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, 2015 by Delphix. All rights reserved.
25 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/fm/fs/zfs.h>
31 #include <sys/spa_impl.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/uberblock_impl.h>
36 #include <sys/metaslab.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
39 #include <sys/space_reftree.h>
42 #include <sys/fs/zfs.h>
45 #include <sys/dsl_scan.h>
48 #include <sys/zfs_ratelimit.h>
51 * When a vdev is added, it will be divided into approximately (but no
52 * more than) this number of metaslabs.
54 int metaslabs_per_vdev
= 200;
57 * Virtual device management.
60 static vdev_ops_t
*vdev_ops_table
[] = {
74 * Given a vdev type, return the appropriate ops vector.
77 vdev_getops(const char *type
)
79 vdev_ops_t
*ops
, **opspp
;
81 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
82 if (strcmp(ops
->vdev_op_type
, type
) == 0)
89 * Default asize function: return the MAX of psize with the asize of
90 * all children. This is what's used by anything other than RAID-Z.
93 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
95 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
99 for (c
= 0; c
< vd
->vdev_children
; c
++) {
100 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
101 asize
= MAX(asize
, csize
);
108 * Get the minimum allocatable size. We define the allocatable size as
109 * the vdev's asize rounded to the nearest metaslab. This allows us to
110 * replace or attach devices which don't have the same physical size but
111 * can still satisfy the same number of allocations.
114 vdev_get_min_asize(vdev_t
*vd
)
116 vdev_t
*pvd
= vd
->vdev_parent
;
119 * If our parent is NULL (inactive spare or cache) or is the root,
120 * just return our own asize.
123 return (vd
->vdev_asize
);
126 * The top-level vdev just returns the allocatable size rounded
127 * to the nearest metaslab.
129 if (vd
== vd
->vdev_top
)
130 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
133 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
134 * so each child must provide at least 1/Nth of its asize.
136 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
137 return ((pvd
->vdev_min_asize
+ pvd
->vdev_children
- 1) /
140 return (pvd
->vdev_min_asize
);
144 vdev_set_min_asize(vdev_t
*vd
)
147 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
149 for (c
= 0; c
< vd
->vdev_children
; c
++)
150 vdev_set_min_asize(vd
->vdev_child
[c
]);
154 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
156 vdev_t
*rvd
= spa
->spa_root_vdev
;
158 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
160 if (vdev
< rvd
->vdev_children
) {
161 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
162 return (rvd
->vdev_child
[vdev
]);
169 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
174 if (vd
->vdev_guid
== guid
)
177 for (c
= 0; c
< vd
->vdev_children
; c
++)
178 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
186 vdev_count_leaves_impl(vdev_t
*vd
)
191 if (vd
->vdev_ops
->vdev_op_leaf
)
194 for (c
= 0; c
< vd
->vdev_children
; c
++)
195 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
201 vdev_count_leaves(spa_t
*spa
)
203 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
207 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
209 size_t oldsize
, newsize
;
210 uint64_t id
= cvd
->vdev_id
;
213 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
214 ASSERT(cvd
->vdev_parent
== NULL
);
216 cvd
->vdev_parent
= pvd
;
221 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
223 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
224 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
225 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
227 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
228 if (pvd
->vdev_child
!= NULL
) {
229 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
230 kmem_free(pvd
->vdev_child
, oldsize
);
233 pvd
->vdev_child
= newchild
;
234 pvd
->vdev_child
[id
] = cvd
;
236 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
237 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
240 * Walk up all ancestors to update guid sum.
242 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
243 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
247 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
250 uint_t id
= cvd
->vdev_id
;
252 ASSERT(cvd
->vdev_parent
== pvd
);
257 ASSERT(id
< pvd
->vdev_children
);
258 ASSERT(pvd
->vdev_child
[id
] == cvd
);
260 pvd
->vdev_child
[id
] = NULL
;
261 cvd
->vdev_parent
= NULL
;
263 for (c
= 0; c
< pvd
->vdev_children
; c
++)
264 if (pvd
->vdev_child
[c
])
267 if (c
== pvd
->vdev_children
) {
268 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
269 pvd
->vdev_child
= NULL
;
270 pvd
->vdev_children
= 0;
274 * Walk up all ancestors to update guid sum.
276 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
277 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
281 * Remove any holes in the child array.
284 vdev_compact_children(vdev_t
*pvd
)
286 vdev_t
**newchild
, *cvd
;
287 int oldc
= pvd
->vdev_children
;
291 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
293 for (c
= newc
= 0; c
< oldc
; c
++)
294 if (pvd
->vdev_child
[c
])
297 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
299 for (c
= newc
= 0; c
< oldc
; c
++) {
300 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
301 newchild
[newc
] = cvd
;
302 cvd
->vdev_id
= newc
++;
306 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
307 pvd
->vdev_child
= newchild
;
308 pvd
->vdev_children
= newc
;
312 * Allocate and minimally initialize a vdev_t.
315 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
320 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
322 if (spa
->spa_root_vdev
== NULL
) {
323 ASSERT(ops
== &vdev_root_ops
);
324 spa
->spa_root_vdev
= vd
;
325 spa
->spa_load_guid
= spa_generate_guid(NULL
);
328 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
329 if (spa
->spa_root_vdev
== vd
) {
331 * The root vdev's guid will also be the pool guid,
332 * which must be unique among all pools.
334 guid
= spa_generate_guid(NULL
);
337 * Any other vdev's guid must be unique within the pool.
339 guid
= spa_generate_guid(spa
);
341 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
346 vd
->vdev_guid
= guid
;
347 vd
->vdev_guid_sum
= guid
;
349 vd
->vdev_state
= VDEV_STATE_CLOSED
;
350 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
353 * Initialize rate limit structs for events. We rate limit ZIO delay
354 * and checksum events so that we don't overwhelm ZED with thousands
355 * of events when a disk is acting up.
357 zfs_ratelimit_init(&vd
->vdev_delay_rl
, DELAYS_PER_SECOND
, 1);
358 zfs_ratelimit_init(&vd
->vdev_checksum_rl
, CHECKSUMS_PER_SECOND
, 1);
360 list_link_init(&vd
->vdev_config_dirty_node
);
361 list_link_init(&vd
->vdev_state_dirty_node
);
362 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
363 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
364 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
365 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
367 for (t
= 0; t
< DTL_TYPES
; t
++) {
368 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
371 txg_list_create(&vd
->vdev_ms_list
,
372 offsetof(struct metaslab
, ms_txg_node
));
373 txg_list_create(&vd
->vdev_dtl_list
,
374 offsetof(struct vdev
, vdev_dtl_node
));
375 vd
->vdev_stat
.vs_timestamp
= gethrtime();
383 * Allocate a new vdev. The 'alloctype' is used to control whether we are
384 * creating a new vdev or loading an existing one - the behavior is slightly
385 * different for each case.
388 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
393 uint64_t guid
= 0, islog
, nparity
;
396 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
398 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
399 return (SET_ERROR(EINVAL
));
401 if ((ops
= vdev_getops(type
)) == NULL
)
402 return (SET_ERROR(EINVAL
));
405 * If this is a load, get the vdev guid from the nvlist.
406 * Otherwise, vdev_alloc_common() will generate one for us.
408 if (alloctype
== VDEV_ALLOC_LOAD
) {
411 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
413 return (SET_ERROR(EINVAL
));
415 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
416 return (SET_ERROR(EINVAL
));
417 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
418 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
419 return (SET_ERROR(EINVAL
));
420 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
421 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
422 return (SET_ERROR(EINVAL
));
423 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
424 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
425 return (SET_ERROR(EINVAL
));
429 * The first allocated vdev must be of type 'root'.
431 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
432 return (SET_ERROR(EINVAL
));
435 * Determine whether we're a log vdev.
438 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
439 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
440 return (SET_ERROR(ENOTSUP
));
442 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
443 return (SET_ERROR(ENOTSUP
));
446 * Set the nparity property for RAID-Z vdevs.
449 if (ops
== &vdev_raidz_ops
) {
450 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
452 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
453 return (SET_ERROR(EINVAL
));
455 * Previous versions could only support 1 or 2 parity
459 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
460 return (SET_ERROR(ENOTSUP
));
462 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
463 return (SET_ERROR(ENOTSUP
));
466 * We require the parity to be specified for SPAs that
467 * support multiple parity levels.
469 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
470 return (SET_ERROR(EINVAL
));
472 * Otherwise, we default to 1 parity device for RAID-Z.
479 ASSERT(nparity
!= -1ULL);
481 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
483 vd
->vdev_islog
= islog
;
484 vd
->vdev_nparity
= nparity
;
486 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
487 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
488 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
489 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
490 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
491 &vd
->vdev_physpath
) == 0)
492 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
494 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
495 &vd
->vdev_enc_sysfs_path
) == 0)
496 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
498 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
499 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
502 * Set the whole_disk property. If it's not specified, leave the value
505 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
506 &vd
->vdev_wholedisk
) != 0)
507 vd
->vdev_wholedisk
= -1ULL;
510 * Look for the 'not present' flag. This will only be set if the device
511 * was not present at the time of import.
513 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
514 &vd
->vdev_not_present
);
517 * Get the alignment requirement.
519 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
522 * Retrieve the vdev creation time.
524 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
528 * If we're a top-level vdev, try to load the allocation parameters.
530 if (parent
&& !parent
->vdev_parent
&&
531 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
532 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
534 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
536 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
538 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
540 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
543 ASSERT0(vd
->vdev_top_zap
);
546 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
547 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
548 alloctype
== VDEV_ALLOC_ADD
||
549 alloctype
== VDEV_ALLOC_SPLIT
||
550 alloctype
== VDEV_ALLOC_ROOTPOOL
);
551 vd
->vdev_mg
= metaslab_group_create(islog
?
552 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
555 if (vd
->vdev_ops
->vdev_op_leaf
&&
556 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
557 (void) nvlist_lookup_uint64(nv
,
558 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
560 ASSERT0(vd
->vdev_leaf_zap
);
564 * If we're a leaf vdev, try to load the DTL object and other state.
567 if (vd
->vdev_ops
->vdev_op_leaf
&&
568 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
569 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
570 if (alloctype
== VDEV_ALLOC_LOAD
) {
571 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
572 &vd
->vdev_dtl_object
);
573 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
577 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
580 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
581 &spare
) == 0 && spare
)
585 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
588 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
589 &vd
->vdev_resilver_txg
);
592 * When importing a pool, we want to ignore the persistent fault
593 * state, as the diagnosis made on another system may not be
594 * valid in the current context. Local vdevs will
595 * remain in the faulted state.
597 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
598 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
600 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
602 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
605 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
609 VDEV_AUX_ERR_EXCEEDED
;
610 if (nvlist_lookup_string(nv
,
611 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
612 strcmp(aux
, "external") == 0)
613 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
619 * Add ourselves to the parent's list of children.
621 vdev_add_child(parent
, vd
);
629 vdev_free(vdev_t
*vd
)
632 spa_t
*spa
= vd
->vdev_spa
;
635 * vdev_free() implies closing the vdev first. This is simpler than
636 * trying to ensure complicated semantics for all callers.
640 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
641 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
646 for (c
= 0; c
< vd
->vdev_children
; c
++)
647 vdev_free(vd
->vdev_child
[c
]);
649 ASSERT(vd
->vdev_child
== NULL
);
650 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
653 * Discard allocation state.
655 if (vd
->vdev_mg
!= NULL
) {
656 vdev_metaslab_fini(vd
);
657 metaslab_group_destroy(vd
->vdev_mg
);
660 ASSERT0(vd
->vdev_stat
.vs_space
);
661 ASSERT0(vd
->vdev_stat
.vs_dspace
);
662 ASSERT0(vd
->vdev_stat
.vs_alloc
);
665 * Remove this vdev from its parent's child list.
667 vdev_remove_child(vd
->vdev_parent
, vd
);
669 ASSERT(vd
->vdev_parent
== NULL
);
672 * Clean up vdev structure.
678 spa_strfree(vd
->vdev_path
);
680 spa_strfree(vd
->vdev_devid
);
681 if (vd
->vdev_physpath
)
682 spa_strfree(vd
->vdev_physpath
);
684 if (vd
->vdev_enc_sysfs_path
)
685 spa_strfree(vd
->vdev_enc_sysfs_path
);
688 spa_strfree(vd
->vdev_fru
);
690 if (vd
->vdev_isspare
)
691 spa_spare_remove(vd
);
692 if (vd
->vdev_isl2cache
)
693 spa_l2cache_remove(vd
);
695 txg_list_destroy(&vd
->vdev_ms_list
);
696 txg_list_destroy(&vd
->vdev_dtl_list
);
698 mutex_enter(&vd
->vdev_dtl_lock
);
699 space_map_close(vd
->vdev_dtl_sm
);
700 for (t
= 0; t
< DTL_TYPES
; t
++) {
701 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
702 range_tree_destroy(vd
->vdev_dtl
[t
]);
704 mutex_exit(&vd
->vdev_dtl_lock
);
706 mutex_destroy(&vd
->vdev_queue_lock
);
707 mutex_destroy(&vd
->vdev_dtl_lock
);
708 mutex_destroy(&vd
->vdev_stat_lock
);
709 mutex_destroy(&vd
->vdev_probe_lock
);
711 if (vd
== spa
->spa_root_vdev
)
712 spa
->spa_root_vdev
= NULL
;
714 kmem_free(vd
, sizeof (vdev_t
));
718 * Transfer top-level vdev state from svd to tvd.
721 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
723 spa_t
*spa
= svd
->vdev_spa
;
728 ASSERT(tvd
== tvd
->vdev_top
);
730 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
731 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
732 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
733 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
734 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
736 svd
->vdev_ms_array
= 0;
737 svd
->vdev_ms_shift
= 0;
738 svd
->vdev_ms_count
= 0;
739 svd
->vdev_top_zap
= 0;
742 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
743 tvd
->vdev_mg
= svd
->vdev_mg
;
744 tvd
->vdev_ms
= svd
->vdev_ms
;
749 if (tvd
->vdev_mg
!= NULL
)
750 tvd
->vdev_mg
->mg_vd
= tvd
;
752 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
753 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
754 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
756 svd
->vdev_stat
.vs_alloc
= 0;
757 svd
->vdev_stat
.vs_space
= 0;
758 svd
->vdev_stat
.vs_dspace
= 0;
760 for (t
= 0; t
< TXG_SIZE
; t
++) {
761 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
762 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
763 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
764 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
765 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
766 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
769 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
770 vdev_config_clean(svd
);
771 vdev_config_dirty(tvd
);
774 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
775 vdev_state_clean(svd
);
776 vdev_state_dirty(tvd
);
779 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
780 svd
->vdev_deflate_ratio
= 0;
782 tvd
->vdev_islog
= svd
->vdev_islog
;
787 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
796 for (c
= 0; c
< vd
->vdev_children
; c
++)
797 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
801 * Add a mirror/replacing vdev above an existing vdev.
804 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
806 spa_t
*spa
= cvd
->vdev_spa
;
807 vdev_t
*pvd
= cvd
->vdev_parent
;
810 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
812 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
814 mvd
->vdev_asize
= cvd
->vdev_asize
;
815 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
816 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
817 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
818 mvd
->vdev_state
= cvd
->vdev_state
;
819 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
821 vdev_remove_child(pvd
, cvd
);
822 vdev_add_child(pvd
, mvd
);
823 cvd
->vdev_id
= mvd
->vdev_children
;
824 vdev_add_child(mvd
, cvd
);
825 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
827 if (mvd
== mvd
->vdev_top
)
828 vdev_top_transfer(cvd
, mvd
);
834 * Remove a 1-way mirror/replacing vdev from the tree.
837 vdev_remove_parent(vdev_t
*cvd
)
839 vdev_t
*mvd
= cvd
->vdev_parent
;
840 vdev_t
*pvd
= mvd
->vdev_parent
;
842 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
844 ASSERT(mvd
->vdev_children
== 1);
845 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
846 mvd
->vdev_ops
== &vdev_replacing_ops
||
847 mvd
->vdev_ops
== &vdev_spare_ops
);
848 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
850 vdev_remove_child(mvd
, cvd
);
851 vdev_remove_child(pvd
, mvd
);
854 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
855 * Otherwise, we could have detached an offline device, and when we
856 * go to import the pool we'll think we have two top-level vdevs,
857 * instead of a different version of the same top-level vdev.
859 if (mvd
->vdev_top
== mvd
) {
860 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
861 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
862 cvd
->vdev_guid
+= guid_delta
;
863 cvd
->vdev_guid_sum
+= guid_delta
;
866 * If pool not set for autoexpand, we need to also preserve
867 * mvd's asize to prevent automatic expansion of cvd.
868 * Otherwise if we are adjusting the mirror by attaching and
869 * detaching children of non-uniform sizes, the mirror could
870 * autoexpand, unexpectedly requiring larger devices to
871 * re-establish the mirror.
873 if (!cvd
->vdev_spa
->spa_autoexpand
)
874 cvd
->vdev_asize
= mvd
->vdev_asize
;
876 cvd
->vdev_id
= mvd
->vdev_id
;
877 vdev_add_child(pvd
, cvd
);
878 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
880 if (cvd
== cvd
->vdev_top
)
881 vdev_top_transfer(mvd
, cvd
);
883 ASSERT(mvd
->vdev_children
== 0);
888 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
890 spa_t
*spa
= vd
->vdev_spa
;
891 objset_t
*mos
= spa
->spa_meta_objset
;
893 uint64_t oldc
= vd
->vdev_ms_count
;
894 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
898 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
901 * This vdev is not being allocated from yet or is a hole.
903 if (vd
->vdev_ms_shift
== 0)
906 ASSERT(!vd
->vdev_ishole
);
909 * Compute the raidz-deflation ratio. Note, we hard-code
910 * in 128k (1 << 17) because it is the "typical" blocksize.
911 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
912 * otherwise it would inconsistently account for existing bp's.
914 vd
->vdev_deflate_ratio
= (1 << 17) /
915 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
917 ASSERT(oldc
<= newc
);
919 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
922 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
923 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
927 vd
->vdev_ms_count
= newc
;
929 for (m
= oldc
; m
< newc
; m
++) {
933 error
= dmu_read(mos
, vd
->vdev_ms_array
,
934 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
940 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
947 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
950 * If the vdev is being removed we don't activate
951 * the metaslabs since we want to ensure that no new
952 * allocations are performed on this device.
954 if (oldc
== 0 && !vd
->vdev_removing
)
955 metaslab_group_activate(vd
->vdev_mg
);
958 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
964 vdev_metaslab_fini(vdev_t
*vd
)
967 uint64_t count
= vd
->vdev_ms_count
;
969 if (vd
->vdev_ms
!= NULL
) {
970 metaslab_group_passivate(vd
->vdev_mg
);
971 for (m
= 0; m
< count
; m
++) {
972 metaslab_t
*msp
= vd
->vdev_ms
[m
];
977 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
981 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
984 typedef struct vdev_probe_stats
{
985 boolean_t vps_readable
;
986 boolean_t vps_writeable
;
988 } vdev_probe_stats_t
;
991 vdev_probe_done(zio_t
*zio
)
993 spa_t
*spa
= zio
->io_spa
;
994 vdev_t
*vd
= zio
->io_vd
;
995 vdev_probe_stats_t
*vps
= zio
->io_private
;
997 ASSERT(vd
->vdev_probe_zio
!= NULL
);
999 if (zio
->io_type
== ZIO_TYPE_READ
) {
1000 if (zio
->io_error
== 0)
1001 vps
->vps_readable
= 1;
1002 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1003 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1004 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1005 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1006 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1008 abd_free(zio
->io_abd
);
1010 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1011 if (zio
->io_error
== 0)
1012 vps
->vps_writeable
= 1;
1013 abd_free(zio
->io_abd
);
1014 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1018 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1019 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1021 if (vdev_readable(vd
) &&
1022 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1025 ASSERT(zio
->io_error
!= 0);
1026 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1027 spa
, vd
, NULL
, 0, 0);
1028 zio
->io_error
= SET_ERROR(ENXIO
);
1031 mutex_enter(&vd
->vdev_probe_lock
);
1032 ASSERT(vd
->vdev_probe_zio
== zio
);
1033 vd
->vdev_probe_zio
= NULL
;
1034 mutex_exit(&vd
->vdev_probe_lock
);
1037 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1038 if (!vdev_accessible(vd
, pio
))
1039 pio
->io_error
= SET_ERROR(ENXIO
);
1041 kmem_free(vps
, sizeof (*vps
));
1046 * Determine whether this device is accessible.
1048 * Read and write to several known locations: the pad regions of each
1049 * vdev label but the first, which we leave alone in case it contains
1053 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1055 spa_t
*spa
= vd
->vdev_spa
;
1056 vdev_probe_stats_t
*vps
= NULL
;
1060 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1063 * Don't probe the probe.
1065 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1069 * To prevent 'probe storms' when a device fails, we create
1070 * just one probe i/o at a time. All zios that want to probe
1071 * this vdev will become parents of the probe io.
1073 mutex_enter(&vd
->vdev_probe_lock
);
1075 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1076 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1078 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1079 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1082 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1084 * vdev_cant_read and vdev_cant_write can only
1085 * transition from TRUE to FALSE when we have the
1086 * SCL_ZIO lock as writer; otherwise they can only
1087 * transition from FALSE to TRUE. This ensures that
1088 * any zio looking at these values can assume that
1089 * failures persist for the life of the I/O. That's
1090 * important because when a device has intermittent
1091 * connectivity problems, we want to ensure that
1092 * they're ascribed to the device (ENXIO) and not
1095 * Since we hold SCL_ZIO as writer here, clear both
1096 * values so the probe can reevaluate from first
1099 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1100 vd
->vdev_cant_read
= B_FALSE
;
1101 vd
->vdev_cant_write
= B_FALSE
;
1104 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1105 vdev_probe_done
, vps
,
1106 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1109 * We can't change the vdev state in this context, so we
1110 * kick off an async task to do it on our behalf.
1113 vd
->vdev_probe_wanted
= B_TRUE
;
1114 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1119 zio_add_child(zio
, pio
);
1121 mutex_exit(&vd
->vdev_probe_lock
);
1124 ASSERT(zio
!= NULL
);
1128 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1129 zio_nowait(zio_read_phys(pio
, vd
,
1130 vdev_label_offset(vd
->vdev_psize
, l
,
1131 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1132 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1133 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1134 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1145 vdev_open_child(void *arg
)
1149 vd
->vdev_open_thread
= curthread
;
1150 vd
->vdev_open_error
= vdev_open(vd
);
1151 vd
->vdev_open_thread
= NULL
;
1155 vdev_uses_zvols(vdev_t
*vd
)
1160 if (zvol_is_zvol(vd
->vdev_path
))
1164 for (c
= 0; c
< vd
->vdev_children
; c
++)
1165 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1172 vdev_open_children(vdev_t
*vd
)
1175 int children
= vd
->vdev_children
;
1179 * in order to handle pools on top of zvols, do the opens
1180 * in a single thread so that the same thread holds the
1181 * spa_namespace_lock
1183 if (vdev_uses_zvols(vd
)) {
1185 for (c
= 0; c
< children
; c
++)
1186 vd
->vdev_child
[c
]->vdev_open_error
=
1187 vdev_open(vd
->vdev_child
[c
]);
1189 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1190 children
, children
, TASKQ_PREPOPULATE
);
1194 for (c
= 0; c
< children
; c
++)
1195 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1196 vd
->vdev_child
[c
], TQ_SLEEP
) != TASKQID_INVALID
);
1201 vd
->vdev_nonrot
= B_TRUE
;
1203 for (c
= 0; c
< children
; c
++)
1204 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1208 * Prepare a virtual device for access.
1211 vdev_open(vdev_t
*vd
)
1213 spa_t
*spa
= vd
->vdev_spa
;
1216 uint64_t max_osize
= 0;
1217 uint64_t asize
, max_asize
, psize
;
1218 uint64_t ashift
= 0;
1221 ASSERT(vd
->vdev_open_thread
== curthread
||
1222 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1223 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1224 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1225 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1227 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1228 vd
->vdev_cant_read
= B_FALSE
;
1229 vd
->vdev_cant_write
= B_FALSE
;
1230 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1233 * If this vdev is not removed, check its fault status. If it's
1234 * faulted, bail out of the open.
1236 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1237 ASSERT(vd
->vdev_children
== 0);
1238 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1239 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1240 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1241 vd
->vdev_label_aux
);
1242 return (SET_ERROR(ENXIO
));
1243 } else if (vd
->vdev_offline
) {
1244 ASSERT(vd
->vdev_children
== 0);
1245 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1246 return (SET_ERROR(ENXIO
));
1249 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1252 * Reset the vdev_reopening flag so that we actually close
1253 * the vdev on error.
1255 vd
->vdev_reopening
= B_FALSE
;
1256 if (zio_injection_enabled
&& error
== 0)
1257 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1260 if (vd
->vdev_removed
&&
1261 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1262 vd
->vdev_removed
= B_FALSE
;
1264 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1265 vd
->vdev_stat
.vs_aux
);
1269 vd
->vdev_removed
= B_FALSE
;
1272 * Recheck the faulted flag now that we have confirmed that
1273 * the vdev is accessible. If we're faulted, bail.
1275 if (vd
->vdev_faulted
) {
1276 ASSERT(vd
->vdev_children
== 0);
1277 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1278 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1279 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1280 vd
->vdev_label_aux
);
1281 return (SET_ERROR(ENXIO
));
1284 if (vd
->vdev_degraded
) {
1285 ASSERT(vd
->vdev_children
== 0);
1286 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1287 VDEV_AUX_ERR_EXCEEDED
);
1289 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1293 * For hole or missing vdevs we just return success.
1295 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1298 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1299 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1300 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1306 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1307 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1309 if (vd
->vdev_children
== 0) {
1310 if (osize
< SPA_MINDEVSIZE
) {
1311 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1312 VDEV_AUX_TOO_SMALL
);
1313 return (SET_ERROR(EOVERFLOW
));
1316 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1317 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1318 VDEV_LABEL_END_SIZE
);
1320 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1321 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1322 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1323 VDEV_AUX_TOO_SMALL
);
1324 return (SET_ERROR(EOVERFLOW
));
1328 max_asize
= max_osize
;
1331 vd
->vdev_psize
= psize
;
1334 * Make sure the allocatable size hasn't shrunk too much.
1336 if (asize
< vd
->vdev_min_asize
) {
1337 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1338 VDEV_AUX_BAD_LABEL
);
1339 return (SET_ERROR(EINVAL
));
1342 if (vd
->vdev_asize
== 0) {
1344 * This is the first-ever open, so use the computed values.
1345 * For compatibility, a different ashift can be requested.
1347 vd
->vdev_asize
= asize
;
1348 vd
->vdev_max_asize
= max_asize
;
1349 if (vd
->vdev_ashift
== 0) {
1350 vd
->vdev_ashift
= ashift
; /* use detected value */
1352 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1353 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1354 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1355 VDEV_AUX_BAD_ASHIFT
);
1356 return (SET_ERROR(EDOM
));
1360 * Detect if the alignment requirement has increased.
1361 * We don't want to make the pool unavailable, just
1362 * post an event instead.
1364 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1365 vd
->vdev_ops
->vdev_op_leaf
) {
1366 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1367 spa
, vd
, NULL
, 0, 0);
1370 vd
->vdev_max_asize
= max_asize
;
1374 * If all children are healthy we update asize if either:
1375 * The asize has increased, due to a device expansion caused by dynamic
1376 * LUN growth or vdev replacement, and automatic expansion is enabled;
1377 * making the additional space available.
1379 * The asize has decreased, due to a device shrink usually caused by a
1380 * vdev replace with a smaller device. This ensures that calculations
1381 * based of max_asize and asize e.g. esize are always valid. It's safe
1382 * to do this as we've already validated that asize is greater than
1385 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1386 ((asize
> vd
->vdev_asize
&&
1387 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1388 (asize
< vd
->vdev_asize
)))
1389 vd
->vdev_asize
= asize
;
1391 vdev_set_min_asize(vd
);
1394 * Ensure we can issue some IO before declaring the
1395 * vdev open for business.
1397 if (vd
->vdev_ops
->vdev_op_leaf
&&
1398 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1399 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1400 VDEV_AUX_ERR_EXCEEDED
);
1405 * Track the min and max ashift values for normal data devices.
1407 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1408 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1409 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1410 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1411 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1412 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1416 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1417 * resilver. But don't do this if we are doing a reopen for a scrub,
1418 * since this would just restart the scrub we are already doing.
1420 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1421 vdev_resilver_needed(vd
, NULL
, NULL
))
1422 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1428 * Called once the vdevs are all opened, this routine validates the label
1429 * contents. This needs to be done before vdev_load() so that we don't
1430 * inadvertently do repair I/Os to the wrong device.
1432 * If 'strict' is false ignore the spa guid check. This is necessary because
1433 * if the machine crashed during a re-guid the new guid might have been written
1434 * to all of the vdev labels, but not the cached config. The strict check
1435 * will be performed when the pool is opened again using the mos config.
1437 * This function will only return failure if one of the vdevs indicates that it
1438 * has since been destroyed or exported. This is only possible if
1439 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1440 * will be updated but the function will return 0.
1443 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1445 spa_t
*spa
= vd
->vdev_spa
;
1447 uint64_t guid
= 0, top_guid
;
1451 for (c
= 0; c
< vd
->vdev_children
; c
++)
1452 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1453 return (SET_ERROR(EBADF
));
1456 * If the device has already failed, or was marked offline, don't do
1457 * any further validation. Otherwise, label I/O will fail and we will
1458 * overwrite the previous state.
1460 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1461 uint64_t aux_guid
= 0;
1463 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1464 spa_last_synced_txg(spa
) : -1ULL;
1466 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1467 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1468 VDEV_AUX_BAD_LABEL
);
1473 * Determine if this vdev has been split off into another
1474 * pool. If so, then refuse to open it.
1476 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1477 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1478 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1479 VDEV_AUX_SPLIT_POOL
);
1484 if (strict
&& (nvlist_lookup_uint64(label
,
1485 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1486 guid
!= spa_guid(spa
))) {
1487 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1488 VDEV_AUX_CORRUPT_DATA
);
1493 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1494 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1499 * If this vdev just became a top-level vdev because its
1500 * sibling was detached, it will have adopted the parent's
1501 * vdev guid -- but the label may or may not be on disk yet.
1502 * Fortunately, either version of the label will have the
1503 * same top guid, so if we're a top-level vdev, we can
1504 * safely compare to that instead.
1506 * If we split this vdev off instead, then we also check the
1507 * original pool's guid. We don't want to consider the vdev
1508 * corrupt if it is partway through a split operation.
1510 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1512 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1514 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1515 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1516 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1517 VDEV_AUX_CORRUPT_DATA
);
1522 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1524 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1525 VDEV_AUX_CORRUPT_DATA
);
1533 * If this is a verbatim import, no need to check the
1534 * state of the pool.
1536 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1537 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1538 state
!= POOL_STATE_ACTIVE
)
1539 return (SET_ERROR(EBADF
));
1542 * If we were able to open and validate a vdev that was
1543 * previously marked permanently unavailable, clear that state
1546 if (vd
->vdev_not_present
)
1547 vd
->vdev_not_present
= 0;
1554 * Close a virtual device.
1557 vdev_close(vdev_t
*vd
)
1559 vdev_t
*pvd
= vd
->vdev_parent
;
1560 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1562 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1565 * If our parent is reopening, then we are as well, unless we are
1568 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1569 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1571 vd
->vdev_ops
->vdev_op_close(vd
);
1573 vdev_cache_purge(vd
);
1576 * We record the previous state before we close it, so that if we are
1577 * doing a reopen(), we don't generate FMA ereports if we notice that
1578 * it's still faulted.
1580 vd
->vdev_prevstate
= vd
->vdev_state
;
1582 if (vd
->vdev_offline
)
1583 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1585 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1586 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1590 vdev_hold(vdev_t
*vd
)
1592 spa_t
*spa
= vd
->vdev_spa
;
1595 ASSERT(spa_is_root(spa
));
1596 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1599 for (c
= 0; c
< vd
->vdev_children
; c
++)
1600 vdev_hold(vd
->vdev_child
[c
]);
1602 if (vd
->vdev_ops
->vdev_op_leaf
)
1603 vd
->vdev_ops
->vdev_op_hold(vd
);
1607 vdev_rele(vdev_t
*vd
)
1611 ASSERT(spa_is_root(vd
->vdev_spa
));
1612 for (c
= 0; c
< vd
->vdev_children
; c
++)
1613 vdev_rele(vd
->vdev_child
[c
]);
1615 if (vd
->vdev_ops
->vdev_op_leaf
)
1616 vd
->vdev_ops
->vdev_op_rele(vd
);
1620 * Reopen all interior vdevs and any unopened leaves. We don't actually
1621 * reopen leaf vdevs which had previously been opened as they might deadlock
1622 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1623 * If the leaf has never been opened then open it, as usual.
1626 vdev_reopen(vdev_t
*vd
)
1628 spa_t
*spa
= vd
->vdev_spa
;
1630 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1632 /* set the reopening flag unless we're taking the vdev offline */
1633 vd
->vdev_reopening
= !vd
->vdev_offline
;
1635 (void) vdev_open(vd
);
1638 * Call vdev_validate() here to make sure we have the same device.
1639 * Otherwise, a device with an invalid label could be successfully
1640 * opened in response to vdev_reopen().
1643 (void) vdev_validate_aux(vd
);
1644 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1645 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1646 !l2arc_vdev_present(vd
))
1647 l2arc_add_vdev(spa
, vd
);
1649 (void) vdev_validate(vd
, B_TRUE
);
1653 * Reassess parent vdev's health.
1655 vdev_propagate_state(vd
);
1659 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1664 * Normally, partial opens (e.g. of a mirror) are allowed.
1665 * For a create, however, we want to fail the request if
1666 * there are any components we can't open.
1668 error
= vdev_open(vd
);
1670 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1672 return (error
? error
: ENXIO
);
1676 * Recursively load DTLs and initialize all labels.
1678 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1679 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1680 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1689 vdev_metaslab_set_size(vdev_t
*vd
)
1692 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1694 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1695 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1699 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1701 ASSERT(vd
== vd
->vdev_top
);
1702 ASSERT(!vd
->vdev_ishole
);
1703 ASSERT(ISP2(flags
));
1704 ASSERT(spa_writeable(vd
->vdev_spa
));
1706 if (flags
& VDD_METASLAB
)
1707 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1709 if (flags
& VDD_DTL
)
1710 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1712 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1716 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1720 for (c
= 0; c
< vd
->vdev_children
; c
++)
1721 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1723 if (vd
->vdev_ops
->vdev_op_leaf
)
1724 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1730 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1731 * the vdev has less than perfect replication. There are four kinds of DTL:
1733 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1735 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1737 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1738 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1739 * txgs that was scrubbed.
1741 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1742 * persistent errors or just some device being offline.
1743 * Unlike the other three, the DTL_OUTAGE map is not generally
1744 * maintained; it's only computed when needed, typically to
1745 * determine whether a device can be detached.
1747 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1748 * either has the data or it doesn't.
1750 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1751 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1752 * if any child is less than fully replicated, then so is its parent.
1753 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1754 * comprising only those txgs which appear in 'maxfaults' or more children;
1755 * those are the txgs we don't have enough replication to read. For example,
1756 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1757 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1758 * two child DTL_MISSING maps.
1760 * It should be clear from the above that to compute the DTLs and outage maps
1761 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1762 * Therefore, that is all we keep on disk. When loading the pool, or after
1763 * a configuration change, we generate all other DTLs from first principles.
1766 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1768 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1770 ASSERT(t
< DTL_TYPES
);
1771 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1772 ASSERT(spa_writeable(vd
->vdev_spa
));
1774 mutex_enter(rt
->rt_lock
);
1775 if (!range_tree_contains(rt
, txg
, size
))
1776 range_tree_add(rt
, txg
, size
);
1777 mutex_exit(rt
->rt_lock
);
1781 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1783 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1784 boolean_t dirty
= B_FALSE
;
1786 ASSERT(t
< DTL_TYPES
);
1787 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1789 mutex_enter(rt
->rt_lock
);
1790 if (range_tree_space(rt
) != 0)
1791 dirty
= range_tree_contains(rt
, txg
, size
);
1792 mutex_exit(rt
->rt_lock
);
1798 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1800 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1803 mutex_enter(rt
->rt_lock
);
1804 empty
= (range_tree_space(rt
) == 0);
1805 mutex_exit(rt
->rt_lock
);
1811 * Returns the lowest txg in the DTL range.
1814 vdev_dtl_min(vdev_t
*vd
)
1818 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1819 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1820 ASSERT0(vd
->vdev_children
);
1822 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1823 return (rs
->rs_start
- 1);
1827 * Returns the highest txg in the DTL.
1830 vdev_dtl_max(vdev_t
*vd
)
1834 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1835 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1836 ASSERT0(vd
->vdev_children
);
1838 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1839 return (rs
->rs_end
);
1843 * Determine if a resilvering vdev should remove any DTL entries from
1844 * its range. If the vdev was resilvering for the entire duration of the
1845 * scan then it should excise that range from its DTLs. Otherwise, this
1846 * vdev is considered partially resilvered and should leave its DTL
1847 * entries intact. The comment in vdev_dtl_reassess() describes how we
1851 vdev_dtl_should_excise(vdev_t
*vd
)
1853 spa_t
*spa
= vd
->vdev_spa
;
1854 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1856 ASSERT0(scn
->scn_phys
.scn_errors
);
1857 ASSERT0(vd
->vdev_children
);
1859 if (vd
->vdev_resilver_txg
== 0 ||
1860 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1864 * When a resilver is initiated the scan will assign the scn_max_txg
1865 * value to the highest txg value that exists in all DTLs. If this
1866 * device's max DTL is not part of this scan (i.e. it is not in
1867 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1870 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1871 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1872 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1873 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1880 * Reassess DTLs after a config change or scrub completion.
1883 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1885 spa_t
*spa
= vd
->vdev_spa
;
1889 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1891 for (c
= 0; c
< vd
->vdev_children
; c
++)
1892 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1893 scrub_txg
, scrub_done
);
1895 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1898 if (vd
->vdev_ops
->vdev_op_leaf
) {
1899 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1901 mutex_enter(&vd
->vdev_dtl_lock
);
1904 * If we've completed a scan cleanly then determine
1905 * if this vdev should remove any DTLs. We only want to
1906 * excise regions on vdevs that were available during
1907 * the entire duration of this scan.
1909 if (scrub_txg
!= 0 &&
1910 (spa
->spa_scrub_started
||
1911 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1912 vdev_dtl_should_excise(vd
)) {
1914 * We completed a scrub up to scrub_txg. If we
1915 * did it without rebooting, then the scrub dtl
1916 * will be valid, so excise the old region and
1917 * fold in the scrub dtl. Otherwise, leave the
1918 * dtl as-is if there was an error.
1920 * There's little trick here: to excise the beginning
1921 * of the DTL_MISSING map, we put it into a reference
1922 * tree and then add a segment with refcnt -1 that
1923 * covers the range [0, scrub_txg). This means
1924 * that each txg in that range has refcnt -1 or 0.
1925 * We then add DTL_SCRUB with a refcnt of 2, so that
1926 * entries in the range [0, scrub_txg) will have a
1927 * positive refcnt -- either 1 or 2. We then convert
1928 * the reference tree into the new DTL_MISSING map.
1930 space_reftree_create(&reftree
);
1931 space_reftree_add_map(&reftree
,
1932 vd
->vdev_dtl
[DTL_MISSING
], 1);
1933 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1934 space_reftree_add_map(&reftree
,
1935 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1936 space_reftree_generate_map(&reftree
,
1937 vd
->vdev_dtl
[DTL_MISSING
], 1);
1938 space_reftree_destroy(&reftree
);
1940 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1941 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1942 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1944 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1945 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1946 if (!vdev_readable(vd
))
1947 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1949 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1950 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1953 * If the vdev was resilvering and no longer has any
1954 * DTLs then reset its resilvering flag and dirty
1955 * the top level so that we persist the change.
1957 if (vd
->vdev_resilver_txg
!= 0 &&
1958 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1959 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
1960 vd
->vdev_resilver_txg
= 0;
1961 vdev_config_dirty(vd
->vdev_top
);
1964 mutex_exit(&vd
->vdev_dtl_lock
);
1967 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1971 mutex_enter(&vd
->vdev_dtl_lock
);
1972 for (t
= 0; t
< DTL_TYPES
; t
++) {
1975 /* account for child's outage in parent's missing map */
1976 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1978 continue; /* leaf vdevs only */
1979 if (t
== DTL_PARTIAL
)
1980 minref
= 1; /* i.e. non-zero */
1981 else if (vd
->vdev_nparity
!= 0)
1982 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1984 minref
= vd
->vdev_children
; /* any kind of mirror */
1985 space_reftree_create(&reftree
);
1986 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1987 vdev_t
*cvd
= vd
->vdev_child
[c
];
1988 mutex_enter(&cvd
->vdev_dtl_lock
);
1989 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1990 mutex_exit(&cvd
->vdev_dtl_lock
);
1992 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1993 space_reftree_destroy(&reftree
);
1995 mutex_exit(&vd
->vdev_dtl_lock
);
1999 vdev_dtl_load(vdev_t
*vd
)
2001 spa_t
*spa
= vd
->vdev_spa
;
2002 objset_t
*mos
= spa
->spa_meta_objset
;
2006 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2007 ASSERT(!vd
->vdev_ishole
);
2009 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2010 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
2013 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2015 mutex_enter(&vd
->vdev_dtl_lock
);
2018 * Now that we've opened the space_map we need to update
2021 space_map_update(vd
->vdev_dtl_sm
);
2023 error
= space_map_load(vd
->vdev_dtl_sm
,
2024 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2025 mutex_exit(&vd
->vdev_dtl_lock
);
2030 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2031 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2040 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2042 spa_t
*spa
= vd
->vdev_spa
;
2044 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2045 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2050 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2052 spa_t
*spa
= vd
->vdev_spa
;
2053 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2054 DMU_OT_NONE
, 0, tx
);
2057 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2064 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2068 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2069 vd
->vdev_ops
!= &vdev_missing_ops
&&
2070 vd
->vdev_ops
!= &vdev_root_ops
&&
2071 !vd
->vdev_top
->vdev_removing
) {
2072 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2073 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2075 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2076 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2079 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2080 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2085 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2087 spa_t
*spa
= vd
->vdev_spa
;
2088 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2089 objset_t
*mos
= spa
->spa_meta_objset
;
2090 range_tree_t
*rtsync
;
2093 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2095 ASSERT(!vd
->vdev_ishole
);
2096 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2098 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2100 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2101 mutex_enter(&vd
->vdev_dtl_lock
);
2102 space_map_free(vd
->vdev_dtl_sm
, tx
);
2103 space_map_close(vd
->vdev_dtl_sm
);
2104 vd
->vdev_dtl_sm
= NULL
;
2105 mutex_exit(&vd
->vdev_dtl_lock
);
2108 * We only destroy the leaf ZAP for detached leaves or for
2109 * removed log devices. Removed data devices handle leaf ZAP
2110 * cleanup later, once cancellation is no longer possible.
2112 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2113 vd
->vdev_top
->vdev_islog
)) {
2114 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2115 vd
->vdev_leaf_zap
= 0;
2122 if (vd
->vdev_dtl_sm
== NULL
) {
2123 uint64_t new_object
;
2125 new_object
= space_map_alloc(mos
, tx
);
2126 VERIFY3U(new_object
, !=, 0);
2128 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2129 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2130 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2133 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2135 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2137 mutex_enter(&rtlock
);
2139 mutex_enter(&vd
->vdev_dtl_lock
);
2140 range_tree_walk(rt
, range_tree_add
, rtsync
);
2141 mutex_exit(&vd
->vdev_dtl_lock
);
2143 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2144 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2145 range_tree_vacate(rtsync
, NULL
, NULL
);
2147 range_tree_destroy(rtsync
);
2149 mutex_exit(&rtlock
);
2150 mutex_destroy(&rtlock
);
2153 * If the object for the space map has changed then dirty
2154 * the top level so that we update the config.
2156 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2157 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2158 "new object %llu", txg
, spa_name(spa
), object
,
2159 space_map_object(vd
->vdev_dtl_sm
));
2160 vdev_config_dirty(vd
->vdev_top
);
2165 mutex_enter(&vd
->vdev_dtl_lock
);
2166 space_map_update(vd
->vdev_dtl_sm
);
2167 mutex_exit(&vd
->vdev_dtl_lock
);
2171 * Determine whether the specified vdev can be offlined/detached/removed
2172 * without losing data.
2175 vdev_dtl_required(vdev_t
*vd
)
2177 spa_t
*spa
= vd
->vdev_spa
;
2178 vdev_t
*tvd
= vd
->vdev_top
;
2179 uint8_t cant_read
= vd
->vdev_cant_read
;
2182 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2184 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2188 * Temporarily mark the device as unreadable, and then determine
2189 * whether this results in any DTL outages in the top-level vdev.
2190 * If not, we can safely offline/detach/remove the device.
2192 vd
->vdev_cant_read
= B_TRUE
;
2193 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2194 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2195 vd
->vdev_cant_read
= cant_read
;
2196 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2198 if (!required
&& zio_injection_enabled
)
2199 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2205 * Determine if resilver is needed, and if so the txg range.
2208 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2210 boolean_t needed
= B_FALSE
;
2211 uint64_t thismin
= UINT64_MAX
;
2212 uint64_t thismax
= 0;
2215 if (vd
->vdev_children
== 0) {
2216 mutex_enter(&vd
->vdev_dtl_lock
);
2217 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2218 vdev_writeable(vd
)) {
2220 thismin
= vdev_dtl_min(vd
);
2221 thismax
= vdev_dtl_max(vd
);
2224 mutex_exit(&vd
->vdev_dtl_lock
);
2226 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2227 vdev_t
*cvd
= vd
->vdev_child
[c
];
2228 uint64_t cmin
, cmax
;
2230 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2231 thismin
= MIN(thismin
, cmin
);
2232 thismax
= MAX(thismax
, cmax
);
2238 if (needed
&& minp
) {
2246 vdev_load(vdev_t
*vd
)
2251 * Recursively load all children.
2253 for (c
= 0; c
< vd
->vdev_children
; c
++)
2254 vdev_load(vd
->vdev_child
[c
]);
2257 * If this is a top-level vdev, initialize its metaslabs.
2259 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2260 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2261 vdev_metaslab_init(vd
, 0) != 0))
2262 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2263 VDEV_AUX_CORRUPT_DATA
);
2265 * If this is a leaf vdev, load its DTL.
2267 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2268 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2269 VDEV_AUX_CORRUPT_DATA
);
2273 * The special vdev case is used for hot spares and l2cache devices. Its
2274 * sole purpose it to set the vdev state for the associated vdev. To do this,
2275 * we make sure that we can open the underlying device, then try to read the
2276 * label, and make sure that the label is sane and that it hasn't been
2277 * repurposed to another pool.
2280 vdev_validate_aux(vdev_t
*vd
)
2283 uint64_t guid
, version
;
2286 if (!vdev_readable(vd
))
2289 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2290 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2291 VDEV_AUX_CORRUPT_DATA
);
2295 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2296 !SPA_VERSION_IS_SUPPORTED(version
) ||
2297 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2298 guid
!= vd
->vdev_guid
||
2299 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2300 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2301 VDEV_AUX_CORRUPT_DATA
);
2307 * We don't actually check the pool state here. If it's in fact in
2308 * use by another pool, we update this fact on the fly when requested.
2315 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2317 spa_t
*spa
= vd
->vdev_spa
;
2318 objset_t
*mos
= spa
->spa_meta_objset
;
2322 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2323 ASSERT(vd
== vd
->vdev_top
);
2324 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2326 if (vd
->vdev_ms
!= NULL
) {
2327 metaslab_group_t
*mg
= vd
->vdev_mg
;
2329 metaslab_group_histogram_verify(mg
);
2330 metaslab_class_histogram_verify(mg
->mg_class
);
2332 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2333 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2335 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2338 mutex_enter(&msp
->ms_lock
);
2340 * If the metaslab was not loaded when the vdev
2341 * was removed then the histogram accounting may
2342 * not be accurate. Update the histogram information
2343 * here so that we ensure that the metaslab group
2344 * and metaslab class are up-to-date.
2346 metaslab_group_histogram_remove(mg
, msp
);
2348 VERIFY0(space_map_allocated(msp
->ms_sm
));
2349 space_map_free(msp
->ms_sm
, tx
);
2350 space_map_close(msp
->ms_sm
);
2352 mutex_exit(&msp
->ms_lock
);
2355 metaslab_group_histogram_verify(mg
);
2356 metaslab_class_histogram_verify(mg
->mg_class
);
2357 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2358 ASSERT0(mg
->mg_histogram
[i
]);
2362 if (vd
->vdev_ms_array
) {
2363 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2364 vd
->vdev_ms_array
= 0;
2367 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2368 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2369 vd
->vdev_top_zap
= 0;
2375 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2378 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2380 ASSERT(!vd
->vdev_ishole
);
2382 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2383 metaslab_sync_done(msp
, txg
);
2386 metaslab_sync_reassess(vd
->vdev_mg
);
2390 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2392 spa_t
*spa
= vd
->vdev_spa
;
2397 ASSERT(!vd
->vdev_ishole
);
2399 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2400 ASSERT(vd
== vd
->vdev_top
);
2401 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2402 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2403 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2404 ASSERT(vd
->vdev_ms_array
!= 0);
2405 vdev_config_dirty(vd
);
2410 * Remove the metadata associated with this vdev once it's empty.
2412 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2413 vdev_remove(vd
, txg
);
2415 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2416 metaslab_sync(msp
, txg
);
2417 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2420 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2421 vdev_dtl_sync(lvd
, txg
);
2423 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2427 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2429 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2433 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2434 * not be opened, and no I/O is attempted.
2437 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2441 spa_vdev_state_enter(spa
, SCL_NONE
);
2443 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2444 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2446 if (!vd
->vdev_ops
->vdev_op_leaf
)
2447 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2452 * We don't directly use the aux state here, but if we do a
2453 * vdev_reopen(), we need this value to be present to remember why we
2456 vd
->vdev_label_aux
= aux
;
2459 * Faulted state takes precedence over degraded.
2461 vd
->vdev_delayed_close
= B_FALSE
;
2462 vd
->vdev_faulted
= 1ULL;
2463 vd
->vdev_degraded
= 0ULL;
2464 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2467 * If this device has the only valid copy of the data, then
2468 * back off and simply mark the vdev as degraded instead.
2470 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2471 vd
->vdev_degraded
= 1ULL;
2472 vd
->vdev_faulted
= 0ULL;
2475 * If we reopen the device and it's not dead, only then do we
2480 if (vdev_readable(vd
))
2481 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2484 return (spa_vdev_state_exit(spa
, vd
, 0));
2488 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2489 * user that something is wrong. The vdev continues to operate as normal as far
2490 * as I/O is concerned.
2493 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2497 spa_vdev_state_enter(spa
, SCL_NONE
);
2499 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2500 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2502 if (!vd
->vdev_ops
->vdev_op_leaf
)
2503 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2506 * If the vdev is already faulted, then don't do anything.
2508 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2509 return (spa_vdev_state_exit(spa
, NULL
, 0));
2511 vd
->vdev_degraded
= 1ULL;
2512 if (!vdev_is_dead(vd
))
2513 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2516 return (spa_vdev_state_exit(spa
, vd
, 0));
2520 * Online the given vdev.
2522 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2523 * spare device should be detached when the device finishes resilvering.
2524 * Second, the online should be treated like a 'test' online case, so no FMA
2525 * events are generated if the device fails to open.
2528 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2530 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2531 boolean_t postevent
= B_FALSE
;
2533 spa_vdev_state_enter(spa
, SCL_NONE
);
2535 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2536 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2538 if (!vd
->vdev_ops
->vdev_op_leaf
)
2539 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2542 (vd
->vdev_offline
== B_TRUE
|| vd
->vdev_tmpoffline
== B_TRUE
) ?
2546 vd
->vdev_offline
= B_FALSE
;
2547 vd
->vdev_tmpoffline
= B_FALSE
;
2548 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2549 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2551 /* XXX - L2ARC 1.0 does not support expansion */
2552 if (!vd
->vdev_aux
) {
2553 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2554 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2558 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2560 if (!vd
->vdev_aux
) {
2561 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2562 pvd
->vdev_expanding
= B_FALSE
;
2566 *newstate
= vd
->vdev_state
;
2567 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2568 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2569 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2570 vd
->vdev_parent
->vdev_child
[0] == vd
)
2571 vd
->vdev_unspare
= B_TRUE
;
2573 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2575 /* XXX - L2ARC 1.0 does not support expansion */
2577 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2578 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2582 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2584 return (spa_vdev_state_exit(spa
, vd
, 0));
2588 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2592 uint64_t generation
;
2593 metaslab_group_t
*mg
;
2596 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2598 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2599 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2601 if (!vd
->vdev_ops
->vdev_op_leaf
)
2602 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2606 generation
= spa
->spa_config_generation
+ 1;
2609 * If the device isn't already offline, try to offline it.
2611 if (!vd
->vdev_offline
) {
2613 * If this device has the only valid copy of some data,
2614 * don't allow it to be offlined. Log devices are always
2617 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2618 vdev_dtl_required(vd
))
2619 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2622 * If the top-level is a slog and it has had allocations
2623 * then proceed. We check that the vdev's metaslab group
2624 * is not NULL since it's possible that we may have just
2625 * added this vdev but not yet initialized its metaslabs.
2627 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2629 * Prevent any future allocations.
2631 metaslab_group_passivate(mg
);
2632 (void) spa_vdev_state_exit(spa
, vd
, 0);
2634 error
= spa_offline_log(spa
);
2636 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2639 * Check to see if the config has changed.
2641 if (error
|| generation
!= spa
->spa_config_generation
) {
2642 metaslab_group_activate(mg
);
2644 return (spa_vdev_state_exit(spa
,
2646 (void) spa_vdev_state_exit(spa
, vd
, 0);
2649 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2653 * Offline this device and reopen its top-level vdev.
2654 * If the top-level vdev is a log device then just offline
2655 * it. Otherwise, if this action results in the top-level
2656 * vdev becoming unusable, undo it and fail the request.
2658 vd
->vdev_offline
= B_TRUE
;
2661 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2662 vdev_is_dead(tvd
)) {
2663 vd
->vdev_offline
= B_FALSE
;
2665 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2669 * Add the device back into the metaslab rotor so that
2670 * once we online the device it's open for business.
2672 if (tvd
->vdev_islog
&& mg
!= NULL
)
2673 metaslab_group_activate(mg
);
2676 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2678 return (spa_vdev_state_exit(spa
, vd
, 0));
2682 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2686 mutex_enter(&spa
->spa_vdev_top_lock
);
2687 error
= vdev_offline_locked(spa
, guid
, flags
);
2688 mutex_exit(&spa
->spa_vdev_top_lock
);
2694 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2695 * vdev_offline(), we assume the spa config is locked. We also clear all
2696 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2699 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2701 vdev_t
*rvd
= spa
->spa_root_vdev
;
2704 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2709 vd
->vdev_stat
.vs_read_errors
= 0;
2710 vd
->vdev_stat
.vs_write_errors
= 0;
2711 vd
->vdev_stat
.vs_checksum_errors
= 0;
2713 for (c
= 0; c
< vd
->vdev_children
; c
++)
2714 vdev_clear(spa
, vd
->vdev_child
[c
]);
2717 * If we're in the FAULTED state or have experienced failed I/O, then
2718 * clear the persistent state and attempt to reopen the device. We
2719 * also mark the vdev config dirty, so that the new faulted state is
2720 * written out to disk.
2722 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2723 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2726 * When reopening in response to a clear event, it may be due to
2727 * a fmadm repair request. In this case, if the device is
2728 * still broken, we want to still post the ereport again.
2730 vd
->vdev_forcefault
= B_TRUE
;
2732 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2733 vd
->vdev_cant_read
= B_FALSE
;
2734 vd
->vdev_cant_write
= B_FALSE
;
2736 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2738 vd
->vdev_forcefault
= B_FALSE
;
2740 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2741 vdev_state_dirty(vd
->vdev_top
);
2743 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2744 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2746 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2750 * When clearing a FMA-diagnosed fault, we always want to
2751 * unspare the device, as we assume that the original spare was
2752 * done in response to the FMA fault.
2754 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2755 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2756 vd
->vdev_parent
->vdev_child
[0] == vd
)
2757 vd
->vdev_unspare
= B_TRUE
;
2761 vdev_is_dead(vdev_t
*vd
)
2764 * Holes and missing devices are always considered "dead".
2765 * This simplifies the code since we don't have to check for
2766 * these types of devices in the various code paths.
2767 * Instead we rely on the fact that we skip over dead devices
2768 * before issuing I/O to them.
2770 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2771 vd
->vdev_ops
== &vdev_missing_ops
);
2775 vdev_readable(vdev_t
*vd
)
2777 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2781 vdev_writeable(vdev_t
*vd
)
2783 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2787 vdev_allocatable(vdev_t
*vd
)
2789 uint64_t state
= vd
->vdev_state
;
2792 * We currently allow allocations from vdevs which may be in the
2793 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2794 * fails to reopen then we'll catch it later when we're holding
2795 * the proper locks. Note that we have to get the vdev state
2796 * in a local variable because although it changes atomically,
2797 * we're asking two separate questions about it.
2799 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2800 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
&&
2801 vd
->vdev_mg
->mg_initialized
);
2805 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2807 ASSERT(zio
->io_vd
== vd
);
2809 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2812 if (zio
->io_type
== ZIO_TYPE_READ
)
2813 return (!vd
->vdev_cant_read
);
2815 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2816 return (!vd
->vdev_cant_write
);
2822 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2825 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2826 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2827 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2830 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2834 * Get extended stats
2837 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2840 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2841 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2842 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2844 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2845 vsx
->vsx_total_histo
[t
][b
] +=
2846 cvsx
->vsx_total_histo
[t
][b
];
2850 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2851 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2852 vsx
->vsx_queue_histo
[t
][b
] +=
2853 cvsx
->vsx_queue_histo
[t
][b
];
2855 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2856 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2858 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2859 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2861 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2862 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2868 * Get statistics for the given vdev.
2871 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2875 * If we're getting stats on the root vdev, aggregate the I/O counts
2876 * over all top-level vdevs (i.e. the direct children of the root).
2878 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2880 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2881 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2884 memset(vsx
, 0, sizeof (*vsx
));
2886 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2887 vdev_t
*cvd
= vd
->vdev_child
[c
];
2888 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2889 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2891 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2893 vdev_get_child_stat(cvd
, vs
, cvs
);
2895 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2900 * We're a leaf. Just copy our ZIO active queue stats in. The
2901 * other leaf stats are updated in vdev_stat_update().
2906 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2908 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2909 vsx
->vsx_active_queue
[t
] =
2910 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2911 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2912 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2918 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2920 vdev_t
*tvd
= vd
->vdev_top
;
2921 mutex_enter(&vd
->vdev_stat_lock
);
2923 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2924 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2925 vs
->vs_state
= vd
->vdev_state
;
2926 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2927 if (vd
->vdev_ops
->vdev_op_leaf
)
2928 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
2929 VDEV_LABEL_END_SIZE
;
2931 * Report expandable space on top-level, non-auxillary devices
2932 * only. The expandable space is reported in terms of metaslab
2933 * sized units since that determines how much space the pool
2936 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
2937 vs
->vs_esize
= P2ALIGN(
2938 vd
->vdev_max_asize
- vd
->vdev_asize
,
2939 1ULL << tvd
->vdev_ms_shift
);
2941 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2942 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
2944 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2948 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
2949 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
2950 mutex_exit(&vd
->vdev_stat_lock
);
2954 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2956 return (vdev_get_stats_ex(vd
, vs
, NULL
));
2960 vdev_clear_stats(vdev_t
*vd
)
2962 mutex_enter(&vd
->vdev_stat_lock
);
2963 vd
->vdev_stat
.vs_space
= 0;
2964 vd
->vdev_stat
.vs_dspace
= 0;
2965 vd
->vdev_stat
.vs_alloc
= 0;
2966 mutex_exit(&vd
->vdev_stat_lock
);
2970 vdev_scan_stat_init(vdev_t
*vd
)
2972 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2975 for (c
= 0; c
< vd
->vdev_children
; c
++)
2976 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2978 mutex_enter(&vd
->vdev_stat_lock
);
2979 vs
->vs_scan_processed
= 0;
2980 mutex_exit(&vd
->vdev_stat_lock
);
2984 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2986 spa_t
*spa
= zio
->io_spa
;
2987 vdev_t
*rvd
= spa
->spa_root_vdev
;
2988 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2990 uint64_t txg
= zio
->io_txg
;
2991 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2992 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
2993 zio_type_t type
= zio
->io_type
;
2994 int flags
= zio
->io_flags
;
2997 * If this i/o is a gang leader, it didn't do any actual work.
2999 if (zio
->io_gang_tree
)
3002 if (zio
->io_error
== 0) {
3004 * If this is a root i/o, don't count it -- we've already
3005 * counted the top-level vdevs, and vdev_get_stats() will
3006 * aggregate them when asked. This reduces contention on
3007 * the root vdev_stat_lock and implicitly handles blocks
3008 * that compress away to holes, for which there is no i/o.
3009 * (Holes never create vdev children, so all the counters
3010 * remain zero, which is what we want.)
3012 * Note: this only applies to successful i/o (io_error == 0)
3013 * because unlike i/o counts, errors are not additive.
3014 * When reading a ditto block, for example, failure of
3015 * one top-level vdev does not imply a root-level error.
3020 ASSERT(vd
== zio
->io_vd
);
3022 if (flags
& ZIO_FLAG_IO_BYPASS
)
3025 mutex_enter(&vd
->vdev_stat_lock
);
3027 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3028 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3029 dsl_scan_phys_t
*scn_phys
=
3030 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3031 uint64_t *processed
= &scn_phys
->scn_processed
;
3034 if (vd
->vdev_ops
->vdev_op_leaf
)
3035 atomic_add_64(processed
, psize
);
3036 vs
->vs_scan_processed
+= psize
;
3039 if (flags
& ZIO_FLAG_SELF_HEAL
)
3040 vs
->vs_self_healed
+= psize
;
3044 * The bytes/ops/histograms are recorded at the leaf level and
3045 * aggregated into the higher level vdevs in vdev_get_stats().
3047 if (vd
->vdev_ops
->vdev_op_leaf
&&
3048 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3051 vs
->vs_bytes
[type
] += psize
;
3053 if (flags
& ZIO_FLAG_DELEGATED
) {
3054 vsx
->vsx_agg_histo
[zio
->io_priority
]
3055 [RQ_HISTO(zio
->io_size
)]++;
3057 vsx
->vsx_ind_histo
[zio
->io_priority
]
3058 [RQ_HISTO(zio
->io_size
)]++;
3061 if (zio
->io_delta
&& zio
->io_delay
) {
3062 vsx
->vsx_queue_histo
[zio
->io_priority
]
3063 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3064 vsx
->vsx_disk_histo
[type
]
3065 [L_HISTO(zio
->io_delay
)]++;
3066 vsx
->vsx_total_histo
[type
]
3067 [L_HISTO(zio
->io_delta
)]++;
3071 mutex_exit(&vd
->vdev_stat_lock
);
3075 if (flags
& ZIO_FLAG_SPECULATIVE
)
3079 * If this is an I/O error that is going to be retried, then ignore the
3080 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3081 * hard errors, when in reality they can happen for any number of
3082 * innocuous reasons (bus resets, MPxIO link failure, etc).
3084 if (zio
->io_error
== EIO
&&
3085 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3089 * Intent logs writes won't propagate their error to the root
3090 * I/O so don't mark these types of failures as pool-level
3093 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3096 mutex_enter(&vd
->vdev_stat_lock
);
3097 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3098 if (zio
->io_error
== ECKSUM
)
3099 vs
->vs_checksum_errors
++;
3101 vs
->vs_read_errors
++;
3103 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3104 vs
->vs_write_errors
++;
3105 mutex_exit(&vd
->vdev_stat_lock
);
3107 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3108 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3109 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3110 spa
->spa_claiming
)) {
3112 * This is either a normal write (not a repair), or it's
3113 * a repair induced by the scrub thread, or it's a repair
3114 * made by zil_claim() during spa_load() in the first txg.
3115 * In the normal case, we commit the DTL change in the same
3116 * txg as the block was born. In the scrub-induced repair
3117 * case, we know that scrubs run in first-pass syncing context,
3118 * so we commit the DTL change in spa_syncing_txg(spa).
3119 * In the zil_claim() case, we commit in spa_first_txg(spa).
3121 * We currently do not make DTL entries for failed spontaneous
3122 * self-healing writes triggered by normal (non-scrubbing)
3123 * reads, because we have no transactional context in which to
3124 * do so -- and it's not clear that it'd be desirable anyway.
3126 if (vd
->vdev_ops
->vdev_op_leaf
) {
3127 uint64_t commit_txg
= txg
;
3128 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3129 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3130 ASSERT(spa_sync_pass(spa
) == 1);
3131 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3132 commit_txg
= spa_syncing_txg(spa
);
3133 } else if (spa
->spa_claiming
) {
3134 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3135 commit_txg
= spa_first_txg(spa
);
3137 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3138 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3140 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3141 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3142 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3145 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3150 * Update the in-core space usage stats for this vdev, its metaslab class,
3151 * and the root vdev.
3154 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3155 int64_t space_delta
)
3157 int64_t dspace_delta
= space_delta
;
3158 spa_t
*spa
= vd
->vdev_spa
;
3159 vdev_t
*rvd
= spa
->spa_root_vdev
;
3160 metaslab_group_t
*mg
= vd
->vdev_mg
;
3161 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3163 ASSERT(vd
== vd
->vdev_top
);
3166 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3167 * factor. We must calculate this here and not at the root vdev
3168 * because the root vdev's psize-to-asize is simply the max of its
3169 * childrens', thus not accurate enough for us.
3171 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3172 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3173 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3174 vd
->vdev_deflate_ratio
;
3176 mutex_enter(&vd
->vdev_stat_lock
);
3177 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3178 vd
->vdev_stat
.vs_space
+= space_delta
;
3179 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3180 mutex_exit(&vd
->vdev_stat_lock
);
3182 if (mc
== spa_normal_class(spa
)) {
3183 mutex_enter(&rvd
->vdev_stat_lock
);
3184 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3185 rvd
->vdev_stat
.vs_space
+= space_delta
;
3186 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3187 mutex_exit(&rvd
->vdev_stat_lock
);
3191 ASSERT(rvd
== vd
->vdev_parent
);
3192 ASSERT(vd
->vdev_ms_count
!= 0);
3194 metaslab_class_space_update(mc
,
3195 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3200 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3201 * so that it will be written out next time the vdev configuration is synced.
3202 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3205 vdev_config_dirty(vdev_t
*vd
)
3207 spa_t
*spa
= vd
->vdev_spa
;
3208 vdev_t
*rvd
= spa
->spa_root_vdev
;
3211 ASSERT(spa_writeable(spa
));
3214 * If this is an aux vdev (as with l2cache and spare devices), then we
3215 * update the vdev config manually and set the sync flag.
3217 if (vd
->vdev_aux
!= NULL
) {
3218 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3222 for (c
= 0; c
< sav
->sav_count
; c
++) {
3223 if (sav
->sav_vdevs
[c
] == vd
)
3227 if (c
== sav
->sav_count
) {
3229 * We're being removed. There's nothing more to do.
3231 ASSERT(sav
->sav_sync
== B_TRUE
);
3235 sav
->sav_sync
= B_TRUE
;
3237 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3238 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3239 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3240 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3246 * Setting the nvlist in the middle if the array is a little
3247 * sketchy, but it will work.
3249 nvlist_free(aux
[c
]);
3250 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3256 * The dirty list is protected by the SCL_CONFIG lock. The caller
3257 * must either hold SCL_CONFIG as writer, or must be the sync thread
3258 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3259 * so this is sufficient to ensure mutual exclusion.
3261 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3262 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3263 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3266 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3267 vdev_config_dirty(rvd
->vdev_child
[c
]);
3269 ASSERT(vd
== vd
->vdev_top
);
3271 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3273 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3278 vdev_config_clean(vdev_t
*vd
)
3280 spa_t
*spa
= vd
->vdev_spa
;
3282 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3283 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3284 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3286 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3287 list_remove(&spa
->spa_config_dirty_list
, vd
);
3291 * Mark a top-level vdev's state as dirty, so that the next pass of
3292 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3293 * the state changes from larger config changes because they require
3294 * much less locking, and are often needed for administrative actions.
3297 vdev_state_dirty(vdev_t
*vd
)
3299 spa_t
*spa
= vd
->vdev_spa
;
3301 ASSERT(spa_writeable(spa
));
3302 ASSERT(vd
== vd
->vdev_top
);
3305 * The state list is protected by the SCL_STATE lock. The caller
3306 * must either hold SCL_STATE as writer, or must be the sync thread
3307 * (which holds SCL_STATE as reader). There's only one sync thread,
3308 * so this is sufficient to ensure mutual exclusion.
3310 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3311 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3312 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3314 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3315 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3319 vdev_state_clean(vdev_t
*vd
)
3321 spa_t
*spa
= vd
->vdev_spa
;
3323 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3324 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3325 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3327 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3328 list_remove(&spa
->spa_state_dirty_list
, vd
);
3332 * Propagate vdev state up from children to parent.
3335 vdev_propagate_state(vdev_t
*vd
)
3337 spa_t
*spa
= vd
->vdev_spa
;
3338 vdev_t
*rvd
= spa
->spa_root_vdev
;
3339 int degraded
= 0, faulted
= 0;
3344 if (vd
->vdev_children
> 0) {
3345 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3346 child
= vd
->vdev_child
[c
];
3349 * Don't factor holes into the decision.
3351 if (child
->vdev_ishole
)
3354 if (!vdev_readable(child
) ||
3355 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3357 * Root special: if there is a top-level log
3358 * device, treat the root vdev as if it were
3361 if (child
->vdev_islog
&& vd
== rvd
)
3365 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3369 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3373 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3376 * Root special: if there is a top-level vdev that cannot be
3377 * opened due to corrupted metadata, then propagate the root
3378 * vdev's aux state as 'corrupt' rather than 'insufficient
3381 if (corrupted
&& vd
== rvd
&&
3382 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3383 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3384 VDEV_AUX_CORRUPT_DATA
);
3387 if (vd
->vdev_parent
)
3388 vdev_propagate_state(vd
->vdev_parent
);
3392 * Set a vdev's state. If this is during an open, we don't update the parent
3393 * state, because we're in the process of opening children depth-first.
3394 * Otherwise, we propagate the change to the parent.
3396 * If this routine places a device in a faulted state, an appropriate ereport is
3400 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3402 uint64_t save_state
;
3403 spa_t
*spa
= vd
->vdev_spa
;
3405 if (state
== vd
->vdev_state
) {
3407 * Since vdev_offline() code path is already in an offline
3408 * state we can miss a statechange event to OFFLINE. Check
3409 * the previous state to catch this condition.
3411 if (vd
->vdev_ops
->vdev_op_leaf
&&
3412 (state
== VDEV_STATE_OFFLINE
) &&
3413 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
3414 /* post an offline state change */
3415 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
3417 vd
->vdev_stat
.vs_aux
= aux
;
3421 save_state
= vd
->vdev_state
;
3423 vd
->vdev_state
= state
;
3424 vd
->vdev_stat
.vs_aux
= aux
;
3427 * If we are setting the vdev state to anything but an open state, then
3428 * always close the underlying device unless the device has requested
3429 * a delayed close (i.e. we're about to remove or fault the device).
3430 * Otherwise, we keep accessible but invalid devices open forever.
3431 * We don't call vdev_close() itself, because that implies some extra
3432 * checks (offline, etc) that we don't want here. This is limited to
3433 * leaf devices, because otherwise closing the device will affect other
3436 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3437 vd
->vdev_ops
->vdev_op_leaf
)
3438 vd
->vdev_ops
->vdev_op_close(vd
);
3440 if (vd
->vdev_removed
&&
3441 state
== VDEV_STATE_CANT_OPEN
&&
3442 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3444 * If the previous state is set to VDEV_STATE_REMOVED, then this
3445 * device was previously marked removed and someone attempted to
3446 * reopen it. If this failed due to a nonexistent device, then
3447 * keep the device in the REMOVED state. We also let this be if
3448 * it is one of our special test online cases, which is only
3449 * attempting to online the device and shouldn't generate an FMA
3452 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3453 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3454 } else if (state
== VDEV_STATE_REMOVED
) {
3455 vd
->vdev_removed
= B_TRUE
;
3456 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3458 * If we fail to open a vdev during an import or recovery, we
3459 * mark it as "not available", which signifies that it was
3460 * never there to begin with. Failure to open such a device
3461 * is not considered an error.
3463 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3464 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3465 vd
->vdev_ops
->vdev_op_leaf
)
3466 vd
->vdev_not_present
= 1;
3469 * Post the appropriate ereport. If the 'prevstate' field is
3470 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3471 * that this is part of a vdev_reopen(). In this case, we don't
3472 * want to post the ereport if the device was already in the
3473 * CANT_OPEN state beforehand.
3475 * If the 'checkremove' flag is set, then this is an attempt to
3476 * online the device in response to an insertion event. If we
3477 * hit this case, then we have detected an insertion event for a
3478 * faulted or offline device that wasn't in the removed state.
3479 * In this scenario, we don't post an ereport because we are
3480 * about to replace the device, or attempt an online with
3481 * vdev_forcefault, which will generate the fault for us.
3483 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3484 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3485 vd
!= spa
->spa_root_vdev
) {
3489 case VDEV_AUX_OPEN_FAILED
:
3490 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3492 case VDEV_AUX_CORRUPT_DATA
:
3493 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3495 case VDEV_AUX_NO_REPLICAS
:
3496 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3498 case VDEV_AUX_BAD_GUID_SUM
:
3499 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3501 case VDEV_AUX_TOO_SMALL
:
3502 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3504 case VDEV_AUX_BAD_LABEL
:
3505 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3507 case VDEV_AUX_BAD_ASHIFT
:
3508 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
3511 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3514 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3517 /* Erase any notion of persistent removed state */
3518 vd
->vdev_removed
= B_FALSE
;
3520 vd
->vdev_removed
= B_FALSE
;
3524 * Notify ZED of any significant state-change on a leaf vdev.
3527 if (vd
->vdev_ops
->vdev_op_leaf
) {
3528 /* preserve original state from a vdev_reopen() */
3529 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
3530 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
3531 (save_state
<= VDEV_STATE_CLOSED
))
3532 save_state
= vd
->vdev_prevstate
;
3534 /* filter out state change due to initial vdev_open */
3535 if (save_state
> VDEV_STATE_CLOSED
)
3536 zfs_post_state_change(spa
, vd
, save_state
);
3539 if (!isopen
&& vd
->vdev_parent
)
3540 vdev_propagate_state(vd
->vdev_parent
);
3544 * Check the vdev configuration to ensure that it's capable of supporting
3548 vdev_is_bootable(vdev_t
*vd
)
3550 #if defined(__sun__) || defined(__sun)
3552 * Currently, we do not support RAID-Z or partial configuration.
3553 * In addition, only a single top-level vdev is allowed and none of the
3554 * leaves can be wholedisks.
3558 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3559 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3561 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3562 vd
->vdev_children
> 1) {
3564 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3565 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3570 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3571 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3574 #endif /* __sun__ || __sun */
3579 * Load the state from the original vdev tree (ovd) which
3580 * we've retrieved from the MOS config object. If the original
3581 * vdev was offline or faulted then we transfer that state to the
3582 * device in the current vdev tree (nvd).
3585 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3589 ASSERT(nvd
->vdev_top
->vdev_islog
);
3590 ASSERT(spa_config_held(nvd
->vdev_spa
,
3591 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3592 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3594 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3595 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3597 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3599 * Restore the persistent vdev state
3601 nvd
->vdev_offline
= ovd
->vdev_offline
;
3602 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3603 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3604 nvd
->vdev_removed
= ovd
->vdev_removed
;
3609 * Determine if a log device has valid content. If the vdev was
3610 * removed or faulted in the MOS config then we know that
3611 * the content on the log device has already been written to the pool.
3614 vdev_log_state_valid(vdev_t
*vd
)
3618 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3622 for (c
= 0; c
< vd
->vdev_children
; c
++)
3623 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3630 * Expand a vdev if possible.
3633 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3635 ASSERT(vd
->vdev_top
== vd
);
3636 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3638 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3639 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3640 vdev_config_dirty(vd
);
3648 vdev_split(vdev_t
*vd
)
3650 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3652 vdev_remove_child(pvd
, vd
);
3653 vdev_compact_children(pvd
);
3655 cvd
= pvd
->vdev_child
[0];
3656 if (pvd
->vdev_children
== 1) {
3657 vdev_remove_parent(cvd
);
3658 cvd
->vdev_splitting
= B_TRUE
;
3660 vdev_propagate_state(cvd
);
3664 vdev_deadman(vdev_t
*vd
)
3668 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3669 vdev_t
*cvd
= vd
->vdev_child
[c
];
3674 if (vd
->vdev_ops
->vdev_op_leaf
) {
3675 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3677 mutex_enter(&vq
->vq_lock
);
3678 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3679 spa_t
*spa
= vd
->vdev_spa
;
3684 * Look at the head of all the pending queues,
3685 * if any I/O has been outstanding for longer than
3686 * the spa_deadman_synctime we log a zevent.
3688 fio
= avl_first(&vq
->vq_active_tree
);
3689 delta
= gethrtime() - fio
->io_timestamp
;
3690 if (delta
> spa_deadman_synctime(spa
)) {
3691 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3692 "delta %lluns, last io %lluns",
3693 fio
->io_timestamp
, delta
,
3694 vq
->vq_io_complete_ts
);
3695 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3696 spa
, vd
, fio
, 0, 0);
3699 mutex_exit(&vq
->vq_lock
);
3703 #if defined(_KERNEL) && defined(HAVE_SPL)
3704 EXPORT_SYMBOL(vdev_fault
);
3705 EXPORT_SYMBOL(vdev_degrade
);
3706 EXPORT_SYMBOL(vdev_online
);
3707 EXPORT_SYMBOL(vdev_offline
);
3708 EXPORT_SYMBOL(vdev_clear
);
3710 module_param(metaslabs_per_vdev
, int, 0644);
3711 MODULE_PARM_DESC(metaslabs_per_vdev
,
3712 "Divide added vdev into approximately (but no more than) this number "