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 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011, 2014 by Delphix. 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>
49 * Virtual device management.
52 static vdev_ops_t
*vdev_ops_table
[] = {
66 * Given a vdev type, return the appropriate ops vector.
69 vdev_getops(const char *type
)
71 vdev_ops_t
*ops
, **opspp
;
73 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
74 if (strcmp(ops
->vdev_op_type
, type
) == 0)
81 * Default asize function: return the MAX of psize with the asize of
82 * all children. This is what's used by anything other than RAID-Z.
85 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
87 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
91 for (c
= 0; c
< vd
->vdev_children
; c
++) {
92 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
93 asize
= MAX(asize
, csize
);
100 * Get the minimum allocatable size. We define the allocatable size as
101 * the vdev's asize rounded to the nearest metaslab. This allows us to
102 * replace or attach devices which don't have the same physical size but
103 * can still satisfy the same number of allocations.
106 vdev_get_min_asize(vdev_t
*vd
)
108 vdev_t
*pvd
= vd
->vdev_parent
;
111 * If our parent is NULL (inactive spare or cache) or is the root,
112 * just return our own asize.
115 return (vd
->vdev_asize
);
118 * The top-level vdev just returns the allocatable size rounded
119 * to the nearest metaslab.
121 if (vd
== vd
->vdev_top
)
122 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
125 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
126 * so each child must provide at least 1/Nth of its asize.
128 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
129 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
131 return (pvd
->vdev_min_asize
);
135 vdev_set_min_asize(vdev_t
*vd
)
138 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
140 for (c
= 0; c
< vd
->vdev_children
; c
++)
141 vdev_set_min_asize(vd
->vdev_child
[c
]);
145 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
147 vdev_t
*rvd
= spa
->spa_root_vdev
;
149 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
151 if (vdev
< rvd
->vdev_children
) {
152 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
153 return (rvd
->vdev_child
[vdev
]);
160 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
165 if (vd
->vdev_guid
== guid
)
168 for (c
= 0; c
< vd
->vdev_children
; c
++)
169 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
177 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
179 size_t oldsize
, newsize
;
180 uint64_t id
= cvd
->vdev_id
;
183 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
184 ASSERT(cvd
->vdev_parent
== NULL
);
186 cvd
->vdev_parent
= pvd
;
191 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
193 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
194 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
195 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
197 newchild
= kmem_alloc(newsize
, KM_PUSHPAGE
);
198 if (pvd
->vdev_child
!= NULL
) {
199 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
200 kmem_free(pvd
->vdev_child
, oldsize
);
203 pvd
->vdev_child
= newchild
;
204 pvd
->vdev_child
[id
] = cvd
;
206 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
207 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
210 * Walk up all ancestors to update guid sum.
212 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
213 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
217 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
220 uint_t id
= cvd
->vdev_id
;
222 ASSERT(cvd
->vdev_parent
== pvd
);
227 ASSERT(id
< pvd
->vdev_children
);
228 ASSERT(pvd
->vdev_child
[id
] == cvd
);
230 pvd
->vdev_child
[id
] = NULL
;
231 cvd
->vdev_parent
= NULL
;
233 for (c
= 0; c
< pvd
->vdev_children
; c
++)
234 if (pvd
->vdev_child
[c
])
237 if (c
== pvd
->vdev_children
) {
238 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
239 pvd
->vdev_child
= NULL
;
240 pvd
->vdev_children
= 0;
244 * Walk up all ancestors to update guid sum.
246 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
247 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
251 * Remove any holes in the child array.
254 vdev_compact_children(vdev_t
*pvd
)
256 vdev_t
**newchild
, *cvd
;
257 int oldc
= pvd
->vdev_children
;
261 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
263 for (c
= newc
= 0; c
< oldc
; c
++)
264 if (pvd
->vdev_child
[c
])
267 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_PUSHPAGE
);
269 for (c
= newc
= 0; c
< oldc
; c
++) {
270 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
271 newchild
[newc
] = cvd
;
272 cvd
->vdev_id
= newc
++;
276 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
277 pvd
->vdev_child
= newchild
;
278 pvd
->vdev_children
= newc
;
282 * Allocate and minimally initialize a vdev_t.
285 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
290 vd
= kmem_zalloc(sizeof (vdev_t
), KM_PUSHPAGE
);
292 if (spa
->spa_root_vdev
== NULL
) {
293 ASSERT(ops
== &vdev_root_ops
);
294 spa
->spa_root_vdev
= vd
;
295 spa
->spa_load_guid
= spa_generate_guid(NULL
);
298 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
299 if (spa
->spa_root_vdev
== vd
) {
301 * The root vdev's guid will also be the pool guid,
302 * which must be unique among all pools.
304 guid
= spa_generate_guid(NULL
);
307 * Any other vdev's guid must be unique within the pool.
309 guid
= spa_generate_guid(spa
);
311 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
316 vd
->vdev_guid
= guid
;
317 vd
->vdev_guid_sum
= guid
;
319 vd
->vdev_state
= VDEV_STATE_CLOSED
;
320 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
322 list_link_init(&vd
->vdev_config_dirty_node
);
323 list_link_init(&vd
->vdev_state_dirty_node
);
324 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
325 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
326 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
327 for (t
= 0; t
< DTL_TYPES
; t
++) {
328 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
331 txg_list_create(&vd
->vdev_ms_list
,
332 offsetof(struct metaslab
, ms_txg_node
));
333 txg_list_create(&vd
->vdev_dtl_list
,
334 offsetof(struct vdev
, vdev_dtl_node
));
335 vd
->vdev_stat
.vs_timestamp
= gethrtime();
343 * Allocate a new vdev. The 'alloctype' is used to control whether we are
344 * creating a new vdev or loading an existing one - the behavior is slightly
345 * different for each case.
348 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
353 uint64_t guid
= 0, islog
, nparity
;
356 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
358 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
359 return (SET_ERROR(EINVAL
));
361 if ((ops
= vdev_getops(type
)) == NULL
)
362 return (SET_ERROR(EINVAL
));
365 * If this is a load, get the vdev guid from the nvlist.
366 * Otherwise, vdev_alloc_common() will generate one for us.
368 if (alloctype
== VDEV_ALLOC_LOAD
) {
371 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
373 return (SET_ERROR(EINVAL
));
375 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
376 return (SET_ERROR(EINVAL
));
377 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
378 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
379 return (SET_ERROR(EINVAL
));
380 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
381 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
382 return (SET_ERROR(EINVAL
));
383 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
384 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
385 return (SET_ERROR(EINVAL
));
389 * The first allocated vdev must be of type 'root'.
391 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
392 return (SET_ERROR(EINVAL
));
395 * Determine whether we're a log vdev.
398 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
399 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
400 return (SET_ERROR(ENOTSUP
));
402 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
403 return (SET_ERROR(ENOTSUP
));
406 * Set the nparity property for RAID-Z vdevs.
409 if (ops
== &vdev_raidz_ops
) {
410 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
412 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
413 return (SET_ERROR(EINVAL
));
415 * Previous versions could only support 1 or 2 parity
419 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
420 return (SET_ERROR(ENOTSUP
));
422 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
423 return (SET_ERROR(ENOTSUP
));
426 * We require the parity to be specified for SPAs that
427 * support multiple parity levels.
429 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
430 return (SET_ERROR(EINVAL
));
432 * Otherwise, we default to 1 parity device for RAID-Z.
439 ASSERT(nparity
!= -1ULL);
441 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
443 vd
->vdev_islog
= islog
;
444 vd
->vdev_nparity
= nparity
;
446 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
447 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
448 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
449 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
450 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
451 &vd
->vdev_physpath
) == 0)
452 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
453 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
454 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
457 * Set the whole_disk property. If it's not specified, leave the value
460 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
461 &vd
->vdev_wholedisk
) != 0)
462 vd
->vdev_wholedisk
= -1ULL;
465 * Look for the 'not present' flag. This will only be set if the device
466 * was not present at the time of import.
468 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
469 &vd
->vdev_not_present
);
472 * Get the alignment requirement.
474 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
477 * Retrieve the vdev creation time.
479 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
483 * If we're a top-level vdev, try to load the allocation parameters.
485 if (parent
&& !parent
->vdev_parent
&&
486 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
487 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
489 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
491 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
493 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
497 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
498 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
499 alloctype
== VDEV_ALLOC_ADD
||
500 alloctype
== VDEV_ALLOC_SPLIT
||
501 alloctype
== VDEV_ALLOC_ROOTPOOL
);
502 vd
->vdev_mg
= metaslab_group_create(islog
?
503 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
507 * If we're a leaf vdev, try to load the DTL object and other state.
509 if (vd
->vdev_ops
->vdev_op_leaf
&&
510 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
511 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
512 if (alloctype
== VDEV_ALLOC_LOAD
) {
513 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
514 &vd
->vdev_dtl_object
);
515 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
519 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
522 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
523 &spare
) == 0 && spare
)
527 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
530 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
531 &vd
->vdev_resilver_txg
);
534 * When importing a pool, we want to ignore the persistent fault
535 * state, as the diagnosis made on another system may not be
536 * valid in the current context. Local vdevs will
537 * remain in the faulted state.
539 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
540 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
542 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
544 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
547 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
551 VDEV_AUX_ERR_EXCEEDED
;
552 if (nvlist_lookup_string(nv
,
553 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
554 strcmp(aux
, "external") == 0)
555 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
561 * Add ourselves to the parent's list of children.
563 vdev_add_child(parent
, vd
);
571 vdev_free(vdev_t
*vd
)
574 spa_t
*spa
= vd
->vdev_spa
;
577 * vdev_free() implies closing the vdev first. This is simpler than
578 * trying to ensure complicated semantics for all callers.
582 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
583 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
588 for (c
= 0; c
< vd
->vdev_children
; c
++)
589 vdev_free(vd
->vdev_child
[c
]);
591 ASSERT(vd
->vdev_child
== NULL
);
592 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
595 * Discard allocation state.
597 if (vd
->vdev_mg
!= NULL
) {
598 vdev_metaslab_fini(vd
);
599 metaslab_group_destroy(vd
->vdev_mg
);
602 ASSERT0(vd
->vdev_stat
.vs_space
);
603 ASSERT0(vd
->vdev_stat
.vs_dspace
);
604 ASSERT0(vd
->vdev_stat
.vs_alloc
);
607 * Remove this vdev from its parent's child list.
609 vdev_remove_child(vd
->vdev_parent
, vd
);
611 ASSERT(vd
->vdev_parent
== NULL
);
614 * Clean up vdev structure.
620 spa_strfree(vd
->vdev_path
);
622 spa_strfree(vd
->vdev_devid
);
623 if (vd
->vdev_physpath
)
624 spa_strfree(vd
->vdev_physpath
);
626 spa_strfree(vd
->vdev_fru
);
628 if (vd
->vdev_isspare
)
629 spa_spare_remove(vd
);
630 if (vd
->vdev_isl2cache
)
631 spa_l2cache_remove(vd
);
633 txg_list_destroy(&vd
->vdev_ms_list
);
634 txg_list_destroy(&vd
->vdev_dtl_list
);
636 mutex_enter(&vd
->vdev_dtl_lock
);
637 space_map_close(vd
->vdev_dtl_sm
);
638 for (t
= 0; t
< DTL_TYPES
; t
++) {
639 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
640 range_tree_destroy(vd
->vdev_dtl
[t
]);
642 mutex_exit(&vd
->vdev_dtl_lock
);
644 mutex_destroy(&vd
->vdev_dtl_lock
);
645 mutex_destroy(&vd
->vdev_stat_lock
);
646 mutex_destroy(&vd
->vdev_probe_lock
);
648 if (vd
== spa
->spa_root_vdev
)
649 spa
->spa_root_vdev
= NULL
;
651 kmem_free(vd
, sizeof (vdev_t
));
655 * Transfer top-level vdev state from svd to tvd.
658 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
660 spa_t
*spa
= svd
->vdev_spa
;
665 ASSERT(tvd
== tvd
->vdev_top
);
667 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
668 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
669 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
671 svd
->vdev_ms_array
= 0;
672 svd
->vdev_ms_shift
= 0;
673 svd
->vdev_ms_count
= 0;
676 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
677 tvd
->vdev_mg
= svd
->vdev_mg
;
678 tvd
->vdev_ms
= svd
->vdev_ms
;
683 if (tvd
->vdev_mg
!= NULL
)
684 tvd
->vdev_mg
->mg_vd
= tvd
;
686 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
687 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
688 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
690 svd
->vdev_stat
.vs_alloc
= 0;
691 svd
->vdev_stat
.vs_space
= 0;
692 svd
->vdev_stat
.vs_dspace
= 0;
694 for (t
= 0; t
< TXG_SIZE
; t
++) {
695 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
696 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
697 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
698 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
699 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
700 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
703 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
704 vdev_config_clean(svd
);
705 vdev_config_dirty(tvd
);
708 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
709 vdev_state_clean(svd
);
710 vdev_state_dirty(tvd
);
713 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
714 svd
->vdev_deflate_ratio
= 0;
716 tvd
->vdev_islog
= svd
->vdev_islog
;
721 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
730 for (c
= 0; c
< vd
->vdev_children
; c
++)
731 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
735 * Add a mirror/replacing vdev above an existing vdev.
738 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
740 spa_t
*spa
= cvd
->vdev_spa
;
741 vdev_t
*pvd
= cvd
->vdev_parent
;
744 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
746 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
748 mvd
->vdev_asize
= cvd
->vdev_asize
;
749 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
750 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
751 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
752 mvd
->vdev_state
= cvd
->vdev_state
;
753 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
755 vdev_remove_child(pvd
, cvd
);
756 vdev_add_child(pvd
, mvd
);
757 cvd
->vdev_id
= mvd
->vdev_children
;
758 vdev_add_child(mvd
, cvd
);
759 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
761 if (mvd
== mvd
->vdev_top
)
762 vdev_top_transfer(cvd
, mvd
);
768 * Remove a 1-way mirror/replacing vdev from the tree.
771 vdev_remove_parent(vdev_t
*cvd
)
773 vdev_t
*mvd
= cvd
->vdev_parent
;
774 vdev_t
*pvd
= mvd
->vdev_parent
;
776 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
778 ASSERT(mvd
->vdev_children
== 1);
779 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
780 mvd
->vdev_ops
== &vdev_replacing_ops
||
781 mvd
->vdev_ops
== &vdev_spare_ops
);
782 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
784 vdev_remove_child(mvd
, cvd
);
785 vdev_remove_child(pvd
, mvd
);
788 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
789 * Otherwise, we could have detached an offline device, and when we
790 * go to import the pool we'll think we have two top-level vdevs,
791 * instead of a different version of the same top-level vdev.
793 if (mvd
->vdev_top
== mvd
) {
794 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
795 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
796 cvd
->vdev_guid
+= guid_delta
;
797 cvd
->vdev_guid_sum
+= guid_delta
;
800 * If pool not set for autoexpand, we need to also preserve
801 * mvd's asize to prevent automatic expansion of cvd.
802 * Otherwise if we are adjusting the mirror by attaching and
803 * detaching children of non-uniform sizes, the mirror could
804 * autoexpand, unexpectedly requiring larger devices to
805 * re-establish the mirror.
807 if (!cvd
->vdev_spa
->spa_autoexpand
)
808 cvd
->vdev_asize
= mvd
->vdev_asize
;
810 cvd
->vdev_id
= mvd
->vdev_id
;
811 vdev_add_child(pvd
, cvd
);
812 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
814 if (cvd
== cvd
->vdev_top
)
815 vdev_top_transfer(mvd
, cvd
);
817 ASSERT(mvd
->vdev_children
== 0);
822 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
824 spa_t
*spa
= vd
->vdev_spa
;
825 objset_t
*mos
= spa
->spa_meta_objset
;
827 uint64_t oldc
= vd
->vdev_ms_count
;
828 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
832 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
835 * This vdev is not being allocated from yet or is a hole.
837 if (vd
->vdev_ms_shift
== 0)
840 ASSERT(!vd
->vdev_ishole
);
843 * Compute the raidz-deflation ratio. Note, we hard-code
844 * in 128k (1 << 17) because it is the current "typical" blocksize.
845 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
846 * or we will inconsistently account for existing bp's.
848 vd
->vdev_deflate_ratio
= (1 << 17) /
849 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
851 ASSERT(oldc
<= newc
);
853 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_PUSHPAGE
| KM_NODEBUG
);
856 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
857 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
861 vd
->vdev_ms_count
= newc
;
863 for (m
= oldc
; m
< newc
; m
++) {
867 error
= dmu_read(mos
, vd
->vdev_ms_array
,
868 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
873 vd
->vdev_ms
[m
] = metaslab_init(vd
->vdev_mg
, m
, object
, txg
);
877 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
880 * If the vdev is being removed we don't activate
881 * the metaslabs since we want to ensure that no new
882 * allocations are performed on this device.
884 if (oldc
== 0 && !vd
->vdev_removing
)
885 metaslab_group_activate(vd
->vdev_mg
);
888 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
894 vdev_metaslab_fini(vdev_t
*vd
)
897 uint64_t count
= vd
->vdev_ms_count
;
899 if (vd
->vdev_ms
!= NULL
) {
900 metaslab_group_passivate(vd
->vdev_mg
);
901 for (m
= 0; m
< count
; m
++) {
902 metaslab_t
*msp
= vd
->vdev_ms
[m
];
907 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
911 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
914 typedef struct vdev_probe_stats
{
915 boolean_t vps_readable
;
916 boolean_t vps_writeable
;
918 } vdev_probe_stats_t
;
921 vdev_probe_done(zio_t
*zio
)
923 spa_t
*spa
= zio
->io_spa
;
924 vdev_t
*vd
= zio
->io_vd
;
925 vdev_probe_stats_t
*vps
= zio
->io_private
;
927 ASSERT(vd
->vdev_probe_zio
!= NULL
);
929 if (zio
->io_type
== ZIO_TYPE_READ
) {
930 if (zio
->io_error
== 0)
931 vps
->vps_readable
= 1;
932 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
933 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
934 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
935 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
936 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
938 zio_buf_free(zio
->io_data
, zio
->io_size
);
940 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
941 if (zio
->io_error
== 0)
942 vps
->vps_writeable
= 1;
943 zio_buf_free(zio
->io_data
, zio
->io_size
);
944 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
947 vd
->vdev_cant_read
|= !vps
->vps_readable
;
948 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
950 if (vdev_readable(vd
) &&
951 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
954 ASSERT(zio
->io_error
!= 0);
955 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
956 spa
, vd
, NULL
, 0, 0);
957 zio
->io_error
= SET_ERROR(ENXIO
);
960 mutex_enter(&vd
->vdev_probe_lock
);
961 ASSERT(vd
->vdev_probe_zio
== zio
);
962 vd
->vdev_probe_zio
= NULL
;
963 mutex_exit(&vd
->vdev_probe_lock
);
965 while ((pio
= zio_walk_parents(zio
)) != NULL
)
966 if (!vdev_accessible(vd
, pio
))
967 pio
->io_error
= SET_ERROR(ENXIO
);
969 kmem_free(vps
, sizeof (*vps
));
974 * Determine whether this device is accessible.
976 * Read and write to several known locations: the pad regions of each
977 * vdev label but the first, which we leave alone in case it contains
981 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
983 spa_t
*spa
= vd
->vdev_spa
;
984 vdev_probe_stats_t
*vps
= NULL
;
988 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
991 * Don't probe the probe.
993 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
997 * To prevent 'probe storms' when a device fails, we create
998 * just one probe i/o at a time. All zios that want to probe
999 * this vdev will become parents of the probe io.
1001 mutex_enter(&vd
->vdev_probe_lock
);
1003 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1004 vps
= kmem_zalloc(sizeof (*vps
), KM_PUSHPAGE
);
1006 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1007 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1010 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1012 * vdev_cant_read and vdev_cant_write can only
1013 * transition from TRUE to FALSE when we have the
1014 * SCL_ZIO lock as writer; otherwise they can only
1015 * transition from FALSE to TRUE. This ensures that
1016 * any zio looking at these values can assume that
1017 * failures persist for the life of the I/O. That's
1018 * important because when a device has intermittent
1019 * connectivity problems, we want to ensure that
1020 * they're ascribed to the device (ENXIO) and not
1023 * Since we hold SCL_ZIO as writer here, clear both
1024 * values so the probe can reevaluate from first
1027 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1028 vd
->vdev_cant_read
= B_FALSE
;
1029 vd
->vdev_cant_write
= B_FALSE
;
1032 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1033 vdev_probe_done
, vps
,
1034 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1037 * We can't change the vdev state in this context, so we
1038 * kick off an async task to do it on our behalf.
1041 vd
->vdev_probe_wanted
= B_TRUE
;
1042 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1047 zio_add_child(zio
, pio
);
1049 mutex_exit(&vd
->vdev_probe_lock
);
1052 ASSERT(zio
!= NULL
);
1056 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1057 zio_nowait(zio_read_phys(pio
, vd
,
1058 vdev_label_offset(vd
->vdev_psize
, l
,
1059 offsetof(vdev_label_t
, vl_pad2
)),
1060 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1061 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1062 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1073 vdev_open_child(void *arg
)
1077 vd
->vdev_open_thread
= curthread
;
1078 vd
->vdev_open_error
= vdev_open(vd
);
1079 vd
->vdev_open_thread
= NULL
;
1083 vdev_uses_zvols(vdev_t
*vd
)
1088 if (zvol_is_zvol(vd
->vdev_path
))
1092 for (c
= 0; c
< vd
->vdev_children
; c
++)
1093 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1100 vdev_open_children(vdev_t
*vd
)
1103 int children
= vd
->vdev_children
;
1107 * in order to handle pools on top of zvols, do the opens
1108 * in a single thread so that the same thread holds the
1109 * spa_namespace_lock
1111 if (vdev_uses_zvols(vd
)) {
1112 for (c
= 0; c
< children
; c
++)
1113 vd
->vdev_child
[c
]->vdev_open_error
=
1114 vdev_open(vd
->vdev_child
[c
]);
1117 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1118 children
, children
, TASKQ_PREPOPULATE
);
1120 for (c
= 0; c
< children
; c
++)
1121 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1128 * Prepare a virtual device for access.
1131 vdev_open(vdev_t
*vd
)
1133 spa_t
*spa
= vd
->vdev_spa
;
1136 uint64_t max_osize
= 0;
1137 uint64_t asize
, max_asize
, psize
;
1138 uint64_t ashift
= 0;
1141 ASSERT(vd
->vdev_open_thread
== curthread
||
1142 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1143 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1144 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1145 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1147 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1148 vd
->vdev_cant_read
= B_FALSE
;
1149 vd
->vdev_cant_write
= B_FALSE
;
1150 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1153 * If this vdev is not removed, check its fault status. If it's
1154 * faulted, bail out of the open.
1156 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1157 ASSERT(vd
->vdev_children
== 0);
1158 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1159 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1160 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1161 vd
->vdev_label_aux
);
1162 return (SET_ERROR(ENXIO
));
1163 } else if (vd
->vdev_offline
) {
1164 ASSERT(vd
->vdev_children
== 0);
1165 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1166 return (SET_ERROR(ENXIO
));
1169 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1172 * Reset the vdev_reopening flag so that we actually close
1173 * the vdev on error.
1175 vd
->vdev_reopening
= B_FALSE
;
1176 if (zio_injection_enabled
&& error
== 0)
1177 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1180 if (vd
->vdev_removed
&&
1181 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1182 vd
->vdev_removed
= B_FALSE
;
1184 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1185 vd
->vdev_stat
.vs_aux
);
1189 vd
->vdev_removed
= B_FALSE
;
1192 * Recheck the faulted flag now that we have confirmed that
1193 * the vdev is accessible. If we're faulted, bail.
1195 if (vd
->vdev_faulted
) {
1196 ASSERT(vd
->vdev_children
== 0);
1197 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1198 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1199 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1200 vd
->vdev_label_aux
);
1201 return (SET_ERROR(ENXIO
));
1204 if (vd
->vdev_degraded
) {
1205 ASSERT(vd
->vdev_children
== 0);
1206 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1207 VDEV_AUX_ERR_EXCEEDED
);
1209 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1213 * For hole or missing vdevs we just return success.
1215 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1218 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1219 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1220 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1226 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1227 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1229 if (vd
->vdev_children
== 0) {
1230 if (osize
< SPA_MINDEVSIZE
) {
1231 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1232 VDEV_AUX_TOO_SMALL
);
1233 return (SET_ERROR(EOVERFLOW
));
1236 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1237 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1238 VDEV_LABEL_END_SIZE
);
1240 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1241 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1242 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1243 VDEV_AUX_TOO_SMALL
);
1244 return (SET_ERROR(EOVERFLOW
));
1248 max_asize
= max_osize
;
1251 vd
->vdev_psize
= psize
;
1254 * Make sure the allocatable size hasn't shrunk.
1256 if (asize
< vd
->vdev_min_asize
) {
1257 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1258 VDEV_AUX_BAD_LABEL
);
1259 return (SET_ERROR(EINVAL
));
1262 if (vd
->vdev_asize
== 0) {
1264 * This is the first-ever open, so use the computed values.
1265 * For compatibility, a different ashift can be requested.
1267 vd
->vdev_asize
= asize
;
1268 vd
->vdev_max_asize
= max_asize
;
1269 if (vd
->vdev_ashift
== 0)
1270 vd
->vdev_ashift
= ashift
;
1273 * Detect if the alignment requirement has increased.
1274 * We don't want to make the pool unavailable, just
1275 * post an event instead.
1277 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1278 vd
->vdev_ops
->vdev_op_leaf
) {
1279 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1280 spa
, vd
, NULL
, 0, 0);
1283 vd
->vdev_max_asize
= max_asize
;
1287 * If all children are healthy and the asize has increased,
1288 * then we've experienced dynamic LUN growth. If automatic
1289 * expansion is enabled then use the additional space.
1291 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1292 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1293 vd
->vdev_asize
= asize
;
1295 vdev_set_min_asize(vd
);
1298 * Ensure we can issue some IO before declaring the
1299 * vdev open for business.
1301 if (vd
->vdev_ops
->vdev_op_leaf
&&
1302 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1303 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1304 VDEV_AUX_ERR_EXCEEDED
);
1309 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1310 * resilver. But don't do this if we are doing a reopen for a scrub,
1311 * since this would just restart the scrub we are already doing.
1313 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1314 vdev_resilver_needed(vd
, NULL
, NULL
))
1315 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1321 * Called once the vdevs are all opened, this routine validates the label
1322 * contents. This needs to be done before vdev_load() so that we don't
1323 * inadvertently do repair I/Os to the wrong device.
1325 * If 'strict' is false ignore the spa guid check. This is necessary because
1326 * if the machine crashed during a re-guid the new guid might have been written
1327 * to all of the vdev labels, but not the cached config. The strict check
1328 * will be performed when the pool is opened again using the mos config.
1330 * This function will only return failure if one of the vdevs indicates that it
1331 * has since been destroyed or exported. This is only possible if
1332 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1333 * will be updated but the function will return 0.
1336 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1338 spa_t
*spa
= vd
->vdev_spa
;
1340 uint64_t guid
= 0, top_guid
;
1344 for (c
= 0; c
< vd
->vdev_children
; c
++)
1345 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1346 return (SET_ERROR(EBADF
));
1349 * If the device has already failed, or was marked offline, don't do
1350 * any further validation. Otherwise, label I/O will fail and we will
1351 * overwrite the previous state.
1353 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1354 uint64_t aux_guid
= 0;
1356 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1357 spa_last_synced_txg(spa
) : -1ULL;
1359 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1360 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1361 VDEV_AUX_BAD_LABEL
);
1366 * Determine if this vdev has been split off into another
1367 * pool. If so, then refuse to open it.
1369 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1370 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1371 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1372 VDEV_AUX_SPLIT_POOL
);
1377 if (strict
&& (nvlist_lookup_uint64(label
,
1378 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1379 guid
!= spa_guid(spa
))) {
1380 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1381 VDEV_AUX_CORRUPT_DATA
);
1386 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1387 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1392 * If this vdev just became a top-level vdev because its
1393 * sibling was detached, it will have adopted the parent's
1394 * vdev guid -- but the label may or may not be on disk yet.
1395 * Fortunately, either version of the label will have the
1396 * same top guid, so if we're a top-level vdev, we can
1397 * safely compare to that instead.
1399 * If we split this vdev off instead, then we also check the
1400 * original pool's guid. We don't want to consider the vdev
1401 * corrupt if it is partway through a split operation.
1403 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1405 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1407 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1408 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1409 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1410 VDEV_AUX_CORRUPT_DATA
);
1415 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1417 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1418 VDEV_AUX_CORRUPT_DATA
);
1426 * If this is a verbatim import, no need to check the
1427 * state of the pool.
1429 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1430 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1431 state
!= POOL_STATE_ACTIVE
)
1432 return (SET_ERROR(EBADF
));
1435 * If we were able to open and validate a vdev that was
1436 * previously marked permanently unavailable, clear that state
1439 if (vd
->vdev_not_present
)
1440 vd
->vdev_not_present
= 0;
1447 * Close a virtual device.
1450 vdev_close(vdev_t
*vd
)
1452 vdev_t
*pvd
= vd
->vdev_parent
;
1453 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1455 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1458 * If our parent is reopening, then we are as well, unless we are
1461 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1462 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1464 vd
->vdev_ops
->vdev_op_close(vd
);
1466 vdev_cache_purge(vd
);
1469 * We record the previous state before we close it, so that if we are
1470 * doing a reopen(), we don't generate FMA ereports if we notice that
1471 * it's still faulted.
1473 vd
->vdev_prevstate
= vd
->vdev_state
;
1475 if (vd
->vdev_offline
)
1476 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1478 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1479 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1483 vdev_hold(vdev_t
*vd
)
1485 spa_t
*spa
= vd
->vdev_spa
;
1488 ASSERT(spa_is_root(spa
));
1489 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1492 for (c
= 0; c
< vd
->vdev_children
; c
++)
1493 vdev_hold(vd
->vdev_child
[c
]);
1495 if (vd
->vdev_ops
->vdev_op_leaf
)
1496 vd
->vdev_ops
->vdev_op_hold(vd
);
1500 vdev_rele(vdev_t
*vd
)
1504 ASSERT(spa_is_root(vd
->vdev_spa
));
1505 for (c
= 0; c
< vd
->vdev_children
; c
++)
1506 vdev_rele(vd
->vdev_child
[c
]);
1508 if (vd
->vdev_ops
->vdev_op_leaf
)
1509 vd
->vdev_ops
->vdev_op_rele(vd
);
1513 * Reopen all interior vdevs and any unopened leaves. We don't actually
1514 * reopen leaf vdevs which had previously been opened as they might deadlock
1515 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1516 * If the leaf has never been opened then open it, as usual.
1519 vdev_reopen(vdev_t
*vd
)
1521 spa_t
*spa
= vd
->vdev_spa
;
1523 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1525 /* set the reopening flag unless we're taking the vdev offline */
1526 vd
->vdev_reopening
= !vd
->vdev_offline
;
1528 (void) vdev_open(vd
);
1531 * Call vdev_validate() here to make sure we have the same device.
1532 * Otherwise, a device with an invalid label could be successfully
1533 * opened in response to vdev_reopen().
1536 (void) vdev_validate_aux(vd
);
1537 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1538 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1539 !l2arc_vdev_present(vd
))
1540 l2arc_add_vdev(spa
, vd
);
1542 (void) vdev_validate(vd
, B_TRUE
);
1546 * Reassess parent vdev's health.
1548 vdev_propagate_state(vd
);
1552 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1557 * Normally, partial opens (e.g. of a mirror) are allowed.
1558 * For a create, however, we want to fail the request if
1559 * there are any components we can't open.
1561 error
= vdev_open(vd
);
1563 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1565 return (error
? error
: ENXIO
);
1569 * Recursively load DTLs and initialize all labels.
1571 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1572 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1573 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1582 vdev_metaslab_set_size(vdev_t
*vd
)
1585 * Aim for roughly 200 metaslabs per vdev.
1587 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ 200);
1588 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1592 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1594 ASSERT(vd
== vd
->vdev_top
);
1595 ASSERT(!vd
->vdev_ishole
);
1596 ASSERT(ISP2(flags
));
1597 ASSERT(spa_writeable(vd
->vdev_spa
));
1599 if (flags
& VDD_METASLAB
)
1600 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1602 if (flags
& VDD_DTL
)
1603 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1605 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1609 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1613 for (c
= 0; c
< vd
->vdev_children
; c
++)
1614 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1616 if (vd
->vdev_ops
->vdev_op_leaf
)
1617 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1623 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1624 * the vdev has less than perfect replication. There are four kinds of DTL:
1626 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1628 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1630 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1631 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1632 * txgs that was scrubbed.
1634 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1635 * persistent errors or just some device being offline.
1636 * Unlike the other three, the DTL_OUTAGE map is not generally
1637 * maintained; it's only computed when needed, typically to
1638 * determine whether a device can be detached.
1640 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1641 * either has the data or it doesn't.
1643 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1644 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1645 * if any child is less than fully replicated, then so is its parent.
1646 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1647 * comprising only those txgs which appear in 'maxfaults' or more children;
1648 * those are the txgs we don't have enough replication to read. For example,
1649 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1650 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1651 * two child DTL_MISSING maps.
1653 * It should be clear from the above that to compute the DTLs and outage maps
1654 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1655 * Therefore, that is all we keep on disk. When loading the pool, or after
1656 * a configuration change, we generate all other DTLs from first principles.
1659 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1661 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1663 ASSERT(t
< DTL_TYPES
);
1664 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1665 ASSERT(spa_writeable(vd
->vdev_spa
));
1667 mutex_enter(rt
->rt_lock
);
1668 if (!range_tree_contains(rt
, txg
, size
))
1669 range_tree_add(rt
, txg
, size
);
1670 mutex_exit(rt
->rt_lock
);
1674 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1676 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1677 boolean_t dirty
= B_FALSE
;
1679 ASSERT(t
< DTL_TYPES
);
1680 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1682 mutex_enter(rt
->rt_lock
);
1683 if (range_tree_space(rt
) != 0)
1684 dirty
= range_tree_contains(rt
, txg
, size
);
1685 mutex_exit(rt
->rt_lock
);
1691 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1693 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1696 mutex_enter(rt
->rt_lock
);
1697 empty
= (range_tree_space(rt
) == 0);
1698 mutex_exit(rt
->rt_lock
);
1704 * Returns the lowest txg in the DTL range.
1707 vdev_dtl_min(vdev_t
*vd
)
1711 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1712 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1713 ASSERT0(vd
->vdev_children
);
1715 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1716 return (rs
->rs_start
- 1);
1720 * Returns the highest txg in the DTL.
1723 vdev_dtl_max(vdev_t
*vd
)
1727 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1728 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1729 ASSERT0(vd
->vdev_children
);
1731 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1732 return (rs
->rs_end
);
1736 * Determine if a resilvering vdev should remove any DTL entries from
1737 * its range. If the vdev was resilvering for the entire duration of the
1738 * scan then it should excise that range from its DTLs. Otherwise, this
1739 * vdev is considered partially resilvered and should leave its DTL
1740 * entries intact. The comment in vdev_dtl_reassess() describes how we
1744 vdev_dtl_should_excise(vdev_t
*vd
)
1746 spa_t
*spa
= vd
->vdev_spa
;
1747 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1749 ASSERT0(scn
->scn_phys
.scn_errors
);
1750 ASSERT0(vd
->vdev_children
);
1752 if (vd
->vdev_resilver_txg
== 0 ||
1753 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1757 * When a resilver is initiated the scan will assign the scn_max_txg
1758 * value to the highest txg value that exists in all DTLs. If this
1759 * device's max DTL is not part of this scan (i.e. it is not in
1760 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1763 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1764 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1765 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1766 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1773 * Reassess DTLs after a config change or scrub completion.
1776 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1778 spa_t
*spa
= vd
->vdev_spa
;
1782 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1784 for (c
= 0; c
< vd
->vdev_children
; c
++)
1785 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1786 scrub_txg
, scrub_done
);
1788 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1791 if (vd
->vdev_ops
->vdev_op_leaf
) {
1792 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1794 mutex_enter(&vd
->vdev_dtl_lock
);
1797 * If we've completed a scan cleanly then determine
1798 * if this vdev should remove any DTLs. We only want to
1799 * excise regions on vdevs that were available during
1800 * the entire duration of this scan.
1802 if (scrub_txg
!= 0 &&
1803 (spa
->spa_scrub_started
||
1804 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1805 vdev_dtl_should_excise(vd
)) {
1807 * We completed a scrub up to scrub_txg. If we
1808 * did it without rebooting, then the scrub dtl
1809 * will be valid, so excise the old region and
1810 * fold in the scrub dtl. Otherwise, leave the
1811 * dtl as-is if there was an error.
1813 * There's little trick here: to excise the beginning
1814 * of the DTL_MISSING map, we put it into a reference
1815 * tree and then add a segment with refcnt -1 that
1816 * covers the range [0, scrub_txg). This means
1817 * that each txg in that range has refcnt -1 or 0.
1818 * We then add DTL_SCRUB with a refcnt of 2, so that
1819 * entries in the range [0, scrub_txg) will have a
1820 * positive refcnt -- either 1 or 2. We then convert
1821 * the reference tree into the new DTL_MISSING map.
1823 space_reftree_create(&reftree
);
1824 space_reftree_add_map(&reftree
,
1825 vd
->vdev_dtl
[DTL_MISSING
], 1);
1826 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1827 space_reftree_add_map(&reftree
,
1828 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1829 space_reftree_generate_map(&reftree
,
1830 vd
->vdev_dtl
[DTL_MISSING
], 1);
1831 space_reftree_destroy(&reftree
);
1833 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1834 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1835 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1837 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1838 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1839 if (!vdev_readable(vd
))
1840 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1842 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1843 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1846 * If the vdev was resilvering and no longer has any
1847 * DTLs then reset its resilvering flag.
1849 if (vd
->vdev_resilver_txg
!= 0 &&
1850 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1851 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0)
1852 vd
->vdev_resilver_txg
= 0;
1854 mutex_exit(&vd
->vdev_dtl_lock
);
1857 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1861 mutex_enter(&vd
->vdev_dtl_lock
);
1862 for (t
= 0; t
< DTL_TYPES
; t
++) {
1865 /* account for child's outage in parent's missing map */
1866 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1868 continue; /* leaf vdevs only */
1869 if (t
== DTL_PARTIAL
)
1870 minref
= 1; /* i.e. non-zero */
1871 else if (vd
->vdev_nparity
!= 0)
1872 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1874 minref
= vd
->vdev_children
; /* any kind of mirror */
1875 space_reftree_create(&reftree
);
1876 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1877 vdev_t
*cvd
= vd
->vdev_child
[c
];
1878 mutex_enter(&cvd
->vdev_dtl_lock
);
1879 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1880 mutex_exit(&cvd
->vdev_dtl_lock
);
1882 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1883 space_reftree_destroy(&reftree
);
1885 mutex_exit(&vd
->vdev_dtl_lock
);
1889 vdev_dtl_load(vdev_t
*vd
)
1891 spa_t
*spa
= vd
->vdev_spa
;
1892 objset_t
*mos
= spa
->spa_meta_objset
;
1896 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
1897 ASSERT(!vd
->vdev_ishole
);
1899 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
1900 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
1903 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
1905 mutex_enter(&vd
->vdev_dtl_lock
);
1908 * Now that we've opened the space_map we need to update
1911 space_map_update(vd
->vdev_dtl_sm
);
1913 error
= space_map_load(vd
->vdev_dtl_sm
,
1914 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
1915 mutex_exit(&vd
->vdev_dtl_lock
);
1920 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1921 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
1930 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1932 spa_t
*spa
= vd
->vdev_spa
;
1933 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
1934 objset_t
*mos
= spa
->spa_meta_objset
;
1935 range_tree_t
*rtsync
;
1938 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
1940 ASSERT(!vd
->vdev_ishole
);
1941 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1943 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1945 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
1946 mutex_enter(&vd
->vdev_dtl_lock
);
1947 space_map_free(vd
->vdev_dtl_sm
, tx
);
1948 space_map_close(vd
->vdev_dtl_sm
);
1949 vd
->vdev_dtl_sm
= NULL
;
1950 mutex_exit(&vd
->vdev_dtl_lock
);
1955 if (vd
->vdev_dtl_sm
== NULL
) {
1956 uint64_t new_object
;
1958 new_object
= space_map_alloc(mos
, tx
);
1959 VERIFY3U(new_object
, !=, 0);
1961 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
1962 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
1963 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
1966 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
1968 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
1970 mutex_enter(&rtlock
);
1972 mutex_enter(&vd
->vdev_dtl_lock
);
1973 range_tree_walk(rt
, range_tree_add
, rtsync
);
1974 mutex_exit(&vd
->vdev_dtl_lock
);
1976 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
1977 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
1978 range_tree_vacate(rtsync
, NULL
, NULL
);
1980 range_tree_destroy(rtsync
);
1982 mutex_exit(&rtlock
);
1983 mutex_destroy(&rtlock
);
1986 * If the object for the space map has changed then dirty
1987 * the top level so that we update the config.
1989 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
1990 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
1991 "new object %llu", txg
, spa_name(spa
), object
,
1992 space_map_object(vd
->vdev_dtl_sm
));
1993 vdev_config_dirty(vd
->vdev_top
);
1998 mutex_enter(&vd
->vdev_dtl_lock
);
1999 space_map_update(vd
->vdev_dtl_sm
);
2000 mutex_exit(&vd
->vdev_dtl_lock
);
2004 * Determine whether the specified vdev can be offlined/detached/removed
2005 * without losing data.
2008 vdev_dtl_required(vdev_t
*vd
)
2010 spa_t
*spa
= vd
->vdev_spa
;
2011 vdev_t
*tvd
= vd
->vdev_top
;
2012 uint8_t cant_read
= vd
->vdev_cant_read
;
2015 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2017 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2021 * Temporarily mark the device as unreadable, and then determine
2022 * whether this results in any DTL outages in the top-level vdev.
2023 * If not, we can safely offline/detach/remove the device.
2025 vd
->vdev_cant_read
= B_TRUE
;
2026 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2027 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2028 vd
->vdev_cant_read
= cant_read
;
2029 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2031 if (!required
&& zio_injection_enabled
)
2032 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2038 * Determine if resilver is needed, and if so the txg range.
2041 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2043 boolean_t needed
= B_FALSE
;
2044 uint64_t thismin
= UINT64_MAX
;
2045 uint64_t thismax
= 0;
2048 if (vd
->vdev_children
== 0) {
2049 mutex_enter(&vd
->vdev_dtl_lock
);
2050 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2051 vdev_writeable(vd
)) {
2053 thismin
= vdev_dtl_min(vd
);
2054 thismax
= vdev_dtl_max(vd
);
2057 mutex_exit(&vd
->vdev_dtl_lock
);
2059 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2060 vdev_t
*cvd
= vd
->vdev_child
[c
];
2061 uint64_t cmin
, cmax
;
2063 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2064 thismin
= MIN(thismin
, cmin
);
2065 thismax
= MAX(thismax
, cmax
);
2071 if (needed
&& minp
) {
2079 vdev_load(vdev_t
*vd
)
2084 * Recursively load all children.
2086 for (c
= 0; c
< vd
->vdev_children
; c
++)
2087 vdev_load(vd
->vdev_child
[c
]);
2090 * If this is a top-level vdev, initialize its metaslabs.
2092 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2093 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2094 vdev_metaslab_init(vd
, 0) != 0))
2095 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2096 VDEV_AUX_CORRUPT_DATA
);
2099 * If this is a leaf vdev, load its DTL.
2101 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2102 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2103 VDEV_AUX_CORRUPT_DATA
);
2107 * The special vdev case is used for hot spares and l2cache devices. Its
2108 * sole purpose it to set the vdev state for the associated vdev. To do this,
2109 * we make sure that we can open the underlying device, then try to read the
2110 * label, and make sure that the label is sane and that it hasn't been
2111 * repurposed to another pool.
2114 vdev_validate_aux(vdev_t
*vd
)
2117 uint64_t guid
, version
;
2120 if (!vdev_readable(vd
))
2123 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2124 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2125 VDEV_AUX_CORRUPT_DATA
);
2129 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2130 !SPA_VERSION_IS_SUPPORTED(version
) ||
2131 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2132 guid
!= vd
->vdev_guid
||
2133 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2134 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2135 VDEV_AUX_CORRUPT_DATA
);
2141 * We don't actually check the pool state here. If it's in fact in
2142 * use by another pool, we update this fact on the fly when requested.
2149 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2151 spa_t
*spa
= vd
->vdev_spa
;
2152 objset_t
*mos
= spa
->spa_meta_objset
;
2156 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2158 if (vd
->vdev_ms
!= NULL
) {
2159 metaslab_group_t
*mg
= vd
->vdev_mg
;
2161 metaslab_group_histogram_verify(mg
);
2162 metaslab_class_histogram_verify(mg
->mg_class
);
2164 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2165 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2167 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2170 mutex_enter(&msp
->ms_lock
);
2172 * If the metaslab was not loaded when the vdev
2173 * was removed then the histogram accounting may
2174 * not be accurate. Update the histogram information
2175 * here so that we ensure that the metaslab group
2176 * and metaslab class are up-to-date.
2178 metaslab_group_histogram_remove(mg
, msp
);
2180 VERIFY0(space_map_allocated(msp
->ms_sm
));
2181 space_map_free(msp
->ms_sm
, tx
);
2182 space_map_close(msp
->ms_sm
);
2184 mutex_exit(&msp
->ms_lock
);
2187 metaslab_group_histogram_verify(mg
);
2188 metaslab_class_histogram_verify(mg
->mg_class
);
2189 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2190 ASSERT0(mg
->mg_histogram
[i
]);
2194 if (vd
->vdev_ms_array
) {
2195 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2196 vd
->vdev_ms_array
= 0;
2202 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2205 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2207 ASSERT(!vd
->vdev_ishole
);
2209 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2210 metaslab_sync_done(msp
, txg
);
2213 metaslab_sync_reassess(vd
->vdev_mg
);
2217 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2219 spa_t
*spa
= vd
->vdev_spa
;
2224 ASSERT(!vd
->vdev_ishole
);
2226 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2227 ASSERT(vd
== vd
->vdev_top
);
2228 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2229 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2230 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2231 ASSERT(vd
->vdev_ms_array
!= 0);
2232 vdev_config_dirty(vd
);
2237 * Remove the metadata associated with this vdev once it's empty.
2239 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2240 vdev_remove(vd
, txg
);
2242 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2243 metaslab_sync(msp
, txg
);
2244 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2247 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2248 vdev_dtl_sync(lvd
, txg
);
2250 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2254 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2256 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2260 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2261 * not be opened, and no I/O is attempted.
2264 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2268 spa_vdev_state_enter(spa
, SCL_NONE
);
2270 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2271 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2273 if (!vd
->vdev_ops
->vdev_op_leaf
)
2274 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2279 * We don't directly use the aux state here, but if we do a
2280 * vdev_reopen(), we need this value to be present to remember why we
2283 vd
->vdev_label_aux
= aux
;
2286 * Faulted state takes precedence over degraded.
2288 vd
->vdev_delayed_close
= B_FALSE
;
2289 vd
->vdev_faulted
= 1ULL;
2290 vd
->vdev_degraded
= 0ULL;
2291 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2294 * If this device has the only valid copy of the data, then
2295 * back off and simply mark the vdev as degraded instead.
2297 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2298 vd
->vdev_degraded
= 1ULL;
2299 vd
->vdev_faulted
= 0ULL;
2302 * If we reopen the device and it's not dead, only then do we
2307 if (vdev_readable(vd
))
2308 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2311 return (spa_vdev_state_exit(spa
, vd
, 0));
2315 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2316 * user that something is wrong. The vdev continues to operate as normal as far
2317 * as I/O is concerned.
2320 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2324 spa_vdev_state_enter(spa
, SCL_NONE
);
2326 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2327 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2329 if (!vd
->vdev_ops
->vdev_op_leaf
)
2330 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2333 * If the vdev is already faulted, then don't do anything.
2335 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2336 return (spa_vdev_state_exit(spa
, NULL
, 0));
2338 vd
->vdev_degraded
= 1ULL;
2339 if (!vdev_is_dead(vd
))
2340 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2343 return (spa_vdev_state_exit(spa
, vd
, 0));
2347 * Online the given vdev.
2349 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2350 * spare device should be detached when the device finishes resilvering.
2351 * Second, the online should be treated like a 'test' online case, so no FMA
2352 * events are generated if the device fails to open.
2355 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2357 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2359 spa_vdev_state_enter(spa
, SCL_NONE
);
2361 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2362 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2364 if (!vd
->vdev_ops
->vdev_op_leaf
)
2365 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2368 vd
->vdev_offline
= B_FALSE
;
2369 vd
->vdev_tmpoffline
= B_FALSE
;
2370 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2371 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2373 /* XXX - L2ARC 1.0 does not support expansion */
2374 if (!vd
->vdev_aux
) {
2375 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2376 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2380 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2382 if (!vd
->vdev_aux
) {
2383 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2384 pvd
->vdev_expanding
= B_FALSE
;
2388 *newstate
= vd
->vdev_state
;
2389 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2390 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2391 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2392 vd
->vdev_parent
->vdev_child
[0] == vd
)
2393 vd
->vdev_unspare
= B_TRUE
;
2395 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2397 /* XXX - L2ARC 1.0 does not support expansion */
2399 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2400 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2402 return (spa_vdev_state_exit(spa
, vd
, 0));
2406 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2410 uint64_t generation
;
2411 metaslab_group_t
*mg
;
2414 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2416 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2417 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2419 if (!vd
->vdev_ops
->vdev_op_leaf
)
2420 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2424 generation
= spa
->spa_config_generation
+ 1;
2427 * If the device isn't already offline, try to offline it.
2429 if (!vd
->vdev_offline
) {
2431 * If this device has the only valid copy of some data,
2432 * don't allow it to be offlined. Log devices are always
2435 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2436 vdev_dtl_required(vd
))
2437 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2440 * If the top-level is a slog and it has had allocations
2441 * then proceed. We check that the vdev's metaslab group
2442 * is not NULL since it's possible that we may have just
2443 * added this vdev but not yet initialized its metaslabs.
2445 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2447 * Prevent any future allocations.
2449 metaslab_group_passivate(mg
);
2450 (void) spa_vdev_state_exit(spa
, vd
, 0);
2452 error
= spa_offline_log(spa
);
2454 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2457 * Check to see if the config has changed.
2459 if (error
|| generation
!= spa
->spa_config_generation
) {
2460 metaslab_group_activate(mg
);
2462 return (spa_vdev_state_exit(spa
,
2464 (void) spa_vdev_state_exit(spa
, vd
, 0);
2467 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2471 * Offline this device and reopen its top-level vdev.
2472 * If the top-level vdev is a log device then just offline
2473 * it. Otherwise, if this action results in the top-level
2474 * vdev becoming unusable, undo it and fail the request.
2476 vd
->vdev_offline
= B_TRUE
;
2479 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2480 vdev_is_dead(tvd
)) {
2481 vd
->vdev_offline
= B_FALSE
;
2483 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2487 * Add the device back into the metaslab rotor so that
2488 * once we online the device it's open for business.
2490 if (tvd
->vdev_islog
&& mg
!= NULL
)
2491 metaslab_group_activate(mg
);
2494 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2496 return (spa_vdev_state_exit(spa
, vd
, 0));
2500 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2504 mutex_enter(&spa
->spa_vdev_top_lock
);
2505 error
= vdev_offline_locked(spa
, guid
, flags
);
2506 mutex_exit(&spa
->spa_vdev_top_lock
);
2512 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2513 * vdev_offline(), we assume the spa config is locked. We also clear all
2514 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2517 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2519 vdev_t
*rvd
= spa
->spa_root_vdev
;
2522 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2527 vd
->vdev_stat
.vs_read_errors
= 0;
2528 vd
->vdev_stat
.vs_write_errors
= 0;
2529 vd
->vdev_stat
.vs_checksum_errors
= 0;
2531 for (c
= 0; c
< vd
->vdev_children
; c
++)
2532 vdev_clear(spa
, vd
->vdev_child
[c
]);
2535 * If we're in the FAULTED state or have experienced failed I/O, then
2536 * clear the persistent state and attempt to reopen the device. We
2537 * also mark the vdev config dirty, so that the new faulted state is
2538 * written out to disk.
2540 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2541 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2544 * When reopening in reponse to a clear event, it may be due to
2545 * a fmadm repair request. In this case, if the device is
2546 * still broken, we want to still post the ereport again.
2548 vd
->vdev_forcefault
= B_TRUE
;
2550 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2551 vd
->vdev_cant_read
= B_FALSE
;
2552 vd
->vdev_cant_write
= B_FALSE
;
2554 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2556 vd
->vdev_forcefault
= B_FALSE
;
2558 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2559 vdev_state_dirty(vd
->vdev_top
);
2561 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2562 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2564 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_CLEAR
);
2568 * When clearing a FMA-diagnosed fault, we always want to
2569 * unspare the device, as we assume that the original spare was
2570 * done in response to the FMA fault.
2572 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2573 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2574 vd
->vdev_parent
->vdev_child
[0] == vd
)
2575 vd
->vdev_unspare
= B_TRUE
;
2579 vdev_is_dead(vdev_t
*vd
)
2582 * Holes and missing devices are always considered "dead".
2583 * This simplifies the code since we don't have to check for
2584 * these types of devices in the various code paths.
2585 * Instead we rely on the fact that we skip over dead devices
2586 * before issuing I/O to them.
2588 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2589 vd
->vdev_ops
== &vdev_missing_ops
);
2593 vdev_readable(vdev_t
*vd
)
2595 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2599 vdev_writeable(vdev_t
*vd
)
2601 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2605 vdev_allocatable(vdev_t
*vd
)
2607 uint64_t state
= vd
->vdev_state
;
2610 * We currently allow allocations from vdevs which may be in the
2611 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2612 * fails to reopen then we'll catch it later when we're holding
2613 * the proper locks. Note that we have to get the vdev state
2614 * in a local variable because although it changes atomically,
2615 * we're asking two separate questions about it.
2617 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2618 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2622 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2624 ASSERT(zio
->io_vd
== vd
);
2626 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2629 if (zio
->io_type
== ZIO_TYPE_READ
)
2630 return (!vd
->vdev_cant_read
);
2632 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2633 return (!vd
->vdev_cant_write
);
2639 * Get statistics for the given vdev.
2642 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2644 spa_t
*spa
= vd
->vdev_spa
;
2645 vdev_t
*rvd
= spa
->spa_root_vdev
;
2648 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2650 mutex_enter(&vd
->vdev_stat_lock
);
2651 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2652 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2653 vs
->vs_state
= vd
->vdev_state
;
2654 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2655 if (vd
->vdev_ops
->vdev_op_leaf
)
2656 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2657 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2658 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
)
2659 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2662 * If we're getting stats on the root vdev, aggregate the I/O counts
2663 * over all top-level vdevs (i.e. the direct children of the root).
2666 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2667 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2668 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2670 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2671 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2672 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2674 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2677 mutex_exit(&vd
->vdev_stat_lock
);
2681 vdev_clear_stats(vdev_t
*vd
)
2683 mutex_enter(&vd
->vdev_stat_lock
);
2684 vd
->vdev_stat
.vs_space
= 0;
2685 vd
->vdev_stat
.vs_dspace
= 0;
2686 vd
->vdev_stat
.vs_alloc
= 0;
2687 mutex_exit(&vd
->vdev_stat_lock
);
2691 vdev_scan_stat_init(vdev_t
*vd
)
2693 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2696 for (c
= 0; c
< vd
->vdev_children
; c
++)
2697 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2699 mutex_enter(&vd
->vdev_stat_lock
);
2700 vs
->vs_scan_processed
= 0;
2701 mutex_exit(&vd
->vdev_stat_lock
);
2705 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2707 spa_t
*spa
= zio
->io_spa
;
2708 vdev_t
*rvd
= spa
->spa_root_vdev
;
2709 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2711 uint64_t txg
= zio
->io_txg
;
2712 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2713 zio_type_t type
= zio
->io_type
;
2714 int flags
= zio
->io_flags
;
2717 * If this i/o is a gang leader, it didn't do any actual work.
2719 if (zio
->io_gang_tree
)
2722 if (zio
->io_error
== 0) {
2724 * If this is a root i/o, don't count it -- we've already
2725 * counted the top-level vdevs, and vdev_get_stats() will
2726 * aggregate them when asked. This reduces contention on
2727 * the root vdev_stat_lock and implicitly handles blocks
2728 * that compress away to holes, for which there is no i/o.
2729 * (Holes never create vdev children, so all the counters
2730 * remain zero, which is what we want.)
2732 * Note: this only applies to successful i/o (io_error == 0)
2733 * because unlike i/o counts, errors are not additive.
2734 * When reading a ditto block, for example, failure of
2735 * one top-level vdev does not imply a root-level error.
2740 ASSERT(vd
== zio
->io_vd
);
2742 if (flags
& ZIO_FLAG_IO_BYPASS
)
2745 mutex_enter(&vd
->vdev_stat_lock
);
2747 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2748 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2749 dsl_scan_phys_t
*scn_phys
=
2750 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2751 uint64_t *processed
= &scn_phys
->scn_processed
;
2754 if (vd
->vdev_ops
->vdev_op_leaf
)
2755 atomic_add_64(processed
, psize
);
2756 vs
->vs_scan_processed
+= psize
;
2759 if (flags
& ZIO_FLAG_SELF_HEAL
)
2760 vs
->vs_self_healed
+= psize
;
2764 vs
->vs_bytes
[type
] += psize
;
2766 mutex_exit(&vd
->vdev_stat_lock
);
2770 if (flags
& ZIO_FLAG_SPECULATIVE
)
2774 * If this is an I/O error that is going to be retried, then ignore the
2775 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2776 * hard errors, when in reality they can happen for any number of
2777 * innocuous reasons (bus resets, MPxIO link failure, etc).
2779 if (zio
->io_error
== EIO
&&
2780 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2784 * Intent logs writes won't propagate their error to the root
2785 * I/O so don't mark these types of failures as pool-level
2788 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2791 mutex_enter(&vd
->vdev_stat_lock
);
2792 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2793 if (zio
->io_error
== ECKSUM
)
2794 vs
->vs_checksum_errors
++;
2796 vs
->vs_read_errors
++;
2798 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2799 vs
->vs_write_errors
++;
2800 mutex_exit(&vd
->vdev_stat_lock
);
2802 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2803 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2804 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2805 spa
->spa_claiming
)) {
2807 * This is either a normal write (not a repair), or it's
2808 * a repair induced by the scrub thread, or it's a repair
2809 * made by zil_claim() during spa_load() in the first txg.
2810 * In the normal case, we commit the DTL change in the same
2811 * txg as the block was born. In the scrub-induced repair
2812 * case, we know that scrubs run in first-pass syncing context,
2813 * so we commit the DTL change in spa_syncing_txg(spa).
2814 * In the zil_claim() case, we commit in spa_first_txg(spa).
2816 * We currently do not make DTL entries for failed spontaneous
2817 * self-healing writes triggered by normal (non-scrubbing)
2818 * reads, because we have no transactional context in which to
2819 * do so -- and it's not clear that it'd be desirable anyway.
2821 if (vd
->vdev_ops
->vdev_op_leaf
) {
2822 uint64_t commit_txg
= txg
;
2823 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2824 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2825 ASSERT(spa_sync_pass(spa
) == 1);
2826 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2827 commit_txg
= spa_syncing_txg(spa
);
2828 } else if (spa
->spa_claiming
) {
2829 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2830 commit_txg
= spa_first_txg(spa
);
2832 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2833 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2835 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2836 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2837 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2840 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2845 * Update the in-core space usage stats for this vdev, its metaslab class,
2846 * and the root vdev.
2849 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2850 int64_t space_delta
)
2852 int64_t dspace_delta
= space_delta
;
2853 spa_t
*spa
= vd
->vdev_spa
;
2854 vdev_t
*rvd
= spa
->spa_root_vdev
;
2855 metaslab_group_t
*mg
= vd
->vdev_mg
;
2856 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2858 ASSERT(vd
== vd
->vdev_top
);
2861 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2862 * factor. We must calculate this here and not at the root vdev
2863 * because the root vdev's psize-to-asize is simply the max of its
2864 * childrens', thus not accurate enough for us.
2866 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2867 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2868 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2869 vd
->vdev_deflate_ratio
;
2871 mutex_enter(&vd
->vdev_stat_lock
);
2872 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2873 vd
->vdev_stat
.vs_space
+= space_delta
;
2874 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2875 mutex_exit(&vd
->vdev_stat_lock
);
2877 if (mc
== spa_normal_class(spa
)) {
2878 mutex_enter(&rvd
->vdev_stat_lock
);
2879 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2880 rvd
->vdev_stat
.vs_space
+= space_delta
;
2881 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2882 mutex_exit(&rvd
->vdev_stat_lock
);
2886 ASSERT(rvd
== vd
->vdev_parent
);
2887 ASSERT(vd
->vdev_ms_count
!= 0);
2889 metaslab_class_space_update(mc
,
2890 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2895 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2896 * so that it will be written out next time the vdev configuration is synced.
2897 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2900 vdev_config_dirty(vdev_t
*vd
)
2902 spa_t
*spa
= vd
->vdev_spa
;
2903 vdev_t
*rvd
= spa
->spa_root_vdev
;
2906 ASSERT(spa_writeable(spa
));
2909 * If this is an aux vdev (as with l2cache and spare devices), then we
2910 * update the vdev config manually and set the sync flag.
2912 if (vd
->vdev_aux
!= NULL
) {
2913 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2917 for (c
= 0; c
< sav
->sav_count
; c
++) {
2918 if (sav
->sav_vdevs
[c
] == vd
)
2922 if (c
== sav
->sav_count
) {
2924 * We're being removed. There's nothing more to do.
2926 ASSERT(sav
->sav_sync
== B_TRUE
);
2930 sav
->sav_sync
= B_TRUE
;
2932 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2933 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2934 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2935 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2941 * Setting the nvlist in the middle if the array is a little
2942 * sketchy, but it will work.
2944 nvlist_free(aux
[c
]);
2945 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
2951 * The dirty list is protected by the SCL_CONFIG lock. The caller
2952 * must either hold SCL_CONFIG as writer, or must be the sync thread
2953 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2954 * so this is sufficient to ensure mutual exclusion.
2956 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2957 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2958 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2961 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2962 vdev_config_dirty(rvd
->vdev_child
[c
]);
2964 ASSERT(vd
== vd
->vdev_top
);
2966 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
2968 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
2973 vdev_config_clean(vdev_t
*vd
)
2975 spa_t
*spa
= vd
->vdev_spa
;
2977 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2978 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2979 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2981 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
2982 list_remove(&spa
->spa_config_dirty_list
, vd
);
2986 * Mark a top-level vdev's state as dirty, so that the next pass of
2987 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2988 * the state changes from larger config changes because they require
2989 * much less locking, and are often needed for administrative actions.
2992 vdev_state_dirty(vdev_t
*vd
)
2994 spa_t
*spa
= vd
->vdev_spa
;
2996 ASSERT(spa_writeable(spa
));
2997 ASSERT(vd
== vd
->vdev_top
);
3000 * The state list is protected by the SCL_STATE lock. The caller
3001 * must either hold SCL_STATE as writer, or must be the sync thread
3002 * (which holds SCL_STATE as reader). There's only one sync thread,
3003 * so this is sufficient to ensure mutual exclusion.
3005 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3006 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3007 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3009 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3010 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3014 vdev_state_clean(vdev_t
*vd
)
3016 spa_t
*spa
= vd
->vdev_spa
;
3018 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3019 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3020 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3022 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3023 list_remove(&spa
->spa_state_dirty_list
, vd
);
3027 * Propagate vdev state up from children to parent.
3030 vdev_propagate_state(vdev_t
*vd
)
3032 spa_t
*spa
= vd
->vdev_spa
;
3033 vdev_t
*rvd
= spa
->spa_root_vdev
;
3034 int degraded
= 0, faulted
= 0;
3039 if (vd
->vdev_children
> 0) {
3040 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3041 child
= vd
->vdev_child
[c
];
3044 * Don't factor holes into the decision.
3046 if (child
->vdev_ishole
)
3049 if (!vdev_readable(child
) ||
3050 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3052 * Root special: if there is a top-level log
3053 * device, treat the root vdev as if it were
3056 if (child
->vdev_islog
&& vd
== rvd
)
3060 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3064 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3068 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3071 * Root special: if there is a top-level vdev that cannot be
3072 * opened due to corrupted metadata, then propagate the root
3073 * vdev's aux state as 'corrupt' rather than 'insufficient
3076 if (corrupted
&& vd
== rvd
&&
3077 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3078 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3079 VDEV_AUX_CORRUPT_DATA
);
3082 if (vd
->vdev_parent
)
3083 vdev_propagate_state(vd
->vdev_parent
);
3087 * Set a vdev's state. If this is during an open, we don't update the parent
3088 * state, because we're in the process of opening children depth-first.
3089 * Otherwise, we propagate the change to the parent.
3091 * If this routine places a device in a faulted state, an appropriate ereport is
3095 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3097 uint64_t save_state
;
3098 spa_t
*spa
= vd
->vdev_spa
;
3100 if (state
== vd
->vdev_state
) {
3101 vd
->vdev_stat
.vs_aux
= aux
;
3105 save_state
= vd
->vdev_state
;
3107 vd
->vdev_state
= state
;
3108 vd
->vdev_stat
.vs_aux
= aux
;
3111 * If we are setting the vdev state to anything but an open state, then
3112 * always close the underlying device unless the device has requested
3113 * a delayed close (i.e. we're about to remove or fault the device).
3114 * Otherwise, we keep accessible but invalid devices open forever.
3115 * We don't call vdev_close() itself, because that implies some extra
3116 * checks (offline, etc) that we don't want here. This is limited to
3117 * leaf devices, because otherwise closing the device will affect other
3120 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3121 vd
->vdev_ops
->vdev_op_leaf
)
3122 vd
->vdev_ops
->vdev_op_close(vd
);
3125 * If we have brought this vdev back into service, we need
3126 * to notify fmd so that it can gracefully repair any outstanding
3127 * cases due to a missing device. We do this in all cases, even those
3128 * that probably don't correlate to a repaired fault. This is sure to
3129 * catch all cases, and we let the zfs-retire agent sort it out. If
3130 * this is a transient state it's OK, as the retire agent will
3131 * double-check the state of the vdev before repairing it.
3133 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
3134 vd
->vdev_prevstate
!= state
)
3135 zfs_post_state_change(spa
, vd
);
3137 if (vd
->vdev_removed
&&
3138 state
== VDEV_STATE_CANT_OPEN
&&
3139 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3141 * If the previous state is set to VDEV_STATE_REMOVED, then this
3142 * device was previously marked removed and someone attempted to
3143 * reopen it. If this failed due to a nonexistent device, then
3144 * keep the device in the REMOVED state. We also let this be if
3145 * it is one of our special test online cases, which is only
3146 * attempting to online the device and shouldn't generate an FMA
3149 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3150 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3151 } else if (state
== VDEV_STATE_REMOVED
) {
3152 vd
->vdev_removed
= B_TRUE
;
3153 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3155 * If we fail to open a vdev during an import or recovery, we
3156 * mark it as "not available", which signifies that it was
3157 * never there to begin with. Failure to open such a device
3158 * is not considered an error.
3160 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3161 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3162 vd
->vdev_ops
->vdev_op_leaf
)
3163 vd
->vdev_not_present
= 1;
3166 * Post the appropriate ereport. If the 'prevstate' field is
3167 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3168 * that this is part of a vdev_reopen(). In this case, we don't
3169 * want to post the ereport if the device was already in the
3170 * CANT_OPEN state beforehand.
3172 * If the 'checkremove' flag is set, then this is an attempt to
3173 * online the device in response to an insertion event. If we
3174 * hit this case, then we have detected an insertion event for a
3175 * faulted or offline device that wasn't in the removed state.
3176 * In this scenario, we don't post an ereport because we are
3177 * about to replace the device, or attempt an online with
3178 * vdev_forcefault, which will generate the fault for us.
3180 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3181 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3182 vd
!= spa
->spa_root_vdev
) {
3186 case VDEV_AUX_OPEN_FAILED
:
3187 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3189 case VDEV_AUX_CORRUPT_DATA
:
3190 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3192 case VDEV_AUX_NO_REPLICAS
:
3193 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3195 case VDEV_AUX_BAD_GUID_SUM
:
3196 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3198 case VDEV_AUX_TOO_SMALL
:
3199 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3201 case VDEV_AUX_BAD_LABEL
:
3202 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3205 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3208 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3211 /* Erase any notion of persistent removed state */
3212 vd
->vdev_removed
= B_FALSE
;
3214 vd
->vdev_removed
= B_FALSE
;
3217 if (!isopen
&& vd
->vdev_parent
)
3218 vdev_propagate_state(vd
->vdev_parent
);
3222 * Check the vdev configuration to ensure that it's capable of supporting
3226 vdev_is_bootable(vdev_t
*vd
)
3228 #if defined(__sun__) || defined(__sun)
3230 * Currently, we do not support RAID-Z or partial configuration.
3231 * In addition, only a single top-level vdev is allowed and none of the
3232 * leaves can be wholedisks.
3236 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3237 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3239 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3240 vd
->vdev_children
> 1) {
3242 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3243 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3246 } else if (vd
->vdev_wholedisk
== 1) {
3250 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3251 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3254 #endif /* __sun__ || __sun */
3259 * Load the state from the original vdev tree (ovd) which
3260 * we've retrieved from the MOS config object. If the original
3261 * vdev was offline or faulted then we transfer that state to the
3262 * device in the current vdev tree (nvd).
3265 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3269 ASSERT(nvd
->vdev_top
->vdev_islog
);
3270 ASSERT(spa_config_held(nvd
->vdev_spa
,
3271 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3272 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3274 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3275 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3277 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3279 * Restore the persistent vdev state
3281 nvd
->vdev_offline
= ovd
->vdev_offline
;
3282 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3283 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3284 nvd
->vdev_removed
= ovd
->vdev_removed
;
3289 * Determine if a log device has valid content. If the vdev was
3290 * removed or faulted in the MOS config then we know that
3291 * the content on the log device has already been written to the pool.
3294 vdev_log_state_valid(vdev_t
*vd
)
3298 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3302 for (c
= 0; c
< vd
->vdev_children
; c
++)
3303 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3310 * Expand a vdev if possible.
3313 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3315 ASSERT(vd
->vdev_top
== vd
);
3316 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3318 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3319 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3320 vdev_config_dirty(vd
);
3328 vdev_split(vdev_t
*vd
)
3330 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3332 vdev_remove_child(pvd
, vd
);
3333 vdev_compact_children(pvd
);
3335 cvd
= pvd
->vdev_child
[0];
3336 if (pvd
->vdev_children
== 1) {
3337 vdev_remove_parent(cvd
);
3338 cvd
->vdev_splitting
= B_TRUE
;
3340 vdev_propagate_state(cvd
);
3344 vdev_deadman(vdev_t
*vd
)
3348 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3349 vdev_t
*cvd
= vd
->vdev_child
[c
];
3354 if (vd
->vdev_ops
->vdev_op_leaf
) {
3355 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3357 mutex_enter(&vq
->vq_lock
);
3358 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3359 spa_t
*spa
= vd
->vdev_spa
;
3364 * Look at the head of all the pending queues,
3365 * if any I/O has been outstanding for longer than
3366 * the spa_deadman_synctime we log a zevent.
3368 fio
= avl_first(&vq
->vq_active_tree
);
3369 delta
= gethrtime() - fio
->io_timestamp
;
3370 if (delta
> spa_deadman_synctime(spa
)) {
3371 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3372 "delta %lluns, last io %lluns",
3373 fio
->io_timestamp
, delta
,
3374 vq
->vq_io_complete_ts
);
3375 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3376 spa
, vd
, fio
, 0, 0);
3379 mutex_exit(&vq
->vq_lock
);
3383 #if defined(_KERNEL) && defined(HAVE_SPL)
3384 EXPORT_SYMBOL(vdev_fault
);
3385 EXPORT_SYMBOL(vdev_degrade
);
3386 EXPORT_SYMBOL(vdev_online
);
3387 EXPORT_SYMBOL(vdev_offline
);
3388 EXPORT_SYMBOL(vdev_clear
);