmp(strval
, "none") == 0)
650 if (strval
[0] != '/') {
651 error
= SET_ERROR(EINVAL
);
655 slash
= strrchr(strval
, '/');
656 ASSERT(slash
!= NULL
);
658 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
659 strcmp(slash
, "/..") == 0)
660 error
= SET_ERROR(EINVAL
);
663 case ZPOOL_PROP_COMMENT
:
664 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
666 for (check
= strval
; *check
!= '\0'; check
++) {
667 if (!isprint(*check
)) {
668 error
= SET_ERROR(EINVAL
);
672 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
673 error
= SET_ERROR(E2BIG
);
676 case ZPOOL_PROP_DEDUPDITTO
:
677 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
678 error
= SET_ERROR(ENOTSUP
);
680 error
= nvpair_value_uint64(elem
, &intval
);
682 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
683 error
= SET_ERROR(EINVAL
);
694 if (!error
&& reset_bootfs
) {
695 error
= nvlist_remove(props
,
696 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
699 error
= nvlist_add_uint64(props
,
700 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
708 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
711 spa_config_dirent_t
*dp
;
713 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
717 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
720 if (cachefile
[0] == '\0')
721 dp
->scd_path
= spa_strdup(spa_config_path
);
722 else if (strcmp(cachefile
, "none") == 0)
725 dp
->scd_path
= spa_strdup(cachefile
);
727 list_insert_head(&spa
->spa_config_list
, dp
);
729 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
733 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
736 nvpair_t
*elem
= NULL
;
737 boolean_t need_sync
= B_FALSE
;
739 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
742 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
743 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
745 if (prop
== ZPOOL_PROP_CACHEFILE
||
746 prop
== ZPOOL_PROP_ALTROOT
||
747 prop
== ZPOOL_PROP_READONLY
)
750 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPOOL_PROP_INVAL
) {
753 if (prop
== ZPOOL_PROP_VERSION
) {
754 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
756 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
757 ver
= SPA_VERSION_FEATURES
;
761 /* Save time if the version is already set. */
762 if (ver
== spa_version(spa
))
766 * In addition to the pool directory object, we might
767 * create the pool properties object, the features for
768 * read object, the features for write object, or the
769 * feature descriptions object.
771 error
= dsl_sync_task(spa
->spa_name
, NULL
,
772 spa_sync_version
, &ver
,
773 6, ZFS_SPACE_CHECK_RESERVED
);
784 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
785 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
792 * If the bootfs property value is dsobj, clear it.
795 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
797 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
798 VERIFY(zap_remove(spa
->spa_meta_objset
,
799 spa
->spa_pool_props_object
,
800 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
807 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
809 ASSERTV(uint64_t *newguid
= arg
);
810 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
811 vdev_t
*rvd
= spa
->spa_root_vdev
;
814 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
815 vdev_state
= rvd
->vdev_state
;
816 spa_config_exit(spa
, SCL_STATE
, FTAG
);
818 if (vdev_state
!= VDEV_STATE_HEALTHY
)
819 return (SET_ERROR(ENXIO
));
821 ASSERT3U(spa_guid(spa
), !=, *newguid
);
827 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
829 uint64_t *newguid
= arg
;
830 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
832 vdev_t
*rvd
= spa
->spa_root_vdev
;
834 oldguid
= spa_guid(spa
);
836 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
837 rvd
->vdev_guid
= *newguid
;
838 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
839 vdev_config_dirty(rvd
);
840 spa_config_exit(spa
, SCL_STATE
, FTAG
);
842 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
847 * Change the GUID for the pool. This is done so that we can later
848 * re-import a pool built from a clone of our own vdevs. We will modify
849 * the root vdev's guid, our own pool guid, and then mark all of our
850 * vdevs dirty. Note that we must make sure that all our vdevs are
851 * online when we do this, or else any vdevs that weren't present
852 * would be orphaned from our pool. We are also going to issue a
853 * sysevent to update any watchers.
856 spa_change_guid(spa_t
*spa
)
861 mutex_enter(&spa
->spa_vdev_top_lock
);
862 mutex_enter(&spa_namespace_lock
);
863 guid
= spa_generate_guid(NULL
);
865 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
866 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
869 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
870 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
873 mutex_exit(&spa_namespace_lock
);
874 mutex_exit(&spa
->spa_vdev_top_lock
);
880 * ==========================================================================
881 * SPA state manipulation (open/create/destroy/import/export)
882 * ==========================================================================
886 spa_error_entry_compare(const void *a
, const void *b
)
888 const spa_error_entry_t
*sa
= (const spa_error_entry_t
*)a
;
889 const spa_error_entry_t
*sb
= (const spa_error_entry_t
*)b
;
892 ret
= memcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
893 sizeof (zbookmark_phys_t
));
895 return (AVL_ISIGN(ret
));
899 * Utility function which retrieves copies of the current logs and
900 * re-initializes them in the process.
903 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
905 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
907 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
908 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
910 avl_create(&spa
->spa_errlist_scrub
,
911 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
912 offsetof(spa_error_entry_t
, se_avl
));
913 avl_create(&spa
->spa_errlist_last
,
914 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
915 offsetof(spa_error_entry_t
, se_avl
));
919 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
921 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
922 enum zti_modes mode
= ztip
->zti_mode
;
923 uint_t value
= ztip
->zti_value
;
924 uint_t count
= ztip
->zti_count
;
925 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
928 boolean_t batch
= B_FALSE
;
930 if (mode
== ZTI_MODE_NULL
) {
932 tqs
->stqs_taskq
= NULL
;
936 ASSERT3U(count
, >, 0);
938 tqs
->stqs_count
= count
;
939 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
943 ASSERT3U(value
, >=, 1);
944 value
= MAX(value
, 1);
945 flags
|= TASKQ_DYNAMIC
;
950 flags
|= TASKQ_THREADS_CPU_PCT
;
951 value
= MIN(zio_taskq_batch_pct
, 100);
955 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
957 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
961 for (uint_t i
= 0; i
< count
; i
++) {
965 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
966 zio_type_name
[t
], zio_taskq_types
[q
], i
);
968 (void) snprintf(name
, sizeof (name
), "%s_%s",
969 zio_type_name
[t
], zio_taskq_types
[q
]);
972 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
974 flags
|= TASKQ_DC_BATCH
;
976 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
977 spa
->spa_proc
, zio_taskq_basedc
, flags
);
979 pri_t pri
= maxclsyspri
;
981 * The write issue taskq can be extremely CPU
982 * intensive. Run it at slightly less important
983 * priority than the other taskqs. Under Linux this
984 * means incrementing the priority value on platforms
985 * like illumos it should be decremented.
987 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
990 tq
= taskq_create_proc(name
, value
, pri
, 50,
991 INT_MAX
, spa
->spa_proc
, flags
);
994 tqs
->stqs_taskq
[i
] = tq
;
999 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
1001 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1003 if (tqs
->stqs_taskq
== NULL
) {
1004 ASSERT3U(tqs
->stqs_count
, ==, 0);
1008 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
1009 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
1010 taskq_destroy(tqs
->stqs_taskq
[i
]);
1013 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
1014 tqs
->stqs_taskq
= NULL
;
1018 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1019 * Note that a type may have multiple discrete taskqs to avoid lock contention
1020 * on the taskq itself. In that case we choose which taskq at random by using
1021 * the low bits of gethrtime().
1024 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1025 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
1027 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1030 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1031 ASSERT3U(tqs
->stqs_count
, !=, 0);
1033 if (tqs
->stqs_count
== 1) {
1034 tq
= tqs
->stqs_taskq
[0];
1036 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1039 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
1043 * Same as spa_taskq_dispatch_ent() but block on the task until completion.
1046 spa_taskq_dispatch_sync(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1047 task_func_t
*func
, void *arg
, uint_t flags
)
1049 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1053 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1054 ASSERT3U(tqs
->stqs_count
, !=, 0);
1056 if (tqs
->stqs_count
== 1) {
1057 tq
= tqs
->stqs_taskq
[0];
1059 tq
= tqs
->stqs_taskq
[((uint64_t)gethrtime()) % tqs
->stqs_count
];
1062 id
= taskq_dispatch(tq
, func
, arg
, flags
);
1064 taskq_wait_id(tq
, id
);
1068 spa_create_zio_taskqs(spa_t
*spa
)
1070 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1071 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1072 spa_taskqs_init(spa
, t
, q
);
1078 * Disabled until spa_thread() can be adapted for Linux.
1080 #undef HAVE_SPA_THREAD
1082 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
1084 spa_thread(void *arg
)
1086 psetid_t zio_taskq_psrset_bind
= PS_NONE
;
1087 callb_cpr_t cprinfo
;
1090 user_t
*pu
= PTOU(curproc
);
1092 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1095 ASSERT(curproc
!= &p0
);
1096 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1097 "zpool-%s", spa
->spa_name
);
1098 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1100 /* bind this thread to the requested psrset */
1101 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1103 mutex_enter(&cpu_lock
);
1104 mutex_enter(&pidlock
);
1105 mutex_enter(&curproc
->p_lock
);
1107 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1108 0, NULL
, NULL
) == 0) {
1109 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1112 "Couldn't bind process for zfs pool \"%s\" to "
1113 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1116 mutex_exit(&curproc
->p_lock
);
1117 mutex_exit(&pidlock
);
1118 mutex_exit(&cpu_lock
);
1122 if (zio_taskq_sysdc
) {
1123 sysdc_thread_enter(curthread
, 100, 0);
1126 spa
->spa_proc
= curproc
;
1127 spa
->spa_did
= curthread
->t_did
;
1129 spa_create_zio_taskqs(spa
);
1131 mutex_enter(&spa
->spa_proc_lock
);
1132 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1134 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1135 cv_broadcast(&spa
->spa_proc_cv
);
1137 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1138 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1139 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1140 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1142 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1143 spa
->spa_proc_state
= SPA_PROC_GONE
;
1144 spa
->spa_proc
= &p0
;
1145 cv_broadcast(&spa
->spa_proc_cv
);
1146 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1148 mutex_enter(&curproc
->p_lock
);
1154 * Activate an uninitialized pool.
1157 spa_activate(spa_t
*spa
, int mode
)
1159 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1161 spa
->spa_state
= POOL_STATE_ACTIVE
;
1162 spa
->spa_mode
= mode
;
1164 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1165 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1167 /* Try to create a covering process */
1168 mutex_enter(&spa
->spa_proc_lock
);
1169 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1170 ASSERT(spa
->spa_proc
== &p0
);
1173 #ifdef HAVE_SPA_THREAD
1174 /* Only create a process if we're going to be around a while. */
1175 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1176 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1178 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1179 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1180 cv_wait(&spa
->spa_proc_cv
,
1181 &spa
->spa_proc_lock
);
1183 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1184 ASSERT(spa
->spa_proc
!= &p0
);
1185 ASSERT(spa
->spa_did
!= 0);
1189 "Couldn't create process for zfs pool \"%s\"\n",
1194 #endif /* HAVE_SPA_THREAD */
1195 mutex_exit(&spa
->spa_proc_lock
);
1197 /* If we didn't create a process, we need to create our taskqs. */
1198 if (spa
->spa_proc
== &p0
) {
1199 spa_create_zio_taskqs(spa
);
1202 for (size_t i
= 0; i
< TXG_SIZE
; i
++)
1203 spa
->spa_txg_zio
[i
] = zio_root(spa
, NULL
, NULL
, 0);
1205 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1206 offsetof(vdev_t
, vdev_config_dirty_node
));
1207 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1208 offsetof(objset_t
, os_evicting_node
));
1209 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1210 offsetof(vdev_t
, vdev_state_dirty_node
));
1212 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1213 offsetof(struct vdev
, vdev_txg_node
));
1215 avl_create(&spa
->spa_errlist_scrub
,
1216 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1217 offsetof(spa_error_entry_t
, se_avl
));
1218 avl_create(&spa
->spa_errlist_last
,
1219 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1220 offsetof(spa_error_entry_t
, se_avl
));
1222 spa_keystore_init(&spa
->spa_keystore
);
1225 * This taskq is used to perform zvol-minor-related tasks
1226 * asynchronously. This has several advantages, including easy
1227 * resolution of various deadlocks (zfsonlinux bug #3681).
1229 * The taskq must be single threaded to ensure tasks are always
1230 * processed in the order in which they were dispatched.
1232 * A taskq per pool allows one to keep the pools independent.
1233 * This way if one pool is suspended, it will not impact another.
1235 * The preferred location to dispatch a zvol minor task is a sync
1236 * task. In this context, there is easy access to the spa_t and minimal
1237 * error handling is required because the sync task must succeed.
1239 spa
->spa_zvol_taskq
= taskq_create("z_zvol", 1, defclsyspri
,
1243 * Taskq dedicated to prefetcher threads: this is used to prevent the
1244 * pool traverse code from monopolizing the global (and limited)
1245 * system_taskq by inappropriately scheduling long running tasks on it.
1247 spa
->spa_prefetch_taskq
= taskq_create("z_prefetch", boot_ncpus
,
1248 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1251 * The taskq to upgrade datasets in this pool. Currently used by
1252 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1254 spa
->spa_upgrade_taskq
= taskq_create("z_upgrade", boot_ncpus
,
1255 defclsyspri
, 1, INT_MAX
, TASKQ_DYNAMIC
);
1259 * Opposite of spa_activate().
1262 spa_deactivate(spa_t
*spa
)
1264 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1265 ASSERT(spa
->spa_dsl_pool
== NULL
);
1266 ASSERT(spa
->spa_root_vdev
== NULL
);
1267 ASSERT(spa
->spa_async_zio_root
== NULL
);
1268 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1270 spa_evicting_os_wait(spa
);
1272 if (spa
->spa_zvol_taskq
) {
1273 taskq_destroy(spa
->spa_zvol_taskq
);
1274 spa
->spa_zvol_taskq
= NULL
;
1277 if (spa
->spa_prefetch_taskq
) {
1278 taskq_destroy(spa
->spa_prefetch_taskq
);
1279 spa
->spa_prefetch_taskq
= NULL
;
1282 if (spa
->spa_upgrade_taskq
) {
1283 taskq_destroy(spa
->spa_upgrade_taskq
);
1284 spa
->spa_upgrade_taskq
= NULL
;
1287 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1289 list_destroy(&spa
->spa_config_dirty_list
);
1290 list_destroy(&spa
->spa_evicting_os_list
);
1291 list_destroy(&spa
->spa_state_dirty_list
);
1293 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
1295 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1296 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1297 spa_taskqs_fini(spa
, t
, q
);
1301 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1302 ASSERT3P(spa
->spa_txg_zio
[i
], !=, NULL
);
1303 VERIFY0(zio_wait(spa
->spa_txg_zio
[i
]));
1304 spa
->spa_txg_zio
[i
] = NULL
;
1307 metaslab_class_destroy(spa
->spa_normal_class
);
1308 spa
->spa_normal_class
= NULL
;
1310 metaslab_class_destroy(spa
->spa_log_class
);
1311 spa
->spa_log_class
= NULL
;
1314 * If this was part of an import or the open otherwise failed, we may
1315 * still have errors left in the queues. Empty them just in case.
1317 spa_errlog_drain(spa
);
1318 avl_destroy(&spa
->spa_errlist_scrub
);
1319 avl_destroy(&spa
->spa_errlist_last
);
1321 spa_keystore_fini(&spa
->spa_keystore
);
1323 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1325 mutex_enter(&spa
->spa_proc_lock
);
1326 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1327 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1328 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1329 cv_broadcast(&spa
->spa_proc_cv
);
1330 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1331 ASSERT(spa
->spa_proc
!= &p0
);
1332 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1334 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1335 spa
->spa_proc_state
= SPA_PROC_NONE
;
1337 ASSERT(spa
->spa_proc
== &p0
);
1338 mutex_exit(&spa
->spa_proc_lock
);
1341 * We want to make sure spa_thread() has actually exited the ZFS
1342 * module, so that the module can't be unloaded out from underneath
1345 if (spa
->spa_did
!= 0) {
1346 thread_join(spa
->spa_did
);
1352 * Verify a pool configuration, and construct the vdev tree appropriately. This
1353 * will create all the necessary vdevs in the appropriate layout, with each vdev
1354 * in the CLOSED state. This will prep the pool before open/creation/import.
1355 * All vdev validation is done by the vdev_alloc() routine.
1358 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1359 uint_t id
, int atype
)
1365 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1368 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1371 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1374 if (error
== ENOENT
)
1380 return (SET_ERROR(EINVAL
));
1383 for (int c
= 0; c
< children
; c
++) {
1385 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1393 ASSERT(*vdp
!= NULL
);
1399 * Opposite of spa_load().
1402 spa_unload(spa_t
*spa
)
1406 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1408 spa_load_note(spa
, "UNLOADING");
1413 spa_async_suspend(spa
);
1418 if (spa
->spa_sync_on
) {
1419 txg_sync_stop(spa
->spa_dsl_pool
);
1420 spa
->spa_sync_on
= B_FALSE
;
1424 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1425 * to call it earlier, before we wait for async i/o to complete.
1426 * This ensures that there is no async metaslab prefetching, by
1427 * calling taskq_wait(mg_taskq).
1429 if (spa
->spa_root_vdev
!= NULL
) {
1430 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1431 for (int c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++)
1432 vdev_metaslab_fini(spa
->spa_root_vdev
->vdev_child
[c
]);
1433 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1436 if (spa
->spa_mmp
.mmp_thread
)
1437 mmp_thread_stop(spa
);
1440 * Wait for any outstanding async I/O to complete.
1442 if (spa
->spa_async_zio_root
!= NULL
) {
1443 for (int i
= 0; i
< max_ncpus
; i
++)
1444 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1445 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1446 spa
->spa_async_zio_root
= NULL
;
1449 if (spa
->spa_vdev_removal
!= NULL
) {
1450 spa_vdev_removal_destroy(spa
->spa_vdev_removal
);
1451 spa
->spa_vdev_removal
= NULL
;
1454 if (spa
->spa_condense_zthr
!= NULL
) {
1455 ASSERT(!zthr_isrunning(spa
->spa_condense_zthr
));
1456 zthr_destroy(spa
->spa_condense_zthr
);
1457 spa
->spa_condense_zthr
= NULL
;
1460 spa_condense_fini(spa
);
1462 bpobj_close(&spa
->spa_deferred_bpobj
);
1464 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1469 if (spa
->spa_root_vdev
)
1470 vdev_free(spa
->spa_root_vdev
);
1471 ASSERT(spa
->spa_root_vdev
== NULL
);
1474 * Close the dsl pool.
1476 if (spa
->spa_dsl_pool
) {
1477 dsl_pool_close(spa
->spa_dsl_pool
);
1478 spa
->spa_dsl_pool
= NULL
;
1479 spa
->spa_meta_objset
= NULL
;
1485 * Drop and purge level 2 cache
1487 spa_l2cache_drop(spa
);
1489 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1490 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1491 if (spa
->spa_spares
.sav_vdevs
) {
1492 kmem_free(spa
->spa_spares
.sav_vdevs
,
1493 spa
->spa_spares
.sav_count
* sizeof (void *));
1494 spa
->spa_spares
.sav_vdevs
= NULL
;
1496 if (spa
->spa_spares
.sav_config
) {
1497 nvlist_free(spa
->spa_spares
.sav_config
);
1498 spa
->spa_spares
.sav_config
= NULL
;
1500 spa
->spa_spares
.sav_count
= 0;
1502 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1503 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1504 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1506 if (spa
->spa_l2cache
.sav_vdevs
) {
1507 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1508 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1509 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1511 if (spa
->spa_l2cache
.sav_config
) {
1512 nvlist_free(spa
->spa_l2cache
.sav_config
);
1513 spa
->spa_l2cache
.sav_config
= NULL
;
1515 spa
->spa_l2cache
.sav_count
= 0;
1517 spa
->spa_async_suspended
= 0;
1519 spa
->spa_indirect_vdevs_loaded
= B_FALSE
;
1521 if (spa
->spa_comment
!= NULL
) {
1522 spa_strfree(spa
->spa_comment
);
1523 spa
->spa_comment
= NULL
;
1526 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1530 * Load (or re-load) the current list of vdevs describing the active spares for
1531 * this pool. When this is called, we have some form of basic information in
1532 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1533 * then re-generate a more complete list including status information.
1536 spa_load_spares(spa_t
*spa
)
1543 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1546 * First, close and free any existing spare vdevs.
1548 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1549 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1551 /* Undo the call to spa_activate() below */
1552 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1553 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1554 spa_spare_remove(tvd
);
1559 if (spa
->spa_spares
.sav_vdevs
)
1560 kmem_free(spa
->spa_spares
.sav_vdevs
,
1561 spa
->spa_spares
.sav_count
* sizeof (void *));
1563 if (spa
->spa_spares
.sav_config
== NULL
)
1566 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1567 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1569 spa
->spa_spares
.sav_count
= (int)nspares
;
1570 spa
->spa_spares
.sav_vdevs
= NULL
;
1576 * Construct the array of vdevs, opening them to get status in the
1577 * process. For each spare, there is potentially two different vdev_t
1578 * structures associated with it: one in the list of spares (used only
1579 * for basic validation purposes) and one in the active vdev
1580 * configuration (if it's spared in). During this phase we open and
1581 * validate each vdev on the spare list. If the vdev also exists in the
1582 * active configuration, then we also mark this vdev as an active spare.
1584 spa
->spa_spares
.sav_vdevs
= kmem_zalloc(nspares
* sizeof (void *),
1586 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1587 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1588 VDEV_ALLOC_SPARE
) == 0);
1591 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1593 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1594 B_FALSE
)) != NULL
) {
1595 if (!tvd
->vdev_isspare
)
1599 * We only mark the spare active if we were successfully
1600 * able to load the vdev. Otherwise, importing a pool
1601 * with a bad active spare would result in strange
1602 * behavior, because multiple pool would think the spare
1603 * is actively in use.
1605 * There is a vulnerability here to an equally bizarre
1606 * circumstance, where a dead active spare is later
1607 * brought back to life (onlined or otherwise). Given
1608 * the rarity of this scenario, and the extra complexity
1609 * it adds, we ignore the possibility.
1611 if (!vdev_is_dead(tvd
))
1612 spa_spare_activate(tvd
);
1616 vd
->vdev_aux
= &spa
->spa_spares
;
1618 if (vdev_open(vd
) != 0)
1621 if (vdev_validate_aux(vd
) == 0)
1626 * Recompute the stashed list of spares, with status information
1629 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1630 DATA_TYPE_NVLIST_ARRAY
) == 0);
1632 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1634 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1635 spares
[i
] = vdev_config_generate(spa
,
1636 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1637 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1638 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1639 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1640 nvlist_free(spares
[i
]);
1641 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1645 * Load (or re-load) the current list of vdevs describing the active l2cache for
1646 * this pool. When this is called, we have some form of basic information in
1647 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1648 * then re-generate a more complete list including status information.
1649 * Devices which are already active have their details maintained, and are
1653 spa_load_l2cache(spa_t
*spa
)
1655 nvlist_t
**l2cache
= NULL
;
1657 int i
, j
, oldnvdevs
;
1659 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1660 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1662 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1664 oldvdevs
= sav
->sav_vdevs
;
1665 oldnvdevs
= sav
->sav_count
;
1666 sav
->sav_vdevs
= NULL
;
1669 if (sav
->sav_config
== NULL
) {
1675 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1676 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1677 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1680 * Process new nvlist of vdevs.
1682 for (i
= 0; i
< nl2cache
; i
++) {
1683 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1687 for (j
= 0; j
< oldnvdevs
; j
++) {
1689 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1691 * Retain previous vdev for add/remove ops.
1699 if (newvdevs
[i
] == NULL
) {
1703 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1704 VDEV_ALLOC_L2CACHE
) == 0);
1709 * Commit this vdev as an l2cache device,
1710 * even if it fails to open.
1712 spa_l2cache_add(vd
);
1717 spa_l2cache_activate(vd
);
1719 if (vdev_open(vd
) != 0)
1722 (void) vdev_validate_aux(vd
);
1724 if (!vdev_is_dead(vd
))
1725 l2arc_add_vdev(spa
, vd
);
1729 sav
->sav_vdevs
= newvdevs
;
1730 sav
->sav_count
= (int)nl2cache
;
1733 * Recompute the stashed list of l2cache devices, with status
1734 * information this time.
1736 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1737 DATA_TYPE_NVLIST_ARRAY
) == 0);
1739 if (sav
->sav_count
> 0)
1740 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *),
1742 for (i
= 0; i
< sav
->sav_count
; i
++)
1743 l2cache
[i
] = vdev_config_generate(spa
,
1744 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1745 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1746 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1750 * Purge vdevs that were dropped
1752 for (i
= 0; i
< oldnvdevs
; i
++) {
1757 ASSERT(vd
->vdev_isl2cache
);
1759 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1760 pool
!= 0ULL && l2arc_vdev_present(vd
))
1761 l2arc_remove_vdev(vd
);
1762 vdev_clear_stats(vd
);
1768 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1770 for (i
= 0; i
< sav
->sav_count
; i
++)
1771 nvlist_free(l2cache
[i
]);
1773 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1777 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1780 char *packed
= NULL
;
1785 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1789 nvsize
= *(uint64_t *)db
->db_data
;
1790 dmu_buf_rele(db
, FTAG
);
1792 packed
= vmem_alloc(nvsize
, KM_SLEEP
);
1793 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1796 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1797 vmem_free(packed
, nvsize
);
1803 * Concrete top-level vdevs that are not missing and are not logs. At every
1804 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1807 spa_healthy_core_tvds(spa_t
*spa
)
1809 vdev_t
*rvd
= spa
->spa_root_vdev
;
1812 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
1813 vdev_t
*vd
= rvd
->vdev_child
[i
];
1816 if (vdev_is_concrete(vd
) && !vdev_is_dead(vd
))
1824 * Checks to see if the given vdev could not be opened, in which case we post a
1825 * sysevent to notify the autoreplace code that the device has been removed.
1828 spa_check_removed(vdev_t
*vd
)
1830 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1831 spa_check_removed(vd
->vdev_child
[c
]);
1833 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1834 vdev_is_concrete(vd
)) {
1835 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1836 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
1841 spa_check_for_missing_logs(spa_t
*spa
)
1843 vdev_t
*rvd
= spa
->spa_root_vdev
;
1846 * If we're doing a normal import, then build up any additional
1847 * diagnostic information about missing log devices.
1848 * We'll pass this up to the user for further processing.
1850 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1851 nvlist_t
**child
, *nv
;
1854 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
*),
1856 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1858 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
1859 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1862 * We consider a device as missing only if it failed
1863 * to open (i.e. offline or faulted is not considered
1866 if (tvd
->vdev_islog
&&
1867 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
1868 child
[idx
++] = vdev_config_generate(spa
, tvd
,
1869 B_FALSE
, VDEV_CONFIG_MISSING
);
1874 fnvlist_add_nvlist_array(nv
,
1875 ZPOOL_CONFIG_CHILDREN
, child
, idx
);
1876 fnvlist_add_nvlist(spa
->spa_load_info
,
1877 ZPOOL_CONFIG_MISSING_DEVICES
, nv
);
1879 for (uint64_t i
= 0; i
< idx
; i
++)
1880 nvlist_free(child
[i
]);
1883 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1886 spa_load_failed(spa
, "some log devices are missing");
1887 vdev_dbgmsg_print_tree(rvd
, 2);
1888 return (SET_ERROR(ENXIO
));
1891 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
1892 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1894 if (tvd
->vdev_islog
&&
1895 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
1896 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1897 spa_load_note(spa
, "some log devices are "
1898 "missing, ZIL is dropped.");
1899 vdev_dbgmsg_print_tree(rvd
, 2);
1909 * Check for missing log devices
1912 spa_check_logs(spa_t
*spa
)
1914 boolean_t rv
= B_FALSE
;
1915 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1917 switch (spa
->spa_log_state
) {
1920 case SPA_LOG_MISSING
:
1921 /* need to recheck in case slog has been restored */
1922 case SPA_LOG_UNKNOWN
:
1923 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1924 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1926 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1933 spa_passivate_log(spa_t
*spa
)
1935 vdev_t
*rvd
= spa
->spa_root_vdev
;
1936 boolean_t slog_found
= B_FALSE
;
1938 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1940 if (!spa_has_slogs(spa
))
1943 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1944 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1945 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1947 if (tvd
->vdev_islog
) {
1948 metaslab_group_passivate(mg
);
1949 slog_found
= B_TRUE
;
1953 return (slog_found
);
1957 spa_activate_log(spa_t
*spa
)
1959 vdev_t
*rvd
= spa
->spa_root_vdev
;
1961 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1963 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1964 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1965 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1967 if (tvd
->vdev_islog
)
1968 metaslab_group_activate(mg
);
1973 spa_reset_logs(spa_t
*spa
)
1977 error
= dmu_objset_find(spa_name(spa
), zil_reset
,
1978 NULL
, DS_FIND_CHILDREN
);
1981 * We successfully offlined the log device, sync out the
1982 * current txg so that the "stubby" block can be removed
1985 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1991 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1993 for (int i
= 0; i
< sav
->sav_count
; i
++)
1994 spa_check_removed(sav
->sav_vdevs
[i
]);
1998 spa_claim_notify(zio_t
*zio
)
2000 spa_t
*spa
= zio
->io_spa
;
2005 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
2006 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
2007 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
2008 mutex_exit(&spa
->spa_props_lock
);
2011 typedef struct spa_load_error
{
2012 uint64_t sle_meta_count
;
2013 uint64_t sle_data_count
;
2017 spa_load_verify_done(zio_t
*zio
)
2019 blkptr_t
*bp
= zio
->io_bp
;
2020 spa_load_error_t
*sle
= zio
->io_private
;
2021 dmu_object_type_t type
= BP_GET_TYPE(bp
);
2022 int error
= zio
->io_error
;
2023 spa_t
*spa
= zio
->io_spa
;
2025 abd_free(zio
->io_abd
);
2027 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
2028 type
!= DMU_OT_INTENT_LOG
)
2029 atomic_inc_64(&sle
->sle_meta_count
);
2031 atomic_inc_64(&sle
->sle_data_count
);
2034 mutex_enter(&spa
->spa_scrub_lock
);
2035 spa
->spa_load_verify_ios
--;
2036 cv_broadcast(&spa
->spa_scrub_io_cv
);
2037 mutex_exit(&spa
->spa_scrub_lock
);
2041 * Maximum number of concurrent scrub i/os to create while verifying
2042 * a pool while importing it.
2044 int spa_load_verify_maxinflight
= 10000;
2045 int spa_load_verify_metadata
= B_TRUE
;
2046 int spa_load_verify_data
= B_TRUE
;
2050 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
2051 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
2053 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
2056 * Note: normally this routine will not be called if
2057 * spa_load_verify_metadata is not set. However, it may be useful
2058 * to manually set the flag after the traversal has begun.
2060 if (!spa_load_verify_metadata
)
2062 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
2066 size_t size
= BP_GET_PSIZE(bp
);
2068 mutex_enter(&spa
->spa_scrub_lock
);
2069 while (spa
->spa_load_verify_ios
>= spa_load_verify_maxinflight
)
2070 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
2071 spa
->spa_load_verify_ios
++;
2072 mutex_exit(&spa
->spa_scrub_lock
);
2074 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
2075 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
2076 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
2077 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
2083 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2085 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
2086 return (SET_ERROR(ENAMETOOLONG
));
2092 spa_load_verify(spa_t
*spa
)
2095 spa_load_error_t sle
= { 0 };
2096 zpool_load_policy_t policy
;
2097 boolean_t verify_ok
= B_FALSE
;
2100 zpool_get_load_policy(spa
->spa_config
, &policy
);
2102 if (policy
.zlp_rewind
& ZPOOL_NEVER_REWIND
)
2105 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
2106 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
2107 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
2109 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
2113 rio
= zio_root(spa
, NULL
, &sle
,
2114 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
2116 if (spa_load_verify_metadata
) {
2117 if (spa
->spa_extreme_rewind
) {
2118 spa_load_note(spa
, "performing a complete scan of the "
2119 "pool since extreme rewind is on. This may take "
2120 "a very long time.\n (spa_load_verify_data=%u, "
2121 "spa_load_verify_metadata=%u)",
2122 spa_load_verify_data
, spa_load_verify_metadata
);
2124 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2125 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
2126 TRAVERSE_NO_DECRYPT
, spa_load_verify_cb
, rio
);
2129 (void) zio_wait(rio
);
2131 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2132 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2134 if (sle
.sle_meta_count
!= 0 || sle
.sle_data_count
!= 0) {
2135 spa_load_note(spa
, "spa_load_verify found %llu metadata errors "
2136 "and %llu data errors", (u_longlong_t
)sle
.sle_meta_count
,
2137 (u_longlong_t
)sle
.sle_data_count
);
2140 if (spa_load_verify_dryrun
||
2141 (!error
&& sle
.sle_meta_count
<= policy
.zlp_maxmeta
&&
2142 sle
.sle_data_count
<= policy
.zlp_maxdata
)) {
2146 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2147 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2149 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2150 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2151 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2152 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2153 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2154 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2155 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2157 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2160 if (spa_load_verify_dryrun
)
2164 if (error
!= ENXIO
&& error
!= EIO
)
2165 error
= SET_ERROR(EIO
);
2169 return (verify_ok
? 0 : EIO
);
2173 * Find a value in the pool props object.
2176 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2178 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2179 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2183 * Find a value in the pool directory object.
2186 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
, boolean_t log_enoent
)
2188 int error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2189 name
, sizeof (uint64_t), 1, val
);
2191 if (error
!= 0 && (error
!= ENOENT
|| log_enoent
)) {
2192 spa_load_failed(spa
, "couldn't get '%s' value in MOS directory "
2193 "[error=%d]", name
, error
);
2200 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2202 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2203 return (SET_ERROR(err
));
2207 spa_spawn_aux_threads(spa_t
*spa
)
2209 ASSERT(spa_writeable(spa
));
2211 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2213 spa_start_indirect_condensing_thread(spa
);
2217 * Fix up config after a partly-completed split. This is done with the
2218 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2219 * pool have that entry in their config, but only the splitting one contains
2220 * a list of all the guids of the vdevs that are being split off.
2222 * This function determines what to do with that list: either rejoin
2223 * all the disks to the pool, or complete the splitting process. To attempt
2224 * the rejoin, each disk that is offlined is marked online again, and
2225 * we do a reopen() call. If the vdev label for every disk that was
2226 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2227 * then we call vdev_split() on each disk, and complete the split.
2229 * Otherwise we leave the config alone, with all the vdevs in place in
2230 * the original pool.
2233 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2240 boolean_t attempt_reopen
;
2242 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2245 /* check that the config is complete */
2246 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2247 &glist
, &gcount
) != 0)
2250 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2252 /* attempt to online all the vdevs & validate */
2253 attempt_reopen
= B_TRUE
;
2254 for (i
= 0; i
< gcount
; i
++) {
2255 if (glist
[i
] == 0) /* vdev is hole */
2258 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2259 if (vd
[i
] == NULL
) {
2261 * Don't bother attempting to reopen the disks;
2262 * just do the split.
2264 attempt_reopen
= B_FALSE
;
2266 /* attempt to re-online it */
2267 vd
[i
]->vdev_offline
= B_FALSE
;
2271 if (attempt_reopen
) {
2272 vdev_reopen(spa
->spa_root_vdev
);
2274 /* check each device to see what state it's in */
2275 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2276 if (vd
[i
] != NULL
&&
2277 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2284 * If every disk has been moved to the new pool, or if we never
2285 * even attempted to look at them, then we split them off for
2288 if (!attempt_reopen
|| gcount
== extracted
) {
2289 for (i
= 0; i
< gcount
; i
++)
2292 vdev_reopen(spa
->spa_root_vdev
);
2295 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2299 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
)
2301 char *ereport
= FM_EREPORT_ZFS_POOL
;
2304 spa
->spa_load_state
= state
;
2306 gethrestime(&spa
->spa_loaded_ts
);
2307 error
= spa_load_impl(spa
, type
, &ereport
, B_FALSE
);
2310 * Don't count references from objsets that are already closed
2311 * and are making their way through the eviction process.
2313 spa_evicting_os_wait(spa
);
2314 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2316 if (error
!= EEXIST
) {
2317 spa
->spa_loaded_ts
.tv_sec
= 0;
2318 spa
->spa_loaded_ts
.tv_nsec
= 0;
2320 if (error
!= EBADF
) {
2321 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, NULL
, 0, 0);
2324 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2332 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2333 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2334 * spa's per-vdev ZAP list.
2337 vdev_count_verify_zaps(vdev_t
*vd
)
2339 spa_t
*spa
= vd
->vdev_spa
;
2342 if (vd
->vdev_top_zap
!= 0) {
2344 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2345 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2347 if (vd
->vdev_leaf_zap
!= 0) {
2349 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2350 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2353 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2354 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2362 * Determine whether the activity check is required.
2365 spa_activity_check_required(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*label
,
2369 uint64_t hostid
= 0;
2370 uint64_t tryconfig_txg
= 0;
2371 uint64_t tryconfig_timestamp
= 0;
2374 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2375 nvinfo
= fnvlist_lookup_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
);
2376 (void) nvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
,
2378 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
2379 &tryconfig_timestamp
);
2382 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_STATE
, &state
);
2385 * Disable the MMP activity check - This is used by zdb which
2386 * is intended to be used on potentially active pools.
2388 if (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
)
2392 * Skip the activity check when the MMP feature is disabled.
2394 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
== 0)
2397 * If the tryconfig_* values are nonzero, they are the results of an
2398 * earlier tryimport. If they match the uberblock we just found, then
2399 * the pool has not changed and we return false so we do not test a
2402 if (tryconfig_txg
&& tryconfig_txg
== ub
->ub_txg
&&
2403 tryconfig_timestamp
&& tryconfig_timestamp
== ub
->ub_timestamp
)
2407 * Allow the activity check to be skipped when importing the pool
2408 * on the same host which last imported it. Since the hostid from
2409 * configuration may be stale use the one read from the label.
2411 if (nvlist_exists(label
, ZPOOL_CONFIG_HOSTID
))
2412 hostid
= fnvlist_lookup_uint64(label
, ZPOOL_CONFIG_HOSTID
);
2414 if (hostid
== spa_get_hostid())
2418 * Skip the activity test when the pool was cleanly exported.
2420 if (state
!= POOL_STATE_ACTIVE
)
2427 * Perform the import activity check. If the user canceled the import or
2428 * we detected activity then fail.
2431 spa_activity_check(spa_t
*spa
, uberblock_t
*ub
, nvlist_t
*config
)
2433 uint64_t import_intervals
= MAX(zfs_multihost_import_intervals
, 1);
2434 uint64_t txg
= ub
->ub_txg
;
2435 uint64_t timestamp
= ub
->ub_timestamp
;
2436 uint64_t import_delay
= NANOSEC
;
2437 hrtime_t import_expire
;
2438 nvlist_t
*mmp_label
= NULL
;
2439 vdev_t
*rvd
= spa
->spa_root_vdev
;
2444 cv_init(&cv
, NULL
, CV_DEFAULT
, NULL
);
2445 mutex_init(&mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2449 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2450 * during the earlier tryimport. If the txg recorded there is 0 then
2451 * the pool is known to be active on another host.
2453 * Otherwise, the pool might be in use on another node. Check for
2454 * changes in the uberblocks on disk if necessary.
2456 if (nvlist_exists(config
, ZPOOL_CONFIG_LOAD_INFO
)) {
2457 nvlist_t
*nvinfo
= fnvlist_lookup_nvlist(config
,
2458 ZPOOL_CONFIG_LOAD_INFO
);
2460 if (nvlist_exists(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) &&
2461 fnvlist_lookup_uint64(nvinfo
, ZPOOL_CONFIG_MMP_TXG
) == 0) {
2462 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2463 error
= SET_ERROR(EREMOTEIO
);
2469 * Preferentially use the zfs_multihost_interval from the node which
2470 * last imported the pool. This value is stored in an MMP uberblock as.
2472 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2474 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
)
2475 import_delay
= MAX(import_delay
, import_intervals
*
2476 ub
->ub_mmp_delay
* MAX(vdev_count_leaves(spa
), 1));
2478 /* Apply a floor using the local default values. */
2479 import_delay
= MAX(import_delay
, import_intervals
*
2480 MSEC2NSEC(MAX(zfs_multihost_interval
, MMP_MIN_INTERVAL
)));
2482 zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu import_intervals=%u "
2483 "leaves=%u", import_delay
, ub
->ub_mmp_delay
, import_intervals
,
2484 vdev_count_leaves(spa
));
2486 /* Add a small random factor in case of simultaneous imports (0-25%) */
2487 import_expire
= gethrtime() + import_delay
+
2488 (import_delay
* spa_get_random(250) / 1000);
2490 while (gethrtime() < import_expire
) {
2491 vdev_uberblock_load(rvd
, ub
, &mmp_label
);
2493 if (txg
!= ub
->ub_txg
|| timestamp
!= ub
->ub_timestamp
) {
2494 error
= SET_ERROR(EREMOTEIO
);
2499 nvlist_free(mmp_label
);
2503 error
= cv_timedwait_sig(&cv
, &mtx
, ddi_get_lbolt() + hz
);
2505 error
= SET_ERROR(EINTR
);
2513 mutex_destroy(&mtx
);
2517 * If the pool is determined to be active store the status in the
2518 * spa->spa_load_info nvlist. If the remote hostname or hostid are
2519 * available from configuration read from disk store them as well.
2520 * This allows 'zpool import' to generate a more useful message.
2522 * ZPOOL_CONFIG_MMP_STATE - observed pool status (mandatory)
2523 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2524 * ZPOOL_CONFIG_MMP_HOSTID - hostid from the active pool
2526 if (error
== EREMOTEIO
) {
2527 char *hostname
= "<unknown>";
2528 uint64_t hostid
= 0;
2531 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTNAME
)) {
2532 hostname
= fnvlist_lookup_string(mmp_label
,
2533 ZPOOL_CONFIG_HOSTNAME
);
2534 fnvlist_add_string(spa
->spa_load_info
,
2535 ZPOOL_CONFIG_MMP_HOSTNAME
, hostname
);
2538 if (nvlist_exists(mmp_label
, ZPOOL_CONFIG_HOSTID
)) {
2539 hostid
= fnvlist_lookup_uint64(mmp_label
,
2540 ZPOOL_CONFIG_HOSTID
);
2541 fnvlist_add_uint64(spa
->spa_load_info
,
2542 ZPOOL_CONFIG_MMP_HOSTID
, hostid
);
2546 fnvlist_add_uint64(spa
->spa_load_info
,
2547 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_ACTIVE
);
2548 fnvlist_add_uint64(spa
->spa_load_info
,
2549 ZPOOL_CONFIG_MMP_TXG
, 0);
2551 error
= spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
);
2555 nvlist_free(mmp_label
);
2561 spa_verify_host(spa_t
*spa
, nvlist_t
*mos_config
)
2565 uint64_t myhostid
= 0;
2567 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
2568 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2569 hostname
= fnvlist_lookup_string(mos_config
,
2570 ZPOOL_CONFIG_HOSTNAME
);
2572 myhostid
= zone_get_hostid(NULL
);
2574 if (hostid
!= 0 && myhostid
!= 0 && hostid
!= myhostid
) {
2575 cmn_err(CE_WARN
, "pool '%s' could not be "
2576 "loaded as it was last accessed by "
2577 "another system (host: %s hostid: 0x%llx). "
2578 "See: http://illumos.org/msg/ZFS-8000-EY",
2579 spa_name(spa
), hostname
, (u_longlong_t
)hostid
);
2580 spa_load_failed(spa
, "hostid verification failed: pool "
2581 "last accessed by host: %s (hostid: 0x%llx)",
2582 hostname
, (u_longlong_t
)hostid
);
2583 return (SET_ERROR(EBADF
));
2591 spa_ld_parse_config(spa_t
*spa
, spa_import_type_t type
)
2594 nvlist_t
*nvtree
, *nvl
, *config
= spa
->spa_config
;
2601 * Versioning wasn't explicitly added to the label until later, so if
2602 * it's not present treat it as the initial version.
2604 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2605 &spa
->spa_ubsync
.ub_version
) != 0)
2606 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2608 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
2609 spa_load_failed(spa
, "invalid config provided: '%s' missing",
2610 ZPOOL_CONFIG_POOL_GUID
);
2611 return (SET_ERROR(EINVAL
));
2614 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
|| spa
->spa_load_state
==
2615 SPA_LOAD_TRYIMPORT
) && spa_guid_exists(pool_guid
, 0)) {
2616 spa_load_failed(spa
, "a pool with guid %llu is already open",
2617 (u_longlong_t
)pool_guid
);
2618 return (SET_ERROR(EEXIST
));
2621 spa
->spa_config_guid
= pool_guid
;
2623 nvlist_free(spa
->spa_load_info
);
2624 spa
->spa_load_info
= fnvlist_alloc();
2626 ASSERT(spa
->spa_comment
== NULL
);
2627 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2628 spa
->spa_comment
= spa_strdup(comment
);
2630 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2631 &spa
->spa_config_txg
);
2633 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) == 0)
2634 spa
->spa_config_splitting
= fnvlist_dup(nvl
);
2636 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
)) {
2637 spa_load_failed(spa
, "invalid config provided: '%s' missing",
2638 ZPOOL_CONFIG_VDEV_TREE
);
2639 return (SET_ERROR(EINVAL
));
2643 * Create "The Godfather" zio to hold all async IOs
2645 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2647 for (int i
= 0; i
< max_ncpus
; i
++) {
2648 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2649 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2650 ZIO_FLAG_GODFATHER
);
2654 * Parse the configuration into a vdev tree. We explicitly set the
2655 * value that will be returned by spa_version() since parsing the
2656 * configuration requires knowing the version number.
2658 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2659 parse
= (type
== SPA_IMPORT_EXISTING
?
2660 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2661 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
2662 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2665 spa_load_failed(spa
, "unable to parse config [error=%d]",
2670 ASSERT(spa
->spa_root_vdev
== rvd
);
2671 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2672 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2674 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2675 ASSERT(spa_guid(spa
) == pool_guid
);
2682 * Recursively open all vdevs in the vdev tree. This function is called twice:
2683 * first with the untrusted config, then with the trusted config.
2686 spa_ld_open_vdevs(spa_t
*spa
)
2691 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2692 * missing/unopenable for the root vdev to be still considered openable.
2694 if (spa
->spa_trust_config
) {
2695 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds
;
2696 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_CACHEFILE
) {
2697 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_cachefile
;
2698 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_SCAN
) {
2699 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_scan
;
2701 spa
->spa_missing_tvds_allowed
= 0;
2704 spa
->spa_missing_tvds_allowed
=
2705 MAX(zfs_max_missing_tvds
, spa
->spa_missing_tvds_allowed
);
2707 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2708 error
= vdev_open(spa
->spa_root_vdev
);
2709 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2711 if (spa
->spa_missing_tvds
!= 0) {
2712 spa_load_note(spa
, "vdev tree has %lld missing top-level "
2713 "vdevs.", (u_longlong_t
)spa
->spa_missing_tvds
);
2714 if (spa
->spa_trust_config
&& (spa
->spa_mode
& FWRITE
)) {
2716 * Although theoretically we could allow users to open
2717 * incomplete pools in RW mode, we'd need to add a lot
2718 * of extra logic (e.g. adjust pool space to account
2719 * for missing vdevs).
2720 * This limitation also prevents users from accidentally
2721 * opening the pool in RW mode during data recovery and
2722 * damaging it further.
2724 spa_load_note(spa
, "pools with missing top-level "
2725 "vdevs can only be opened in read-only mode.");
2726 error
= SET_ERROR(ENXIO
);
2728 spa_load_note(spa
, "current settings allow for maximum "
2729 "%lld missing top-level vdevs at this stage.",
2730 (u_longlong_t
)spa
->spa_missing_tvds_allowed
);
2734 spa_load_failed(spa
, "unable to open vdev tree [error=%d]",
2737 if (spa
->spa_missing_tvds
!= 0 || error
!= 0)
2738 vdev_dbgmsg_print_tree(spa
->spa_root_vdev
, 2);
2744 * We need to validate the vdev labels against the configuration that
2745 * we have in hand. This function is called twice: first with an untrusted
2746 * config, then with a trusted config. The validation is more strict when the
2747 * config is trusted.
2750 spa_ld_validate_vdevs(spa_t
*spa
)
2753 vdev_t
*rvd
= spa
->spa_root_vdev
;
2755 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2756 error
= vdev_validate(rvd
);
2757 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2760 spa_load_failed(spa
, "vdev_validate failed [error=%d]", error
);
2764 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
2765 spa_load_failed(spa
, "cannot open vdev tree after invalidating "
2767 vdev_dbgmsg_print_tree(rvd
, 2);
2768 return (SET_ERROR(ENXIO
));
2775 spa_ld_select_uberblock(spa_t
*spa
, spa_import_type_t type
)
2777 vdev_t
*rvd
= spa
->spa_root_vdev
;
2779 uberblock_t
*ub
= &spa
->spa_uberblock
;
2780 boolean_t activity_check
= B_FALSE
;
2783 * Find the best uberblock.
2785 vdev_uberblock_load(rvd
, ub
, &label
);
2788 * If we weren't able to find a single valid uberblock, return failure.
2790 if (ub
->ub_txg
== 0) {
2792 spa_load_failed(spa
, "no valid uberblock found");
2793 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2796 spa_load_note(spa
, "using uberblock with txg=%llu",
2797 (u_longlong_t
)ub
->ub_txg
);
2801 * For pools which have the multihost property on determine if the
2802 * pool is truly inactive and can be safely imported. Prevent
2803 * hosts which don't have a hostid set from importing the pool.
2805 activity_check
= spa_activity_check_required(spa
, ub
, label
,
2807 if (activity_check
) {
2808 if (ub
->ub_mmp_magic
== MMP_MAGIC
&& ub
->ub_mmp_delay
&&
2809 spa_get_hostid() == 0) {
2811 fnvlist_add_uint64(spa
->spa_load_info
,
2812 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
2813 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
2816 int error
= spa_activity_check(spa
, ub
, spa
->spa_config
);
2822 fnvlist_add_uint64(spa
->spa_load_info
,
2823 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_INACTIVE
);
2824 fnvlist_add_uint64(spa
->spa_load_info
,
2825 ZPOOL_CONFIG_MMP_TXG
, ub
->ub_txg
);
2829 * If the pool has an unsupported version we can't open it.
2831 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2833 spa_load_failed(spa
, "version %llu is not supported",
2834 (u_longlong_t
)ub
->ub_version
);
2835 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2838 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2842 * If we weren't able to find what's necessary for reading the
2843 * MOS in the label, return failure.
2845 if (label
== NULL
) {
2846 spa_load_failed(spa
, "label config unavailable");
2847 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2851 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
2854 spa_load_failed(spa
, "invalid label: '%s' missing",
2855 ZPOOL_CONFIG_FEATURES_FOR_READ
);
2856 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2861 * Update our in-core representation with the definitive values
2864 nvlist_free(spa
->spa_label_features
);
2865 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2871 * Look through entries in the label nvlist's features_for_read. If
2872 * there is a feature listed there which we don't understand then we
2873 * cannot open a pool.
2875 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2876 nvlist_t
*unsup_feat
;
2878 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2881 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
2883 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2884 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2885 VERIFY(nvlist_add_string(unsup_feat
,
2886 nvpair_name(nvp
), "") == 0);
2890 if (!nvlist_empty(unsup_feat
)) {
2891 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2892 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2893 nvlist_free(unsup_feat
);
2894 spa_load_failed(spa
, "some features are unsupported");
2895 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2899 nvlist_free(unsup_feat
);
2902 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2903 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2904 spa_try_repair(spa
, spa
->spa_config
);
2905 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2906 nvlist_free(spa
->spa_config_splitting
);
2907 spa
->spa_config_splitting
= NULL
;
2911 * Initialize internal SPA structures.
2913 spa
->spa_state
= POOL_STATE_ACTIVE
;
2914 spa
->spa_ubsync
= spa
->spa_uberblock
;
2915 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2916 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2917 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2918 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2919 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2920 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2926 spa_ld_open_rootbp(spa_t
*spa
)
2929 vdev_t
*rvd
= spa
->spa_root_vdev
;
2931 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2933 spa_load_failed(spa
, "unable to open rootbp in dsl_pool_init "
2934 "[error=%d]", error
);
2935 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2937 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2943 spa_ld_load_trusted_config(spa_t
*spa
, spa_import_type_t type
,
2944 boolean_t reloading
)
2946 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
2947 nvlist_t
*nv
, *mos_config
, *policy
;
2948 int error
= 0, copy_error
;
2949 uint64_t healthy_tvds
, healthy_tvds_mos
;
2950 uint64_t mos_config_txg
;
2952 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
, B_TRUE
)
2954 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2957 * If we're assembling a pool from a split, the config provided is
2958 * already trusted so there is nothing to do.
2960 if (type
== SPA_IMPORT_ASSEMBLE
)
2963 healthy_tvds
= spa_healthy_core_tvds(spa
);
2965 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
)
2967 spa_load_failed(spa
, "unable to retrieve MOS config");
2968 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2972 * If we are doing an open, pool owner wasn't verified yet, thus do
2973 * the verification here.
2975 if (spa
->spa_load_state
== SPA_LOAD_OPEN
) {
2976 error
= spa_verify_host(spa
, mos_config
);
2978 nvlist_free(mos_config
);
2983 nv
= fnvlist_lookup_nvlist(mos_config
, ZPOOL_CONFIG_VDEV_TREE
);
2985 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2988 * Build a new vdev tree from the trusted config
2990 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
2993 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2994 * obtained by scanning /dev/dsk, then it will have the right vdev
2995 * paths. We update the trusted MOS config with this information.
2996 * We first try to copy the paths with vdev_copy_path_strict, which
2997 * succeeds only when both configs have exactly the same vdev tree.
2998 * If that fails, we fall back to a more flexible method that has a
2999 * best effort policy.
3001 copy_error
= vdev_copy_path_strict(rvd
, mrvd
);
3002 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3003 spa_load_note(spa
, "provided vdev tree:");
3004 vdev_dbgmsg_print_tree(rvd
, 2);
3005 spa_load_note(spa
, "MOS vdev tree:");
3006 vdev_dbgmsg_print_tree(mrvd
, 2);
3008 if (copy_error
!= 0) {
3009 spa_load_note(spa
, "vdev_copy_path_strict failed, falling "
3010 "back to vdev_copy_path_relaxed");
3011 vdev_copy_path_relaxed(rvd
, mrvd
);
3016 spa
->spa_root_vdev
= mrvd
;
3018 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3021 * We will use spa_config if we decide to reload the spa or if spa_load
3022 * fails and we rewind. We must thus regenerate the config using the
3023 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3024 * pass settings on how to load the pool and is not stored in the MOS.
3025 * We copy it over to our new, trusted config.
3027 mos_config_txg
= fnvlist_lookup_uint64(mos_config
,
3028 ZPOOL_CONFIG_POOL_TXG
);
3029 nvlist_free(mos_config
);
3030 mos_config
= spa_config_generate(spa
, NULL
, mos_config_txg
, B_FALSE
);
3031 if (nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_LOAD_POLICY
,
3033 fnvlist_add_nvlist(mos_config
, ZPOOL_LOAD_POLICY
, policy
);
3034 spa_config_set(spa
, mos_config
);
3035 spa
->spa_config_source
= SPA_CONFIG_SRC_MOS
;
3038 * Now that we got the config from the MOS, we should be more strict
3039 * in checking blkptrs and can make assumptions about the consistency
3040 * of the vdev tree. spa_trust_config must be set to true before opening
3041 * vdevs in order for them to be writeable.
3043 spa
->spa_trust_config
= B_TRUE
;
3046 * Open and validate the new vdev tree
3048 error
= spa_ld_open_vdevs(spa
);
3052 error
= spa_ld_validate_vdevs(spa
);
3056 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
3057 spa_load_note(spa
, "final vdev tree:");
3058 vdev_dbgmsg_print_tree(rvd
, 2);
3061 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
&&
3062 !spa
->spa_extreme_rewind
&& zfs_max_missing_tvds
== 0) {
3064 * Sanity check to make sure that we are indeed loading the
3065 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3066 * in the config provided and they happened to be the only ones
3067 * to have the latest uberblock, we could involuntarily perform
3068 * an extreme rewind.
3070 healthy_tvds_mos
= spa_healthy_core_tvds(spa
);
3071 if (healthy_tvds_mos
- healthy_tvds
>=
3072 SPA_SYNC_MIN_VDEVS
) {
3073 spa_load_note(spa
, "config provided misses too many "
3074 "top-level vdevs compared to MOS (%lld vs %lld). ",
3075 (u_longlong_t
)healthy_tvds
,
3076 (u_longlong_t
)healthy_tvds_mos
);
3077 spa_load_note(spa
, "vdev tree:");
3078 vdev_dbgmsg_print_tree(rvd
, 2);
3080 spa_load_failed(spa
, "config was already "
3081 "provided from MOS. Aborting.");
3082 return (spa_vdev_err(rvd
,
3083 VDEV_AUX_CORRUPT_DATA
, EIO
));
3085 spa_load_note(spa
, "spa must be reloaded using MOS "
3087 return (SET_ERROR(EAGAIN
));
3091 error
= spa_check_for_missing_logs(spa
);
3093 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
3095 if (rvd
->vdev_guid_sum
!= spa
->spa_uberblock
.ub_guid_sum
) {
3096 spa_load_failed(spa
, "uberblock guid sum doesn't match MOS "
3097 "guid sum (%llu != %llu)",
3098 (u_longlong_t
)spa
->spa_uberblock
.ub_guid_sum
,
3099 (u_longlong_t
)rvd
->vdev_guid_sum
);
3100 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
3108 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
3111 vdev_t
*rvd
= spa
->spa_root_vdev
;
3114 * Everything that we read before spa_remove_init() must be stored
3115 * on concreted vdevs. Therefore we do this as early as possible.
3117 error
= spa_remove_init(spa
);
3119 spa_load_failed(spa
, "spa_remove_init failed [error=%d]",
3121 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3125 * Retrieve information needed to condense indirect vdev mappings.
3127 error
= spa_condense_init(spa
);
3129 spa_load_failed(spa
, "spa_condense_init failed [error=%d]",
3131 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3138 spa_ld_check_features(spa_t
*spa
, boolean_t
*missing_feat_writep
)
3141 vdev_t
*rvd
= spa
->spa_root_vdev
;
3143 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
3144 boolean_t missing_feat_read
= B_FALSE
;
3145 nvlist_t
*unsup_feat
, *enabled_feat
;
3147 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
3148 &spa
->spa_feat_for_read_obj
, B_TRUE
) != 0) {
3149 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3152 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
3153 &spa
->spa_feat_for_write_obj
, B_TRUE
) != 0) {
3154 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3157 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
3158 &spa
->spa_feat_desc_obj
, B_TRUE
) != 0) {
3159 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3162 enabled_feat
= fnvlist_alloc();
3163 unsup_feat
= fnvlist_alloc();
3165 if (!spa_features_check(spa
, B_FALSE
,
3166 unsup_feat
, enabled_feat
))
3167 missing_feat_read
= B_TRUE
;
3169 if (spa_writeable(spa
) ||
3170 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) {
3171 if (!spa_features_check(spa
, B_TRUE
,
3172 unsup_feat
, enabled_feat
)) {
3173 *missing_feat_writep
= B_TRUE
;
3177 fnvlist_add_nvlist(spa
->spa_load_info
,
3178 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
3180 if (!nvlist_empty(unsup_feat
)) {
3181 fnvlist_add_nvlist(spa
->spa_load_info
,
3182 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
3185 fnvlist_free(enabled_feat
);
3186 fnvlist_free(unsup_feat
);
3188 if (!missing_feat_read
) {
3189 fnvlist_add_boolean(spa
->spa_load_info
,
3190 ZPOOL_CONFIG_CAN_RDONLY
);
3194 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3195 * twofold: to determine whether the pool is available for
3196 * import in read-write mode and (if it is not) whether the
3197 * pool is available for import in read-only mode. If the pool
3198 * is available for import in read-write mode, it is displayed
3199 * as available in userland; if it is not available for import
3200 * in read-only mode, it is displayed as unavailable in
3201 * userland. If the pool is available for import in read-only
3202 * mode but not read-write mode, it is displayed as unavailable
3203 * in userland with a special note that the pool is actually
3204 * available for open in read-only mode.
3206 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3207 * missing a feature for write, we must first determine whether
3208 * the pool can be opened read-only before returning to
3209 * userland in order to know whether to display the
3210 * abovementioned note.
3212 if (missing_feat_read
|| (*missing_feat_writep
&&
3213 spa_writeable(spa
))) {
3214 spa_load_failed(spa
, "pool uses unsupported features");
3215 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
3220 * Load refcounts for ZFS features from disk into an in-memory
3221 * cache during SPA initialization.
3223 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
3226 error
= feature_get_refcount_from_disk(spa
,
3227 &spa_feature_table
[i
], &refcount
);
3229 spa
->spa_feat_refcount_cache
[i
] = refcount
;
3230 } else if (error
== ENOTSUP
) {
3231 spa
->spa_feat_refcount_cache
[i
] =
3232 SPA_FEATURE_DISABLED
;
3234 spa_load_failed(spa
, "error getting refcount "
3235 "for feature %s [error=%d]",
3236 spa_feature_table
[i
].fi_guid
, error
);
3237 return (spa_vdev_err(rvd
,
3238 VDEV_AUX_CORRUPT_DATA
, EIO
));
3243 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
3244 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
3245 &spa
->spa_feat_enabled_txg_obj
, B_TRUE
) != 0)
3246 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3253 spa_ld_load_special_directories(spa_t
*spa
)
3256 vdev_t
*rvd
= spa
->spa_root_vdev
;
3258 spa
->spa_is_initializing
= B_TRUE
;
3259 error
= dsl_pool_open(spa
->spa_dsl_pool
);
3260 spa
->spa_is_initializing
= B_FALSE
;
3262 spa_load_failed(spa
, "dsl_pool_open failed [error=%d]", error
);
3263 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3270 spa_ld_get_props(spa_t
*spa
)
3274 vdev_t
*rvd
= spa
->spa_root_vdev
;
3276 /* Grab the checksum salt from the MOS. */
3277 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
3278 DMU_POOL_CHECKSUM_SALT
, 1,
3279 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
3280 spa
->spa_cksum_salt
.zcs_bytes
);
3281 if (error
== ENOENT
) {
3282 /* Generate a new salt for subsequent use */
3283 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
3284 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
3285 } else if (error
!= 0) {
3286 spa_load_failed(spa
, "unable to retrieve checksum salt from "
3287 "MOS [error=%d]", error
);
3288 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3291 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
, B_TRUE
) != 0)
3292 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3293 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
3295 spa_load_failed(spa
, "error opening deferred-frees bpobj "
3296 "[error=%d]", error
);
3297 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3301 * Load the bit that tells us to use the new accounting function
3302 * (raid-z deflation). If we have an older pool, this will not
3305 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
, B_FALSE
);
3306 if (error
!= 0 && error
!= ENOENT
)
3307 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3309 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
3310 &spa
->spa_creation_version
, B_FALSE
);
3311 if (error
!= 0 && error
!= ENOENT
)
3312 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3315 * Load the persistent error log. If we have an older pool, this will
3318 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
,
3320 if (error
!= 0 && error
!= ENOENT
)
3321 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3323 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
3324 &spa
->spa_errlog_scrub
, B_FALSE
);
3325 if (error
!= 0 && error
!= ENOENT
)
3326 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3329 * Load the history object. If we have an older pool, this
3330 * will not be present.
3332 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
, B_FALSE
);
3333 if (error
!= 0 && error
!= ENOENT
)
3334 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3337 * Load the per-vdev ZAP map. If we have an older pool, this will not
3338 * be present; in this case, defer its creation to a later time to
3339 * avoid dirtying the MOS this early / out of sync context. See
3340 * spa_sync_config_object.
3343 /* The sentinel is only available in the MOS config. */
3344 nvlist_t
*mos_config
;
3345 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0) {
3346 spa_load_failed(spa
, "unable to retrieve MOS config");
3347 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3350 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
3351 &spa
->spa_all_vdev_zaps
, B_FALSE
);
3353 if (error
== ENOENT
) {
3354 VERIFY(!nvlist_exists(mos_config
,
3355 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
3356 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
3357 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3358 } else if (error
!= 0) {
3359 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3360 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
3362 * An older version of ZFS overwrote the sentinel value, so
3363 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3364 * destruction to later; see spa_sync_config_object.
3366 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
3368 * We're assuming that no vdevs have had their ZAPs created
3369 * before this. Better be sure of it.
3371 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3373 nvlist_free(mos_config
);
3375 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3377 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
,
3379 if (error
&& error
!= ENOENT
)
3380 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3383 uint64_t autoreplace
;
3385 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3386 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3387 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3388 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3389 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3390 spa_prop_find(spa
, ZPOOL_PROP_MULTIHOST
, &spa
->spa_multihost
);
3391 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3392 &spa
->spa_dedup_ditto
);
3394 spa
->spa_autoreplace
= (autoreplace
!= 0);
3398 * If we are importing a pool with missing top-level vdevs,
3399 * we enforce that the pool doesn't panic or get suspended on
3400 * error since the likelihood of missing data is extremely high.
3402 if (spa
->spa_missing_tvds
> 0 &&
3403 spa
->spa_failmode
!= ZIO_FAILURE_MODE_CONTINUE
&&
3404 spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
3405 spa_load_note(spa
, "forcing failmode to 'continue' "
3406 "as some top level vdevs are missing");
3407 spa
->spa_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
3414 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
3417 vdev_t
*rvd
= spa
->spa_root_vdev
;
3420 * If we're assembling the pool from the split-off vdevs of
3421 * an existing pool, we don't want to attach the spares & cache
3426 * Load any hot spares for this pool.
3428 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
,
3430 if (error
!= 0 && error
!= ENOENT
)
3431 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3432 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3433 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3434 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3435 &spa
->spa_spares
.sav_config
) != 0) {
3436 spa_load_failed(spa
, "error loading spares nvlist");
3437 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3440 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3441 spa_load_spares(spa
);
3442 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3443 } else if (error
== 0) {
3444 spa
->spa_spares
.sav_sync
= B_TRUE
;
3448 * Load any level 2 ARC devices for this pool.
3450 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3451 &spa
->spa_l2cache
.sav_object
, B_FALSE
);
3452 if (error
!= 0 && error
!= ENOENT
)
3453 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3454 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3455 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3456 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3457 &spa
->spa_l2cache
.sav_config
) != 0) {
3458 spa_load_failed(spa
, "error loading l2cache nvlist");
3459 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3462 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3463 spa_load_l2cache(spa
);
3464 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3465 } else if (error
== 0) {
3466 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3473 spa_ld_load_vdev_metadata(spa_t
*spa
)
3476 vdev_t
*rvd
= spa
->spa_root_vdev
;
3479 * If the 'multihost' property is set, then never allow a pool to
3480 * be imported when the system hostid is zero. The exception to
3481 * this rule is zdb which is always allowed to access pools.
3483 if (spa_multihost(spa
) && spa_get_hostid() == 0 &&
3484 (spa
->spa_import_flags
& ZFS_IMPORT_SKIP_MMP
) == 0) {
3485 fnvlist_add_uint64(spa
->spa_load_info
,
3486 ZPOOL_CONFIG_MMP_STATE
, MMP_STATE_NO_HOSTID
);
3487 return (spa_vdev_err(rvd
, VDEV_AUX_ACTIVE
, EREMOTEIO
));
3491 * If the 'autoreplace' property is set, then post a resource notifying
3492 * the ZFS DE that it should not issue any faults for unopenable
3493 * devices. We also iterate over the vdevs, and post a sysevent for any
3494 * unopenable vdevs so that the normal autoreplace handler can take
3497 if (spa
->spa_autoreplace
&& spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
3498 spa_check_removed(spa
->spa_root_vdev
);
3500 * For the import case, this is done in spa_import(), because
3501 * at this point we're using the spare definitions from
3502 * the MOS config, not necessarily from the userland config.
3504 if (spa
->spa_load_state
!= SPA_LOAD_IMPORT
) {
3505 spa_aux_check_removed(&spa
->spa_spares
);
3506 spa_aux_check_removed(&spa
->spa_l2cache
);
3511 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3513 error
= vdev_load(rvd
);
3515 spa_load_failed(spa
, "vdev_load failed [error=%d]", error
);
3516 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3520 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3522 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3523 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3524 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3530 spa_ld_load_dedup_tables(spa_t
*spa
)
3533 vdev_t
*rvd
= spa
->spa_root_vdev
;
3535 error
= ddt_load(spa
);
3537 spa_load_failed(spa
, "ddt_load failed [error=%d]", error
);
3538 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3545 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
3547 vdev_t
*rvd
= spa
->spa_root_vdev
;
3549 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
)) {
3550 boolean_t missing
= spa_check_logs(spa
);
3552 if (spa
->spa_missing_tvds
!= 0) {
3553 spa_load_note(spa
, "spa_check_logs failed "
3554 "so dropping the logs");
3556 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3557 spa_load_failed(spa
, "spa_check_logs failed");
3558 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
,
3568 spa_ld_verify_pool_data(spa_t
*spa
)
3571 vdev_t
*rvd
= spa
->spa_root_vdev
;
3574 * We've successfully opened the pool, verify that we're ready
3575 * to start pushing transactions.
3577 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
3578 error
= spa_load_verify(spa
);
3580 spa_load_failed(spa
, "spa_load_verify failed "
3581 "[error=%d]", error
);
3582 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3591 spa_ld_claim_log_blocks(spa_t
*spa
)
3594 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3597 * Claim log blocks that haven't been committed yet.
3598 * This must all happen in a single txg.
3599 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3600 * invoked from zil_claim_log_block()'s i/o done callback.
3601 * Price of rollback is that we abandon the log.
3603 spa
->spa_claiming
= B_TRUE
;
3605 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3606 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3607 zil_claim
, tx
, DS_FIND_CHILDREN
);
3610 spa
->spa_claiming
= B_FALSE
;
3612 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3616 spa_ld_check_for_config_update(spa_t
*spa
, uint64_t config_cache_txg
,
3617 boolean_t reloading
)
3619 vdev_t
*rvd
= spa
->spa_root_vdev
;
3620 int need_update
= B_FALSE
;
3623 * If the config cache is stale, or we have uninitialized
3624 * metaslabs (see spa_vdev_add()), then update the config.
3626 * If this is a verbatim import, trust the current
3627 * in-core spa_config and update the disk labels.
3629 if (reloading
|| config_cache_txg
!= spa
->spa_config_txg
||
3630 spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3631 spa
->spa_load_state
== SPA_LOAD_RECOVER
||
3632 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3633 need_update
= B_TRUE
;
3635 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
3636 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3637 need_update
= B_TRUE
;
3640 * Update the config cache asychronously in case we're the
3641 * root pool, in which case the config cache isn't writable yet.
3644 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3648 spa_ld_prepare_for_reload(spa_t
*spa
)
3650 int mode
= spa
->spa_mode
;
3651 int async_suspended
= spa
->spa_async_suspended
;
3654 spa_deactivate(spa
);
3655 spa_activate(spa
, mode
);
3658 * We save the value of spa_async_suspended as it gets reset to 0 by
3659 * spa_unload(). We want to restore it back to the original value before
3660 * returning as we might be calling spa_async_resume() later.
3662 spa
->spa_async_suspended
= async_suspended
;
3666 * Load an existing storage pool, using the config provided. This config
3667 * describes which vdevs are part of the pool and is later validated against
3668 * partial configs present in each vdev's label and an entire copy of the
3669 * config stored in the MOS.
3672 spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
,
3673 boolean_t reloading
)
3676 boolean_t missing_feat_write
= B_FALSE
;
3678 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
3679 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
3682 * Never trust the config that is provided unless we are assembling
3683 * a pool following a split.
3684 * This means don't trust blkptrs and the vdev tree in general. This
3685 * also effectively puts the spa in read-only mode since
3686 * spa_writeable() checks for spa_trust_config to be true.
3687 * We will later load a trusted config from the MOS.
3689 if (type
!= SPA_IMPORT_ASSEMBLE
)
3690 spa
->spa_trust_config
= B_FALSE
;
3693 spa_load_note(spa
, "RELOADING");
3695 spa_load_note(spa
, "LOADING");
3698 * Parse the config provided to create a vdev tree.
3700 error
= spa_ld_parse_config(spa
, type
);
3705 * Now that we have the vdev tree, try to open each vdev. This involves
3706 * opening the underlying physical device, retrieving its geometry and
3707 * probing the vdev with a dummy I/O. The state of each vdev will be set
3708 * based on the success of those operations. After this we'll be ready
3709 * to read from the vdevs.
3711 error
= spa_ld_open_vdevs(spa
);
3716 * Read the label of each vdev and make sure that the GUIDs stored
3717 * there match the GUIDs in the config provided.
3718 * If we're assembling a new pool that's been split off from an
3719 * existing pool, the labels haven't yet been updated so we skip
3720 * validation for now.
3722 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3723 error
= spa_ld_validate_vdevs(spa
);
3729 * Read vdev labels to find the best uberblock (i.e. latest, unless
3730 * spa_load_max_txg is set) and store it in spa_uberblock. We get the
3731 * list of features required to read blkptrs in the MOS from the vdev
3732 * label with the best uberblock and verify that our version of zfs
3733 * supports them all.
3735 error
= spa_ld_select_uberblock(spa
, type
);
3740 * Pass that uberblock to the dsl_pool layer which will open the root
3741 * blkptr. This blkptr points to the latest version of the MOS and will
3742 * allow us to read its contents.
3744 error
= spa_ld_open_rootbp(spa
);
3749 * Retrieve the trusted config stored in the MOS and use it to create
3750 * a new, exact version of the vdev tree, then reopen all vdevs.
3752 error
= spa_ld_load_trusted_config(spa
, type
, reloading
);
3753 if (error
== EAGAIN
) {
3756 * Redo the loading process with the trusted config if it is
3757 * too different from the untrusted config.
3759 spa_ld_prepare_for_reload(spa
);
3760 return (spa_load_impl(spa
, type
, ereport
, B_TRUE
));
3761 } else if (error
!= 0) {
3766 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3767 * from the pool and their contents were re-mapped to other vdevs. Note
3768 * that everything that we read before this step must have been
3769 * rewritten on concrete vdevs after the last device removal was
3770 * initiated. Otherwise we could be reading from indirect vdevs before
3771 * we have loaded their mappings.
3773 error
= spa_ld_open_indirect_vdev_metadata(spa
);
3778 * Retrieve the full list of active features from the MOS and check if
3779 * they are all supported.
3781 error
= spa_ld_check_features(spa
, &missing_feat_write
);
3786 * Load several special directories from the MOS needed by the dsl_pool
3789 error
= spa_ld_load_special_directories(spa
);
3794 * Retrieve pool properties from the MOS.
3796 error
= spa_ld_get_props(spa
);
3801 * Retrieve the list of auxiliary devices - cache devices and spares -
3804 error
= spa_ld_open_aux_vdevs(spa
, type
);
3809 * Load the metadata for all vdevs. Also check if unopenable devices
3810 * should be autoreplaced.
3812 error
= spa_ld_load_vdev_metadata(spa
);
3816 error
= spa_ld_load_dedup_tables(spa
);
3821 * Verify the logs now to make sure we don't have any unexpected errors
3822 * when we claim log blocks later.
3824 error
= spa_ld_verify_logs(spa
, type
, ereport
);
3828 if (missing_feat_write
) {
3829 ASSERT(spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
);
3832 * At this point, we know that we can open the pool in
3833 * read-only mode but not read-write mode. We now have enough
3834 * information and can return to userland.
3836 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
3841 * Traverse the last txgs to make sure the pool was left off in a safe
3842 * state. When performing an extreme rewind, we verify the whole pool,
3843 * which can take a very long time.
3845 error
= spa_ld_verify_pool_data(spa
);
3850 * Calculate the deflated space for the pool. This must be done before
3851 * we write anything to the pool because we'd need to update the space
3852 * accounting using the deflated sizes.
3854 spa_update_dspace(spa
);
3857 * We have now retrieved all the information we needed to open the
3858 * pool. If we are importing the pool in read-write mode, a few
3859 * additional steps must be performed to finish the import.
3861 if (spa_writeable(spa
) && (spa
->spa_load_state
== SPA_LOAD_RECOVER
||
3862 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3863 uint64_t config_cache_txg
= spa
->spa_config_txg
;
3865 ASSERT(spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
);
3868 * Traverse the ZIL and claim all blocks.
3870 spa_ld_claim_log_blocks(spa
);
3873 * Kick-off the syncing thread.
3875 spa
->spa_sync_on
= B_TRUE
;
3876 txg_sync_start(spa
->spa_dsl_pool
);
3877 mmp_thread_start(spa
);
3880 * Wait for all claims to sync. We sync up to the highest
3881 * claimed log block birth time so that claimed log blocks
3882 * don't appear to be from the future. spa_claim_max_txg
3883 * will have been set for us by ZIL traversal operations
3886 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3889 * Check if we need to request an update of the config. On the
3890 * next sync, we would update the config stored in vdev labels
3891 * and the cachefile (by default /etc/zfs/zpool.cache).
3893 spa_ld_check_for_config_update(spa
, config_cache_txg
,
3897 * Check all DTLs to see if anything needs resilvering.
3899 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3900 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
))
3901 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3904 * Log the fact that we booted up (so that we can detect if
3905 * we rebooted in the middle of an operation).
3907 spa_history_log_version(spa
, "open", NULL
);
3910 * Delete any inconsistent datasets.
3912 (void) dmu_objset_find(spa_name(spa
),
3913 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3916 * Clean up any stale temporary dataset userrefs.
3918 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3920 spa_restart_removal(spa
);
3922 spa_spawn_aux_threads(spa
);
3925 spa_load_note(spa
, "LOADED");
3931 spa_load_retry(spa_t
*spa
, spa_load_state_t state
)
3933 int mode
= spa
->spa_mode
;
3936 spa_deactivate(spa
);
3938 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3940 spa_activate(spa
, mode
);
3941 spa_async_suspend(spa
);
3943 spa_load_note(spa
, "spa_load_retry: rewind, max txg: %llu",
3944 (u_longlong_t
)spa
->spa_load_max_txg
);
3946 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
));
3950 * If spa_load() fails this function will try loading prior txg's. If
3951 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3952 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3953 * function will not rewind the pool and will return the same error as
3957 spa_load_best(spa_t
*spa
, spa_load_state_t state
, uint64_t max_request
,
3960 nvlist_t
*loadinfo
= NULL
;
3961 nvlist_t
*config
= NULL
;
3962 int load_error
, rewind_error
;
3963 uint64_t safe_rewind_txg
;
3966 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3967 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3968 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3970 spa
->spa_load_max_txg
= max_request
;
3971 if (max_request
!= UINT64_MAX
)
3972 spa
->spa_extreme_rewind
= B_TRUE
;
3975 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
);
3976 if (load_error
== 0)
3979 if (spa
->spa_root_vdev
!= NULL
)
3980 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3982 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3983 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3985 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3986 nvlist_free(config
);
3987 return (load_error
);
3990 if (state
== SPA_LOAD_RECOVER
) {
3991 /* Price of rolling back is discarding txgs, including log */
3992 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3995 * If we aren't rolling back save the load info from our first
3996 * import attempt so that we can restore it after attempting
3999 loadinfo
= spa
->spa_load_info
;
4000 spa
->spa_load_info
= fnvlist_alloc();
4003 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
4004 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
4005 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
4006 TXG_INITIAL
: safe_rewind_txg
;
4009 * Continue as long as we're finding errors, we're still within
4010 * the acceptable rewind range, and we're still finding uberblocks
4012 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
4013 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
4014 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
4015 spa
->spa_extreme_rewind
= B_TRUE
;
4016 rewind_error
= spa_load_retry(spa
, state
);
4019 spa
->spa_extreme_rewind
= B_FALSE
;
4020 spa
->spa_load_max_txg
= UINT64_MAX
;
4022 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
4023 spa_config_set(spa
, config
);
4025 nvlist_free(config
);
4027 if (state
== SPA_LOAD_RECOVER
) {
4028 ASSERT3P(loadinfo
, ==, NULL
);
4029 return (rewind_error
);
4031 /* Store the rewind info as part of the initial load info */
4032 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
4033 spa
->spa_load_info
);
4035 /* Restore the initial load info */
4036 fnvlist_free(spa
->spa_load_info
);
4037 spa
->spa_load_info
= loadinfo
;
4039 return (load_error
);
4046 * The import case is identical to an open except that the configuration is sent
4047 * down from userland, instead of grabbed from the configuration cache. For the
4048 * case of an open, the pool configuration will exist in the
4049 * POOL_STATE_UNINITIALIZED state.
4051 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4052 * the same time open the pool, without having to keep around the spa_t in some
4056 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
4060 spa_load_state_t state
= SPA_LOAD_OPEN
;
4062 int locked
= B_FALSE
;
4063 int firstopen
= B_FALSE
;
4068 * As disgusting as this is, we need to support recursive calls to this
4069 * function because dsl_dir_open() is called during spa_load(), and ends
4070 * up calling spa_open() again. The real fix is to figure out how to
4071 * avoid dsl_dir_open() calling this in the first place.
4073 if (MUTEX_NOT_HELD(&spa_namespace_lock
)) {
4074 mutex_enter(&spa_namespace_lock
);
4078 if ((spa
= spa_lookup(pool
)) == NULL
) {
4080 mutex_exit(&spa_namespace_lock
);
4081 return (SET_ERROR(ENOENT
));
4084 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
4085 zpool_load_policy_t policy
;
4089 zpool_get_load_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
4091 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
4092 state
= SPA_LOAD_RECOVER
;
4094 spa_activate(spa
, spa_mode_global
);
4096 if (state
!= SPA_LOAD_RECOVER
)
4097 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4098 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
4100 zfs_dbgmsg("spa_open_common: opening %s", pool
);
4101 error
= spa_load_best(spa
, state
, policy
.zlp_txg
,
4104 if (error
== EBADF
) {
4106 * If vdev_validate() returns failure (indicated by
4107 * EBADF), it indicates that one of the vdevs indicates
4108 * that the pool has been exported or destroyed. If
4109 * this is the case, the config cache is out of sync and
4110 * we should remove the pool from the namespace.
4113 spa_deactivate(spa
);
4114 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
4117 mutex_exit(&spa_namespace_lock
);
4118 return (SET_ERROR(ENOENT
));
4123 * We can't open the pool, but we still have useful
4124 * information: the state of each vdev after the
4125 * attempted vdev_open(). Return this to the user.
4127 if (config
!= NULL
&& spa
->spa_config
) {
4128 VERIFY(nvlist_dup(spa
->spa_config
, config
,
4130 VERIFY(nvlist_add_nvlist(*config
,
4131 ZPOOL_CONFIG_LOAD_INFO
,
4132 spa
->spa_load_info
) == 0);
4135 spa_deactivate(spa
);
4136 spa
->spa_last_open_failed
= error
;
4138 mutex_exit(&spa_namespace_lock
);
4144 spa_open_ref(spa
, tag
);
4147 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4150 * If we've recovered the pool, pass back any information we
4151 * gathered while doing the load.
4153 if (state
== SPA_LOAD_RECOVER
) {
4154 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
4155 spa
->spa_load_info
) == 0);
4159 spa
->spa_last_open_failed
= 0;
4160 spa
->spa_last_ubsync_txg
= 0;
4161 spa
->spa_load_txg
= 0;
4162 mutex_exit(&spa_namespace_lock
);
4166 zvol_create_minors(spa
, spa_name(spa
), B_TRUE
);
4174 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
4177 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
4181 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
4183 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
4187 * Lookup the given spa_t, incrementing the inject count in the process,
4188 * preventing it from being exported or destroyed.
4191 spa_inject_addref(char *name
)
4195 mutex_enter(&spa_namespace_lock
);
4196 if ((spa
= spa_lookup(name
)) == NULL
) {
4197 mutex_exit(&spa_namespace_lock
);
4200 spa
->spa_inject_ref
++;
4201 mutex_exit(&spa_namespace_lock
);
4207 spa_inject_delref(spa_t
*spa
)
4209 mutex_enter(&spa_namespace_lock
);
4210 spa
->spa_inject_ref
--;
4211 mutex_exit(&spa_namespace_lock
);
4215 * Add spares device information to the nvlist.
4218 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
4228 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
4230 if (spa
->spa_spares
.sav_count
== 0)
4233 VERIFY(nvlist_lookup_nvlist(config
,
4234 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
4235 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
4236 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
4238 VERIFY(nvlist_add_nvlist_array(nvroot
,
4239 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4240 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
4241 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
4244 * Go through and find any spares which have since been
4245 * repurposed as an active spare. If this is the case, update
4246 * their status appropriately.
4248 for (i
= 0; i
< nspares
; i
++) {
4249 VERIFY(nvlist_lookup_uint64(spares
[i
],
4250 ZPOOL_CONFIG_GUID
, &guid
) == 0);
4251 if (spa_spare_exists(guid
, &pool
, NULL
) &&
4253 VERIFY(nvlist_lookup_uint64_array(
4254 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
4255 (uint64_t **)&vs
, &vsc
) == 0);
4256 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
4257 vs
->vs_aux
= VDEV_AUX_SPARED
;
4264 * Add l2cache device information to the nvlist, including vdev stats.
4267 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
4270 uint_t i
, j
, nl2cache
;
4277 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
4279 if (spa
->spa_l2cache
.sav_count
== 0)
4282 VERIFY(nvlist_lookup_nvlist(config
,
4283 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
4284 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
4285 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
4286 if (nl2cache
!= 0) {
4287 VERIFY(nvlist_add_nvlist_array(nvroot
,
4288 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4289 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
4290 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
4293 * Update level 2 cache device stats.
4296 for (i
= 0; i
< nl2cache
; i
++) {
4297 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
4298 ZPOOL_CONFIG_GUID
, &guid
) == 0);
4301 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
4303 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
4304 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
4310 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
4311 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
4313 vdev_get_stats(vd
, vs
);
4314 vdev_config_generate_stats(vd
, l2cache
[i
]);
4321 spa_feature_stats_from_disk(spa_t
*spa
, nvlist_t
*features
)
4326 if (spa
->spa_feat_for_read_obj
!= 0) {
4327 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
4328 spa
->spa_feat_for_read_obj
);
4329 zap_cursor_retrieve(&zc
, &za
) == 0;
4330 zap_cursor_advance(&zc
)) {
4331 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
4332 za
.za_num_integers
== 1);
4333 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
4334 za
.za_first_integer
));
4336 zap_cursor_fini(&zc
);
4339 if (spa
->spa_feat_for_write_obj
!= 0) {
4340 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
4341 spa
->spa_feat_for_write_obj
);
4342 zap_cursor_retrieve(&zc
, &za
) == 0;
4343 zap_cursor_advance(&zc
)) {
4344 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
4345 za
.za_num_integers
== 1);
4346 VERIFY0(nvlist_add_uint64(features
, za
.za_name
,
4347 za
.za_first_integer
));
4349 zap_cursor_fini(&zc
);
4354 spa_feature_stats_from_cache(spa_t
*spa
, nvlist_t
*features
)
4358 for (i
= 0; i
< SPA_FEATURES
; i
++) {
4359 zfeature_info_t feature
= spa_feature_table
[i
];
4362 if (feature_get_refcount(spa
, &feature
, &refcount
) != 0)
4365 VERIFY0(nvlist_add_uint64(features
, feature
.fi_guid
, refcount
));
4370 * Store a list of pool features and their reference counts in the
4373 * The first time this is called on a spa, allocate a new nvlist, fetch
4374 * the pool features and reference counts from disk, then save the list
4375 * in the spa. In subsequent calls on the same spa use the saved nvlist
4376 * and refresh its values from the cached reference counts. This
4377 * ensures we don't block here on I/O on a suspended pool so 'zpool
4378 * clear' can resume the pool.
4381 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
4385 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
4387 mutex_enter(&spa
->spa_feat_stats_lock
);
4388 features
= spa
->spa_feat_stats
;
4390 if (features
!= NULL
) {
4391 spa_feature_stats_from_cache(spa
, features
);
4393 VERIFY0(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
));
4394 spa
->spa_feat_stats
= features
;
4395 spa_feature_stats_from_disk(spa
, features
);
4398 VERIFY0(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
4401 mutex_exit(&spa
->spa_feat_stats_lock
);
4405 spa_get_stats(const char *name
, nvlist_t
**config
,
4406 char *altroot
, size_t buflen
)
4412 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
4416 * This still leaves a window of inconsistency where the spares
4417 * or l2cache devices could change and the config would be
4418 * self-inconsistent.
4420 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4422 if (*config
!= NULL
) {
4423 uint64_t loadtimes
[2];
4425 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
4426 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
4427 VERIFY(nvlist_add_uint64_array(*config
,
4428 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
4430 VERIFY(nvlist_add_uint64(*config
,
4431 ZPOOL_CONFIG_ERRCOUNT
,
4432 spa_get_errlog_size(spa
)) == 0);
4434 if (spa_suspended(spa
)) {
4435 VERIFY(nvlist_add_uint64(*config
,
4436 ZPOOL_CONFIG_SUSPENDED
,
4437 spa
->spa_failmode
) == 0);
4438 VERIFY(nvlist_add_uint64(*config
,
4439 ZPOOL_CONFIG_SUSPENDED_REASON
,
4440 spa
->spa_suspended
) == 0);
4443 spa_add_spares(spa
, *config
);
4444 spa_add_l2cache(spa
, *config
);
4445 spa_add_feature_stats(spa
, *config
);
4450 * We want to get the alternate root even for faulted pools, so we cheat
4451 * and call spa_lookup() directly.
4455 mutex_enter(&spa_namespace_lock
);
4456 spa
= spa_lookup(name
);
4458 spa_altroot(spa
, altroot
, buflen
);
4462 mutex_exit(&spa_namespace_lock
);
4464 spa_altroot(spa
, altroot
, buflen
);
4469 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4470 spa_close(spa
, FTAG
);
4477 * Validate that the auxiliary device array is well formed. We must have an
4478 * array of nvlists, each which describes a valid leaf vdev. If this is an
4479 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4480 * specified, as long as they are well-formed.
4483 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
4484 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
4485 vdev_labeltype_t label
)
4492 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4495 * It's acceptable to have no devs specified.
4497 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
4501 return (SET_ERROR(EINVAL
));
4504 * Make sure the pool is formatted with a version that supports this
4507 if (spa_version(spa
) < version
)
4508 return (SET_ERROR(ENOTSUP
));
4511 * Set the pending device list so we correctly handle device in-use
4514 sav
->sav_pending
= dev
;
4515 sav
->sav_npending
= ndev
;
4517 for (i
= 0; i
< ndev
; i
++) {
4518 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
4522 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4524 error
= SET_ERROR(EINVAL
);
4530 if ((error
= vdev_open(vd
)) == 0 &&
4531 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
4532 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
4533 vd
->vdev_guid
) == 0);
4539 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
4546 sav
->sav_pending
= NULL
;
4547 sav
->sav_npending
= 0;
4552 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
4556 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4558 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
4559 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
4560 VDEV_LABEL_SPARE
)) != 0) {
4564 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
4565 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
4566 VDEV_LABEL_L2CACHE
));
4570 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
4575 if (sav
->sav_config
!= NULL
) {
4581 * Generate new dev list by concatenating with the
4584 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
4585 &olddevs
, &oldndevs
) == 0);
4587 newdevs
= kmem_alloc(sizeof (void *) *
4588 (ndevs
+ oldndevs
), KM_SLEEP
);
4589 for (i
= 0; i
< oldndevs
; i
++)
4590 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
4592 for (i
= 0; i
< ndevs
; i
++)
4593 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
4596 VERIFY(nvlist_remove(sav
->sav_config
, config
,
4597 DATA_TYPE_NVLIST_ARRAY
) == 0);
4599 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
4600 config
, newdevs
, ndevs
+ oldndevs
) == 0);
4601 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
4602 nvlist_free(newdevs
[i
]);
4603 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
4606 * Generate a new dev list.
4608 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
4610 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
4616 * Stop and drop level 2 ARC devices
4619 spa_l2cache_drop(spa_t
*spa
)
4623 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4625 for (i
= 0; i
< sav
->sav_count
; i
++) {
4628 vd
= sav
->sav_vdevs
[i
];
4631 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
4632 pool
!= 0ULL && l2arc_vdev_present(vd
))
4633 l2arc_remove_vdev(vd
);
4638 * Verify encryption parameters for spa creation. If we are encrypting, we must
4639 * have the encryption feature flag enabled.
4642 spa_create_check_encryption_params(dsl_crypto_params_t
*dcp
,
4643 boolean_t has_encryption
)
4645 if (dcp
->cp_crypt
!= ZIO_CRYPT_OFF
&&
4646 dcp
->cp_crypt
!= ZIO_CRYPT_INHERIT
&&
4648 return (SET_ERROR(ENOTSUP
));
4650 return (dmu_objset_create_crypt_check(NULL
, dcp
));
4657 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
4658 nvlist_t
*zplprops
, dsl_crypto_params_t
*dcp
)
4661 char *altroot
= NULL
;
4666 uint64_t txg
= TXG_INITIAL
;
4667 nvlist_t
**spares
, **l2cache
;
4668 uint_t nspares
, nl2cache
;
4669 uint64_t version
, obj
, root_dsobj
= 0;
4670 boolean_t has_features
;
4671 boolean_t has_encryption
;
4677 if (nvlist_lookup_string(props
, "tname", &poolname
) != 0)
4678 poolname
= (char *)pool
;
4681 * If this pool already exists, return failure.
4683 mutex_enter(&spa_namespace_lock
);
4684 if (spa_lookup(poolname
) != NULL
) {
4685 mutex_exit(&spa_namespace_lock
);
4686 return (SET_ERROR(EEXIST
));
4690 * Allocate a new spa_t structure.
4692 nvl
= fnvlist_alloc();
4693 fnvlist_add_string(nvl
, ZPOOL_CONFIG_POOL_NAME
, pool
);
4694 (void) nvlist_lookup_string(props
,
4695 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4696 spa
= spa_add(poolname
, nvl
, altroot
);
4698 spa_activate(spa
, spa_mode_global
);
4700 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4701 spa_deactivate(spa
);
4703 mutex_exit(&spa_namespace_lock
);
4708 * Temporary pool names should never be written to disk.
4710 if (poolname
!= pool
)
4711 spa
->spa_import_flags
|= ZFS_IMPORT_TEMP_NAME
;
4713 has_features
= B_FALSE
;
4714 has_encryption
= B_FALSE
;
4715 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
4716 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4717 if (zpool_prop_feature(nvpair_name(elem
))) {
4718 has_features
= B_TRUE
;
4720 feat_name
= strchr(nvpair_name(elem
), '@') + 1;
4721 VERIFY0(zfeature_lookup_name(feat_name
, &feat
));
4722 if (feat
== SPA_FEATURE_ENCRYPTION
)
4723 has_encryption
= B_TRUE
;
4727 /* verify encryption params, if they were provided */
4729 error
= spa_create_check_encryption_params(dcp
, has_encryption
);
4731 spa_deactivate(spa
);
4733 mutex_exit(&spa_namespace_lock
);
4738 if (has_features
|| nvlist_lookup_uint64(props
,
4739 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4740 version
= SPA_VERSION
;
4742 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4744 spa
->spa_first_txg
= txg
;
4745 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4746 spa
->spa_uberblock
.ub_version
= version
;
4747 spa
->spa_ubsync
= spa
->spa_uberblock
;
4748 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4749 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
4750 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
4751 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
4754 * Create "The Godfather" zio to hold all async IOs
4756 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4758 for (int i
= 0; i
< max_ncpus
; i
++) {
4759 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4760 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4761 ZIO_FLAG_GODFATHER
);
4765 * Create the root vdev.
4767 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4769 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4771 ASSERT(error
!= 0 || rvd
!= NULL
);
4772 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4774 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4775 error
= SET_ERROR(EINVAL
);
4778 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4779 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4780 VDEV_ALLOC_ADD
)) == 0) {
4781 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
4782 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4783 vdev_expand(rvd
->vdev_child
[c
], txg
);
4787 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4791 spa_deactivate(spa
);
4793 mutex_exit(&spa_namespace_lock
);
4798 * Get the list of spares, if specified.
4800 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4801 &spares
, &nspares
) == 0) {
4802 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4804 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4805 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4806 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4807 spa_load_spares(spa
);
4808 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4809 spa
->spa_spares
.sav_sync
= B_TRUE
;
4813 * Get the list of level 2 cache devices, if specified.
4815 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4816 &l2cache
, &nl2cache
) == 0) {
4817 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4818 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4819 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4820 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4821 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4822 spa_load_l2cache(spa
);
4823 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4824 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4827 spa
->spa_is_initializing
= B_TRUE
;
4828 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, dcp
, txg
);
4829 spa
->spa_is_initializing
= B_FALSE
;
4832 * Create DDTs (dedup tables).
4836 spa_update_dspace(spa
);
4838 tx
= dmu_tx_create_assigned(dp
, txg
);
4841 * Create the pool's history object.
4843 if (version
>= SPA_VERSION_ZPOOL_HISTORY
&& !spa
->spa_history
)
4844 spa_history_create_obj(spa
, tx
);
4846 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4847 spa_history_log_version(spa
, "create", tx
);
4850 * Create the pool config object.
4852 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4853 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4854 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4856 if (zap_add(spa
->spa_meta_objset
,
4857 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4858 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4859 cmn_err(CE_PANIC
, "failed to add pool config");
4862 if (zap_add(spa
->spa_meta_objset
,
4863 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4864 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4865 cmn_err(CE_PANIC
, "failed to add pool version");
4868 /* Newly created pools with the right version are always deflated. */
4869 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4870 spa
->spa_deflate
= TRUE
;
4871 if (zap_add(spa
->spa_meta_objset
,
4872 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4873 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4874 cmn_err(CE_PANIC
, "failed to add deflate");
4879 * Create the deferred-free bpobj. Turn off compression
4880 * because sync-to-convergence takes longer if the blocksize
4883 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4884 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4885 ZIO_COMPRESS_OFF
, tx
);
4886 if (zap_add(spa
->spa_meta_objset
,
4887 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4888 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4889 cmn_err(CE_PANIC
, "failed to add bpobj");
4891 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4892 spa
->spa_meta_objset
, obj
));
4895 * Generate some random noise for salted checksums to operate on.
4897 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4898 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4901 * Set pool properties.
4903 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4904 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4905 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4906 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4907 spa
->spa_multihost
= zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST
);
4909 if (props
!= NULL
) {
4910 spa_configfile_set(spa
, props
, B_FALSE
);
4911 spa_sync_props(props
, tx
);
4917 * If the root dataset is encrypted we will need to create key mappings
4918 * for the zio layer before we start to write any data to disk and hold
4919 * them until after the first txg has been synced. Waiting for the first
4920 * transaction to complete also ensures that our bean counters are
4921 * appropriately updated.
4923 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0) {
4924 root_dsobj
= dsl_dir_phys(dp
->dp_root_dir
)->dd_head_dataset_obj
;
4925 VERIFY0(spa_keystore_create_mapping_impl(spa
, root_dsobj
,
4926 dp
->dp_root_dir
, FTAG
));
4929 spa
->spa_sync_on
= B_TRUE
;
4931 mmp_thread_start(spa
);
4932 txg_wait_synced(dp
, txg
);
4934 if (dp
->dp_root_dir
->dd_crypto_obj
!= 0)
4935 VERIFY0(spa_keystore_remove_mapping(spa
, root_dsobj
, FTAG
));
4937 spa_spawn_aux_threads(spa
);
4939 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
4942 * Don't count references from objsets that are already closed
4943 * and are making their way through the eviction process.
4945 spa_evicting_os_wait(spa
);
4946 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4947 spa
->spa_load_state
= SPA_LOAD_NONE
;
4949 mutex_exit(&spa_namespace_lock
);
4955 * Import a non-root pool into the system.
4958 spa_import(char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4961 char *altroot
= NULL
;
4962 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4963 zpool_load_policy_t policy
;
4964 uint64_t mode
= spa_mode_global
;
4965 uint64_t readonly
= B_FALSE
;
4968 nvlist_t
**spares
, **l2cache
;
4969 uint_t nspares
, nl2cache
;
4972 * If a pool with this name exists, return failure.
4974 mutex_enter(&spa_namespace_lock
);
4975 if (spa_lookup(pool
) != NULL
) {
4976 mutex_exit(&spa_namespace_lock
);
4977 return (SET_ERROR(EEXIST
));
4981 * Create and initialize the spa structure.
4983 (void) nvlist_lookup_string(props
,
4984 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4985 (void) nvlist_lookup_uint64(props
,
4986 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4989 spa
= spa_add(pool
, config
, altroot
);
4990 spa
->spa_import_flags
= flags
;
4993 * Verbatim import - Take a pool and insert it into the namespace
4994 * as if it had been loaded at boot.
4996 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4998 spa_configfile_set(spa
, props
, B_FALSE
);
5000 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
5001 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
5002 zfs_dbgmsg("spa_import: verbatim import of %s", pool
);
5003 mutex_exit(&spa_namespace_lock
);
5007 spa_activate(spa
, mode
);
5010 * Don't start async tasks until we know everything is healthy.
5012 spa_async_suspend(spa
);
5014 zpool_get_load_policy(config
, &policy
);
5015 if (policy
.zlp_rewind
& ZPOOL_DO_REWIND
)
5016 state
= SPA_LOAD_RECOVER
;
5018 spa
->spa_config_source
= SPA_CONFIG_SRC_TRYIMPORT
;
5020 if (state
!= SPA_LOAD_RECOVER
) {
5021 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
5022 zfs_dbgmsg("spa_import: importing %s", pool
);
5024 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5025 "(RECOVERY MODE)", pool
, (longlong_t
)policy
.zlp_txg
);
5027 error
= spa_load_best(spa
, state
, policy
.zlp_txg
, policy
.zlp_rewind
);
5030 * Propagate anything learned while loading the pool and pass it
5031 * back to caller (i.e. rewind info, missing devices, etc).
5033 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
5034 spa
->spa_load_info
) == 0);
5036 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5038 * Toss any existing sparelist, as it doesn't have any validity
5039 * anymore, and conflicts with spa_has_spare().
5041 if (spa
->spa_spares
.sav_config
) {
5042 nvlist_free(spa
->spa_spares
.sav_config
);
5043 spa
->spa_spares
.sav_config
= NULL
;
5044 spa_load_spares(spa
);
5046 if (spa
->spa_l2cache
.sav_config
) {
5047 nvlist_free(spa
->spa_l2cache
.sav_config
);
5048 spa
->spa_l2cache
.sav_config
= NULL
;
5049 spa_load_l2cache(spa
);
5052 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
5054 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5057 spa_configfile_set(spa
, props
, B_FALSE
);
5059 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
5060 (error
= spa_prop_set(spa
, props
)))) {
5062 spa_deactivate(spa
);
5064 mutex_exit(&spa_namespace_lock
);
5068 spa_async_resume(spa
);
5071 * Override any spares and level 2 cache devices as specified by
5072 * the user, as these may have correct device names/devids, etc.
5074 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
5075 &spares
, &nspares
) == 0) {
5076 if (spa
->spa_spares
.sav_config
)
5077 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
5078 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5080 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
5081 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5082 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
5083 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
5084 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5085 spa_load_spares(spa
);
5086 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5087 spa
->spa_spares
.sav_sync
= B_TRUE
;
5089 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
5090 &l2cache
, &nl2cache
) == 0) {
5091 if (spa
->spa_l2cache
.sav_config
)
5092 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
5093 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
5095 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
5096 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5097 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
5098 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
5099 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5100 spa_load_l2cache(spa
);
5101 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5102 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5106 * Check for any removed devices.
5108 if (spa
->spa_autoreplace
) {
5109 spa_aux_check_removed(&spa
->spa_spares
);
5110 spa_aux_check_removed(&spa
->spa_l2cache
);
5113 if (spa_writeable(spa
)) {
5115 * Update the config cache to include the newly-imported pool.
5117 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5121 * It's possible that the pool was expanded while it was exported.
5122 * We kick off an async task to handle this for us.
5124 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
5126 spa_history_log_version(spa
, "import", NULL
);
5128 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
5130 zvol_create_minors(spa
, pool
, B_TRUE
);
5132 mutex_exit(&spa_namespace_lock
);
5138 spa_tryimport(nvlist_t
*tryconfig
)
5140 nvlist_t
*config
= NULL
;
5141 char *poolname
, *cachefile
;
5145 zpool_load_policy_t policy
;
5147 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
5150 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
5154 * Create and initialize the spa structure.
5156 mutex_enter(&spa_namespace_lock
);
5157 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
5158 spa_activate(spa
, FREAD
);
5161 * Rewind pool if a max txg was provided.
5163 zpool_get_load_policy(spa
->spa_config
, &policy
);
5164 if (policy
.zlp_txg
!= UINT64_MAX
) {
5165 spa
->spa_load_max_txg
= policy
.zlp_txg
;
5166 spa
->spa_extreme_rewind
= B_TRUE
;
5167 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
5168 poolname
, (longlong_t
)policy
.zlp_txg
);
5170 zfs_dbgmsg("spa_tryimport: importing %s", poolname
);
5173 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_CACHEFILE
, &cachefile
)
5175 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile
);
5176 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
5178 spa
->spa_config_source
= SPA_CONFIG_SRC_SCAN
;
5181 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
);
5184 * If 'tryconfig' was at least parsable, return the current config.
5186 if (spa
->spa_root_vdev
!= NULL
) {
5187 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
5188 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
5190 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5192 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
5193 spa
->spa_uberblock
.ub_timestamp
) == 0);
5194 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
5195 spa
->spa_load_info
) == 0);
5196 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_ERRATA
,
5197 spa
->spa_errata
) == 0);
5200 * If the bootfs property exists on this pool then we
5201 * copy it out so that external consumers can tell which
5202 * pools are bootable.
5204 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
5205 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
5208 * We have to play games with the name since the
5209 * pool was opened as TRYIMPORT_NAME.
5211 if (dsl_dsobj_to_dsname(spa_name(spa
),
5212 spa
->spa_bootfs
, tmpname
) == 0) {
5216 dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
5218 cp
= strchr(tmpname
, '/');
5220 (void) strlcpy(dsname
, tmpname
,
5223 (void) snprintf(dsname
, MAXPATHLEN
,
5224 "%s/%s", poolname
, ++cp
);
5226 VERIFY(nvlist_add_string(config
,
5227 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
5228 kmem_free(dsname
, MAXPATHLEN
);
5230 kmem_free(tmpname
, MAXPATHLEN
);
5234 * Add the list of hot spares and level 2 cache devices.
5236 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5237 spa_add_spares(spa
, config
);
5238 spa_add_l2cache(spa
, config
);
5239 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5243 spa_deactivate(spa
);
5245 mutex_exit(&spa_namespace_lock
);
5251 * Pool export/destroy
5253 * The act of destroying or exporting a pool is very simple. We make sure there
5254 * is no more pending I/O and any references to the pool are gone. Then, we
5255 * update the pool state and sync all the labels to disk, removing the
5256 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5257 * we don't sync the labels or remove the configuration cache.
5260 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
5261 boolean_t force
, boolean_t hardforce
)
5268 if (!(spa_mode_global
& FWRITE
))
5269 return (SET_ERROR(EROFS
));
5271 mutex_enter(&spa_namespace_lock
);
5272 if ((spa
= spa_lookup(pool
)) == NULL
) {
5273 mutex_exit(&spa_namespace_lock
);
5274 return (SET_ERROR(ENOENT
));
5278 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5279 * reacquire the namespace lock, and see if we can export.
5281 spa_open_ref(spa
, FTAG
);
5282 mutex_exit(&spa_namespace_lock
);
5283 spa_async_suspend(spa
);
5284 if (spa
->spa_zvol_taskq
) {
5285 zvol_remove_minors(spa
, spa_name(spa
), B_TRUE
);
5286 taskq_wait(spa
->spa_zvol_taskq
);
5288 mutex_enter(&spa_namespace_lock
);
5289 spa_close(spa
, FTAG
);
5291 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
5294 * The pool will be in core if it's openable, in which case we can
5295 * modify its state. Objsets may be open only because they're dirty,
5296 * so we have to force it to sync before checking spa_refcnt.
5298 if (spa
->spa_sync_on
) {
5299 txg_wait_synced(spa
->spa_dsl_pool
, 0);
5300 spa_evicting_os_wait(spa
);
5304 * A pool cannot be exported or destroyed if there are active
5305 * references. If we are resetting a pool, allow references by
5306 * fault injection handlers.
5308 if (!spa_refcount_zero(spa
) ||
5309 (spa
->spa_inject_ref
!= 0 &&
5310 new_state
!= POOL_STATE_UNINITIALIZED
)) {
5311 spa_async_resume(spa
);
5312 mutex_exit(&spa_namespace_lock
);
5313 return (SET_ERROR(EBUSY
));
5316 if (spa
->spa_sync_on
) {
5318 * A pool cannot be exported if it has an active shared spare.
5319 * This is to prevent other pools stealing the active spare
5320 * from an exported pool. At user's own will, such pool can
5321 * be forcedly exported.
5323 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
5324 spa_has_active_shared_spare(spa
)) {
5325 spa_async_resume(spa
);
5326 mutex_exit(&spa_namespace_lock
);
5327 return (SET_ERROR(EXDEV
));
5331 * We want this to be reflected on every label,
5332 * so mark them all dirty. spa_unload() will do the
5333 * final sync that pushes these changes out.
5335 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
5336 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5337 spa
->spa_state
= new_state
;
5338 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
5340 vdev_config_dirty(spa
->spa_root_vdev
);
5341 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5346 if (new_state
== POOL_STATE_DESTROYED
)
5347 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
5348 else if (new_state
== POOL_STATE_EXPORTED
)
5349 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_EXPORT
);
5351 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
5353 spa_deactivate(spa
);
5356 if (oldconfig
&& spa
->spa_config
)
5357 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
5359 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
5361 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
5364 mutex_exit(&spa_namespace_lock
);
5370 * Destroy a storage pool.
5373 spa_destroy(char *pool
)
5375 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
5380 * Export a storage pool.
5383 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
5384 boolean_t hardforce
)
5386 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
5391 * Similar to spa_export(), this unloads the spa_t without actually removing it
5392 * from the namespace in any way.
5395 spa_reset(char *pool
)
5397 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
5402 * ==========================================================================
5403 * Device manipulation
5404 * ==========================================================================
5408 * Add a device to a storage pool.
5411 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
5415 vdev_t
*rvd
= spa
->spa_root_vdev
;
5417 nvlist_t
**spares
, **l2cache
;
5418 uint_t nspares
, nl2cache
;
5420 ASSERT(spa_writeable(spa
));
5422 txg
= spa_vdev_enter(spa
);
5424 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
5425 VDEV_ALLOC_ADD
)) != 0)
5426 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5428 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
5430 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
5434 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
5438 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
5439 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5441 if (vd
->vdev_children
!= 0 &&
5442 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
5443 return (spa_vdev_exit(spa
, vd
, txg
, error
));
5446 * We must validate the spares and l2cache devices after checking the
5447 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5449 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
5450 return (spa_vdev_exit(spa
, vd
, txg
, error
));
5453 * If we are in the middle of a device removal, we can only add
5454 * devices which match the existing devices in the pool.
5455 * If we are in the middle of a removal, or have some indirect
5456 * vdevs, we can not add raidz toplevels.
5458 if (spa
->spa_vdev_removal
!= NULL
||
5459 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
5460 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5461 tvd
= vd
->vdev_child
[c
];
5462 if (spa
->spa_vdev_removal
!= NULL
&&
5463 tvd
->vdev_ashift
!= spa
->spa_max_ashift
) {
5464 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5466 /* Fail if top level vdev is raidz */
5467 if (tvd
->vdev_ops
== &vdev_raidz_ops
) {
5468 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5471 * Need the top level mirror to be
5472 * a mirror of leaf vdevs only
5474 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
5475 for (uint64_t cid
= 0;
5476 cid
< tvd
->vdev_children
; cid
++) {
5477 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
5478 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
5479 return (spa_vdev_exit(spa
, vd
,
5487 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5490 * Set the vdev id to the first hole, if one exists.
5492 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
5493 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
5494 vdev_free(rvd
->vdev_child
[id
]);
5498 tvd
= vd
->vdev_child
[c
];
5499 vdev_remove_child(vd
, tvd
);
5501 vdev_add_child(rvd
, tvd
);
5502 vdev_config_dirty(tvd
);
5506 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
5507 ZPOOL_CONFIG_SPARES
);
5508 spa_load_spares(spa
);
5509 spa
->spa_spares
.sav_sync
= B_TRUE
;
5512 if (nl2cache
!= 0) {
5513 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
5514 ZPOOL_CONFIG_L2CACHE
);
5515 spa_load_l2cache(spa
);
5516 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5520 * We have to be careful when adding new vdevs to an existing pool.
5521 * If other threads start allocating from these vdevs before we
5522 * sync the config cache, and we lose power, then upon reboot we may
5523 * fail to open the pool because there are DVAs that the config cache
5524 * can't translate. Therefore, we first add the vdevs without
5525 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5526 * and then let spa_config_update() initialize the new metaslabs.
5528 * spa_load() checks for added-but-not-initialized vdevs, so that
5529 * if we lose power at any point in this sequence, the remaining
5530 * steps will be completed the next time we load the pool.
5532 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
5534 mutex_enter(&spa_namespace_lock
);
5535 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5536 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
5537 mutex_exit(&spa_namespace_lock
);
5543 * Attach a device to a mirror. The arguments are the path to any device
5544 * in the mirror, and the nvroot for the new device. If the path specifies
5545 * a device that is not mirrored, we automatically insert the mirror vdev.
5547 * If 'replacing' is specified, the new device is intended to replace the
5548 * existing device; in this case the two devices are made into their own
5549 * mirror using the 'replacing' vdev, which is functionally identical to
5550 * the mirror vdev (it actually reuses all the same ops) but has a few
5551 * extra rules: you can't attach to it after it's been created, and upon
5552 * completion of resilvering, the first disk (the one being replaced)
5553 * is automatically detached.
5556 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
5558 uint64_t txg
, dtl_max_txg
;
5559 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5560 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
5562 char *oldvdpath
, *newvdpath
;
5566 ASSERT(spa_writeable(spa
));
5568 txg
= spa_vdev_enter(spa
);
5570 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5572 if (spa
->spa_vdev_removal
!= NULL
)
5573 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5576 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5578 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
5579 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5581 pvd
= oldvd
->vdev_parent
;
5583 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
5584 VDEV_ALLOC_ATTACH
)) != 0)
5585 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5587 if (newrootvd
->vdev_children
!= 1)
5588 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
5590 newvd
= newrootvd
->vdev_child
[0];
5592 if (!newvd
->vdev_ops
->vdev_op_leaf
)
5593 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
5595 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
5596 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
5599 * Spares can't replace logs
5601 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
5602 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5606 * For attach, the only allowable parent is a mirror or the root
5609 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5610 pvd
->vdev_ops
!= &vdev_root_ops
)
5611 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5613 pvops
= &vdev_mirror_ops
;
5616 * Active hot spares can only be replaced by inactive hot
5619 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5620 oldvd
->vdev_isspare
&&
5621 !spa_has_spare(spa
, newvd
->vdev_guid
))
5622 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5625 * If the source is a hot spare, and the parent isn't already a
5626 * spare, then we want to create a new hot spare. Otherwise, we
5627 * want to create a replacing vdev. The user is not allowed to
5628 * attach to a spared vdev child unless the 'isspare' state is
5629 * the same (spare replaces spare, non-spare replaces
5632 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
5633 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
5634 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5635 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5636 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
5637 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5640 if (newvd
->vdev_isspare
)
5641 pvops
= &vdev_spare_ops
;
5643 pvops
= &vdev_replacing_ops
;
5647 * Make sure the new device is big enough.
5649 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
5650 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
5653 * The new device cannot have a higher alignment requirement
5654 * than the top-level vdev.
5656 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
5657 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
5660 * If this is an in-place replacement, update oldvd's path and devid
5661 * to make it distinguishable from newvd, and unopenable from now on.
5663 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
5664 spa_strfree(oldvd
->vdev_path
);
5665 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
5667 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
5668 newvd
->vdev_path
, "old");
5669 if (oldvd
->vdev_devid
!= NULL
) {
5670 spa_strfree(oldvd
->vdev_devid
);
5671 oldvd
->vdev_devid
= NULL
;
5675 /* mark the device being resilvered */
5676 newvd
->vdev_resilver_txg
= txg
;
5679 * If the parent is not a mirror, or if we're replacing, insert the new
5680 * mirror/replacing/spare vdev above oldvd.
5682 if (pvd
->vdev_ops
!= pvops
)
5683 pvd
= vdev_add_parent(oldvd
, pvops
);
5685 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
5686 ASSERT(pvd
->vdev_ops
== pvops
);
5687 ASSERT(oldvd
->vdev_parent
== pvd
);
5690 * Extract the new device from its root and add it to pvd.
5692 vdev_remove_child(newrootvd
, newvd
);
5693 newvd
->vdev_id
= pvd
->vdev_children
;
5694 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
5695 vdev_add_child(pvd
, newvd
);
5698 * Reevaluate the parent vdev state.
5700 vdev_propagate_state(pvd
);
5702 tvd
= newvd
->vdev_top
;
5703 ASSERT(pvd
->vdev_top
== tvd
);
5704 ASSERT(tvd
->vdev_parent
== rvd
);
5706 vdev_config_dirty(tvd
);
5709 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5710 * for any dmu_sync-ed blocks. It will propagate upward when
5711 * spa_vdev_exit() calls vdev_dtl_reassess().
5713 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
5715 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
5716 dtl_max_txg
- TXG_INITIAL
);
5718 if (newvd
->vdev_isspare
) {
5719 spa_spare_activate(newvd
);
5720 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
5723 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
5724 newvdpath
= spa_strdup(newvd
->vdev_path
);
5725 newvd_isspare
= newvd
->vdev_isspare
;
5728 * Mark newvd's DTL dirty in this txg.
5730 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
5733 * Schedule the resilver to restart in the future. We do this to
5734 * ensure that dmu_sync-ed blocks have been stitched into the
5735 * respective datasets.
5737 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
5739 if (spa
->spa_bootfs
)
5740 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
5742 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
5747 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
5749 spa_history_log_internal(spa
, "vdev attach", NULL
,
5750 "%s vdev=%s %s vdev=%s",
5751 replacing
&& newvd_isspare
? "spare in" :
5752 replacing
? "replace" : "attach", newvdpath
,
5753 replacing
? "for" : "to", oldvdpath
);
5755 spa_strfree(oldvdpath
);
5756 spa_strfree(newvdpath
);
5762 * Detach a device from a mirror or replacing vdev.
5764 * If 'replace_done' is specified, only detach if the parent
5765 * is a replacing vdev.
5768 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5772 ASSERTV(vdev_t
*rvd
= spa
->spa_root_vdev
);
5773 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5774 boolean_t unspare
= B_FALSE
;
5775 uint64_t unspare_guid
= 0;
5778 ASSERT(spa_writeable(spa
));
5780 txg
= spa_vdev_enter(spa
);
5782 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5785 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5787 if (!vd
->vdev_ops
->vdev_op_leaf
)
5788 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5790 pvd
= vd
->vdev_parent
;
5793 * If the parent/child relationship is not as expected, don't do it.
5794 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5795 * vdev that's replacing B with C. The user's intent in replacing
5796 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5797 * the replace by detaching C, the expected behavior is to end up
5798 * M(A,B). But suppose that right after deciding to detach C,
5799 * the replacement of B completes. We would have M(A,C), and then
5800 * ask to detach C, which would leave us with just A -- not what
5801 * the user wanted. To prevent this, we make sure that the
5802 * parent/child relationship hasn't changed -- in this example,
5803 * that C's parent is still the replacing vdev R.
5805 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5806 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5809 * Only 'replacing' or 'spare' vdevs can be replaced.
5811 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5812 pvd
->vdev_ops
!= &vdev_spare_ops
)
5813 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5815 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5816 spa_version(spa
) >= SPA_VERSION_SPARES
);
5819 * Only mirror, replacing, and spare vdevs support detach.
5821 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5822 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5823 pvd
->vdev_ops
!= &vdev_spare_ops
)
5824 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5827 * If this device has the only valid copy of some data,
5828 * we cannot safely detach it.
5830 if (vdev_dtl_required(vd
))
5831 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5833 ASSERT(pvd
->vdev_children
>= 2);
5836 * If we are detaching the second disk from a replacing vdev, then
5837 * check to see if we changed the original vdev's path to have "/old"
5838 * at the end in spa_vdev_attach(). If so, undo that change now.
5840 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5841 vd
->vdev_path
!= NULL
) {
5842 size_t len
= strlen(vd
->vdev_path
);
5844 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
5845 cvd
= pvd
->vdev_child
[c
];
5847 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5850 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5851 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5852 spa_strfree(cvd
->vdev_path
);
5853 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5860 * If we are detaching the original disk from a spare, then it implies
5861 * that the spare should become a real disk, and be removed from the
5862 * active spare list for the pool.
5864 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5866 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5870 * Erase the disk labels so the disk can be used for other things.
5871 * This must be done after all other error cases are handled,
5872 * but before we disembowel vd (so we can still do I/O to it).
5873 * But if we can't do it, don't treat the error as fatal --
5874 * it may be that the unwritability of the disk is the reason
5875 * it's being detached!
5877 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5880 * Remove vd from its parent and compact the parent's children.
5882 vdev_remove_child(pvd
, vd
);
5883 vdev_compact_children(pvd
);
5886 * Remember one of the remaining children so we can get tvd below.
5888 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5891 * If we need to remove the remaining child from the list of hot spares,
5892 * do it now, marking the vdev as no longer a spare in the process.
5893 * We must do this before vdev_remove_parent(), because that can
5894 * change the GUID if it creates a new toplevel GUID. For a similar
5895 * reason, we must remove the spare now, in the same txg as the detach;
5896 * otherwise someone could attach a new sibling, change the GUID, and
5897 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5900 ASSERT(cvd
->vdev_isspare
);
5901 spa_spare_remove(cvd
);
5902 unspare_guid
= cvd
->vdev_guid
;
5903 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5904 cvd
->vdev_unspare
= B_TRUE
;
5908 * If the parent mirror/replacing vdev only has one child,
5909 * the parent is no longer needed. Remove it from the tree.
5911 if (pvd
->vdev_children
== 1) {
5912 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5913 cvd
->vdev_unspare
= B_FALSE
;
5914 vdev_remove_parent(cvd
);
5919 * We don't set tvd until now because the parent we just removed
5920 * may have been the previous top-level vdev.
5922 tvd
= cvd
->vdev_top
;
5923 ASSERT(tvd
->vdev_parent
== rvd
);
5926 * Reevaluate the parent vdev state.
5928 vdev_propagate_state(cvd
);
5931 * If the 'autoexpand' property is set on the pool then automatically
5932 * try to expand the size of the pool. For example if the device we
5933 * just detached was smaller than the others, it may be possible to
5934 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5935 * first so that we can obtain the updated sizes of the leaf vdevs.
5937 if (spa
->spa_autoexpand
) {
5939 vdev_expand(tvd
, txg
);
5942 vdev_config_dirty(tvd
);
5945 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5946 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5947 * But first make sure we're not on any *other* txg's DTL list, to
5948 * prevent vd from being accessed after it's freed.
5950 vdpath
= spa_strdup(vd
->vdev_path
? vd
->vdev_path
: "none");
5951 for (int t
= 0; t
< TXG_SIZE
; t
++)
5952 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5953 vd
->vdev_detached
= B_TRUE
;
5954 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5956 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5958 /* hang on to the spa before we release the lock */
5959 spa_open_ref(spa
, FTAG
);
5961 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5963 spa_history_log_internal(spa
, "detach", NULL
,
5965 spa_strfree(vdpath
);
5968 * If this was the removal of the original device in a hot spare vdev,
5969 * then we want to go through and remove the device from the hot spare
5970 * list of every other pool.
5973 spa_t
*altspa
= NULL
;
5975 mutex_enter(&spa_namespace_lock
);
5976 while ((altspa
= spa_next(altspa
)) != NULL
) {
5977 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5981 spa_open_ref(altspa
, FTAG
);
5982 mutex_exit(&spa_namespace_lock
);
5983 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5984 mutex_enter(&spa_namespace_lock
);
5985 spa_close(altspa
, FTAG
);
5987 mutex_exit(&spa_namespace_lock
);
5989 /* search the rest of the vdevs for spares to remove */
5990 spa_vdev_resilver_done(spa
);
5993 /* all done with the spa; OK to release */
5994 mutex_enter(&spa_namespace_lock
);
5995 spa_close(spa
, FTAG
);
5996 mutex_exit(&spa_namespace_lock
);
6002 * Split a set of devices from their mirrors, and create a new pool from them.
6005 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
6006 nvlist_t
*props
, boolean_t exp
)
6009 uint64_t txg
, *glist
;
6011 uint_t c
, children
, lastlog
;
6012 nvlist_t
**child
, *nvl
, *tmp
;
6014 char *altroot
= NULL
;
6015 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
6016 boolean_t activate_slog
;
6018 ASSERT(spa_writeable(spa
));
6020 txg
= spa_vdev_enter(spa
);
6022 /* clear the log and flush everything up to now */
6023 activate_slog
= spa_passivate_log(spa
);
6024 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
6025 error
= spa_reset_logs(spa
);
6026 txg
= spa_vdev_config_enter(spa
);
6029 spa_activate_log(spa
);
6032 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6034 /* check new spa name before going any further */
6035 if (spa_lookup(newname
) != NULL
)
6036 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
6039 * scan through all the children to ensure they're all mirrors
6041 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
6042 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
6044 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
6046 /* first, check to ensure we've got the right child count */
6047 rvd
= spa
->spa_root_vdev
;
6049 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
6050 vdev_t
*vd
= rvd
->vdev_child
[c
];
6052 /* don't count the holes & logs as children */
6053 if (vd
->vdev_islog
|| !vdev_is_concrete(vd
)) {
6061 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
6062 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
6064 /* next, ensure no spare or cache devices are part of the split */
6065 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
6066 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
6067 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
6069 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
6070 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
6072 /* then, loop over each vdev and validate it */
6073 for (c
= 0; c
< children
; c
++) {
6074 uint64_t is_hole
= 0;
6076 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
6080 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
6081 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
6084 error
= SET_ERROR(EINVAL
);
6089 /* which disk is going to be split? */
6090 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
6092 error
= SET_ERROR(EINVAL
);
6096 /* look it up in the spa */
6097 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
6098 if (vml
[c
] == NULL
) {
6099 error
= SET_ERROR(ENODEV
);
6103 /* make sure there's nothing stopping the split */
6104 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
6105 vml
[c
]->vdev_islog
||
6106 !vdev_is_concrete(vml
[c
]) ||
6107 vml
[c
]->vdev_isspare
||
6108 vml
[c
]->vdev_isl2cache
||
6109 !vdev_writeable(vml
[c
]) ||
6110 vml
[c
]->vdev_children
!= 0 ||
6111 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
6112 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
6113 error
= SET_ERROR(EINVAL
);
6117 if (vdev_dtl_required(vml
[c
])) {
6118 error
= SET_ERROR(EBUSY
);
6122 /* we need certain info from the top level */
6123 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
6124 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
6125 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
6126 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
6127 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
6128 vml
[c
]->vdev_top
->vdev_asize
) == 0);
6129 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
6130 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
6132 /* transfer per-vdev ZAPs */
6133 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
6134 VERIFY0(nvlist_add_uint64(child
[c
],
6135 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
6137 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
6138 VERIFY0(nvlist_add_uint64(child
[c
],
6139 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
6140 vml
[c
]->vdev_parent
->vdev_top_zap
));
6144 kmem_free(vml
, children
* sizeof (vdev_t
*));
6145 kmem_free(glist
, children
* sizeof (uint64_t));
6146 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
6149 /* stop writers from using the disks */
6150 for (c
= 0; c
< children
; c
++) {
6152 vml
[c
]->vdev_offline
= B_TRUE
;
6154 vdev_reopen(spa
->spa_root_vdev
);
6157 * Temporarily record the splitting vdevs in the spa config. This
6158 * will disappear once the config is regenerated.
6160 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6161 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
6162 glist
, children
) == 0);
6163 kmem_free(glist
, children
* sizeof (uint64_t));
6165 mutex_enter(&spa
->spa_props_lock
);
6166 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
6168 mutex_exit(&spa
->spa_props_lock
);
6169 spa
->spa_config_splitting
= nvl
;
6170 vdev_config_dirty(spa
->spa_root_vdev
);
6172 /* configure and create the new pool */
6173 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
6174 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
6175 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
6176 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6177 spa_version(spa
)) == 0);
6178 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
6179 spa
->spa_config_txg
) == 0);
6180 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
6181 spa_generate_guid(NULL
)) == 0);
6182 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
6183 (void) nvlist_lookup_string(props
,
6184 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
6186 /* add the new pool to the namespace */
6187 newspa
= spa_add(newname
, config
, altroot
);
6188 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
6189 newspa
->spa_config_txg
= spa
->spa_config_txg
;
6190 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
6192 /* release the spa config lock, retaining the namespace lock */
6193 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
6195 if (zio_injection_enabled
)
6196 zio_handle_panic_injection(spa
, FTAG
, 1);
6198 spa_activate(newspa
, spa_mode_global
);
6199 spa_async_suspend(newspa
);
6201 newspa
->spa_config_source
= SPA_CONFIG_SRC_SPLIT
;
6203 /* create the new pool from the disks of the original pool */
6204 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
);
6208 /* if that worked, generate a real config for the new pool */
6209 if (newspa
->spa_root_vdev
!= NULL
) {
6210 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
6211 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6212 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
6213 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
6214 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
6219 if (props
!= NULL
) {
6220 spa_configfile_set(newspa
, props
, B_FALSE
);
6221 error
= spa_prop_set(newspa
, props
);
6226 /* flush everything */
6227 txg
= spa_vdev_config_enter(newspa
);
6228 vdev_config_dirty(newspa
->spa_root_vdev
);
6229 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
6231 if (zio_injection_enabled
)
6232 zio_handle_panic_injection(spa
, FTAG
, 2);
6234 spa_async_resume(newspa
);
6236 /* finally, update the original pool's config */
6237 txg
= spa_vdev_config_enter(spa
);
6238 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
6239 error
= dmu_tx_assign(tx
, TXG_WAIT
);
6242 for (c
= 0; c
< children
; c
++) {
6243 if (vml
[c
] != NULL
) {
6246 spa_history_log_internal(spa
, "detach", tx
,
6247 "vdev=%s", vml
[c
]->vdev_path
);
6252 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
6253 vdev_config_dirty(spa
->spa_root_vdev
);
6254 spa
->spa_config_splitting
= NULL
;
6258 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
6260 if (zio_injection_enabled
)
6261 zio_handle_panic_injection(spa
, FTAG
, 3);
6263 /* split is complete; log a history record */
6264 spa_history_log_internal(newspa
, "split", NULL
,
6265 "from pool %s", spa_name(spa
));
6267 kmem_free(vml
, children
* sizeof (vdev_t
*));
6269 /* if we're not going to mount the filesystems in userland, export */
6271 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
6278 spa_deactivate(newspa
);
6281 txg
= spa_vdev_config_enter(spa
);
6283 /* re-online all offlined disks */
6284 for (c
= 0; c
< children
; c
++) {
6286 vml
[c
]->vdev_offline
= B_FALSE
;
6288 vdev_reopen(spa
->spa_root_vdev
);
6290 nvlist_free(spa
->spa_config_splitting
);
6291 spa
->spa_config_splitting
= NULL
;
6292 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
6294 kmem_free(vml
, children
* sizeof (vdev_t
*));
6299 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6300 * currently spared, so we can detach it.
6303 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
6305 vdev_t
*newvd
, *oldvd
;
6307 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6308 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
6314 * Check for a completed replacement. We always consider the first
6315 * vdev in the list to be the oldest vdev, and the last one to be
6316 * the newest (see spa_vdev_attach() for how that works). In
6317 * the case where the newest vdev is faulted, we will not automatically
6318 * remove it after a resilver completes. This is OK as it will require
6319 * user intervention to determine which disk the admin wishes to keep.
6321 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
6322 ASSERT(vd
->vdev_children
> 1);
6324 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
6325 oldvd
= vd
->vdev_child
[0];
6327 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
6328 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
6329 !vdev_dtl_required(oldvd
))
6334 * Check for a completed resilver with the 'unspare' flag set.
6336 if (vd
->vdev_ops
== &vdev_spare_ops
) {
6337 vdev_t
*first
= vd
->vdev_child
[0];
6338 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
6340 if (last
->vdev_unspare
) {
6343 } else if (first
->vdev_unspare
) {
6350 if (oldvd
!= NULL
&&
6351 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
6352 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
6353 !vdev_dtl_required(oldvd
))
6357 * If there are more than two spares attached to a disk,
6358 * and those spares are not required, then we want to
6359 * attempt to free them up now so that they can be used
6360 * by other pools. Once we're back down to a single
6361 * disk+spare, we stop removing them.
6363 if (vd
->vdev_children
> 2) {
6364 newvd
= vd
->vdev_child
[1];
6366 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
6367 vdev_dtl_empty(last
, DTL_MISSING
) &&
6368 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
6369 !vdev_dtl_required(newvd
))
6378 spa_vdev_resilver_done(spa_t
*spa
)
6380 vdev_t
*vd
, *pvd
, *ppvd
;
6381 uint64_t guid
, sguid
, pguid
, ppguid
;
6383 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6385 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
6386 pvd
= vd
->vdev_parent
;
6387 ppvd
= pvd
->vdev_parent
;
6388 guid
= vd
->vdev_guid
;
6389 pguid
= pvd
->vdev_guid
;
6390 ppguid
= ppvd
->vdev_guid
;
6393 * If we have just finished replacing a hot spared device, then
6394 * we need to detach the parent's first child (the original hot
6397 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
6398 ppvd
->vdev_children
== 2) {
6399 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
6400 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
6402 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
6404 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6405 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
6407 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
6409 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6412 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6416 * Update the stored path or FRU for this vdev.
6419 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
6423 boolean_t sync
= B_FALSE
;
6425 ASSERT(spa_writeable(spa
));
6427 spa_vdev_state_enter(spa
, SCL_ALL
);
6429 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
6430 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
6432 if (!vd
->vdev_ops
->vdev_op_leaf
)
6433 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
6436 if (strcmp(value
, vd
->vdev_path
) != 0) {
6437 spa_strfree(vd
->vdev_path
);
6438 vd
->vdev_path
= spa_strdup(value
);
6442 if (vd
->vdev_fru
== NULL
) {
6443 vd
->vdev_fru
= spa_strdup(value
);
6445 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
6446 spa_strfree(vd
->vdev_fru
);
6447 vd
->vdev_fru
= spa_strdup(value
);
6452 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
6456 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
6458 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
6462 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
6464 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
6468 * ==========================================================================
6470 * ==========================================================================
6473 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
6475 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6477 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6478 return (SET_ERROR(EBUSY
));
6480 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
6484 spa_scan_stop(spa_t
*spa
)
6486 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6487 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6488 return (SET_ERROR(EBUSY
));
6489 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
6493 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
6495 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6497 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
6498 return (SET_ERROR(ENOTSUP
));
6501 * If a resilver was requested, but there is no DTL on a
6502 * writeable leaf device, we have nothing to do.
6504 if (func
== POOL_SCAN_RESILVER
&&
6505 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
6506 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6510 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6514 * ==========================================================================
6515 * SPA async task processing
6516 * ==========================================================================
6520 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6522 if (vd
->vdev_remove_wanted
) {
6523 vd
->vdev_remove_wanted
= B_FALSE
;
6524 vd
->vdev_delayed_close
= B_FALSE
;
6525 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6528 * We want to clear the stats, but we don't want to do a full
6529 * vdev_clear() as that will cause us to throw away
6530 * degraded/faulted state as well as attempt to reopen the
6531 * device, all of which is a waste.
6533 vd
->vdev_stat
.vs_read_errors
= 0;
6534 vd
->vdev_stat
.vs_write_errors
= 0;
6535 vd
->vdev_stat
.vs_checksum_errors
= 0;
6537 vdev_state_dirty(vd
->vdev_top
);
6540 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6541 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6545 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6547 if (vd
->vdev_probe_wanted
) {
6548 vd
->vdev_probe_wanted
= B_FALSE
;
6549 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6552 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6553 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6557 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6559 if (!spa
->spa_autoexpand
)
6562 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6563 vdev_t
*cvd
= vd
->vdev_child
[c
];
6564 spa_async_autoexpand(spa
, cvd
);
6567 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6570 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_AUTOEXPAND
);
6574 spa_async_thread(void *arg
)
6576 spa_t
*spa
= (spa_t
*)arg
;
6579 ASSERT(spa
->spa_sync_on
);
6581 mutex_enter(&spa
->spa_async_lock
);
6582 tasks
= spa
->spa_async_tasks
;
6583 spa
->spa_async_tasks
= 0;
6584 mutex_exit(&spa
->spa_async_lock
);
6587 * See if the config needs to be updated.
6589 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6590 uint64_t old_space
, new_space
;
6592 mutex_enter(&spa_namespace_lock
);
6593 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6594 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6595 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6596 mutex_exit(&spa_namespace_lock
);
6599 * If the pool grew as a result of the config update,
6600 * then log an internal history event.
6602 if (new_space
!= old_space
) {
6603 spa_history_log_internal(spa
, "vdev online", NULL
,
6604 "pool '%s' size: %llu(+%llu)",
6605 spa_name(spa
), new_space
, new_space
- old_space
);
6610 * See if any devices need to be marked REMOVED.
6612 if (tasks
& SPA_ASYNC_REMOVE
) {
6613 spa_vdev_state_enter(spa
, SCL_NONE
);
6614 spa_async_remove(spa
, spa
->spa_root_vdev
);
6615 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6616 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6617 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6618 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6619 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6622 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6623 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6624 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6625 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6629 * See if any devices need to be probed.
6631 if (tasks
& SPA_ASYNC_PROBE
) {
6632 spa_vdev_state_enter(spa
, SCL_NONE
);
6633 spa_async_probe(spa
, spa
->spa_root_vdev
);
6634 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6638 * If any devices are done replacing, detach them.
6640 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6641 spa_vdev_resilver_done(spa
);
6644 * Kick off a resilver.
6646 if (tasks
& SPA_ASYNC_RESILVER
)
6647 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6650 * Let the world know that we're done.
6652 mutex_enter(&spa
->spa_async_lock
);
6653 spa
->spa_async_thread
= NULL
;
6654 cv_broadcast(&spa
->spa_async_cv
);
6655 mutex_exit(&spa
->spa_async_lock
);
6660 spa_async_suspend(spa_t
*spa
)
6662 mutex_enter(&spa
->spa_async_lock
);
6663 spa
->spa_async_suspended
++;
6664 while (spa
->spa_async_thread
!= NULL
)
6665 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6666 mutex_exit(&spa
->spa_async_lock
);
6668 spa_vdev_remove_suspend(spa
);
6670 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
6671 if (condense_thread
!= NULL
&& zthr_isrunning(condense_thread
))
6672 VERIFY0(zthr_cancel(condense_thread
));
6676 spa_async_resume(spa_t
*spa
)
6678 mutex_enter(&spa
->spa_async_lock
);
6679 ASSERT(spa
->spa_async_suspended
!= 0);
6680 spa
->spa_async_suspended
--;
6681 mutex_exit(&spa
->spa_async_lock
);
6682 spa_restart_removal(spa
);
6684 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
6685 if (condense_thread
!= NULL
&& !zthr_isrunning(condense_thread
))
6686 zthr_resume(condense_thread
);
6690 spa_async_tasks_pending(spa_t
*spa
)
6692 uint_t non_config_tasks
;
6694 boolean_t config_task_suspended
;
6696 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6697 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6698 if (spa
->spa_ccw_fail_time
== 0) {
6699 config_task_suspended
= B_FALSE
;
6701 config_task_suspended
=
6702 (gethrtime() - spa
->spa_ccw_fail_time
) <
6703 ((hrtime_t
)zfs_ccw_retry_interval
* NANOSEC
);
6706 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6710 spa_async_dispatch(spa_t
*spa
)
6712 mutex_enter(&spa
->spa_async_lock
);
6713 if (spa_async_tasks_pending(spa
) &&
6714 !spa
->spa_async_suspended
&&
6715 spa
->spa_async_thread
== NULL
&&
6717 spa
->spa_async_thread
= thread_create(NULL
, 0,
6718 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6719 mutex_exit(&spa
->spa_async_lock
);
6723 spa_async_request(spa_t
*spa
, int task
)
6725 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6726 mutex_enter(&spa
->spa_async_lock
);
6727 spa
->spa_async_tasks
|= task
;
6728 mutex_exit(&spa
->spa_async_lock
);
6732 * ==========================================================================
6733 * SPA syncing routines
6734 * ==========================================================================
6738 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6741 bpobj_enqueue(bpo
, bp
, tx
);
6746 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6750 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6756 * Note: this simple function is not inlined to make it easier to dtrace the
6757 * amount of time spent syncing frees.
6760 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6762 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6763 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6764 VERIFY(zio_wait(zio
) == 0);
6768 * Note: this simple function is not inlined to make it easier to dtrace the
6769 * amount of time spent syncing deferred frees.
6772 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6774 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6775 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6776 spa_free_sync_cb
, zio
, tx
), ==, 0);
6777 VERIFY0(zio_wait(zio
));
6781 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6783 char *packed
= NULL
;
6788 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6791 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6792 * information. This avoids the dmu_buf_will_dirty() path and
6793 * saves us a pre-read to get data we don't actually care about.
6795 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6796 packed
= vmem_alloc(bufsize
, KM_SLEEP
);
6798 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6800 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6802 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6804 vmem_free(packed
, bufsize
);
6806 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6807 dmu_buf_will_dirty(db
, tx
);
6808 *(uint64_t *)db
->db_data
= nvsize
;
6809 dmu_buf_rele(db
, FTAG
);
6813 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6814 const char *config
, const char *entry
)
6824 * Update the MOS nvlist describing the list of available devices.
6825 * spa_validate_aux() will have already made sure this nvlist is
6826 * valid and the vdevs are labeled appropriately.
6828 if (sav
->sav_object
== 0) {
6829 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6830 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6831 sizeof (uint64_t), tx
);
6832 VERIFY(zap_update(spa
->spa_meta_objset
,
6833 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6834 &sav
->sav_object
, tx
) == 0);
6837 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6838 if (sav
->sav_count
== 0) {
6839 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6841 list
= kmem_alloc(sav
->sav_count
*sizeof (void *), KM_SLEEP
);
6842 for (i
= 0; i
< sav
->sav_count
; i
++)
6843 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6844 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6845 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6846 sav
->sav_count
) == 0);
6847 for (i
= 0; i
< sav
->sav_count
; i
++)
6848 nvlist_free(list
[i
]);
6849 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6852 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6853 nvlist_free(nvroot
);
6855 sav
->sav_sync
= B_FALSE
;
6859 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6860 * The all-vdev ZAP must be empty.
6863 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6865 spa_t
*spa
= vd
->vdev_spa
;
6867 if (vd
->vdev_top_zap
!= 0) {
6868 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6869 vd
->vdev_top_zap
, tx
));
6871 if (vd
->vdev_leaf_zap
!= 0) {
6872 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6873 vd
->vdev_leaf_zap
, tx
));
6875 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
6876 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6881 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6886 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6887 * its config may not be dirty but we still need to build per-vdev ZAPs.
6888 * Similarly, if the pool is being assembled (e.g. after a split), we
6889 * need to rebuild the AVZ although the config may not be dirty.
6891 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6892 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6895 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6897 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6898 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6899 spa
->spa_all_vdev_zaps
!= 0);
6901 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6902 /* Make and build the new AVZ */
6903 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6904 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6905 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6907 /* Diff old AVZ with new one */
6911 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6912 spa
->spa_all_vdev_zaps
);
6913 zap_cursor_retrieve(&zc
, &za
) == 0;
6914 zap_cursor_advance(&zc
)) {
6915 uint64_t vdzap
= za
.za_first_integer
;
6916 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6919 * ZAP is listed in old AVZ but not in new one;
6922 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6927 zap_cursor_fini(&zc
);
6929 /* Destroy the old AVZ */
6930 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6931 spa
->spa_all_vdev_zaps
, tx
));
6933 /* Replace the old AVZ in the dir obj with the new one */
6934 VERIFY0(zap_update(spa
->spa_meta_objset
,
6935 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6936 sizeof (new_avz
), 1, &new_avz
, tx
));
6938 spa
->spa_all_vdev_zaps
= new_avz
;
6939 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6943 /* Walk through the AVZ and destroy all listed ZAPs */
6944 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6945 spa
->spa_all_vdev_zaps
);
6946 zap_cursor_retrieve(&zc
, &za
) == 0;
6947 zap_cursor_advance(&zc
)) {
6948 uint64_t zap
= za
.za_first_integer
;
6949 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6952 zap_cursor_fini(&zc
);
6954 /* Destroy and unlink the AVZ itself */
6955 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6956 spa
->spa_all_vdev_zaps
, tx
));
6957 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6958 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6959 spa
->spa_all_vdev_zaps
= 0;
6962 if (spa
->spa_all_vdev_zaps
== 0) {
6963 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6964 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6965 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6967 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6969 /* Create ZAPs for vdevs that don't have them. */
6970 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6972 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6973 dmu_tx_get_txg(tx
), B_FALSE
);
6976 * If we're upgrading the spa version then make sure that
6977 * the config object gets updated with the correct version.
6979 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6980 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6981 spa
->spa_uberblock
.ub_version
);
6983 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6985 nvlist_free(spa
->spa_config_syncing
);
6986 spa
->spa_config_syncing
= config
;
6988 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6992 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6994 uint64_t *versionp
= arg
;
6995 uint64_t version
= *versionp
;
6996 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6999 * Setting the version is special cased when first creating the pool.
7001 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
7003 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
7004 ASSERT(version
>= spa_version(spa
));
7006 spa
->spa_uberblock
.ub_version
= version
;
7007 vdev_config_dirty(spa
->spa_root_vdev
);
7008 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
7012 * Set zpool properties.
7015 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
7017 nvlist_t
*nvp
= arg
;
7018 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
7019 objset_t
*mos
= spa
->spa_meta_objset
;
7020 nvpair_t
*elem
= NULL
;
7022 mutex_enter(&spa
->spa_props_lock
);
7024 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
7026 char *strval
, *fname
;
7028 const char *propname
;
7029 zprop_type_t proptype
;
7032 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
7033 case ZPOOL_PROP_INVAL
:
7035 * We checked this earlier in spa_prop_validate().
7037 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
7039 fname
= strchr(nvpair_name(elem
), '@') + 1;
7040 VERIFY0(zfeature_lookup_name(fname
, &fid
));
7042 spa_feature_enable(spa
, fid
, tx
);
7043 spa_history_log_internal(spa
, "set", tx
,
7044 "%s=enabled", nvpair_name(elem
));
7047 case ZPOOL_PROP_VERSION
:
7048 intval
= fnvpair_value_uint64(elem
);
7050 * The version is synced separately before other
7051 * properties and should be correct by now.
7053 ASSERT3U(spa_version(spa
), >=, intval
);
7056 case ZPOOL_PROP_ALTROOT
:
7058 * 'altroot' is a non-persistent property. It should
7059 * have been set temporarily at creation or import time.
7061 ASSERT(spa
->spa_root
!= NULL
);
7064 case ZPOOL_PROP_READONLY
:
7065 case ZPOOL_PROP_CACHEFILE
:
7067 * 'readonly' and 'cachefile' are also non-persisitent
7071 case ZPOOL_PROP_COMMENT
:
7072 strval
= fnvpair_value_string(elem
);
7073 if (spa
->spa_comment
!= NULL
)
7074 spa_strfree(spa
->spa_comment
);
7075 spa
->spa_comment
= spa_strdup(strval
);
7077 * We need to dirty the configuration on all the vdevs
7078 * so that their labels get updated. It's unnecessary
7079 * to do this for pool creation since the vdev's
7080 * configuration has already been dirtied.
7082 if (tx
->tx_txg
!= TXG_INITIAL
)
7083 vdev_config_dirty(spa
->spa_root_vdev
);
7084 spa_history_log_internal(spa
, "set", tx
,
7085 "%s=%s", nvpair_name(elem
), strval
);
7089 * Set pool property values in the poolprops mos object.
7091 if (spa
->spa_pool_props_object
== 0) {
7092 spa
->spa_pool_props_object
=
7093 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
7094 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
7098 /* normalize the property name */
7099 propname
= zpool_prop_to_name(prop
);
7100 proptype
= zpool_prop_get_type(prop
);
7102 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
7103 ASSERT(proptype
== PROP_TYPE_STRING
);
7104 strval
= fnvpair_value_string(elem
);
7105 VERIFY0(zap_update(mos
,
7106 spa
->spa_pool_props_object
, propname
,
7107 1, strlen(strval
) + 1, strval
, tx
));
7108 spa_history_log_internal(spa
, "set", tx
,
7109 "%s=%s", nvpair_name(elem
), strval
);
7110 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
7111 intval
= fnvpair_value_uint64(elem
);
7113 if (proptype
== PROP_TYPE_INDEX
) {
7115 VERIFY0(zpool_prop_index_to_string(
7116 prop
, intval
, &unused
));
7118 VERIFY0(zap_update(mos
,
7119 spa
->spa_pool_props_object
, propname
,
7120 8, 1, &intval
, tx
));
7121 spa_history_log_internal(spa
, "set", tx
,
7122 "%s=%lld", nvpair_name(elem
), intval
);
7124 ASSERT(0); /* not allowed */
7128 case ZPOOL_PROP_DELEGATION
:
7129 spa
->spa_delegation
= intval
;
7131 case ZPOOL_PROP_BOOTFS
:
7132 spa
->spa_bootfs
= intval
;
7134 case ZPOOL_PROP_FAILUREMODE
:
7135 spa
->spa_failmode
= intval
;
7137 case ZPOOL_PROP_AUTOEXPAND
:
7138 spa
->spa_autoexpand
= intval
;
7139 if (tx
->tx_txg
!= TXG_INITIAL
)
7140 spa_async_request(spa
,
7141 SPA_ASYNC_AUTOEXPAND
);
7143 case ZPOOL_PROP_MULTIHOST
:
7144 spa
->spa_multihost
= intval
;
7146 case ZPOOL_PROP_DEDUPDITTO
:
7147 spa
->spa_dedup_ditto
= intval
;
7156 mutex_exit(&spa
->spa_props_lock
);
7160 * Perform one-time upgrade on-disk changes. spa_version() does not
7161 * reflect the new version this txg, so there must be no changes this
7162 * txg to anything that the upgrade code depends on after it executes.
7163 * Therefore this must be called after dsl_pool_sync() does the sync
7167 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
7169 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7171 ASSERT(spa
->spa_sync_pass
== 1);
7173 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
7175 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
7176 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
7177 dsl_pool_create_origin(dp
, tx
);
7179 /* Keeping the origin open increases spa_minref */
7180 spa
->spa_minref
+= 3;
7183 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
7184 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
7185 dsl_pool_upgrade_clones(dp
, tx
);
7188 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
7189 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
7190 dsl_pool_upgrade_dir_clones(dp
, tx
);
7192 /* Keeping the freedir open increases spa_minref */
7193 spa
->spa_minref
+= 3;
7196 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
7197 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
7198 spa_feature_create_zap_objects(spa
, tx
);
7202 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
7203 * when possibility to use lz4 compression for metadata was added
7204 * Old pools that have this feature enabled must be upgraded to have
7205 * this feature active
7207 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
7208 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
7209 SPA_FEATURE_LZ4_COMPRESS
);
7210 boolean_t lz4_ac
= spa_feature_is_active(spa
,
7211 SPA_FEATURE_LZ4_COMPRESS
);
7213 if (lz4_en
&& !lz4_ac
)
7214 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
7218 * If we haven't written the salt, do so now. Note that the
7219 * feature may not be activated yet, but that's fine since
7220 * the presence of this ZAP entry is backwards compatible.
7222 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
7223 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
7224 VERIFY0(zap_add(spa
->spa_meta_objset
,
7225 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
7226 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
7227 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
7230 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
7234 vdev_indirect_state_sync_verify(vdev_t
*vd
)
7236 ASSERTV(vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
);
7237 ASSERTV(vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
);
7239 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
7240 ASSERT(vim
!= NULL
);
7241 ASSERT(vib
!= NULL
);
7244 if (vdev_obsolete_sm_object(vd
) != 0) {
7245 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
7246 ASSERT(vd
->vdev_removing
||
7247 vd
->vdev_ops
== &vdev_indirect_ops
);
7248 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
7249 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
7251 ASSERT3U(vdev_obsolete_sm_object(vd
), ==,
7252 space_map_object(vd
->vdev_obsolete_sm
));
7253 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
7254 space_map_allocated(vd
->vdev_obsolete_sm
));
7256 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
7259 * Since frees / remaps to an indirect vdev can only
7260 * happen in syncing context, the obsolete segments
7261 * tree must be empty when we start syncing.
7263 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
7267 * Sync the specified transaction group. New blocks may be dirtied as
7268 * part of the process, so we iterate until it converges.
7271 spa_sync(spa_t
*spa
, uint64_t txg
)
7273 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7274 objset_t
*mos
= spa
->spa_meta_objset
;
7275 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
7276 vdev_t
*rvd
= spa
->spa_root_vdev
;
7280 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
7281 zfs_vdev_queue_depth_pct
/ 100;
7283 VERIFY(spa_writeable(spa
));
7286 * Wait for i/os issued in open context that need to complete
7287 * before this txg syncs.
7289 VERIFY0(zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]));
7290 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
, 0);
7293 * Lock out configuration changes.
7295 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
7297 spa
->spa_syncing_txg
= txg
;
7298 spa
->spa_sync_pass
= 0;
7300 mutex_enter(&spa
->spa_alloc_lock
);
7301 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7302 mutex_exit(&spa
->spa_alloc_lock
);
7305 * If there are any pending vdev state changes, convert them
7306 * into config changes that go out with this transaction group.
7308 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
7309 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
7311 * We need the write lock here because, for aux vdevs,
7312 * calling vdev_config_dirty() modifies sav_config.
7313 * This is ugly and will become unnecessary when we
7314 * eliminate the aux vdev wart by integrating all vdevs
7315 * into the root vdev tree.
7317 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7318 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
7319 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
7320 vdev_state_clean(vd
);
7321 vdev_config_dirty(vd
);
7323 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7324 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7326 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7328 tx
= dmu_tx_create_assigned(dp
, txg
);
7330 spa
->spa_sync_starttime
= gethrtime();
7331 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
7332 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
7333 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
7334 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
7337 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
7338 * set spa_deflate if we have no raid-z vdevs.
7340 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
7341 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
7344 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
7345 vd
= rvd
->vdev_child
[i
];
7346 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
7349 if (i
== rvd
->vdev_children
) {
7350 spa
->spa_deflate
= TRUE
;
7351 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
7352 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
7353 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
7358 * Set the top-level vdev's max queue depth. Evaluate each
7359 * top-level's async write queue depth in case it changed.
7360 * The max queue depth will not change in the middle of syncing
7363 uint64_t queue_depth_total
= 0;
7364 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
7365 vdev_t
*tvd
= rvd
->vdev_child
[c
];
7366 metaslab_group_t
*mg
= tvd
->vdev_mg
;
7368 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
7369 !metaslab_group_initialized(mg
))
7373 * It is safe to do a lock-free check here because only async
7374 * allocations look at mg_max_alloc_queue_depth, and async
7375 * allocations all happen from spa_sync().
7377 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
7378 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
7379 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
7381 metaslab_class_t
*mc
= spa_normal_class(spa
);
7382 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
7383 mc
->mc_alloc_max_slots
= queue_depth_total
;
7384 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
7386 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
7387 max_queue_depth
* rvd
->vdev_children
);
7389 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
7390 vdev_t
*vd
= rvd
->vdev_child
[c
];
7391 vdev_indirect_state_sync_verify(vd
);
7393 if (vdev_indirect_should_condense(vd
)) {
7394 spa_condense_indirect_start_sync(vd
, tx
);
7400 * Iterate to convergence.
7403 int pass
= ++spa
->spa_sync_pass
;
7405 spa_sync_config_object(spa
, tx
);
7406 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
7407 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
7408 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
7409 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
7410 spa_errlog_sync(spa
, txg
);
7411 dsl_pool_sync(dp
, txg
);
7413 if (pass
< zfs_sync_pass_deferred_free
) {
7414 spa_sync_frees(spa
, free_bpl
, tx
);
7417 * We can not defer frees in pass 1, because
7418 * we sync the deferred frees later in pass 1.
7420 ASSERT3U(pass
, >, 1);
7421 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
7422 &spa
->spa_deferred_bpobj
, tx
);
7426 dsl_scan_sync(dp
, tx
);
7428 if (spa
->spa_vdev_removal
!= NULL
)
7431 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
7436 spa_sync_upgrades(spa
, tx
);
7438 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
7440 * Note: We need to check if the MOS is dirty
7441 * because we could have marked the MOS dirty
7442 * without updating the uberblock (e.g. if we
7443 * have sync tasks but no dirty user data). We
7444 * need to check the uberblock's rootbp because
7445 * it is updated if we have synced out dirty
7446 * data (though in this case the MOS will most
7447 * likely also be dirty due to second order
7448 * effects, we don't want to rely on that here).
7450 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
7451 !dmu_objset_is_dirty(mos
, txg
)) {
7453 * Nothing changed on the first pass,
7454 * therefore this TXG is a no-op. Avoid
7455 * syncing deferred frees, so that we
7456 * can keep this TXG as a no-op.
7458 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
7460 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7461 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
7464 spa_sync_deferred_frees(spa
, tx
);
7467 } while (dmu_objset_is_dirty(mos
, txg
));
7470 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
7472 * Make sure that the number of ZAPs for all the vdevs matches
7473 * the number of ZAPs in the per-vdev ZAP list. This only gets
7474 * called if the config is dirty; otherwise there may be
7475 * outstanding AVZ operations that weren't completed in
7476 * spa_sync_config_object.
7478 uint64_t all_vdev_zap_entry_count
;
7479 ASSERT0(zap_count(spa
->spa_meta_objset
,
7480 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
7481 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
7482 all_vdev_zap_entry_count
);
7486 if (spa
->spa_vdev_removal
!= NULL
) {
7487 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
7491 * Rewrite the vdev configuration (which includes the uberblock)
7492 * to commit the transaction group.
7494 * If there are no dirty vdevs, we sync the uberblock to a few
7495 * random top-level vdevs that are known to be visible in the
7496 * config cache (see spa_vdev_add() for a complete description).
7497 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7501 * We hold SCL_STATE to prevent vdev open/close/etc.
7502 * while we're attempting to write the vdev labels.
7504 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
7506 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
7507 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
];
7509 int children
= rvd
->vdev_children
;
7510 int c0
= spa_get_random(children
);
7512 for (int c
= 0; c
< children
; c
++) {
7513 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
7514 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
7515 !vdev_is_concrete(vd
))
7517 svd
[svdcount
++] = vd
;
7518 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
7521 error
= vdev_config_sync(svd
, svdcount
, txg
);
7523 error
= vdev_config_sync(rvd
->vdev_child
,
7524 rvd
->vdev_children
, txg
);
7528 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
7530 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7534 zio_suspend(spa
, NULL
, ZIO_SUSPEND_IOERR
);
7535 zio_resume_wait(spa
);
7539 taskq_cancel_id(system_delay_taskq
, spa
->spa_deadman_tqid
);
7540 spa
->spa_deadman_tqid
= 0;
7543 * Clear the dirty config list.
7545 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
7546 vdev_config_clean(vd
);
7549 * Now that the new config has synced transactionally,
7550 * let it become visible to the config cache.
7552 if (spa
->spa_config_syncing
!= NULL
) {
7553 spa_config_set(spa
, spa
->spa_config_syncing
);
7554 spa
->spa_config_txg
= txg
;
7555 spa
->spa_config_syncing
= NULL
;
7558 dsl_pool_sync_done(dp
, txg
);
7560 mutex_enter(&spa
->spa_alloc_lock
);
7561 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7562 mutex_exit(&spa
->spa_alloc_lock
);
7565 * Update usable space statistics.
7567 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
))))
7568 vdev_sync_done(vd
, txg
);
7570 spa_update_dspace(spa
);
7573 * It had better be the case that we didn't dirty anything
7574 * since vdev_config_sync().
7576 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7577 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7578 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7580 spa
->spa_sync_pass
= 0;
7583 * Update the last synced uberblock here. We want to do this at
7584 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7585 * will be guaranteed that all the processing associated with
7586 * that txg has been completed.
7588 spa
->spa_ubsync
= spa
->spa_uberblock
;
7589 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7591 spa_handle_ignored_writes(spa
);
7594 * If any async tasks have been requested, kick them off.
7596 spa_async_dispatch(spa
);
7600 * Sync all pools. We don't want to hold the namespace lock across these
7601 * operations, so we take a reference on the spa_t and drop the lock during the
7605 spa_sync_allpools(void)
7608 mutex_enter(&spa_namespace_lock
);
7609 while ((spa
= spa_next(spa
)) != NULL
) {
7610 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7611 !spa_writeable(spa
) || spa_suspended(spa
))
7613 spa_open_ref(spa
, FTAG
);
7614 mutex_exit(&spa_namespace_lock
);
7615 txg_wait_synced(spa_get_dsl(spa
), 0);
7616 mutex_enter(&spa_namespace_lock
);
7617 spa_close(spa
, FTAG
);
7619 mutex_exit(&spa_namespace_lock
);
7623 * ==========================================================================
7624 * Miscellaneous routines
7625 * ==========================================================================
7629 * Remove all pools in the system.
7637 * Remove all cached state. All pools should be closed now,
7638 * so every spa in the AVL tree should be unreferenced.
7640 mutex_enter(&spa_namespace_lock
);
7641 while ((spa
= spa_next(NULL
)) != NULL
) {
7643 * Stop async tasks. The async thread may need to detach
7644 * a device that's been replaced, which requires grabbing
7645 * spa_namespace_lock, so we must drop it here.
7647 spa_open_ref(spa
, FTAG
);
7648 mutex_exit(&spa_namespace_lock
);
7649 spa_async_suspend(spa
);
7650 mutex_enter(&spa_namespace_lock
);
7651 spa_close(spa
, FTAG
);
7653 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7655 spa_deactivate(spa
);
7659 mutex_exit(&spa_namespace_lock
);
7663 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7668 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7672 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7673 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7674 if (vd
->vdev_guid
== guid
)
7678 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7679 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7680 if (vd
->vdev_guid
== guid
)
7689 spa_upgrade(spa_t
*spa
, uint64_t version
)
7691 ASSERT(spa_writeable(spa
));
7693 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7696 * This should only be called for a non-faulted pool, and since a
7697 * future version would result in an unopenable pool, this shouldn't be
7700 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7701 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7703 spa
->spa_uberblock
.ub_version
= version
;
7704 vdev_config_dirty(spa
->spa_root_vdev
);
7706 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7708 txg_wait_synced(spa_get_dsl(spa
), 0);
7712 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7716 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7718 for (i
= 0; i
< sav
->sav_count
; i
++)
7719 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7722 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7723 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7724 &spareguid
) == 0 && spareguid
== guid
)
7732 * Check if a pool has an active shared spare device.
7733 * Note: reference count of an active spare is 2, as a spare and as a replace
7736 spa_has_active_shared_spare(spa_t
*spa
)
7740 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7742 for (i
= 0; i
< sav
->sav_count
; i
++) {
7743 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7744 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7753 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7755 sysevent_t
*ev
= NULL
;
7759 resource
= zfs_event_create(spa
, vd
, FM_SYSEVENT_CLASS
, name
, hist_nvl
);
7761 ev
= kmem_alloc(sizeof (sysevent_t
), KM_SLEEP
);
7762 ev
->resource
= resource
;
7769 spa_event_post(sysevent_t
*ev
)
7773 zfs_zevent_post(ev
->resource
, NULL
, zfs_zevent_post_cb
);
7774 kmem_free(ev
, sizeof (*ev
));
7780 * Post a zevent corresponding to the given sysevent. The 'name' must be one
7781 * of the event definitions in sys/sysevent/eventdefs.h. The payload will be
7782 * filled in from the spa and (optionally) the vdev. This doesn't do anything
7783 * in the userland libzpool, as we don't want consumers to misinterpret ztest
7784 * or zdb as real changes.
7787 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7789 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));
7792 #if defined(_KERNEL)
7793 /* state manipulation functions */
7794 EXPORT_SYMBOL(spa_open
);
7795 EXPORT_SYMBOL(spa_open_rewind
);
7796 EXPORT_SYMBOL(spa_get_stats
);
7797 EXPORT_SYMBOL(spa_create
);
7798 EXPORT_SYMBOL(spa_import
);
7799 EXPORT_SYMBOL(spa_tryimport
);
7800 EXPORT_SYMBOL(spa_destroy
);
7801 EXPORT_SYMBOL(spa_export
);
7802 EXPORT_SYMBOL(spa_reset
);
7803 EXPORT_SYMBOL(spa_async_request
);
7804 EXPORT_SYMBOL(spa_async_suspend
);
7805 EXPORT_SYMBOL(spa_async_resume
);
7806 EXPORT_SYMBOL(spa_inject_addref
);
7807 EXPORT_SYMBOL(spa_inject_delref
);
7808 EXPORT_SYMBOL(spa_scan_stat_init
);
7809 EXPORT_SYMBOL(spa_scan_get_stats
);
7811 /* device maniion */
7812 EXPORT_SYMBOL(spa_vdev_add
);
7813 EXPORT_SYMBOL(spa_vdev_attach
);
7814 EXPORT_SYMBOL(spa_vdev_detach
);
7815 EXPORT_SYMBOL(spa_vdev_setpath
);
7816 EXPORT_SYMBOL(spa_vdev_setfru
);
7817 EXPORT_SYMBOL(spa_vdev_split_mirror
);
7819 /* spare statech is global across all pools) */
7820 EXPORT_SYMBOL(spa_spare_add
);
7821 EXPORT_SYMBOL(spa_spare_remove
);
7822 EXPORT_SYMBOL(spa_spare_exists
);
7823 EXPORT_SYMBOL(spa_spare_activate
);
7825 /* L2ARC statech is global across all pools) */
7826 EXPORT_SYMBOL(spa_l2cache_add
);
7827 EXPORT_SYMBOL(spa_l2cache_remove
);
7828 EXPORT_SYMBOL(spa_l2cache_exists
);
7829 EXPORT_SYMBOL(spa_l2cache_activate
);
7830 EXPORT_SYMBOL(spa_l2cache_drop
);
7833 EXPORT_SYMBOL(spa_scan
);
7834 EXPORT_SYMBOL(spa_scan_stop
);
7837 EXPORT_SYMBOL(spa_sync
); /* only for DMU use */
7838 EXPORT_SYMBOL(spa_sync_allpools
);
7841 EXPORT_SYMBOL(spa_prop_set
);
7842 EXPORT_SYMBOL(spa_prop_get
);
7843 EXPORT_SYMBOL(spa_prop_clear_bootfs
);
7845 /* asynchronous event notification */
7846 EXPORT_SYMBOL(spa_event_notify
);
7849 #if defined(_KERNEL)
7850 module_param(spa_load_verify_maxinflight
, int, 0644);
7851 MODULE_PARM_DESC(spa_load_verify_maxinflight
,
7852 "Max concurrent traversal I/Os while verifying pool during import -X");
7854 module_param(spa_load_verify_metadata
, int, 0644);
7855 MODULE_PARM_DESC(spa_load_verify_metadata
,
7856 "Set to traverse metadata on pool import");
7858 module_param(spa_load_verify_data
, int, 0644);
7859 MODULE_PARM_DESC(spa_load_verify_data
,
7860 "Set to traverse data on pool import");
7862 module_param(spa_load_print_vdev_tree
, int, 0644);
7863 MODULE_PARM_DESC(spa_load_print_vdev_tree
,
7864 "Print vdev tree to zfs_dbgmsg during pool import");
7867 module_param(zio_taskq_batch_pct
, uint
, 0444);
7868 MODULE_PARM_DESC(zio_taskq_batch_pct
,
7869 "Percentage of CPUs to run an IO worker thread");
7872 module_param(zfs_max_missing_tvds
, ulong
, 0644);
7873 MODULE_PARM_DESC(zfs_max_missing_tvds
,
7874 "Allow importing pool with up to this number of missing top-level vdevs"
7875 " (in read-only mode)");