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, 2015 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 * When a vdev is added, it will be divided into approximately (but no
50 * more than) this number of metaslabs.
52 int metaslabs_per_vdev
= 200;
55 * Virtual device management.
58 static vdev_ops_t
*vdev_ops_table
[] = {
72 * Given a vdev type, return the appropriate ops vector.
75 vdev_getops(const char *type
)
77 vdev_ops_t
*ops
, **opspp
;
79 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
80 if (strcmp(ops
->vdev_op_type
, type
) == 0)
87 * Default asize function: return the MAX of psize with the asize of
88 * all children. This is what's used by anything other than RAID-Z.
91 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
93 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
97 for (c
= 0; c
< vd
->vdev_children
; c
++) {
98 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
99 asize
= MAX(asize
, csize
);
106 * Get the minimum allocatable size. We define the allocatable size as
107 * the vdev's asize rounded to the nearest metaslab. This allows us to
108 * replace or attach devices which don't have the same physical size but
109 * can still satisfy the same number of allocations.
112 vdev_get_min_asize(vdev_t
*vd
)
114 vdev_t
*pvd
= vd
->vdev_parent
;
117 * If our parent is NULL (inactive spare or cache) or is the root,
118 * just return our own asize.
121 return (vd
->vdev_asize
);
124 * The top-level vdev just returns the allocatable size rounded
125 * to the nearest metaslab.
127 if (vd
== vd
->vdev_top
)
128 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
131 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
132 * so each child must provide at least 1/Nth of its asize.
134 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
135 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
137 return (pvd
->vdev_min_asize
);
141 vdev_set_min_asize(vdev_t
*vd
)
144 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
146 for (c
= 0; c
< vd
->vdev_children
; c
++)
147 vdev_set_min_asize(vd
->vdev_child
[c
]);
151 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
153 vdev_t
*rvd
= spa
->spa_root_vdev
;
155 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
157 if (vdev
< rvd
->vdev_children
) {
158 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
159 return (rvd
->vdev_child
[vdev
]);
166 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
171 if (vd
->vdev_guid
== guid
)
174 for (c
= 0; c
< vd
->vdev_children
; c
++)
175 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
183 vdev_count_leaves_impl(vdev_t
*vd
)
188 if (vd
->vdev_ops
->vdev_op_leaf
)
191 for (c
= 0; c
< vd
->vdev_children
; c
++)
192 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
198 vdev_count_leaves(spa_t
*spa
)
200 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
204 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
206 size_t oldsize
, newsize
;
207 uint64_t id
= cvd
->vdev_id
;
210 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
211 ASSERT(cvd
->vdev_parent
== NULL
);
213 cvd
->vdev_parent
= pvd
;
218 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
220 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
221 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
222 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
224 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
225 if (pvd
->vdev_child
!= NULL
) {
226 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
227 kmem_free(pvd
->vdev_child
, oldsize
);
230 pvd
->vdev_child
= newchild
;
231 pvd
->vdev_child
[id
] = cvd
;
233 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
234 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
237 * Walk up all ancestors to update guid sum.
239 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
240 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
244 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
247 uint_t id
= cvd
->vdev_id
;
249 ASSERT(cvd
->vdev_parent
== pvd
);
254 ASSERT(id
< pvd
->vdev_children
);
255 ASSERT(pvd
->vdev_child
[id
] == cvd
);
257 pvd
->vdev_child
[id
] = NULL
;
258 cvd
->vdev_parent
= NULL
;
260 for (c
= 0; c
< pvd
->vdev_children
; c
++)
261 if (pvd
->vdev_child
[c
])
264 if (c
== pvd
->vdev_children
) {
265 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
266 pvd
->vdev_child
= NULL
;
267 pvd
->vdev_children
= 0;
271 * Walk up all ancestors to update guid sum.
273 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
274 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
278 * Remove any holes in the child array.
281 vdev_compact_children(vdev_t
*pvd
)
283 vdev_t
**newchild
, *cvd
;
284 int oldc
= pvd
->vdev_children
;
288 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
290 for (c
= newc
= 0; c
< oldc
; c
++)
291 if (pvd
->vdev_child
[c
])
294 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
296 for (c
= newc
= 0; c
< oldc
; c
++) {
297 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
298 newchild
[newc
] = cvd
;
299 cvd
->vdev_id
= newc
++;
303 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
304 pvd
->vdev_child
= newchild
;
305 pvd
->vdev_children
= newc
;
309 * Allocate and minimally initialize a vdev_t.
312 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
317 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
319 if (spa
->spa_root_vdev
== NULL
) {
320 ASSERT(ops
== &vdev_root_ops
);
321 spa
->spa_root_vdev
= vd
;
322 spa
->spa_load_guid
= spa_generate_guid(NULL
);
325 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
326 if (spa
->spa_root_vdev
== vd
) {
328 * The root vdev's guid will also be the pool guid,
329 * which must be unique among all pools.
331 guid
= spa_generate_guid(NULL
);
334 * Any other vdev's guid must be unique within the pool.
336 guid
= spa_generate_guid(spa
);
338 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
343 vd
->vdev_guid
= guid
;
344 vd
->vdev_guid_sum
= guid
;
346 vd
->vdev_state
= VDEV_STATE_CLOSED
;
347 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
349 list_link_init(&vd
->vdev_config_dirty_node
);
350 list_link_init(&vd
->vdev_state_dirty_node
);
351 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
352 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
353 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
354 for (t
= 0; t
< DTL_TYPES
; t
++) {
355 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
358 txg_list_create(&vd
->vdev_ms_list
,
359 offsetof(struct metaslab
, ms_txg_node
));
360 txg_list_create(&vd
->vdev_dtl_list
,
361 offsetof(struct vdev
, vdev_dtl_node
));
362 vd
->vdev_stat
.vs_timestamp
= gethrtime();
370 * Allocate a new vdev. The 'alloctype' is used to control whether we are
371 * creating a new vdev or loading an existing one - the behavior is slightly
372 * different for each case.
375 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
380 uint64_t guid
= 0, islog
, nparity
;
383 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
385 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
386 return (SET_ERROR(EINVAL
));
388 if ((ops
= vdev_getops(type
)) == NULL
)
389 return (SET_ERROR(EINVAL
));
392 * If this is a load, get the vdev guid from the nvlist.
393 * Otherwise, vdev_alloc_common() will generate one for us.
395 if (alloctype
== VDEV_ALLOC_LOAD
) {
398 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
400 return (SET_ERROR(EINVAL
));
402 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
403 return (SET_ERROR(EINVAL
));
404 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
405 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
406 return (SET_ERROR(EINVAL
));
407 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
408 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
409 return (SET_ERROR(EINVAL
));
410 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
411 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
412 return (SET_ERROR(EINVAL
));
416 * The first allocated vdev must be of type 'root'.
418 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
419 return (SET_ERROR(EINVAL
));
422 * Determine whether we're a log vdev.
425 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
426 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
427 return (SET_ERROR(ENOTSUP
));
429 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
430 return (SET_ERROR(ENOTSUP
));
433 * Set the nparity property for RAID-Z vdevs.
436 if (ops
== &vdev_raidz_ops
) {
437 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
439 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
440 return (SET_ERROR(EINVAL
));
442 * Previous versions could only support 1 or 2 parity
446 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
447 return (SET_ERROR(ENOTSUP
));
449 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
450 return (SET_ERROR(ENOTSUP
));
453 * We require the parity to be specified for SPAs that
454 * support multiple parity levels.
456 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
457 return (SET_ERROR(EINVAL
));
459 * Otherwise, we default to 1 parity device for RAID-Z.
466 ASSERT(nparity
!= -1ULL);
468 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
470 vd
->vdev_islog
= islog
;
471 vd
->vdev_nparity
= nparity
;
473 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
474 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
475 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
476 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
477 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
478 &vd
->vdev_physpath
) == 0)
479 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
480 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
481 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
484 * Set the whole_disk property. If it's not specified, leave the value
487 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
488 &vd
->vdev_wholedisk
) != 0)
489 vd
->vdev_wholedisk
= -1ULL;
492 * Look for the 'not present' flag. This will only be set if the device
493 * was not present at the time of import.
495 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
496 &vd
->vdev_not_present
);
499 * Get the alignment requirement.
501 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
504 * Retrieve the vdev creation time.
506 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
510 * If we're a top-level vdev, try to load the allocation parameters.
512 if (parent
&& !parent
->vdev_parent
&&
513 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
514 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
516 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
518 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
520 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
524 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
525 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
526 alloctype
== VDEV_ALLOC_ADD
||
527 alloctype
== VDEV_ALLOC_SPLIT
||
528 alloctype
== VDEV_ALLOC_ROOTPOOL
);
529 vd
->vdev_mg
= metaslab_group_create(islog
?
530 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
534 * If we're a leaf vdev, try to load the DTL object and other state.
536 if (vd
->vdev_ops
->vdev_op_leaf
&&
537 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
538 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
539 if (alloctype
== VDEV_ALLOC_LOAD
) {
540 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
541 &vd
->vdev_dtl_object
);
542 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
546 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
549 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
550 &spare
) == 0 && spare
)
554 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
557 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
558 &vd
->vdev_resilver_txg
);
561 * When importing a pool, we want to ignore the persistent fault
562 * state, as the diagnosis made on another system may not be
563 * valid in the current context. Local vdevs will
564 * remain in the faulted state.
566 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
567 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
569 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
571 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
574 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
578 VDEV_AUX_ERR_EXCEEDED
;
579 if (nvlist_lookup_string(nv
,
580 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
581 strcmp(aux
, "external") == 0)
582 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
588 * Add ourselves to the parent's list of children.
590 vdev_add_child(parent
, vd
);
598 vdev_free(vdev_t
*vd
)
601 spa_t
*spa
= vd
->vdev_spa
;
604 * vdev_free() implies closing the vdev first. This is simpler than
605 * trying to ensure complicated semantics for all callers.
609 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
610 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
615 for (c
= 0; c
< vd
->vdev_children
; c
++)
616 vdev_free(vd
->vdev_child
[c
]);
618 ASSERT(vd
->vdev_child
== NULL
);
619 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
622 * Discard allocation state.
624 if (vd
->vdev_mg
!= NULL
) {
625 vdev_metaslab_fini(vd
);
626 metaslab_group_destroy(vd
->vdev_mg
);
629 ASSERT0(vd
->vdev_stat
.vs_space
);
630 ASSERT0(vd
->vdev_stat
.vs_dspace
);
631 ASSERT0(vd
->vdev_stat
.vs_alloc
);
634 * Remove this vdev from its parent's child list.
636 vdev_remove_child(vd
->vdev_parent
, vd
);
638 ASSERT(vd
->vdev_parent
== NULL
);
641 * Clean up vdev structure.
647 spa_strfree(vd
->vdev_path
);
649 spa_strfree(vd
->vdev_devid
);
650 if (vd
->vdev_physpath
)
651 spa_strfree(vd
->vdev_physpath
);
653 spa_strfree(vd
->vdev_fru
);
655 if (vd
->vdev_isspare
)
656 spa_spare_remove(vd
);
657 if (vd
->vdev_isl2cache
)
658 spa_l2cache_remove(vd
);
660 txg_list_destroy(&vd
->vdev_ms_list
);
661 txg_list_destroy(&vd
->vdev_dtl_list
);
663 mutex_enter(&vd
->vdev_dtl_lock
);
664 space_map_close(vd
->vdev_dtl_sm
);
665 for (t
= 0; t
< DTL_TYPES
; t
++) {
666 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
667 range_tree_destroy(vd
->vdev_dtl
[t
]);
669 mutex_exit(&vd
->vdev_dtl_lock
);
671 mutex_destroy(&vd
->vdev_dtl_lock
);
672 mutex_destroy(&vd
->vdev_stat_lock
);
673 mutex_destroy(&vd
->vdev_probe_lock
);
675 if (vd
== spa
->spa_root_vdev
)
676 spa
->spa_root_vdev
= NULL
;
678 kmem_free(vd
, sizeof (vdev_t
));
682 * Transfer top-level vdev state from svd to tvd.
685 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
687 spa_t
*spa
= svd
->vdev_spa
;
692 ASSERT(tvd
== tvd
->vdev_top
);
694 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
695 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
696 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
698 svd
->vdev_ms_array
= 0;
699 svd
->vdev_ms_shift
= 0;
700 svd
->vdev_ms_count
= 0;
703 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
704 tvd
->vdev_mg
= svd
->vdev_mg
;
705 tvd
->vdev_ms
= svd
->vdev_ms
;
710 if (tvd
->vdev_mg
!= NULL
)
711 tvd
->vdev_mg
->mg_vd
= tvd
;
713 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
714 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
715 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
717 svd
->vdev_stat
.vs_alloc
= 0;
718 svd
->vdev_stat
.vs_space
= 0;
719 svd
->vdev_stat
.vs_dspace
= 0;
721 for (t
= 0; t
< TXG_SIZE
; t
++) {
722 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
723 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
724 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
725 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
726 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
727 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
730 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
731 vdev_config_clean(svd
);
732 vdev_config_dirty(tvd
);
735 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
736 vdev_state_clean(svd
);
737 vdev_state_dirty(tvd
);
740 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
741 svd
->vdev_deflate_ratio
= 0;
743 tvd
->vdev_islog
= svd
->vdev_islog
;
748 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
757 for (c
= 0; c
< vd
->vdev_children
; c
++)
758 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
762 * Add a mirror/replacing vdev above an existing vdev.
765 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
767 spa_t
*spa
= cvd
->vdev_spa
;
768 vdev_t
*pvd
= cvd
->vdev_parent
;
771 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
773 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
775 mvd
->vdev_asize
= cvd
->vdev_asize
;
776 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
777 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
778 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
779 mvd
->vdev_state
= cvd
->vdev_state
;
780 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
782 vdev_remove_child(pvd
, cvd
);
783 vdev_add_child(pvd
, mvd
);
784 cvd
->vdev_id
= mvd
->vdev_children
;
785 vdev_add_child(mvd
, cvd
);
786 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
788 if (mvd
== mvd
->vdev_top
)
789 vdev_top_transfer(cvd
, mvd
);
795 * Remove a 1-way mirror/replacing vdev from the tree.
798 vdev_remove_parent(vdev_t
*cvd
)
800 vdev_t
*mvd
= cvd
->vdev_parent
;
801 vdev_t
*pvd
= mvd
->vdev_parent
;
803 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
805 ASSERT(mvd
->vdev_children
== 1);
806 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
807 mvd
->vdev_ops
== &vdev_replacing_ops
||
808 mvd
->vdev_ops
== &vdev_spare_ops
);
809 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
811 vdev_remove_child(mvd
, cvd
);
812 vdev_remove_child(pvd
, mvd
);
815 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
816 * Otherwise, we could have detached an offline device, and when we
817 * go to import the pool we'll think we have two top-level vdevs,
818 * instead of a different version of the same top-level vdev.
820 if (mvd
->vdev_top
== mvd
) {
821 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
822 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
823 cvd
->vdev_guid
+= guid_delta
;
824 cvd
->vdev_guid_sum
+= guid_delta
;
827 * If pool not set for autoexpand, we need to also preserve
828 * mvd's asize to prevent automatic expansion of cvd.
829 * Otherwise if we are adjusting the mirror by attaching and
830 * detaching children of non-uniform sizes, the mirror could
831 * autoexpand, unexpectedly requiring larger devices to
832 * re-establish the mirror.
834 if (!cvd
->vdev_spa
->spa_autoexpand
)
835 cvd
->vdev_asize
= mvd
->vdev_asize
;
837 cvd
->vdev_id
= mvd
->vdev_id
;
838 vdev_add_child(pvd
, cvd
);
839 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
841 if (cvd
== cvd
->vdev_top
)
842 vdev_top_transfer(mvd
, cvd
);
844 ASSERT(mvd
->vdev_children
== 0);
849 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
851 spa_t
*spa
= vd
->vdev_spa
;
852 objset_t
*mos
= spa
->spa_meta_objset
;
854 uint64_t oldc
= vd
->vdev_ms_count
;
855 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
859 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
862 * This vdev is not being allocated from yet or is a hole.
864 if (vd
->vdev_ms_shift
== 0)
867 ASSERT(!vd
->vdev_ishole
);
870 * Compute the raidz-deflation ratio. Note, we hard-code
871 * in 128k (1 << 17) because it is the "typical" blocksize.
872 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
873 * otherwise it would inconsistently account for existing bp's.
875 vd
->vdev_deflate_ratio
= (1 << 17) /
876 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
878 ASSERT(oldc
<= newc
);
880 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
883 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
884 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
888 vd
->vdev_ms_count
= newc
;
890 for (m
= oldc
; m
< newc
; m
++) {
894 error
= dmu_read(mos
, vd
->vdev_ms_array
,
895 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
901 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
908 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
911 * If the vdev is being removed we don't activate
912 * the metaslabs since we want to ensure that no new
913 * allocations are performed on this device.
915 if (oldc
== 0 && !vd
->vdev_removing
)
916 metaslab_group_activate(vd
->vdev_mg
);
919 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
925 vdev_metaslab_fini(vdev_t
*vd
)
928 uint64_t count
= vd
->vdev_ms_count
;
930 if (vd
->vdev_ms
!= NULL
) {
931 metaslab_group_passivate(vd
->vdev_mg
);
932 for (m
= 0; m
< count
; m
++) {
933 metaslab_t
*msp
= vd
->vdev_ms
[m
];
938 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
942 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
945 typedef struct vdev_probe_stats
{
946 boolean_t vps_readable
;
947 boolean_t vps_writeable
;
949 } vdev_probe_stats_t
;
952 vdev_probe_done(zio_t
*zio
)
954 spa_t
*spa
= zio
->io_spa
;
955 vdev_t
*vd
= zio
->io_vd
;
956 vdev_probe_stats_t
*vps
= zio
->io_private
;
958 ASSERT(vd
->vdev_probe_zio
!= NULL
);
960 if (zio
->io_type
== ZIO_TYPE_READ
) {
961 if (zio
->io_error
== 0)
962 vps
->vps_readable
= 1;
963 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
964 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
965 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
966 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
967 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
969 zio_buf_free(zio
->io_data
, zio
->io_size
);
971 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
972 if (zio
->io_error
== 0)
973 vps
->vps_writeable
= 1;
974 zio_buf_free(zio
->io_data
, zio
->io_size
);
975 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
978 vd
->vdev_cant_read
|= !vps
->vps_readable
;
979 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
981 if (vdev_readable(vd
) &&
982 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
985 ASSERT(zio
->io_error
!= 0);
986 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
987 spa
, vd
, NULL
, 0, 0);
988 zio
->io_error
= SET_ERROR(ENXIO
);
991 mutex_enter(&vd
->vdev_probe_lock
);
992 ASSERT(vd
->vdev_probe_zio
== zio
);
993 vd
->vdev_probe_zio
= NULL
;
994 mutex_exit(&vd
->vdev_probe_lock
);
996 while ((pio
= zio_walk_parents(zio
)) != NULL
)
997 if (!vdev_accessible(vd
, pio
))
998 pio
->io_error
= SET_ERROR(ENXIO
);
1000 kmem_free(vps
, sizeof (*vps
));
1005 * Determine whether this device is accessible.
1007 * Read and write to several known locations: the pad regions of each
1008 * vdev label but the first, which we leave alone in case it contains
1012 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1014 spa_t
*spa
= vd
->vdev_spa
;
1015 vdev_probe_stats_t
*vps
= NULL
;
1019 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1022 * Don't probe the probe.
1024 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1028 * To prevent 'probe storms' when a device fails, we create
1029 * just one probe i/o at a time. All zios that want to probe
1030 * this vdev will become parents of the probe io.
1032 mutex_enter(&vd
->vdev_probe_lock
);
1034 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1035 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1037 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1038 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1041 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1043 * vdev_cant_read and vdev_cant_write can only
1044 * transition from TRUE to FALSE when we have the
1045 * SCL_ZIO lock as writer; otherwise they can only
1046 * transition from FALSE to TRUE. This ensures that
1047 * any zio looking at these values can assume that
1048 * failures persist for the life of the I/O. That's
1049 * important because when a device has intermittent
1050 * connectivity problems, we want to ensure that
1051 * they're ascribed to the device (ENXIO) and not
1054 * Since we hold SCL_ZIO as writer here, clear both
1055 * values so the probe can reevaluate from first
1058 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1059 vd
->vdev_cant_read
= B_FALSE
;
1060 vd
->vdev_cant_write
= B_FALSE
;
1063 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1064 vdev_probe_done
, vps
,
1065 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1068 * We can't change the vdev state in this context, so we
1069 * kick off an async task to do it on our behalf.
1072 vd
->vdev_probe_wanted
= B_TRUE
;
1073 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1078 zio_add_child(zio
, pio
);
1080 mutex_exit(&vd
->vdev_probe_lock
);
1083 ASSERT(zio
!= NULL
);
1087 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1088 zio_nowait(zio_read_phys(pio
, vd
,
1089 vdev_label_offset(vd
->vdev_psize
, l
,
1090 offsetof(vdev_label_t
, vl_pad2
)),
1091 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1092 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1093 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1104 vdev_open_child(void *arg
)
1108 vd
->vdev_open_thread
= curthread
;
1109 vd
->vdev_open_error
= vdev_open(vd
);
1110 vd
->vdev_open_thread
= NULL
;
1111 vd
->vdev_parent
->vdev_nonrot
&= vd
->vdev_nonrot
;
1115 vdev_uses_zvols(vdev_t
*vd
)
1120 if (zvol_is_zvol(vd
->vdev_path
))
1124 for (c
= 0; c
< vd
->vdev_children
; c
++)
1125 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1132 vdev_open_children(vdev_t
*vd
)
1135 int children
= vd
->vdev_children
;
1138 vd
->vdev_nonrot
= B_TRUE
;
1141 * in order to handle pools on top of zvols, do the opens
1142 * in a single thread so that the same thread holds the
1143 * spa_namespace_lock
1145 if (vdev_uses_zvols(vd
)) {
1146 for (c
= 0; c
< children
; c
++) {
1147 vd
->vdev_child
[c
]->vdev_open_error
=
1148 vdev_open(vd
->vdev_child
[c
]);
1149 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1153 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1154 children
, children
, TASKQ_PREPOPULATE
);
1156 for (c
= 0; c
< children
; c
++)
1157 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1162 for (c
= 0; c
< children
; c
++)
1163 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1167 * Prepare a virtual device for access.
1170 vdev_open(vdev_t
*vd
)
1172 spa_t
*spa
= vd
->vdev_spa
;
1175 uint64_t max_osize
= 0;
1176 uint64_t asize
, max_asize
, psize
;
1177 uint64_t ashift
= 0;
1180 ASSERT(vd
->vdev_open_thread
== curthread
||
1181 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1182 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1183 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1184 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1186 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1187 vd
->vdev_cant_read
= B_FALSE
;
1188 vd
->vdev_cant_write
= B_FALSE
;
1189 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1192 * If this vdev is not removed, check its fault status. If it's
1193 * faulted, bail out of the open.
1195 if (!vd
->vdev_removed
&& 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
));
1202 } else if (vd
->vdev_offline
) {
1203 ASSERT(vd
->vdev_children
== 0);
1204 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1205 return (SET_ERROR(ENXIO
));
1208 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1211 * Reset the vdev_reopening flag so that we actually close
1212 * the vdev on error.
1214 vd
->vdev_reopening
= B_FALSE
;
1215 if (zio_injection_enabled
&& error
== 0)
1216 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1219 if (vd
->vdev_removed
&&
1220 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1221 vd
->vdev_removed
= B_FALSE
;
1223 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1224 vd
->vdev_stat
.vs_aux
);
1228 vd
->vdev_removed
= B_FALSE
;
1231 * Recheck the faulted flag now that we have confirmed that
1232 * the vdev is accessible. If we're faulted, bail.
1234 if (vd
->vdev_faulted
) {
1235 ASSERT(vd
->vdev_children
== 0);
1236 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1237 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1238 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1239 vd
->vdev_label_aux
);
1240 return (SET_ERROR(ENXIO
));
1243 if (vd
->vdev_degraded
) {
1244 ASSERT(vd
->vdev_children
== 0);
1245 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1246 VDEV_AUX_ERR_EXCEEDED
);
1248 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1252 * For hole or missing vdevs we just return success.
1254 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1257 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1258 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1259 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1265 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1266 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1268 if (vd
->vdev_children
== 0) {
1269 if (osize
< SPA_MINDEVSIZE
) {
1270 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1271 VDEV_AUX_TOO_SMALL
);
1272 return (SET_ERROR(EOVERFLOW
));
1275 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1276 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1277 VDEV_LABEL_END_SIZE
);
1279 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1280 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1281 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1282 VDEV_AUX_TOO_SMALL
);
1283 return (SET_ERROR(EOVERFLOW
));
1287 max_asize
= max_osize
;
1290 vd
->vdev_psize
= psize
;
1293 * Make sure the allocatable size hasn't shrunk.
1295 if (asize
< vd
->vdev_min_asize
) {
1296 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1297 VDEV_AUX_BAD_LABEL
);
1298 return (SET_ERROR(EINVAL
));
1301 if (vd
->vdev_asize
== 0) {
1303 * This is the first-ever open, so use the computed values.
1304 * For compatibility, a different ashift can be requested.
1306 vd
->vdev_asize
= asize
;
1307 vd
->vdev_max_asize
= max_asize
;
1308 if (vd
->vdev_ashift
== 0)
1309 vd
->vdev_ashift
= ashift
;
1312 * Detect if the alignment requirement has increased.
1313 * We don't want to make the pool unavailable, just
1314 * post an event instead.
1316 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1317 vd
->vdev_ops
->vdev_op_leaf
) {
1318 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1319 spa
, vd
, NULL
, 0, 0);
1322 vd
->vdev_max_asize
= max_asize
;
1326 * If all children are healthy and the asize has increased,
1327 * then we've experienced dynamic LUN growth. If automatic
1328 * expansion is enabled then use the additional space.
1330 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1331 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1332 vd
->vdev_asize
= asize
;
1334 vdev_set_min_asize(vd
);
1337 * Ensure we can issue some IO before declaring the
1338 * vdev open for business.
1340 if (vd
->vdev_ops
->vdev_op_leaf
&&
1341 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1342 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1343 VDEV_AUX_ERR_EXCEEDED
);
1348 * Track the min and max ashift values for normal data devices.
1350 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1351 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1352 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1353 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1354 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1355 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1359 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1360 * resilver. But don't do this if we are doing a reopen for a scrub,
1361 * since this would just restart the scrub we are already doing.
1363 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1364 vdev_resilver_needed(vd
, NULL
, NULL
))
1365 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1371 * Called once the vdevs are all opened, this routine validates the label
1372 * contents. This needs to be done before vdev_load() so that we don't
1373 * inadvertently do repair I/Os to the wrong device.
1375 * If 'strict' is false ignore the spa guid check. This is necessary because
1376 * if the machine crashed during a re-guid the new guid might have been written
1377 * to all of the vdev labels, but not the cached config. The strict check
1378 * will be performed when the pool is opened again using the mos config.
1380 * This function will only return failure if one of the vdevs indicates that it
1381 * has since been destroyed or exported. This is only possible if
1382 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1383 * will be updated but the function will return 0.
1386 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1388 spa_t
*spa
= vd
->vdev_spa
;
1390 uint64_t guid
= 0, top_guid
;
1394 for (c
= 0; c
< vd
->vdev_children
; c
++)
1395 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1396 return (SET_ERROR(EBADF
));
1399 * If the device has already failed, or was marked offline, don't do
1400 * any further validation. Otherwise, label I/O will fail and we will
1401 * overwrite the previous state.
1403 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1404 uint64_t aux_guid
= 0;
1406 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1407 spa_last_synced_txg(spa
) : -1ULL;
1409 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1410 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1411 VDEV_AUX_BAD_LABEL
);
1416 * Determine if this vdev has been split off into another
1417 * pool. If so, then refuse to open it.
1419 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1420 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1421 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1422 VDEV_AUX_SPLIT_POOL
);
1427 if (strict
&& (nvlist_lookup_uint64(label
,
1428 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1429 guid
!= spa_guid(spa
))) {
1430 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1431 VDEV_AUX_CORRUPT_DATA
);
1436 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1437 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1442 * If this vdev just became a top-level vdev because its
1443 * sibling was detached, it will have adopted the parent's
1444 * vdev guid -- but the label may or may not be on disk yet.
1445 * Fortunately, either version of the label will have the
1446 * same top guid, so if we're a top-level vdev, we can
1447 * safely compare to that instead.
1449 * If we split this vdev off instead, then we also check the
1450 * original pool's guid. We don't want to consider the vdev
1451 * corrupt if it is partway through a split operation.
1453 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1455 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1457 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1458 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1459 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1460 VDEV_AUX_CORRUPT_DATA
);
1465 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1467 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1468 VDEV_AUX_CORRUPT_DATA
);
1476 * If this is a verbatim import, no need to check the
1477 * state of the pool.
1479 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1480 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1481 state
!= POOL_STATE_ACTIVE
)
1482 return (SET_ERROR(EBADF
));
1485 * If we were able to open and validate a vdev that was
1486 * previously marked permanently unavailable, clear that state
1489 if (vd
->vdev_not_present
)
1490 vd
->vdev_not_present
= 0;
1497 * Close a virtual device.
1500 vdev_close(vdev_t
*vd
)
1502 vdev_t
*pvd
= vd
->vdev_parent
;
1503 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1505 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1508 * If our parent is reopening, then we are as well, unless we are
1511 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1512 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1514 vd
->vdev_ops
->vdev_op_close(vd
);
1516 vdev_cache_purge(vd
);
1519 * We record the previous state before we close it, so that if we are
1520 * doing a reopen(), we don't generate FMA ereports if we notice that
1521 * it's still faulted.
1523 vd
->vdev_prevstate
= vd
->vdev_state
;
1525 if (vd
->vdev_offline
)
1526 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1528 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1529 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1533 vdev_hold(vdev_t
*vd
)
1535 spa_t
*spa
= vd
->vdev_spa
;
1538 ASSERT(spa_is_root(spa
));
1539 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1542 for (c
= 0; c
< vd
->vdev_children
; c
++)
1543 vdev_hold(vd
->vdev_child
[c
]);
1545 if (vd
->vdev_ops
->vdev_op_leaf
)
1546 vd
->vdev_ops
->vdev_op_hold(vd
);
1550 vdev_rele(vdev_t
*vd
)
1554 ASSERT(spa_is_root(vd
->vdev_spa
));
1555 for (c
= 0; c
< vd
->vdev_children
; c
++)
1556 vdev_rele(vd
->vdev_child
[c
]);
1558 if (vd
->vdev_ops
->vdev_op_leaf
)
1559 vd
->vdev_ops
->vdev_op_rele(vd
);
1563 * Reopen all interior vdevs and any unopened leaves. We don't actually
1564 * reopen leaf vdevs which had previously been opened as they might deadlock
1565 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1566 * If the leaf has never been opened then open it, as usual.
1569 vdev_reopen(vdev_t
*vd
)
1571 spa_t
*spa
= vd
->vdev_spa
;
1573 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1575 /* set the reopening flag unless we're taking the vdev offline */
1576 vd
->vdev_reopening
= !vd
->vdev_offline
;
1578 (void) vdev_open(vd
);
1581 * Call vdev_validate() here to make sure we have the same device.
1582 * Otherwise, a device with an invalid label could be successfully
1583 * opened in response to vdev_reopen().
1586 (void) vdev_validate_aux(vd
);
1587 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1588 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1589 !l2arc_vdev_present(vd
))
1590 l2arc_add_vdev(spa
, vd
);
1592 (void) vdev_validate(vd
, B_TRUE
);
1596 * Reassess parent vdev's health.
1598 vdev_propagate_state(vd
);
1602 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1607 * Normally, partial opens (e.g. of a mirror) are allowed.
1608 * For a create, however, we want to fail the request if
1609 * there are any components we can't open.
1611 error
= vdev_open(vd
);
1613 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1615 return (error
? error
: ENXIO
);
1619 * Recursively load DTLs and initialize all labels.
1621 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1622 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1623 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1632 vdev_metaslab_set_size(vdev_t
*vd
)
1635 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1637 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1638 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1642 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1644 ASSERT(vd
== vd
->vdev_top
);
1645 ASSERT(!vd
->vdev_ishole
);
1646 ASSERT(ISP2(flags
));
1647 ASSERT(spa_writeable(vd
->vdev_spa
));
1649 if (flags
& VDD_METASLAB
)
1650 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1652 if (flags
& VDD_DTL
)
1653 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1655 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1659 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1663 for (c
= 0; c
< vd
->vdev_children
; c
++)
1664 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1666 if (vd
->vdev_ops
->vdev_op_leaf
)
1667 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1673 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1674 * the vdev has less than perfect replication. There are four kinds of DTL:
1676 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1678 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1680 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1681 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1682 * txgs that was scrubbed.
1684 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1685 * persistent errors or just some device being offline.
1686 * Unlike the other three, the DTL_OUTAGE map is not generally
1687 * maintained; it's only computed when needed, typically to
1688 * determine whether a device can be detached.
1690 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1691 * either has the data or it doesn't.
1693 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1694 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1695 * if any child is less than fully replicated, then so is its parent.
1696 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1697 * comprising only those txgs which appear in 'maxfaults' or more children;
1698 * those are the txgs we don't have enough replication to read. For example,
1699 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1700 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1701 * two child DTL_MISSING maps.
1703 * It should be clear from the above that to compute the DTLs and outage maps
1704 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1705 * Therefore, that is all we keep on disk. When loading the pool, or after
1706 * a configuration change, we generate all other DTLs from first principles.
1709 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1711 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1713 ASSERT(t
< DTL_TYPES
);
1714 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1715 ASSERT(spa_writeable(vd
->vdev_spa
));
1717 mutex_enter(rt
->rt_lock
);
1718 if (!range_tree_contains(rt
, txg
, size
))
1719 range_tree_add(rt
, txg
, size
);
1720 mutex_exit(rt
->rt_lock
);
1724 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1726 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1727 boolean_t dirty
= B_FALSE
;
1729 ASSERT(t
< DTL_TYPES
);
1730 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1732 mutex_enter(rt
->rt_lock
);
1733 if (range_tree_space(rt
) != 0)
1734 dirty
= range_tree_contains(rt
, txg
, size
);
1735 mutex_exit(rt
->rt_lock
);
1741 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1743 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1746 mutex_enter(rt
->rt_lock
);
1747 empty
= (range_tree_space(rt
) == 0);
1748 mutex_exit(rt
->rt_lock
);
1754 * Returns the lowest txg in the DTL range.
1757 vdev_dtl_min(vdev_t
*vd
)
1761 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1762 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1763 ASSERT0(vd
->vdev_children
);
1765 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1766 return (rs
->rs_start
- 1);
1770 * Returns the highest txg in the DTL.
1773 vdev_dtl_max(vdev_t
*vd
)
1777 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1778 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1779 ASSERT0(vd
->vdev_children
);
1781 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1782 return (rs
->rs_end
);
1786 * Determine if a resilvering vdev should remove any DTL entries from
1787 * its range. If the vdev was resilvering for the entire duration of the
1788 * scan then it should excise that range from its DTLs. Otherwise, this
1789 * vdev is considered partially resilvered and should leave its DTL
1790 * entries intact. The comment in vdev_dtl_reassess() describes how we
1794 vdev_dtl_should_excise(vdev_t
*vd
)
1796 spa_t
*spa
= vd
->vdev_spa
;
1797 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1799 ASSERT0(scn
->scn_phys
.scn_errors
);
1800 ASSERT0(vd
->vdev_children
);
1802 if (vd
->vdev_resilver_txg
== 0 ||
1803 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1807 * When a resilver is initiated the scan will assign the scn_max_txg
1808 * value to the highest txg value that exists in all DTLs. If this
1809 * device's max DTL is not part of this scan (i.e. it is not in
1810 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1813 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1814 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1815 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1816 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1823 * Reassess DTLs after a config change or scrub completion.
1826 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1828 spa_t
*spa
= vd
->vdev_spa
;
1832 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1834 for (c
= 0; c
< vd
->vdev_children
; c
++)
1835 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1836 scrub_txg
, scrub_done
);
1838 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1841 if (vd
->vdev_ops
->vdev_op_leaf
) {
1842 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1844 mutex_enter(&vd
->vdev_dtl_lock
);
1847 * If we've completed a scan cleanly then determine
1848 * if this vdev should remove any DTLs. We only want to
1849 * excise regions on vdevs that were available during
1850 * the entire duration of this scan.
1852 if (scrub_txg
!= 0 &&
1853 (spa
->spa_scrub_started
||
1854 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1855 vdev_dtl_should_excise(vd
)) {
1857 * We completed a scrub up to scrub_txg. If we
1858 * did it without rebooting, then the scrub dtl
1859 * will be valid, so excise the old region and
1860 * fold in the scrub dtl. Otherwise, leave the
1861 * dtl as-is if there was an error.
1863 * There's little trick here: to excise the beginning
1864 * of the DTL_MISSING map, we put it into a reference
1865 * tree and then add a segment with refcnt -1 that
1866 * covers the range [0, scrub_txg). This means
1867 * that each txg in that range has refcnt -1 or 0.
1868 * We then add DTL_SCRUB with a refcnt of 2, so that
1869 * entries in the range [0, scrub_txg) will have a
1870 * positive refcnt -- either 1 or 2. We then convert
1871 * the reference tree into the new DTL_MISSING map.
1873 space_reftree_create(&reftree
);
1874 space_reftree_add_map(&reftree
,
1875 vd
->vdev_dtl
[DTL_MISSING
], 1);
1876 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1877 space_reftree_add_map(&reftree
,
1878 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1879 space_reftree_generate_map(&reftree
,
1880 vd
->vdev_dtl
[DTL_MISSING
], 1);
1881 space_reftree_destroy(&reftree
);
1883 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1884 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1885 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1887 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1888 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1889 if (!vdev_readable(vd
))
1890 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1892 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1893 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1896 * If the vdev was resilvering and no longer has any
1897 * DTLs then reset its resilvering flag and dirty
1898 * the top level so that we persist the change.
1900 if (vd
->vdev_resilver_txg
!= 0 &&
1901 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1902 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
1903 vd
->vdev_resilver_txg
= 0;
1904 vdev_config_dirty(vd
->vdev_top
);
1907 mutex_exit(&vd
->vdev_dtl_lock
);
1910 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1914 mutex_enter(&vd
->vdev_dtl_lock
);
1915 for (t
= 0; t
< DTL_TYPES
; t
++) {
1918 /* account for child's outage in parent's missing map */
1919 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1921 continue; /* leaf vdevs only */
1922 if (t
== DTL_PARTIAL
)
1923 minref
= 1; /* i.e. non-zero */
1924 else if (vd
->vdev_nparity
!= 0)
1925 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1927 minref
= vd
->vdev_children
; /* any kind of mirror */
1928 space_reftree_create(&reftree
);
1929 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1930 vdev_t
*cvd
= vd
->vdev_child
[c
];
1931 mutex_enter(&cvd
->vdev_dtl_lock
);
1932 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1933 mutex_exit(&cvd
->vdev_dtl_lock
);
1935 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1936 space_reftree_destroy(&reftree
);
1938 mutex_exit(&vd
->vdev_dtl_lock
);
1942 vdev_dtl_load(vdev_t
*vd
)
1944 spa_t
*spa
= vd
->vdev_spa
;
1945 objset_t
*mos
= spa
->spa_meta_objset
;
1949 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
1950 ASSERT(!vd
->vdev_ishole
);
1952 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
1953 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
1956 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
1958 mutex_enter(&vd
->vdev_dtl_lock
);
1961 * Now that we've opened the space_map we need to update
1964 space_map_update(vd
->vdev_dtl_sm
);
1966 error
= space_map_load(vd
->vdev_dtl_sm
,
1967 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
1968 mutex_exit(&vd
->vdev_dtl_lock
);
1973 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1974 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
1983 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1985 spa_t
*spa
= vd
->vdev_spa
;
1986 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
1987 objset_t
*mos
= spa
->spa_meta_objset
;
1988 range_tree_t
*rtsync
;
1991 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
1993 ASSERT(!vd
->vdev_ishole
);
1994 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1996 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1998 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
1999 mutex_enter(&vd
->vdev_dtl_lock
);
2000 space_map_free(vd
->vdev_dtl_sm
, tx
);
2001 space_map_close(vd
->vdev_dtl_sm
);
2002 vd
->vdev_dtl_sm
= NULL
;
2003 mutex_exit(&vd
->vdev_dtl_lock
);
2008 if (vd
->vdev_dtl_sm
== NULL
) {
2009 uint64_t new_object
;
2011 new_object
= space_map_alloc(mos
, tx
);
2012 VERIFY3U(new_object
, !=, 0);
2014 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2015 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2016 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2019 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2021 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2023 mutex_enter(&rtlock
);
2025 mutex_enter(&vd
->vdev_dtl_lock
);
2026 range_tree_walk(rt
, range_tree_add
, rtsync
);
2027 mutex_exit(&vd
->vdev_dtl_lock
);
2029 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2030 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2031 range_tree_vacate(rtsync
, NULL
, NULL
);
2033 range_tree_destroy(rtsync
);
2035 mutex_exit(&rtlock
);
2036 mutex_destroy(&rtlock
);
2039 * If the object for the space map has changed then dirty
2040 * the top level so that we update the config.
2042 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2043 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2044 "new object %llu", txg
, spa_name(spa
), object
,
2045 space_map_object(vd
->vdev_dtl_sm
));
2046 vdev_config_dirty(vd
->vdev_top
);
2051 mutex_enter(&vd
->vdev_dtl_lock
);
2052 space_map_update(vd
->vdev_dtl_sm
);
2053 mutex_exit(&vd
->vdev_dtl_lock
);
2057 * Determine whether the specified vdev can be offlined/detached/removed
2058 * without losing data.
2061 vdev_dtl_required(vdev_t
*vd
)
2063 spa_t
*spa
= vd
->vdev_spa
;
2064 vdev_t
*tvd
= vd
->vdev_top
;
2065 uint8_t cant_read
= vd
->vdev_cant_read
;
2068 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2070 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2074 * Temporarily mark the device as unreadable, and then determine
2075 * whether this results in any DTL outages in the top-level vdev.
2076 * If not, we can safely offline/detach/remove the device.
2078 vd
->vdev_cant_read
= B_TRUE
;
2079 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2080 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2081 vd
->vdev_cant_read
= cant_read
;
2082 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2084 if (!required
&& zio_injection_enabled
)
2085 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2091 * Determine if resilver is needed, and if so the txg range.
2094 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2096 boolean_t needed
= B_FALSE
;
2097 uint64_t thismin
= UINT64_MAX
;
2098 uint64_t thismax
= 0;
2101 if (vd
->vdev_children
== 0) {
2102 mutex_enter(&vd
->vdev_dtl_lock
);
2103 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2104 vdev_writeable(vd
)) {
2106 thismin
= vdev_dtl_min(vd
);
2107 thismax
= vdev_dtl_max(vd
);
2110 mutex_exit(&vd
->vdev_dtl_lock
);
2112 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2113 vdev_t
*cvd
= vd
->vdev_child
[c
];
2114 uint64_t cmin
, cmax
;
2116 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2117 thismin
= MIN(thismin
, cmin
);
2118 thismax
= MAX(thismax
, cmax
);
2124 if (needed
&& minp
) {
2132 vdev_load(vdev_t
*vd
)
2137 * Recursively load all children.
2139 for (c
= 0; c
< vd
->vdev_children
; c
++)
2140 vdev_load(vd
->vdev_child
[c
]);
2143 * If this is a top-level vdev, initialize its metaslabs.
2145 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2146 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2147 vdev_metaslab_init(vd
, 0) != 0))
2148 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2149 VDEV_AUX_CORRUPT_DATA
);
2152 * If this is a leaf vdev, load its DTL.
2154 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2155 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2156 VDEV_AUX_CORRUPT_DATA
);
2160 * The special vdev case is used for hot spares and l2cache devices. Its
2161 * sole purpose it to set the vdev state for the associated vdev. To do this,
2162 * we make sure that we can open the underlying device, then try to read the
2163 * label, and make sure that the label is sane and that it hasn't been
2164 * repurposed to another pool.
2167 vdev_validate_aux(vdev_t
*vd
)
2170 uint64_t guid
, version
;
2173 if (!vdev_readable(vd
))
2176 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2177 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2178 VDEV_AUX_CORRUPT_DATA
);
2182 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2183 !SPA_VERSION_IS_SUPPORTED(version
) ||
2184 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2185 guid
!= vd
->vdev_guid
||
2186 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2187 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2188 VDEV_AUX_CORRUPT_DATA
);
2194 * We don't actually check the pool state here. If it's in fact in
2195 * use by another pool, we update this fact on the fly when requested.
2202 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2204 spa_t
*spa
= vd
->vdev_spa
;
2205 objset_t
*mos
= spa
->spa_meta_objset
;
2209 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2211 if (vd
->vdev_ms
!= NULL
) {
2212 metaslab_group_t
*mg
= vd
->vdev_mg
;
2214 metaslab_group_histogram_verify(mg
);
2215 metaslab_class_histogram_verify(mg
->mg_class
);
2217 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2218 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2220 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2223 mutex_enter(&msp
->ms_lock
);
2225 * If the metaslab was not loaded when the vdev
2226 * was removed then the histogram accounting may
2227 * not be accurate. Update the histogram information
2228 * here so that we ensure that the metaslab group
2229 * and metaslab class are up-to-date.
2231 metaslab_group_histogram_remove(mg
, msp
);
2233 VERIFY0(space_map_allocated(msp
->ms_sm
));
2234 space_map_free(msp
->ms_sm
, tx
);
2235 space_map_close(msp
->ms_sm
);
2237 mutex_exit(&msp
->ms_lock
);
2240 metaslab_group_histogram_verify(mg
);
2241 metaslab_class_histogram_verify(mg
->mg_class
);
2242 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2243 ASSERT0(mg
->mg_histogram
[i
]);
2247 if (vd
->vdev_ms_array
) {
2248 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2249 vd
->vdev_ms_array
= 0;
2255 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2258 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2260 ASSERT(!vd
->vdev_ishole
);
2262 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2263 metaslab_sync_done(msp
, txg
);
2266 metaslab_sync_reassess(vd
->vdev_mg
);
2270 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2272 spa_t
*spa
= vd
->vdev_spa
;
2277 ASSERT(!vd
->vdev_ishole
);
2279 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2280 ASSERT(vd
== vd
->vdev_top
);
2281 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2282 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2283 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2284 ASSERT(vd
->vdev_ms_array
!= 0);
2285 vdev_config_dirty(vd
);
2290 * Remove the metadata associated with this vdev once it's empty.
2292 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2293 vdev_remove(vd
, txg
);
2295 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2296 metaslab_sync(msp
, txg
);
2297 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2300 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2301 vdev_dtl_sync(lvd
, txg
);
2303 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2307 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2309 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2313 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2314 * not be opened, and no I/O is attempted.
2317 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2321 spa_vdev_state_enter(spa
, SCL_NONE
);
2323 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2324 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2326 if (!vd
->vdev_ops
->vdev_op_leaf
)
2327 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2332 * We don't directly use the aux state here, but if we do a
2333 * vdev_reopen(), we need this value to be present to remember why we
2336 vd
->vdev_label_aux
= aux
;
2339 * Faulted state takes precedence over degraded.
2341 vd
->vdev_delayed_close
= B_FALSE
;
2342 vd
->vdev_faulted
= 1ULL;
2343 vd
->vdev_degraded
= 0ULL;
2344 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2347 * If this device has the only valid copy of the data, then
2348 * back off and simply mark the vdev as degraded instead.
2350 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2351 vd
->vdev_degraded
= 1ULL;
2352 vd
->vdev_faulted
= 0ULL;
2355 * If we reopen the device and it's not dead, only then do we
2360 if (vdev_readable(vd
))
2361 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2364 return (spa_vdev_state_exit(spa
, vd
, 0));
2368 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2369 * user that something is wrong. The vdev continues to operate as normal as far
2370 * as I/O is concerned.
2373 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2377 spa_vdev_state_enter(spa
, SCL_NONE
);
2379 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2380 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2382 if (!vd
->vdev_ops
->vdev_op_leaf
)
2383 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2386 * If the vdev is already faulted, then don't do anything.
2388 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2389 return (spa_vdev_state_exit(spa
, NULL
, 0));
2391 vd
->vdev_degraded
= 1ULL;
2392 if (!vdev_is_dead(vd
))
2393 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2396 return (spa_vdev_state_exit(spa
, vd
, 0));
2400 * Online the given vdev.
2402 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2403 * spare device should be detached when the device finishes resilvering.
2404 * Second, the online should be treated like a 'test' online case, so no FMA
2405 * events are generated if the device fails to open.
2408 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2410 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2412 spa_vdev_state_enter(spa
, SCL_NONE
);
2414 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2415 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2417 if (!vd
->vdev_ops
->vdev_op_leaf
)
2418 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2421 vd
->vdev_offline
= B_FALSE
;
2422 vd
->vdev_tmpoffline
= B_FALSE
;
2423 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2424 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2426 /* XXX - L2ARC 1.0 does not support expansion */
2427 if (!vd
->vdev_aux
) {
2428 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2429 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2433 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2435 if (!vd
->vdev_aux
) {
2436 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2437 pvd
->vdev_expanding
= B_FALSE
;
2441 *newstate
= vd
->vdev_state
;
2442 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2443 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2444 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2445 vd
->vdev_parent
->vdev_child
[0] == vd
)
2446 vd
->vdev_unspare
= B_TRUE
;
2448 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2450 /* XXX - L2ARC 1.0 does not support expansion */
2452 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2453 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2455 return (spa_vdev_state_exit(spa
, vd
, 0));
2459 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2463 uint64_t generation
;
2464 metaslab_group_t
*mg
;
2467 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2469 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2470 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2472 if (!vd
->vdev_ops
->vdev_op_leaf
)
2473 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2477 generation
= spa
->spa_config_generation
+ 1;
2480 * If the device isn't already offline, try to offline it.
2482 if (!vd
->vdev_offline
) {
2484 * If this device has the only valid copy of some data,
2485 * don't allow it to be offlined. Log devices are always
2488 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2489 vdev_dtl_required(vd
))
2490 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2493 * If the top-level is a slog and it has had allocations
2494 * then proceed. We check that the vdev's metaslab group
2495 * is not NULL since it's possible that we may have just
2496 * added this vdev but not yet initialized its metaslabs.
2498 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2500 * Prevent any future allocations.
2502 metaslab_group_passivate(mg
);
2503 (void) spa_vdev_state_exit(spa
, vd
, 0);
2505 error
= spa_offline_log(spa
);
2507 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2510 * Check to see if the config has changed.
2512 if (error
|| generation
!= spa
->spa_config_generation
) {
2513 metaslab_group_activate(mg
);
2515 return (spa_vdev_state_exit(spa
,
2517 (void) spa_vdev_state_exit(spa
, vd
, 0);
2520 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2524 * Offline this device and reopen its top-level vdev.
2525 * If the top-level vdev is a log device then just offline
2526 * it. Otherwise, if this action results in the top-level
2527 * vdev becoming unusable, undo it and fail the request.
2529 vd
->vdev_offline
= B_TRUE
;
2532 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2533 vdev_is_dead(tvd
)) {
2534 vd
->vdev_offline
= B_FALSE
;
2536 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2540 * Add the device back into the metaslab rotor so that
2541 * once we online the device it's open for business.
2543 if (tvd
->vdev_islog
&& mg
!= NULL
)
2544 metaslab_group_activate(mg
);
2547 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2549 return (spa_vdev_state_exit(spa
, vd
, 0));
2553 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2557 mutex_enter(&spa
->spa_vdev_top_lock
);
2558 error
= vdev_offline_locked(spa
, guid
, flags
);
2559 mutex_exit(&spa
->spa_vdev_top_lock
);
2565 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2566 * vdev_offline(), we assume the spa config is locked. We also clear all
2567 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2570 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2572 vdev_t
*rvd
= spa
->spa_root_vdev
;
2575 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2580 vd
->vdev_stat
.vs_read_errors
= 0;
2581 vd
->vdev_stat
.vs_write_errors
= 0;
2582 vd
->vdev_stat
.vs_checksum_errors
= 0;
2584 for (c
= 0; c
< vd
->vdev_children
; c
++)
2585 vdev_clear(spa
, vd
->vdev_child
[c
]);
2588 * If we're in the FAULTED state or have experienced failed I/O, then
2589 * clear the persistent state and attempt to reopen the device. We
2590 * also mark the vdev config dirty, so that the new faulted state is
2591 * written out to disk.
2593 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2594 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2597 * When reopening in reponse to a clear event, it may be due to
2598 * a fmadm repair request. In this case, if the device is
2599 * still broken, we want to still post the ereport again.
2601 vd
->vdev_forcefault
= B_TRUE
;
2603 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2604 vd
->vdev_cant_read
= B_FALSE
;
2605 vd
->vdev_cant_write
= B_FALSE
;
2607 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2609 vd
->vdev_forcefault
= B_FALSE
;
2611 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2612 vdev_state_dirty(vd
->vdev_top
);
2614 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2615 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2617 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_CLEAR
);
2621 * When clearing a FMA-diagnosed fault, we always want to
2622 * unspare the device, as we assume that the original spare was
2623 * done in response to the FMA fault.
2625 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2626 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2627 vd
->vdev_parent
->vdev_child
[0] == vd
)
2628 vd
->vdev_unspare
= B_TRUE
;
2632 vdev_is_dead(vdev_t
*vd
)
2635 * Holes and missing devices are always considered "dead".
2636 * This simplifies the code since we don't have to check for
2637 * these types of devices in the various code paths.
2638 * Instead we rely on the fact that we skip over dead devices
2639 * before issuing I/O to them.
2641 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2642 vd
->vdev_ops
== &vdev_missing_ops
);
2646 vdev_readable(vdev_t
*vd
)
2648 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2652 vdev_writeable(vdev_t
*vd
)
2654 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2658 vdev_allocatable(vdev_t
*vd
)
2660 uint64_t state
= vd
->vdev_state
;
2663 * We currently allow allocations from vdevs which may be in the
2664 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2665 * fails to reopen then we'll catch it later when we're holding
2666 * the proper locks. Note that we have to get the vdev state
2667 * in a local variable because although it changes atomically,
2668 * we're asking two separate questions about it.
2670 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2671 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2675 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2677 ASSERT(zio
->io_vd
== vd
);
2679 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2682 if (zio
->io_type
== ZIO_TYPE_READ
)
2683 return (!vd
->vdev_cant_read
);
2685 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2686 return (!vd
->vdev_cant_write
);
2692 * Get statistics for the given vdev.
2695 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2697 spa_t
*spa
= vd
->vdev_spa
;
2698 vdev_t
*rvd
= spa
->spa_root_vdev
;
2701 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2703 mutex_enter(&vd
->vdev_stat_lock
);
2704 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2705 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2706 vs
->vs_state
= vd
->vdev_state
;
2707 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2708 if (vd
->vdev_ops
->vdev_op_leaf
)
2709 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2710 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2711 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&& !vd
->vdev_ishole
) {
2712 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2716 * If we're getting stats on the root vdev, aggregate the I/O counts
2717 * over all top-level vdevs (i.e. the direct children of the root).
2720 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2721 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2722 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2724 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2725 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2726 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2728 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2731 mutex_exit(&vd
->vdev_stat_lock
);
2735 vdev_clear_stats(vdev_t
*vd
)
2737 mutex_enter(&vd
->vdev_stat_lock
);
2738 vd
->vdev_stat
.vs_space
= 0;
2739 vd
->vdev_stat
.vs_dspace
= 0;
2740 vd
->vdev_stat
.vs_alloc
= 0;
2741 mutex_exit(&vd
->vdev_stat_lock
);
2745 vdev_scan_stat_init(vdev_t
*vd
)
2747 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2750 for (c
= 0; c
< vd
->vdev_children
; c
++)
2751 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2753 mutex_enter(&vd
->vdev_stat_lock
);
2754 vs
->vs_scan_processed
= 0;
2755 mutex_exit(&vd
->vdev_stat_lock
);
2759 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2761 spa_t
*spa
= zio
->io_spa
;
2762 vdev_t
*rvd
= spa
->spa_root_vdev
;
2763 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2765 uint64_t txg
= zio
->io_txg
;
2766 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2767 zio_type_t type
= zio
->io_type
;
2768 int flags
= zio
->io_flags
;
2771 * If this i/o is a gang leader, it didn't do any actual work.
2773 if (zio
->io_gang_tree
)
2776 if (zio
->io_error
== 0) {
2778 * If this is a root i/o, don't count it -- we've already
2779 * counted the top-level vdevs, and vdev_get_stats() will
2780 * aggregate them when asked. This reduces contention on
2781 * the root vdev_stat_lock and implicitly handles blocks
2782 * that compress away to holes, for which there is no i/o.
2783 * (Holes never create vdev children, so all the counters
2784 * remain zero, which is what we want.)
2786 * Note: this only applies to successful i/o (io_error == 0)
2787 * because unlike i/o counts, errors are not additive.
2788 * When reading a ditto block, for example, failure of
2789 * one top-level vdev does not imply a root-level error.
2794 ASSERT(vd
== zio
->io_vd
);
2796 if (flags
& ZIO_FLAG_IO_BYPASS
)
2799 mutex_enter(&vd
->vdev_stat_lock
);
2801 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2802 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2803 dsl_scan_phys_t
*scn_phys
=
2804 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2805 uint64_t *processed
= &scn_phys
->scn_processed
;
2808 if (vd
->vdev_ops
->vdev_op_leaf
)
2809 atomic_add_64(processed
, psize
);
2810 vs
->vs_scan_processed
+= psize
;
2813 if (flags
& ZIO_FLAG_SELF_HEAL
)
2814 vs
->vs_self_healed
+= psize
;
2818 vs
->vs_bytes
[type
] += psize
;
2820 mutex_exit(&vd
->vdev_stat_lock
);
2824 if (flags
& ZIO_FLAG_SPECULATIVE
)
2828 * If this is an I/O error that is going to be retried, then ignore the
2829 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2830 * hard errors, when in reality they can happen for any number of
2831 * innocuous reasons (bus resets, MPxIO link failure, etc).
2833 if (zio
->io_error
== EIO
&&
2834 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2838 * Intent logs writes won't propagate their error to the root
2839 * I/O so don't mark these types of failures as pool-level
2842 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2845 mutex_enter(&vd
->vdev_stat_lock
);
2846 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2847 if (zio
->io_error
== ECKSUM
)
2848 vs
->vs_checksum_errors
++;
2850 vs
->vs_read_errors
++;
2852 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2853 vs
->vs_write_errors
++;
2854 mutex_exit(&vd
->vdev_stat_lock
);
2856 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2857 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2858 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2859 spa
->spa_claiming
)) {
2861 * This is either a normal write (not a repair), or it's
2862 * a repair induced by the scrub thread, or it's a repair
2863 * made by zil_claim() during spa_load() in the first txg.
2864 * In the normal case, we commit the DTL change in the same
2865 * txg as the block was born. In the scrub-induced repair
2866 * case, we know that scrubs run in first-pass syncing context,
2867 * so we commit the DTL change in spa_syncing_txg(spa).
2868 * In the zil_claim() case, we commit in spa_first_txg(spa).
2870 * We currently do not make DTL entries for failed spontaneous
2871 * self-healing writes triggered by normal (non-scrubbing)
2872 * reads, because we have no transactional context in which to
2873 * do so -- and it's not clear that it'd be desirable anyway.
2875 if (vd
->vdev_ops
->vdev_op_leaf
) {
2876 uint64_t commit_txg
= txg
;
2877 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2878 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2879 ASSERT(spa_sync_pass(spa
) == 1);
2880 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2881 commit_txg
= spa_syncing_txg(spa
);
2882 } else if (spa
->spa_claiming
) {
2883 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2884 commit_txg
= spa_first_txg(spa
);
2886 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2887 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2889 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2890 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2891 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2894 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2899 * Update the in-core space usage stats for this vdev, its metaslab class,
2900 * and the root vdev.
2903 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2904 int64_t space_delta
)
2906 int64_t dspace_delta
= space_delta
;
2907 spa_t
*spa
= vd
->vdev_spa
;
2908 vdev_t
*rvd
= spa
->spa_root_vdev
;
2909 metaslab_group_t
*mg
= vd
->vdev_mg
;
2910 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2912 ASSERT(vd
== vd
->vdev_top
);
2915 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2916 * factor. We must calculate this here and not at the root vdev
2917 * because the root vdev's psize-to-asize is simply the max of its
2918 * childrens', thus not accurate enough for us.
2920 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2921 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2922 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2923 vd
->vdev_deflate_ratio
;
2925 mutex_enter(&vd
->vdev_stat_lock
);
2926 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2927 vd
->vdev_stat
.vs_space
+= space_delta
;
2928 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2929 mutex_exit(&vd
->vdev_stat_lock
);
2931 if (mc
== spa_normal_class(spa
)) {
2932 mutex_enter(&rvd
->vdev_stat_lock
);
2933 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2934 rvd
->vdev_stat
.vs_space
+= space_delta
;
2935 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2936 mutex_exit(&rvd
->vdev_stat_lock
);
2940 ASSERT(rvd
== vd
->vdev_parent
);
2941 ASSERT(vd
->vdev_ms_count
!= 0);
2943 metaslab_class_space_update(mc
,
2944 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2949 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2950 * so that it will be written out next time the vdev configuration is synced.
2951 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2954 vdev_config_dirty(vdev_t
*vd
)
2956 spa_t
*spa
= vd
->vdev_spa
;
2957 vdev_t
*rvd
= spa
->spa_root_vdev
;
2960 ASSERT(spa_writeable(spa
));
2963 * If this is an aux vdev (as with l2cache and spare devices), then we
2964 * update the vdev config manually and set the sync flag.
2966 if (vd
->vdev_aux
!= NULL
) {
2967 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2971 for (c
= 0; c
< sav
->sav_count
; c
++) {
2972 if (sav
->sav_vdevs
[c
] == vd
)
2976 if (c
== sav
->sav_count
) {
2978 * We're being removed. There's nothing more to do.
2980 ASSERT(sav
->sav_sync
== B_TRUE
);
2984 sav
->sav_sync
= B_TRUE
;
2986 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2987 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2988 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2989 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2995 * Setting the nvlist in the middle if the array is a little
2996 * sketchy, but it will work.
2998 nvlist_free(aux
[c
]);
2999 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3005 * The dirty list is protected by the SCL_CONFIG lock. The caller
3006 * must either hold SCL_CONFIG as writer, or must be the sync thread
3007 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3008 * so this is sufficient to ensure mutual exclusion.
3010 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3011 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3012 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3015 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3016 vdev_config_dirty(rvd
->vdev_child
[c
]);
3018 ASSERT(vd
== vd
->vdev_top
);
3020 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3022 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3027 vdev_config_clean(vdev_t
*vd
)
3029 spa_t
*spa
= vd
->vdev_spa
;
3031 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3032 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3033 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3035 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3036 list_remove(&spa
->spa_config_dirty_list
, vd
);
3040 * Mark a top-level vdev's state as dirty, so that the next pass of
3041 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3042 * the state changes from larger config changes because they require
3043 * much less locking, and are often needed for administrative actions.
3046 vdev_state_dirty(vdev_t
*vd
)
3048 spa_t
*spa
= vd
->vdev_spa
;
3050 ASSERT(spa_writeable(spa
));
3051 ASSERT(vd
== vd
->vdev_top
);
3054 * The state list is protected by the SCL_STATE lock. The caller
3055 * must either hold SCL_STATE as writer, or must be the sync thread
3056 * (which holds SCL_STATE as reader). There's only one sync thread,
3057 * so this is sufficient to ensure mutual exclusion.
3059 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3060 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3061 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3063 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3064 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3068 vdev_state_clean(vdev_t
*vd
)
3070 spa_t
*spa
= vd
->vdev_spa
;
3072 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3073 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3074 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3076 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3077 list_remove(&spa
->spa_state_dirty_list
, vd
);
3081 * Propagate vdev state up from children to parent.
3084 vdev_propagate_state(vdev_t
*vd
)
3086 spa_t
*spa
= vd
->vdev_spa
;
3087 vdev_t
*rvd
= spa
->spa_root_vdev
;
3088 int degraded
= 0, faulted
= 0;
3093 if (vd
->vdev_children
> 0) {
3094 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3095 child
= vd
->vdev_child
[c
];
3098 * Don't factor holes into the decision.
3100 if (child
->vdev_ishole
)
3103 if (!vdev_readable(child
) ||
3104 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3106 * Root special: if there is a top-level log
3107 * device, treat the root vdev as if it were
3110 if (child
->vdev_islog
&& vd
== rvd
)
3114 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3118 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3122 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3125 * Root special: if there is a top-level vdev that cannot be
3126 * opened due to corrupted metadata, then propagate the root
3127 * vdev's aux state as 'corrupt' rather than 'insufficient
3130 if (corrupted
&& vd
== rvd
&&
3131 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3132 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3133 VDEV_AUX_CORRUPT_DATA
);
3136 if (vd
->vdev_parent
)
3137 vdev_propagate_state(vd
->vdev_parent
);
3141 * Set a vdev's state. If this is during an open, we don't update the parent
3142 * state, because we're in the process of opening children depth-first.
3143 * Otherwise, we propagate the change to the parent.
3145 * If this routine places a device in a faulted state, an appropriate ereport is
3149 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3151 uint64_t save_state
;
3152 spa_t
*spa
= vd
->vdev_spa
;
3154 if (state
== vd
->vdev_state
) {
3155 vd
->vdev_stat
.vs_aux
= aux
;
3159 save_state
= vd
->vdev_state
;
3161 vd
->vdev_state
= state
;
3162 vd
->vdev_stat
.vs_aux
= aux
;
3165 * If we are setting the vdev state to anything but an open state, then
3166 * always close the underlying device unless the device has requested
3167 * a delayed close (i.e. we're about to remove or fault the device).
3168 * Otherwise, we keep accessible but invalid devices open forever.
3169 * We don't call vdev_close() itself, because that implies some extra
3170 * checks (offline, etc) that we don't want here. This is limited to
3171 * leaf devices, because otherwise closing the device will affect other
3174 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3175 vd
->vdev_ops
->vdev_op_leaf
)
3176 vd
->vdev_ops
->vdev_op_close(vd
);
3179 * If we have brought this vdev back into service, we need
3180 * to notify fmd so that it can gracefully repair any outstanding
3181 * cases due to a missing device. We do this in all cases, even those
3182 * that probably don't correlate to a repaired fault. This is sure to
3183 * catch all cases, and we let the zfs-retire agent sort it out. If
3184 * this is a transient state it's OK, as the retire agent will
3185 * double-check the state of the vdev before repairing it.
3187 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
3188 vd
->vdev_prevstate
!= state
)
3189 zfs_post_state_change(spa
, vd
);
3191 if (vd
->vdev_removed
&&
3192 state
== VDEV_STATE_CANT_OPEN
&&
3193 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3195 * If the previous state is set to VDEV_STATE_REMOVED, then this
3196 * device was previously marked removed and someone attempted to
3197 * reopen it. If this failed due to a nonexistent device, then
3198 * keep the device in the REMOVED state. We also let this be if
3199 * it is one of our special test online cases, which is only
3200 * attempting to online the device and shouldn't generate an FMA
3203 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3204 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3205 } else if (state
== VDEV_STATE_REMOVED
) {
3206 vd
->vdev_removed
= B_TRUE
;
3207 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3209 * If we fail to open a vdev during an import or recovery, we
3210 * mark it as "not available", which signifies that it was
3211 * never there to begin with. Failure to open such a device
3212 * is not considered an error.
3214 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3215 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3216 vd
->vdev_ops
->vdev_op_leaf
)
3217 vd
->vdev_not_present
= 1;
3220 * Post the appropriate ereport. If the 'prevstate' field is
3221 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3222 * that this is part of a vdev_reopen(). In this case, we don't
3223 * want to post the ereport if the device was already in the
3224 * CANT_OPEN state beforehand.
3226 * If the 'checkremove' flag is set, then this is an attempt to
3227 * online the device in response to an insertion event. If we
3228 * hit this case, then we have detected an insertion event for a
3229 * faulted or offline device that wasn't in the removed state.
3230 * In this scenario, we don't post an ereport because we are
3231 * about to replace the device, or attempt an online with
3232 * vdev_forcefault, which will generate the fault for us.
3234 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3235 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3236 vd
!= spa
->spa_root_vdev
) {
3240 case VDEV_AUX_OPEN_FAILED
:
3241 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3243 case VDEV_AUX_CORRUPT_DATA
:
3244 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3246 case VDEV_AUX_NO_REPLICAS
:
3247 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3249 case VDEV_AUX_BAD_GUID_SUM
:
3250 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3252 case VDEV_AUX_TOO_SMALL
:
3253 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3255 case VDEV_AUX_BAD_LABEL
:
3256 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3259 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3262 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3265 /* Erase any notion of persistent removed state */
3266 vd
->vdev_removed
= B_FALSE
;
3268 vd
->vdev_removed
= B_FALSE
;
3271 if (!isopen
&& vd
->vdev_parent
)
3272 vdev_propagate_state(vd
->vdev_parent
);
3276 * Check the vdev configuration to ensure that it's capable of supporting
3280 vdev_is_bootable(vdev_t
*vd
)
3282 #if defined(__sun__) || defined(__sun)
3284 * Currently, we do not support RAID-Z or partial configuration.
3285 * In addition, only a single top-level vdev is allowed and none of the
3286 * leaves can be wholedisks.
3290 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3291 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3293 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3294 vd
->vdev_children
> 1) {
3296 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3297 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3300 } else if (vd
->vdev_wholedisk
== 1) {
3304 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3305 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3308 #endif /* __sun__ || __sun */
3313 * Load the state from the original vdev tree (ovd) which
3314 * we've retrieved from the MOS config object. If the original
3315 * vdev was offline or faulted then we transfer that state to the
3316 * device in the current vdev tree (nvd).
3319 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3323 ASSERT(nvd
->vdev_top
->vdev_islog
);
3324 ASSERT(spa_config_held(nvd
->vdev_spa
,
3325 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3326 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3328 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3329 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3331 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3333 * Restore the persistent vdev state
3335 nvd
->vdev_offline
= ovd
->vdev_offline
;
3336 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3337 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3338 nvd
->vdev_removed
= ovd
->vdev_removed
;
3343 * Determine if a log device has valid content. If the vdev was
3344 * removed or faulted in the MOS config then we know that
3345 * the content on the log device has already been written to the pool.
3348 vdev_log_state_valid(vdev_t
*vd
)
3352 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3356 for (c
= 0; c
< vd
->vdev_children
; c
++)
3357 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3364 * Expand a vdev if possible.
3367 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3369 ASSERT(vd
->vdev_top
== vd
);
3370 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3372 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3373 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3374 vdev_config_dirty(vd
);
3382 vdev_split(vdev_t
*vd
)
3384 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3386 vdev_remove_child(pvd
, vd
);
3387 vdev_compact_children(pvd
);
3389 cvd
= pvd
->vdev_child
[0];
3390 if (pvd
->vdev_children
== 1) {
3391 vdev_remove_parent(cvd
);
3392 cvd
->vdev_splitting
= B_TRUE
;
3394 vdev_propagate_state(cvd
);
3398 vdev_deadman(vdev_t
*vd
)
3402 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3403 vdev_t
*cvd
= vd
->vdev_child
[c
];
3408 if (vd
->vdev_ops
->vdev_op_leaf
) {
3409 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3411 mutex_enter(&vq
->vq_lock
);
3412 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3413 spa_t
*spa
= vd
->vdev_spa
;
3418 * Look at the head of all the pending queues,
3419 * if any I/O has been outstanding for longer than
3420 * the spa_deadman_synctime we log a zevent.
3422 fio
= avl_first(&vq
->vq_active_tree
);
3423 delta
= gethrtime() - fio
->io_timestamp
;
3424 if (delta
> spa_deadman_synctime(spa
)) {
3425 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3426 "delta %lluns, last io %lluns",
3427 fio
->io_timestamp
, delta
,
3428 vq
->vq_io_complete_ts
);
3429 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3430 spa
, vd
, fio
, 0, 0);
3433 mutex_exit(&vq
->vq_lock
);
3437 #if defined(_KERNEL) && defined(HAVE_SPL)
3438 EXPORT_SYMBOL(vdev_fault
);
3439 EXPORT_SYMBOL(vdev_degrade
);
3440 EXPORT_SYMBOL(vdev_online
);
3441 EXPORT_SYMBOL(vdev_offline
);
3442 EXPORT_SYMBOL(vdev_clear
);
3444 module_param(metaslabs_per_vdev
, int, 0644);
3445 MODULE_PARM_DESC(metaslabs_per_vdev
,
3446 "Divide added vdev into approximately (but no more than) this number "