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, 2018 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2016 Toomas Soome <tsoome@me.com>
30 * Copyright 2017 Joyent, Inc.
31 * Copyright (c) 2017 Datto Inc.
32 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
36 * SPA: Storage Pool Allocator
38 * This file contains all the routines used when modifying on-disk SPA state.
39 * This includes opening, importing, destroying, exporting a pool, and syncing a
43 #include <sys/zfs_context.h>
44 #include <sys/fm/fs/zfs.h>
45 #include <sys/spa_impl.h>
47 #include <sys/zio_checksum.h>
49 #include <sys/dmu_tx.h>
53 #include <sys/vdev_impl.h>
54 #include <sys/vdev_removal.h>
55 #include <sys/vdev_indirect_mapping.h>
56 #include <sys/vdev_indirect_births.h>
57 #include <sys/metaslab.h>
58 #include <sys/metaslab_impl.h>
59 #include <sys/uberblock_impl.h>
62 #include <sys/bpobj.h>
63 #include <sys/dmu_traverse.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/unique.h>
66 #include <sys/dsl_pool.h>
67 #include <sys/dsl_dataset.h>
68 #include <sys/dsl_dir.h>
69 #include <sys/dsl_prop.h>
70 #include <sys/dsl_synctask.h>
71 #include <sys/fs/zfs.h>
73 #include <sys/callb.h>
74 #include <sys/systeminfo.h>
75 #include <sys/spa_boot.h>
76 #include <sys/zfs_ioctl.h>
77 #include <sys/dsl_scan.h>
78 #include <sys/zfeature.h>
79 #include <sys/dsl_destroy.h>
83 #include <sys/bootprops.h>
84 #include <sys/callb.h>
85 #include <sys/cpupart.h>
87 #include <sys/sysdc.h>
92 #include "zfs_comutil.h"
95 * The interval, in seconds, at which failed configuration cache file writes
98 int zfs_ccw_retry_interval
= 300;
100 typedef enum zti_modes
{
101 ZTI_MODE_FIXED
, /* value is # of threads (min 1) */
102 ZTI_MODE_BATCH
, /* cpu-intensive; value is ignored */
103 ZTI_MODE_NULL
, /* don't create a taskq */
107 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
108 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
109 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
111 #define ZTI_N(n) ZTI_P(n, 1)
112 #define ZTI_ONE ZTI_N(1)
114 typedef struct zio_taskq_info
{
115 zti_modes_t zti_mode
;
120 static const char *const zio_taskq_types
[ZIO_TASKQ_TYPES
] = {
121 "issue", "issue_high", "intr", "intr_high"
125 * This table defines the taskq settings for each ZFS I/O type. When
126 * initializing a pool, we use this table to create an appropriately sized
127 * taskq. Some operations are low volume and therefore have a small, static
128 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
129 * macros. Other operations process a large amount of data; the ZTI_BATCH
130 * macro causes us to create a taskq oriented for throughput. Some operations
131 * are so high frequency and short-lived that the taskq itself can become a a
132 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
133 * additional degree of parallelism specified by the number of threads per-
134 * taskq and the number of taskqs; when dispatching an event in this case, the
135 * particular taskq is chosen at random.
137 * The different taskq priorities are to handle the different contexts (issue
138 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
139 * need to be handled with minimum delay.
141 const zio_taskq_info_t zio_taskqs
[ZIO_TYPES
][ZIO_TASKQ_TYPES
] = {
142 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
143 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* NULL */
144 { ZTI_N(8), ZTI_NULL
, ZTI_P(12, 8), ZTI_NULL
}, /* READ */
145 { ZTI_BATCH
, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
146 { ZTI_P(12, 8), ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* FREE */
147 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* CLAIM */
148 { ZTI_ONE
, ZTI_NULL
, ZTI_ONE
, ZTI_NULL
}, /* IOCTL */
151 static void spa_sync_version(void *arg
, dmu_tx_t
*tx
);
152 static void spa_sync_props(void *arg
, dmu_tx_t
*tx
);
153 static boolean_t
spa_has_active_shared_spare(spa_t
*spa
);
154 static int spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
,
155 boolean_t reloading
);
156 static void spa_vdev_resilver_done(spa_t
*spa
);
158 uint_t zio_taskq_batch_pct
= 75; /* 1 thread per cpu in pset */
159 id_t zio_taskq_psrset_bind
= PS_NONE
;
160 boolean_t zio_taskq_sysdc
= B_TRUE
; /* use SDC scheduling class */
161 uint_t zio_taskq_basedc
= 80; /* base duty cycle */
163 boolean_t spa_create_process
= B_TRUE
; /* no process ==> no sysdc */
164 extern int zfs_sync_pass_deferred_free
;
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 boolean_t 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 uint64_t 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 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 src
= ZPROP_SRC_DEFAULT
;
298 src
= ZPROP_SRC_LOCAL
;
299 spa_prop_add_list(*nvp
, ZPOOL_PROP_VERSION
, NULL
, version
, src
);
304 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
305 * when opening pools before this version freedir will be NULL.
307 if (pool
->dp_free_dir
!= NULL
) {
308 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
, NULL
,
309 dsl_dir_phys(pool
->dp_free_dir
)->dd_used_bytes
,
312 spa_prop_add_list(*nvp
, ZPOOL_PROP_FREEING
,
316 if (pool
->dp_leak_dir
!= NULL
) {
317 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
, NULL
,
318 dsl_dir_phys(pool
->dp_leak_dir
)->dd_used_bytes
,
321 spa_prop_add_list(*nvp
, ZPOOL_PROP_LEAKED
,
326 spa_prop_add_list(*nvp
, ZPOOL_PROP_GUID
, NULL
, spa_guid(spa
), src
);
328 if (spa
->spa_comment
!= NULL
) {
329 spa_prop_add_list(*nvp
, ZPOOL_PROP_COMMENT
, spa
->spa_comment
,
333 if (spa
->spa_root
!= NULL
)
334 spa_prop_add_list(*nvp
, ZPOOL_PROP_ALTROOT
, spa
->spa_root
,
337 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
338 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
339 MIN(zfs_max_recordsize
, SPA_MAXBLOCKSIZE
), ZPROP_SRC_NONE
);
341 spa_prop_add_list(*nvp
, ZPOOL_PROP_MAXBLOCKSIZE
, NULL
,
342 SPA_OLD_MAXBLOCKSIZE
, ZPROP_SRC_NONE
);
345 if ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
346 if (dp
->scd_path
== NULL
) {
347 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
348 "none", 0, ZPROP_SRC_LOCAL
);
349 } else if (strcmp(dp
->scd_path
, spa_config_path
) != 0) {
350 spa_prop_add_list(*nvp
, ZPOOL_PROP_CACHEFILE
,
351 dp
->scd_path
, 0, ZPROP_SRC_LOCAL
);
357 * Get zpool property values.
360 spa_prop_get(spa_t
*spa
, nvlist_t
**nvp
)
362 objset_t
*mos
= spa
->spa_meta_objset
;
367 VERIFY(nvlist_alloc(nvp
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
369 mutex_enter(&spa
->spa_props_lock
);
372 * Get properties from the spa config.
374 spa_prop_get_config(spa
, nvp
);
376 /* If no pool property object, no more prop to get. */
377 if (mos
== NULL
|| spa
->spa_pool_props_object
== 0) {
378 mutex_exit(&spa
->spa_props_lock
);
383 * Get properties from the MOS pool property object.
385 for (zap_cursor_init(&zc
, mos
, spa
->spa_pool_props_object
);
386 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
387 zap_cursor_advance(&zc
)) {
390 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
393 if ((prop
= zpool_name_to_prop(za
.za_name
)) == ZPOOL_PROP_INVAL
)
396 switch (za
.za_integer_length
) {
398 /* integer property */
399 if (za
.za_first_integer
!=
400 zpool_prop_default_numeric(prop
))
401 src
= ZPROP_SRC_LOCAL
;
403 if (prop
== ZPOOL_PROP_BOOTFS
) {
405 dsl_dataset_t
*ds
= NULL
;
407 dp
= spa_get_dsl(spa
);
408 dsl_pool_config_enter(dp
, FTAG
);
409 if (err
= dsl_dataset_hold_obj(dp
,
410 za
.za_first_integer
, FTAG
, &ds
)) {
411 dsl_pool_config_exit(dp
, FTAG
);
415 strval
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
,
417 dsl_dataset_name(ds
, strval
);
418 dsl_dataset_rele(ds
, FTAG
);
419 dsl_pool_config_exit(dp
, FTAG
);
422 intval
= za
.za_first_integer
;
425 spa_prop_add_list(*nvp
, prop
, strval
, intval
, src
);
428 kmem_free(strval
, ZFS_MAX_DATASET_NAME_LEN
);
433 /* string property */
434 strval
= kmem_alloc(za
.za_num_integers
, KM_SLEEP
);
435 err
= zap_lookup(mos
, spa
->spa_pool_props_object
,
436 za
.za_name
, 1, za
.za_num_integers
, strval
);
438 kmem_free(strval
, za
.za_num_integers
);
441 spa_prop_add_list(*nvp
, prop
, strval
, 0, src
);
442 kmem_free(strval
, za
.za_num_integers
);
449 zap_cursor_fini(&zc
);
450 mutex_exit(&spa
->spa_props_lock
);
452 if (err
&& err
!= ENOENT
) {
462 * Validate the given pool properties nvlist and modify the list
463 * for the property values to be set.
466 spa_prop_validate(spa_t
*spa
, nvlist_t
*props
)
469 int error
= 0, reset_bootfs
= 0;
471 boolean_t has_feature
= B_FALSE
;
474 while ((elem
= nvlist_next_nvpair(props
, elem
)) != NULL
) {
476 char *strval
, *slash
, *check
, *fname
;
477 const char *propname
= nvpair_name(elem
);
478 zpool_prop_t prop
= zpool_name_to_prop(propname
);
481 case ZPOOL_PROP_INVAL
:
482 if (!zpool_prop_feature(propname
)) {
483 error
= SET_ERROR(EINVAL
);
488 * Sanitize the input.
490 if (nvpair_type(elem
) != DATA_TYPE_UINT64
) {
491 error
= SET_ERROR(EINVAL
);
495 if (nvpair_value_uint64(elem
, &intval
) != 0) {
496 error
= SET_ERROR(EINVAL
);
501 error
= SET_ERROR(EINVAL
);
505 fname
= strchr(propname
, '@') + 1;
506 if (zfeature_lookup_name(fname
, NULL
) != 0) {
507 error
= SET_ERROR(EINVAL
);
511 has_feature
= B_TRUE
;
514 case ZPOOL_PROP_VERSION
:
515 error
= nvpair_value_uint64(elem
, &intval
);
517 (intval
< spa_version(spa
) ||
518 intval
> SPA_VERSION_BEFORE_FEATURES
||
520 error
= SET_ERROR(EINVAL
);
523 case ZPOOL_PROP_DELEGATION
:
524 case ZPOOL_PROP_AUTOREPLACE
:
525 case ZPOOL_PROP_LISTSNAPS
:
526 case ZPOOL_PROP_AUTOEXPAND
:
527 error
= nvpair_value_uint64(elem
, &intval
);
528 if (!error
&& intval
> 1)
529 error
= SET_ERROR(EINVAL
);
532 case ZPOOL_PROP_BOOTFS
:
534 * If the pool version is less than SPA_VERSION_BOOTFS,
535 * or the pool is still being created (version == 0),
536 * the bootfs property cannot be set.
538 if (spa_version(spa
) < SPA_VERSION_BOOTFS
) {
539 error
= SET_ERROR(ENOTSUP
);
544 * Make sure the vdev config is bootable
546 if (!vdev_is_bootable(spa
->spa_root_vdev
)) {
547 error
= SET_ERROR(ENOTSUP
);
553 error
= nvpair_value_string(elem
, &strval
);
559 if (strval
== NULL
|| strval
[0] == '\0') {
560 objnum
= zpool_prop_default_numeric(
565 if (error
= dmu_objset_hold(strval
, FTAG
, &os
))
569 * Must be ZPL, and its property settings
570 * must be supported by GRUB (compression
571 * is not gzip, and large blocks are not used).
574 if (dmu_objset_type(os
) != DMU_OST_ZFS
) {
575 error
= SET_ERROR(ENOTSUP
);
577 dsl_prop_get_int_ds(dmu_objset_ds(os
),
578 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
580 !BOOTFS_COMPRESS_VALID(propval
)) {
581 error
= SET_ERROR(ENOTSUP
);
583 objnum
= dmu_objset_id(os
);
585 dmu_objset_rele(os
, FTAG
);
589 case ZPOOL_PROP_FAILUREMODE
:
590 error
= nvpair_value_uint64(elem
, &intval
);
591 if (!error
&& (intval
< ZIO_FAILURE_MODE_WAIT
||
592 intval
> ZIO_FAILURE_MODE_PANIC
))
593 error
= SET_ERROR(EINVAL
);
596 * This is a special case which only occurs when
597 * the pool has completely failed. This allows
598 * the user to change the in-core failmode property
599 * without syncing it out to disk (I/Os might
600 * currently be blocked). We do this by returning
601 * EIO to the caller (spa_prop_set) to trick it
602 * into thinking we encountered a property validation
605 if (!error
&& spa_suspended(spa
)) {
606 spa
->spa_failmode
= intval
;
607 error
= SET_ERROR(EIO
);
611 case ZPOOL_PROP_CACHEFILE
:
612 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
615 if (strval
[0] == '\0')
618 if (strcmp(strval
, "none") == 0)
621 if (strval
[0] != '/') {
622 error
= SET_ERROR(EINVAL
);
626 slash
= strrchr(strval
, '/');
627 ASSERT(slash
!= NULL
);
629 if (slash
[1] == '\0' || strcmp(slash
, "/.") == 0 ||
630 strcmp(slash
, "/..") == 0)
631 error
= SET_ERROR(EINVAL
);
634 case ZPOOL_PROP_COMMENT
:
635 if ((error
= nvpair_value_string(elem
, &strval
)) != 0)
637 for (check
= strval
; *check
!= '\0'; check
++) {
639 * The kernel doesn't have an easy isprint()
640 * check. For this kernel check, we merely
641 * check ASCII apart from DEL. Fix this if
642 * there is an easy-to-use kernel isprint().
644 if (*check
>= 0x7f) {
645 error
= SET_ERROR(EINVAL
);
649 if (strlen(strval
) > ZPROP_MAX_COMMENT
)
653 case ZPOOL_PROP_DEDUPDITTO
:
654 if (spa_version(spa
) < SPA_VERSION_DEDUP
)
655 error
= SET_ERROR(ENOTSUP
);
657 error
= nvpair_value_uint64(elem
, &intval
);
659 intval
!= 0 && intval
< ZIO_DEDUPDITTO_MIN
)
660 error
= SET_ERROR(EINVAL
);
668 if (!error
&& reset_bootfs
) {
669 error
= nvlist_remove(props
,
670 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), DATA_TYPE_STRING
);
673 error
= nvlist_add_uint64(props
,
674 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), objnum
);
682 spa_configfile_set(spa_t
*spa
, nvlist_t
*nvp
, boolean_t need_sync
)
685 spa_config_dirent_t
*dp
;
687 if (nvlist_lookup_string(nvp
, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE
),
691 dp
= kmem_alloc(sizeof (spa_config_dirent_t
),
694 if (cachefile
[0] == '\0')
695 dp
->scd_path
= spa_strdup(spa_config_path
);
696 else if (strcmp(cachefile
, "none") == 0)
699 dp
->scd_path
= spa_strdup(cachefile
);
701 list_insert_head(&spa
->spa_config_list
, dp
);
703 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
707 spa_prop_set(spa_t
*spa
, nvlist_t
*nvp
)
710 nvpair_t
*elem
= NULL
;
711 boolean_t need_sync
= B_FALSE
;
713 if ((error
= spa_prop_validate(spa
, nvp
)) != 0)
716 while ((elem
= nvlist_next_nvpair(nvp
, elem
)) != NULL
) {
717 zpool_prop_t prop
= zpool_name_to_prop(nvpair_name(elem
));
719 if (prop
== ZPOOL_PROP_CACHEFILE
||
720 prop
== ZPOOL_PROP_ALTROOT
||
721 prop
== ZPOOL_PROP_READONLY
)
724 if (prop
== ZPOOL_PROP_VERSION
|| prop
== ZPOOL_PROP_INVAL
) {
727 if (prop
== ZPOOL_PROP_VERSION
) {
728 VERIFY(nvpair_value_uint64(elem
, &ver
) == 0);
730 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
731 ver
= SPA_VERSION_FEATURES
;
735 /* Save time if the version is already set. */
736 if (ver
== spa_version(spa
))
740 * In addition to the pool directory object, we might
741 * create the pool properties object, the features for
742 * read object, the features for write object, or the
743 * feature descriptions object.
745 error
= dsl_sync_task(spa
->spa_name
, NULL
,
746 spa_sync_version
, &ver
,
747 6, ZFS_SPACE_CHECK_RESERVED
);
758 return (dsl_sync_task(spa
->spa_name
, NULL
, spa_sync_props
,
759 nvp
, 6, ZFS_SPACE_CHECK_RESERVED
));
766 * If the bootfs property value is dsobj, clear it.
769 spa_prop_clear_bootfs(spa_t
*spa
, uint64_t dsobj
, dmu_tx_t
*tx
)
771 if (spa
->spa_bootfs
== dsobj
&& spa
->spa_pool_props_object
!= 0) {
772 VERIFY(zap_remove(spa
->spa_meta_objset
,
773 spa
->spa_pool_props_object
,
774 zpool_prop_to_name(ZPOOL_PROP_BOOTFS
), tx
) == 0);
781 spa_change_guid_check(void *arg
, dmu_tx_t
*tx
)
783 uint64_t *newguid
= arg
;
784 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
785 vdev_t
*rvd
= spa
->spa_root_vdev
;
788 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
789 vdev_state
= rvd
->vdev_state
;
790 spa_config_exit(spa
, SCL_STATE
, FTAG
);
792 if (vdev_state
!= VDEV_STATE_HEALTHY
)
793 return (SET_ERROR(ENXIO
));
795 ASSERT3U(spa_guid(spa
), !=, *newguid
);
801 spa_change_guid_sync(void *arg
, dmu_tx_t
*tx
)
803 uint64_t *newguid
= arg
;
804 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
806 vdev_t
*rvd
= spa
->spa_root_vdev
;
808 oldguid
= spa_guid(spa
);
810 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
811 rvd
->vdev_guid
= *newguid
;
812 rvd
->vdev_guid_sum
+= (*newguid
- oldguid
);
813 vdev_config_dirty(rvd
);
814 spa_config_exit(spa
, SCL_STATE
, FTAG
);
816 spa_history_log_internal(spa
, "guid change", tx
, "old=%llu new=%llu",
821 * Change the GUID for the pool. This is done so that we can later
822 * re-import a pool built from a clone of our own vdevs. We will modify
823 * the root vdev's guid, our own pool guid, and then mark all of our
824 * vdevs dirty. Note that we must make sure that all our vdevs are
825 * online when we do this, or else any vdevs that weren't present
826 * would be orphaned from our pool. We are also going to issue a
827 * sysevent to update any watchers.
830 spa_change_guid(spa_t
*spa
)
835 mutex_enter(&spa
->spa_vdev_top_lock
);
836 mutex_enter(&spa_namespace_lock
);
837 guid
= spa_generate_guid(NULL
);
839 error
= dsl_sync_task(spa
->spa_name
, spa_change_guid_check
,
840 spa_change_guid_sync
, &guid
, 5, ZFS_SPACE_CHECK_RESERVED
);
843 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
844 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_REGUID
);
847 mutex_exit(&spa_namespace_lock
);
848 mutex_exit(&spa
->spa_vdev_top_lock
);
854 * ==========================================================================
855 * SPA state manipulation (open/create/destroy/import/export)
856 * ==========================================================================
860 spa_error_entry_compare(const void *a
, const void *b
)
862 spa_error_entry_t
*sa
= (spa_error_entry_t
*)a
;
863 spa_error_entry_t
*sb
= (spa_error_entry_t
*)b
;
866 ret
= bcmp(&sa
->se_bookmark
, &sb
->se_bookmark
,
867 sizeof (zbookmark_phys_t
));
878 * Utility function which retrieves copies of the current logs and
879 * re-initializes them in the process.
882 spa_get_errlists(spa_t
*spa
, avl_tree_t
*last
, avl_tree_t
*scrub
)
884 ASSERT(MUTEX_HELD(&spa
->spa_errlist_lock
));
886 bcopy(&spa
->spa_errlist_last
, last
, sizeof (avl_tree_t
));
887 bcopy(&spa
->spa_errlist_scrub
, scrub
, sizeof (avl_tree_t
));
889 avl_create(&spa
->spa_errlist_scrub
,
890 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
891 offsetof(spa_error_entry_t
, se_avl
));
892 avl_create(&spa
->spa_errlist_last
,
893 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
894 offsetof(spa_error_entry_t
, se_avl
));
898 spa_taskqs_init(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
900 const zio_taskq_info_t
*ztip
= &zio_taskqs
[t
][q
];
901 enum zti_modes mode
= ztip
->zti_mode
;
902 uint_t value
= ztip
->zti_value
;
903 uint_t count
= ztip
->zti_count
;
904 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
907 boolean_t batch
= B_FALSE
;
909 if (mode
== ZTI_MODE_NULL
) {
911 tqs
->stqs_taskq
= NULL
;
915 ASSERT3U(count
, >, 0);
917 tqs
->stqs_count
= count
;
918 tqs
->stqs_taskq
= kmem_alloc(count
* sizeof (taskq_t
*), KM_SLEEP
);
922 ASSERT3U(value
, >=, 1);
923 value
= MAX(value
, 1);
928 flags
|= TASKQ_THREADS_CPU_PCT
;
929 value
= zio_taskq_batch_pct
;
933 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
935 zio_type_name
[t
], zio_taskq_types
[q
], mode
, value
);
939 for (uint_t i
= 0; i
< count
; i
++) {
943 (void) snprintf(name
, sizeof (name
), "%s_%s_%u",
944 zio_type_name
[t
], zio_taskq_types
[q
], i
);
946 (void) snprintf(name
, sizeof (name
), "%s_%s",
947 zio_type_name
[t
], zio_taskq_types
[q
]);
950 if (zio_taskq_sysdc
&& spa
->spa_proc
!= &p0
) {
952 flags
|= TASKQ_DC_BATCH
;
954 tq
= taskq_create_sysdc(name
, value
, 50, INT_MAX
,
955 spa
->spa_proc
, zio_taskq_basedc
, flags
);
957 pri_t pri
= maxclsyspri
;
959 * The write issue taskq can be extremely CPU
960 * intensive. Run it at slightly lower priority
961 * than the other taskqs.
963 if (t
== ZIO_TYPE_WRITE
&& q
== ZIO_TASKQ_ISSUE
)
966 tq
= taskq_create_proc(name
, value
, pri
, 50,
967 INT_MAX
, spa
->spa_proc
, flags
);
970 tqs
->stqs_taskq
[i
] = tq
;
975 spa_taskqs_fini(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
)
977 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
979 if (tqs
->stqs_taskq
== NULL
) {
980 ASSERT0(tqs
->stqs_count
);
984 for (uint_t i
= 0; i
< tqs
->stqs_count
; i
++) {
985 ASSERT3P(tqs
->stqs_taskq
[i
], !=, NULL
);
986 taskq_destroy(tqs
->stqs_taskq
[i
]);
989 kmem_free(tqs
->stqs_taskq
, tqs
->stqs_count
* sizeof (taskq_t
*));
990 tqs
->stqs_taskq
= NULL
;
994 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
995 * Note that a type may have multiple discrete taskqs to avoid lock contention
996 * on the taskq itself. In that case we choose which taskq at random by using
997 * the low bits of gethrtime().
1000 spa_taskq_dispatch_ent(spa_t
*spa
, zio_type_t t
, zio_taskq_type_t q
,
1001 task_func_t
*func
, void *arg
, uint_t flags
, taskq_ent_t
*ent
)
1003 spa_taskqs_t
*tqs
= &spa
->spa_zio_taskq
[t
][q
];
1006 ASSERT3P(tqs
->stqs_taskq
, !=, NULL
);
1007 ASSERT3U(tqs
->stqs_count
, !=, 0);
1009 if (tqs
->stqs_count
== 1) {
1010 tq
= tqs
->stqs_taskq
[0];
1012 tq
= tqs
->stqs_taskq
[gethrtime() % tqs
->stqs_count
];
1015 taskq_dispatch_ent(tq
, func
, arg
, flags
, ent
);
1019 spa_create_zio_taskqs(spa_t
*spa
)
1021 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1022 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1023 spa_taskqs_init(spa
, t
, q
);
1030 spa_thread(void *arg
)
1032 callb_cpr_t cprinfo
;
1035 user_t
*pu
= PTOU(curproc
);
1037 CALLB_CPR_INIT(&cprinfo
, &spa
->spa_proc_lock
, callb_generic_cpr
,
1040 ASSERT(curproc
!= &p0
);
1041 (void) snprintf(pu
->u_psargs
, sizeof (pu
->u_psargs
),
1042 "zpool-%s", spa
->spa_name
);
1043 (void) strlcpy(pu
->u_comm
, pu
->u_psargs
, sizeof (pu
->u_comm
));
1045 /* bind this thread to the requested psrset */
1046 if (zio_taskq_psrset_bind
!= PS_NONE
) {
1048 mutex_enter(&cpu_lock
);
1049 mutex_enter(&pidlock
);
1050 mutex_enter(&curproc
->p_lock
);
1052 if (cpupart_bind_thread(curthread
, zio_taskq_psrset_bind
,
1053 0, NULL
, NULL
) == 0) {
1054 curthread
->t_bind_pset
= zio_taskq_psrset_bind
;
1057 "Couldn't bind process for zfs pool \"%s\" to "
1058 "pset %d\n", spa
->spa_name
, zio_taskq_psrset_bind
);
1061 mutex_exit(&curproc
->p_lock
);
1062 mutex_exit(&pidlock
);
1063 mutex_exit(&cpu_lock
);
1067 if (zio_taskq_sysdc
) {
1068 sysdc_thread_enter(curthread
, 100, 0);
1071 spa
->spa_proc
= curproc
;
1072 spa
->spa_did
= curthread
->t_did
;
1074 spa_create_zio_taskqs(spa
);
1076 mutex_enter(&spa
->spa_proc_lock
);
1077 ASSERT(spa
->spa_proc_state
== SPA_PROC_CREATED
);
1079 spa
->spa_proc_state
= SPA_PROC_ACTIVE
;
1080 cv_broadcast(&spa
->spa_proc_cv
);
1082 CALLB_CPR_SAFE_BEGIN(&cprinfo
);
1083 while (spa
->spa_proc_state
== SPA_PROC_ACTIVE
)
1084 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1085 CALLB_CPR_SAFE_END(&cprinfo
, &spa
->spa_proc_lock
);
1087 ASSERT(spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
);
1088 spa
->spa_proc_state
= SPA_PROC_GONE
;
1089 spa
->spa_proc
= &p0
;
1090 cv_broadcast(&spa
->spa_proc_cv
);
1091 CALLB_CPR_EXIT(&cprinfo
); /* drops spa_proc_lock */
1093 mutex_enter(&curproc
->p_lock
);
1099 * Activate an uninitialized pool.
1102 spa_activate(spa_t
*spa
, int mode
)
1104 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
1106 spa
->spa_state
= POOL_STATE_ACTIVE
;
1107 spa
->spa_mode
= mode
;
1109 spa
->spa_normal_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1110 spa
->spa_log_class
= metaslab_class_create(spa
, zfs_metaslab_ops
);
1112 /* Try to create a covering process */
1113 mutex_enter(&spa
->spa_proc_lock
);
1114 ASSERT(spa
->spa_proc_state
== SPA_PROC_NONE
);
1115 ASSERT(spa
->spa_proc
== &p0
);
1118 /* Only create a process if we're going to be around a while. */
1119 if (spa_create_process
&& strcmp(spa
->spa_name
, TRYIMPORT_NAME
) != 0) {
1120 if (newproc(spa_thread
, (caddr_t
)spa
, syscid
, maxclsyspri
,
1122 spa
->spa_proc_state
= SPA_PROC_CREATED
;
1123 while (spa
->spa_proc_state
== SPA_PROC_CREATED
) {
1124 cv_wait(&spa
->spa_proc_cv
,
1125 &spa
->spa_proc_lock
);
1127 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1128 ASSERT(spa
->spa_proc
!= &p0
);
1129 ASSERT(spa
->spa_did
!= 0);
1133 "Couldn't create process for zfs pool \"%s\"\n",
1138 mutex_exit(&spa
->spa_proc_lock
);
1140 /* If we didn't create a process, we need to create our taskqs. */
1141 if (spa
->spa_proc
== &p0
) {
1142 spa_create_zio_taskqs(spa
);
1145 for (size_t i
= 0; i
< TXG_SIZE
; i
++)
1146 spa
->spa_txg_zio
[i
] = zio_root(spa
, NULL
, NULL
, 0);
1148 list_create(&spa
->spa_config_dirty_list
, sizeof (vdev_t
),
1149 offsetof(vdev_t
, vdev_config_dirty_node
));
1150 list_create(&spa
->spa_evicting_os_list
, sizeof (objset_t
),
1151 offsetof(objset_t
, os_evicting_node
));
1152 list_create(&spa
->spa_state_dirty_list
, sizeof (vdev_t
),
1153 offsetof(vdev_t
, vdev_state_dirty_node
));
1155 txg_list_create(&spa
->spa_vdev_txg_list
, spa
,
1156 offsetof(struct vdev
, vdev_txg_node
));
1158 avl_create(&spa
->spa_errlist_scrub
,
1159 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1160 offsetof(spa_error_entry_t
, se_avl
));
1161 avl_create(&spa
->spa_errlist_last
,
1162 spa_error_entry_compare
, sizeof (spa_error_entry_t
),
1163 offsetof(spa_error_entry_t
, se_avl
));
1167 * Opposite of spa_activate().
1170 spa_deactivate(spa_t
*spa
)
1172 ASSERT(spa
->spa_sync_on
== B_FALSE
);
1173 ASSERT(spa
->spa_dsl_pool
== NULL
);
1174 ASSERT(spa
->spa_root_vdev
== NULL
);
1175 ASSERT(spa
->spa_async_zio_root
== NULL
);
1176 ASSERT(spa
->spa_state
!= POOL_STATE_UNINITIALIZED
);
1178 spa_evicting_os_wait(spa
);
1180 txg_list_destroy(&spa
->spa_vdev_txg_list
);
1182 list_destroy(&spa
->spa_config_dirty_list
);
1183 list_destroy(&spa
->spa_evicting_os_list
);
1184 list_destroy(&spa
->spa_state_dirty_list
);
1186 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
1187 for (int q
= 0; q
< ZIO_TASKQ_TYPES
; q
++) {
1188 spa_taskqs_fini(spa
, t
, q
);
1192 for (size_t i
= 0; i
< TXG_SIZE
; i
++) {
1193 ASSERT3P(spa
->spa_txg_zio
[i
], !=, NULL
);
1194 VERIFY0(zio_wait(spa
->spa_txg_zio
[i
]));
1195 spa
->spa_txg_zio
[i
] = NULL
;
1198 metaslab_class_destroy(spa
->spa_normal_class
);
1199 spa
->spa_normal_class
= NULL
;
1201 metaslab_class_destroy(spa
->spa_log_class
);
1202 spa
->spa_log_class
= NULL
;
1205 * If this was part of an import or the open otherwise failed, we may
1206 * still have errors left in the queues. Empty them just in case.
1208 spa_errlog_drain(spa
);
1210 avl_destroy(&spa
->spa_errlist_scrub
);
1211 avl_destroy(&spa
->spa_errlist_last
);
1213 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
1215 mutex_enter(&spa
->spa_proc_lock
);
1216 if (spa
->spa_proc_state
!= SPA_PROC_NONE
) {
1217 ASSERT(spa
->spa_proc_state
== SPA_PROC_ACTIVE
);
1218 spa
->spa_proc_state
= SPA_PROC_DEACTIVATE
;
1219 cv_broadcast(&spa
->spa_proc_cv
);
1220 while (spa
->spa_proc_state
== SPA_PROC_DEACTIVATE
) {
1221 ASSERT(spa
->spa_proc
!= &p0
);
1222 cv_wait(&spa
->spa_proc_cv
, &spa
->spa_proc_lock
);
1224 ASSERT(spa
->spa_proc_state
== SPA_PROC_GONE
);
1225 spa
->spa_proc_state
= SPA_PROC_NONE
;
1227 ASSERT(spa
->spa_proc
== &p0
);
1228 mutex_exit(&spa
->spa_proc_lock
);
1231 * We want to make sure spa_thread() has actually exited the ZFS
1232 * module, so that the module can't be unloaded out from underneath
1235 if (spa
->spa_did
!= 0) {
1236 thread_join(spa
->spa_did
);
1242 * Verify a pool configuration, and construct the vdev tree appropriately. This
1243 * will create all the necessary vdevs in the appropriate layout, with each vdev
1244 * in the CLOSED state. This will prep the pool before open/creation/import.
1245 * All vdev validation is done by the vdev_alloc() routine.
1248 spa_config_parse(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
,
1249 uint_t id
, int atype
)
1255 if ((error
= vdev_alloc(spa
, vdp
, nv
, parent
, id
, atype
)) != 0)
1258 if ((*vdp
)->vdev_ops
->vdev_op_leaf
)
1261 error
= nvlist_lookup_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
1264 if (error
== ENOENT
)
1270 return (SET_ERROR(EINVAL
));
1273 for (int c
= 0; c
< children
; c
++) {
1275 if ((error
= spa_config_parse(spa
, &vd
, child
[c
], *vdp
, c
,
1283 ASSERT(*vdp
!= NULL
);
1289 * Opposite of spa_load().
1292 spa_unload(spa_t
*spa
)
1296 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1298 spa_load_note(spa
, "UNLOADING");
1303 spa_async_suspend(spa
);
1308 if (spa
->spa_sync_on
) {
1309 txg_sync_stop(spa
->spa_dsl_pool
);
1310 spa
->spa_sync_on
= B_FALSE
;
1314 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1315 * to call it earlier, before we wait for async i/o to complete.
1316 * This ensures that there is no async metaslab prefetching, by
1317 * calling taskq_wait(mg_taskq).
1319 if (spa
->spa_root_vdev
!= NULL
) {
1320 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1321 for (int c
= 0; c
< spa
->spa_root_vdev
->vdev_children
; c
++)
1322 vdev_metaslab_fini(spa
->spa_root_vdev
->vdev_child
[c
]);
1323 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1327 * Wait for any outstanding async I/O to complete.
1329 if (spa
->spa_async_zio_root
!= NULL
) {
1330 for (int i
= 0; i
< max_ncpus
; i
++)
1331 (void) zio_wait(spa
->spa_async_zio_root
[i
]);
1332 kmem_free(spa
->spa_async_zio_root
, max_ncpus
* sizeof (void *));
1333 spa
->spa_async_zio_root
= NULL
;
1336 if (spa
->spa_vdev_removal
!= NULL
) {
1337 spa_vdev_removal_destroy(spa
->spa_vdev_removal
);
1338 spa
->spa_vdev_removal
= NULL
;
1341 if (spa
->spa_condense_zthr
!= NULL
) {
1342 ASSERT(!zthr_isrunning(spa
->spa_condense_zthr
));
1343 zthr_destroy(spa
->spa_condense_zthr
);
1344 spa
->spa_condense_zthr
= NULL
;
1347 spa_condense_fini(spa
);
1349 bpobj_close(&spa
->spa_deferred_bpobj
);
1351 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1356 if (spa
->spa_root_vdev
)
1357 vdev_free(spa
->spa_root_vdev
);
1358 ASSERT(spa
->spa_root_vdev
== NULL
);
1361 * Close the dsl pool.
1363 if (spa
->spa_dsl_pool
) {
1364 dsl_pool_close(spa
->spa_dsl_pool
);
1365 spa
->spa_dsl_pool
= NULL
;
1366 spa
->spa_meta_objset
= NULL
;
1372 * Drop and purge level 2 cache
1374 spa_l2cache_drop(spa
);
1376 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1377 vdev_free(spa
->spa_spares
.sav_vdevs
[i
]);
1378 if (spa
->spa_spares
.sav_vdevs
) {
1379 kmem_free(spa
->spa_spares
.sav_vdevs
,
1380 spa
->spa_spares
.sav_count
* sizeof (void *));
1381 spa
->spa_spares
.sav_vdevs
= NULL
;
1383 if (spa
->spa_spares
.sav_config
) {
1384 nvlist_free(spa
->spa_spares
.sav_config
);
1385 spa
->spa_spares
.sav_config
= NULL
;
1387 spa
->spa_spares
.sav_count
= 0;
1389 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
1390 vdev_clear_stats(spa
->spa_l2cache
.sav_vdevs
[i
]);
1391 vdev_free(spa
->spa_l2cache
.sav_vdevs
[i
]);
1393 if (spa
->spa_l2cache
.sav_vdevs
) {
1394 kmem_free(spa
->spa_l2cache
.sav_vdevs
,
1395 spa
->spa_l2cache
.sav_count
* sizeof (void *));
1396 spa
->spa_l2cache
.sav_vdevs
= NULL
;
1398 if (spa
->spa_l2cache
.sav_config
) {
1399 nvlist_free(spa
->spa_l2cache
.sav_config
);
1400 spa
->spa_l2cache
.sav_config
= NULL
;
1402 spa
->spa_l2cache
.sav_count
= 0;
1404 spa
->spa_async_suspended
= 0;
1406 spa
->spa_indirect_vdevs_loaded
= B_FALSE
;
1408 if (spa
->spa_comment
!= NULL
) {
1409 spa_strfree(spa
->spa_comment
);
1410 spa
->spa_comment
= NULL
;
1413 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1417 * Load (or re-load) the current list of vdevs describing the active spares for
1418 * this pool. When this is called, we have some form of basic information in
1419 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1420 * then re-generate a more complete list including status information.
1423 spa_load_spares(spa_t
*spa
)
1430 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1433 * First, close and free any existing spare vdevs.
1435 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1436 vd
= spa
->spa_spares
.sav_vdevs
[i
];
1438 /* Undo the call to spa_activate() below */
1439 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1440 B_FALSE
)) != NULL
&& tvd
->vdev_isspare
)
1441 spa_spare_remove(tvd
);
1446 if (spa
->spa_spares
.sav_vdevs
)
1447 kmem_free(spa
->spa_spares
.sav_vdevs
,
1448 spa
->spa_spares
.sav_count
* sizeof (void *));
1450 if (spa
->spa_spares
.sav_config
== NULL
)
1453 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
1454 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
1456 spa
->spa_spares
.sav_count
= (int)nspares
;
1457 spa
->spa_spares
.sav_vdevs
= NULL
;
1463 * Construct the array of vdevs, opening them to get status in the
1464 * process. For each spare, there is potentially two different vdev_t
1465 * structures associated with it: one in the list of spares (used only
1466 * for basic validation purposes) and one in the active vdev
1467 * configuration (if it's spared in). During this phase we open and
1468 * validate each vdev on the spare list. If the vdev also exists in the
1469 * active configuration, then we also mark this vdev as an active spare.
1471 spa
->spa_spares
.sav_vdevs
= kmem_alloc(nspares
* sizeof (void *),
1473 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
1474 VERIFY(spa_config_parse(spa
, &vd
, spares
[i
], NULL
, 0,
1475 VDEV_ALLOC_SPARE
) == 0);
1478 spa
->spa_spares
.sav_vdevs
[i
] = vd
;
1480 if ((tvd
= spa_lookup_by_guid(spa
, vd
->vdev_guid
,
1481 B_FALSE
)) != NULL
) {
1482 if (!tvd
->vdev_isspare
)
1486 * We only mark the spare active if we were successfully
1487 * able to load the vdev. Otherwise, importing a pool
1488 * with a bad active spare would result in strange
1489 * behavior, because multiple pool would think the spare
1490 * is actively in use.
1492 * There is a vulnerability here to an equally bizarre
1493 * circumstance, where a dead active spare is later
1494 * brought back to life (onlined or otherwise). Given
1495 * the rarity of this scenario, and the extra complexity
1496 * it adds, we ignore the possibility.
1498 if (!vdev_is_dead(tvd
))
1499 spa_spare_activate(tvd
);
1503 vd
->vdev_aux
= &spa
->spa_spares
;
1505 if (vdev_open(vd
) != 0)
1508 if (vdev_validate_aux(vd
) == 0)
1513 * Recompute the stashed list of spares, with status information
1516 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
, ZPOOL_CONFIG_SPARES
,
1517 DATA_TYPE_NVLIST_ARRAY
) == 0);
1519 spares
= kmem_alloc(spa
->spa_spares
.sav_count
* sizeof (void *),
1521 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1522 spares
[i
] = vdev_config_generate(spa
,
1523 spa
->spa_spares
.sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_SPARE
);
1524 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
1525 ZPOOL_CONFIG_SPARES
, spares
, spa
->spa_spares
.sav_count
) == 0);
1526 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
1527 nvlist_free(spares
[i
]);
1528 kmem_free(spares
, spa
->spa_spares
.sav_count
* sizeof (void *));
1532 * Load (or re-load) the current list of vdevs describing the active l2cache for
1533 * this pool. When this is called, we have some form of basic information in
1534 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1535 * then re-generate a more complete list including status information.
1536 * Devices which are already active have their details maintained, and are
1540 spa_load_l2cache(spa_t
*spa
)
1544 int i
, j
, oldnvdevs
;
1546 vdev_t
*vd
, **oldvdevs
, **newvdevs
;
1547 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
1549 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1551 if (sav
->sav_config
!= NULL
) {
1552 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
1553 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
1554 newvdevs
= kmem_alloc(nl2cache
* sizeof (void *), KM_SLEEP
);
1560 oldvdevs
= sav
->sav_vdevs
;
1561 oldnvdevs
= sav
->sav_count
;
1562 sav
->sav_vdevs
= NULL
;
1566 * Process new nvlist of vdevs.
1568 for (i
= 0; i
< nl2cache
; i
++) {
1569 VERIFY(nvlist_lookup_uint64(l2cache
[i
], ZPOOL_CONFIG_GUID
,
1573 for (j
= 0; j
< oldnvdevs
; j
++) {
1575 if (vd
!= NULL
&& guid
== vd
->vdev_guid
) {
1577 * Retain previous vdev for add/remove ops.
1585 if (newvdevs
[i
] == NULL
) {
1589 VERIFY(spa_config_parse(spa
, &vd
, l2cache
[i
], NULL
, 0,
1590 VDEV_ALLOC_L2CACHE
) == 0);
1595 * Commit this vdev as an l2cache device,
1596 * even if it fails to open.
1598 spa_l2cache_add(vd
);
1603 spa_l2cache_activate(vd
);
1605 if (vdev_open(vd
) != 0)
1608 (void) vdev_validate_aux(vd
);
1610 if (!vdev_is_dead(vd
))
1611 l2arc_add_vdev(spa
, vd
);
1616 * Purge vdevs that were dropped
1618 for (i
= 0; i
< oldnvdevs
; i
++) {
1623 ASSERT(vd
->vdev_isl2cache
);
1625 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
1626 pool
!= 0ULL && l2arc_vdev_present(vd
))
1627 l2arc_remove_vdev(vd
);
1628 vdev_clear_stats(vd
);
1634 kmem_free(oldvdevs
, oldnvdevs
* sizeof (void *));
1636 if (sav
->sav_config
== NULL
)
1639 sav
->sav_vdevs
= newvdevs
;
1640 sav
->sav_count
= (int)nl2cache
;
1643 * Recompute the stashed list of l2cache devices, with status
1644 * information this time.
1646 VERIFY(nvlist_remove(sav
->sav_config
, ZPOOL_CONFIG_L2CACHE
,
1647 DATA_TYPE_NVLIST_ARRAY
) == 0);
1649 l2cache
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
1650 for (i
= 0; i
< sav
->sav_count
; i
++)
1651 l2cache
[i
] = vdev_config_generate(spa
,
1652 sav
->sav_vdevs
[i
], B_TRUE
, VDEV_CONFIG_L2CACHE
);
1653 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
1654 ZPOOL_CONFIG_L2CACHE
, l2cache
, sav
->sav_count
) == 0);
1656 for (i
= 0; i
< sav
->sav_count
; i
++)
1657 nvlist_free(l2cache
[i
]);
1659 kmem_free(l2cache
, sav
->sav_count
* sizeof (void *));
1663 load_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
**value
)
1666 char *packed
= NULL
;
1671 error
= dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
);
1675 nvsize
= *(uint64_t *)db
->db_data
;
1676 dmu_buf_rele(db
, FTAG
);
1678 packed
= kmem_alloc(nvsize
, KM_SLEEP
);
1679 error
= dmu_read(spa
->spa_meta_objset
, obj
, 0, nvsize
, packed
,
1682 error
= nvlist_unpack(packed
, nvsize
, value
, 0);
1683 kmem_free(packed
, nvsize
);
1689 * Concrete top-level vdevs that are not missing and are not logs. At every
1690 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1693 spa_healthy_core_tvds(spa_t
*spa
)
1695 vdev_t
*rvd
= spa
->spa_root_vdev
;
1698 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
1699 vdev_t
*vd
= rvd
->vdev_child
[i
];
1702 if (vdev_is_concrete(vd
) && !vdev_is_dead(vd
))
1710 * Checks to see if the given vdev could not be opened, in which case we post a
1711 * sysevent to notify the autoreplace code that the device has been removed.
1714 spa_check_removed(vdev_t
*vd
)
1716 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1717 spa_check_removed(vd
->vdev_child
[c
]);
1719 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_is_dead(vd
) &&
1720 vdev_is_concrete(vd
)) {
1721 zfs_post_autoreplace(vd
->vdev_spa
, vd
);
1722 spa_event_notify(vd
->vdev_spa
, vd
, NULL
, ESC_ZFS_VDEV_CHECK
);
1727 spa_check_for_missing_logs(spa_t
*spa
)
1729 vdev_t
*rvd
= spa
->spa_root_vdev
;
1732 * If we're doing a normal import, then build up any additional
1733 * diagnostic information about missing log devices.
1734 * We'll pass this up to the user for further processing.
1736 if (!(spa
->spa_import_flags
& ZFS_IMPORT_MISSING_LOG
)) {
1737 nvlist_t
**child
, *nv
;
1740 child
= kmem_alloc(rvd
->vdev_children
* sizeof (nvlist_t
**),
1742 VERIFY(nvlist_alloc(&nv
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1744 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
1745 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1748 * We consider a device as missing only if it failed
1749 * to open (i.e. offline or faulted is not considered
1752 if (tvd
->vdev_islog
&&
1753 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
1754 child
[idx
++] = vdev_config_generate(spa
, tvd
,
1755 B_FALSE
, VDEV_CONFIG_MISSING
);
1760 fnvlist_add_nvlist_array(nv
,
1761 ZPOOL_CONFIG_CHILDREN
, child
, idx
);
1762 fnvlist_add_nvlist(spa
->spa_load_info
,
1763 ZPOOL_CONFIG_MISSING_DEVICES
, nv
);
1765 for (uint64_t i
= 0; i
< idx
; i
++)
1766 nvlist_free(child
[i
]);
1769 kmem_free(child
, rvd
->vdev_children
* sizeof (char **));
1772 spa_load_failed(spa
, "some log devices are missing");
1773 return (SET_ERROR(ENXIO
));
1776 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
1777 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1779 if (tvd
->vdev_islog
&&
1780 tvd
->vdev_state
== VDEV_STATE_CANT_OPEN
) {
1781 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
1782 spa_load_note(spa
, "some log devices are "
1783 "missing, ZIL is dropped.");
1793 * Check for missing log devices
1796 spa_check_logs(spa_t
*spa
)
1798 boolean_t rv
= B_FALSE
;
1799 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1801 switch (spa
->spa_log_state
) {
1802 case SPA_LOG_MISSING
:
1803 /* need to recheck in case slog has been restored */
1804 case SPA_LOG_UNKNOWN
:
1805 rv
= (dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
1806 zil_check_log_chain
, NULL
, DS_FIND_CHILDREN
) != 0);
1808 spa_set_log_state(spa
, SPA_LOG_MISSING
);
1815 spa_passivate_log(spa_t
*spa
)
1817 vdev_t
*rvd
= spa
->spa_root_vdev
;
1818 boolean_t slog_found
= B_FALSE
;
1820 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1822 if (!spa_has_slogs(spa
))
1825 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1826 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1827 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1829 if (tvd
->vdev_islog
) {
1830 metaslab_group_passivate(mg
);
1831 slog_found
= B_TRUE
;
1835 return (slog_found
);
1839 spa_activate_log(spa_t
*spa
)
1841 vdev_t
*rvd
= spa
->spa_root_vdev
;
1843 ASSERT(spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1845 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
1846 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1847 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1849 if (tvd
->vdev_islog
)
1850 metaslab_group_activate(mg
);
1855 spa_reset_logs(spa_t
*spa
)
1859 error
= dmu_objset_find(spa_name(spa
), zil_reset
,
1860 NULL
, DS_FIND_CHILDREN
);
1863 * We successfully offlined the log device, sync out the
1864 * current txg so that the "stubby" block can be removed
1867 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1873 spa_aux_check_removed(spa_aux_vdev_t
*sav
)
1875 for (int i
= 0; i
< sav
->sav_count
; i
++)
1876 spa_check_removed(sav
->sav_vdevs
[i
]);
1880 spa_claim_notify(zio_t
*zio
)
1882 spa_t
*spa
= zio
->io_spa
;
1887 mutex_enter(&spa
->spa_props_lock
); /* any mutex will do */
1888 if (spa
->spa_claim_max_txg
< zio
->io_bp
->blk_birth
)
1889 spa
->spa_claim_max_txg
= zio
->io_bp
->blk_birth
;
1890 mutex_exit(&spa
->spa_props_lock
);
1893 typedef struct spa_load_error
{
1894 uint64_t sle_meta_count
;
1895 uint64_t sle_data_count
;
1899 spa_load_verify_done(zio_t
*zio
)
1901 blkptr_t
*bp
= zio
->io_bp
;
1902 spa_load_error_t
*sle
= zio
->io_private
;
1903 dmu_object_type_t type
= BP_GET_TYPE(bp
);
1904 int error
= zio
->io_error
;
1905 spa_t
*spa
= zio
->io_spa
;
1907 abd_free(zio
->io_abd
);
1909 if ((BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
)) &&
1910 type
!= DMU_OT_INTENT_LOG
)
1911 atomic_inc_64(&sle
->sle_meta_count
);
1913 atomic_inc_64(&sle
->sle_data_count
);
1916 mutex_enter(&spa
->spa_scrub_lock
);
1917 spa
->spa_scrub_inflight
--;
1918 cv_broadcast(&spa
->spa_scrub_io_cv
);
1919 mutex_exit(&spa
->spa_scrub_lock
);
1923 * Maximum number of concurrent scrub i/os to create while verifying
1924 * a pool while importing it.
1926 int spa_load_verify_maxinflight
= 10000;
1927 boolean_t spa_load_verify_metadata
= B_TRUE
;
1928 boolean_t spa_load_verify_data
= B_TRUE
;
1932 spa_load_verify_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
1933 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
1935 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
1938 * Note: normally this routine will not be called if
1939 * spa_load_verify_metadata is not set. However, it may be useful
1940 * to manually set the flag after the traversal has begun.
1942 if (!spa_load_verify_metadata
)
1944 if (!BP_IS_METADATA(bp
) && !spa_load_verify_data
)
1948 size_t size
= BP_GET_PSIZE(bp
);
1950 mutex_enter(&spa
->spa_scrub_lock
);
1951 while (spa
->spa_scrub_inflight
>= spa_load_verify_maxinflight
)
1952 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
1953 spa
->spa_scrub_inflight
++;
1954 mutex_exit(&spa
->spa_scrub_lock
);
1956 zio_nowait(zio_read(rio
, spa
, bp
, abd_alloc_for_io(size
, B_FALSE
), size
,
1957 spa_load_verify_done
, rio
->io_private
, ZIO_PRIORITY_SCRUB
,
1958 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_CANFAIL
|
1959 ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
, zb
));
1965 verify_dataset_name_len(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
1967 if (dsl_dataset_namelen(ds
) >= ZFS_MAX_DATASET_NAME_LEN
)
1968 return (SET_ERROR(ENAMETOOLONG
));
1974 spa_load_verify(spa_t
*spa
)
1977 spa_load_error_t sle
= { 0 };
1978 zpool_rewind_policy_t policy
;
1979 boolean_t verify_ok
= B_FALSE
;
1982 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
1984 if (policy
.zrp_request
& ZPOOL_NEVER_REWIND
)
1987 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
1988 error
= dmu_objset_find_dp(spa
->spa_dsl_pool
,
1989 spa
->spa_dsl_pool
->dp_root_dir_obj
, verify_dataset_name_len
, NULL
,
1991 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
1995 rio
= zio_root(spa
, NULL
, &sle
,
1996 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
);
1998 if (spa_load_verify_metadata
) {
1999 if (spa
->spa_extreme_rewind
) {
2000 spa_load_note(spa
, "performing a complete scan of the "
2001 "pool since extreme rewind is on. This may take "
2002 "a very long time.\n (spa_load_verify_data=%u, "
2003 "spa_load_verify_metadata=%u)",
2004 spa_load_verify_data
, spa_load_verify_metadata
);
2006 error
= traverse_pool(spa
, spa
->spa_verify_min_txg
,
2007 TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
,
2008 spa_load_verify_cb
, rio
);
2011 (void) zio_wait(rio
);
2013 spa
->spa_load_meta_errors
= sle
.sle_meta_count
;
2014 spa
->spa_load_data_errors
= sle
.sle_data_count
;
2016 if (sle
.sle_meta_count
!= 0 || sle
.sle_data_count
!= 0) {
2017 spa_load_note(spa
, "spa_load_verify found %llu metadata errors "
2018 "and %llu data errors", (u_longlong_t
)sle
.sle_meta_count
,
2019 (u_longlong_t
)sle
.sle_data_count
);
2022 if (spa_load_verify_dryrun
||
2023 (!error
&& sle
.sle_meta_count
<= policy
.zrp_maxmeta
&&
2024 sle
.sle_data_count
<= policy
.zrp_maxdata
)) {
2028 spa
->spa_load_txg
= spa
->spa_uberblock
.ub_txg
;
2029 spa
->spa_load_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
2031 loss
= spa
->spa_last_ubsync_txg_ts
- spa
->spa_load_txg_ts
;
2032 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2033 ZPOOL_CONFIG_LOAD_TIME
, spa
->spa_load_txg_ts
) == 0);
2034 VERIFY(nvlist_add_int64(spa
->spa_load_info
,
2035 ZPOOL_CONFIG_REWIND_TIME
, loss
) == 0);
2036 VERIFY(nvlist_add_uint64(spa
->spa_load_info
,
2037 ZPOOL_CONFIG_LOAD_DATA_ERRORS
, sle
.sle_data_count
) == 0);
2039 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
;
2042 if (spa_load_verify_dryrun
)
2046 if (error
!= ENXIO
&& error
!= EIO
)
2047 error
= SET_ERROR(EIO
);
2051 return (verify_ok
? 0 : EIO
);
2055 * Find a value in the pool props object.
2058 spa_prop_find(spa_t
*spa
, zpool_prop_t prop
, uint64_t *val
)
2060 (void) zap_lookup(spa
->spa_meta_objset
, spa
->spa_pool_props_object
,
2061 zpool_prop_to_name(prop
), sizeof (uint64_t), 1, val
);
2065 * Find a value in the pool directory object.
2068 spa_dir_prop(spa_t
*spa
, const char *name
, uint64_t *val
, boolean_t log_enoent
)
2070 int error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2071 name
, sizeof (uint64_t), 1, val
);
2073 if (error
!= 0 && (error
!= ENOENT
|| log_enoent
)) {
2074 spa_load_failed(spa
, "couldn't get '%s' value in MOS directory "
2075 "[error=%d]", name
, error
);
2082 spa_vdev_err(vdev_t
*vdev
, vdev_aux_t aux
, int err
)
2084 vdev_set_state(vdev
, B_TRUE
, VDEV_STATE_CANT_OPEN
, aux
);
2085 return (SET_ERROR(err
));
2089 spa_spawn_aux_threads(spa_t
*spa
)
2091 ASSERT(spa_writeable(spa
));
2093 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
2095 spa_start_indirect_condensing_thread(spa
);
2099 * Fix up config after a partly-completed split. This is done with the
2100 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2101 * pool have that entry in their config, but only the splitting one contains
2102 * a list of all the guids of the vdevs that are being split off.
2104 * This function determines what to do with that list: either rejoin
2105 * all the disks to the pool, or complete the splitting process. To attempt
2106 * the rejoin, each disk that is offlined is marked online again, and
2107 * we do a reopen() call. If the vdev label for every disk that was
2108 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2109 * then we call vdev_split() on each disk, and complete the split.
2111 * Otherwise we leave the config alone, with all the vdevs in place in
2112 * the original pool.
2115 spa_try_repair(spa_t
*spa
, nvlist_t
*config
)
2122 boolean_t attempt_reopen
;
2124 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) != 0)
2127 /* check that the config is complete */
2128 if (nvlist_lookup_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
2129 &glist
, &gcount
) != 0)
2132 vd
= kmem_zalloc(gcount
* sizeof (vdev_t
*), KM_SLEEP
);
2134 /* attempt to online all the vdevs & validate */
2135 attempt_reopen
= B_TRUE
;
2136 for (i
= 0; i
< gcount
; i
++) {
2137 if (glist
[i
] == 0) /* vdev is hole */
2140 vd
[i
] = spa_lookup_by_guid(spa
, glist
[i
], B_FALSE
);
2141 if (vd
[i
] == NULL
) {
2143 * Don't bother attempting to reopen the disks;
2144 * just do the split.
2146 attempt_reopen
= B_FALSE
;
2148 /* attempt to re-online it */
2149 vd
[i
]->vdev_offline
= B_FALSE
;
2153 if (attempt_reopen
) {
2154 vdev_reopen(spa
->spa_root_vdev
);
2156 /* check each device to see what state it's in */
2157 for (extracted
= 0, i
= 0; i
< gcount
; i
++) {
2158 if (vd
[i
] != NULL
&&
2159 vd
[i
]->vdev_stat
.vs_aux
!= VDEV_AUX_SPLIT_POOL
)
2166 * If every disk has been moved to the new pool, or if we never
2167 * even attempted to look at them, then we split them off for
2170 if (!attempt_reopen
|| gcount
== extracted
) {
2171 for (i
= 0; i
< gcount
; i
++)
2174 vdev_reopen(spa
->spa_root_vdev
);
2177 kmem_free(vd
, gcount
* sizeof (vdev_t
*));
2181 spa_load(spa_t
*spa
, spa_load_state_t state
, spa_import_type_t type
)
2183 char *ereport
= FM_EREPORT_ZFS_POOL
;
2186 spa
->spa_load_state
= state
;
2188 gethrestime(&spa
->spa_loaded_ts
);
2189 error
= spa_load_impl(spa
, type
, &ereport
, B_FALSE
);
2192 * Don't count references from objsets that are already closed
2193 * and are making their way through the eviction process.
2195 spa_evicting_os_wait(spa
);
2196 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
2198 if (error
!= EEXIST
) {
2199 spa
->spa_loaded_ts
.tv_sec
= 0;
2200 spa
->spa_loaded_ts
.tv_nsec
= 0;
2202 if (error
!= EBADF
) {
2203 zfs_ereport_post(ereport
, spa
, NULL
, NULL
, 0, 0);
2206 spa
->spa_load_state
= error
? SPA_LOAD_ERROR
: SPA_LOAD_NONE
;
2213 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2214 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2215 * spa's per-vdev ZAP list.
2218 vdev_count_verify_zaps(vdev_t
*vd
)
2220 spa_t
*spa
= vd
->vdev_spa
;
2222 if (vd
->vdev_top_zap
!= 0) {
2224 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2225 spa
->spa_all_vdev_zaps
, vd
->vdev_top_zap
));
2227 if (vd
->vdev_leaf_zap
!= 0) {
2229 ASSERT0(zap_lookup_int(spa
->spa_meta_objset
,
2230 spa
->spa_all_vdev_zaps
, vd
->vdev_leaf_zap
));
2233 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2234 total
+= vdev_count_verify_zaps(vd
->vdev_child
[i
]);
2241 spa_verify_host(spa_t
*spa
, nvlist_t
*mos_config
)
2245 uint64_t myhostid
= 0;
2247 if (!spa_is_root(spa
) && nvlist_lookup_uint64(mos_config
,
2248 ZPOOL_CONFIG_HOSTID
, &hostid
) == 0) {
2249 hostname
= fnvlist_lookup_string(mos_config
,
2250 ZPOOL_CONFIG_HOSTNAME
);
2252 myhostid
= zone_get_hostid(NULL
);
2254 if (hostid
!= 0 && myhostid
!= 0 && hostid
!= myhostid
) {
2255 cmn_err(CE_WARN
, "pool '%s' could not be "
2256 "loaded as it was last accessed by "
2257 "another system (host: %s hostid: 0x%llx). "
2258 "See: http://illumos.org/msg/ZFS-8000-EY",
2259 spa_name(spa
), hostname
, (u_longlong_t
)hostid
);
2260 spa_load_failed(spa
, "hostid verification failed: pool "
2261 "last accessed by host: %s (hostid: 0x%llx)",
2262 hostname
, (u_longlong_t
)hostid
);
2263 return (SET_ERROR(EBADF
));
2271 spa_ld_parse_config(spa_t
*spa
, spa_import_type_t type
)
2274 nvlist_t
*nvtree
, *nvl
, *config
= spa
->spa_config
;
2281 * Versioning wasn't explicitly added to the label until later, so if
2282 * it's not present treat it as the initial version.
2284 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
2285 &spa
->spa_ubsync
.ub_version
) != 0)
2286 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
2288 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
, &pool_guid
)) {
2289 spa_load_failed(spa
, "invalid config provided: '%s' missing",
2290 ZPOOL_CONFIG_POOL_GUID
);
2291 return (SET_ERROR(EINVAL
));
2294 if ((spa
->spa_load_state
== SPA_LOAD_IMPORT
|| spa
->spa_load_state
==
2295 SPA_LOAD_TRYIMPORT
) && spa_guid_exists(pool_guid
, 0)) {
2296 spa_load_failed(spa
, "a pool with guid %llu is already open",
2297 (u_longlong_t
)pool_guid
);
2298 return (SET_ERROR(EEXIST
));
2301 spa
->spa_config_guid
= pool_guid
;
2303 nvlist_free(spa
->spa_load_info
);
2304 spa
->spa_load_info
= fnvlist_alloc();
2306 ASSERT(spa
->spa_comment
== NULL
);
2307 if (nvlist_lookup_string(config
, ZPOOL_CONFIG_COMMENT
, &comment
) == 0)
2308 spa
->spa_comment
= spa_strdup(comment
);
2310 (void) nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
2311 &spa
->spa_config_txg
);
2313 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_SPLIT
, &nvl
) == 0)
2314 spa
->spa_config_splitting
= fnvlist_dup(nvl
);
2316 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvtree
)) {
2317 spa_load_failed(spa
, "invalid config provided: '%s' missing",
2318 ZPOOL_CONFIG_VDEV_TREE
);
2319 return (SET_ERROR(EINVAL
));
2323 * Create "The Godfather" zio to hold all async IOs
2325 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
2327 for (int i
= 0; i
< max_ncpus
; i
++) {
2328 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
2329 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
2330 ZIO_FLAG_GODFATHER
);
2334 * Parse the configuration into a vdev tree. We explicitly set the
2335 * value that will be returned by spa_version() since parsing the
2336 * configuration requires knowing the version number.
2338 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2339 parse
= (type
== SPA_IMPORT_EXISTING
?
2340 VDEV_ALLOC_LOAD
: VDEV_ALLOC_SPLIT
);
2341 error
= spa_config_parse(spa
, &rvd
, nvtree
, NULL
, 0, parse
);
2342 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2345 spa_load_failed(spa
, "unable to parse config [error=%d]",
2350 ASSERT(spa
->spa_root_vdev
== rvd
);
2351 ASSERT3U(spa
->spa_min_ashift
, >=, SPA_MINBLOCKSHIFT
);
2352 ASSERT3U(spa
->spa_max_ashift
, <=, SPA_MAXBLOCKSHIFT
);
2354 if (type
!= SPA_IMPORT_ASSEMBLE
) {
2355 ASSERT(spa_guid(spa
) == pool_guid
);
2362 * Recursively open all vdevs in the vdev tree. This function is called twice:
2363 * first with the untrusted config, then with the trusted config.
2366 spa_ld_open_vdevs(spa_t
*spa
)
2371 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2372 * missing/unopenable for the root vdev to be still considered openable.
2374 if (spa
->spa_trust_config
) {
2375 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds
;
2376 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_CACHEFILE
) {
2377 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_cachefile
;
2378 } else if (spa
->spa_config_source
== SPA_CONFIG_SRC_SCAN
) {
2379 spa
->spa_missing_tvds_allowed
= zfs_max_missing_tvds_scan
;
2381 spa
->spa_missing_tvds_allowed
= 0;
2384 spa
->spa_missing_tvds_allowed
=
2385 MAX(zfs_max_missing_tvds
, spa
->spa_missing_tvds_allowed
);
2387 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2388 error
= vdev_open(spa
->spa_root_vdev
);
2389 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2391 if (spa
->spa_missing_tvds
!= 0) {
2392 spa_load_note(spa
, "vdev tree has %lld missing top-level "
2393 "vdevs.", (u_longlong_t
)spa
->spa_missing_tvds
);
2394 if (spa
->spa_trust_config
&& (spa
->spa_mode
& FWRITE
)) {
2396 * Although theoretically we could allow users to open
2397 * incomplete pools in RW mode, we'd need to add a lot
2398 * of extra logic (e.g. adjust pool space to account
2399 * for missing vdevs).
2400 * This limitation also prevents users from accidentally
2401 * opening the pool in RW mode during data recovery and
2402 * damaging it further.
2404 spa_load_note(spa
, "pools with missing top-level "
2405 "vdevs can only be opened in read-only mode.");
2406 error
= SET_ERROR(ENXIO
);
2408 spa_load_note(spa
, "current settings allow for maximum "
2409 "%lld missing top-level vdevs at this stage.",
2410 (u_longlong_t
)spa
->spa_missing_tvds_allowed
);
2414 spa_load_failed(spa
, "unable to open vdev tree [error=%d]",
2417 if (spa
->spa_missing_tvds
!= 0 || error
!= 0)
2418 vdev_dbgmsg_print_tree(spa
->spa_root_vdev
, 2);
2424 * We need to validate the vdev labels against the configuration that
2425 * we have in hand. This function is called twice: first with an untrusted
2426 * config, then with a trusted config. The validation is more strict when the
2427 * config is trusted.
2430 spa_ld_validate_vdevs(spa_t
*spa
)
2433 vdev_t
*rvd
= spa
->spa_root_vdev
;
2435 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2436 error
= vdev_validate(rvd
);
2437 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2440 spa_load_failed(spa
, "vdev_validate failed [error=%d]", error
);
2444 if (rvd
->vdev_state
<= VDEV_STATE_CANT_OPEN
) {
2445 spa_load_failed(spa
, "cannot open vdev tree after invalidating "
2447 vdev_dbgmsg_print_tree(rvd
, 2);
2448 return (SET_ERROR(ENXIO
));
2455 spa_ld_select_uberblock(spa_t
*spa
, spa_import_type_t type
)
2457 vdev_t
*rvd
= spa
->spa_root_vdev
;
2459 uberblock_t
*ub
= &spa
->spa_uberblock
;
2462 * Find the best uberblock.
2464 vdev_uberblock_load(rvd
, ub
, &label
);
2467 * If we weren't able to find a single valid uberblock, return failure.
2469 if (ub
->ub_txg
== 0) {
2471 spa_load_failed(spa
, "no valid uberblock found");
2472 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, ENXIO
));
2475 spa_load_note(spa
, "using uberblock with txg=%llu",
2476 (u_longlong_t
)ub
->ub_txg
);
2479 * If the pool has an unsupported version we can't open it.
2481 if (!SPA_VERSION_IS_SUPPORTED(ub
->ub_version
)) {
2483 spa_load_failed(spa
, "version %llu is not supported",
2484 (u_longlong_t
)ub
->ub_version
);
2485 return (spa_vdev_err(rvd
, VDEV_AUX_VERSION_NEWER
, ENOTSUP
));
2488 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2492 * If we weren't able to find what's necessary for reading the
2493 * MOS in the label, return failure.
2495 if (label
== NULL
) {
2496 spa_load_failed(spa
, "label config unavailable");
2497 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2501 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
2504 spa_load_failed(spa
, "invalid label: '%s' missing",
2505 ZPOOL_CONFIG_FEATURES_FOR_READ
);
2506 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
2511 * Update our in-core representation with the definitive values
2514 nvlist_free(spa
->spa_label_features
);
2515 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
, 0) == 0);
2521 * Look through entries in the label nvlist's features_for_read. If
2522 * there is a feature listed there which we don't understand then we
2523 * cannot open a pool.
2525 if (ub
->ub_version
>= SPA_VERSION_FEATURES
) {
2526 nvlist_t
*unsup_feat
;
2528 VERIFY(nvlist_alloc(&unsup_feat
, NV_UNIQUE_NAME
, KM_SLEEP
) ==
2531 for (nvpair_t
*nvp
= nvlist_next_nvpair(spa
->spa_label_features
,
2533 nvp
= nvlist_next_nvpair(spa
->spa_label_features
, nvp
)) {
2534 if (!zfeature_is_supported(nvpair_name(nvp
))) {
2535 VERIFY(nvlist_add_string(unsup_feat
,
2536 nvpair_name(nvp
), "") == 0);
2540 if (!nvlist_empty(unsup_feat
)) {
2541 VERIFY(nvlist_add_nvlist(spa
->spa_load_info
,
2542 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
) == 0);
2543 nvlist_free(unsup_feat
);
2544 spa_load_failed(spa
, "some features are unsupported");
2545 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2549 nvlist_free(unsup_feat
);
2552 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa
->spa_config_splitting
) {
2553 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2554 spa_try_repair(spa
, spa
->spa_config
);
2555 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2556 nvlist_free(spa
->spa_config_splitting
);
2557 spa
->spa_config_splitting
= NULL
;
2561 * Initialize internal SPA structures.
2563 spa
->spa_state
= POOL_STATE_ACTIVE
;
2564 spa
->spa_ubsync
= spa
->spa_uberblock
;
2565 spa
->spa_verify_min_txg
= spa
->spa_extreme_rewind
?
2566 TXG_INITIAL
- 1 : spa_last_synced_txg(spa
) - TXG_DEFER_SIZE
- 1;
2567 spa
->spa_first_txg
= spa
->spa_last_ubsync_txg
?
2568 spa
->spa_last_ubsync_txg
: spa_last_synced_txg(spa
) + 1;
2569 spa
->spa_claim_max_txg
= spa
->spa_first_txg
;
2570 spa
->spa_prev_software_version
= ub
->ub_software_version
;
2576 spa_ld_open_rootbp(spa_t
*spa
)
2579 vdev_t
*rvd
= spa
->spa_root_vdev
;
2581 error
= dsl_pool_init(spa
, spa
->spa_first_txg
, &spa
->spa_dsl_pool
);
2583 spa_load_failed(spa
, "unable to open rootbp in dsl_pool_init "
2584 "[error=%d]", error
);
2585 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2587 spa
->spa_meta_objset
= spa
->spa_dsl_pool
->dp_meta_objset
;
2593 spa_ld_load_trusted_config(spa_t
*spa
, spa_import_type_t type
,
2594 boolean_t reloading
)
2596 vdev_t
*mrvd
, *rvd
= spa
->spa_root_vdev
;
2597 nvlist_t
*nv
, *mos_config
, *policy
;
2598 int error
= 0, copy_error
;
2599 uint64_t healthy_tvds
, healthy_tvds_mos
;
2600 uint64_t mos_config_txg
;
2602 if (spa_dir_prop(spa
, DMU_POOL_CONFIG
, &spa
->spa_config_object
, B_TRUE
)
2604 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2607 * If we're assembling a pool from a split, the config provided is
2608 * already trusted so there is nothing to do.
2610 if (type
== SPA_IMPORT_ASSEMBLE
)
2613 healthy_tvds
= spa_healthy_core_tvds(spa
);
2615 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
)
2617 spa_load_failed(spa
, "unable to retrieve MOS config");
2618 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2622 * If we are doing an open, pool owner wasn't verified yet, thus do
2623 * the verification here.
2625 if (spa
->spa_load_state
== SPA_LOAD_OPEN
) {
2626 error
= spa_verify_host(spa
, mos_config
);
2628 nvlist_free(mos_config
);
2633 nv
= fnvlist_lookup_nvlist(mos_config
, ZPOOL_CONFIG_VDEV_TREE
);
2635 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2638 * Build a new vdev tree from the trusted config
2640 VERIFY(spa_config_parse(spa
, &mrvd
, nv
, NULL
, 0, VDEV_ALLOC_LOAD
) == 0);
2643 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2644 * obtained by scanning /dev/dsk, then it will have the right vdev
2645 * paths. We update the trusted MOS config with this information.
2646 * We first try to copy the paths with vdev_copy_path_strict, which
2647 * succeeds only when both configs have exactly the same vdev tree.
2648 * If that fails, we fall back to a more flexible method that has a
2649 * best effort policy.
2651 copy_error
= vdev_copy_path_strict(rvd
, mrvd
);
2652 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
2653 spa_load_note(spa
, "provided vdev tree:");
2654 vdev_dbgmsg_print_tree(rvd
, 2);
2655 spa_load_note(spa
, "MOS vdev tree:");
2656 vdev_dbgmsg_print_tree(mrvd
, 2);
2658 if (copy_error
!= 0) {
2659 spa_load_note(spa
, "vdev_copy_path_strict failed, falling "
2660 "back to vdev_copy_path_relaxed");
2661 vdev_copy_path_relaxed(rvd
, mrvd
);
2666 spa
->spa_root_vdev
= mrvd
;
2668 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2671 * We will use spa_config if we decide to reload the spa or if spa_load
2672 * fails and we rewind. We must thus regenerate the config using the
2673 * MOS information with the updated paths. Rewind policy is an import
2674 * setting and is not in the MOS. We copy it over to our new, trusted
2677 mos_config_txg
= fnvlist_lookup_uint64(mos_config
,
2678 ZPOOL_CONFIG_POOL_TXG
);
2679 nvlist_free(mos_config
);
2680 mos_config
= spa_config_generate(spa
, NULL
, mos_config_txg
, B_FALSE
);
2681 if (nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_REWIND_POLICY
,
2683 fnvlist_add_nvlist(mos_config
, ZPOOL_REWIND_POLICY
, policy
);
2684 spa_config_set(spa
, mos_config
);
2685 spa
->spa_config_source
= SPA_CONFIG_SRC_MOS
;
2688 * Now that we got the config from the MOS, we should be more strict
2689 * in checking blkptrs and can make assumptions about the consistency
2690 * of the vdev tree. spa_trust_config must be set to true before opening
2691 * vdevs in order for them to be writeable.
2693 spa
->spa_trust_config
= B_TRUE
;
2696 * Open and validate the new vdev tree
2698 error
= spa_ld_open_vdevs(spa
);
2702 error
= spa_ld_validate_vdevs(spa
);
2706 if (copy_error
!= 0 || spa_load_print_vdev_tree
) {
2707 spa_load_note(spa
, "final vdev tree:");
2708 vdev_dbgmsg_print_tree(rvd
, 2);
2711 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
&&
2712 !spa
->spa_extreme_rewind
&& zfs_max_missing_tvds
== 0) {
2714 * Sanity check to make sure that we are indeed loading the
2715 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
2716 * in the config provided and they happened to be the only ones
2717 * to have the latest uberblock, we could involuntarily perform
2718 * an extreme rewind.
2720 healthy_tvds_mos
= spa_healthy_core_tvds(spa
);
2721 if (healthy_tvds_mos
- healthy_tvds
>=
2722 SPA_SYNC_MIN_VDEVS
) {
2723 spa_load_note(spa
, "config provided misses too many "
2724 "top-level vdevs compared to MOS (%lld vs %lld). ",
2725 (u_longlong_t
)healthy_tvds
,
2726 (u_longlong_t
)healthy_tvds_mos
);
2727 spa_load_note(spa
, "vdev tree:");
2728 vdev_dbgmsg_print_tree(rvd
, 2);
2730 spa_load_failed(spa
, "config was already "
2731 "provided from MOS. Aborting.");
2732 return (spa_vdev_err(rvd
,
2733 VDEV_AUX_CORRUPT_DATA
, EIO
));
2735 spa_load_note(spa
, "spa must be reloaded using MOS "
2737 return (SET_ERROR(EAGAIN
));
2741 error
= spa_check_for_missing_logs(spa
);
2743 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
, ENXIO
));
2745 if (rvd
->vdev_guid_sum
!= spa
->spa_uberblock
.ub_guid_sum
) {
2746 spa_load_failed(spa
, "uberblock guid sum doesn't match MOS "
2747 "guid sum (%llu != %llu)",
2748 (u_longlong_t
)spa
->spa_uberblock
.ub_guid_sum
,
2749 (u_longlong_t
)rvd
->vdev_guid_sum
);
2750 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_GUID_SUM
,
2758 spa_ld_open_indirect_vdev_metadata(spa_t
*spa
)
2761 vdev_t
*rvd
= spa
->spa_root_vdev
;
2764 * Everything that we read before spa_remove_init() must be stored
2765 * on concreted vdevs. Therefore we do this as early as possible.
2767 error
= spa_remove_init(spa
);
2769 spa_load_failed(spa
, "spa_remove_init failed [error=%d]",
2771 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2775 * Retrieve information needed to condense indirect vdev mappings.
2777 error
= spa_condense_init(spa
);
2779 spa_load_failed(spa
, "spa_condense_init failed [error=%d]",
2781 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
2788 spa_ld_check_features(spa_t
*spa
, boolean_t
*missing_feat_writep
)
2791 vdev_t
*rvd
= spa
->spa_root_vdev
;
2793 if (spa_version(spa
) >= SPA_VERSION_FEATURES
) {
2794 boolean_t missing_feat_read
= B_FALSE
;
2795 nvlist_t
*unsup_feat
, *enabled_feat
;
2797 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_READ
,
2798 &spa
->spa_feat_for_read_obj
, B_TRUE
) != 0) {
2799 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2802 if (spa_dir_prop(spa
, DMU_POOL_FEATURES_FOR_WRITE
,
2803 &spa
->spa_feat_for_write_obj
, B_TRUE
) != 0) {
2804 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2807 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_DESCRIPTIONS
,
2808 &spa
->spa_feat_desc_obj
, B_TRUE
) != 0) {
2809 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2812 enabled_feat
= fnvlist_alloc();
2813 unsup_feat
= fnvlist_alloc();
2815 if (!spa_features_check(spa
, B_FALSE
,
2816 unsup_feat
, enabled_feat
))
2817 missing_feat_read
= B_TRUE
;
2819 if (spa_writeable(spa
) ||
2820 spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
) {
2821 if (!spa_features_check(spa
, B_TRUE
,
2822 unsup_feat
, enabled_feat
)) {
2823 *missing_feat_writep
= B_TRUE
;
2827 fnvlist_add_nvlist(spa
->spa_load_info
,
2828 ZPOOL_CONFIG_ENABLED_FEAT
, enabled_feat
);
2830 if (!nvlist_empty(unsup_feat
)) {
2831 fnvlist_add_nvlist(spa
->spa_load_info
,
2832 ZPOOL_CONFIG_UNSUP_FEAT
, unsup_feat
);
2835 fnvlist_free(enabled_feat
);
2836 fnvlist_free(unsup_feat
);
2838 if (!missing_feat_read
) {
2839 fnvlist_add_boolean(spa
->spa_load_info
,
2840 ZPOOL_CONFIG_CAN_RDONLY
);
2844 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2845 * twofold: to determine whether the pool is available for
2846 * import in read-write mode and (if it is not) whether the
2847 * pool is available for import in read-only mode. If the pool
2848 * is available for import in read-write mode, it is displayed
2849 * as available in userland; if it is not available for import
2850 * in read-only mode, it is displayed as unavailable in
2851 * userland. If the pool is available for import in read-only
2852 * mode but not read-write mode, it is displayed as unavailable
2853 * in userland with a special note that the pool is actually
2854 * available for open in read-only mode.
2856 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2857 * missing a feature for write, we must first determine whether
2858 * the pool can be opened read-only before returning to
2859 * userland in order to know whether to display the
2860 * abovementioned note.
2862 if (missing_feat_read
|| (*missing_feat_writep
&&
2863 spa_writeable(spa
))) {
2864 spa_load_failed(spa
, "pool uses unsupported features");
2865 return (spa_vdev_err(rvd
, VDEV_AUX_UNSUP_FEAT
,
2870 * Load refcounts for ZFS features from disk into an in-memory
2871 * cache during SPA initialization.
2873 for (spa_feature_t i
= 0; i
< SPA_FEATURES
; i
++) {
2876 error
= feature_get_refcount_from_disk(spa
,
2877 &spa_feature_table
[i
], &refcount
);
2879 spa
->spa_feat_refcount_cache
[i
] = refcount
;
2880 } else if (error
== ENOTSUP
) {
2881 spa
->spa_feat_refcount_cache
[i
] =
2882 SPA_FEATURE_DISABLED
;
2884 spa_load_failed(spa
, "error getting refcount "
2885 "for feature %s [error=%d]",
2886 spa_feature_table
[i
].fi_guid
, error
);
2887 return (spa_vdev_err(rvd
,
2888 VDEV_AUX_CORRUPT_DATA
, EIO
));
2893 if (spa_feature_is_active(spa
, SPA_FEATURE_ENABLED_TXG
)) {
2894 if (spa_dir_prop(spa
, DMU_POOL_FEATURE_ENABLED_TXG
,
2895 &spa
->spa_feat_enabled_txg_obj
, B_TRUE
) != 0)
2896 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2903 spa_ld_load_special_directories(spa_t
*spa
)
2906 vdev_t
*rvd
= spa
->spa_root_vdev
;
2908 spa
->spa_is_initializing
= B_TRUE
;
2909 error
= dsl_pool_open(spa
->spa_dsl_pool
);
2910 spa
->spa_is_initializing
= B_FALSE
;
2912 spa_load_failed(spa
, "dsl_pool_open failed [error=%d]", error
);
2913 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2920 spa_ld_get_props(spa_t
*spa
)
2924 vdev_t
*rvd
= spa
->spa_root_vdev
;
2926 /* Grab the secret checksum salt from the MOS. */
2927 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
2928 DMU_POOL_CHECKSUM_SALT
, 1,
2929 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
2930 spa
->spa_cksum_salt
.zcs_bytes
);
2931 if (error
== ENOENT
) {
2932 /* Generate a new salt for subsequent use */
2933 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
2934 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
2935 } else if (error
!= 0) {
2936 spa_load_failed(spa
, "unable to retrieve checksum salt from "
2937 "MOS [error=%d]", error
);
2938 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2941 if (spa_dir_prop(spa
, DMU_POOL_SYNC_BPOBJ
, &obj
, B_TRUE
) != 0)
2942 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2943 error
= bpobj_open(&spa
->spa_deferred_bpobj
, spa
->spa_meta_objset
, obj
);
2945 spa_load_failed(spa
, "error opening deferred-frees bpobj "
2946 "[error=%d]", error
);
2947 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2951 * Load the bit that tells us to use the new accounting function
2952 * (raid-z deflation). If we have an older pool, this will not
2955 error
= spa_dir_prop(spa
, DMU_POOL_DEFLATE
, &spa
->spa_deflate
, B_FALSE
);
2956 if (error
!= 0 && error
!= ENOENT
)
2957 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2959 error
= spa_dir_prop(spa
, DMU_POOL_CREATION_VERSION
,
2960 &spa
->spa_creation_version
, B_FALSE
);
2961 if (error
!= 0 && error
!= ENOENT
)
2962 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2965 * Load the persistent error log. If we have an older pool, this will
2968 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_LAST
, &spa
->spa_errlog_last
,
2970 if (error
!= 0 && error
!= ENOENT
)
2971 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2973 error
= spa_dir_prop(spa
, DMU_POOL_ERRLOG_SCRUB
,
2974 &spa
->spa_errlog_scrub
, B_FALSE
);
2975 if (error
!= 0 && error
!= ENOENT
)
2976 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2979 * Load the history object. If we have an older pool, this
2980 * will not be present.
2982 error
= spa_dir_prop(spa
, DMU_POOL_HISTORY
, &spa
->spa_history
, B_FALSE
);
2983 if (error
!= 0 && error
!= ENOENT
)
2984 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
2987 * Load the per-vdev ZAP map. If we have an older pool, this will not
2988 * be present; in this case, defer its creation to a later time to
2989 * avoid dirtying the MOS this early / out of sync context. See
2990 * spa_sync_config_object.
2993 /* The sentinel is only available in the MOS config. */
2994 nvlist_t
*mos_config
;
2995 if (load_nvlist(spa
, spa
->spa_config_object
, &mos_config
) != 0) {
2996 spa_load_failed(spa
, "unable to retrieve MOS config");
2997 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3000 error
= spa_dir_prop(spa
, DMU_POOL_VDEV_ZAP_MAP
,
3001 &spa
->spa_all_vdev_zaps
, B_FALSE
);
3003 if (error
== ENOENT
) {
3004 VERIFY(!nvlist_exists(mos_config
,
3005 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
3006 spa
->spa_avz_action
= AVZ_ACTION_INITIALIZE
;
3007 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3008 } else if (error
!= 0) {
3009 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3010 } else if (!nvlist_exists(mos_config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
)) {
3012 * An older version of ZFS overwrote the sentinel value, so
3013 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3014 * destruction to later; see spa_sync_config_object.
3016 spa
->spa_avz_action
= AVZ_ACTION_DESTROY
;
3018 * We're assuming that no vdevs have had their ZAPs created
3019 * before this. Better be sure of it.
3021 ASSERT0(vdev_count_verify_zaps(spa
->spa_root_vdev
));
3023 nvlist_free(mos_config
);
3025 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
3027 error
= spa_dir_prop(spa
, DMU_POOL_PROPS
, &spa
->spa_pool_props_object
,
3029 if (error
&& error
!= ENOENT
)
3030 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3033 uint64_t autoreplace
;
3035 spa_prop_find(spa
, ZPOOL_PROP_BOOTFS
, &spa
->spa_bootfs
);
3036 spa_prop_find(spa
, ZPOOL_PROP_AUTOREPLACE
, &autoreplace
);
3037 spa_prop_find(spa
, ZPOOL_PROP_DELEGATION
, &spa
->spa_delegation
);
3038 spa_prop_find(spa
, ZPOOL_PROP_FAILUREMODE
, &spa
->spa_failmode
);
3039 spa_prop_find(spa
, ZPOOL_PROP_AUTOEXPAND
, &spa
->spa_autoexpand
);
3040 spa_prop_find(spa
, ZPOOL_PROP_DEDUPDITTO
,
3041 &spa
->spa_dedup_ditto
);
3043 spa
->spa_autoreplace
= (autoreplace
!= 0);
3047 * If we are importing a pool with missing top-level vdevs,
3048 * we enforce that the pool doesn't panic or get suspended on
3049 * error since the likelihood of missing data is extremely high.
3051 if (spa
->spa_missing_tvds
> 0 &&
3052 spa
->spa_failmode
!= ZIO_FAILURE_MODE_CONTINUE
&&
3053 spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
3054 spa_load_note(spa
, "forcing failmode to 'continue' "
3055 "as some top level vdevs are missing");
3056 spa
->spa_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
3063 spa_ld_open_aux_vdevs(spa_t
*spa
, spa_import_type_t type
)
3066 vdev_t
*rvd
= spa
->spa_root_vdev
;
3069 * If we're assembling the pool from the split-off vdevs of
3070 * an existing pool, we don't want to attach the spares & cache
3075 * Load any hot spares for this pool.
3077 error
= spa_dir_prop(spa
, DMU_POOL_SPARES
, &spa
->spa_spares
.sav_object
,
3079 if (error
!= 0 && error
!= ENOENT
)
3080 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3081 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3082 ASSERT(spa_version(spa
) >= SPA_VERSION_SPARES
);
3083 if (load_nvlist(spa
, spa
->spa_spares
.sav_object
,
3084 &spa
->spa_spares
.sav_config
) != 0) {
3085 spa_load_failed(spa
, "error loading spares nvlist");
3086 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3089 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3090 spa_load_spares(spa
);
3091 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3092 } else if (error
== 0) {
3093 spa
->spa_spares
.sav_sync
= B_TRUE
;
3097 * Load any level 2 ARC devices for this pool.
3099 error
= spa_dir_prop(spa
, DMU_POOL_L2CACHE
,
3100 &spa
->spa_l2cache
.sav_object
, B_FALSE
);
3101 if (error
!= 0 && error
!= ENOENT
)
3102 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3103 if (error
== 0 && type
!= SPA_IMPORT_ASSEMBLE
) {
3104 ASSERT(spa_version(spa
) >= SPA_VERSION_L2CACHE
);
3105 if (load_nvlist(spa
, spa
->spa_l2cache
.sav_object
,
3106 &spa
->spa_l2cache
.sav_config
) != 0) {
3107 spa_load_failed(spa
, "error loading l2cache nvlist");
3108 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3111 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3112 spa_load_l2cache(spa
);
3113 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3114 } else if (error
== 0) {
3115 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
3122 spa_ld_load_vdev_metadata(spa_t
*spa
)
3125 vdev_t
*rvd
= spa
->spa_root_vdev
;
3128 * If the 'autoreplace' property is set, then post a resource notifying
3129 * the ZFS DE that it should not issue any faults for unopenable
3130 * devices. We also iterate over the vdevs, and post a sysevent for any
3131 * unopenable vdevs so that the normal autoreplace handler can take
3134 if (spa
->spa_autoreplace
&& spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
3135 spa_check_removed(spa
->spa_root_vdev
);
3137 * For the import case, this is done in spa_import(), because
3138 * at this point we're using the spare definitions from
3139 * the MOS config, not necessarily from the userland config.
3141 if (spa
->spa_load_state
!= SPA_LOAD_IMPORT
) {
3142 spa_aux_check_removed(&spa
->spa_spares
);
3143 spa_aux_check_removed(&spa
->spa_l2cache
);
3148 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3150 error
= vdev_load(rvd
);
3152 spa_load_failed(spa
, "vdev_load failed [error=%d]", error
);
3153 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, error
));
3157 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3159 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3160 vdev_dtl_reassess(rvd
, 0, 0, B_FALSE
);
3161 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3167 spa_ld_load_dedup_tables(spa_t
*spa
)
3170 vdev_t
*rvd
= spa
->spa_root_vdev
;
3172 error
= ddt_load(spa
);
3174 spa_load_failed(spa
, "ddt_load failed [error=%d]", error
);
3175 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
, EIO
));
3182 spa_ld_verify_logs(spa_t
*spa
, spa_import_type_t type
, char **ereport
)
3184 vdev_t
*rvd
= spa
->spa_root_vdev
;
3186 if (type
!= SPA_IMPORT_ASSEMBLE
&& spa_writeable(spa
)) {
3187 boolean_t missing
= spa_check_logs(spa
);
3189 if (spa
->spa_missing_tvds
!= 0) {
3190 spa_load_note(spa
, "spa_check_logs failed "
3191 "so dropping the logs");
3193 *ereport
= FM_EREPORT_ZFS_LOG_REPLAY
;
3194 spa_load_failed(spa
, "spa_check_logs failed");
3195 return (spa_vdev_err(rvd
, VDEV_AUX_BAD_LOG
,
3205 spa_ld_verify_pool_data(spa_t
*spa
)
3208 vdev_t
*rvd
= spa
->spa_root_vdev
;
3211 * We've successfully opened the pool, verify that we're ready
3212 * to start pushing transactions.
3214 if (spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
) {
3215 error
= spa_load_verify(spa
);
3217 spa_load_failed(spa
, "spa_load_verify failed "
3218 "[error=%d]", error
);
3219 return (spa_vdev_err(rvd
, VDEV_AUX_CORRUPT_DATA
,
3228 spa_ld_claim_log_blocks(spa_t
*spa
)
3231 dsl_pool_t
*dp
= spa_get_dsl(spa
);
3234 * Claim log blocks that haven't been committed yet.
3235 * This must all happen in a single txg.
3236 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3237 * invoked from zil_claim_log_block()'s i/o done callback.
3238 * Price of rollback is that we abandon the log.
3240 spa
->spa_claiming
= B_TRUE
;
3242 tx
= dmu_tx_create_assigned(dp
, spa_first_txg(spa
));
3243 (void) dmu_objset_find_dp(dp
, dp
->dp_root_dir_obj
,
3244 zil_claim
, tx
, DS_FIND_CHILDREN
);
3247 spa
->spa_claiming
= B_FALSE
;
3249 spa_set_log_state(spa
, SPA_LOG_GOOD
);
3253 spa_ld_check_for_config_update(spa_t
*spa
, uint64_t config_cache_txg
,
3254 boolean_t reloading
)
3256 vdev_t
*rvd
= spa
->spa_root_vdev
;
3257 int need_update
= B_FALSE
;
3260 * If the config cache is stale, or we have uninitialized
3261 * metaslabs (see spa_vdev_add()), then update the config.
3263 * If this is a verbatim import, trust the current
3264 * in-core spa_config and update the disk labels.
3266 if (reloading
|| config_cache_txg
!= spa
->spa_config_txg
||
3267 spa
->spa_load_state
== SPA_LOAD_IMPORT
||
3268 spa
->spa_load_state
== SPA_LOAD_RECOVER
||
3269 (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
))
3270 need_update
= B_TRUE
;
3272 for (int c
= 0; c
< rvd
->vdev_children
; c
++)
3273 if (rvd
->vdev_child
[c
]->vdev_ms_array
== 0)
3274 need_update
= B_TRUE
;
3277 * Update the config cache asychronously in case we're the
3278 * root pool, in which case the config cache isn't writable yet.
3281 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3285 spa_ld_prepare_for_reload(spa_t
*spa
)
3287 int mode
= spa
->spa_mode
;
3288 int async_suspended
= spa
->spa_async_suspended
;
3291 spa_deactivate(spa
);
3292 spa_activate(spa
, mode
);
3295 * We save the value of spa_async_suspended as it gets reset to 0 by
3296 * spa_unload(). We want to restore it back to the original value before
3297 * returning as we might be calling spa_async_resume() later.
3299 spa
->spa_async_suspended
= async_suspended
;
3303 * Load an existing storage pool, using the config provided. This config
3304 * describes which vdevs are part of the pool and is later validated against
3305 * partial configs present in each vdev's label and an entire copy of the
3306 * config stored in the MOS.
3309 spa_load_impl(spa_t
*spa
, spa_import_type_t type
, char **ereport
,
3310 boolean_t reloading
)
3313 boolean_t missing_feat_write
= B_FALSE
;
3315 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
3316 ASSERT(spa
->spa_config_source
!= SPA_CONFIG_SRC_NONE
);
3319 * Never trust the config that is provided unless we are assembling
3320 * a pool following a split.
3321 * This means don't trust blkptrs and the vdev tree in general. This
3322 * also effectively puts the spa in read-only mode since
3323 * spa_writeable() checks for spa_trust_config to be true.
3324 * We will later load a trusted config from the MOS.
3326 if (type
!= SPA_IMPORT_ASSEMBLE
)
3327 spa
->spa_trust_config
= B_FALSE
;
3330 spa_load_note(spa
, "RELOADING");
3332 spa_load_note(spa
, "LOADING");
3335 * Parse the config provided to create a vdev tree.
3337 error
= spa_ld_parse_config(spa
, type
);
3342 * Now that we have the vdev tree, try to open each vdev. This involves
3343 * opening the underlying physical device, retrieving its geometry and
3344 * probing the vdev with a dummy I/O. The state of each vdev will be set
3345 * based on the success of those operations. After this we'll be ready
3346 * to read from the vdevs.
3348 error
= spa_ld_open_vdevs(spa
);
3353 * Read the label of each vdev and make sure that the GUIDs stored
3354 * there match the GUIDs in the config provided.
3355 * If we're assembling a new pool that's been split off from an
3356 * existing pool, the labels haven't yet been updated so we skip
3357 * validation for now.
3359 if (type
!= SPA_IMPORT_ASSEMBLE
) {
3360 error
= spa_ld_validate_vdevs(spa
);
3366 * Read vdev labels to find the best uberblock (i.e. latest, unless
3367 * spa_load_max_txg is set) and store it in spa_uberblock. We get the
3368 * list of features required to read blkptrs in the MOS from the vdev
3369 * label with the best uberblock and verify that our version of zfs
3370 * supports them all.
3372 error
= spa_ld_select_uberblock(spa
, type
);
3377 * Pass that uberblock to the dsl_pool layer which will open the root
3378 * blkptr. This blkptr points to the latest version of the MOS and will
3379 * allow us to read its contents.
3381 error
= spa_ld_open_rootbp(spa
);
3386 * Retrieve the trusted config stored in the MOS and use it to create
3387 * a new, exact version of the vdev tree, then reopen all vdevs.
3389 error
= spa_ld_load_trusted_config(spa
, type
, reloading
);
3390 if (error
== EAGAIN
) {
3393 * Redo the loading process with the trusted config if it is
3394 * too different from the untrusted config.
3396 spa_ld_prepare_for_reload(spa
);
3397 return (spa_load_impl(spa
, type
, ereport
, B_TRUE
));
3398 } else if (error
!= 0) {
3403 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3404 * from the pool and their contents were re-mapped to other vdevs. Note
3405 * that everything that we read before this step must have been
3406 * rewritten on concrete vdevs after the last device removal was
3407 * initiated. Otherwise we could be reading from indirect vdevs before
3408 * we have loaded their mappings.
3410 error
= spa_ld_open_indirect_vdev_metadata(spa
);
3415 * Retrieve the full list of active features from the MOS and check if
3416 * they are all supported.
3418 error
= spa_ld_check_features(spa
, &missing_feat_write
);
3423 * Load several special directories from the MOS needed by the dsl_pool
3426 error
= spa_ld_load_special_directories(spa
);
3431 * Retrieve pool properties from the MOS.
3433 error
= spa_ld_get_props(spa
);
3438 * Retrieve the list of auxiliary devices - cache devices and spares -
3441 error
= spa_ld_open_aux_vdevs(spa
, type
);
3446 * Load the metadata for all vdevs. Also check if unopenable devices
3447 * should be autoreplaced.
3449 error
= spa_ld_load_vdev_metadata(spa
);
3453 error
= spa_ld_load_dedup_tables(spa
);
3458 * Verify the logs now to make sure we don't have any unexpected errors
3459 * when we claim log blocks later.
3461 error
= spa_ld_verify_logs(spa
, type
, ereport
);
3465 if (missing_feat_write
) {
3466 ASSERT(spa
->spa_load_state
== SPA_LOAD_TRYIMPORT
);
3469 * At this point, we know that we can open the pool in
3470 * read-only mode but not read-write mode. We now have enough
3471 * information and can return to userland.
3473 return (spa_vdev_err(spa
->spa_root_vdev
, VDEV_AUX_UNSUP_FEAT
,
3478 * Traverse the last txgs to make sure the pool was left off in a safe
3479 * state. When performing an extreme rewind, we verify the whole pool,
3480 * which can take a very long time.
3482 error
= spa_ld_verify_pool_data(spa
);
3487 * Calculate the deflated space for the pool. This must be done before
3488 * we write anything to the pool because we'd need to update the space
3489 * accounting using the deflated sizes.
3491 spa_update_dspace(spa
);
3494 * We have now retrieved all the information we needed to open the
3495 * pool. If we are importing the pool in read-write mode, a few
3496 * additional steps must be performed to finish the import.
3498 if (spa_writeable(spa
) && (spa
->spa_load_state
== SPA_LOAD_RECOVER
||
3499 spa
->spa_load_max_txg
== UINT64_MAX
)) {
3500 uint64_t config_cache_txg
= spa
->spa_config_txg
;
3502 ASSERT(spa
->spa_load_state
!= SPA_LOAD_TRYIMPORT
);
3505 * Traverse the ZIL and claim all blocks.
3507 spa_ld_claim_log_blocks(spa
);
3510 * Kick-off the syncing thread.
3512 spa
->spa_sync_on
= B_TRUE
;
3513 txg_sync_start(spa
->spa_dsl_pool
);
3516 * Wait for all claims to sync. We sync up to the highest
3517 * claimed log block birth time so that claimed log blocks
3518 * don't appear to be from the future. spa_claim_max_txg
3519 * will have been set for us by ZIL traversal operations
3522 txg_wait_synced(spa
->spa_dsl_pool
, spa
->spa_claim_max_txg
);
3525 * Check if we need to request an update of the config. On the
3526 * next sync, we would update the config stored in vdev labels
3527 * and the cachefile (by default /etc/zfs/zpool.cache).
3529 spa_ld_check_for_config_update(spa
, config_cache_txg
,
3533 * Check all DTLs to see if anything needs resilvering.
3535 if (!dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3536 vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
))
3537 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3540 * Log the fact that we booted up (so that we can detect if
3541 * we rebooted in the middle of an operation).
3543 spa_history_log_version(spa
, "open");
3546 * Delete any inconsistent datasets.
3548 (void) dmu_objset_find(spa_name(spa
),
3549 dsl_destroy_inconsistent
, NULL
, DS_FIND_CHILDREN
);
3552 * Clean up any stale temporary dataset userrefs.
3554 dsl_pool_clean_tmp_userrefs(spa
->spa_dsl_pool
);
3556 spa_restart_removal(spa
);
3558 spa_spawn_aux_threads(spa
);
3561 spa_load_note(spa
, "LOADED");
3567 spa_load_retry(spa_t
*spa
, spa_load_state_t state
)
3569 int mode
= spa
->spa_mode
;
3572 spa_deactivate(spa
);
3574 spa
->spa_load_max_txg
= spa
->spa_uberblock
.ub_txg
- 1;
3576 spa_activate(spa
, mode
);
3577 spa_async_suspend(spa
);
3579 spa_load_note(spa
, "spa_load_retry: rewind, max txg: %llu",
3580 (u_longlong_t
)spa
->spa_load_max_txg
);
3582 return (spa_load(spa
, state
, SPA_IMPORT_EXISTING
));
3586 * If spa_load() fails this function will try loading prior txg's. If
3587 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3588 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3589 * function will not rewind the pool and will return the same error as
3593 spa_load_best(spa_t
*spa
, spa_load_state_t state
, uint64_t max_request
,
3596 nvlist_t
*loadinfo
= NULL
;
3597 nvlist_t
*config
= NULL
;
3598 int load_error
, rewind_error
;
3599 uint64_t safe_rewind_txg
;
3602 if (spa
->spa_load_txg
&& state
== SPA_LOAD_RECOVER
) {
3603 spa
->spa_load_max_txg
= spa
->spa_load_txg
;
3604 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3606 spa
->spa_load_max_txg
= max_request
;
3607 if (max_request
!= UINT64_MAX
)
3608 spa
->spa_extreme_rewind
= B_TRUE
;
3611 load_error
= rewind_error
= spa_load(spa
, state
, SPA_IMPORT_EXISTING
);
3612 if (load_error
== 0)
3615 if (spa
->spa_root_vdev
!= NULL
)
3616 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3618 spa
->spa_last_ubsync_txg
= spa
->spa_uberblock
.ub_txg
;
3619 spa
->spa_last_ubsync_txg_ts
= spa
->spa_uberblock
.ub_timestamp
;
3621 if (rewind_flags
& ZPOOL_NEVER_REWIND
) {
3622 nvlist_free(config
);
3623 return (load_error
);
3626 if (state
== SPA_LOAD_RECOVER
) {
3627 /* Price of rolling back is discarding txgs, including log */
3628 spa_set_log_state(spa
, SPA_LOG_CLEAR
);
3631 * If we aren't rolling back save the load info from our first
3632 * import attempt so that we can restore it after attempting
3635 loadinfo
= spa
->spa_load_info
;
3636 spa
->spa_load_info
= fnvlist_alloc();
3639 spa
->spa_load_max_txg
= spa
->spa_last_ubsync_txg
;
3640 safe_rewind_txg
= spa
->spa_last_ubsync_txg
- TXG_DEFER_SIZE
;
3641 min_txg
= (rewind_flags
& ZPOOL_EXTREME_REWIND
) ?
3642 TXG_INITIAL
: safe_rewind_txg
;
3645 * Continue as long as we're finding errors, we're still within
3646 * the acceptable rewind range, and we're still finding uberblocks
3648 while (rewind_error
&& spa
->spa_uberblock
.ub_txg
>= min_txg
&&
3649 spa
->spa_uberblock
.ub_txg
<= spa
->spa_load_max_txg
) {
3650 if (spa
->spa_load_max_txg
< safe_rewind_txg
)
3651 spa
->spa_extreme_rewind
= B_TRUE
;
3652 rewind_error
= spa_load_retry(spa
, state
);
3655 spa
->spa_extreme_rewind
= B_FALSE
;
3656 spa
->spa_load_max_txg
= UINT64_MAX
;
3658 if (config
&& (rewind_error
|| state
!= SPA_LOAD_RECOVER
))
3659 spa_config_set(spa
, config
);
3661 nvlist_free(config
);
3663 if (state
== SPA_LOAD_RECOVER
) {
3664 ASSERT3P(loadinfo
, ==, NULL
);
3665 return (rewind_error
);
3667 /* Store the rewind info as part of the initial load info */
3668 fnvlist_add_nvlist(loadinfo
, ZPOOL_CONFIG_REWIND_INFO
,
3669 spa
->spa_load_info
);
3671 /* Restore the initial load info */
3672 fnvlist_free(spa
->spa_load_info
);
3673 spa
->spa_load_info
= loadinfo
;
3675 return (load_error
);
3682 * The import case is identical to an open except that the configuration is sent
3683 * down from userland, instead of grabbed from the configuration cache. For the
3684 * case of an open, the pool configuration will exist in the
3685 * POOL_STATE_UNINITIALIZED state.
3687 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3688 * the same time open the pool, without having to keep around the spa_t in some
3692 spa_open_common(const char *pool
, spa_t
**spapp
, void *tag
, nvlist_t
*nvpolicy
,
3696 spa_load_state_t state
= SPA_LOAD_OPEN
;
3698 int locked
= B_FALSE
;
3703 * As disgusting as this is, we need to support recursive calls to this
3704 * function because dsl_dir_open() is called during spa_load(), and ends
3705 * up calling spa_open() again. The real fix is to figure out how to
3706 * avoid dsl_dir_open() calling this in the first place.
3708 if (mutex_owner(&spa_namespace_lock
) != curthread
) {
3709 mutex_enter(&spa_namespace_lock
);
3713 if ((spa
= spa_lookup(pool
)) == NULL
) {
3715 mutex_exit(&spa_namespace_lock
);
3716 return (SET_ERROR(ENOENT
));
3719 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
) {
3720 zpool_rewind_policy_t policy
;
3722 zpool_get_rewind_policy(nvpolicy
? nvpolicy
: spa
->spa_config
,
3724 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
3725 state
= SPA_LOAD_RECOVER
;
3727 spa_activate(spa
, spa_mode_global
);
3729 if (state
!= SPA_LOAD_RECOVER
)
3730 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
3731 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
3733 zfs_dbgmsg("spa_open_common: opening %s", pool
);
3734 error
= spa_load_best(spa
, state
, policy
.zrp_txg
,
3735 policy
.zrp_request
);
3737 if (error
== EBADF
) {
3739 * If vdev_validate() returns failure (indicated by
3740 * EBADF), it indicates that one of the vdevs indicates
3741 * that the pool has been exported or destroyed. If
3742 * this is the case, the config cache is out of sync and
3743 * we should remove the pool from the namespace.
3746 spa_deactivate(spa
);
3747 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
3750 mutex_exit(&spa_namespace_lock
);
3751 return (SET_ERROR(ENOENT
));
3756 * We can't open the pool, but we still have useful
3757 * information: the state of each vdev after the
3758 * attempted vdev_open(). Return this to the user.
3760 if (config
!= NULL
&& spa
->spa_config
) {
3761 VERIFY(nvlist_dup(spa
->spa_config
, config
,
3763 VERIFY(nvlist_add_nvlist(*config
,
3764 ZPOOL_CONFIG_LOAD_INFO
,
3765 spa
->spa_load_info
) == 0);
3768 spa_deactivate(spa
);
3769 spa
->spa_last_open_failed
= error
;
3771 mutex_exit(&spa_namespace_lock
);
3777 spa_open_ref(spa
, tag
);
3780 *config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
3783 * If we've recovered the pool, pass back any information we
3784 * gathered while doing the load.
3786 if (state
== SPA_LOAD_RECOVER
) {
3787 VERIFY(nvlist_add_nvlist(*config
, ZPOOL_CONFIG_LOAD_INFO
,
3788 spa
->spa_load_info
) == 0);
3792 spa
->spa_last_open_failed
= 0;
3793 spa
->spa_last_ubsync_txg
= 0;
3794 spa
->spa_load_txg
= 0;
3795 mutex_exit(&spa_namespace_lock
);
3804 spa_open_rewind(const char *name
, spa_t
**spapp
, void *tag
, nvlist_t
*policy
,
3807 return (spa_open_common(name
, spapp
, tag
, policy
, config
));
3811 spa_open(const char *name
, spa_t
**spapp
, void *tag
)
3813 return (spa_open_common(name
, spapp
, tag
, NULL
, NULL
));
3817 * Lookup the given spa_t, incrementing the inject count in the process,
3818 * preventing it from being exported or destroyed.
3821 spa_inject_addref(char *name
)
3825 mutex_enter(&spa_namespace_lock
);
3826 if ((spa
= spa_lookup(name
)) == NULL
) {
3827 mutex_exit(&spa_namespace_lock
);
3830 spa
->spa_inject_ref
++;
3831 mutex_exit(&spa_namespace_lock
);
3837 spa_inject_delref(spa_t
*spa
)
3839 mutex_enter(&spa_namespace_lock
);
3840 spa
->spa_inject_ref
--;
3841 mutex_exit(&spa_namespace_lock
);
3845 * Add spares device information to the nvlist.
3848 spa_add_spares(spa_t
*spa
, nvlist_t
*config
)
3858 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3860 if (spa
->spa_spares
.sav_count
== 0)
3863 VERIFY(nvlist_lookup_nvlist(config
,
3864 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3865 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_spares
.sav_config
,
3866 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3868 VERIFY(nvlist_add_nvlist_array(nvroot
,
3869 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
3870 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3871 ZPOOL_CONFIG_SPARES
, &spares
, &nspares
) == 0);
3874 * Go through and find any spares which have since been
3875 * repurposed as an active spare. If this is the case, update
3876 * their status appropriately.
3878 for (i
= 0; i
< nspares
; i
++) {
3879 VERIFY(nvlist_lookup_uint64(spares
[i
],
3880 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3881 if (spa_spare_exists(guid
, &pool
, NULL
) &&
3883 VERIFY(nvlist_lookup_uint64_array(
3884 spares
[i
], ZPOOL_CONFIG_VDEV_STATS
,
3885 (uint64_t **)&vs
, &vsc
) == 0);
3886 vs
->vs_state
= VDEV_STATE_CANT_OPEN
;
3887 vs
->vs_aux
= VDEV_AUX_SPARED
;
3894 * Add l2cache device information to the nvlist, including vdev stats.
3897 spa_add_l2cache(spa_t
*spa
, nvlist_t
*config
)
3900 uint_t i
, j
, nl2cache
;
3907 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3909 if (spa
->spa_l2cache
.sav_count
== 0)
3912 VERIFY(nvlist_lookup_nvlist(config
,
3913 ZPOOL_CONFIG_VDEV_TREE
, &nvroot
) == 0);
3914 VERIFY(nvlist_lookup_nvlist_array(spa
->spa_l2cache
.sav_config
,
3915 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3916 if (nl2cache
!= 0) {
3917 VERIFY(nvlist_add_nvlist_array(nvroot
,
3918 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
3919 VERIFY(nvlist_lookup_nvlist_array(nvroot
,
3920 ZPOOL_CONFIG_L2CACHE
, &l2cache
, &nl2cache
) == 0);
3923 * Update level 2 cache device stats.
3926 for (i
= 0; i
< nl2cache
; i
++) {
3927 VERIFY(nvlist_lookup_uint64(l2cache
[i
],
3928 ZPOOL_CONFIG_GUID
, &guid
) == 0);
3931 for (j
= 0; j
< spa
->spa_l2cache
.sav_count
; j
++) {
3933 spa
->spa_l2cache
.sav_vdevs
[j
]->vdev_guid
) {
3934 vd
= spa
->spa_l2cache
.sav_vdevs
[j
];
3940 VERIFY(nvlist_lookup_uint64_array(l2cache
[i
],
3941 ZPOOL_CONFIG_VDEV_STATS
, (uint64_t **)&vs
, &vsc
)
3943 vdev_get_stats(vd
, vs
);
3949 spa_add_feature_stats(spa_t
*spa
, nvlist_t
*config
)
3955 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_READER
));
3956 VERIFY(nvlist_alloc(&features
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
3958 if (spa
->spa_feat_for_read_obj
!= 0) {
3959 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3960 spa
->spa_feat_for_read_obj
);
3961 zap_cursor_retrieve(&zc
, &za
) == 0;
3962 zap_cursor_advance(&zc
)) {
3963 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3964 za
.za_num_integers
== 1);
3965 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
3966 za
.za_first_integer
));
3968 zap_cursor_fini(&zc
);
3971 if (spa
->spa_feat_for_write_obj
!= 0) {
3972 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
3973 spa
->spa_feat_for_write_obj
);
3974 zap_cursor_retrieve(&zc
, &za
) == 0;
3975 zap_cursor_advance(&zc
)) {
3976 ASSERT(za
.za_integer_length
== sizeof (uint64_t) &&
3977 za
.za_num_integers
== 1);
3978 VERIFY3U(0, ==, nvlist_add_uint64(features
, za
.za_name
,
3979 za
.za_first_integer
));
3981 zap_cursor_fini(&zc
);
3984 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_FEATURE_STATS
,
3986 nvlist_free(features
);
3990 spa_get_stats(const char *name
, nvlist_t
**config
,
3991 char *altroot
, size_t buflen
)
3997 error
= spa_open_common(name
, &spa
, FTAG
, NULL
, config
);
4001 * This still leaves a window of inconsistency where the spares
4002 * or l2cache devices could change and the config would be
4003 * self-inconsistent.
4005 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4007 if (*config
!= NULL
) {
4008 uint64_t loadtimes
[2];
4010 loadtimes
[0] = spa
->spa_loaded_ts
.tv_sec
;
4011 loadtimes
[1] = spa
->spa_loaded_ts
.tv_nsec
;
4012 VERIFY(nvlist_add_uint64_array(*config
,
4013 ZPOOL_CONFIG_LOADED_TIME
, loadtimes
, 2) == 0);
4015 VERIFY(nvlist_add_uint64(*config
,
4016 ZPOOL_CONFIG_ERRCOUNT
,
4017 spa_get_errlog_size(spa
)) == 0);
4019 if (spa_suspended(spa
))
4020 VERIFY(nvlist_add_uint64(*config
,
4021 ZPOOL_CONFIG_SUSPENDED
,
4022 spa
->spa_failmode
) == 0);
4024 spa_add_spares(spa
, *config
);
4025 spa_add_l2cache(spa
, *config
);
4026 spa_add_feature_stats(spa
, *config
);
4031 * We want to get the alternate root even for faulted pools, so we cheat
4032 * and call spa_lookup() directly.
4036 mutex_enter(&spa_namespace_lock
);
4037 spa
= spa_lookup(name
);
4039 spa_altroot(spa
, altroot
, buflen
);
4043 mutex_exit(&spa_namespace_lock
);
4045 spa_altroot(spa
, altroot
, buflen
);
4050 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4051 spa_close(spa
, FTAG
);
4058 * Validate that the auxiliary device array is well formed. We must have an
4059 * array of nvlists, each which describes a valid leaf vdev. If this is an
4060 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4061 * specified, as long as they are well-formed.
4064 spa_validate_aux_devs(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
,
4065 spa_aux_vdev_t
*sav
, const char *config
, uint64_t version
,
4066 vdev_labeltype_t label
)
4073 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4076 * It's acceptable to have no devs specified.
4078 if (nvlist_lookup_nvlist_array(nvroot
, config
, &dev
, &ndev
) != 0)
4082 return (SET_ERROR(EINVAL
));
4085 * Make sure the pool is formatted with a version that supports this
4088 if (spa_version(spa
) < version
)
4089 return (SET_ERROR(ENOTSUP
));
4092 * Set the pending device list so we correctly handle device in-use
4095 sav
->sav_pending
= dev
;
4096 sav
->sav_npending
= ndev
;
4098 for (i
= 0; i
< ndev
; i
++) {
4099 if ((error
= spa_config_parse(spa
, &vd
, dev
[i
], NULL
, 0,
4103 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4105 error
= SET_ERROR(EINVAL
);
4110 * The L2ARC currently only supports disk devices in
4111 * kernel context. For user-level testing, we allow it.
4114 if ((strcmp(config
, ZPOOL_CONFIG_L2CACHE
) == 0) &&
4115 strcmp(vd
->vdev_ops
->vdev_op_type
, VDEV_TYPE_DISK
) != 0) {
4116 error
= SET_ERROR(ENOTBLK
);
4123 if ((error
= vdev_open(vd
)) == 0 &&
4124 (error
= vdev_label_init(vd
, crtxg
, label
)) == 0) {
4125 VERIFY(nvlist_add_uint64(dev
[i
], ZPOOL_CONFIG_GUID
,
4126 vd
->vdev_guid
) == 0);
4132 (mode
!= VDEV_ALLOC_SPARE
&& mode
!= VDEV_ALLOC_L2CACHE
))
4139 sav
->sav_pending
= NULL
;
4140 sav
->sav_npending
= 0;
4145 spa_validate_aux(spa_t
*spa
, nvlist_t
*nvroot
, uint64_t crtxg
, int mode
)
4149 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4151 if ((error
= spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
4152 &spa
->spa_spares
, ZPOOL_CONFIG_SPARES
, SPA_VERSION_SPARES
,
4153 VDEV_LABEL_SPARE
)) != 0) {
4157 return (spa_validate_aux_devs(spa
, nvroot
, crtxg
, mode
,
4158 &spa
->spa_l2cache
, ZPOOL_CONFIG_L2CACHE
, SPA_VERSION_L2CACHE
,
4159 VDEV_LABEL_L2CACHE
));
4163 spa_set_aux_vdevs(spa_aux_vdev_t
*sav
, nvlist_t
**devs
, int ndevs
,
4168 if (sav
->sav_config
!= NULL
) {
4174 * Generate new dev list by concatentating with the
4177 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
, config
,
4178 &olddevs
, &oldndevs
) == 0);
4180 newdevs
= kmem_alloc(sizeof (void *) *
4181 (ndevs
+ oldndevs
), KM_SLEEP
);
4182 for (i
= 0; i
< oldndevs
; i
++)
4183 VERIFY(nvlist_dup(olddevs
[i
], &newdevs
[i
],
4185 for (i
= 0; i
< ndevs
; i
++)
4186 VERIFY(nvlist_dup(devs
[i
], &newdevs
[i
+ oldndevs
],
4189 VERIFY(nvlist_remove(sav
->sav_config
, config
,
4190 DATA_TYPE_NVLIST_ARRAY
) == 0);
4192 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
,
4193 config
, newdevs
, ndevs
+ oldndevs
) == 0);
4194 for (i
= 0; i
< oldndevs
+ ndevs
; i
++)
4195 nvlist_free(newdevs
[i
]);
4196 kmem_free(newdevs
, (oldndevs
+ ndevs
) * sizeof (void *));
4199 * Generate a new dev list.
4201 VERIFY(nvlist_alloc(&sav
->sav_config
, NV_UNIQUE_NAME
,
4203 VERIFY(nvlist_add_nvlist_array(sav
->sav_config
, config
,
4209 * Stop and drop level 2 ARC devices
4212 spa_l2cache_drop(spa_t
*spa
)
4216 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4218 for (i
= 0; i
< sav
->sav_count
; i
++) {
4221 vd
= sav
->sav_vdevs
[i
];
4224 if (spa_l2cache_exists(vd
->vdev_guid
, &pool
) &&
4225 pool
!= 0ULL && l2arc_vdev_present(vd
))
4226 l2arc_remove_vdev(vd
);
4234 spa_create(const char *pool
, nvlist_t
*nvroot
, nvlist_t
*props
,
4238 char *altroot
= NULL
;
4243 uint64_t txg
= TXG_INITIAL
;
4244 nvlist_t
**spares
, **l2cache
;
4245 uint_t nspares
, nl2cache
;
4246 uint64_t version
, obj
;
4247 boolean_t has_features
;
4250 * If this pool already exists, return failure.
4252 mutex_enter(&spa_namespace_lock
);
4253 if (spa_lookup(pool
) != NULL
) {
4254 mutex_exit(&spa_namespace_lock
);
4255 return (SET_ERROR(EEXIST
));
4259 * Allocate a new spa_t structure.
4261 (void) nvlist_lookup_string(props
,
4262 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4263 spa
= spa_add(pool
, NULL
, altroot
);
4264 spa_activate(spa
, spa_mode_global
);
4266 if (props
&& (error
= spa_prop_validate(spa
, props
))) {
4267 spa_deactivate(spa
);
4269 mutex_exit(&spa_namespace_lock
);
4273 has_features
= B_FALSE
;
4274 for (nvpair_t
*elem
= nvlist_next_nvpair(props
, NULL
);
4275 elem
!= NULL
; elem
= nvlist_next_nvpair(props
, elem
)) {
4276 if (zpool_prop_feature(nvpair_name(elem
)))
4277 has_features
= B_TRUE
;
4280 if (has_features
|| nvlist_lookup_uint64(props
,
4281 zpool_prop_to_name(ZPOOL_PROP_VERSION
), &version
) != 0) {
4282 version
= SPA_VERSION
;
4284 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
4286 spa
->spa_first_txg
= txg
;
4287 spa
->spa_uberblock
.ub_txg
= txg
- 1;
4288 spa
->spa_uberblock
.ub_version
= version
;
4289 spa
->spa_ubsync
= spa
->spa_uberblock
;
4290 spa
->spa_load_state
= SPA_LOAD_CREATE
;
4291 spa
->spa_removing_phys
.sr_state
= DSS_NONE
;
4292 spa
->spa_removing_phys
.sr_removing_vdev
= -1;
4293 spa
->spa_removing_phys
.sr_prev_indirect_vdev
= -1;
4296 * Create "The Godfather" zio to hold all async IOs
4298 spa
->spa_async_zio_root
= kmem_alloc(max_ncpus
* sizeof (void *),
4300 for (int i
= 0; i
< max_ncpus
; i
++) {
4301 spa
->spa_async_zio_root
[i
] = zio_root(spa
, NULL
, NULL
,
4302 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
4303 ZIO_FLAG_GODFATHER
);
4307 * Create the root vdev.
4309 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4311 error
= spa_config_parse(spa
, &rvd
, nvroot
, NULL
, 0, VDEV_ALLOC_ADD
);
4313 ASSERT(error
!= 0 || rvd
!= NULL
);
4314 ASSERT(error
!= 0 || spa
->spa_root_vdev
== rvd
);
4316 if (error
== 0 && !zfs_allocatable_devs(nvroot
))
4317 error
= SET_ERROR(EINVAL
);
4320 (error
= vdev_create(rvd
, txg
, B_FALSE
)) == 0 &&
4321 (error
= spa_validate_aux(spa
, nvroot
, txg
,
4322 VDEV_ALLOC_ADD
)) == 0) {
4323 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
4324 vdev_metaslab_set_size(rvd
->vdev_child
[c
]);
4325 vdev_expand(rvd
->vdev_child
[c
], txg
);
4329 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4333 spa_deactivate(spa
);
4335 mutex_exit(&spa_namespace_lock
);
4340 * Get the list of spares, if specified.
4342 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4343 &spares
, &nspares
) == 0) {
4344 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
, NV_UNIQUE_NAME
,
4346 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4347 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4348 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4349 spa_load_spares(spa
);
4350 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4351 spa
->spa_spares
.sav_sync
= B_TRUE
;
4355 * Get the list of level 2 cache devices, if specified.
4357 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4358 &l2cache
, &nl2cache
) == 0) {
4359 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4360 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4361 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4362 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4363 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4364 spa_load_l2cache(spa
);
4365 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4366 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4369 spa
->spa_is_initializing
= B_TRUE
;
4370 spa
->spa_dsl_pool
= dp
= dsl_pool_create(spa
, zplprops
, txg
);
4371 spa
->spa_meta_objset
= dp
->dp_meta_objset
;
4372 spa
->spa_is_initializing
= B_FALSE
;
4375 * Create DDTs (dedup tables).
4379 spa_update_dspace(spa
);
4381 tx
= dmu_tx_create_assigned(dp
, txg
);
4384 * Create the pool config object.
4386 spa
->spa_config_object
= dmu_object_alloc(spa
->spa_meta_objset
,
4387 DMU_OT_PACKED_NVLIST
, SPA_CONFIG_BLOCKSIZE
,
4388 DMU_OT_PACKED_NVLIST_SIZE
, sizeof (uint64_t), tx
);
4390 if (zap_add(spa
->spa_meta_objset
,
4391 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CONFIG
,
4392 sizeof (uint64_t), 1, &spa
->spa_config_object
, tx
) != 0) {
4393 cmn_err(CE_PANIC
, "failed to add pool config");
4396 if (spa_version(spa
) >= SPA_VERSION_FEATURES
)
4397 spa_feature_create_zap_objects(spa
, tx
);
4399 if (zap_add(spa
->spa_meta_objset
,
4400 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CREATION_VERSION
,
4401 sizeof (uint64_t), 1, &version
, tx
) != 0) {
4402 cmn_err(CE_PANIC
, "failed to add pool version");
4405 /* Newly created pools with the right version are always deflated. */
4406 if (version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
4407 spa
->spa_deflate
= TRUE
;
4408 if (zap_add(spa
->spa_meta_objset
,
4409 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
4410 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
) != 0) {
4411 cmn_err(CE_PANIC
, "failed to add deflate");
4416 * Create the deferred-free bpobj. Turn off compression
4417 * because sync-to-convergence takes longer if the blocksize
4420 obj
= bpobj_alloc(spa
->spa_meta_objset
, 1 << 14, tx
);
4421 dmu_object_set_compress(spa
->spa_meta_objset
, obj
,
4422 ZIO_COMPRESS_OFF
, tx
);
4423 if (zap_add(spa
->spa_meta_objset
,
4424 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_SYNC_BPOBJ
,
4425 sizeof (uint64_t), 1, &obj
, tx
) != 0) {
4426 cmn_err(CE_PANIC
, "failed to add bpobj");
4428 VERIFY3U(0, ==, bpobj_open(&spa
->spa_deferred_bpobj
,
4429 spa
->spa_meta_objset
, obj
));
4432 * Create the pool's history object.
4434 if (version
>= SPA_VERSION_ZPOOL_HISTORY
)
4435 spa_history_create_obj(spa
, tx
);
4438 * Generate some random noise for salted checksums to operate on.
4440 (void) random_get_pseudo_bytes(spa
->spa_cksum_salt
.zcs_bytes
,
4441 sizeof (spa
->spa_cksum_salt
.zcs_bytes
));
4444 * Set pool properties.
4446 spa
->spa_bootfs
= zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS
);
4447 spa
->spa_delegation
= zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION
);
4448 spa
->spa_failmode
= zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE
);
4449 spa
->spa_autoexpand
= zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND
);
4451 if (props
!= NULL
) {
4452 spa_configfile_set(spa
, props
, B_FALSE
);
4453 spa_sync_props(props
, tx
);
4458 spa
->spa_sync_on
= B_TRUE
;
4459 txg_sync_start(spa
->spa_dsl_pool
);
4462 * We explicitly wait for the first transaction to complete so that our
4463 * bean counters are appropriately updated.
4465 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
4467 spa_spawn_aux_threads(spa
);
4469 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
4470 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_CREATE
);
4472 spa_history_log_version(spa
, "create");
4475 * Don't count references from objsets that are already closed
4476 * and are making their way through the eviction process.
4478 spa_evicting_os_wait(spa
);
4479 spa
->spa_minref
= refcount_count(&spa
->spa_refcount
);
4480 spa
->spa_load_state
= SPA_LOAD_NONE
;
4482 mutex_exit(&spa_namespace_lock
);
4489 * Get the root pool information from the root disk, then import the root pool
4490 * during the system boot up time.
4492 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t
**);
4495 spa_generate_rootconf(char *devpath
, char *devid
, uint64_t *guid
)
4498 nvlist_t
*nvtop
, *nvroot
;
4501 if (vdev_disk_read_rootlabel(devpath
, devid
, &config
) != 0)
4505 * Add this top-level vdev to the child array.
4507 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4509 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
4511 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_GUID
, guid
) == 0);
4514 * Put this pool's top-level vdevs into a root vdev.
4516 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4517 VERIFY(nvlist_add_string(nvroot
, ZPOOL_CONFIG_TYPE
,
4518 VDEV_TYPE_ROOT
) == 0);
4519 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_ID
, 0ULL) == 0);
4520 VERIFY(nvlist_add_uint64(nvroot
, ZPOOL_CONFIG_GUID
, pgid
) == 0);
4521 VERIFY(nvlist_add_nvlist_array(nvroot
, ZPOOL_CONFIG_CHILDREN
,
4525 * Replace the existing vdev_tree with the new root vdev in
4526 * this pool's configuration (remove the old, add the new).
4528 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, nvroot
) == 0);
4529 nvlist_free(nvroot
);
4534 * Walk the vdev tree and see if we can find a device with "better"
4535 * configuration. A configuration is "better" if the label on that
4536 * device has a more recent txg.
4539 spa_alt_rootvdev(vdev_t
*vd
, vdev_t
**avd
, uint64_t *txg
)
4541 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4542 spa_alt_rootvdev(vd
->vdev_child
[c
], avd
, txg
);
4544 if (vd
->vdev_ops
->vdev_op_leaf
) {
4548 if (vdev_disk_read_rootlabel(vd
->vdev_physpath
, vd
->vdev_devid
,
4552 VERIFY(nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
4556 * Do we have a better boot device?
4558 if (label_txg
> *txg
) {
4567 * Import a root pool.
4569 * For x86. devpath_list will consist of devid and/or physpath name of
4570 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
4571 * The GRUB "findroot" command will return the vdev we should boot.
4573 * For Sparc, devpath_list consists the physpath name of the booting device
4574 * no matter the rootpool is a single device pool or a mirrored pool.
4576 * "/pci@1f,0/ide@d/disk@0,0:a"
4579 spa_import_rootpool(char *devpath
, char *devid
)
4582 vdev_t
*rvd
, *bvd
, *avd
= NULL
;
4583 nvlist_t
*config
, *nvtop
;
4589 * Read the label from the boot device and generate a configuration.
4591 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
4592 #if defined(_OBP) && defined(_KERNEL)
4593 if (config
== NULL
) {
4594 if (strstr(devpath
, "/iscsi/ssd") != NULL
) {
4596 get_iscsi_bootpath_phy(devpath
);
4597 config
= spa_generate_rootconf(devpath
, devid
, &guid
);
4601 if (config
== NULL
) {
4602 cmn_err(CE_NOTE
, "Cannot read the pool label from '%s'",
4604 return (SET_ERROR(EIO
));
4607 VERIFY(nvlist_lookup_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4609 VERIFY(nvlist_lookup_uint64(config
, ZPOOL_CONFIG_POOL_TXG
, &txg
) == 0);
4611 mutex_enter(&spa_namespace_lock
);
4612 if ((spa
= spa_lookup(pname
)) != NULL
) {
4614 * Remove the existing root pool from the namespace so that we
4615 * can replace it with the correct config we just read in.
4620 spa
= spa_add(pname
, config
, NULL
);
4621 spa
->spa_is_root
= B_TRUE
;
4622 spa
->spa_import_flags
= ZFS_IMPORT_VERBATIM
;
4623 if (nvlist_lookup_uint64(config
, ZPOOL_CONFIG_VERSION
,
4624 &spa
->spa_ubsync
.ub_version
) != 0)
4625 spa
->spa_ubsync
.ub_version
= SPA_VERSION_INITIAL
;
4628 * Build up a vdev tree based on the boot device's label config.
4630 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4632 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4633 error
= spa_config_parse(spa
, &rvd
, nvtop
, NULL
, 0,
4634 VDEV_ALLOC_ROOTPOOL
);
4635 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4637 mutex_exit(&spa_namespace_lock
);
4638 nvlist_free(config
);
4639 cmn_err(CE_NOTE
, "Can not parse the config for pool '%s'",
4645 * Get the boot vdev.
4647 if ((bvd
= vdev_lookup_by_guid(rvd
, guid
)) == NULL
) {
4648 cmn_err(CE_NOTE
, "Can not find the boot vdev for guid %llu",
4649 (u_longlong_t
)guid
);
4650 error
= SET_ERROR(ENOENT
);
4655 * Determine if there is a better boot device.
4658 spa_alt_rootvdev(rvd
, &avd
, &txg
);
4660 cmn_err(CE_NOTE
, "The boot device is 'degraded'. Please "
4661 "try booting from '%s'", avd
->vdev_path
);
4662 error
= SET_ERROR(EINVAL
);
4667 * If the boot device is part of a spare vdev then ensure that
4668 * we're booting off the active spare.
4670 if (bvd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4671 !bvd
->vdev_isspare
) {
4672 cmn_err(CE_NOTE
, "The boot device is currently spared. Please "
4673 "try booting from '%s'",
4675 vdev_child
[bvd
->vdev_parent
->vdev_children
- 1]->vdev_path
);
4676 error
= SET_ERROR(EINVAL
);
4682 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4684 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4685 mutex_exit(&spa_namespace_lock
);
4687 nvlist_free(config
);
4694 * Import a non-root pool into the system.
4697 spa_import(const char *pool
, nvlist_t
*config
, nvlist_t
*props
, uint64_t flags
)
4700 char *altroot
= NULL
;
4701 spa_load_state_t state
= SPA_LOAD_IMPORT
;
4702 zpool_rewind_policy_t policy
;
4703 uint64_t mode
= spa_mode_global
;
4704 uint64_t readonly
= B_FALSE
;
4707 nvlist_t
**spares
, **l2cache
;
4708 uint_t nspares
, nl2cache
;
4711 * If a pool with this name exists, return failure.
4713 mutex_enter(&spa_namespace_lock
);
4714 if (spa_lookup(pool
) != NULL
) {
4715 mutex_exit(&spa_namespace_lock
);
4716 return (SET_ERROR(EEXIST
));
4720 * Create and initialize the spa structure.
4722 (void) nvlist_lookup_string(props
,
4723 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
4724 (void) nvlist_lookup_uint64(props
,
4725 zpool_prop_to_name(ZPOOL_PROP_READONLY
), &readonly
);
4728 spa
= spa_add(pool
, config
, altroot
);
4729 spa
->spa_import_flags
= flags
;
4732 * Verbatim import - Take a pool and insert it into the namespace
4733 * as if it had been loaded at boot.
4735 if (spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) {
4737 spa_configfile_set(spa
, props
, B_FALSE
);
4739 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
4740 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4741 zfs_dbgmsg("spa_import: verbatim import of %s", pool
);
4742 mutex_exit(&spa_namespace_lock
);
4746 spa_activate(spa
, mode
);
4749 * Don't start async tasks until we know everything is healthy.
4751 spa_async_suspend(spa
);
4753 zpool_get_rewind_policy(config
, &policy
);
4754 if (policy
.zrp_request
& ZPOOL_DO_REWIND
)
4755 state
= SPA_LOAD_RECOVER
;
4757 spa
->spa_config_source
= SPA_CONFIG_SRC_TRYIMPORT
;
4759 if (state
!= SPA_LOAD_RECOVER
) {
4760 spa
->spa_last_ubsync_txg
= spa
->spa_load_txg
= 0;
4761 zfs_dbgmsg("spa_import: importing %s", pool
);
4763 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
4764 "(RECOVERY MODE)", pool
, (longlong_t
)policy
.zrp_txg
);
4766 error
= spa_load_best(spa
, state
, policy
.zrp_txg
, policy
.zrp_request
);
4769 * Propagate anything learned while loading the pool and pass it
4770 * back to caller (i.e. rewind info, missing devices, etc).
4772 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4773 spa
->spa_load_info
) == 0);
4775 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4777 * Toss any existing sparelist, as it doesn't have any validity
4778 * anymore, and conflicts with spa_has_spare().
4780 if (spa
->spa_spares
.sav_config
) {
4781 nvlist_free(spa
->spa_spares
.sav_config
);
4782 spa
->spa_spares
.sav_config
= NULL
;
4783 spa_load_spares(spa
);
4785 if (spa
->spa_l2cache
.sav_config
) {
4786 nvlist_free(spa
->spa_l2cache
.sav_config
);
4787 spa
->spa_l2cache
.sav_config
= NULL
;
4788 spa_load_l2cache(spa
);
4791 VERIFY(nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
,
4794 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4797 error
= spa_validate_aux(spa
, nvroot
, -1ULL,
4798 VDEV_ALLOC_L2CACHE
);
4799 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4802 spa_configfile_set(spa
, props
, B_FALSE
);
4804 if (error
!= 0 || (props
&& spa_writeable(spa
) &&
4805 (error
= spa_prop_set(spa
, props
)))) {
4807 spa_deactivate(spa
);
4809 mutex_exit(&spa_namespace_lock
);
4813 spa_async_resume(spa
);
4816 * Override any spares and level 2 cache devices as specified by
4817 * the user, as these may have correct device names/devids, etc.
4819 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
,
4820 &spares
, &nspares
) == 0) {
4821 if (spa
->spa_spares
.sav_config
)
4822 VERIFY(nvlist_remove(spa
->spa_spares
.sav_config
,
4823 ZPOOL_CONFIG_SPARES
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4825 VERIFY(nvlist_alloc(&spa
->spa_spares
.sav_config
,
4826 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4827 VERIFY(nvlist_add_nvlist_array(spa
->spa_spares
.sav_config
,
4828 ZPOOL_CONFIG_SPARES
, spares
, nspares
) == 0);
4829 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4830 spa_load_spares(spa
);
4831 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4832 spa
->spa_spares
.sav_sync
= B_TRUE
;
4834 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
,
4835 &l2cache
, &nl2cache
) == 0) {
4836 if (spa
->spa_l2cache
.sav_config
)
4837 VERIFY(nvlist_remove(spa
->spa_l2cache
.sav_config
,
4838 ZPOOL_CONFIG_L2CACHE
, DATA_TYPE_NVLIST_ARRAY
) == 0);
4840 VERIFY(nvlist_alloc(&spa
->spa_l2cache
.sav_config
,
4841 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
4842 VERIFY(nvlist_add_nvlist_array(spa
->spa_l2cache
.sav_config
,
4843 ZPOOL_CONFIG_L2CACHE
, l2cache
, nl2cache
) == 0);
4844 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
4845 spa_load_l2cache(spa
);
4846 spa_config_exit(spa
, SCL_ALL
, FTAG
);
4847 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
4851 * Check for any removed devices.
4853 if (spa
->spa_autoreplace
) {
4854 spa_aux_check_removed(&spa
->spa_spares
);
4855 spa_aux_check_removed(&spa
->spa_l2cache
);
4858 if (spa_writeable(spa
)) {
4860 * Update the config cache to include the newly-imported pool.
4862 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
4866 * It's possible that the pool was expanded while it was exported.
4867 * We kick off an async task to handle this for us.
4869 spa_async_request(spa
, SPA_ASYNC_AUTOEXPAND
);
4871 spa_history_log_version(spa
, "import");
4873 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_IMPORT
);
4875 mutex_exit(&spa_namespace_lock
);
4881 spa_tryimport(nvlist_t
*tryconfig
)
4883 nvlist_t
*config
= NULL
;
4884 char *poolname
, *cachefile
;
4888 zpool_rewind_policy_t policy
;
4890 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_POOL_NAME
, &poolname
))
4893 if (nvlist_lookup_uint64(tryconfig
, ZPOOL_CONFIG_POOL_STATE
, &state
))
4897 * Create and initialize the spa structure.
4899 mutex_enter(&spa_namespace_lock
);
4900 spa
= spa_add(TRYIMPORT_NAME
, tryconfig
, NULL
);
4901 spa_activate(spa
, FREAD
);
4904 * Rewind pool if a max txg was provided. Note that even though we
4905 * retrieve the complete rewind policy, only the rewind txg is relevant
4908 zpool_get_rewind_policy(spa
->spa_config
, &policy
);
4909 if (policy
.zrp_txg
!= UINT64_MAX
) {
4910 spa
->spa_load_max_txg
= policy
.zrp_txg
;
4911 spa
->spa_extreme_rewind
= B_TRUE
;
4912 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
4913 poolname
, (longlong_t
)policy
.zrp_txg
);
4915 zfs_dbgmsg("spa_tryimport: importing %s", poolname
);
4918 if (nvlist_lookup_string(tryconfig
, ZPOOL_CONFIG_CACHEFILE
, &cachefile
)
4920 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile
);
4921 spa
->spa_config_source
= SPA_CONFIG_SRC_CACHEFILE
;
4923 spa
->spa_config_source
= SPA_CONFIG_SRC_SCAN
;
4926 error
= spa_load(spa
, SPA_LOAD_TRYIMPORT
, SPA_IMPORT_EXISTING
);
4929 * If 'tryconfig' was at least parsable, return the current config.
4931 if (spa
->spa_root_vdev
!= NULL
) {
4932 config
= spa_config_generate(spa
, NULL
, -1ULL, B_TRUE
);
4933 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
,
4935 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
4937 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_TIMESTAMP
,
4938 spa
->spa_uberblock
.ub_timestamp
) == 0);
4939 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_LOAD_INFO
,
4940 spa
->spa_load_info
) == 0);
4943 * If the bootfs property exists on this pool then we
4944 * copy it out so that external consumers can tell which
4945 * pools are bootable.
4947 if ((!error
|| error
== EEXIST
) && spa
->spa_bootfs
) {
4948 char *tmpname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4951 * We have to play games with the name since the
4952 * pool was opened as TRYIMPORT_NAME.
4954 if (dsl_dsobj_to_dsname(spa_name(spa
),
4955 spa
->spa_bootfs
, tmpname
) == 0) {
4957 char *dsname
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
4959 cp
= strchr(tmpname
, '/');
4961 (void) strlcpy(dsname
, tmpname
,
4964 (void) snprintf(dsname
, MAXPATHLEN
,
4965 "%s/%s", poolname
, ++cp
);
4967 VERIFY(nvlist_add_string(config
,
4968 ZPOOL_CONFIG_BOOTFS
, dsname
) == 0);
4969 kmem_free(dsname
, MAXPATHLEN
);
4971 kmem_free(tmpname
, MAXPATHLEN
);
4975 * Add the list of hot spares and level 2 cache devices.
4977 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
4978 spa_add_spares(spa
, config
);
4979 spa_add_l2cache(spa
, config
);
4980 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
4984 spa_deactivate(spa
);
4986 mutex_exit(&spa_namespace_lock
);
4992 * Pool export/destroy
4994 * The act of destroying or exporting a pool is very simple. We make sure there
4995 * is no more pending I/O and any references to the pool are gone. Then, we
4996 * update the pool state and sync all the labels to disk, removing the
4997 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4998 * we don't sync the labels or remove the configuration cache.
5001 spa_export_common(char *pool
, int new_state
, nvlist_t
**oldconfig
,
5002 boolean_t force
, boolean_t hardforce
)
5009 if (!(spa_mode_global
& FWRITE
))
5010 return (SET_ERROR(EROFS
));
5012 mutex_enter(&spa_namespace_lock
);
5013 if ((spa
= spa_lookup(pool
)) == NULL
) {
5014 mutex_exit(&spa_namespace_lock
);
5015 return (SET_ERROR(ENOENT
));
5019 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5020 * reacquire the namespace lock, and see if we can export.
5022 spa_open_ref(spa
, FTAG
);
5023 mutex_exit(&spa_namespace_lock
);
5024 spa_async_suspend(spa
);
5025 mutex_enter(&spa_namespace_lock
);
5026 spa_close(spa
, FTAG
);
5029 * The pool will be in core if it's openable,
5030 * in which case we can modify its state.
5032 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
&& spa
->spa_sync_on
) {
5034 * Objsets may be open only because they're dirty, so we
5035 * have to force it to sync before checking spa_refcnt.
5037 txg_wait_synced(spa
->spa_dsl_pool
, 0);
5038 spa_evicting_os_wait(spa
);
5041 * A pool cannot be exported or destroyed if there are active
5042 * references. If we are resetting a pool, allow references by
5043 * fault injection handlers.
5045 if (!spa_refcount_zero(spa
) ||
5046 (spa
->spa_inject_ref
!= 0 &&
5047 new_state
!= POOL_STATE_UNINITIALIZED
)) {
5048 spa_async_resume(spa
);
5049 mutex_exit(&spa_namespace_lock
);
5050 return (SET_ERROR(EBUSY
));
5054 * A pool cannot be exported if it has an active shared spare.
5055 * This is to prevent other pools stealing the active spare
5056 * from an exported pool. At user's own will, such pool can
5057 * be forcedly exported.
5059 if (!force
&& new_state
== POOL_STATE_EXPORTED
&&
5060 spa_has_active_shared_spare(spa
)) {
5061 spa_async_resume(spa
);
5062 mutex_exit(&spa_namespace_lock
);
5063 return (SET_ERROR(EXDEV
));
5067 * We want this to be reflected on every label,
5068 * so mark them all dirty. spa_unload() will do the
5069 * final sync that pushes these changes out.
5071 if (new_state
!= POOL_STATE_UNINITIALIZED
&& !hardforce
) {
5072 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
5073 spa
->spa_state
= new_state
;
5074 spa
->spa_final_txg
= spa_last_synced_txg(spa
) +
5076 vdev_config_dirty(spa
->spa_root_vdev
);
5077 spa_config_exit(spa
, SCL_ALL
, FTAG
);
5081 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_POOL_DESTROY
);
5083 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
5085 spa_deactivate(spa
);
5088 if (oldconfig
&& spa
->spa_config
)
5089 VERIFY(nvlist_dup(spa
->spa_config
, oldconfig
, 0) == 0);
5091 if (new_state
!= POOL_STATE_UNINITIALIZED
) {
5093 spa_write_cachefile(spa
, B_TRUE
, B_TRUE
);
5096 mutex_exit(&spa_namespace_lock
);
5102 * Destroy a storage pool.
5105 spa_destroy(char *pool
)
5107 return (spa_export_common(pool
, POOL_STATE_DESTROYED
, NULL
,
5112 * Export a storage pool.
5115 spa_export(char *pool
, nvlist_t
**oldconfig
, boolean_t force
,
5116 boolean_t hardforce
)
5118 return (spa_export_common(pool
, POOL_STATE_EXPORTED
, oldconfig
,
5123 * Similar to spa_export(), this unloads the spa_t without actually removing it
5124 * from the namespace in any way.
5127 spa_reset(char *pool
)
5129 return (spa_export_common(pool
, POOL_STATE_UNINITIALIZED
, NULL
,
5134 * ==========================================================================
5135 * Device manipulation
5136 * ==========================================================================
5140 * Add a device to a storage pool.
5143 spa_vdev_add(spa_t
*spa
, nvlist_t
*nvroot
)
5147 vdev_t
*rvd
= spa
->spa_root_vdev
;
5149 nvlist_t
**spares
, **l2cache
;
5150 uint_t nspares
, nl2cache
;
5152 ASSERT(spa_writeable(spa
));
5154 txg
= spa_vdev_enter(spa
);
5156 if ((error
= spa_config_parse(spa
, &vd
, nvroot
, NULL
, 0,
5157 VDEV_ALLOC_ADD
)) != 0)
5158 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5160 spa
->spa_pending_vdev
= vd
; /* spa_vdev_exit() will clear this */
5162 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_SPARES
, &spares
,
5166 if (nvlist_lookup_nvlist_array(nvroot
, ZPOOL_CONFIG_L2CACHE
, &l2cache
,
5170 if (vd
->vdev_children
== 0 && nspares
== 0 && nl2cache
== 0)
5171 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5173 if (vd
->vdev_children
!= 0 &&
5174 (error
= vdev_create(vd
, txg
, B_FALSE
)) != 0)
5175 return (spa_vdev_exit(spa
, vd
, txg
, error
));
5178 * We must validate the spares and l2cache devices after checking the
5179 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5181 if ((error
= spa_validate_aux(spa
, nvroot
, txg
, VDEV_ALLOC_ADD
)) != 0)
5182 return (spa_vdev_exit(spa
, vd
, txg
, error
));
5185 * If we are in the middle of a device removal, we can only add
5186 * devices which match the existing devices in the pool.
5187 * If we are in the middle of a removal, or have some indirect
5188 * vdevs, we can not add raidz toplevels.
5190 if (spa
->spa_vdev_removal
!= NULL
||
5191 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
5192 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5193 tvd
= vd
->vdev_child
[c
];
5194 if (spa
->spa_vdev_removal
!= NULL
&&
5196 spa
->spa_vdev_removal
->svr_vdev
->vdev_ashift
) {
5197 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5199 /* Fail if top level vdev is raidz */
5200 if (tvd
->vdev_ops
== &vdev_raidz_ops
) {
5201 return (spa_vdev_exit(spa
, vd
, txg
, EINVAL
));
5204 * Need the top level mirror to be
5205 * a mirror of leaf vdevs only
5207 if (tvd
->vdev_ops
== &vdev_mirror_ops
) {
5208 for (uint64_t cid
= 0;
5209 cid
< tvd
->vdev_children
; cid
++) {
5210 vdev_t
*cvd
= tvd
->vdev_child
[cid
];
5211 if (!cvd
->vdev_ops
->vdev_op_leaf
) {
5212 return (spa_vdev_exit(spa
, vd
,
5220 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5223 * Set the vdev id to the first hole, if one exists.
5225 for (id
= 0; id
< rvd
->vdev_children
; id
++) {
5226 if (rvd
->vdev_child
[id
]->vdev_ishole
) {
5227 vdev_free(rvd
->vdev_child
[id
]);
5231 tvd
= vd
->vdev_child
[c
];
5232 vdev_remove_child(vd
, tvd
);
5234 vdev_add_child(rvd
, tvd
);
5235 vdev_config_dirty(tvd
);
5239 spa_set_aux_vdevs(&spa
->spa_spares
, spares
, nspares
,
5240 ZPOOL_CONFIG_SPARES
);
5241 spa_load_spares(spa
);
5242 spa
->spa_spares
.sav_sync
= B_TRUE
;
5245 if (nl2cache
!= 0) {
5246 spa_set_aux_vdevs(&spa
->spa_l2cache
, l2cache
, nl2cache
,
5247 ZPOOL_CONFIG_L2CACHE
);
5248 spa_load_l2cache(spa
);
5249 spa
->spa_l2cache
.sav_sync
= B_TRUE
;
5253 * We have to be careful when adding new vdevs to an existing pool.
5254 * If other threads start allocating from these vdevs before we
5255 * sync the config cache, and we lose power, then upon reboot we may
5256 * fail to open the pool because there are DVAs that the config cache
5257 * can't translate. Therefore, we first add the vdevs without
5258 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5259 * and then let spa_config_update() initialize the new metaslabs.
5261 * spa_load() checks for added-but-not-initialized vdevs, so that
5262 * if we lose power at any point in this sequence, the remaining
5263 * steps will be completed the next time we load the pool.
5265 (void) spa_vdev_exit(spa
, vd
, txg
, 0);
5267 mutex_enter(&spa_namespace_lock
);
5268 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
5269 spa_event_notify(spa
, NULL
, NULL
, ESC_ZFS_VDEV_ADD
);
5270 mutex_exit(&spa_namespace_lock
);
5276 * Attach a device to a mirror. The arguments are the path to any device
5277 * in the mirror, and the nvroot for the new device. If the path specifies
5278 * a device that is not mirrored, we automatically insert the mirror vdev.
5280 * If 'replacing' is specified, the new device is intended to replace the
5281 * existing device; in this case the two devices are made into their own
5282 * mirror using the 'replacing' vdev, which is functionally identical to
5283 * the mirror vdev (it actually reuses all the same ops) but has a few
5284 * extra rules: you can't attach to it after it's been created, and upon
5285 * completion of resilvering, the first disk (the one being replaced)
5286 * is automatically detached.
5289 spa_vdev_attach(spa_t
*spa
, uint64_t guid
, nvlist_t
*nvroot
, int replacing
)
5291 uint64_t txg
, dtl_max_txg
;
5292 vdev_t
*rvd
= spa
->spa_root_vdev
;
5293 vdev_t
*oldvd
, *newvd
, *newrootvd
, *pvd
, *tvd
;
5295 char *oldvdpath
, *newvdpath
;
5299 ASSERT(spa_writeable(spa
));
5301 txg
= spa_vdev_enter(spa
);
5303 oldvd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5305 if (spa
->spa_vdev_removal
!= NULL
||
5306 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
5307 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5311 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5313 if (!oldvd
->vdev_ops
->vdev_op_leaf
)
5314 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5316 pvd
= oldvd
->vdev_parent
;
5318 if ((error
= spa_config_parse(spa
, &newrootvd
, nvroot
, NULL
, 0,
5319 VDEV_ALLOC_ATTACH
)) != 0)
5320 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5322 if (newrootvd
->vdev_children
!= 1)
5323 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
5325 newvd
= newrootvd
->vdev_child
[0];
5327 if (!newvd
->vdev_ops
->vdev_op_leaf
)
5328 return (spa_vdev_exit(spa
, newrootvd
, txg
, EINVAL
));
5330 if ((error
= vdev_create(newrootvd
, txg
, replacing
)) != 0)
5331 return (spa_vdev_exit(spa
, newrootvd
, txg
, error
));
5334 * Spares can't replace logs
5336 if (oldvd
->vdev_top
->vdev_islog
&& newvd
->vdev_isspare
)
5337 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5341 * For attach, the only allowable parent is a mirror or the root
5344 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5345 pvd
->vdev_ops
!= &vdev_root_ops
)
5346 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5348 pvops
= &vdev_mirror_ops
;
5351 * Active hot spares can only be replaced by inactive hot
5354 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5355 oldvd
->vdev_isspare
&&
5356 !spa_has_spare(spa
, newvd
->vdev_guid
))
5357 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5360 * If the source is a hot spare, and the parent isn't already a
5361 * spare, then we want to create a new hot spare. Otherwise, we
5362 * want to create a replacing vdev. The user is not allowed to
5363 * attach to a spared vdev child unless the 'isspare' state is
5364 * the same (spare replaces spare, non-spare replaces
5367 if (pvd
->vdev_ops
== &vdev_replacing_ops
&&
5368 spa_version(spa
) < SPA_VERSION_MULTI_REPLACE
) {
5369 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5370 } else if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5371 newvd
->vdev_isspare
!= oldvd
->vdev_isspare
) {
5372 return (spa_vdev_exit(spa
, newrootvd
, txg
, ENOTSUP
));
5375 if (newvd
->vdev_isspare
)
5376 pvops
= &vdev_spare_ops
;
5378 pvops
= &vdev_replacing_ops
;
5382 * Make sure the new device is big enough.
5384 if (newvd
->vdev_asize
< vdev_get_min_asize(oldvd
))
5385 return (spa_vdev_exit(spa
, newrootvd
, txg
, EOVERFLOW
));
5388 * The new device cannot have a higher alignment requirement
5389 * than the top-level vdev.
5391 if (newvd
->vdev_ashift
> oldvd
->vdev_top
->vdev_ashift
)
5392 return (spa_vdev_exit(spa
, newrootvd
, txg
, EDOM
));
5395 * If this is an in-place replacement, update oldvd's path and devid
5396 * to make it distinguishable from newvd, and unopenable from now on.
5398 if (strcmp(oldvd
->vdev_path
, newvd
->vdev_path
) == 0) {
5399 spa_strfree(oldvd
->vdev_path
);
5400 oldvd
->vdev_path
= kmem_alloc(strlen(newvd
->vdev_path
) + 5,
5402 (void) sprintf(oldvd
->vdev_path
, "%s/%s",
5403 newvd
->vdev_path
, "old");
5404 if (oldvd
->vdev_devid
!= NULL
) {
5405 spa_strfree(oldvd
->vdev_devid
);
5406 oldvd
->vdev_devid
= NULL
;
5410 /* mark the device being resilvered */
5411 newvd
->vdev_resilver_txg
= txg
;
5414 * If the parent is not a mirror, or if we're replacing, insert the new
5415 * mirror/replacing/spare vdev above oldvd.
5417 if (pvd
->vdev_ops
!= pvops
)
5418 pvd
= vdev_add_parent(oldvd
, pvops
);
5420 ASSERT(pvd
->vdev_top
->vdev_parent
== rvd
);
5421 ASSERT(pvd
->vdev_ops
== pvops
);
5422 ASSERT(oldvd
->vdev_parent
== pvd
);
5425 * Extract the new device from its root and add it to pvd.
5427 vdev_remove_child(newrootvd
, newvd
);
5428 newvd
->vdev_id
= pvd
->vdev_children
;
5429 newvd
->vdev_crtxg
= oldvd
->vdev_crtxg
;
5430 vdev_add_child(pvd
, newvd
);
5432 tvd
= newvd
->vdev_top
;
5433 ASSERT(pvd
->vdev_top
== tvd
);
5434 ASSERT(tvd
->vdev_parent
== rvd
);
5436 vdev_config_dirty(tvd
);
5439 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5440 * for any dmu_sync-ed blocks. It will propagate upward when
5441 * spa_vdev_exit() calls vdev_dtl_reassess().
5443 dtl_max_txg
= txg
+ TXG_CONCURRENT_STATES
;
5445 vdev_dtl_dirty(newvd
, DTL_MISSING
, TXG_INITIAL
,
5446 dtl_max_txg
- TXG_INITIAL
);
5448 if (newvd
->vdev_isspare
) {
5449 spa_spare_activate(newvd
);
5450 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_SPARE
);
5453 oldvdpath
= spa_strdup(oldvd
->vdev_path
);
5454 newvdpath
= spa_strdup(newvd
->vdev_path
);
5455 newvd_isspare
= newvd
->vdev_isspare
;
5458 * Mark newvd's DTL dirty in this txg.
5460 vdev_dirty(tvd
, VDD_DTL
, newvd
, txg
);
5463 * Schedule the resilver to restart in the future. We do this to
5464 * ensure that dmu_sync-ed blocks have been stitched into the
5465 * respective datasets.
5467 dsl_resilver_restart(spa
->spa_dsl_pool
, dtl_max_txg
);
5469 if (spa
->spa_bootfs
)
5470 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_BOOTFS_VDEV_ATTACH
);
5472 spa_event_notify(spa
, newvd
, NULL
, ESC_ZFS_VDEV_ATTACH
);
5477 (void) spa_vdev_exit(spa
, newrootvd
, dtl_max_txg
, 0);
5479 spa_history_log_internal(spa
, "vdev attach", NULL
,
5480 "%s vdev=%s %s vdev=%s",
5481 replacing
&& newvd_isspare
? "spare in" :
5482 replacing
? "replace" : "attach", newvdpath
,
5483 replacing
? "for" : "to", oldvdpath
);
5485 spa_strfree(oldvdpath
);
5486 spa_strfree(newvdpath
);
5492 * Detach a device from a mirror or replacing vdev.
5494 * If 'replace_done' is specified, only detach if the parent
5495 * is a replacing vdev.
5498 spa_vdev_detach(spa_t
*spa
, uint64_t guid
, uint64_t pguid
, int replace_done
)
5502 vdev_t
*rvd
= spa
->spa_root_vdev
;
5503 vdev_t
*vd
, *pvd
, *cvd
, *tvd
;
5504 boolean_t unspare
= B_FALSE
;
5505 uint64_t unspare_guid
= 0;
5508 ASSERT(spa_writeable(spa
));
5510 txg
= spa_vdev_enter(spa
);
5512 vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
5515 return (spa_vdev_exit(spa
, NULL
, txg
, ENODEV
));
5517 if (!vd
->vdev_ops
->vdev_op_leaf
)
5518 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5520 pvd
= vd
->vdev_parent
;
5523 * If the parent/child relationship is not as expected, don't do it.
5524 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5525 * vdev that's replacing B with C. The user's intent in replacing
5526 * is to go from M(A,B) to M(A,C). If the user decides to cancel
5527 * the replace by detaching C, the expected behavior is to end up
5528 * M(A,B). But suppose that right after deciding to detach C,
5529 * the replacement of B completes. We would have M(A,C), and then
5530 * ask to detach C, which would leave us with just A -- not what
5531 * the user wanted. To prevent this, we make sure that the
5532 * parent/child relationship hasn't changed -- in this example,
5533 * that C's parent is still the replacing vdev R.
5535 if (pvd
->vdev_guid
!= pguid
&& pguid
!= 0)
5536 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5539 * Only 'replacing' or 'spare' vdevs can be replaced.
5541 if (replace_done
&& pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5542 pvd
->vdev_ops
!= &vdev_spare_ops
)
5543 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5545 ASSERT(pvd
->vdev_ops
!= &vdev_spare_ops
||
5546 spa_version(spa
) >= SPA_VERSION_SPARES
);
5549 * Only mirror, replacing, and spare vdevs support detach.
5551 if (pvd
->vdev_ops
!= &vdev_replacing_ops
&&
5552 pvd
->vdev_ops
!= &vdev_mirror_ops
&&
5553 pvd
->vdev_ops
!= &vdev_spare_ops
)
5554 return (spa_vdev_exit(spa
, NULL
, txg
, ENOTSUP
));
5557 * If this device has the only valid copy of some data,
5558 * we cannot safely detach it.
5560 if (vdev_dtl_required(vd
))
5561 return (spa_vdev_exit(spa
, NULL
, txg
, EBUSY
));
5563 ASSERT(pvd
->vdev_children
>= 2);
5566 * If we are detaching the second disk from a replacing vdev, then
5567 * check to see if we changed the original vdev's path to have "/old"
5568 * at the end in spa_vdev_attach(). If so, undo that change now.
5570 if (pvd
->vdev_ops
== &vdev_replacing_ops
&& vd
->vdev_id
> 0 &&
5571 vd
->vdev_path
!= NULL
) {
5572 size_t len
= strlen(vd
->vdev_path
);
5574 for (int c
= 0; c
< pvd
->vdev_children
; c
++) {
5575 cvd
= pvd
->vdev_child
[c
];
5577 if (cvd
== vd
|| cvd
->vdev_path
== NULL
)
5580 if (strncmp(cvd
->vdev_path
, vd
->vdev_path
, len
) == 0 &&
5581 strcmp(cvd
->vdev_path
+ len
, "/old") == 0) {
5582 spa_strfree(cvd
->vdev_path
);
5583 cvd
->vdev_path
= spa_strdup(vd
->vdev_path
);
5590 * If we are detaching the original disk from a spare, then it implies
5591 * that the spare should become a real disk, and be removed from the
5592 * active spare list for the pool.
5594 if (pvd
->vdev_ops
== &vdev_spare_ops
&&
5596 pvd
->vdev_child
[pvd
->vdev_children
- 1]->vdev_isspare
)
5600 * Erase the disk labels so the disk can be used for other things.
5601 * This must be done after all other error cases are handled,
5602 * but before we disembowel vd (so we can still do I/O to it).
5603 * But if we can't do it, don't treat the error as fatal --
5604 * it may be that the unwritability of the disk is the reason
5605 * it's being detached!
5607 error
= vdev_label_init(vd
, 0, VDEV_LABEL_REMOVE
);
5610 * Remove vd from its parent and compact the parent's children.
5612 vdev_remove_child(pvd
, vd
);
5613 vdev_compact_children(pvd
);
5616 * Remember one of the remaining children so we can get tvd below.
5618 cvd
= pvd
->vdev_child
[pvd
->vdev_children
- 1];
5621 * If we need to remove the remaining child from the list of hot spares,
5622 * do it now, marking the vdev as no longer a spare in the process.
5623 * We must do this before vdev_remove_parent(), because that can
5624 * change the GUID if it creates a new toplevel GUID. For a similar
5625 * reason, we must remove the spare now, in the same txg as the detach;
5626 * otherwise someone could attach a new sibling, change the GUID, and
5627 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
5630 ASSERT(cvd
->vdev_isspare
);
5631 spa_spare_remove(cvd
);
5632 unspare_guid
= cvd
->vdev_guid
;
5633 (void) spa_vdev_remove(spa
, unspare_guid
, B_TRUE
);
5634 cvd
->vdev_unspare
= B_TRUE
;
5638 * If the parent mirror/replacing vdev only has one child,
5639 * the parent is no longer needed. Remove it from the tree.
5641 if (pvd
->vdev_children
== 1) {
5642 if (pvd
->vdev_ops
== &vdev_spare_ops
)
5643 cvd
->vdev_unspare
= B_FALSE
;
5644 vdev_remove_parent(cvd
);
5649 * We don't set tvd until now because the parent we just removed
5650 * may have been the previous top-level vdev.
5652 tvd
= cvd
->vdev_top
;
5653 ASSERT(tvd
->vdev_parent
== rvd
);
5656 * Reevaluate the parent vdev state.
5658 vdev_propagate_state(cvd
);
5661 * If the 'autoexpand' property is set on the pool then automatically
5662 * try to expand the size of the pool. For example if the device we
5663 * just detached was smaller than the others, it may be possible to
5664 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
5665 * first so that we can obtain the updated sizes of the leaf vdevs.
5667 if (spa
->spa_autoexpand
) {
5669 vdev_expand(tvd
, txg
);
5672 vdev_config_dirty(tvd
);
5675 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5676 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5677 * But first make sure we're not on any *other* txg's DTL list, to
5678 * prevent vd from being accessed after it's freed.
5680 vdpath
= spa_strdup(vd
->vdev_path
);
5681 for (int t
= 0; t
< TXG_SIZE
; t
++)
5682 (void) txg_list_remove_this(&tvd
->vdev_dtl_list
, vd
, t
);
5683 vd
->vdev_detached
= B_TRUE
;
5684 vdev_dirty(tvd
, VDD_DTL
, vd
, txg
);
5686 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_REMOVE
);
5688 /* hang on to the spa before we release the lock */
5689 spa_open_ref(spa
, FTAG
);
5691 error
= spa_vdev_exit(spa
, vd
, txg
, 0);
5693 spa_history_log_internal(spa
, "detach", NULL
,
5695 spa_strfree(vdpath
);
5698 * If this was the removal of the original device in a hot spare vdev,
5699 * then we want to go through and remove the device from the hot spare
5700 * list of every other pool.
5703 spa_t
*altspa
= NULL
;
5705 mutex_enter(&spa_namespace_lock
);
5706 while ((altspa
= spa_next(altspa
)) != NULL
) {
5707 if (altspa
->spa_state
!= POOL_STATE_ACTIVE
||
5711 spa_open_ref(altspa
, FTAG
);
5712 mutex_exit(&spa_namespace_lock
);
5713 (void) spa_vdev_remove(altspa
, unspare_guid
, B_TRUE
);
5714 mutex_enter(&spa_namespace_lock
);
5715 spa_close(altspa
, FTAG
);
5717 mutex_exit(&spa_namespace_lock
);
5719 /* search the rest of the vdevs for spares to remove */
5720 spa_vdev_resilver_done(spa
);
5723 /* all done with the spa; OK to release */
5724 mutex_enter(&spa_namespace_lock
);
5725 spa_close(spa
, FTAG
);
5726 mutex_exit(&spa_namespace_lock
);
5732 * Split a set of devices from their mirrors, and create a new pool from them.
5735 spa_vdev_split_mirror(spa_t
*spa
, char *newname
, nvlist_t
*config
,
5736 nvlist_t
*props
, boolean_t exp
)
5739 uint64_t txg
, *glist
;
5741 uint_t c
, children
, lastlog
;
5742 nvlist_t
**child
, *nvl
, *tmp
;
5744 char *altroot
= NULL
;
5745 vdev_t
*rvd
, **vml
= NULL
; /* vdev modify list */
5746 boolean_t activate_slog
;
5748 ASSERT(spa_writeable(spa
));
5750 txg
= spa_vdev_enter(spa
);
5752 /* clear the log and flush everything up to now */
5753 activate_slog
= spa_passivate_log(spa
);
5754 (void) spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5755 error
= spa_reset_logs(spa
);
5756 txg
= spa_vdev_config_enter(spa
);
5759 spa_activate_log(spa
);
5762 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5764 /* check new spa name before going any further */
5765 if (spa_lookup(newname
) != NULL
)
5766 return (spa_vdev_exit(spa
, NULL
, txg
, EEXIST
));
5769 * scan through all the children to ensure they're all mirrors
5771 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
) != 0 ||
5772 nvlist_lookup_nvlist_array(nvl
, ZPOOL_CONFIG_CHILDREN
, &child
,
5774 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5776 /* first, check to ensure we've got the right child count */
5777 rvd
= spa
->spa_root_vdev
;
5779 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
5780 vdev_t
*vd
= rvd
->vdev_child
[c
];
5782 /* don't count the holes & logs as children */
5783 if (vd
->vdev_islog
|| !vdev_is_concrete(vd
)) {
5791 if (children
!= (lastlog
!= 0 ? lastlog
: rvd
->vdev_children
))
5792 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5794 /* next, ensure no spare or cache devices are part of the split */
5795 if (nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_SPARES
, &tmp
) == 0 ||
5796 nvlist_lookup_nvlist(nvl
, ZPOOL_CONFIG_L2CACHE
, &tmp
) == 0)
5797 return (spa_vdev_exit(spa
, NULL
, txg
, EINVAL
));
5799 vml
= kmem_zalloc(children
* sizeof (vdev_t
*), KM_SLEEP
);
5800 glist
= kmem_zalloc(children
* sizeof (uint64_t), KM_SLEEP
);
5802 /* then, loop over each vdev and validate it */
5803 for (c
= 0; c
< children
; c
++) {
5804 uint64_t is_hole
= 0;
5806 (void) nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_IS_HOLE
,
5810 if (spa
->spa_root_vdev
->vdev_child
[c
]->vdev_ishole
||
5811 spa
->spa_root_vdev
->vdev_child
[c
]->vdev_islog
) {
5814 error
= SET_ERROR(EINVAL
);
5819 /* which disk is going to be split? */
5820 if (nvlist_lookup_uint64(child
[c
], ZPOOL_CONFIG_GUID
,
5822 error
= SET_ERROR(EINVAL
);
5826 /* look it up in the spa */
5827 vml
[c
] = spa_lookup_by_guid(spa
, glist
[c
], B_FALSE
);
5828 if (vml
[c
] == NULL
) {
5829 error
= SET_ERROR(ENODEV
);
5833 /* make sure there's nothing stopping the split */
5834 if (vml
[c
]->vdev_parent
->vdev_ops
!= &vdev_mirror_ops
||
5835 vml
[c
]->vdev_islog
||
5836 !vdev_is_concrete(vml
[c
]) ||
5837 vml
[c
]->vdev_isspare
||
5838 vml
[c
]->vdev_isl2cache
||
5839 !vdev_writeable(vml
[c
]) ||
5840 vml
[c
]->vdev_children
!= 0 ||
5841 vml
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
||
5842 c
!= spa
->spa_root_vdev
->vdev_child
[c
]->vdev_id
) {
5843 error
= SET_ERROR(EINVAL
);
5847 if (vdev_dtl_required(vml
[c
])) {
5848 error
= SET_ERROR(EBUSY
);
5852 /* we need certain info from the top level */
5853 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_ARRAY
,
5854 vml
[c
]->vdev_top
->vdev_ms_array
) == 0);
5855 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_METASLAB_SHIFT
,
5856 vml
[c
]->vdev_top
->vdev_ms_shift
) == 0);
5857 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASIZE
,
5858 vml
[c
]->vdev_top
->vdev_asize
) == 0);
5859 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_ASHIFT
,
5860 vml
[c
]->vdev_top
->vdev_ashift
) == 0);
5862 /* transfer per-vdev ZAPs */
5863 ASSERT3U(vml
[c
]->vdev_leaf_zap
, !=, 0);
5864 VERIFY0(nvlist_add_uint64(child
[c
],
5865 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, vml
[c
]->vdev_leaf_zap
));
5867 ASSERT3U(vml
[c
]->vdev_top
->vdev_top_zap
, !=, 0);
5868 VERIFY0(nvlist_add_uint64(child
[c
],
5869 ZPOOL_CONFIG_VDEV_TOP_ZAP
,
5870 vml
[c
]->vdev_parent
->vdev_top_zap
));
5874 kmem_free(vml
, children
* sizeof (vdev_t
*));
5875 kmem_free(glist
, children
* sizeof (uint64_t));
5876 return (spa_vdev_exit(spa
, NULL
, txg
, error
));
5879 /* stop writers from using the disks */
5880 for (c
= 0; c
< children
; c
++) {
5882 vml
[c
]->vdev_offline
= B_TRUE
;
5884 vdev_reopen(spa
->spa_root_vdev
);
5887 * Temporarily record the splitting vdevs in the spa config. This
5888 * will disappear once the config is regenerated.
5890 VERIFY(nvlist_alloc(&nvl
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5891 VERIFY(nvlist_add_uint64_array(nvl
, ZPOOL_CONFIG_SPLIT_LIST
,
5892 glist
, children
) == 0);
5893 kmem_free(glist
, children
* sizeof (uint64_t));
5895 mutex_enter(&spa
->spa_props_lock
);
5896 VERIFY(nvlist_add_nvlist(spa
->spa_config
, ZPOOL_CONFIG_SPLIT
,
5898 mutex_exit(&spa
->spa_props_lock
);
5899 spa
->spa_config_splitting
= nvl
;
5900 vdev_config_dirty(spa
->spa_root_vdev
);
5902 /* configure and create the new pool */
5903 VERIFY(nvlist_add_string(config
, ZPOOL_CONFIG_POOL_NAME
, newname
) == 0);
5904 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_STATE
,
5905 exp
? POOL_STATE_EXPORTED
: POOL_STATE_ACTIVE
) == 0);
5906 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
5907 spa_version(spa
)) == 0);
5908 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_TXG
,
5909 spa
->spa_config_txg
) == 0);
5910 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_POOL_GUID
,
5911 spa_generate_guid(NULL
)) == 0);
5912 VERIFY0(nvlist_add_boolean(config
, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS
));
5913 (void) nvlist_lookup_string(props
,
5914 zpool_prop_to_name(ZPOOL_PROP_ALTROOT
), &altroot
);
5916 /* add the new pool to the namespace */
5917 newspa
= spa_add(newname
, config
, altroot
);
5918 newspa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5919 newspa
->spa_config_txg
= spa
->spa_config_txg
;
5920 spa_set_log_state(newspa
, SPA_LOG_CLEAR
);
5922 /* release the spa config lock, retaining the namespace lock */
5923 spa_vdev_config_exit(spa
, NULL
, txg
, 0, FTAG
);
5925 if (zio_injection_enabled
)
5926 zio_handle_panic_injection(spa
, FTAG
, 1);
5928 spa_activate(newspa
, spa_mode_global
);
5929 spa_async_suspend(newspa
);
5931 newspa
->spa_config_source
= SPA_CONFIG_SRC_SPLIT
;
5933 /* create the new pool from the disks of the original pool */
5934 error
= spa_load(newspa
, SPA_LOAD_IMPORT
, SPA_IMPORT_ASSEMBLE
);
5938 /* if that worked, generate a real config for the new pool */
5939 if (newspa
->spa_root_vdev
!= NULL
) {
5940 VERIFY(nvlist_alloc(&newspa
->spa_config_splitting
,
5941 NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
5942 VERIFY(nvlist_add_uint64(newspa
->spa_config_splitting
,
5943 ZPOOL_CONFIG_SPLIT_GUID
, spa_guid(spa
)) == 0);
5944 spa_config_set(newspa
, spa_config_generate(newspa
, NULL
, -1ULL,
5949 if (props
!= NULL
) {
5950 spa_configfile_set(newspa
, props
, B_FALSE
);
5951 error
= spa_prop_set(newspa
, props
);
5956 /* flush everything */
5957 txg
= spa_vdev_config_enter(newspa
);
5958 vdev_config_dirty(newspa
->spa_root_vdev
);
5959 (void) spa_vdev_config_exit(newspa
, NULL
, txg
, 0, FTAG
);
5961 if (zio_injection_enabled
)
5962 zio_handle_panic_injection(spa
, FTAG
, 2);
5964 spa_async_resume(newspa
);
5966 /* finally, update the original pool's config */
5967 txg
= spa_vdev_config_enter(spa
);
5968 tx
= dmu_tx_create_dd(spa_get_dsl(spa
)->dp_mos_dir
);
5969 error
= dmu_tx_assign(tx
, TXG_WAIT
);
5972 for (c
= 0; c
< children
; c
++) {
5973 if (vml
[c
] != NULL
) {
5976 spa_history_log_internal(spa
, "detach", tx
,
5977 "vdev=%s", vml
[c
]->vdev_path
);
5982 spa
->spa_avz_action
= AVZ_ACTION_REBUILD
;
5983 vdev_config_dirty(spa
->spa_root_vdev
);
5984 spa
->spa_config_splitting
= NULL
;
5988 (void) spa_vdev_exit(spa
, NULL
, txg
, 0);
5990 if (zio_injection_enabled
)
5991 zio_handle_panic_injection(spa
, FTAG
, 3);
5993 /* split is complete; log a history record */
5994 spa_history_log_internal(newspa
, "split", NULL
,
5995 "from pool %s", spa_name(spa
));
5997 kmem_free(vml
, children
* sizeof (vdev_t
*));
5999 /* if we're not going to mount the filesystems in userland, export */
6001 error
= spa_export_common(newname
, POOL_STATE_EXPORTED
, NULL
,
6008 spa_deactivate(newspa
);
6011 txg
= spa_vdev_config_enter(spa
);
6013 /* re-online all offlined disks */
6014 for (c
= 0; c
< children
; c
++) {
6016 vml
[c
]->vdev_offline
= B_FALSE
;
6018 vdev_reopen(spa
->spa_root_vdev
);
6020 nvlist_free(spa
->spa_config_splitting
);
6021 spa
->spa_config_splitting
= NULL
;
6022 (void) spa_vdev_exit(spa
, NULL
, txg
, error
);
6024 kmem_free(vml
, children
* sizeof (vdev_t
*));
6029 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6030 * currently spared, so we can detach it.
6033 spa_vdev_resilver_done_hunt(vdev_t
*vd
)
6035 vdev_t
*newvd
, *oldvd
;
6037 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6038 oldvd
= spa_vdev_resilver_done_hunt(vd
->vdev_child
[c
]);
6044 * Check for a completed replacement. We always consider the first
6045 * vdev in the list to be the oldest vdev, and the last one to be
6046 * the newest (see spa_vdev_attach() for how that works). In
6047 * the case where the newest vdev is faulted, we will not automatically
6048 * remove it after a resilver completes. This is OK as it will require
6049 * user intervention to determine which disk the admin wishes to keep.
6051 if (vd
->vdev_ops
== &vdev_replacing_ops
) {
6052 ASSERT(vd
->vdev_children
> 1);
6054 newvd
= vd
->vdev_child
[vd
->vdev_children
- 1];
6055 oldvd
= vd
->vdev_child
[0];
6057 if (vdev_dtl_empty(newvd
, DTL_MISSING
) &&
6058 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
6059 !vdev_dtl_required(oldvd
))
6064 * Check for a completed resilver with the 'unspare' flag set.
6066 if (vd
->vdev_ops
== &vdev_spare_ops
) {
6067 vdev_t
*first
= vd
->vdev_child
[0];
6068 vdev_t
*last
= vd
->vdev_child
[vd
->vdev_children
- 1];
6070 if (last
->vdev_unspare
) {
6073 } else if (first
->vdev_unspare
) {
6080 if (oldvd
!= NULL
&&
6081 vdev_dtl_empty(newvd
, DTL_MISSING
) &&
6082 vdev_dtl_empty(newvd
, DTL_OUTAGE
) &&
6083 !vdev_dtl_required(oldvd
))
6087 * If there are more than two spares attached to a disk,
6088 * and those spares are not required, then we want to
6089 * attempt to free them up now so that they can be used
6090 * by other pools. Once we're back down to a single
6091 * disk+spare, we stop removing them.
6093 if (vd
->vdev_children
> 2) {
6094 newvd
= vd
->vdev_child
[1];
6096 if (newvd
->vdev_isspare
&& last
->vdev_isspare
&&
6097 vdev_dtl_empty(last
, DTL_MISSING
) &&
6098 vdev_dtl_empty(last
, DTL_OUTAGE
) &&
6099 !vdev_dtl_required(newvd
))
6108 spa_vdev_resilver_done(spa_t
*spa
)
6110 vdev_t
*vd
, *pvd
, *ppvd
;
6111 uint64_t guid
, sguid
, pguid
, ppguid
;
6113 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6115 while ((vd
= spa_vdev_resilver_done_hunt(spa
->spa_root_vdev
)) != NULL
) {
6116 pvd
= vd
->vdev_parent
;
6117 ppvd
= pvd
->vdev_parent
;
6118 guid
= vd
->vdev_guid
;
6119 pguid
= pvd
->vdev_guid
;
6120 ppguid
= ppvd
->vdev_guid
;
6123 * If we have just finished replacing a hot spared device, then
6124 * we need to detach the parent's first child (the original hot
6127 if (ppvd
->vdev_ops
== &vdev_spare_ops
&& pvd
->vdev_id
== 0 &&
6128 ppvd
->vdev_children
== 2) {
6129 ASSERT(pvd
->vdev_ops
== &vdev_replacing_ops
);
6130 sguid
= ppvd
->vdev_child
[1]->vdev_guid
;
6132 ASSERT(vd
->vdev_resilver_txg
== 0 || !vdev_dtl_required(vd
));
6134 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6135 if (spa_vdev_detach(spa
, guid
, pguid
, B_TRUE
) != 0)
6137 if (sguid
&& spa_vdev_detach(spa
, sguid
, ppguid
, B_TRUE
) != 0)
6139 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
6142 spa_config_exit(spa
, SCL_ALL
, FTAG
);
6146 * Update the stored path or FRU for this vdev.
6149 spa_vdev_set_common(spa_t
*spa
, uint64_t guid
, const char *value
,
6153 boolean_t sync
= B_FALSE
;
6155 ASSERT(spa_writeable(spa
));
6157 spa_vdev_state_enter(spa
, SCL_ALL
);
6159 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
6160 return (spa_vdev_state_exit(spa
, NULL
, ENOENT
));
6162 if (!vd
->vdev_ops
->vdev_op_leaf
)
6163 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
6166 if (strcmp(value
, vd
->vdev_path
) != 0) {
6167 spa_strfree(vd
->vdev_path
);
6168 vd
->vdev_path
= spa_strdup(value
);
6172 if (vd
->vdev_fru
== NULL
) {
6173 vd
->vdev_fru
= spa_strdup(value
);
6175 } else if (strcmp(value
, vd
->vdev_fru
) != 0) {
6176 spa_strfree(vd
->vdev_fru
);
6177 vd
->vdev_fru
= spa_strdup(value
);
6182 return (spa_vdev_state_exit(spa
, sync
? vd
: NULL
, 0));
6186 spa_vdev_setpath(spa_t
*spa
, uint64_t guid
, const char *newpath
)
6188 return (spa_vdev_set_common(spa
, guid
, newpath
, B_TRUE
));
6192 spa_vdev_setfru(spa_t
*spa
, uint64_t guid
, const char *newfru
)
6194 return (spa_vdev_set_common(spa
, guid
, newfru
, B_FALSE
));
6198 * ==========================================================================
6200 * ==========================================================================
6203 spa_scrub_pause_resume(spa_t
*spa
, pool_scrub_cmd_t cmd
)
6205 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6207 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6208 return (SET_ERROR(EBUSY
));
6210 return (dsl_scrub_set_pause_resume(spa
->spa_dsl_pool
, cmd
));
6214 spa_scan_stop(spa_t
*spa
)
6216 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6217 if (dsl_scan_resilvering(spa
->spa_dsl_pool
))
6218 return (SET_ERROR(EBUSY
));
6219 return (dsl_scan_cancel(spa
->spa_dsl_pool
));
6223 spa_scan(spa_t
*spa
, pool_scan_func_t func
)
6225 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == 0);
6227 if (func
>= POOL_SCAN_FUNCS
|| func
== POOL_SCAN_NONE
)
6228 return (SET_ERROR(ENOTSUP
));
6231 * If a resilver was requested, but there is no DTL on a
6232 * writeable leaf device, we have nothing to do.
6234 if (func
== POOL_SCAN_RESILVER
&&
6235 !vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
)) {
6236 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
6240 return (dsl_scan(spa
->spa_dsl_pool
, func
));
6244 * ==========================================================================
6245 * SPA async task processing
6246 * ==========================================================================
6250 spa_async_remove(spa_t
*spa
, vdev_t
*vd
)
6252 if (vd
->vdev_remove_wanted
) {
6253 vd
->vdev_remove_wanted
= B_FALSE
;
6254 vd
->vdev_delayed_close
= B_FALSE
;
6255 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_REMOVED
, VDEV_AUX_NONE
);
6258 * We want to clear the stats, but we don't want to do a full
6259 * vdev_clear() as that will cause us to throw away
6260 * degraded/faulted state as well as attempt to reopen the
6261 * device, all of which is a waste.
6263 vd
->vdev_stat
.vs_read_errors
= 0;
6264 vd
->vdev_stat
.vs_write_errors
= 0;
6265 vd
->vdev_stat
.vs_checksum_errors
= 0;
6267 vdev_state_dirty(vd
->vdev_top
);
6270 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6271 spa_async_remove(spa
, vd
->vdev_child
[c
]);
6275 spa_async_probe(spa_t
*spa
, vdev_t
*vd
)
6277 if (vd
->vdev_probe_wanted
) {
6278 vd
->vdev_probe_wanted
= B_FALSE
;
6279 vdev_reopen(vd
); /* vdev_open() does the actual probe */
6282 for (int c
= 0; c
< vd
->vdev_children
; c
++)
6283 spa_async_probe(spa
, vd
->vdev_child
[c
]);
6287 spa_async_autoexpand(spa_t
*spa
, vdev_t
*vd
)
6293 if (!spa
->spa_autoexpand
)
6296 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
6297 vdev_t
*cvd
= vd
->vdev_child
[c
];
6298 spa_async_autoexpand(spa
, cvd
);
6301 if (!vd
->vdev_ops
->vdev_op_leaf
|| vd
->vdev_physpath
== NULL
)
6304 physpath
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
6305 (void) snprintf(physpath
, MAXPATHLEN
, "/devices%s", vd
->vdev_physpath
);
6307 VERIFY(nvlist_alloc(&attr
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6308 VERIFY(nvlist_add_string(attr
, DEV_PHYS_PATH
, physpath
) == 0);
6310 (void) ddi_log_sysevent(zfs_dip
, SUNW_VENDOR
, EC_DEV_STATUS
,
6311 ESC_DEV_DLE
, attr
, &eid
, DDI_SLEEP
);
6314 kmem_free(physpath
, MAXPATHLEN
);
6318 spa_async_thread(void *arg
)
6320 spa_t
*spa
= (spa_t
*)arg
;
6323 ASSERT(spa
->spa_sync_on
);
6325 mutex_enter(&spa
->spa_async_lock
);
6326 tasks
= spa
->spa_async_tasks
;
6327 spa
->spa_async_tasks
= 0;
6328 mutex_exit(&spa
->spa_async_lock
);
6331 * See if the config needs to be updated.
6333 if (tasks
& SPA_ASYNC_CONFIG_UPDATE
) {
6334 uint64_t old_space
, new_space
;
6336 mutex_enter(&spa_namespace_lock
);
6337 old_space
= metaslab_class_get_space(spa_normal_class(spa
));
6338 spa_config_update(spa
, SPA_CONFIG_UPDATE_POOL
);
6339 new_space
= metaslab_class_get_space(spa_normal_class(spa
));
6340 mutex_exit(&spa_namespace_lock
);
6343 * If the pool grew as a result of the config update,
6344 * then log an internal history event.
6346 if (new_space
!= old_space
) {
6347 spa_history_log_internal(spa
, "vdev online", NULL
,
6348 "pool '%s' size: %llu(+%llu)",
6349 spa_name(spa
), new_space
, new_space
- old_space
);
6354 * See if any devices need to be marked REMOVED.
6356 if (tasks
& SPA_ASYNC_REMOVE
) {
6357 spa_vdev_state_enter(spa
, SCL_NONE
);
6358 spa_async_remove(spa
, spa
->spa_root_vdev
);
6359 for (int i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++)
6360 spa_async_remove(spa
, spa
->spa_l2cache
.sav_vdevs
[i
]);
6361 for (int i
= 0; i
< spa
->spa_spares
.sav_count
; i
++)
6362 spa_async_remove(spa
, spa
->spa_spares
.sav_vdevs
[i
]);
6363 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6366 if ((tasks
& SPA_ASYNC_AUTOEXPAND
) && !spa_suspended(spa
)) {
6367 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6368 spa_async_autoexpand(spa
, spa
->spa_root_vdev
);
6369 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6373 * See if any devices need to be probed.
6375 if (tasks
& SPA_ASYNC_PROBE
) {
6376 spa_vdev_state_enter(spa
, SCL_NONE
);
6377 spa_async_probe(spa
, spa
->spa_root_vdev
);
6378 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6382 * If any devices are done replacing, detach them.
6384 if (tasks
& SPA_ASYNC_RESILVER_DONE
)
6385 spa_vdev_resilver_done(spa
);
6388 * Kick off a resilver.
6390 if (tasks
& SPA_ASYNC_RESILVER
)
6391 dsl_resilver_restart(spa
->spa_dsl_pool
, 0);
6394 * Let the world know that we're done.
6396 mutex_enter(&spa
->spa_async_lock
);
6397 spa
->spa_async_thread
= NULL
;
6398 cv_broadcast(&spa
->spa_async_cv
);
6399 mutex_exit(&spa
->spa_async_lock
);
6404 spa_async_suspend(spa_t
*spa
)
6406 mutex_enter(&spa
->spa_async_lock
);
6407 spa
->spa_async_suspended
++;
6408 while (spa
->spa_async_thread
!= NULL
)
6409 cv_wait(&spa
->spa_async_cv
, &spa
->spa_async_lock
);
6410 mutex_exit(&spa
->spa_async_lock
);
6412 spa_vdev_remove_suspend(spa
);
6414 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
6415 if (condense_thread
!= NULL
&& zthr_isrunning(condense_thread
))
6416 VERIFY0(zthr_cancel(condense_thread
));
6420 spa_async_resume(spa_t
*spa
)
6422 mutex_enter(&spa
->spa_async_lock
);
6423 ASSERT(spa
->spa_async_suspended
!= 0);
6424 spa
->spa_async_suspended
--;
6425 mutex_exit(&spa
->spa_async_lock
);
6426 spa_restart_removal(spa
);
6428 zthr_t
*condense_thread
= spa
->spa_condense_zthr
;
6429 if (condense_thread
!= NULL
&& !zthr_isrunning(condense_thread
))
6430 zthr_resume(condense_thread
);
6434 spa_async_tasks_pending(spa_t
*spa
)
6436 uint_t non_config_tasks
;
6438 boolean_t config_task_suspended
;
6440 non_config_tasks
= spa
->spa_async_tasks
& ~SPA_ASYNC_CONFIG_UPDATE
;
6441 config_task
= spa
->spa_async_tasks
& SPA_ASYNC_CONFIG_UPDATE
;
6442 if (spa
->spa_ccw_fail_time
== 0) {
6443 config_task_suspended
= B_FALSE
;
6445 config_task_suspended
=
6446 (gethrtime() - spa
->spa_ccw_fail_time
) <
6447 (zfs_ccw_retry_interval
* NANOSEC
);
6450 return (non_config_tasks
|| (config_task
&& !config_task_suspended
));
6454 spa_async_dispatch(spa_t
*spa
)
6456 mutex_enter(&spa
->spa_async_lock
);
6457 if (spa_async_tasks_pending(spa
) &&
6458 !spa
->spa_async_suspended
&&
6459 spa
->spa_async_thread
== NULL
&&
6461 spa
->spa_async_thread
= thread_create(NULL
, 0,
6462 spa_async_thread
, spa
, 0, &p0
, TS_RUN
, maxclsyspri
);
6463 mutex_exit(&spa
->spa_async_lock
);
6467 spa_async_request(spa_t
*spa
, int task
)
6469 zfs_dbgmsg("spa=%s async request task=%u", spa
->spa_name
, task
);
6470 mutex_enter(&spa
->spa_async_lock
);
6471 spa
->spa_async_tasks
|= task
;
6472 mutex_exit(&spa
->spa_async_lock
);
6476 * ==========================================================================
6477 * SPA syncing routines
6478 * ==========================================================================
6482 bpobj_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6485 bpobj_enqueue(bpo
, bp
, tx
);
6490 spa_free_sync_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6494 zio_nowait(zio_free_sync(zio
, zio
->io_spa
, dmu_tx_get_txg(tx
), bp
,
6500 * Note: this simple function is not inlined to make it easier to dtrace the
6501 * amount of time spent syncing frees.
6504 spa_sync_frees(spa_t
*spa
, bplist_t
*bpl
, dmu_tx_t
*tx
)
6506 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6507 bplist_iterate(bpl
, spa_free_sync_cb
, zio
, tx
);
6508 VERIFY(zio_wait(zio
) == 0);
6512 * Note: this simple function is not inlined to make it easier to dtrace the
6513 * amount of time spent syncing deferred frees.
6516 spa_sync_deferred_frees(spa_t
*spa
, dmu_tx_t
*tx
)
6518 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, 0);
6519 VERIFY3U(bpobj_iterate(&spa
->spa_deferred_bpobj
,
6520 spa_free_sync_cb
, zio
, tx
), ==, 0);
6521 VERIFY0(zio_wait(zio
));
6526 spa_sync_nvlist(spa_t
*spa
, uint64_t obj
, nvlist_t
*nv
, dmu_tx_t
*tx
)
6528 char *packed
= NULL
;
6533 VERIFY(nvlist_size(nv
, &nvsize
, NV_ENCODE_XDR
) == 0);
6536 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6537 * information. This avoids the dmu_buf_will_dirty() path and
6538 * saves us a pre-read to get data we don't actually care about.
6540 bufsize
= P2ROUNDUP((uint64_t)nvsize
, SPA_CONFIG_BLOCKSIZE
);
6541 packed
= kmem_alloc(bufsize
, KM_SLEEP
);
6543 VERIFY(nvlist_pack(nv
, &packed
, &nvsize
, NV_ENCODE_XDR
,
6545 bzero(packed
+ nvsize
, bufsize
- nvsize
);
6547 dmu_write(spa
->spa_meta_objset
, obj
, 0, bufsize
, packed
, tx
);
6549 kmem_free(packed
, bufsize
);
6551 VERIFY(0 == dmu_bonus_hold(spa
->spa_meta_objset
, obj
, FTAG
, &db
));
6552 dmu_buf_will_dirty(db
, tx
);
6553 *(uint64_t *)db
->db_data
= nvsize
;
6554 dmu_buf_rele(db
, FTAG
);
6558 spa_sync_aux_dev(spa_t
*spa
, spa_aux_vdev_t
*sav
, dmu_tx_t
*tx
,
6559 const char *config
, const char *entry
)
6569 * Update the MOS nvlist describing the list of available devices.
6570 * spa_validate_aux() will have already made sure this nvlist is
6571 * valid and the vdevs are labeled appropriately.
6573 if (sav
->sav_object
== 0) {
6574 sav
->sav_object
= dmu_object_alloc(spa
->spa_meta_objset
,
6575 DMU_OT_PACKED_NVLIST
, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE
,
6576 sizeof (uint64_t), tx
);
6577 VERIFY(zap_update(spa
->spa_meta_objset
,
6578 DMU_POOL_DIRECTORY_OBJECT
, entry
, sizeof (uint64_t), 1,
6579 &sav
->sav_object
, tx
) == 0);
6582 VERIFY(nvlist_alloc(&nvroot
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
6583 if (sav
->sav_count
== 0) {
6584 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, NULL
, 0) == 0);
6586 list
= kmem_alloc(sav
->sav_count
* sizeof (void *), KM_SLEEP
);
6587 for (i
= 0; i
< sav
->sav_count
; i
++)
6588 list
[i
] = vdev_config_generate(spa
, sav
->sav_vdevs
[i
],
6589 B_FALSE
, VDEV_CONFIG_L2CACHE
);
6590 VERIFY(nvlist_add_nvlist_array(nvroot
, config
, list
,
6591 sav
->sav_count
) == 0);
6592 for (i
= 0; i
< sav
->sav_count
; i
++)
6593 nvlist_free(list
[i
]);
6594 kmem_free(list
, sav
->sav_count
* sizeof (void *));
6597 spa_sync_nvlist(spa
, sav
->sav_object
, nvroot
, tx
);
6598 nvlist_free(nvroot
);
6600 sav
->sav_sync
= B_FALSE
;
6604 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6605 * The all-vdev ZAP must be empty.
6608 spa_avz_build(vdev_t
*vd
, uint64_t avz
, dmu_tx_t
*tx
)
6610 spa_t
*spa
= vd
->vdev_spa
;
6611 if (vd
->vdev_top_zap
!= 0) {
6612 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6613 vd
->vdev_top_zap
, tx
));
6615 if (vd
->vdev_leaf_zap
!= 0) {
6616 VERIFY0(zap_add_int(spa
->spa_meta_objset
, avz
,
6617 vd
->vdev_leaf_zap
, tx
));
6619 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
6620 spa_avz_build(vd
->vdev_child
[i
], avz
, tx
);
6625 spa_sync_config_object(spa_t
*spa
, dmu_tx_t
*tx
)
6630 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6631 * its config may not be dirty but we still need to build per-vdev ZAPs.
6632 * Similarly, if the pool is being assembled (e.g. after a split), we
6633 * need to rebuild the AVZ although the config may not be dirty.
6635 if (list_is_empty(&spa
->spa_config_dirty_list
) &&
6636 spa
->spa_avz_action
== AVZ_ACTION_NONE
)
6639 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6641 ASSERT(spa
->spa_avz_action
== AVZ_ACTION_NONE
||
6642 spa
->spa_avz_action
== AVZ_ACTION_INITIALIZE
||
6643 spa
->spa_all_vdev_zaps
!= 0);
6645 if (spa
->spa_avz_action
== AVZ_ACTION_REBUILD
) {
6646 /* Make and build the new AVZ */
6647 uint64_t new_avz
= zap_create(spa
->spa_meta_objset
,
6648 DMU_OTN_ZAP_METADATA
, DMU_OT_NONE
, 0, tx
);
6649 spa_avz_build(spa
->spa_root_vdev
, new_avz
, tx
);
6651 /* Diff old AVZ with new one */
6655 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6656 spa
->spa_all_vdev_zaps
);
6657 zap_cursor_retrieve(&zc
, &za
) == 0;
6658 zap_cursor_advance(&zc
)) {
6659 uint64_t vdzap
= za
.za_first_integer
;
6660 if (zap_lookup_int(spa
->spa_meta_objset
, new_avz
,
6663 * ZAP is listed in old AVZ but not in new one;
6666 VERIFY0(zap_destroy(spa
->spa_meta_objset
, vdzap
,
6671 zap_cursor_fini(&zc
);
6673 /* Destroy the old AVZ */
6674 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6675 spa
->spa_all_vdev_zaps
, tx
));
6677 /* Replace the old AVZ in the dir obj with the new one */
6678 VERIFY0(zap_update(spa
->spa_meta_objset
,
6679 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
,
6680 sizeof (new_avz
), 1, &new_avz
, tx
));
6682 spa
->spa_all_vdev_zaps
= new_avz
;
6683 } else if (spa
->spa_avz_action
== AVZ_ACTION_DESTROY
) {
6687 /* Walk through the AVZ and destroy all listed ZAPs */
6688 for (zap_cursor_init(&zc
, spa
->spa_meta_objset
,
6689 spa
->spa_all_vdev_zaps
);
6690 zap_cursor_retrieve(&zc
, &za
) == 0;
6691 zap_cursor_advance(&zc
)) {
6692 uint64_t zap
= za
.za_first_integer
;
6693 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zap
, tx
));
6696 zap_cursor_fini(&zc
);
6698 /* Destroy and unlink the AVZ itself */
6699 VERIFY0(zap_destroy(spa
->spa_meta_objset
,
6700 spa
->spa_all_vdev_zaps
, tx
));
6701 VERIFY0(zap_remove(spa
->spa_meta_objset
,
6702 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_VDEV_ZAP_MAP
, tx
));
6703 spa
->spa_all_vdev_zaps
= 0;
6706 if (spa
->spa_all_vdev_zaps
== 0) {
6707 spa
->spa_all_vdev_zaps
= zap_create_link(spa
->spa_meta_objset
,
6708 DMU_OTN_ZAP_METADATA
, DMU_POOL_DIRECTORY_OBJECT
,
6709 DMU_POOL_VDEV_ZAP_MAP
, tx
);
6711 spa
->spa_avz_action
= AVZ_ACTION_NONE
;
6713 /* Create ZAPs for vdevs that don't have them. */
6714 vdev_construct_zaps(spa
->spa_root_vdev
, tx
);
6716 config
= spa_config_generate(spa
, spa
->spa_root_vdev
,
6717 dmu_tx_get_txg(tx
), B_FALSE
);
6720 * If we're upgrading the spa version then make sure that
6721 * the config object gets updated with the correct version.
6723 if (spa
->spa_ubsync
.ub_version
< spa
->spa_uberblock
.ub_version
)
6724 fnvlist_add_uint64(config
, ZPOOL_CONFIG_VERSION
,
6725 spa
->spa_uberblock
.ub_version
);
6727 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6729 nvlist_free(spa
->spa_config_syncing
);
6730 spa
->spa_config_syncing
= config
;
6732 spa_sync_nvlist(spa
, spa
->spa_config_object
, config
, tx
);
6736 spa_sync_version(void *arg
, dmu_tx_t
*tx
)
6738 uint64_t *versionp
= arg
;
6739 uint64_t version
= *versionp
;
6740 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6743 * Setting the version is special cased when first creating the pool.
6745 ASSERT(tx
->tx_txg
!= TXG_INITIAL
);
6747 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
6748 ASSERT(version
>= spa_version(spa
));
6750 spa
->spa_uberblock
.ub_version
= version
;
6751 vdev_config_dirty(spa
->spa_root_vdev
);
6752 spa_history_log_internal(spa
, "set", tx
, "version=%lld", version
);
6756 * Set zpool properties.
6759 spa_sync_props(void *arg
, dmu_tx_t
*tx
)
6761 nvlist_t
*nvp
= arg
;
6762 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
6763 objset_t
*mos
= spa
->spa_meta_objset
;
6764 nvpair_t
*elem
= NULL
;
6766 mutex_enter(&spa
->spa_props_lock
);
6768 while ((elem
= nvlist_next_nvpair(nvp
, elem
))) {
6770 char *strval
, *fname
;
6772 const char *propname
;
6773 zprop_type_t proptype
;
6776 switch (prop
= zpool_name_to_prop(nvpair_name(elem
))) {
6777 case ZPOOL_PROP_INVAL
:
6779 * We checked this earlier in spa_prop_validate().
6781 ASSERT(zpool_prop_feature(nvpair_name(elem
)));
6783 fname
= strchr(nvpair_name(elem
), '@') + 1;
6784 VERIFY0(zfeature_lookup_name(fname
, &fid
));
6786 spa_feature_enable(spa
, fid
, tx
);
6787 spa_history_log_internal(spa
, "set", tx
,
6788 "%s=enabled", nvpair_name(elem
));
6791 case ZPOOL_PROP_VERSION
:
6792 intval
= fnvpair_value_uint64(elem
);
6794 * The version is synced seperatly before other
6795 * properties and should be correct by now.
6797 ASSERT3U(spa_version(spa
), >=, intval
);
6800 case ZPOOL_PROP_ALTROOT
:
6802 * 'altroot' is a non-persistent property. It should
6803 * have been set temporarily at creation or import time.
6805 ASSERT(spa
->spa_root
!= NULL
);
6808 case ZPOOL_PROP_READONLY
:
6809 case ZPOOL_PROP_CACHEFILE
:
6811 * 'readonly' and 'cachefile' are also non-persisitent
6815 case ZPOOL_PROP_COMMENT
:
6816 strval
= fnvpair_value_string(elem
);
6817 if (spa
->spa_comment
!= NULL
)
6818 spa_strfree(spa
->spa_comment
);
6819 spa
->spa_comment
= spa_strdup(strval
);
6821 * We need to dirty the configuration on all the vdevs
6822 * so that their labels get updated. It's unnecessary
6823 * to do this for pool creation since the vdev's
6824 * configuratoin has already been dirtied.
6826 if (tx
->tx_txg
!= TXG_INITIAL
)
6827 vdev_config_dirty(spa
->spa_root_vdev
);
6828 spa_history_log_internal(spa
, "set", tx
,
6829 "%s=%s", nvpair_name(elem
), strval
);
6833 * Set pool property values in the poolprops mos object.
6835 if (spa
->spa_pool_props_object
== 0) {
6836 spa
->spa_pool_props_object
=
6837 zap_create_link(mos
, DMU_OT_POOL_PROPS
,
6838 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_PROPS
,
6842 /* normalize the property name */
6843 propname
= zpool_prop_to_name(prop
);
6844 proptype
= zpool_prop_get_type(prop
);
6846 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
6847 ASSERT(proptype
== PROP_TYPE_STRING
);
6848 strval
= fnvpair_value_string(elem
);
6849 VERIFY0(zap_update(mos
,
6850 spa
->spa_pool_props_object
, propname
,
6851 1, strlen(strval
) + 1, strval
, tx
));
6852 spa_history_log_internal(spa
, "set", tx
,
6853 "%s=%s", nvpair_name(elem
), strval
);
6854 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
6855 intval
= fnvpair_value_uint64(elem
);
6857 if (proptype
== PROP_TYPE_INDEX
) {
6859 VERIFY0(zpool_prop_index_to_string(
6860 prop
, intval
, &unused
));
6862 VERIFY0(zap_update(mos
,
6863 spa
->spa_pool_props_object
, propname
,
6864 8, 1, &intval
, tx
));
6865 spa_history_log_internal(spa
, "set", tx
,
6866 "%s=%lld", nvpair_name(elem
), intval
);
6868 ASSERT(0); /* not allowed */
6872 case ZPOOL_PROP_DELEGATION
:
6873 spa
->spa_delegation
= intval
;
6875 case ZPOOL_PROP_BOOTFS
:
6876 spa
->spa_bootfs
= intval
;
6878 case ZPOOL_PROP_FAILUREMODE
:
6879 spa
->spa_failmode
= intval
;
6881 case ZPOOL_PROP_AUTOEXPAND
:
6882 spa
->spa_autoexpand
= intval
;
6883 if (tx
->tx_txg
!= TXG_INITIAL
)
6884 spa_async_request(spa
,
6885 SPA_ASYNC_AUTOEXPAND
);
6887 case ZPOOL_PROP_DEDUPDITTO
:
6888 spa
->spa_dedup_ditto
= intval
;
6897 mutex_exit(&spa
->spa_props_lock
);
6901 * Perform one-time upgrade on-disk changes. spa_version() does not
6902 * reflect the new version this txg, so there must be no changes this
6903 * txg to anything that the upgrade code depends on after it executes.
6904 * Therefore this must be called after dsl_pool_sync() does the sync
6908 spa_sync_upgrades(spa_t
*spa
, dmu_tx_t
*tx
)
6910 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6912 ASSERT(spa
->spa_sync_pass
== 1);
6914 rrw_enter(&dp
->dp_config_rwlock
, RW_WRITER
, FTAG
);
6916 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_ORIGIN
&&
6917 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_ORIGIN
) {
6918 dsl_pool_create_origin(dp
, tx
);
6920 /* Keeping the origin open increases spa_minref */
6921 spa
->spa_minref
+= 3;
6924 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_NEXT_CLONES
&&
6925 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_NEXT_CLONES
) {
6926 dsl_pool_upgrade_clones(dp
, tx
);
6929 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_DIR_CLONES
&&
6930 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_DIR_CLONES
) {
6931 dsl_pool_upgrade_dir_clones(dp
, tx
);
6933 /* Keeping the freedir open increases spa_minref */
6934 spa
->spa_minref
+= 3;
6937 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_FEATURES
&&
6938 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6939 spa_feature_create_zap_objects(spa
, tx
);
6943 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6944 * when possibility to use lz4 compression for metadata was added
6945 * Old pools that have this feature enabled must be upgraded to have
6946 * this feature active
6948 if (spa
->spa_uberblock
.ub_version
>= SPA_VERSION_FEATURES
) {
6949 boolean_t lz4_en
= spa_feature_is_enabled(spa
,
6950 SPA_FEATURE_LZ4_COMPRESS
);
6951 boolean_t lz4_ac
= spa_feature_is_active(spa
,
6952 SPA_FEATURE_LZ4_COMPRESS
);
6954 if (lz4_en
&& !lz4_ac
)
6955 spa_feature_incr(spa
, SPA_FEATURE_LZ4_COMPRESS
, tx
);
6959 * If we haven't written the salt, do so now. Note that the
6960 * feature may not be activated yet, but that's fine since
6961 * the presence of this ZAP entry is backwards compatible.
6963 if (zap_contains(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
6964 DMU_POOL_CHECKSUM_SALT
) == ENOENT
) {
6965 VERIFY0(zap_add(spa
->spa_meta_objset
,
6966 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_CHECKSUM_SALT
, 1,
6967 sizeof (spa
->spa_cksum_salt
.zcs_bytes
),
6968 spa
->spa_cksum_salt
.zcs_bytes
, tx
));
6971 rrw_exit(&dp
->dp_config_rwlock
, FTAG
);
6975 vdev_indirect_state_sync_verify(vdev_t
*vd
)
6977 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6978 vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
;
6980 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
6981 ASSERT(vim
!= NULL
);
6982 ASSERT(vib
!= NULL
);
6985 if (vdev_obsolete_sm_object(vd
) != 0) {
6986 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
6987 ASSERT(vd
->vdev_removing
||
6988 vd
->vdev_ops
== &vdev_indirect_ops
);
6989 ASSERT(vdev_indirect_mapping_num_entries(vim
) > 0);
6990 ASSERT(vdev_indirect_mapping_bytes_mapped(vim
) > 0);
6992 ASSERT3U(vdev_obsolete_sm_object(vd
), ==,
6993 space_map_object(vd
->vdev_obsolete_sm
));
6994 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim
), >=,
6995 space_map_allocated(vd
->vdev_obsolete_sm
));
6997 ASSERT(vd
->vdev_obsolete_segments
!= NULL
);
7000 * Since frees / remaps to an indirect vdev can only
7001 * happen in syncing context, the obsolete segments
7002 * tree must be empty when we start syncing.
7004 ASSERT0(range_tree_space(vd
->vdev_obsolete_segments
));
7008 * Sync the specified transaction group. New blocks may be dirtied as
7009 * part of the process, so we iterate until it converges.
7012 spa_sync(spa_t
*spa
, uint64_t txg
)
7014 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7015 objset_t
*mos
= spa
->spa_meta_objset
;
7016 bplist_t
*free_bpl
= &spa
->spa_free_bplist
[txg
& TXG_MASK
];
7017 vdev_t
*rvd
= spa
->spa_root_vdev
;
7021 uint32_t max_queue_depth
= zfs_vdev_async_write_max_active
*
7022 zfs_vdev_queue_depth_pct
/ 100;
7024 VERIFY(spa_writeable(spa
));
7027 * Wait for i/os issued in open context that need to complete
7028 * before this txg syncs.
7030 VERIFY0(zio_wait(spa
->spa_txg_zio
[txg
& TXG_MASK
]));
7031 spa
->spa_txg_zio
[txg
& TXG_MASK
] = zio_root(spa
, NULL
, NULL
, 0);
7034 * Lock out configuration changes.
7036 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
7038 spa
->spa_syncing_txg
= txg
;
7039 spa
->spa_sync_pass
= 0;
7041 mutex_enter(&spa
->spa_alloc_lock
);
7042 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7043 mutex_exit(&spa
->spa_alloc_lock
);
7046 * If there are any pending vdev state changes, convert them
7047 * into config changes that go out with this transaction group.
7049 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
7050 while (list_head(&spa
->spa_state_dirty_list
) != NULL
) {
7052 * We need the write lock here because, for aux vdevs,
7053 * calling vdev_config_dirty() modifies sav_config.
7054 * This is ugly and will become unnecessary when we
7055 * eliminate the aux vdev wart by integrating all vdevs
7056 * into the root vdev tree.
7058 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7059 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_WRITER
);
7060 while ((vd
= list_head(&spa
->spa_state_dirty_list
)) != NULL
) {
7061 vdev_state_clean(vd
);
7062 vdev_config_dirty(vd
);
7064 spa_config_exit(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
);
7065 spa_config_enter(spa
, SCL_CONFIG
| SCL_STATE
, FTAG
, RW_READER
);
7067 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7069 tx
= dmu_tx_create_assigned(dp
, txg
);
7071 spa
->spa_sync_starttime
= gethrtime();
7072 VERIFY(cyclic_reprogram(spa
->spa_deadman_cycid
,
7073 spa
->spa_sync_starttime
+ spa
->spa_deadman_synctime
));
7076 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
7077 * set spa_deflate if we have no raid-z vdevs.
7079 if (spa
->spa_ubsync
.ub_version
< SPA_VERSION_RAIDZ_DEFLATE
&&
7080 spa
->spa_uberblock
.ub_version
>= SPA_VERSION_RAIDZ_DEFLATE
) {
7083 for (i
= 0; i
< rvd
->vdev_children
; i
++) {
7084 vd
= rvd
->vdev_child
[i
];
7085 if (vd
->vdev_deflate_ratio
!= SPA_MINBLOCKSIZE
)
7088 if (i
== rvd
->vdev_children
) {
7089 spa
->spa_deflate
= TRUE
;
7090 VERIFY(0 == zap_add(spa
->spa_meta_objset
,
7091 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_DEFLATE
,
7092 sizeof (uint64_t), 1, &spa
->spa_deflate
, tx
));
7097 * Set the top-level vdev's max queue depth. Evaluate each
7098 * top-level's async write queue depth in case it changed.
7099 * The max queue depth will not change in the middle of syncing
7102 uint64_t queue_depth_total
= 0;
7103 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
7104 vdev_t
*tvd
= rvd
->vdev_child
[c
];
7105 metaslab_group_t
*mg
= tvd
->vdev_mg
;
7107 if (mg
== NULL
|| mg
->mg_class
!= spa_normal_class(spa
) ||
7108 !metaslab_group_initialized(mg
))
7112 * It is safe to do a lock-free check here because only async
7113 * allocations look at mg_max_alloc_queue_depth, and async
7114 * allocations all happen from spa_sync().
7116 ASSERT0(refcount_count(&mg
->mg_alloc_queue_depth
));
7117 mg
->mg_max_alloc_queue_depth
= max_queue_depth
;
7118 queue_depth_total
+= mg
->mg_max_alloc_queue_depth
;
7120 metaslab_class_t
*mc
= spa_normal_class(spa
);
7121 ASSERT0(refcount_count(&mc
->mc_alloc_slots
));
7122 mc
->mc_alloc_max_slots
= queue_depth_total
;
7123 mc
->mc_alloc_throttle_enabled
= zio_dva_throttle_enabled
;
7125 ASSERT3U(mc
->mc_alloc_max_slots
, <=,
7126 max_queue_depth
* rvd
->vdev_children
);
7128 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
7129 vdev_t
*vd
= rvd
->vdev_child
[c
];
7130 vdev_indirect_state_sync_verify(vd
);
7132 if (vdev_indirect_should_condense(vd
)) {
7133 spa_condense_indirect_start_sync(vd
, tx
);
7139 * Iterate to convergence.
7142 int pass
= ++spa
->spa_sync_pass
;
7144 spa_sync_config_object(spa
, tx
);
7145 spa_sync_aux_dev(spa
, &spa
->spa_spares
, tx
,
7146 ZPOOL_CONFIG_SPARES
, DMU_POOL_SPARES
);
7147 spa_sync_aux_dev(spa
, &spa
->spa_l2cache
, tx
,
7148 ZPOOL_CONFIG_L2CACHE
, DMU_POOL_L2CACHE
);
7149 spa_errlog_sync(spa
, txg
);
7150 dsl_pool_sync(dp
, txg
);
7152 if (pass
< zfs_sync_pass_deferred_free
) {
7153 spa_sync_frees(spa
, free_bpl
, tx
);
7156 * We can not defer frees in pass 1, because
7157 * we sync the deferred frees later in pass 1.
7159 ASSERT3U(pass
, >, 1);
7160 bplist_iterate(free_bpl
, bpobj_enqueue_cb
,
7161 &spa
->spa_deferred_bpobj
, tx
);
7165 dsl_scan_sync(dp
, tx
);
7167 if (spa
->spa_vdev_removal
!= NULL
)
7170 while ((vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, txg
))
7175 spa_sync_upgrades(spa
, tx
);
7177 spa
->spa_uberblock
.ub_rootbp
.blk_birth
);
7179 * Note: We need to check if the MOS is dirty
7180 * because we could have marked the MOS dirty
7181 * without updating the uberblock (e.g. if we
7182 * have sync tasks but no dirty user data). We
7183 * need to check the uberblock's rootbp because
7184 * it is updated if we have synced out dirty
7185 * data (though in this case the MOS will most
7186 * likely also be dirty due to second order
7187 * effects, we don't want to rely on that here).
7189 if (spa
->spa_uberblock
.ub_rootbp
.blk_birth
< txg
&&
7190 !dmu_objset_is_dirty(mos
, txg
)) {
7192 * Nothing changed on the first pass,
7193 * therefore this TXG is a no-op. Avoid
7194 * syncing deferred frees, so that we
7195 * can keep this TXG as a no-op.
7197 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
,
7199 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7200 ASSERT(txg_list_empty(&dp
->dp_sync_tasks
, txg
));
7203 spa_sync_deferred_frees(spa
, tx
);
7206 } while (dmu_objset_is_dirty(mos
, txg
));
7208 if (!list_is_empty(&spa
->spa_config_dirty_list
)) {
7210 * Make sure that the number of ZAPs for all the vdevs matches
7211 * the number of ZAPs in the per-vdev ZAP list. This only gets
7212 * called if the config is dirty; otherwise there may be
7213 * outstanding AVZ operations that weren't completed in
7214 * spa_sync_config_object.
7216 uint64_t all_vdev_zap_entry_count
;
7217 ASSERT0(zap_count(spa
->spa_meta_objset
,
7218 spa
->spa_all_vdev_zaps
, &all_vdev_zap_entry_count
));
7219 ASSERT3U(vdev_count_verify_zaps(spa
->spa_root_vdev
), ==,
7220 all_vdev_zap_entry_count
);
7223 if (spa
->spa_vdev_removal
!= NULL
) {
7224 ASSERT0(spa
->spa_vdev_removal
->svr_bytes_done
[txg
& TXG_MASK
]);
7228 * Rewrite the vdev configuration (which includes the uberblock)
7229 * to commit the transaction group.
7231 * If there are no dirty vdevs, we sync the uberblock to a few
7232 * random top-level vdevs that are known to be visible in the
7233 * config cache (see spa_vdev_add() for a complete description).
7234 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7238 * We hold SCL_STATE to prevent vdev open/close/etc.
7239 * while we're attempting to write the vdev labels.
7241 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
7243 if (list_is_empty(&spa
->spa_config_dirty_list
)) {
7244 vdev_t
*svd
[SPA_SYNC_MIN_VDEVS
];
7246 int children
= rvd
->vdev_children
;
7247 int c0
= spa_get_random(children
);
7249 for (int c
= 0; c
< children
; c
++) {
7250 vd
= rvd
->vdev_child
[(c0
+ c
) % children
];
7251 if (vd
->vdev_ms_array
== 0 || vd
->vdev_islog
||
7252 !vdev_is_concrete(vd
))
7254 svd
[svdcount
++] = vd
;
7255 if (svdcount
== SPA_SYNC_MIN_VDEVS
)
7258 error
= vdev_config_sync(svd
, svdcount
, txg
);
7260 error
= vdev_config_sync(rvd
->vdev_child
,
7261 rvd
->vdev_children
, txg
);
7265 spa
->spa_last_synced_guid
= rvd
->vdev_guid
;
7267 spa_config_exit(spa
, SCL_STATE
, FTAG
);
7271 zio_suspend(spa
, NULL
);
7272 zio_resume_wait(spa
);
7276 VERIFY(cyclic_reprogram(spa
->spa_deadman_cycid
, CY_INFINITY
));
7279 * Clear the dirty config list.
7281 while ((vd
= list_head(&spa
->spa_config_dirty_list
)) != NULL
)
7282 vdev_config_clean(vd
);
7285 * Now that the new config has synced transactionally,
7286 * let it become visible to the config cache.
7288 if (spa
->spa_config_syncing
!= NULL
) {
7289 spa_config_set(spa
, spa
->spa_config_syncing
);
7290 spa
->spa_config_txg
= txg
;
7291 spa
->spa_config_syncing
= NULL
;
7294 dsl_pool_sync_done(dp
, txg
);
7296 mutex_enter(&spa
->spa_alloc_lock
);
7297 VERIFY0(avl_numnodes(&spa
->spa_alloc_tree
));
7298 mutex_exit(&spa
->spa_alloc_lock
);
7301 * Update usable space statistics.
7303 while (vd
= txg_list_remove(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)))
7304 vdev_sync_done(vd
, txg
);
7306 spa_update_dspace(spa
);
7309 * It had better be the case that we didn't dirty anything
7310 * since vdev_config_sync().
7312 ASSERT(txg_list_empty(&dp
->dp_dirty_datasets
, txg
));
7313 ASSERT(txg_list_empty(&dp
->dp_dirty_dirs
, txg
));
7314 ASSERT(txg_list_empty(&spa
->spa_vdev_txg_list
, txg
));
7316 spa
->spa_sync_pass
= 0;
7319 * Update the last synced uberblock here. We want to do this at
7320 * the end of spa_sync() so that consumers of spa_last_synced_txg()
7321 * will be guaranteed that all the processing associated with
7322 * that txg has been completed.
7324 spa
->spa_ubsync
= spa
->spa_uberblock
;
7325 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7327 spa_handle_ignored_writes(spa
);
7330 * If any async tasks have been requested, kick them off.
7332 spa_async_dispatch(spa
);
7336 * Sync all pools. We don't want to hold the namespace lock across these
7337 * operations, so we take a reference on the spa_t and drop the lock during the
7341 spa_sync_allpools(void)
7344 mutex_enter(&spa_namespace_lock
);
7345 while ((spa
= spa_next(spa
)) != NULL
) {
7346 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
7347 !spa_writeable(spa
) || spa_suspended(spa
))
7349 spa_open_ref(spa
, FTAG
);
7350 mutex_exit(&spa_namespace_lock
);
7351 txg_wait_synced(spa_get_dsl(spa
), 0);
7352 mutex_enter(&spa_namespace_lock
);
7353 spa_close(spa
, FTAG
);
7355 mutex_exit(&spa_namespace_lock
);
7359 * ==========================================================================
7360 * Miscellaneous routines
7361 * ==========================================================================
7365 * Remove all pools in the system.
7373 * Remove all cached state. All pools should be closed now,
7374 * so every spa in the AVL tree should be unreferenced.
7376 mutex_enter(&spa_namespace_lock
);
7377 while ((spa
= spa_next(NULL
)) != NULL
) {
7379 * Stop async tasks. The async thread may need to detach
7380 * a device that's been replaced, which requires grabbing
7381 * spa_namespace_lock, so we must drop it here.
7383 spa_open_ref(spa
, FTAG
);
7384 mutex_exit(&spa_namespace_lock
);
7385 spa_async_suspend(spa
);
7386 mutex_enter(&spa_namespace_lock
);
7387 spa_close(spa
, FTAG
);
7389 if (spa
->spa_state
!= POOL_STATE_UNINITIALIZED
) {
7391 spa_deactivate(spa
);
7395 mutex_exit(&spa_namespace_lock
);
7399 spa_lookup_by_guid(spa_t
*spa
, uint64_t guid
, boolean_t aux
)
7404 if ((vd
= vdev_lookup_by_guid(spa
->spa_root_vdev
, guid
)) != NULL
)
7408 for (i
= 0; i
< spa
->spa_l2cache
.sav_count
; i
++) {
7409 vd
= spa
->spa_l2cache
.sav_vdevs
[i
];
7410 if (vd
->vdev_guid
== guid
)
7414 for (i
= 0; i
< spa
->spa_spares
.sav_count
; i
++) {
7415 vd
= spa
->spa_spares
.sav_vdevs
[i
];
7416 if (vd
->vdev_guid
== guid
)
7425 spa_upgrade(spa_t
*spa
, uint64_t version
)
7427 ASSERT(spa_writeable(spa
));
7429 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
7432 * This should only be called for a non-faulted pool, and since a
7433 * future version would result in an unopenable pool, this shouldn't be
7436 ASSERT(SPA_VERSION_IS_SUPPORTED(spa
->spa_uberblock
.ub_version
));
7437 ASSERT3U(version
, >=, spa
->spa_uberblock
.ub_version
);
7439 spa
->spa_uberblock
.ub_version
= version
;
7440 vdev_config_dirty(spa
->spa_root_vdev
);
7442 spa_config_exit(spa
, SCL_ALL
, FTAG
);
7444 txg_wait_synced(spa_get_dsl(spa
), 0);
7448 spa_has_spare(spa_t
*spa
, uint64_t guid
)
7452 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7454 for (i
= 0; i
< sav
->sav_count
; i
++)
7455 if (sav
->sav_vdevs
[i
]->vdev_guid
== guid
)
7458 for (i
= 0; i
< sav
->sav_npending
; i
++) {
7459 if (nvlist_lookup_uint64(sav
->sav_pending
[i
], ZPOOL_CONFIG_GUID
,
7460 &spareguid
) == 0 && spareguid
== guid
)
7468 * Check if a pool has an active shared spare device.
7469 * Note: reference count of an active spare is 2, as a spare and as a replace
7472 spa_has_active_shared_spare(spa_t
*spa
)
7476 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
7478 for (i
= 0; i
< sav
->sav_count
; i
++) {
7479 if (spa_spare_exists(sav
->sav_vdevs
[i
]->vdev_guid
, &pool
,
7480 &refcnt
) && pool
!= 0ULL && pool
== spa_guid(spa
) &&
7489 spa_event_create(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7491 sysevent_t
*ev
= NULL
;
7493 sysevent_attr_list_t
*attr
= NULL
;
7494 sysevent_value_t value
;
7496 ev
= sysevent_alloc(EC_ZFS
, (char *)name
, SUNW_KERN_PUB
"zfs",
7500 value
.value_type
= SE_DATA_TYPE_STRING
;
7501 value
.value
.sv_string
= spa_name(spa
);
7502 if (sysevent_add_attr(&attr
, ZFS_EV_POOL_NAME
, &value
, SE_SLEEP
) != 0)
7505 value
.value_type
= SE_DATA_TYPE_UINT64
;
7506 value
.value
.sv_uint64
= spa_guid(spa
);
7507 if (sysevent_add_attr(&attr
, ZFS_EV_POOL_GUID
, &value
, SE_SLEEP
) != 0)
7511 value
.value_type
= SE_DATA_TYPE_UINT64
;
7512 value
.value
.sv_uint64
= vd
->vdev_guid
;
7513 if (sysevent_add_attr(&attr
, ZFS_EV_VDEV_GUID
, &value
,
7517 if (vd
->vdev_path
) {
7518 value
.value_type
= SE_DATA_TYPE_STRING
;
7519 value
.value
.sv_string
= vd
->vdev_path
;
7520 if (sysevent_add_attr(&attr
, ZFS_EV_VDEV_PATH
,
7521 &value
, SE_SLEEP
) != 0)
7526 if (hist_nvl
!= NULL
) {
7527 fnvlist_merge((nvlist_t
*)attr
, hist_nvl
);
7530 if (sysevent_attach_attributes(ev
, attr
) != 0)
7536 sysevent_free_attr(attr
);
7543 spa_event_post(sysevent_t
*ev
)
7548 (void) log_sysevent(ev
, SE_SLEEP
, &eid
);
7554 spa_event_discard(sysevent_t
*ev
)
7562 * Post a sysevent corresponding to the given event. The 'name' must be one of
7563 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
7564 * filled in from the spa and (optionally) the vdev and history nvl. This
7565 * doesn't do anything in the userland libzpool, as we don't want consumers to
7566 * misinterpret ztest or zdb as real changes.
7569 spa_event_notify(spa_t
*spa
, vdev_t
*vd
, nvlist_t
*hist_nvl
, const char *name
)
7571 spa_event_post(spa_event_create(spa
, vd
, hist_nvl
, name
));