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
;
209 spa_t
*spa
= cvd
->vdev_spa
;
211 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
212 ASSERT(cvd
->vdev_parent
== NULL
);
214 cvd
->vdev_parent
= pvd
;
219 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
221 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
222 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
223 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
225 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
226 if (pvd
->vdev_child
!= NULL
) {
227 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
228 kmem_free(pvd
->vdev_child
, oldsize
);
231 pvd
->vdev_child
= newchild
;
232 pvd
->vdev_child
[id
] = cvd
;
234 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
235 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
238 * Walk up all ancestors to update guid sum.
240 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
241 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
245 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
248 uint_t id
= cvd
->vdev_id
;
250 ASSERT(cvd
->vdev_parent
== pvd
);
255 ASSERT(id
< pvd
->vdev_children
);
256 ASSERT(pvd
->vdev_child
[id
] == cvd
);
258 pvd
->vdev_child
[id
] = NULL
;
259 cvd
->vdev_parent
= NULL
;
261 for (c
= 0; c
< pvd
->vdev_children
; c
++)
262 if (pvd
->vdev_child
[c
])
265 if (c
== pvd
->vdev_children
) {
266 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
267 pvd
->vdev_child
= NULL
;
268 pvd
->vdev_children
= 0;
272 * Walk up all ancestors to update guid sum.
274 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
275 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
279 * Remove any holes in the child array.
282 vdev_compact_children(vdev_t
*pvd
)
284 vdev_t
**newchild
, *cvd
;
285 int oldc
= pvd
->vdev_children
;
289 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
291 for (c
= newc
= 0; c
< oldc
; c
++)
292 if (pvd
->vdev_child
[c
])
295 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
297 for (c
= newc
= 0; c
< oldc
; c
++) {
298 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
299 newchild
[newc
] = cvd
;
300 cvd
->vdev_id
= newc
++;
304 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
305 pvd
->vdev_child
= newchild
;
306 pvd
->vdev_children
= newc
;
310 * Allocate and minimally initialize a vdev_t.
313 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
318 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
320 if (spa
->spa_root_vdev
== NULL
) {
321 ASSERT(ops
== &vdev_root_ops
);
322 spa
->spa_root_vdev
= vd
;
323 spa
->spa_load_guid
= spa_generate_guid(NULL
);
326 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
327 if (spa
->spa_root_vdev
== vd
) {
329 * The root vdev's guid will also be the pool guid,
330 * which must be unique among all pools.
332 guid
= spa_generate_guid(NULL
);
335 * Any other vdev's guid must be unique within the pool.
337 guid
= spa_generate_guid(spa
);
339 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
344 vd
->vdev_guid
= guid
;
345 vd
->vdev_guid_sum
= guid
;
347 vd
->vdev_state
= VDEV_STATE_CLOSED
;
348 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
350 list_link_init(&vd
->vdev_config_dirty_node
);
351 list_link_init(&vd
->vdev_state_dirty_node
);
352 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
353 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
354 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
355 for (t
= 0; t
< DTL_TYPES
; t
++) {
356 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
359 txg_list_create(&vd
->vdev_ms_list
,
360 offsetof(struct metaslab
, ms_txg_node
));
361 txg_list_create(&vd
->vdev_dtl_list
,
362 offsetof(struct vdev
, vdev_dtl_node
));
363 vd
->vdev_stat
.vs_timestamp
= gethrtime();
371 * Allocate a new vdev. The 'alloctype' is used to control whether we are
372 * creating a new vdev or loading an existing one - the behavior is slightly
373 * different for each case.
376 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
381 uint64_t guid
= 0, islog
, nparity
;
384 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
386 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
387 return (SET_ERROR(EINVAL
));
389 if ((ops
= vdev_getops(type
)) == NULL
)
390 return (SET_ERROR(EINVAL
));
393 * If this is a load, get the vdev guid from the nvlist.
394 * Otherwise, vdev_alloc_common() will generate one for us.
396 if (alloctype
== VDEV_ALLOC_LOAD
) {
399 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
401 return (SET_ERROR(EINVAL
));
403 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
404 return (SET_ERROR(EINVAL
));
405 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
406 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
407 return (SET_ERROR(EINVAL
));
408 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
409 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
410 return (SET_ERROR(EINVAL
));
411 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
412 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
413 return (SET_ERROR(EINVAL
));
417 * The first allocated vdev must be of type 'root'.
419 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
420 return (SET_ERROR(EINVAL
));
423 * Determine whether we're a log vdev.
426 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
427 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
428 return (SET_ERROR(ENOTSUP
));
430 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
431 return (SET_ERROR(ENOTSUP
));
434 * Set the nparity property for RAID-Z vdevs.
437 if (ops
== &vdev_raidz_ops
) {
438 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
440 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
441 return (SET_ERROR(EINVAL
));
443 * Previous versions could only support 1 or 2 parity
447 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
448 return (SET_ERROR(ENOTSUP
));
450 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
451 return (SET_ERROR(ENOTSUP
));
454 * We require the parity to be specified for SPAs that
455 * support multiple parity levels.
457 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
458 return (SET_ERROR(EINVAL
));
460 * Otherwise, we default to 1 parity device for RAID-Z.
467 ASSERT(nparity
!= -1ULL);
469 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
471 vd
->vdev_islog
= islog
;
472 vd
->vdev_nparity
= nparity
;
474 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
475 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
476 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
477 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
478 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
479 &vd
->vdev_physpath
) == 0)
480 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
481 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
482 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
485 * Set the whole_disk property. If it's not specified, leave the value
488 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
489 &vd
->vdev_wholedisk
) != 0)
490 vd
->vdev_wholedisk
= -1ULL;
493 * Look for the 'not present' flag. This will only be set if the device
494 * was not present at the time of import.
496 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
497 &vd
->vdev_not_present
);
500 * Get the alignment requirement.
502 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
505 * Retrieve the vdev creation time.
507 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
511 * If we're a top-level vdev, try to load the allocation parameters.
513 if (parent
&& !parent
->vdev_parent
&&
514 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
515 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
517 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
519 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
521 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
525 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
526 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
527 alloctype
== VDEV_ALLOC_ADD
||
528 alloctype
== VDEV_ALLOC_SPLIT
||
529 alloctype
== VDEV_ALLOC_ROOTPOOL
);
530 vd
->vdev_mg
= metaslab_group_create(islog
?
531 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
535 * If we're a leaf vdev, try to load the DTL object and other state.
537 if (vd
->vdev_ops
->vdev_op_leaf
&&
538 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
539 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
540 if (alloctype
== VDEV_ALLOC_LOAD
) {
541 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
542 &vd
->vdev_dtl_object
);
543 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
547 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
550 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
551 &spare
) == 0 && spare
)
555 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
558 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
559 &vd
->vdev_resilver_txg
);
562 * When importing a pool, we want to ignore the persistent fault
563 * state, as the diagnosis made on another system may not be
564 * valid in the current context. Local vdevs will
565 * remain in the faulted state.
567 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
568 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
570 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
572 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
575 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
579 VDEV_AUX_ERR_EXCEEDED
;
580 if (nvlist_lookup_string(nv
,
581 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
582 strcmp(aux
, "external") == 0)
583 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
589 * Add ourselves to the parent's list of children.
591 vdev_add_child(parent
, vd
);
599 vdev_free(vdev_t
*vd
)
602 spa_t
*spa
= vd
->vdev_spa
;
605 * vdev_free() implies closing the vdev first. This is simpler than
606 * trying to ensure complicated semantics for all callers.
610 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
611 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
616 for (c
= 0; c
< vd
->vdev_children
; c
++)
617 vdev_free(vd
->vdev_child
[c
]);
619 ASSERT(vd
->vdev_child
== NULL
);
620 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
623 * Discard allocation state.
625 if (vd
->vdev_mg
!= NULL
) {
626 vdev_metaslab_fini(vd
);
627 metaslab_group_destroy(vd
->vdev_mg
);
630 ASSERT0(vd
->vdev_stat
.vs_space
);
631 ASSERT0(vd
->vdev_stat
.vs_dspace
);
632 ASSERT0(vd
->vdev_stat
.vs_alloc
);
635 * Remove this vdev from its parent's child list.
637 vdev_remove_child(vd
->vdev_parent
, vd
);
639 ASSERT(vd
->vdev_parent
== NULL
);
642 * Clean up vdev structure.
648 spa_strfree(vd
->vdev_path
);
650 spa_strfree(vd
->vdev_devid
);
651 if (vd
->vdev_physpath
)
652 spa_strfree(vd
->vdev_physpath
);
654 spa_strfree(vd
->vdev_fru
);
656 if (vd
->vdev_isspare
)
657 spa_spare_remove(vd
);
658 if (vd
->vdev_isl2cache
)
659 spa_l2cache_remove(vd
);
661 txg_list_destroy(&vd
->vdev_ms_list
);
662 txg_list_destroy(&vd
->vdev_dtl_list
);
664 mutex_enter(&vd
->vdev_dtl_lock
);
665 space_map_close(vd
->vdev_dtl_sm
);
666 for (t
= 0; t
< DTL_TYPES
; t
++) {
667 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
668 range_tree_destroy(vd
->vdev_dtl
[t
]);
670 mutex_exit(&vd
->vdev_dtl_lock
);
672 mutex_destroy(&vd
->vdev_dtl_lock
);
673 mutex_destroy(&vd
->vdev_stat_lock
);
674 mutex_destroy(&vd
->vdev_probe_lock
);
676 if (vd
== spa
->spa_root_vdev
)
677 spa
->spa_root_vdev
= NULL
;
679 kmem_free(vd
, sizeof (vdev_t
));
683 * Transfer top-level vdev state from svd to tvd.
686 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
688 spa_t
*spa
= svd
->vdev_spa
;
693 ASSERT(tvd
== tvd
->vdev_top
);
695 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
696 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
697 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
699 svd
->vdev_ms_array
= 0;
700 svd
->vdev_ms_shift
= 0;
701 svd
->vdev_ms_count
= 0;
704 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
705 tvd
->vdev_mg
= svd
->vdev_mg
;
706 tvd
->vdev_ms
= svd
->vdev_ms
;
711 if (tvd
->vdev_mg
!= NULL
)
712 tvd
->vdev_mg
->mg_vd
= tvd
;
714 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
715 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
716 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
718 svd
->vdev_stat
.vs_alloc
= 0;
719 svd
->vdev_stat
.vs_space
= 0;
720 svd
->vdev_stat
.vs_dspace
= 0;
722 for (t
= 0; t
< TXG_SIZE
; t
++) {
723 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
724 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
725 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
726 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
727 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
728 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
731 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
732 vdev_config_clean(svd
);
733 vdev_config_dirty(tvd
);
736 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
737 vdev_state_clean(svd
);
738 vdev_state_dirty(tvd
);
741 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
742 svd
->vdev_deflate_ratio
= 0;
744 tvd
->vdev_islog
= svd
->vdev_islog
;
749 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
758 for (c
= 0; c
< vd
->vdev_children
; c
++)
759 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
763 * Add a mirror/replacing vdev above an existing vdev.
766 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
768 spa_t
*spa
= cvd
->vdev_spa
;
769 vdev_t
*pvd
= cvd
->vdev_parent
;
772 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
774 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
776 mvd
->vdev_asize
= cvd
->vdev_asize
;
777 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
778 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
779 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
780 mvd
->vdev_state
= cvd
->vdev_state
;
781 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
783 vdev_remove_child(pvd
, cvd
);
784 vdev_add_child(pvd
, mvd
);
785 cvd
->vdev_id
= mvd
->vdev_children
;
786 vdev_add_child(mvd
, cvd
);
787 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
789 if (mvd
== mvd
->vdev_top
)
790 vdev_top_transfer(cvd
, mvd
);
796 * Remove a 1-way mirror/replacing vdev from the tree.
799 vdev_remove_parent(vdev_t
*cvd
)
801 vdev_t
*mvd
= cvd
->vdev_parent
;
802 vdev_t
*pvd
= mvd
->vdev_parent
;
804 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
806 ASSERT(mvd
->vdev_children
== 1);
807 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
808 mvd
->vdev_ops
== &vdev_replacing_ops
||
809 mvd
->vdev_ops
== &vdev_spare_ops
);
810 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
812 vdev_remove_child(mvd
, cvd
);
813 vdev_remove_child(pvd
, mvd
);
816 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
817 * Otherwise, we could have detached an offline device, and when we
818 * go to import the pool we'll think we have two top-level vdevs,
819 * instead of a different version of the same top-level vdev.
821 if (mvd
->vdev_top
== mvd
) {
822 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
823 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
824 cvd
->vdev_guid
+= guid_delta
;
825 cvd
->vdev_guid_sum
+= guid_delta
;
828 * If pool not set for autoexpand, we need to also preserve
829 * mvd's asize to prevent automatic expansion of cvd.
830 * Otherwise if we are adjusting the mirror by attaching and
831 * detaching children of non-uniform sizes, the mirror could
832 * autoexpand, unexpectedly requiring larger devices to
833 * re-establish the mirror.
835 if (!cvd
->vdev_spa
->spa_autoexpand
)
836 cvd
->vdev_asize
= mvd
->vdev_asize
;
838 cvd
->vdev_id
= mvd
->vdev_id
;
839 vdev_add_child(pvd
, cvd
);
840 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
842 if (cvd
== cvd
->vdev_top
)
843 vdev_top_transfer(mvd
, cvd
);
845 ASSERT(mvd
->vdev_children
== 0);
850 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
852 spa_t
*spa
= vd
->vdev_spa
;
853 objset_t
*mos
= spa
->spa_meta_objset
;
855 uint64_t oldc
= vd
->vdev_ms_count
;
856 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
860 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
863 * This vdev is not being allocated from yet or is a hole.
865 if (vd
->vdev_ms_shift
== 0)
868 ASSERT(!vd
->vdev_ishole
);
871 * Compute the raidz-deflation ratio. Note, we hard-code
872 * in 128k (1 << 17) because it is the "typical" blocksize.
873 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
874 * otherwise it would inconsistently account for existing bp's.
876 vd
->vdev_deflate_ratio
= (1 << 17) /
877 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
879 ASSERT(oldc
<= newc
);
881 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
884 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
885 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
889 vd
->vdev_ms_count
= newc
;
891 for (m
= oldc
; m
< newc
; m
++) {
895 error
= dmu_read(mos
, vd
->vdev_ms_array
,
896 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
902 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
909 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
912 * If the vdev is being removed we don't activate
913 * the metaslabs since we want to ensure that no new
914 * allocations are performed on this device.
916 if (oldc
== 0 && !vd
->vdev_removing
)
917 metaslab_group_activate(vd
->vdev_mg
);
920 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
926 vdev_metaslab_fini(vdev_t
*vd
)
929 uint64_t count
= vd
->vdev_ms_count
;
931 if (vd
->vdev_ms
!= NULL
) {
932 metaslab_group_passivate(vd
->vdev_mg
);
933 for (m
= 0; m
< count
; m
++) {
934 metaslab_t
*msp
= vd
->vdev_ms
[m
];
939 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
943 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
946 typedef struct vdev_probe_stats
{
947 boolean_t vps_readable
;
948 boolean_t vps_writeable
;
950 } vdev_probe_stats_t
;
953 vdev_probe_done(zio_t
*zio
)
955 spa_t
*spa
= zio
->io_spa
;
956 vdev_t
*vd
= zio
->io_vd
;
957 vdev_probe_stats_t
*vps
= zio
->io_private
;
959 ASSERT(vd
->vdev_probe_zio
!= NULL
);
961 if (zio
->io_type
== ZIO_TYPE_READ
) {
962 if (zio
->io_error
== 0)
963 vps
->vps_readable
= 1;
964 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
965 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
966 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
967 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
968 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
970 zio_buf_free(zio
->io_data
, zio
->io_size
);
972 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
973 if (zio
->io_error
== 0)
974 vps
->vps_writeable
= 1;
975 zio_buf_free(zio
->io_data
, zio
->io_size
);
976 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
979 vd
->vdev_cant_read
|= !vps
->vps_readable
;
980 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
982 if (vdev_readable(vd
) &&
983 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
986 ASSERT(zio
->io_error
!= 0);
987 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
988 spa
, vd
, NULL
, 0, 0);
989 zio
->io_error
= SET_ERROR(ENXIO
);
992 mutex_enter(&vd
->vdev_probe_lock
);
993 ASSERT(vd
->vdev_probe_zio
== zio
);
994 vd
->vdev_probe_zio
= NULL
;
995 mutex_exit(&vd
->vdev_probe_lock
);
997 while ((pio
= zio_walk_parents(zio
)) != NULL
)
998 if (!vdev_accessible(vd
, pio
))
999 pio
->io_error
= SET_ERROR(ENXIO
);
1001 kmem_free(vps
, sizeof (*vps
));
1006 * Determine whether this device is accessible.
1008 * Read and write to several known locations: the pad regions of each
1009 * vdev label but the first, which we leave alone in case it contains
1013 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1015 spa_t
*spa
= vd
->vdev_spa
;
1016 vdev_probe_stats_t
*vps
= NULL
;
1020 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1023 * Don't probe the probe.
1025 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1029 * To prevent 'probe storms' when a device fails, we create
1030 * just one probe i/o at a time. All zios that want to probe
1031 * this vdev will become parents of the probe io.
1033 mutex_enter(&vd
->vdev_probe_lock
);
1035 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1036 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1038 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1039 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1042 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1044 * vdev_cant_read and vdev_cant_write can only
1045 * transition from TRUE to FALSE when we have the
1046 * SCL_ZIO lock as writer; otherwise they can only
1047 * transition from FALSE to TRUE. This ensures that
1048 * any zio looking at these values can assume that
1049 * failures persist for the life of the I/O. That's
1050 * important because when a device has intermittent
1051 * connectivity problems, we want to ensure that
1052 * they're ascribed to the device (ENXIO) and not
1055 * Since we hold SCL_ZIO as writer here, clear both
1056 * values so the probe can reevaluate from first
1059 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1060 vd
->vdev_cant_read
= B_FALSE
;
1061 vd
->vdev_cant_write
= B_FALSE
;
1064 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1065 vdev_probe_done
, vps
,
1066 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1069 * We can't change the vdev state in this context, so we
1070 * kick off an async task to do it on our behalf.
1073 vd
->vdev_probe_wanted
= B_TRUE
;
1074 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1079 zio_add_child(zio
, pio
);
1081 mutex_exit(&vd
->vdev_probe_lock
);
1084 ASSERT(zio
!= NULL
);
1088 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1089 zio_nowait(zio_read_phys(pio
, vd
,
1090 vdev_label_offset(vd
->vdev_psize
, l
,
1091 offsetof(vdev_label_t
, vl_pad2
)),
1092 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1093 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1094 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1105 vdev_open_child(void *arg
)
1109 vd
->vdev_open_thread
= curthread
;
1110 vd
->vdev_open_error
= vdev_open(vd
);
1111 vd
->vdev_open_thread
= NULL
;
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
;
1139 * in order to handle pools on top of zvols, do the opens
1140 * in a single thread so that the same thread holds the
1141 * spa_namespace_lock
1143 if (vdev_uses_zvols(vd
)) {
1144 for (c
= 0; c
< children
; c
++)
1145 vd
->vdev_child
[c
]->vdev_open_error
=
1146 vdev_open(vd
->vdev_child
[c
]);
1149 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1150 children
, children
, TASKQ_PREPOPULATE
);
1152 for (c
= 0; c
< children
; c
++)
1153 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1160 * Prepare a virtual device for access.
1163 vdev_open(vdev_t
*vd
)
1165 spa_t
*spa
= vd
->vdev_spa
;
1168 uint64_t max_osize
= 0;
1169 uint64_t asize
, max_asize
, psize
;
1170 uint64_t ashift
= 0;
1173 ASSERT(vd
->vdev_open_thread
== curthread
||
1174 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1175 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1176 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1177 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1179 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1180 vd
->vdev_cant_read
= B_FALSE
;
1181 vd
->vdev_cant_write
= B_FALSE
;
1182 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1185 * If this vdev is not removed, check its fault status. If it's
1186 * faulted, bail out of the open.
1188 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1189 ASSERT(vd
->vdev_children
== 0);
1190 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1191 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1192 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1193 vd
->vdev_label_aux
);
1194 return (SET_ERROR(ENXIO
));
1195 } else if (vd
->vdev_offline
) {
1196 ASSERT(vd
->vdev_children
== 0);
1197 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1198 return (SET_ERROR(ENXIO
));
1201 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1204 * Reset the vdev_reopening flag so that we actually close
1205 * the vdev on error.
1207 vd
->vdev_reopening
= B_FALSE
;
1208 if (zio_injection_enabled
&& error
== 0)
1209 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1212 if (vd
->vdev_removed
&&
1213 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1214 vd
->vdev_removed
= B_FALSE
;
1216 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1217 vd
->vdev_stat
.vs_aux
);
1221 vd
->vdev_removed
= B_FALSE
;
1224 * Recheck the faulted flag now that we have confirmed that
1225 * the vdev is accessible. If we're faulted, bail.
1227 if (vd
->vdev_faulted
) {
1228 ASSERT(vd
->vdev_children
== 0);
1229 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1230 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1231 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1232 vd
->vdev_label_aux
);
1233 return (SET_ERROR(ENXIO
));
1236 if (vd
->vdev_degraded
) {
1237 ASSERT(vd
->vdev_children
== 0);
1238 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1239 VDEV_AUX_ERR_EXCEEDED
);
1241 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1245 * For hole or missing vdevs we just return success.
1247 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1250 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1251 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1252 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1258 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1259 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1261 if (vd
->vdev_children
== 0) {
1262 if (osize
< SPA_MINDEVSIZE
) {
1263 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1264 VDEV_AUX_TOO_SMALL
);
1265 return (SET_ERROR(EOVERFLOW
));
1268 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1269 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1270 VDEV_LABEL_END_SIZE
);
1272 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1273 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1274 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1275 VDEV_AUX_TOO_SMALL
);
1276 return (SET_ERROR(EOVERFLOW
));
1280 max_asize
= max_osize
;
1283 vd
->vdev_psize
= psize
;
1286 * Make sure the allocatable size hasn't shrunk.
1288 if (asize
< vd
->vdev_min_asize
) {
1289 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1290 VDEV_AUX_BAD_LABEL
);
1291 return (SET_ERROR(EINVAL
));
1294 if (vd
->vdev_asize
== 0) {
1296 * This is the first-ever open, so use the computed values.
1297 * For compatibility, a different ashift can be requested.
1299 vd
->vdev_asize
= asize
;
1300 vd
->vdev_max_asize
= max_asize
;
1301 if (vd
->vdev_ashift
== 0)
1302 vd
->vdev_ashift
= ashift
;
1305 * Detect if the alignment requirement has increased.
1306 * We don't want to make the pool unavailable, just
1307 * post an event instead.
1309 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1310 vd
->vdev_ops
->vdev_op_leaf
) {
1311 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1312 spa
, vd
, NULL
, 0, 0);
1315 vd
->vdev_max_asize
= max_asize
;
1319 * If all children are healthy and the asize has increased,
1320 * then we've experienced dynamic LUN growth. If automatic
1321 * expansion is enabled then use the additional space.
1323 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1324 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1325 vd
->vdev_asize
= asize
;
1327 vdev_set_min_asize(vd
);
1330 * Ensure we can issue some IO before declaring the
1331 * vdev open for business.
1333 if (vd
->vdev_ops
->vdev_op_leaf
&&
1334 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1335 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1336 VDEV_AUX_ERR_EXCEEDED
);
1341 * Track the min and max ashift values for normal data devices.
1343 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1344 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1345 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1346 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1347 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1348 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1352 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1353 * resilver. But don't do this if we are doing a reopen for a scrub,
1354 * since this would just restart the scrub we are already doing.
1356 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1357 vdev_resilver_needed(vd
, NULL
, NULL
))
1358 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1364 * Called once the vdevs are all opened, this routine validates the label
1365 * contents. This needs to be done before vdev_load() so that we don't
1366 * inadvertently do repair I/Os to the wrong device.
1368 * If 'strict' is false ignore the spa guid check. This is necessary because
1369 * if the machine crashed during a re-guid the new guid might have been written
1370 * to all of the vdev labels, but not the cached config. The strict check
1371 * will be performed when the pool is opened again using the mos config.
1373 * This function will only return failure if one of the vdevs indicates that it
1374 * has since been destroyed or exported. This is only possible if
1375 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1376 * will be updated but the function will return 0.
1379 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1381 spa_t
*spa
= vd
->vdev_spa
;
1383 uint64_t guid
= 0, top_guid
;
1387 for (c
= 0; c
< vd
->vdev_children
; c
++)
1388 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1389 return (SET_ERROR(EBADF
));
1392 * If the device has already failed, or was marked offline, don't do
1393 * any further validation. Otherwise, label I/O will fail and we will
1394 * overwrite the previous state.
1396 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1397 uint64_t aux_guid
= 0;
1399 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1400 spa_last_synced_txg(spa
) : -1ULL;
1402 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1403 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1404 VDEV_AUX_BAD_LABEL
);
1409 * Determine if this vdev has been split off into another
1410 * pool. If so, then refuse to open it.
1412 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1413 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1414 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1415 VDEV_AUX_SPLIT_POOL
);
1420 if (strict
&& (nvlist_lookup_uint64(label
,
1421 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1422 guid
!= spa_guid(spa
))) {
1423 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1424 VDEV_AUX_CORRUPT_DATA
);
1429 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1430 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1435 * If this vdev just became a top-level vdev because its
1436 * sibling was detached, it will have adopted the parent's
1437 * vdev guid -- but the label may or may not be on disk yet.
1438 * Fortunately, either version of the label will have the
1439 * same top guid, so if we're a top-level vdev, we can
1440 * safely compare to that instead.
1442 * If we split this vdev off instead, then we also check the
1443 * original pool's guid. We don't want to consider the vdev
1444 * corrupt if it is partway through a split operation.
1446 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1448 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1450 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1451 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1452 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1453 VDEV_AUX_CORRUPT_DATA
);
1458 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1460 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1461 VDEV_AUX_CORRUPT_DATA
);
1469 * If this is a verbatim import, no need to check the
1470 * state of the pool.
1472 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1473 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1474 state
!= POOL_STATE_ACTIVE
)
1475 return (SET_ERROR(EBADF
));
1478 * If we were able to open and validate a vdev that was
1479 * previously marked permanently unavailable, clear that state
1482 if (vd
->vdev_not_present
)
1483 vd
->vdev_not_present
= 0;
1490 * Close a virtual device.
1493 vdev_close(vdev_t
*vd
)
1495 vdev_t
*pvd
= vd
->vdev_parent
;
1496 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1498 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1501 * If our parent is reopening, then we are as well, unless we are
1504 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1505 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1507 vd
->vdev_ops
->vdev_op_close(vd
);
1509 vdev_cache_purge(vd
);
1512 * We record the previous state before we close it, so that if we are
1513 * doing a reopen(), we don't generate FMA ereports if we notice that
1514 * it's still faulted.
1516 vd
->vdev_prevstate
= vd
->vdev_state
;
1518 if (vd
->vdev_offline
)
1519 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1521 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1522 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1526 vdev_hold(vdev_t
*vd
)
1528 spa_t
*spa
= vd
->vdev_spa
;
1531 ASSERT(spa_is_root(spa
));
1532 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1535 for (c
= 0; c
< vd
->vdev_children
; c
++)
1536 vdev_hold(vd
->vdev_child
[c
]);
1538 if (vd
->vdev_ops
->vdev_op_leaf
)
1539 vd
->vdev_ops
->vdev_op_hold(vd
);
1543 vdev_rele(vdev_t
*vd
)
1547 ASSERT(spa_is_root(vd
->vdev_spa
));
1548 for (c
= 0; c
< vd
->vdev_children
; c
++)
1549 vdev_rele(vd
->vdev_child
[c
]);
1551 if (vd
->vdev_ops
->vdev_op_leaf
)
1552 vd
->vdev_ops
->vdev_op_rele(vd
);
1556 * Reopen all interior vdevs and any unopened leaves. We don't actually
1557 * reopen leaf vdevs which had previously been opened as they might deadlock
1558 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1559 * If the leaf has never been opened then open it, as usual.
1562 vdev_reopen(vdev_t
*vd
)
1564 spa_t
*spa
= vd
->vdev_spa
;
1566 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1568 /* set the reopening flag unless we're taking the vdev offline */
1569 vd
->vdev_reopening
= !vd
->vdev_offline
;
1571 (void) vdev_open(vd
);
1574 * Call vdev_validate() here to make sure we have the same device.
1575 * Otherwise, a device with an invalid label could be successfully
1576 * opened in response to vdev_reopen().
1579 (void) vdev_validate_aux(vd
);
1580 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1581 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1582 !l2arc_vdev_present(vd
))
1583 l2arc_add_vdev(spa
, vd
);
1585 (void) vdev_validate(vd
, B_TRUE
);
1589 * Reassess parent vdev's health.
1591 vdev_propagate_state(vd
);
1595 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1600 * Normally, partial opens (e.g. of a mirror) are allowed.
1601 * For a create, however, we want to fail the request if
1602 * there are any components we can't open.
1604 error
= vdev_open(vd
);
1606 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1608 return (error
? error
: ENXIO
);
1612 * Recursively load DTLs and initialize all labels.
1614 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1615 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1616 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1625 vdev_metaslab_set_size(vdev_t
*vd
)
1628 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1630 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1631 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1635 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1637 ASSERT(vd
== vd
->vdev_top
);
1638 ASSERT(!vd
->vdev_ishole
);
1639 ASSERT(ISP2(flags
));
1640 ASSERT(spa_writeable(vd
->vdev_spa
));
1642 if (flags
& VDD_METASLAB
)
1643 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1645 if (flags
& VDD_DTL
)
1646 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1648 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1652 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1656 for (c
= 0; c
< vd
->vdev_children
; c
++)
1657 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1659 if (vd
->vdev_ops
->vdev_op_leaf
)
1660 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1666 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1667 * the vdev has less than perfect replication. There are four kinds of DTL:
1669 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1671 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1673 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1674 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1675 * txgs that was scrubbed.
1677 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1678 * persistent errors or just some device being offline.
1679 * Unlike the other three, the DTL_OUTAGE map is not generally
1680 * maintained; it's only computed when needed, typically to
1681 * determine whether a device can be detached.
1683 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1684 * either has the data or it doesn't.
1686 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1687 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1688 * if any child is less than fully replicated, then so is its parent.
1689 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1690 * comprising only those txgs which appear in 'maxfaults' or more children;
1691 * those are the txgs we don't have enough replication to read. For example,
1692 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1693 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1694 * two child DTL_MISSING maps.
1696 * It should be clear from the above that to compute the DTLs and outage maps
1697 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1698 * Therefore, that is all we keep on disk. When loading the pool, or after
1699 * a configuration change, we generate all other DTLs from first principles.
1702 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1704 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1706 ASSERT(t
< DTL_TYPES
);
1707 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1708 ASSERT(spa_writeable(vd
->vdev_spa
));
1710 mutex_enter(rt
->rt_lock
);
1711 if (!range_tree_contains(rt
, txg
, size
))
1712 range_tree_add(rt
, txg
, size
);
1713 mutex_exit(rt
->rt_lock
);
1717 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1719 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1720 boolean_t dirty
= B_FALSE
;
1722 ASSERT(t
< DTL_TYPES
);
1723 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1725 mutex_enter(rt
->rt_lock
);
1726 if (range_tree_space(rt
) != 0)
1727 dirty
= range_tree_contains(rt
, txg
, size
);
1728 mutex_exit(rt
->rt_lock
);
1734 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1736 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1739 mutex_enter(rt
->rt_lock
);
1740 empty
= (range_tree_space(rt
) == 0);
1741 mutex_exit(rt
->rt_lock
);
1747 * Returns the lowest txg in the DTL range.
1750 vdev_dtl_min(vdev_t
*vd
)
1754 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1755 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1756 ASSERT0(vd
->vdev_children
);
1758 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1759 return (rs
->rs_start
- 1);
1763 * Returns the highest txg in the DTL.
1766 vdev_dtl_max(vdev_t
*vd
)
1770 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1771 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1772 ASSERT0(vd
->vdev_children
);
1774 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1775 return (rs
->rs_end
);
1779 * Determine if a resilvering vdev should remove any DTL entries from
1780 * its range. If the vdev was resilvering for the entire duration of the
1781 * scan then it should excise that range from its DTLs. Otherwise, this
1782 * vdev is considered partially resilvered and should leave its DTL
1783 * entries intact. The comment in vdev_dtl_reassess() describes how we
1787 vdev_dtl_should_excise(vdev_t
*vd
)
1789 spa_t
*spa
= vd
->vdev_spa
;
1790 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1792 ASSERT0(scn
->scn_phys
.scn_errors
);
1793 ASSERT0(vd
->vdev_children
);
1795 if (vd
->vdev_resilver_txg
== 0 ||
1796 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1800 * When a resilver is initiated the scan will assign the scn_max_txg
1801 * value to the highest txg value that exists in all DTLs. If this
1802 * device's max DTL is not part of this scan (i.e. it is not in
1803 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1806 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1807 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1808 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1809 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1816 * Reassess DTLs after a config change or scrub completion.
1819 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1821 spa_t
*spa
= vd
->vdev_spa
;
1825 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1827 for (c
= 0; c
< vd
->vdev_children
; c
++)
1828 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1829 scrub_txg
, scrub_done
);
1831 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1834 if (vd
->vdev_ops
->vdev_op_leaf
) {
1835 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1837 mutex_enter(&vd
->vdev_dtl_lock
);
1840 * If we've completed a scan cleanly then determine
1841 * if this vdev should remove any DTLs. We only want to
1842 * excise regions on vdevs that were available during
1843 * the entire duration of this scan.
1845 if (scrub_txg
!= 0 &&
1846 (spa
->spa_scrub_started
||
1847 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1848 vdev_dtl_should_excise(vd
)) {
1850 * We completed a scrub up to scrub_txg. If we
1851 * did it without rebooting, then the scrub dtl
1852 * will be valid, so excise the old region and
1853 * fold in the scrub dtl. Otherwise, leave the
1854 * dtl as-is if there was an error.
1856 * There's little trick here: to excise the beginning
1857 * of the DTL_MISSING map, we put it into a reference
1858 * tree and then add a segment with refcnt -1 that
1859 * covers the range [0, scrub_txg). This means
1860 * that each txg in that range has refcnt -1 or 0.
1861 * We then add DTL_SCRUB with a refcnt of 2, so that
1862 * entries in the range [0, scrub_txg) will have a
1863 * positive refcnt -- either 1 or 2. We then convert
1864 * the reference tree into the new DTL_MISSING map.
1866 space_reftree_create(&reftree
);
1867 space_reftree_add_map(&reftree
,
1868 vd
->vdev_dtl
[DTL_MISSING
], 1);
1869 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1870 space_reftree_add_map(&reftree
,
1871 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1872 space_reftree_generate_map(&reftree
,
1873 vd
->vdev_dtl
[DTL_MISSING
], 1);
1874 space_reftree_destroy(&reftree
);
1876 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1877 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1878 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1880 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1881 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1882 if (!vdev_readable(vd
))
1883 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1885 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1886 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1889 * If the vdev was resilvering and no longer has any
1890 * DTLs then reset its resilvering flag.
1892 if (vd
->vdev_resilver_txg
!= 0 &&
1893 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1894 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0)
1895 vd
->vdev_resilver_txg
= 0;
1897 mutex_exit(&vd
->vdev_dtl_lock
);
1900 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1904 mutex_enter(&vd
->vdev_dtl_lock
);
1905 for (t
= 0; t
< DTL_TYPES
; t
++) {
1908 /* account for child's outage in parent's missing map */
1909 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1911 continue; /* leaf vdevs only */
1912 if (t
== DTL_PARTIAL
)
1913 minref
= 1; /* i.e. non-zero */
1914 else if (vd
->vdev_nparity
!= 0)
1915 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1917 minref
= vd
->vdev_children
; /* any kind of mirror */
1918 space_reftree_create(&reftree
);
1919 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1920 vdev_t
*cvd
= vd
->vdev_child
[c
];
1921 mutex_enter(&cvd
->vdev_dtl_lock
);
1922 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1923 mutex_exit(&cvd
->vdev_dtl_lock
);
1925 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1926 space_reftree_destroy(&reftree
);
1928 mutex_exit(&vd
->vdev_dtl_lock
);
1932 vdev_dtl_load(vdev_t
*vd
)
1934 spa_t
*spa
= vd
->vdev_spa
;
1935 objset_t
*mos
= spa
->spa_meta_objset
;
1939 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
1940 ASSERT(!vd
->vdev_ishole
);
1942 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
1943 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
1946 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
1948 mutex_enter(&vd
->vdev_dtl_lock
);
1951 * Now that we've opened the space_map we need to update
1954 space_map_update(vd
->vdev_dtl_sm
);
1956 error
= space_map_load(vd
->vdev_dtl_sm
,
1957 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
1958 mutex_exit(&vd
->vdev_dtl_lock
);
1963 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1964 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
1973 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1975 spa_t
*spa
= vd
->vdev_spa
;
1976 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
1977 objset_t
*mos
= spa
->spa_meta_objset
;
1978 range_tree_t
*rtsync
;
1981 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
1983 ASSERT(!vd
->vdev_ishole
);
1984 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1986 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1988 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
1989 mutex_enter(&vd
->vdev_dtl_lock
);
1990 space_map_free(vd
->vdev_dtl_sm
, tx
);
1991 space_map_close(vd
->vdev_dtl_sm
);
1992 vd
->vdev_dtl_sm
= NULL
;
1993 mutex_exit(&vd
->vdev_dtl_lock
);
1998 if (vd
->vdev_dtl_sm
== NULL
) {
1999 uint64_t new_object
;
2001 new_object
= space_map_alloc(mos
, tx
);
2002 VERIFY3U(new_object
, !=, 0);
2004 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2005 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2006 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2009 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2011 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2013 mutex_enter(&rtlock
);
2015 mutex_enter(&vd
->vdev_dtl_lock
);
2016 range_tree_walk(rt
, range_tree_add
, rtsync
);
2017 mutex_exit(&vd
->vdev_dtl_lock
);
2019 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2020 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2021 range_tree_vacate(rtsync
, NULL
, NULL
);
2023 range_tree_destroy(rtsync
);
2025 mutex_exit(&rtlock
);
2026 mutex_destroy(&rtlock
);
2029 * If the object for the space map has changed then dirty
2030 * the top level so that we update the config.
2032 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2033 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2034 "new object %llu", txg
, spa_name(spa
), object
,
2035 space_map_object(vd
->vdev_dtl_sm
));
2036 vdev_config_dirty(vd
->vdev_top
);
2041 mutex_enter(&vd
->vdev_dtl_lock
);
2042 space_map_update(vd
->vdev_dtl_sm
);
2043 mutex_exit(&vd
->vdev_dtl_lock
);
2047 * Determine whether the specified vdev can be offlined/detached/removed
2048 * without losing data.
2051 vdev_dtl_required(vdev_t
*vd
)
2053 spa_t
*spa
= vd
->vdev_spa
;
2054 vdev_t
*tvd
= vd
->vdev_top
;
2055 uint8_t cant_read
= vd
->vdev_cant_read
;
2058 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2060 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2064 * Temporarily mark the device as unreadable, and then determine
2065 * whether this results in any DTL outages in the top-level vdev.
2066 * If not, we can safely offline/detach/remove the device.
2068 vd
->vdev_cant_read
= B_TRUE
;
2069 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2070 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2071 vd
->vdev_cant_read
= cant_read
;
2072 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2074 if (!required
&& zio_injection_enabled
)
2075 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2081 * Determine if resilver is needed, and if so the txg range.
2084 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2086 boolean_t needed
= B_FALSE
;
2087 uint64_t thismin
= UINT64_MAX
;
2088 uint64_t thismax
= 0;
2091 if (vd
->vdev_children
== 0) {
2092 mutex_enter(&vd
->vdev_dtl_lock
);
2093 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2094 vdev_writeable(vd
)) {
2096 thismin
= vdev_dtl_min(vd
);
2097 thismax
= vdev_dtl_max(vd
);
2100 mutex_exit(&vd
->vdev_dtl_lock
);
2102 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2103 vdev_t
*cvd
= vd
->vdev_child
[c
];
2104 uint64_t cmin
, cmax
;
2106 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2107 thismin
= MIN(thismin
, cmin
);
2108 thismax
= MAX(thismax
, cmax
);
2114 if (needed
&& minp
) {
2122 vdev_load(vdev_t
*vd
)
2127 * Recursively load all children.
2129 for (c
= 0; c
< vd
->vdev_children
; c
++)
2130 vdev_load(vd
->vdev_child
[c
]);
2133 * If this is a top-level vdev, initialize its metaslabs.
2135 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2136 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2137 vdev_metaslab_init(vd
, 0) != 0))
2138 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2139 VDEV_AUX_CORRUPT_DATA
);
2142 * If this is a leaf vdev, load its DTL.
2144 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2145 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2146 VDEV_AUX_CORRUPT_DATA
);
2150 * The special vdev case is used for hot spares and l2cache devices. Its
2151 * sole purpose it to set the vdev state for the associated vdev. To do this,
2152 * we make sure that we can open the underlying device, then try to read the
2153 * label, and make sure that the label is sane and that it hasn't been
2154 * repurposed to another pool.
2157 vdev_validate_aux(vdev_t
*vd
)
2160 uint64_t guid
, version
;
2163 if (!vdev_readable(vd
))
2166 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2167 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2168 VDEV_AUX_CORRUPT_DATA
);
2172 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2173 !SPA_VERSION_IS_SUPPORTED(version
) ||
2174 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2175 guid
!= vd
->vdev_guid
||
2176 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2177 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2178 VDEV_AUX_CORRUPT_DATA
);
2184 * We don't actually check the pool state here. If it's in fact in
2185 * use by another pool, we update this fact on the fly when requested.
2192 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2194 spa_t
*spa
= vd
->vdev_spa
;
2195 objset_t
*mos
= spa
->spa_meta_objset
;
2199 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2201 if (vd
->vdev_ms
!= NULL
) {
2202 metaslab_group_t
*mg
= vd
->vdev_mg
;
2204 metaslab_group_histogram_verify(mg
);
2205 metaslab_class_histogram_verify(mg
->mg_class
);
2207 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2208 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2210 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2213 mutex_enter(&msp
->ms_lock
);
2215 * If the metaslab was not loaded when the vdev
2216 * was removed then the histogram accounting may
2217 * not be accurate. Update the histogram information
2218 * here so that we ensure that the metaslab group
2219 * and metaslab class are up-to-date.
2221 metaslab_group_histogram_remove(mg
, msp
);
2223 VERIFY0(space_map_allocated(msp
->ms_sm
));
2224 space_map_free(msp
->ms_sm
, tx
);
2225 space_map_close(msp
->ms_sm
);
2227 mutex_exit(&msp
->ms_lock
);
2230 metaslab_group_histogram_verify(mg
);
2231 metaslab_class_histogram_verify(mg
->mg_class
);
2232 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2233 ASSERT0(mg
->mg_histogram
[i
]);
2237 if (vd
->vdev_ms_array
) {
2238 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2239 vd
->vdev_ms_array
= 0;
2245 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2248 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2250 ASSERT(!vd
->vdev_ishole
);
2252 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2253 metaslab_sync_done(msp
, txg
);
2256 metaslab_sync_reassess(vd
->vdev_mg
);
2260 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2262 spa_t
*spa
= vd
->vdev_spa
;
2267 ASSERT(!vd
->vdev_ishole
);
2269 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2270 ASSERT(vd
== vd
->vdev_top
);
2271 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2272 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2273 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2274 ASSERT(vd
->vdev_ms_array
!= 0);
2275 vdev_config_dirty(vd
);
2280 * Remove the metadata associated with this vdev once it's empty.
2282 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2283 vdev_remove(vd
, txg
);
2285 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2286 metaslab_sync(msp
, txg
);
2287 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2290 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2291 vdev_dtl_sync(lvd
, txg
);
2293 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2297 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2299 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2303 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2304 * not be opened, and no I/O is attempted.
2307 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2311 spa_vdev_state_enter(spa
, SCL_NONE
);
2313 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2314 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2316 if (!vd
->vdev_ops
->vdev_op_leaf
)
2317 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2322 * We don't directly use the aux state here, but if we do a
2323 * vdev_reopen(), we need this value to be present to remember why we
2326 vd
->vdev_label_aux
= aux
;
2329 * Faulted state takes precedence over degraded.
2331 vd
->vdev_delayed_close
= B_FALSE
;
2332 vd
->vdev_faulted
= 1ULL;
2333 vd
->vdev_degraded
= 0ULL;
2334 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2337 * If this device has the only valid copy of the data, then
2338 * back off and simply mark the vdev as degraded instead.
2340 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2341 vd
->vdev_degraded
= 1ULL;
2342 vd
->vdev_faulted
= 0ULL;
2345 * If we reopen the device and it's not dead, only then do we
2350 if (vdev_readable(vd
))
2351 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2354 return (spa_vdev_state_exit(spa
, vd
, 0));
2358 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2359 * user that something is wrong. The vdev continues to operate as normal as far
2360 * as I/O is concerned.
2363 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2367 spa_vdev_state_enter(spa
, SCL_NONE
);
2369 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2370 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2372 if (!vd
->vdev_ops
->vdev_op_leaf
)
2373 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2376 * If the vdev is already faulted, then don't do anything.
2378 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2379 return (spa_vdev_state_exit(spa
, NULL
, 0));
2381 vd
->vdev_degraded
= 1ULL;
2382 if (!vdev_is_dead(vd
))
2383 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2386 return (spa_vdev_state_exit(spa
, vd
, 0));
2390 * Online the given vdev.
2392 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2393 * spare device should be detached when the device finishes resilvering.
2394 * Second, the online should be treated like a 'test' online case, so no FMA
2395 * events are generated if the device fails to open.
2398 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2400 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2402 spa_vdev_state_enter(spa
, SCL_NONE
);
2404 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2405 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2407 if (!vd
->vdev_ops
->vdev_op_leaf
)
2408 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2411 vd
->vdev_offline
= B_FALSE
;
2412 vd
->vdev_tmpoffline
= B_FALSE
;
2413 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2414 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2416 /* XXX - L2ARC 1.0 does not support expansion */
2417 if (!vd
->vdev_aux
) {
2418 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2419 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2423 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2425 if (!vd
->vdev_aux
) {
2426 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2427 pvd
->vdev_expanding
= B_FALSE
;
2431 *newstate
= vd
->vdev_state
;
2432 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2433 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2434 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2435 vd
->vdev_parent
->vdev_child
[0] == vd
)
2436 vd
->vdev_unspare
= B_TRUE
;
2438 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2440 /* XXX - L2ARC 1.0 does not support expansion */
2442 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2443 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2445 return (spa_vdev_state_exit(spa
, vd
, 0));
2449 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2453 uint64_t generation
;
2454 metaslab_group_t
*mg
;
2457 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2459 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2460 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2462 if (!vd
->vdev_ops
->vdev_op_leaf
)
2463 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2467 generation
= spa
->spa_config_generation
+ 1;
2470 * If the device isn't already offline, try to offline it.
2472 if (!vd
->vdev_offline
) {
2474 * If this device has the only valid copy of some data,
2475 * don't allow it to be offlined. Log devices are always
2478 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2479 vdev_dtl_required(vd
))
2480 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2483 * If the top-level is a slog and it has had allocations
2484 * then proceed. We check that the vdev's metaslab group
2485 * is not NULL since it's possible that we may have just
2486 * added this vdev but not yet initialized its metaslabs.
2488 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2490 * Prevent any future allocations.
2492 metaslab_group_passivate(mg
);
2493 (void) spa_vdev_state_exit(spa
, vd
, 0);
2495 error
= spa_offline_log(spa
);
2497 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2500 * Check to see if the config has changed.
2502 if (error
|| generation
!= spa
->spa_config_generation
) {
2503 metaslab_group_activate(mg
);
2505 return (spa_vdev_state_exit(spa
,
2507 (void) spa_vdev_state_exit(spa
, vd
, 0);
2510 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2514 * Offline this device and reopen its top-level vdev.
2515 * If the top-level vdev is a log device then just offline
2516 * it. Otherwise, if this action results in the top-level
2517 * vdev becoming unusable, undo it and fail the request.
2519 vd
->vdev_offline
= B_TRUE
;
2522 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2523 vdev_is_dead(tvd
)) {
2524 vd
->vdev_offline
= B_FALSE
;
2526 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2530 * Add the device back into the metaslab rotor so that
2531 * once we online the device it's open for business.
2533 if (tvd
->vdev_islog
&& mg
!= NULL
)
2534 metaslab_group_activate(mg
);
2537 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2539 return (spa_vdev_state_exit(spa
, vd
, 0));
2543 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2547 mutex_enter(&spa
->spa_vdev_top_lock
);
2548 error
= vdev_offline_locked(spa
, guid
, flags
);
2549 mutex_exit(&spa
->spa_vdev_top_lock
);
2555 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2556 * vdev_offline(), we assume the spa config is locked. We also clear all
2557 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2560 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2562 vdev_t
*rvd
= spa
->spa_root_vdev
;
2565 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2570 vd
->vdev_stat
.vs_read_errors
= 0;
2571 vd
->vdev_stat
.vs_write_errors
= 0;
2572 vd
->vdev_stat
.vs_checksum_errors
= 0;
2574 for (c
= 0; c
< vd
->vdev_children
; c
++)
2575 vdev_clear(spa
, vd
->vdev_child
[c
]);
2578 * If we're in the FAULTED state or have experienced failed I/O, then
2579 * clear the persistent state and attempt to reopen the device. We
2580 * also mark the vdev config dirty, so that the new faulted state is
2581 * written out to disk.
2583 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2584 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2587 * When reopening in reponse to a clear event, it may be due to
2588 * a fmadm repair request. In this case, if the device is
2589 * still broken, we want to still post the ereport again.
2591 vd
->vdev_forcefault
= B_TRUE
;
2593 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2594 vd
->vdev_cant_read
= B_FALSE
;
2595 vd
->vdev_cant_write
= B_FALSE
;
2597 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2599 vd
->vdev_forcefault
= B_FALSE
;
2601 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2602 vdev_state_dirty(vd
->vdev_top
);
2604 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2605 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2607 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_CLEAR
);
2611 * When clearing a FMA-diagnosed fault, we always want to
2612 * unspare the device, as we assume that the original spare was
2613 * done in response to the FMA fault.
2615 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2616 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2617 vd
->vdev_parent
->vdev_child
[0] == vd
)
2618 vd
->vdev_unspare
= B_TRUE
;
2622 vdev_is_dead(vdev_t
*vd
)
2625 * Holes and missing devices are always considered "dead".
2626 * This simplifies the code since we don't have to check for
2627 * these types of devices in the various code paths.
2628 * Instead we rely on the fact that we skip over dead devices
2629 * before issuing I/O to them.
2631 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2632 vd
->vdev_ops
== &vdev_missing_ops
);
2636 vdev_readable(vdev_t
*vd
)
2638 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2642 vdev_writeable(vdev_t
*vd
)
2644 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2648 vdev_allocatable(vdev_t
*vd
)
2650 uint64_t state
= vd
->vdev_state
;
2653 * We currently allow allocations from vdevs which may be in the
2654 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2655 * fails to reopen then we'll catch it later when we're holding
2656 * the proper locks. Note that we have to get the vdev state
2657 * in a local variable because although it changes atomically,
2658 * we're asking two separate questions about it.
2660 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2661 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2665 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2667 ASSERT(zio
->io_vd
== vd
);
2669 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2672 if (zio
->io_type
== ZIO_TYPE_READ
)
2673 return (!vd
->vdev_cant_read
);
2675 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2676 return (!vd
->vdev_cant_write
);
2682 * Get statistics for the given vdev.
2685 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2687 spa_t
*spa
= vd
->vdev_spa
;
2688 vdev_t
*rvd
= spa
->spa_root_vdev
;
2691 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2693 mutex_enter(&vd
->vdev_stat_lock
);
2694 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2695 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2696 vs
->vs_state
= vd
->vdev_state
;
2697 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2698 if (vd
->vdev_ops
->vdev_op_leaf
)
2699 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2700 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2701 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&& !vd
->vdev_ishole
) {
2702 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2706 * If we're getting stats on the root vdev, aggregate the I/O counts
2707 * over all top-level vdevs (i.e. the direct children of the root).
2710 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2711 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2712 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2714 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2715 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2716 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2718 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2721 mutex_exit(&vd
->vdev_stat_lock
);
2725 vdev_clear_stats(vdev_t
*vd
)
2727 mutex_enter(&vd
->vdev_stat_lock
);
2728 vd
->vdev_stat
.vs_space
= 0;
2729 vd
->vdev_stat
.vs_dspace
= 0;
2730 vd
->vdev_stat
.vs_alloc
= 0;
2731 mutex_exit(&vd
->vdev_stat_lock
);
2735 vdev_scan_stat_init(vdev_t
*vd
)
2737 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2740 for (c
= 0; c
< vd
->vdev_children
; c
++)
2741 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2743 mutex_enter(&vd
->vdev_stat_lock
);
2744 vs
->vs_scan_processed
= 0;
2745 mutex_exit(&vd
->vdev_stat_lock
);
2749 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2751 spa_t
*spa
= zio
->io_spa
;
2752 vdev_t
*rvd
= spa
->spa_root_vdev
;
2753 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2755 uint64_t txg
= zio
->io_txg
;
2756 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2757 zio_type_t type
= zio
->io_type
;
2758 int flags
= zio
->io_flags
;
2761 * If this i/o is a gang leader, it didn't do any actual work.
2763 if (zio
->io_gang_tree
)
2766 if (zio
->io_error
== 0) {
2768 * If this is a root i/o, don't count it -- we've already
2769 * counted the top-level vdevs, and vdev_get_stats() will
2770 * aggregate them when asked. This reduces contention on
2771 * the root vdev_stat_lock and implicitly handles blocks
2772 * that compress away to holes, for which there is no i/o.
2773 * (Holes never create vdev children, so all the counters
2774 * remain zero, which is what we want.)
2776 * Note: this only applies to successful i/o (io_error == 0)
2777 * because unlike i/o counts, errors are not additive.
2778 * When reading a ditto block, for example, failure of
2779 * one top-level vdev does not imply a root-level error.
2784 ASSERT(vd
== zio
->io_vd
);
2786 if (flags
& ZIO_FLAG_IO_BYPASS
)
2789 mutex_enter(&vd
->vdev_stat_lock
);
2791 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2792 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2793 dsl_scan_phys_t
*scn_phys
=
2794 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2795 uint64_t *processed
= &scn_phys
->scn_processed
;
2798 if (vd
->vdev_ops
->vdev_op_leaf
)
2799 atomic_add_64(processed
, psize
);
2800 vs
->vs_scan_processed
+= psize
;
2803 if (flags
& ZIO_FLAG_SELF_HEAL
)
2804 vs
->vs_self_healed
+= psize
;
2808 vs
->vs_bytes
[type
] += psize
;
2810 mutex_exit(&vd
->vdev_stat_lock
);
2814 if (flags
& ZIO_FLAG_SPECULATIVE
)
2818 * If this is an I/O error that is going to be retried, then ignore the
2819 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2820 * hard errors, when in reality they can happen for any number of
2821 * innocuous reasons (bus resets, MPxIO link failure, etc).
2823 if (zio
->io_error
== EIO
&&
2824 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2828 * Intent logs writes won't propagate their error to the root
2829 * I/O so don't mark these types of failures as pool-level
2832 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2835 mutex_enter(&vd
->vdev_stat_lock
);
2836 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2837 if (zio
->io_error
== ECKSUM
)
2838 vs
->vs_checksum_errors
++;
2840 vs
->vs_read_errors
++;
2842 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2843 vs
->vs_write_errors
++;
2844 mutex_exit(&vd
->vdev_stat_lock
);
2846 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2847 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2848 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2849 spa
->spa_claiming
)) {
2851 * This is either a normal write (not a repair), or it's
2852 * a repair induced by the scrub thread, or it's a repair
2853 * made by zil_claim() during spa_load() in the first txg.
2854 * In the normal case, we commit the DTL change in the same
2855 * txg as the block was born. In the scrub-induced repair
2856 * case, we know that scrubs run in first-pass syncing context,
2857 * so we commit the DTL change in spa_syncing_txg(spa).
2858 * In the zil_claim() case, we commit in spa_first_txg(spa).
2860 * We currently do not make DTL entries for failed spontaneous
2861 * self-healing writes triggered by normal (non-scrubbing)
2862 * reads, because we have no transactional context in which to
2863 * do so -- and it's not clear that it'd be desirable anyway.
2865 if (vd
->vdev_ops
->vdev_op_leaf
) {
2866 uint64_t commit_txg
= txg
;
2867 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2868 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2869 ASSERT(spa_sync_pass(spa
) == 1);
2870 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2871 commit_txg
= spa_syncing_txg(spa
);
2872 } else if (spa
->spa_claiming
) {
2873 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2874 commit_txg
= spa_first_txg(spa
);
2876 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2877 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2879 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2880 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2881 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2884 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2889 * Update the in-core space usage stats for this vdev, its metaslab class,
2890 * and the root vdev.
2893 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2894 int64_t space_delta
)
2896 int64_t dspace_delta
= space_delta
;
2897 spa_t
*spa
= vd
->vdev_spa
;
2898 vdev_t
*rvd
= spa
->spa_root_vdev
;
2899 metaslab_group_t
*mg
= vd
->vdev_mg
;
2900 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2902 ASSERT(vd
== vd
->vdev_top
);
2905 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2906 * factor. We must calculate this here and not at the root vdev
2907 * because the root vdev's psize-to-asize is simply the max of its
2908 * childrens', thus not accurate enough for us.
2910 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2911 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2912 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2913 vd
->vdev_deflate_ratio
;
2915 mutex_enter(&vd
->vdev_stat_lock
);
2916 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2917 vd
->vdev_stat
.vs_space
+= space_delta
;
2918 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2919 mutex_exit(&vd
->vdev_stat_lock
);
2921 if (mc
== spa_normal_class(spa
)) {
2922 mutex_enter(&rvd
->vdev_stat_lock
);
2923 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2924 rvd
->vdev_stat
.vs_space
+= space_delta
;
2925 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2926 mutex_exit(&rvd
->vdev_stat_lock
);
2930 ASSERT(rvd
== vd
->vdev_parent
);
2931 ASSERT(vd
->vdev_ms_count
!= 0);
2933 metaslab_class_space_update(mc
,
2934 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2939 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2940 * so that it will be written out next time the vdev configuration is synced.
2941 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2944 vdev_config_dirty(vdev_t
*vd
)
2946 spa_t
*spa
= vd
->vdev_spa
;
2947 vdev_t
*rvd
= spa
->spa_root_vdev
;
2950 ASSERT(spa_writeable(spa
));
2953 * If this is an aux vdev (as with l2cache and spare devices), then we
2954 * update the vdev config manually and set the sync flag.
2956 if (vd
->vdev_aux
!= NULL
) {
2957 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2961 for (c
= 0; c
< sav
->sav_count
; c
++) {
2962 if (sav
->sav_vdevs
[c
] == vd
)
2966 if (c
== sav
->sav_count
) {
2968 * We're being removed. There's nothing more to do.
2970 ASSERT(sav
->sav_sync
== B_TRUE
);
2974 sav
->sav_sync
= B_TRUE
;
2976 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2977 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2978 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2979 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2985 * Setting the nvlist in the middle if the array is a little
2986 * sketchy, but it will work.
2988 nvlist_free(aux
[c
]);
2989 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
2995 * The dirty list is protected by the SCL_CONFIG lock. The caller
2996 * must either hold SCL_CONFIG as writer, or must be the sync thread
2997 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2998 * so this is sufficient to ensure mutual exclusion.
3000 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3001 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3002 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3005 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3006 vdev_config_dirty(rvd
->vdev_child
[c
]);
3008 ASSERT(vd
== vd
->vdev_top
);
3010 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3012 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3017 vdev_config_clean(vdev_t
*vd
)
3019 spa_t
*spa
= vd
->vdev_spa
;
3021 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3022 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3023 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3025 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3026 list_remove(&spa
->spa_config_dirty_list
, vd
);
3030 * Mark a top-level vdev's state as dirty, so that the next pass of
3031 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3032 * the state changes from larger config changes because they require
3033 * much less locking, and are often needed for administrative actions.
3036 vdev_state_dirty(vdev_t
*vd
)
3038 spa_t
*spa
= vd
->vdev_spa
;
3040 ASSERT(spa_writeable(spa
));
3041 ASSERT(vd
== vd
->vdev_top
);
3044 * The state list is protected by the SCL_STATE lock. The caller
3045 * must either hold SCL_STATE as writer, or must be the sync thread
3046 * (which holds SCL_STATE as reader). There's only one sync thread,
3047 * so this is sufficient to ensure mutual exclusion.
3049 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3050 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3051 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3053 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3054 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3058 vdev_state_clean(vdev_t
*vd
)
3060 spa_t
*spa
= vd
->vdev_spa
;
3062 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3063 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3064 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3066 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3067 list_remove(&spa
->spa_state_dirty_list
, vd
);
3071 * Propagate vdev state up from children to parent.
3074 vdev_propagate_state(vdev_t
*vd
)
3076 spa_t
*spa
= vd
->vdev_spa
;
3077 vdev_t
*rvd
= spa
->spa_root_vdev
;
3078 int degraded
= 0, faulted
= 0;
3083 if (vd
->vdev_children
> 0) {
3084 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3085 child
= vd
->vdev_child
[c
];
3088 * Don't factor holes into the decision.
3090 if (child
->vdev_ishole
)
3093 if (!vdev_readable(child
) ||
3094 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3096 * Root special: if there is a top-level log
3097 * device, treat the root vdev as if it were
3100 if (child
->vdev_islog
&& vd
== rvd
)
3104 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3108 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3112 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3115 * Root special: if there is a top-level vdev that cannot be
3116 * opened due to corrupted metadata, then propagate the root
3117 * vdev's aux state as 'corrupt' rather than 'insufficient
3120 if (corrupted
&& vd
== rvd
&&
3121 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3122 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3123 VDEV_AUX_CORRUPT_DATA
);
3126 if (vd
->vdev_parent
)
3127 vdev_propagate_state(vd
->vdev_parent
);
3131 * Set a vdev's state. If this is during an open, we don't update the parent
3132 * state, because we're in the process of opening children depth-first.
3133 * Otherwise, we propagate the change to the parent.
3135 * If this routine places a device in a faulted state, an appropriate ereport is
3139 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3141 uint64_t save_state
;
3142 spa_t
*spa
= vd
->vdev_spa
;
3144 if (state
== vd
->vdev_state
) {
3145 vd
->vdev_stat
.vs_aux
= aux
;
3149 save_state
= vd
->vdev_state
;
3151 vd
->vdev_state
= state
;
3152 vd
->vdev_stat
.vs_aux
= aux
;
3155 * If we are setting the vdev state to anything but an open state, then
3156 * always close the underlying device unless the device has requested
3157 * a delayed close (i.e. we're about to remove or fault the device).
3158 * Otherwise, we keep accessible but invalid devices open forever.
3159 * We don't call vdev_close() itself, because that implies some extra
3160 * checks (offline, etc) that we don't want here. This is limited to
3161 * leaf devices, because otherwise closing the device will affect other
3164 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3165 vd
->vdev_ops
->vdev_op_leaf
)
3166 vd
->vdev_ops
->vdev_op_close(vd
);
3169 * If we have brought this vdev back into service, we need
3170 * to notify fmd so that it can gracefully repair any outstanding
3171 * cases due to a missing device. We do this in all cases, even those
3172 * that probably don't correlate to a repaired fault. This is sure to
3173 * catch all cases, and we let the zfs-retire agent sort it out. If
3174 * this is a transient state it's OK, as the retire agent will
3175 * double-check the state of the vdev before repairing it.
3177 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
3178 vd
->vdev_prevstate
!= state
)
3179 zfs_post_state_change(spa
, vd
);
3181 if (vd
->vdev_removed
&&
3182 state
== VDEV_STATE_CANT_OPEN
&&
3183 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3185 * If the previous state is set to VDEV_STATE_REMOVED, then this
3186 * device was previously marked removed and someone attempted to
3187 * reopen it. If this failed due to a nonexistent device, then
3188 * keep the device in the REMOVED state. We also let this be if
3189 * it is one of our special test online cases, which is only
3190 * attempting to online the device and shouldn't generate an FMA
3193 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3194 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3195 } else if (state
== VDEV_STATE_REMOVED
) {
3196 vd
->vdev_removed
= B_TRUE
;
3197 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3199 * If we fail to open a vdev during an import or recovery, we
3200 * mark it as "not available", which signifies that it was
3201 * never there to begin with. Failure to open such a device
3202 * is not considered an error.
3204 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3205 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3206 vd
->vdev_ops
->vdev_op_leaf
)
3207 vd
->vdev_not_present
= 1;
3210 * Post the appropriate ereport. If the 'prevstate' field is
3211 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3212 * that this is part of a vdev_reopen(). In this case, we don't
3213 * want to post the ereport if the device was already in the
3214 * CANT_OPEN state beforehand.
3216 * If the 'checkremove' flag is set, then this is an attempt to
3217 * online the device in response to an insertion event. If we
3218 * hit this case, then we have detected an insertion event for a
3219 * faulted or offline device that wasn't in the removed state.
3220 * In this scenario, we don't post an ereport because we are
3221 * about to replace the device, or attempt an online with
3222 * vdev_forcefault, which will generate the fault for us.
3224 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3225 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3226 vd
!= spa
->spa_root_vdev
) {
3230 case VDEV_AUX_OPEN_FAILED
:
3231 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3233 case VDEV_AUX_CORRUPT_DATA
:
3234 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3236 case VDEV_AUX_NO_REPLICAS
:
3237 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3239 case VDEV_AUX_BAD_GUID_SUM
:
3240 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3242 case VDEV_AUX_TOO_SMALL
:
3243 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3245 case VDEV_AUX_BAD_LABEL
:
3246 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3249 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3252 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3255 /* Erase any notion of persistent removed state */
3256 vd
->vdev_removed
= B_FALSE
;
3258 vd
->vdev_removed
= B_FALSE
;
3261 if (!isopen
&& vd
->vdev_parent
)
3262 vdev_propagate_state(vd
->vdev_parent
);
3266 * Check the vdev configuration to ensure that it's capable of supporting
3270 vdev_is_bootable(vdev_t
*vd
)
3272 #if defined(__sun__) || defined(__sun)
3274 * Currently, we do not support RAID-Z or partial configuration.
3275 * In addition, only a single top-level vdev is allowed and none of the
3276 * leaves can be wholedisks.
3280 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3281 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3283 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3284 vd
->vdev_children
> 1) {
3286 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3287 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3290 } else if (vd
->vdev_wholedisk
== 1) {
3294 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3295 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3298 #endif /* __sun__ || __sun */
3303 * Load the state from the original vdev tree (ovd) which
3304 * we've retrieved from the MOS config object. If the original
3305 * vdev was offline or faulted then we transfer that state to the
3306 * device in the current vdev tree (nvd).
3309 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3313 ASSERT(nvd
->vdev_top
->vdev_islog
);
3314 ASSERT(spa_config_held(nvd
->vdev_spa
,
3315 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3316 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3318 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3319 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3321 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3323 * Restore the persistent vdev state
3325 nvd
->vdev_offline
= ovd
->vdev_offline
;
3326 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3327 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3328 nvd
->vdev_removed
= ovd
->vdev_removed
;
3333 * Determine if a log device has valid content. If the vdev was
3334 * removed or faulted in the MOS config then we know that
3335 * the content on the log device has already been written to the pool.
3338 vdev_log_state_valid(vdev_t
*vd
)
3342 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3346 for (c
= 0; c
< vd
->vdev_children
; c
++)
3347 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3354 * Expand a vdev if possible.
3357 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3359 ASSERT(vd
->vdev_top
== vd
);
3360 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3362 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3363 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3364 vdev_config_dirty(vd
);
3372 vdev_split(vdev_t
*vd
)
3374 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3376 vdev_remove_child(pvd
, vd
);
3377 vdev_compact_children(pvd
);
3379 cvd
= pvd
->vdev_child
[0];
3380 if (pvd
->vdev_children
== 1) {
3381 vdev_remove_parent(cvd
);
3382 cvd
->vdev_splitting
= B_TRUE
;
3384 vdev_propagate_state(cvd
);
3388 vdev_deadman(vdev_t
*vd
)
3392 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3393 vdev_t
*cvd
= vd
->vdev_child
[c
];
3398 if (vd
->vdev_ops
->vdev_op_leaf
) {
3399 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3401 mutex_enter(&vq
->vq_lock
);
3402 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3403 spa_t
*spa
= vd
->vdev_spa
;
3408 * Look at the head of all the pending queues,
3409 * if any I/O has been outstanding for longer than
3410 * the spa_deadman_synctime we log a zevent.
3412 fio
= avl_first(&vq
->vq_active_tree
);
3413 delta
= gethrtime() - fio
->io_timestamp
;
3414 if (delta
> spa_deadman_synctime(spa
)) {
3415 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3416 "delta %lluns, last io %lluns",
3417 fio
->io_timestamp
, delta
,
3418 vq
->vq_io_complete_ts
);
3419 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3420 spa
, vd
, fio
, 0, 0);
3423 mutex_exit(&vq
->vq_lock
);
3427 #if defined(_KERNEL) && defined(HAVE_SPL)
3428 EXPORT_SYMBOL(vdev_fault
);
3429 EXPORT_SYMBOL(vdev_degrade
);
3430 EXPORT_SYMBOL(vdev_online
);
3431 EXPORT_SYMBOL(vdev_offline
);
3432 EXPORT_SYMBOL(vdev_clear
);
3434 module_param(metaslabs_per_vdev
, int, 0644);
3435 MODULE_PARM_DESC(metaslabs_per_vdev
,
3436 "Divide added vdev into approximately (but no more than) this number "