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>
47 #include <sys/zfs_ratelimit.h>
50 * When a vdev is added, it will be divided into approximately (but no
51 * more than) this number of metaslabs.
53 int metaslabs_per_vdev
= 200;
56 * Virtual device management.
59 static vdev_ops_t
*vdev_ops_table
[] = {
73 * Given a vdev type, return the appropriate ops vector.
76 vdev_getops(const char *type
)
78 vdev_ops_t
*ops
, **opspp
;
80 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
81 if (strcmp(ops
->vdev_op_type
, type
) == 0)
88 * Default asize function: return the MAX of psize with the asize of
89 * all children. This is what's used by anything other than RAID-Z.
92 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
94 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
98 for (c
= 0; c
< vd
->vdev_children
; c
++) {
99 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
100 asize
= MAX(asize
, csize
);
107 * Get the minimum allocatable size. We define the allocatable size as
108 * the vdev's asize rounded to the nearest metaslab. This allows us to
109 * replace or attach devices which don't have the same physical size but
110 * can still satisfy the same number of allocations.
113 vdev_get_min_asize(vdev_t
*vd
)
115 vdev_t
*pvd
= vd
->vdev_parent
;
118 * If our parent is NULL (inactive spare or cache) or is the root,
119 * just return our own asize.
122 return (vd
->vdev_asize
);
125 * The top-level vdev just returns the allocatable size rounded
126 * to the nearest metaslab.
128 if (vd
== vd
->vdev_top
)
129 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
132 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
133 * so each child must provide at least 1/Nth of its asize.
135 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
136 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
138 return (pvd
->vdev_min_asize
);
142 vdev_set_min_asize(vdev_t
*vd
)
145 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
147 for (c
= 0; c
< vd
->vdev_children
; c
++)
148 vdev_set_min_asize(vd
->vdev_child
[c
]);
152 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
154 vdev_t
*rvd
= spa
->spa_root_vdev
;
156 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
158 if (vdev
< rvd
->vdev_children
) {
159 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
160 return (rvd
->vdev_child
[vdev
]);
167 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
172 if (vd
->vdev_guid
== guid
)
175 for (c
= 0; c
< vd
->vdev_children
; c
++)
176 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
184 vdev_count_leaves_impl(vdev_t
*vd
)
189 if (vd
->vdev_ops
->vdev_op_leaf
)
192 for (c
= 0; c
< vd
->vdev_children
; c
++)
193 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
199 vdev_count_leaves(spa_t
*spa
)
201 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
205 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
207 size_t oldsize
, newsize
;
208 uint64_t id
= cvd
->vdev_id
;
211 ASSERT(spa_config_held(cvd
->vdev_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
);
351 * Initialize rate limit structs for events. We rate limit ZIO delay
352 * and checksum events so that we don't overwhelm ZED with thousands
353 * of events when a disk is acting up.
355 zfs_ratelimit_init(&vd
->vdev_delay_rl
, DELAYS_PER_SECOND
, 1);
356 zfs_ratelimit_init(&vd
->vdev_checksum_rl
, CHECKSUMS_PER_SECOND
, 1);
358 list_link_init(&vd
->vdev_config_dirty_node
);
359 list_link_init(&vd
->vdev_state_dirty_node
);
360 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
361 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
362 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
363 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
365 for (t
= 0; t
< DTL_TYPES
; t
++) {
366 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
369 txg_list_create(&vd
->vdev_ms_list
,
370 offsetof(struct metaslab
, ms_txg_node
));
371 txg_list_create(&vd
->vdev_dtl_list
,
372 offsetof(struct vdev
, vdev_dtl_node
));
373 vd
->vdev_stat
.vs_timestamp
= gethrtime();
381 * Allocate a new vdev. The 'alloctype' is used to control whether we are
382 * creating a new vdev or loading an existing one - the behavior is slightly
383 * different for each case.
386 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
391 uint64_t guid
= 0, islog
, nparity
;
394 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
396 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
397 return (SET_ERROR(EINVAL
));
399 if ((ops
= vdev_getops(type
)) == NULL
)
400 return (SET_ERROR(EINVAL
));
403 * If this is a load, get the vdev guid from the nvlist.
404 * Otherwise, vdev_alloc_common() will generate one for us.
406 if (alloctype
== VDEV_ALLOC_LOAD
) {
409 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
411 return (SET_ERROR(EINVAL
));
413 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
414 return (SET_ERROR(EINVAL
));
415 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
416 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
417 return (SET_ERROR(EINVAL
));
418 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
419 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
420 return (SET_ERROR(EINVAL
));
421 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
422 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
423 return (SET_ERROR(EINVAL
));
427 * The first allocated vdev must be of type 'root'.
429 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
430 return (SET_ERROR(EINVAL
));
433 * Determine whether we're a log vdev.
436 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
437 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
438 return (SET_ERROR(ENOTSUP
));
440 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
441 return (SET_ERROR(ENOTSUP
));
444 * Set the nparity property for RAID-Z vdevs.
447 if (ops
== &vdev_raidz_ops
) {
448 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
450 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
451 return (SET_ERROR(EINVAL
));
453 * Previous versions could only support 1 or 2 parity
457 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
458 return (SET_ERROR(ENOTSUP
));
460 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
461 return (SET_ERROR(ENOTSUP
));
464 * We require the parity to be specified for SPAs that
465 * support multiple parity levels.
467 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
468 return (SET_ERROR(EINVAL
));
470 * Otherwise, we default to 1 parity device for RAID-Z.
477 ASSERT(nparity
!= -1ULL);
479 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
481 vd
->vdev_islog
= islog
;
482 vd
->vdev_nparity
= nparity
;
484 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
485 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
486 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
487 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
488 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
489 &vd
->vdev_physpath
) == 0)
490 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
491 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
492 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
495 * Set the whole_disk property. If it's not specified, leave the value
498 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
499 &vd
->vdev_wholedisk
) != 0)
500 vd
->vdev_wholedisk
= -1ULL;
503 * Look for the 'not present' flag. This will only be set if the device
504 * was not present at the time of import.
506 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
507 &vd
->vdev_not_present
);
510 * Get the alignment requirement.
512 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
515 * Retrieve the vdev creation time.
517 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
521 * If we're a top-level vdev, try to load the allocation parameters.
523 if (parent
&& !parent
->vdev_parent
&&
524 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
525 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
527 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
529 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
531 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
533 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
536 ASSERT0(vd
->vdev_top_zap
);
539 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
540 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
541 alloctype
== VDEV_ALLOC_ADD
||
542 alloctype
== VDEV_ALLOC_SPLIT
||
543 alloctype
== VDEV_ALLOC_ROOTPOOL
);
544 vd
->vdev_mg
= metaslab_group_create(islog
?
545 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
548 if (vd
->vdev_ops
->vdev_op_leaf
&&
549 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
550 (void) nvlist_lookup_uint64(nv
,
551 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
553 ASSERT0(vd
->vdev_leaf_zap
);
557 * If we're a leaf vdev, try to load the DTL object and other state.
560 if (vd
->vdev_ops
->vdev_op_leaf
&&
561 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
562 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
563 if (alloctype
== VDEV_ALLOC_LOAD
) {
564 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
565 &vd
->vdev_dtl_object
);
566 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
570 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
573 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
574 &spare
) == 0 && spare
)
578 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
581 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
582 &vd
->vdev_resilver_txg
);
585 * When importing a pool, we want to ignore the persistent fault
586 * state, as the diagnosis made on another system may not be
587 * valid in the current context. Local vdevs will
588 * remain in the faulted state.
590 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
591 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
593 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
595 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
598 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
602 VDEV_AUX_ERR_EXCEEDED
;
603 if (nvlist_lookup_string(nv
,
604 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
605 strcmp(aux
, "external") == 0)
606 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
612 * Add ourselves to the parent's list of children.
614 vdev_add_child(parent
, vd
);
622 vdev_free(vdev_t
*vd
)
625 spa_t
*spa
= vd
->vdev_spa
;
628 * vdev_free() implies closing the vdev first. This is simpler than
629 * trying to ensure complicated semantics for all callers.
633 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
634 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
639 for (c
= 0; c
< vd
->vdev_children
; c
++)
640 vdev_free(vd
->vdev_child
[c
]);
642 ASSERT(vd
->vdev_child
== NULL
);
643 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
646 * Discard allocation state.
648 if (vd
->vdev_mg
!= NULL
) {
649 vdev_metaslab_fini(vd
);
650 metaslab_group_destroy(vd
->vdev_mg
);
653 ASSERT0(vd
->vdev_stat
.vs_space
);
654 ASSERT0(vd
->vdev_stat
.vs_dspace
);
655 ASSERT0(vd
->vdev_stat
.vs_alloc
);
658 * Remove this vdev from its parent's child list.
660 vdev_remove_child(vd
->vdev_parent
, vd
);
662 ASSERT(vd
->vdev_parent
== NULL
);
665 * Clean up vdev structure.
671 spa_strfree(vd
->vdev_path
);
673 spa_strfree(vd
->vdev_devid
);
674 if (vd
->vdev_physpath
)
675 spa_strfree(vd
->vdev_physpath
);
677 spa_strfree(vd
->vdev_fru
);
679 if (vd
->vdev_isspare
)
680 spa_spare_remove(vd
);
681 if (vd
->vdev_isl2cache
)
682 spa_l2cache_remove(vd
);
684 txg_list_destroy(&vd
->vdev_ms_list
);
685 txg_list_destroy(&vd
->vdev_dtl_list
);
687 mutex_enter(&vd
->vdev_dtl_lock
);
688 space_map_close(vd
->vdev_dtl_sm
);
689 for (t
= 0; t
< DTL_TYPES
; t
++) {
690 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
691 range_tree_destroy(vd
->vdev_dtl
[t
]);
693 mutex_exit(&vd
->vdev_dtl_lock
);
695 mutex_destroy(&vd
->vdev_queue_lock
);
696 mutex_destroy(&vd
->vdev_dtl_lock
);
697 mutex_destroy(&vd
->vdev_stat_lock
);
698 mutex_destroy(&vd
->vdev_probe_lock
);
700 if (vd
== spa
->spa_root_vdev
)
701 spa
->spa_root_vdev
= NULL
;
703 kmem_free(vd
, sizeof (vdev_t
));
707 * Transfer top-level vdev state from svd to tvd.
710 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
712 spa_t
*spa
= svd
->vdev_spa
;
717 ASSERT(tvd
== tvd
->vdev_top
);
719 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
720 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
721 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
722 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
723 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
725 svd
->vdev_ms_array
= 0;
726 svd
->vdev_ms_shift
= 0;
727 svd
->vdev_ms_count
= 0;
728 svd
->vdev_top_zap
= 0;
731 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
732 tvd
->vdev_mg
= svd
->vdev_mg
;
733 tvd
->vdev_ms
= svd
->vdev_ms
;
738 if (tvd
->vdev_mg
!= NULL
)
739 tvd
->vdev_mg
->mg_vd
= tvd
;
741 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
742 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
743 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
745 svd
->vdev_stat
.vs_alloc
= 0;
746 svd
->vdev_stat
.vs_space
= 0;
747 svd
->vdev_stat
.vs_dspace
= 0;
749 for (t
= 0; t
< TXG_SIZE
; t
++) {
750 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
751 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
752 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
753 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
754 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
755 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
758 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
759 vdev_config_clean(svd
);
760 vdev_config_dirty(tvd
);
763 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
764 vdev_state_clean(svd
);
765 vdev_state_dirty(tvd
);
768 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
769 svd
->vdev_deflate_ratio
= 0;
771 tvd
->vdev_islog
= svd
->vdev_islog
;
776 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
785 for (c
= 0; c
< vd
->vdev_children
; c
++)
786 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
790 * Add a mirror/replacing vdev above an existing vdev.
793 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
795 spa_t
*spa
= cvd
->vdev_spa
;
796 vdev_t
*pvd
= cvd
->vdev_parent
;
799 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
801 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
803 mvd
->vdev_asize
= cvd
->vdev_asize
;
804 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
805 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
806 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
807 mvd
->vdev_state
= cvd
->vdev_state
;
808 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
810 vdev_remove_child(pvd
, cvd
);
811 vdev_add_child(pvd
, mvd
);
812 cvd
->vdev_id
= mvd
->vdev_children
;
813 vdev_add_child(mvd
, cvd
);
814 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
816 if (mvd
== mvd
->vdev_top
)
817 vdev_top_transfer(cvd
, mvd
);
823 * Remove a 1-way mirror/replacing vdev from the tree.
826 vdev_remove_parent(vdev_t
*cvd
)
828 vdev_t
*mvd
= cvd
->vdev_parent
;
829 vdev_t
*pvd
= mvd
->vdev_parent
;
831 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
833 ASSERT(mvd
->vdev_children
== 1);
834 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
835 mvd
->vdev_ops
== &vdev_replacing_ops
||
836 mvd
->vdev_ops
== &vdev_spare_ops
);
837 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
839 vdev_remove_child(mvd
, cvd
);
840 vdev_remove_child(pvd
, mvd
);
843 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
844 * Otherwise, we could have detached an offline device, and when we
845 * go to import the pool we'll think we have two top-level vdevs,
846 * instead of a different version of the same top-level vdev.
848 if (mvd
->vdev_top
== mvd
) {
849 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
850 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
851 cvd
->vdev_guid
+= guid_delta
;
852 cvd
->vdev_guid_sum
+= guid_delta
;
855 * If pool not set for autoexpand, we need to also preserve
856 * mvd's asize to prevent automatic expansion of cvd.
857 * Otherwise if we are adjusting the mirror by attaching and
858 * detaching children of non-uniform sizes, the mirror could
859 * autoexpand, unexpectedly requiring larger devices to
860 * re-establish the mirror.
862 if (!cvd
->vdev_spa
->spa_autoexpand
)
863 cvd
->vdev_asize
= mvd
->vdev_asize
;
865 cvd
->vdev_id
= mvd
->vdev_id
;
866 vdev_add_child(pvd
, cvd
);
867 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
869 if (cvd
== cvd
->vdev_top
)
870 vdev_top_transfer(mvd
, cvd
);
872 ASSERT(mvd
->vdev_children
== 0);
877 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
879 spa_t
*spa
= vd
->vdev_spa
;
880 objset_t
*mos
= spa
->spa_meta_objset
;
882 uint64_t oldc
= vd
->vdev_ms_count
;
883 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
887 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
890 * This vdev is not being allocated from yet or is a hole.
892 if (vd
->vdev_ms_shift
== 0)
895 ASSERT(!vd
->vdev_ishole
);
898 * Compute the raidz-deflation ratio. Note, we hard-code
899 * in 128k (1 << 17) because it is the "typical" blocksize.
900 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
901 * otherwise it would inconsistently account for existing bp's.
903 vd
->vdev_deflate_ratio
= (1 << 17) /
904 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
906 ASSERT(oldc
<= newc
);
908 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
911 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
912 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
916 vd
->vdev_ms_count
= newc
;
918 for (m
= oldc
; m
< newc
; m
++) {
922 error
= dmu_read(mos
, vd
->vdev_ms_array
,
923 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
929 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
936 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
939 * If the vdev is being removed we don't activate
940 * the metaslabs since we want to ensure that no new
941 * allocations are performed on this device.
943 if (oldc
== 0 && !vd
->vdev_removing
)
944 metaslab_group_activate(vd
->vdev_mg
);
947 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
953 vdev_metaslab_fini(vdev_t
*vd
)
956 uint64_t count
= vd
->vdev_ms_count
;
958 if (vd
->vdev_ms
!= NULL
) {
959 metaslab_group_passivate(vd
->vdev_mg
);
960 for (m
= 0; m
< count
; m
++) {
961 metaslab_t
*msp
= vd
->vdev_ms
[m
];
966 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
970 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
973 typedef struct vdev_probe_stats
{
974 boolean_t vps_readable
;
975 boolean_t vps_writeable
;
977 } vdev_probe_stats_t
;
980 vdev_probe_done(zio_t
*zio
)
982 spa_t
*spa
= zio
->io_spa
;
983 vdev_t
*vd
= zio
->io_vd
;
984 vdev_probe_stats_t
*vps
= zio
->io_private
;
986 ASSERT(vd
->vdev_probe_zio
!= NULL
);
988 if (zio
->io_type
== ZIO_TYPE_READ
) {
989 if (zio
->io_error
== 0)
990 vps
->vps_readable
= 1;
991 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
992 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
993 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
994 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
995 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
997 zio_buf_free(zio
->io_data
, zio
->io_size
);
999 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1000 if (zio
->io_error
== 0)
1001 vps
->vps_writeable
= 1;
1002 zio_buf_free(zio
->io_data
, zio
->io_size
);
1003 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1007 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1008 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1010 if (vdev_readable(vd
) &&
1011 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1014 ASSERT(zio
->io_error
!= 0);
1015 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1016 spa
, vd
, NULL
, 0, 0);
1017 zio
->io_error
= SET_ERROR(ENXIO
);
1020 mutex_enter(&vd
->vdev_probe_lock
);
1021 ASSERT(vd
->vdev_probe_zio
== zio
);
1022 vd
->vdev_probe_zio
= NULL
;
1023 mutex_exit(&vd
->vdev_probe_lock
);
1026 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1027 if (!vdev_accessible(vd
, pio
))
1028 pio
->io_error
= SET_ERROR(ENXIO
);
1030 kmem_free(vps
, sizeof (*vps
));
1035 * Determine whether this device is accessible.
1037 * Read and write to several known locations: the pad regions of each
1038 * vdev label but the first, which we leave alone in case it contains
1042 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1044 spa_t
*spa
= vd
->vdev_spa
;
1045 vdev_probe_stats_t
*vps
= NULL
;
1049 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1052 * Don't probe the probe.
1054 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1058 * To prevent 'probe storms' when a device fails, we create
1059 * just one probe i/o at a time. All zios that want to probe
1060 * this vdev will become parents of the probe io.
1062 mutex_enter(&vd
->vdev_probe_lock
);
1064 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1065 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1067 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1068 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1071 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1073 * vdev_cant_read and vdev_cant_write can only
1074 * transition from TRUE to FALSE when we have the
1075 * SCL_ZIO lock as writer; otherwise they can only
1076 * transition from FALSE to TRUE. This ensures that
1077 * any zio looking at these values can assume that
1078 * failures persist for the life of the I/O. That's
1079 * important because when a device has intermittent
1080 * connectivity problems, we want to ensure that
1081 * they're ascribed to the device (ENXIO) and not
1084 * Since we hold SCL_ZIO as writer here, clear both
1085 * values so the probe can reevaluate from first
1088 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1089 vd
->vdev_cant_read
= B_FALSE
;
1090 vd
->vdev_cant_write
= B_FALSE
;
1093 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1094 vdev_probe_done
, vps
,
1095 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1098 * We can't change the vdev state in this context, so we
1099 * kick off an async task to do it on our behalf.
1102 vd
->vdev_probe_wanted
= B_TRUE
;
1103 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1108 zio_add_child(zio
, pio
);
1110 mutex_exit(&vd
->vdev_probe_lock
);
1113 ASSERT(zio
!= NULL
);
1117 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1118 zio_nowait(zio_read_phys(pio
, vd
,
1119 vdev_label_offset(vd
->vdev_psize
, l
,
1120 offsetof(vdev_label_t
, vl_pad2
)),
1121 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1122 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1123 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1134 vdev_open_child(void *arg
)
1138 vd
->vdev_open_thread
= curthread
;
1139 vd
->vdev_open_error
= vdev_open(vd
);
1140 vd
->vdev_open_thread
= NULL
;
1144 vdev_uses_zvols(vdev_t
*vd
)
1149 if (zvol_is_zvol(vd
->vdev_path
))
1153 for (c
= 0; c
< vd
->vdev_children
; c
++)
1154 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1161 vdev_open_children(vdev_t
*vd
)
1164 int children
= vd
->vdev_children
;
1168 * in order to handle pools on top of zvols, do the opens
1169 * in a single thread so that the same thread holds the
1170 * spa_namespace_lock
1172 if (vdev_uses_zvols(vd
)) {
1173 for (c
= 0; c
< children
; c
++)
1174 vd
->vdev_child
[c
]->vdev_open_error
=
1175 vdev_open(vd
->vdev_child
[c
]);
1177 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1178 children
, children
, TASKQ_PREPOPULATE
);
1180 for (c
= 0; c
< children
; c
++)
1181 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1182 vd
->vdev_child
[c
], TQ_SLEEP
) != 0);
1187 vd
->vdev_nonrot
= B_TRUE
;
1189 for (c
= 0; c
< children
; c
++)
1190 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1194 * Prepare a virtual device for access.
1197 vdev_open(vdev_t
*vd
)
1199 spa_t
*spa
= vd
->vdev_spa
;
1202 uint64_t max_osize
= 0;
1203 uint64_t asize
, max_asize
, psize
;
1204 uint64_t ashift
= 0;
1207 ASSERT(vd
->vdev_open_thread
== curthread
||
1208 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1209 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1210 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1211 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1213 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1214 vd
->vdev_cant_read
= B_FALSE
;
1215 vd
->vdev_cant_write
= B_FALSE
;
1216 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1219 * If this vdev is not removed, check its fault status. If it's
1220 * faulted, bail out of the open.
1222 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1223 ASSERT(vd
->vdev_children
== 0);
1224 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1225 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1226 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1227 vd
->vdev_label_aux
);
1228 return (SET_ERROR(ENXIO
));
1229 } else if (vd
->vdev_offline
) {
1230 ASSERT(vd
->vdev_children
== 0);
1231 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1232 return (SET_ERROR(ENXIO
));
1235 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1238 * Reset the vdev_reopening flag so that we actually close
1239 * the vdev on error.
1241 vd
->vdev_reopening
= B_FALSE
;
1242 if (zio_injection_enabled
&& error
== 0)
1243 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1246 if (vd
->vdev_removed
&&
1247 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1248 vd
->vdev_removed
= B_FALSE
;
1250 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1251 vd
->vdev_stat
.vs_aux
);
1255 vd
->vdev_removed
= B_FALSE
;
1258 * Recheck the faulted flag now that we have confirmed that
1259 * the vdev is accessible. If we're faulted, bail.
1261 if (vd
->vdev_faulted
) {
1262 ASSERT(vd
->vdev_children
== 0);
1263 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1264 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1265 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1266 vd
->vdev_label_aux
);
1267 return (SET_ERROR(ENXIO
));
1270 if (vd
->vdev_degraded
) {
1271 ASSERT(vd
->vdev_children
== 0);
1272 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1273 VDEV_AUX_ERR_EXCEEDED
);
1275 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1279 * For hole or missing vdevs we just return success.
1281 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1284 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1285 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1286 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1292 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1293 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1295 if (vd
->vdev_children
== 0) {
1296 if (osize
< SPA_MINDEVSIZE
) {
1297 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1298 VDEV_AUX_TOO_SMALL
);
1299 return (SET_ERROR(EOVERFLOW
));
1302 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1303 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1304 VDEV_LABEL_END_SIZE
);
1306 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1307 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1308 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1309 VDEV_AUX_TOO_SMALL
);
1310 return (SET_ERROR(EOVERFLOW
));
1314 max_asize
= max_osize
;
1317 vd
->vdev_psize
= psize
;
1320 * Make sure the allocatable size hasn't shrunk.
1322 if (asize
< vd
->vdev_min_asize
) {
1323 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1324 VDEV_AUX_BAD_LABEL
);
1325 return (SET_ERROR(EINVAL
));
1328 if (vd
->vdev_asize
== 0) {
1330 * This is the first-ever open, so use the computed values.
1331 * For compatibility, a different ashift can be requested.
1333 vd
->vdev_asize
= asize
;
1334 vd
->vdev_max_asize
= max_asize
;
1335 if (vd
->vdev_ashift
== 0)
1336 vd
->vdev_ashift
= ashift
;
1339 * Detect if the alignment requirement has increased.
1340 * We don't want to make the pool unavailable, just
1341 * post an event instead.
1343 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1344 vd
->vdev_ops
->vdev_op_leaf
) {
1345 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1346 spa
, vd
, NULL
, 0, 0);
1349 vd
->vdev_max_asize
= max_asize
;
1353 * If all children are healthy and the asize has increased,
1354 * then we've experienced dynamic LUN growth. If automatic
1355 * expansion is enabled then use the additional space.
1357 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1358 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1359 vd
->vdev_asize
= asize
;
1361 vdev_set_min_asize(vd
);
1364 * Ensure we can issue some IO before declaring the
1365 * vdev open for business.
1367 if (vd
->vdev_ops
->vdev_op_leaf
&&
1368 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1369 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1370 VDEV_AUX_ERR_EXCEEDED
);
1375 * Track the min and max ashift values for normal data devices.
1377 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1378 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1379 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1380 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1381 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1382 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1386 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1387 * resilver. But don't do this if we are doing a reopen for a scrub,
1388 * since this would just restart the scrub we are already doing.
1390 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1391 vdev_resilver_needed(vd
, NULL
, NULL
))
1392 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1398 * Called once the vdevs are all opened, this routine validates the label
1399 * contents. This needs to be done before vdev_load() so that we don't
1400 * inadvertently do repair I/Os to the wrong device.
1402 * If 'strict' is false ignore the spa guid check. This is necessary because
1403 * if the machine crashed during a re-guid the new guid might have been written
1404 * to all of the vdev labels, but not the cached config. The strict check
1405 * will be performed when the pool is opened again using the mos config.
1407 * This function will only return failure if one of the vdevs indicates that it
1408 * has since been destroyed or exported. This is only possible if
1409 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1410 * will be updated but the function will return 0.
1413 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1415 spa_t
*spa
= vd
->vdev_spa
;
1417 uint64_t guid
= 0, top_guid
;
1421 for (c
= 0; c
< vd
->vdev_children
; c
++)
1422 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1423 return (SET_ERROR(EBADF
));
1426 * If the device has already failed, or was marked offline, don't do
1427 * any further validation. Otherwise, label I/O will fail and we will
1428 * overwrite the previous state.
1430 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1431 uint64_t aux_guid
= 0;
1433 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1434 spa_last_synced_txg(spa
) : -1ULL;
1436 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1437 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1438 VDEV_AUX_BAD_LABEL
);
1443 * Determine if this vdev has been split off into another
1444 * pool. If so, then refuse to open it.
1446 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1447 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1448 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1449 VDEV_AUX_SPLIT_POOL
);
1454 if (strict
&& (nvlist_lookup_uint64(label
,
1455 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1456 guid
!= spa_guid(spa
))) {
1457 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1458 VDEV_AUX_CORRUPT_DATA
);
1463 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1464 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1469 * If this vdev just became a top-level vdev because its
1470 * sibling was detached, it will have adopted the parent's
1471 * vdev guid -- but the label may or may not be on disk yet.
1472 * Fortunately, either version of the label will have the
1473 * same top guid, so if we're a top-level vdev, we can
1474 * safely compare to that instead.
1476 * If we split this vdev off instead, then we also check the
1477 * original pool's guid. We don't want to consider the vdev
1478 * corrupt if it is partway through a split operation.
1480 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1482 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1484 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1485 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1486 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1487 VDEV_AUX_CORRUPT_DATA
);
1492 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1494 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1495 VDEV_AUX_CORRUPT_DATA
);
1503 * If this is a verbatim import, no need to check the
1504 * state of the pool.
1506 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1507 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1508 state
!= POOL_STATE_ACTIVE
)
1509 return (SET_ERROR(EBADF
));
1512 * If we were able to open and validate a vdev that was
1513 * previously marked permanently unavailable, clear that state
1516 if (vd
->vdev_not_present
)
1517 vd
->vdev_not_present
= 0;
1524 * Close a virtual device.
1527 vdev_close(vdev_t
*vd
)
1529 vdev_t
*pvd
= vd
->vdev_parent
;
1530 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1532 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1535 * If our parent is reopening, then we are as well, unless we are
1538 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1539 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1541 vd
->vdev_ops
->vdev_op_close(vd
);
1543 vdev_cache_purge(vd
);
1546 * We record the previous state before we close it, so that if we are
1547 * doing a reopen(), we don't generate FMA ereports if we notice that
1548 * it's still faulted.
1550 vd
->vdev_prevstate
= vd
->vdev_state
;
1552 if (vd
->vdev_offline
)
1553 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1555 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1556 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1560 vdev_hold(vdev_t
*vd
)
1562 spa_t
*spa
= vd
->vdev_spa
;
1565 ASSERT(spa_is_root(spa
));
1566 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1569 for (c
= 0; c
< vd
->vdev_children
; c
++)
1570 vdev_hold(vd
->vdev_child
[c
]);
1572 if (vd
->vdev_ops
->vdev_op_leaf
)
1573 vd
->vdev_ops
->vdev_op_hold(vd
);
1577 vdev_rele(vdev_t
*vd
)
1581 ASSERT(spa_is_root(vd
->vdev_spa
));
1582 for (c
= 0; c
< vd
->vdev_children
; c
++)
1583 vdev_rele(vd
->vdev_child
[c
]);
1585 if (vd
->vdev_ops
->vdev_op_leaf
)
1586 vd
->vdev_ops
->vdev_op_rele(vd
);
1590 * Reopen all interior vdevs and any unopened leaves. We don't actually
1591 * reopen leaf vdevs which had previously been opened as they might deadlock
1592 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1593 * If the leaf has never been opened then open it, as usual.
1596 vdev_reopen(vdev_t
*vd
)
1598 spa_t
*spa
= vd
->vdev_spa
;
1600 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1602 /* set the reopening flag unless we're taking the vdev offline */
1603 vd
->vdev_reopening
= !vd
->vdev_offline
;
1605 (void) vdev_open(vd
);
1608 * Call vdev_validate() here to make sure we have the same device.
1609 * Otherwise, a device with an invalid label could be successfully
1610 * opened in response to vdev_reopen().
1613 (void) vdev_validate_aux(vd
);
1614 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1615 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1616 !l2arc_vdev_present(vd
))
1617 l2arc_add_vdev(spa
, vd
);
1619 (void) vdev_validate(vd
, B_TRUE
);
1623 * Reassess parent vdev's health.
1625 vdev_propagate_state(vd
);
1629 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1634 * Normally, partial opens (e.g. of a mirror) are allowed.
1635 * For a create, however, we want to fail the request if
1636 * there are any components we can't open.
1638 error
= vdev_open(vd
);
1640 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1642 return (error
? error
: ENXIO
);
1646 * Recursively load DTLs and initialize all labels.
1648 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1649 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1650 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1659 vdev_metaslab_set_size(vdev_t
*vd
)
1662 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1664 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1665 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1669 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1671 ASSERT(vd
== vd
->vdev_top
);
1672 ASSERT(!vd
->vdev_ishole
);
1673 ASSERT(ISP2(flags
));
1674 ASSERT(spa_writeable(vd
->vdev_spa
));
1676 if (flags
& VDD_METASLAB
)
1677 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1679 if (flags
& VDD_DTL
)
1680 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1682 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1686 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1690 for (c
= 0; c
< vd
->vdev_children
; c
++)
1691 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1693 if (vd
->vdev_ops
->vdev_op_leaf
)
1694 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1700 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1701 * the vdev has less than perfect replication. There are four kinds of DTL:
1703 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1705 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1707 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1708 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1709 * txgs that was scrubbed.
1711 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1712 * persistent errors or just some device being offline.
1713 * Unlike the other three, the DTL_OUTAGE map is not generally
1714 * maintained; it's only computed when needed, typically to
1715 * determine whether a device can be detached.
1717 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1718 * either has the data or it doesn't.
1720 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1721 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1722 * if any child is less than fully replicated, then so is its parent.
1723 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1724 * comprising only those txgs which appear in 'maxfaults' or more children;
1725 * those are the txgs we don't have enough replication to read. For example,
1726 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1727 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1728 * two child DTL_MISSING maps.
1730 * It should be clear from the above that to compute the DTLs and outage maps
1731 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1732 * Therefore, that is all we keep on disk. When loading the pool, or after
1733 * a configuration change, we generate all other DTLs from first principles.
1736 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1738 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1740 ASSERT(t
< DTL_TYPES
);
1741 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1742 ASSERT(spa_writeable(vd
->vdev_spa
));
1744 mutex_enter(rt
->rt_lock
);
1745 if (!range_tree_contains(rt
, txg
, size
))
1746 range_tree_add(rt
, txg
, size
);
1747 mutex_exit(rt
->rt_lock
);
1751 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1753 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1754 boolean_t dirty
= B_FALSE
;
1756 ASSERT(t
< DTL_TYPES
);
1757 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1759 mutex_enter(rt
->rt_lock
);
1760 if (range_tree_space(rt
) != 0)
1761 dirty
= range_tree_contains(rt
, txg
, size
);
1762 mutex_exit(rt
->rt_lock
);
1768 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1770 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1773 mutex_enter(rt
->rt_lock
);
1774 empty
= (range_tree_space(rt
) == 0);
1775 mutex_exit(rt
->rt_lock
);
1781 * Returns the lowest txg in the DTL range.
1784 vdev_dtl_min(vdev_t
*vd
)
1788 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1789 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1790 ASSERT0(vd
->vdev_children
);
1792 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1793 return (rs
->rs_start
- 1);
1797 * Returns the highest txg in the DTL.
1800 vdev_dtl_max(vdev_t
*vd
)
1804 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1805 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1806 ASSERT0(vd
->vdev_children
);
1808 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1809 return (rs
->rs_end
);
1813 * Determine if a resilvering vdev should remove any DTL entries from
1814 * its range. If the vdev was resilvering for the entire duration of the
1815 * scan then it should excise that range from its DTLs. Otherwise, this
1816 * vdev is considered partially resilvered and should leave its DTL
1817 * entries intact. The comment in vdev_dtl_reassess() describes how we
1821 vdev_dtl_should_excise(vdev_t
*vd
)
1823 spa_t
*spa
= vd
->vdev_spa
;
1824 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1826 ASSERT0(scn
->scn_phys
.scn_errors
);
1827 ASSERT0(vd
->vdev_children
);
1829 if (vd
->vdev_resilver_txg
== 0 ||
1830 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1834 * When a resilver is initiated the scan will assign the scn_max_txg
1835 * value to the highest txg value that exists in all DTLs. If this
1836 * device's max DTL is not part of this scan (i.e. it is not in
1837 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1840 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1841 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1842 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1843 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1850 * Reassess DTLs after a config change or scrub completion.
1853 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1855 spa_t
*spa
= vd
->vdev_spa
;
1859 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1861 for (c
= 0; c
< vd
->vdev_children
; c
++)
1862 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1863 scrub_txg
, scrub_done
);
1865 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1868 if (vd
->vdev_ops
->vdev_op_leaf
) {
1869 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1871 mutex_enter(&vd
->vdev_dtl_lock
);
1874 * If we've completed a scan cleanly then determine
1875 * if this vdev should remove any DTLs. We only want to
1876 * excise regions on vdevs that were available during
1877 * the entire duration of this scan.
1879 if (scrub_txg
!= 0 &&
1880 (spa
->spa_scrub_started
||
1881 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1882 vdev_dtl_should_excise(vd
)) {
1884 * We completed a scrub up to scrub_txg. If we
1885 * did it without rebooting, then the scrub dtl
1886 * will be valid, so excise the old region and
1887 * fold in the scrub dtl. Otherwise, leave the
1888 * dtl as-is if there was an error.
1890 * There's little trick here: to excise the beginning
1891 * of the DTL_MISSING map, we put it into a reference
1892 * tree and then add a segment with refcnt -1 that
1893 * covers the range [0, scrub_txg). This means
1894 * that each txg in that range has refcnt -1 or 0.
1895 * We then add DTL_SCRUB with a refcnt of 2, so that
1896 * entries in the range [0, scrub_txg) will have a
1897 * positive refcnt -- either 1 or 2. We then convert
1898 * the reference tree into the new DTL_MISSING map.
1900 space_reftree_create(&reftree
);
1901 space_reftree_add_map(&reftree
,
1902 vd
->vdev_dtl
[DTL_MISSING
], 1);
1903 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1904 space_reftree_add_map(&reftree
,
1905 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1906 space_reftree_generate_map(&reftree
,
1907 vd
->vdev_dtl
[DTL_MISSING
], 1);
1908 space_reftree_destroy(&reftree
);
1910 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1911 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1912 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1914 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1915 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1916 if (!vdev_readable(vd
))
1917 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1919 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1920 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1923 * If the vdev was resilvering and no longer has any
1924 * DTLs then reset its resilvering flag and dirty
1925 * the top level so that we persist the change.
1927 if (vd
->vdev_resilver_txg
!= 0 &&
1928 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1929 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
1930 vd
->vdev_resilver_txg
= 0;
1931 vdev_config_dirty(vd
->vdev_top
);
1934 mutex_exit(&vd
->vdev_dtl_lock
);
1937 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1941 mutex_enter(&vd
->vdev_dtl_lock
);
1942 for (t
= 0; t
< DTL_TYPES
; t
++) {
1945 /* account for child's outage in parent's missing map */
1946 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1948 continue; /* leaf vdevs only */
1949 if (t
== DTL_PARTIAL
)
1950 minref
= 1; /* i.e. non-zero */
1951 else if (vd
->vdev_nparity
!= 0)
1952 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1954 minref
= vd
->vdev_children
; /* any kind of mirror */
1955 space_reftree_create(&reftree
);
1956 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1957 vdev_t
*cvd
= vd
->vdev_child
[c
];
1958 mutex_enter(&cvd
->vdev_dtl_lock
);
1959 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1960 mutex_exit(&cvd
->vdev_dtl_lock
);
1962 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1963 space_reftree_destroy(&reftree
);
1965 mutex_exit(&vd
->vdev_dtl_lock
);
1969 vdev_dtl_load(vdev_t
*vd
)
1971 spa_t
*spa
= vd
->vdev_spa
;
1972 objset_t
*mos
= spa
->spa_meta_objset
;
1976 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
1977 ASSERT(!vd
->vdev_ishole
);
1979 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
1980 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
1983 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
1985 mutex_enter(&vd
->vdev_dtl_lock
);
1988 * Now that we've opened the space_map we need to update
1991 space_map_update(vd
->vdev_dtl_sm
);
1993 error
= space_map_load(vd
->vdev_dtl_sm
,
1994 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
1995 mutex_exit(&vd
->vdev_dtl_lock
);
2000 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2001 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2010 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2012 spa_t
*spa
= vd
->vdev_spa
;
2014 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2015 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2020 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2022 spa_t
*spa
= vd
->vdev_spa
;
2023 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2024 DMU_OT_NONE
, 0, tx
);
2027 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2034 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2038 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2039 vd
->vdev_ops
!= &vdev_missing_ops
&&
2040 vd
->vdev_ops
!= &vdev_root_ops
&&
2041 !vd
->vdev_top
->vdev_removing
) {
2042 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2043 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2045 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2046 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2049 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2050 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2055 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2057 spa_t
*spa
= vd
->vdev_spa
;
2058 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2059 objset_t
*mos
= spa
->spa_meta_objset
;
2060 range_tree_t
*rtsync
;
2063 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2065 ASSERT(!vd
->vdev_ishole
);
2066 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2068 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2070 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2071 mutex_enter(&vd
->vdev_dtl_lock
);
2072 space_map_free(vd
->vdev_dtl_sm
, tx
);
2073 space_map_close(vd
->vdev_dtl_sm
);
2074 vd
->vdev_dtl_sm
= NULL
;
2075 mutex_exit(&vd
->vdev_dtl_lock
);
2078 * We only destroy the leaf ZAP for detached leaves or for
2079 * removed log devices. Removed data devices handle leaf ZAP
2080 * cleanup later, once cancellation is no longer possible.
2082 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2083 vd
->vdev_top
->vdev_islog
)) {
2084 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2085 vd
->vdev_leaf_zap
= 0;
2092 if (vd
->vdev_dtl_sm
== NULL
) {
2093 uint64_t new_object
;
2095 new_object
= space_map_alloc(mos
, tx
);
2096 VERIFY3U(new_object
, !=, 0);
2098 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2099 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2100 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2103 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2105 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2107 mutex_enter(&rtlock
);
2109 mutex_enter(&vd
->vdev_dtl_lock
);
2110 range_tree_walk(rt
, range_tree_add
, rtsync
);
2111 mutex_exit(&vd
->vdev_dtl_lock
);
2113 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2114 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2115 range_tree_vacate(rtsync
, NULL
, NULL
);
2117 range_tree_destroy(rtsync
);
2119 mutex_exit(&rtlock
);
2120 mutex_destroy(&rtlock
);
2123 * If the object for the space map has changed then dirty
2124 * the top level so that we update the config.
2126 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2127 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2128 "new object %llu", txg
, spa_name(spa
), object
,
2129 space_map_object(vd
->vdev_dtl_sm
));
2130 vdev_config_dirty(vd
->vdev_top
);
2135 mutex_enter(&vd
->vdev_dtl_lock
);
2136 space_map_update(vd
->vdev_dtl_sm
);
2137 mutex_exit(&vd
->vdev_dtl_lock
);
2141 * Determine whether the specified vdev can be offlined/detached/removed
2142 * without losing data.
2145 vdev_dtl_required(vdev_t
*vd
)
2147 spa_t
*spa
= vd
->vdev_spa
;
2148 vdev_t
*tvd
= vd
->vdev_top
;
2149 uint8_t cant_read
= vd
->vdev_cant_read
;
2152 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2154 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2158 * Temporarily mark the device as unreadable, and then determine
2159 * whether this results in any DTL outages in the top-level vdev.
2160 * If not, we can safely offline/detach/remove the device.
2162 vd
->vdev_cant_read
= B_TRUE
;
2163 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2164 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2165 vd
->vdev_cant_read
= cant_read
;
2166 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2168 if (!required
&& zio_injection_enabled
)
2169 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2175 * Determine if resilver is needed, and if so the txg range.
2178 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2180 boolean_t needed
= B_FALSE
;
2181 uint64_t thismin
= UINT64_MAX
;
2182 uint64_t thismax
= 0;
2185 if (vd
->vdev_children
== 0) {
2186 mutex_enter(&vd
->vdev_dtl_lock
);
2187 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2188 vdev_writeable(vd
)) {
2190 thismin
= vdev_dtl_min(vd
);
2191 thismax
= vdev_dtl_max(vd
);
2194 mutex_exit(&vd
->vdev_dtl_lock
);
2196 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2197 vdev_t
*cvd
= vd
->vdev_child
[c
];
2198 uint64_t cmin
, cmax
;
2200 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2201 thismin
= MIN(thismin
, cmin
);
2202 thismax
= MAX(thismax
, cmax
);
2208 if (needed
&& minp
) {
2216 vdev_load(vdev_t
*vd
)
2221 * Recursively load all children.
2223 for (c
= 0; c
< vd
->vdev_children
; c
++)
2224 vdev_load(vd
->vdev_child
[c
]);
2227 * If this is a top-level vdev, initialize its metaslabs.
2229 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2230 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2231 vdev_metaslab_init(vd
, 0) != 0))
2232 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2233 VDEV_AUX_CORRUPT_DATA
);
2235 * If this is a leaf vdev, load its DTL.
2237 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2238 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2239 VDEV_AUX_CORRUPT_DATA
);
2243 * The special vdev case is used for hot spares and l2cache devices. Its
2244 * sole purpose it to set the vdev state for the associated vdev. To do this,
2245 * we make sure that we can open the underlying device, then try to read the
2246 * label, and make sure that the label is sane and that it hasn't been
2247 * repurposed to another pool.
2250 vdev_validate_aux(vdev_t
*vd
)
2253 uint64_t guid
, version
;
2256 if (!vdev_readable(vd
))
2259 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2260 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2261 VDEV_AUX_CORRUPT_DATA
);
2265 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2266 !SPA_VERSION_IS_SUPPORTED(version
) ||
2267 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2268 guid
!= vd
->vdev_guid
||
2269 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2270 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2271 VDEV_AUX_CORRUPT_DATA
);
2277 * We don't actually check the pool state here. If it's in fact in
2278 * use by another pool, we update this fact on the fly when requested.
2285 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2287 spa_t
*spa
= vd
->vdev_spa
;
2288 objset_t
*mos
= spa
->spa_meta_objset
;
2292 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2293 ASSERT(vd
== vd
->vdev_top
);
2294 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2296 if (vd
->vdev_ms
!= NULL
) {
2297 metaslab_group_t
*mg
= vd
->vdev_mg
;
2299 metaslab_group_histogram_verify(mg
);
2300 metaslab_class_histogram_verify(mg
->mg_class
);
2302 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2303 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2305 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2308 mutex_enter(&msp
->ms_lock
);
2310 * If the metaslab was not loaded when the vdev
2311 * was removed then the histogram accounting may
2312 * not be accurate. Update the histogram information
2313 * here so that we ensure that the metaslab group
2314 * and metaslab class are up-to-date.
2316 metaslab_group_histogram_remove(mg
, msp
);
2318 VERIFY0(space_map_allocated(msp
->ms_sm
));
2319 space_map_free(msp
->ms_sm
, tx
);
2320 space_map_close(msp
->ms_sm
);
2322 mutex_exit(&msp
->ms_lock
);
2325 metaslab_group_histogram_verify(mg
);
2326 metaslab_class_histogram_verify(mg
->mg_class
);
2327 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2328 ASSERT0(mg
->mg_histogram
[i
]);
2332 if (vd
->vdev_ms_array
) {
2333 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2334 vd
->vdev_ms_array
= 0;
2337 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2338 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2339 vd
->vdev_top_zap
= 0;
2345 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2348 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2350 ASSERT(!vd
->vdev_ishole
);
2352 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2353 metaslab_sync_done(msp
, txg
);
2356 metaslab_sync_reassess(vd
->vdev_mg
);
2360 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2362 spa_t
*spa
= vd
->vdev_spa
;
2367 ASSERT(!vd
->vdev_ishole
);
2369 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2370 ASSERT(vd
== vd
->vdev_top
);
2371 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2372 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2373 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2374 ASSERT(vd
->vdev_ms_array
!= 0);
2375 vdev_config_dirty(vd
);
2380 * Remove the metadata associated with this vdev once it's empty.
2382 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2383 vdev_remove(vd
, txg
);
2385 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2386 metaslab_sync(msp
, txg
);
2387 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2390 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2391 vdev_dtl_sync(lvd
, txg
);
2393 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2397 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2399 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2403 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2404 * not be opened, and no I/O is attempted.
2407 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2411 spa_vdev_state_enter(spa
, SCL_NONE
);
2413 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2414 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2416 if (!vd
->vdev_ops
->vdev_op_leaf
)
2417 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2422 * We don't directly use the aux state here, but if we do a
2423 * vdev_reopen(), we need this value to be present to remember why we
2426 vd
->vdev_label_aux
= aux
;
2429 * Faulted state takes precedence over degraded.
2431 vd
->vdev_delayed_close
= B_FALSE
;
2432 vd
->vdev_faulted
= 1ULL;
2433 vd
->vdev_degraded
= 0ULL;
2434 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2437 * If this device has the only valid copy of the data, then
2438 * back off and simply mark the vdev as degraded instead.
2440 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2441 vd
->vdev_degraded
= 1ULL;
2442 vd
->vdev_faulted
= 0ULL;
2445 * If we reopen the device and it's not dead, only then do we
2450 if (vdev_readable(vd
))
2451 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2454 return (spa_vdev_state_exit(spa
, vd
, 0));
2458 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2459 * user that something is wrong. The vdev continues to operate as normal as far
2460 * as I/O is concerned.
2463 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2467 spa_vdev_state_enter(spa
, SCL_NONE
);
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
));
2476 * If the vdev is already faulted, then don't do anything.
2478 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2479 return (spa_vdev_state_exit(spa
, NULL
, 0));
2481 vd
->vdev_degraded
= 1ULL;
2482 if (!vdev_is_dead(vd
))
2483 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2486 return (spa_vdev_state_exit(spa
, vd
, 0));
2490 * Online the given vdev.
2492 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2493 * spare device should be detached when the device finishes resilvering.
2494 * Second, the online should be treated like a 'test' online case, so no FMA
2495 * events are generated if the device fails to open.
2498 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2500 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2501 boolean_t postevent
= B_FALSE
;
2503 spa_vdev_state_enter(spa
, SCL_NONE
);
2505 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2506 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2508 if (!vd
->vdev_ops
->vdev_op_leaf
)
2509 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2512 (vd
->vdev_offline
== B_TRUE
|| vd
->vdev_tmpoffline
== B_TRUE
) ?
2516 vd
->vdev_offline
= B_FALSE
;
2517 vd
->vdev_tmpoffline
= B_FALSE
;
2518 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2519 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2521 /* XXX - L2ARC 1.0 does not support expansion */
2522 if (!vd
->vdev_aux
) {
2523 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2524 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2528 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2530 if (!vd
->vdev_aux
) {
2531 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2532 pvd
->vdev_expanding
= B_FALSE
;
2536 *newstate
= vd
->vdev_state
;
2537 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2538 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2539 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2540 vd
->vdev_parent
->vdev_child
[0] == vd
)
2541 vd
->vdev_unspare
= B_TRUE
;
2543 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2545 /* XXX - L2ARC 1.0 does not support expansion */
2547 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2548 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2552 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2554 return (spa_vdev_state_exit(spa
, vd
, 0));
2558 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2562 uint64_t generation
;
2563 metaslab_group_t
*mg
;
2566 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2568 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2569 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2571 if (!vd
->vdev_ops
->vdev_op_leaf
)
2572 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2576 generation
= spa
->spa_config_generation
+ 1;
2579 * If the device isn't already offline, try to offline it.
2581 if (!vd
->vdev_offline
) {
2583 * If this device has the only valid copy of some data,
2584 * don't allow it to be offlined. Log devices are always
2587 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2588 vdev_dtl_required(vd
))
2589 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2592 * If the top-level is a slog and it has had allocations
2593 * then proceed. We check that the vdev's metaslab group
2594 * is not NULL since it's possible that we may have just
2595 * added this vdev but not yet initialized its metaslabs.
2597 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2599 * Prevent any future allocations.
2601 metaslab_group_passivate(mg
);
2602 (void) spa_vdev_state_exit(spa
, vd
, 0);
2604 error
= spa_offline_log(spa
);
2606 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2609 * Check to see if the config has changed.
2611 if (error
|| generation
!= spa
->spa_config_generation
) {
2612 metaslab_group_activate(mg
);
2614 return (spa_vdev_state_exit(spa
,
2616 (void) spa_vdev_state_exit(spa
, vd
, 0);
2619 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2623 * Offline this device and reopen its top-level vdev.
2624 * If the top-level vdev is a log device then just offline
2625 * it. Otherwise, if this action results in the top-level
2626 * vdev becoming unusable, undo it and fail the request.
2628 vd
->vdev_offline
= B_TRUE
;
2631 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2632 vdev_is_dead(tvd
)) {
2633 vd
->vdev_offline
= B_FALSE
;
2635 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2639 * Add the device back into the metaslab rotor so that
2640 * once we online the device it's open for business.
2642 if (tvd
->vdev_islog
&& mg
!= NULL
)
2643 metaslab_group_activate(mg
);
2646 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2648 return (spa_vdev_state_exit(spa
, vd
, 0));
2652 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2656 mutex_enter(&spa
->spa_vdev_top_lock
);
2657 error
= vdev_offline_locked(spa
, guid
, flags
);
2658 mutex_exit(&spa
->spa_vdev_top_lock
);
2664 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2665 * vdev_offline(), we assume the spa config is locked. We also clear all
2666 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2669 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2671 vdev_t
*rvd
= spa
->spa_root_vdev
;
2674 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2679 vd
->vdev_stat
.vs_read_errors
= 0;
2680 vd
->vdev_stat
.vs_write_errors
= 0;
2681 vd
->vdev_stat
.vs_checksum_errors
= 0;
2683 for (c
= 0; c
< vd
->vdev_children
; c
++)
2684 vdev_clear(spa
, vd
->vdev_child
[c
]);
2687 * If we're in the FAULTED state or have experienced failed I/O, then
2688 * clear the persistent state and attempt to reopen the device. We
2689 * also mark the vdev config dirty, so that the new faulted state is
2690 * written out to disk.
2692 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2693 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2696 * When reopening in reponse to a clear event, it may be due to
2697 * a fmadm repair request. In this case, if the device is
2698 * still broken, we want to still post the ereport again.
2700 vd
->vdev_forcefault
= B_TRUE
;
2702 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2703 vd
->vdev_cant_read
= B_FALSE
;
2704 vd
->vdev_cant_write
= B_FALSE
;
2706 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2708 vd
->vdev_forcefault
= B_FALSE
;
2710 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2711 vdev_state_dirty(vd
->vdev_top
);
2713 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2714 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2716 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2720 * When clearing a FMA-diagnosed fault, we always want to
2721 * unspare the device, as we assume that the original spare was
2722 * done in response to the FMA fault.
2724 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2725 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2726 vd
->vdev_parent
->vdev_child
[0] == vd
)
2727 vd
->vdev_unspare
= B_TRUE
;
2731 vdev_is_dead(vdev_t
*vd
)
2734 * Holes and missing devices are always considered "dead".
2735 * This simplifies the code since we don't have to check for
2736 * these types of devices in the various code paths.
2737 * Instead we rely on the fact that we skip over dead devices
2738 * before issuing I/O to them.
2740 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2741 vd
->vdev_ops
== &vdev_missing_ops
);
2745 vdev_readable(vdev_t
*vd
)
2747 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2751 vdev_writeable(vdev_t
*vd
)
2753 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2757 vdev_allocatable(vdev_t
*vd
)
2759 uint64_t state
= vd
->vdev_state
;
2762 * We currently allow allocations from vdevs which may be in the
2763 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2764 * fails to reopen then we'll catch it later when we're holding
2765 * the proper locks. Note that we have to get the vdev state
2766 * in a local variable because although it changes atomically,
2767 * we're asking two separate questions about it.
2769 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2770 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
&&
2771 vd
->vdev_mg
->mg_initialized
);
2775 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2777 ASSERT(zio
->io_vd
== vd
);
2779 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2782 if (zio
->io_type
== ZIO_TYPE_READ
)
2783 return (!vd
->vdev_cant_read
);
2785 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2786 return (!vd
->vdev_cant_write
);
2792 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2795 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2796 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2797 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2800 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2804 * Get extended stats
2807 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2810 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2811 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2812 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2814 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2815 vsx
->vsx_total_histo
[t
][b
] +=
2816 cvsx
->vsx_total_histo
[t
][b
];
2820 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2821 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2822 vsx
->vsx_queue_histo
[t
][b
] +=
2823 cvsx
->vsx_queue_histo
[t
][b
];
2825 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2826 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2828 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2829 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2831 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2832 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2838 * Get statistics for the given vdev.
2841 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2845 * If we're getting stats on the root vdev, aggregate the I/O counts
2846 * over all top-level vdevs (i.e. the direct children of the root).
2848 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2850 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2851 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2854 memset(vsx
, 0, sizeof (*vsx
));
2856 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2857 vdev_t
*cvd
= vd
->vdev_child
[c
];
2858 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2859 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2861 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2863 vdev_get_child_stat(cvd
, vs
, cvs
);
2865 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2870 * We're a leaf. Just copy our ZIO active queue stats in. The
2871 * other leaf stats are updated in vdev_stat_update().
2876 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2878 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2879 vsx
->vsx_active_queue
[t
] =
2880 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2881 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2882 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2888 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2890 mutex_enter(&vd
->vdev_stat_lock
);
2892 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2893 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2894 vs
->vs_state
= vd
->vdev_state
;
2895 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2896 if (vd
->vdev_ops
->vdev_op_leaf
)
2897 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
2898 VDEV_LABEL_END_SIZE
;
2899 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2900 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
2902 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2906 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
2907 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
2908 mutex_exit(&vd
->vdev_stat_lock
);
2912 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2914 return (vdev_get_stats_ex(vd
, vs
, NULL
));
2918 vdev_clear_stats(vdev_t
*vd
)
2920 mutex_enter(&vd
->vdev_stat_lock
);
2921 vd
->vdev_stat
.vs_space
= 0;
2922 vd
->vdev_stat
.vs_dspace
= 0;
2923 vd
->vdev_stat
.vs_alloc
= 0;
2924 mutex_exit(&vd
->vdev_stat_lock
);
2928 vdev_scan_stat_init(vdev_t
*vd
)
2930 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2933 for (c
= 0; c
< vd
->vdev_children
; c
++)
2934 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2936 mutex_enter(&vd
->vdev_stat_lock
);
2937 vs
->vs_scan_processed
= 0;
2938 mutex_exit(&vd
->vdev_stat_lock
);
2942 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2944 spa_t
*spa
= zio
->io_spa
;
2945 vdev_t
*rvd
= spa
->spa_root_vdev
;
2946 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2948 uint64_t txg
= zio
->io_txg
;
2949 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2950 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
2951 zio_type_t type
= zio
->io_type
;
2952 int flags
= zio
->io_flags
;
2955 * If this i/o is a gang leader, it didn't do any actual work.
2957 if (zio
->io_gang_tree
)
2960 if (zio
->io_error
== 0) {
2962 * If this is a root i/o, don't count it -- we've already
2963 * counted the top-level vdevs, and vdev_get_stats() will
2964 * aggregate them when asked. This reduces contention on
2965 * the root vdev_stat_lock and implicitly handles blocks
2966 * that compress away to holes, for which there is no i/o.
2967 * (Holes never create vdev children, so all the counters
2968 * remain zero, which is what we want.)
2970 * Note: this only applies to successful i/o (io_error == 0)
2971 * because unlike i/o counts, errors are not additive.
2972 * When reading a ditto block, for example, failure of
2973 * one top-level vdev does not imply a root-level error.
2978 ASSERT(vd
== zio
->io_vd
);
2980 if (flags
& ZIO_FLAG_IO_BYPASS
)
2983 mutex_enter(&vd
->vdev_stat_lock
);
2985 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2986 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2987 dsl_scan_phys_t
*scn_phys
=
2988 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2989 uint64_t *processed
= &scn_phys
->scn_processed
;
2992 if (vd
->vdev_ops
->vdev_op_leaf
)
2993 atomic_add_64(processed
, psize
);
2994 vs
->vs_scan_processed
+= psize
;
2997 if (flags
& ZIO_FLAG_SELF_HEAL
)
2998 vs
->vs_self_healed
+= psize
;
3002 * The bytes/ops/histograms are recorded at the leaf level and
3003 * aggregated into the higher level vdevs in vdev_get_stats().
3005 if (vd
->vdev_ops
->vdev_op_leaf
&&
3006 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3009 vs
->vs_bytes
[type
] += psize
;
3011 if (flags
& ZIO_FLAG_DELEGATED
) {
3012 vsx
->vsx_agg_histo
[zio
->io_priority
]
3013 [RQ_HISTO(zio
->io_size
)]++;
3015 vsx
->vsx_ind_histo
[zio
->io_priority
]
3016 [RQ_HISTO(zio
->io_size
)]++;
3019 if (zio
->io_delta
&& zio
->io_delay
) {
3020 vsx
->vsx_queue_histo
[zio
->io_priority
]
3021 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3022 vsx
->vsx_disk_histo
[type
]
3023 [L_HISTO(zio
->io_delay
)]++;
3024 vsx
->vsx_total_histo
[type
]
3025 [L_HISTO(zio
->io_delta
)]++;
3029 mutex_exit(&vd
->vdev_stat_lock
);
3033 if (flags
& ZIO_FLAG_SPECULATIVE
)
3037 * If this is an I/O error that is going to be retried, then ignore the
3038 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3039 * hard errors, when in reality they can happen for any number of
3040 * innocuous reasons (bus resets, MPxIO link failure, etc).
3042 if (zio
->io_error
== EIO
&&
3043 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3047 * Intent logs writes won't propagate their error to the root
3048 * I/O so don't mark these types of failures as pool-level
3051 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3054 mutex_enter(&vd
->vdev_stat_lock
);
3055 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3056 if (zio
->io_error
== ECKSUM
)
3057 vs
->vs_checksum_errors
++;
3059 vs
->vs_read_errors
++;
3061 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3062 vs
->vs_write_errors
++;
3063 mutex_exit(&vd
->vdev_stat_lock
);
3065 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3066 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3067 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3068 spa
->spa_claiming
)) {
3070 * This is either a normal write (not a repair), or it's
3071 * a repair induced by the scrub thread, or it's a repair
3072 * made by zil_claim() during spa_load() in the first txg.
3073 * In the normal case, we commit the DTL change in the same
3074 * txg as the block was born. In the scrub-induced repair
3075 * case, we know that scrubs run in first-pass syncing context,
3076 * so we commit the DTL change in spa_syncing_txg(spa).
3077 * In the zil_claim() case, we commit in spa_first_txg(spa).
3079 * We currently do not make DTL entries for failed spontaneous
3080 * self-healing writes triggered by normal (non-scrubbing)
3081 * reads, because we have no transactional context in which to
3082 * do so -- and it's not clear that it'd be desirable anyway.
3084 if (vd
->vdev_ops
->vdev_op_leaf
) {
3085 uint64_t commit_txg
= txg
;
3086 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3087 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3088 ASSERT(spa_sync_pass(spa
) == 1);
3089 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3090 commit_txg
= spa_syncing_txg(spa
);
3091 } else if (spa
->spa_claiming
) {
3092 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3093 commit_txg
= spa_first_txg(spa
);
3095 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3096 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3098 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3099 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3100 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3103 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3108 * Update the in-core space usage stats for this vdev, its metaslab class,
3109 * and the root vdev.
3112 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3113 int64_t space_delta
)
3115 int64_t dspace_delta
= space_delta
;
3116 spa_t
*spa
= vd
->vdev_spa
;
3117 vdev_t
*rvd
= spa
->spa_root_vdev
;
3118 metaslab_group_t
*mg
= vd
->vdev_mg
;
3119 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3121 ASSERT(vd
== vd
->vdev_top
);
3124 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3125 * factor. We must calculate this here and not at the root vdev
3126 * because the root vdev's psize-to-asize is simply the max of its
3127 * childrens', thus not accurate enough for us.
3129 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3130 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3131 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3132 vd
->vdev_deflate_ratio
;
3134 mutex_enter(&vd
->vdev_stat_lock
);
3135 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3136 vd
->vdev_stat
.vs_space
+= space_delta
;
3137 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3138 mutex_exit(&vd
->vdev_stat_lock
);
3140 if (mc
== spa_normal_class(spa
)) {
3141 mutex_enter(&rvd
->vdev_stat_lock
);
3142 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3143 rvd
->vdev_stat
.vs_space
+= space_delta
;
3144 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3145 mutex_exit(&rvd
->vdev_stat_lock
);
3149 ASSERT(rvd
== vd
->vdev_parent
);
3150 ASSERT(vd
->vdev_ms_count
!= 0);
3152 metaslab_class_space_update(mc
,
3153 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3158 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3159 * so that it will be written out next time the vdev configuration is synced.
3160 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3163 vdev_config_dirty(vdev_t
*vd
)
3165 spa_t
*spa
= vd
->vdev_spa
;
3166 vdev_t
*rvd
= spa
->spa_root_vdev
;
3169 ASSERT(spa_writeable(spa
));
3172 * If this is an aux vdev (as with l2cache and spare devices), then we
3173 * update the vdev config manually and set the sync flag.
3175 if (vd
->vdev_aux
!= NULL
) {
3176 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3180 for (c
= 0; c
< sav
->sav_count
; c
++) {
3181 if (sav
->sav_vdevs
[c
] == vd
)
3185 if (c
== sav
->sav_count
) {
3187 * We're being removed. There's nothing more to do.
3189 ASSERT(sav
->sav_sync
== B_TRUE
);
3193 sav
->sav_sync
= B_TRUE
;
3195 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3196 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3197 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3198 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3204 * Setting the nvlist in the middle if the array is a little
3205 * sketchy, but it will work.
3207 nvlist_free(aux
[c
]);
3208 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3214 * The dirty list is protected by the SCL_CONFIG lock. The caller
3215 * must either hold SCL_CONFIG as writer, or must be the sync thread
3216 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3217 * so this is sufficient to ensure mutual exclusion.
3219 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3220 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3221 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3224 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3225 vdev_config_dirty(rvd
->vdev_child
[c
]);
3227 ASSERT(vd
== vd
->vdev_top
);
3229 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3231 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3236 vdev_config_clean(vdev_t
*vd
)
3238 spa_t
*spa
= vd
->vdev_spa
;
3240 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3241 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3242 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3244 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3245 list_remove(&spa
->spa_config_dirty_list
, vd
);
3249 * Mark a top-level vdev's state as dirty, so that the next pass of
3250 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3251 * the state changes from larger config changes because they require
3252 * much less locking, and are often needed for administrative actions.
3255 vdev_state_dirty(vdev_t
*vd
)
3257 spa_t
*spa
= vd
->vdev_spa
;
3259 ASSERT(spa_writeable(spa
));
3260 ASSERT(vd
== vd
->vdev_top
);
3263 * The state list is protected by the SCL_STATE lock. The caller
3264 * must either hold SCL_STATE as writer, or must be the sync thread
3265 * (which holds SCL_STATE as reader). There's only one sync thread,
3266 * so this is sufficient to ensure mutual exclusion.
3268 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3269 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3270 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3272 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3273 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3277 vdev_state_clean(vdev_t
*vd
)
3279 spa_t
*spa
= vd
->vdev_spa
;
3281 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3282 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3283 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3285 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3286 list_remove(&spa
->spa_state_dirty_list
, vd
);
3290 * Propagate vdev state up from children to parent.
3293 vdev_propagate_state(vdev_t
*vd
)
3295 spa_t
*spa
= vd
->vdev_spa
;
3296 vdev_t
*rvd
= spa
->spa_root_vdev
;
3297 int degraded
= 0, faulted
= 0;
3302 if (vd
->vdev_children
> 0) {
3303 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3304 child
= vd
->vdev_child
[c
];
3307 * Don't factor holes into the decision.
3309 if (child
->vdev_ishole
)
3312 if (!vdev_readable(child
) ||
3313 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3315 * Root special: if there is a top-level log
3316 * device, treat the root vdev as if it were
3319 if (child
->vdev_islog
&& vd
== rvd
)
3323 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3327 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3331 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3334 * Root special: if there is a top-level vdev that cannot be
3335 * opened due to corrupted metadata, then propagate the root
3336 * vdev's aux state as 'corrupt' rather than 'insufficient
3339 if (corrupted
&& vd
== rvd
&&
3340 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3341 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3342 VDEV_AUX_CORRUPT_DATA
);
3345 if (vd
->vdev_parent
)
3346 vdev_propagate_state(vd
->vdev_parent
);
3350 * Set a vdev's state. If this is during an open, we don't update the parent
3351 * state, because we're in the process of opening children depth-first.
3352 * Otherwise, we propagate the change to the parent.
3354 * If this routine places a device in a faulted state, an appropriate ereport is
3358 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3360 uint64_t save_state
;
3361 spa_t
*spa
= vd
->vdev_spa
;
3363 if (state
== vd
->vdev_state
) {
3364 vd
->vdev_stat
.vs_aux
= aux
;
3368 save_state
= vd
->vdev_state
;
3370 vd
->vdev_state
= state
;
3371 vd
->vdev_stat
.vs_aux
= aux
;
3374 * If we are setting the vdev state to anything but an open state, then
3375 * always close the underlying device unless the device has requested
3376 * a delayed close (i.e. we're about to remove or fault the device).
3377 * Otherwise, we keep accessible but invalid devices open forever.
3378 * We don't call vdev_close() itself, because that implies some extra
3379 * checks (offline, etc) that we don't want here. This is limited to
3380 * leaf devices, because otherwise closing the device will affect other
3383 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3384 vd
->vdev_ops
->vdev_op_leaf
)
3385 vd
->vdev_ops
->vdev_op_close(vd
);
3387 if (vd
->vdev_removed
&&
3388 state
== VDEV_STATE_CANT_OPEN
&&
3389 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3391 * If the previous state is set to VDEV_STATE_REMOVED, then this
3392 * device was previously marked removed and someone attempted to
3393 * reopen it. If this failed due to a nonexistent device, then
3394 * keep the device in the REMOVED state. We also let this be if
3395 * it is one of our special test online cases, which is only
3396 * attempting to online the device and shouldn't generate an FMA
3399 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3400 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3401 } else if (state
== VDEV_STATE_REMOVED
) {
3402 vd
->vdev_removed
= B_TRUE
;
3403 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3405 * If we fail to open a vdev during an import or recovery, we
3406 * mark it as "not available", which signifies that it was
3407 * never there to begin with. Failure to open such a device
3408 * is not considered an error.
3410 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3411 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3412 vd
->vdev_ops
->vdev_op_leaf
)
3413 vd
->vdev_not_present
= 1;
3416 * Post the appropriate ereport. If the 'prevstate' field is
3417 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3418 * that this is part of a vdev_reopen(). In this case, we don't
3419 * want to post the ereport if the device was already in the
3420 * CANT_OPEN state beforehand.
3422 * If the 'checkremove' flag is set, then this is an attempt to
3423 * online the device in response to an insertion event. If we
3424 * hit this case, then we have detected an insertion event for a
3425 * faulted or offline device that wasn't in the removed state.
3426 * In this scenario, we don't post an ereport because we are
3427 * about to replace the device, or attempt an online with
3428 * vdev_forcefault, which will generate the fault for us.
3430 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3431 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3432 vd
!= spa
->spa_root_vdev
) {
3436 case VDEV_AUX_OPEN_FAILED
:
3437 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3439 case VDEV_AUX_CORRUPT_DATA
:
3440 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3442 case VDEV_AUX_NO_REPLICAS
:
3443 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3445 case VDEV_AUX_BAD_GUID_SUM
:
3446 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3448 case VDEV_AUX_TOO_SMALL
:
3449 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3451 case VDEV_AUX_BAD_LABEL
:
3452 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3455 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3458 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3461 /* Erase any notion of persistent removed state */
3462 vd
->vdev_removed
= B_FALSE
;
3464 vd
->vdev_removed
= B_FALSE
;
3468 * Notify ZED of any significant state-change on a leaf vdev.
3471 if (vd
->vdev_ops
->vdev_op_leaf
) {
3472 /* preserve original state from a vdev_reopen() */
3473 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
3474 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
3475 (save_state
<= VDEV_STATE_CLOSED
))
3476 save_state
= vd
->vdev_prevstate
;
3478 /* filter out state change due to initial vdev_open */
3479 if (save_state
> VDEV_STATE_CLOSED
)
3480 zfs_post_state_change(spa
, vd
, save_state
);
3483 if (!isopen
&& vd
->vdev_parent
)
3484 vdev_propagate_state(vd
->vdev_parent
);
3488 * Check the vdev configuration to ensure that it's capable of supporting
3492 vdev_is_bootable(vdev_t
*vd
)
3494 #if defined(__sun__) || defined(__sun)
3496 * Currently, we do not support RAID-Z or partial configuration.
3497 * In addition, only a single top-level vdev is allowed and none of the
3498 * leaves can be wholedisks.
3502 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3503 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3505 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3506 vd
->vdev_children
> 1) {
3508 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3509 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3512 } else if (vd
->vdev_wholedisk
== 1) {
3516 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3517 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3520 #endif /* __sun__ || __sun */
3525 * Load the state from the original vdev tree (ovd) which
3526 * we've retrieved from the MOS config object. If the original
3527 * vdev was offline or faulted then we transfer that state to the
3528 * device in the current vdev tree (nvd).
3531 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3535 ASSERT(nvd
->vdev_top
->vdev_islog
);
3536 ASSERT(spa_config_held(nvd
->vdev_spa
,
3537 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3538 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3540 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3541 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3543 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3545 * Restore the persistent vdev state
3547 nvd
->vdev_offline
= ovd
->vdev_offline
;
3548 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3549 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3550 nvd
->vdev_removed
= ovd
->vdev_removed
;
3555 * Determine if a log device has valid content. If the vdev was
3556 * removed or faulted in the MOS config then we know that
3557 * the content on the log device has already been written to the pool.
3560 vdev_log_state_valid(vdev_t
*vd
)
3564 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3568 for (c
= 0; c
< vd
->vdev_children
; c
++)
3569 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3576 * Expand a vdev if possible.
3579 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3581 ASSERT(vd
->vdev_top
== vd
);
3582 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3584 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3585 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3586 vdev_config_dirty(vd
);
3594 vdev_split(vdev_t
*vd
)
3596 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3598 vdev_remove_child(pvd
, vd
);
3599 vdev_compact_children(pvd
);
3601 cvd
= pvd
->vdev_child
[0];
3602 if (pvd
->vdev_children
== 1) {
3603 vdev_remove_parent(cvd
);
3604 cvd
->vdev_splitting
= B_TRUE
;
3606 vdev_propagate_state(cvd
);
3610 vdev_deadman(vdev_t
*vd
)
3614 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3615 vdev_t
*cvd
= vd
->vdev_child
[c
];
3620 if (vd
->vdev_ops
->vdev_op_leaf
) {
3621 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3623 mutex_enter(&vq
->vq_lock
);
3624 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3625 spa_t
*spa
= vd
->vdev_spa
;
3630 * Look at the head of all the pending queues,
3631 * if any I/O has been outstanding for longer than
3632 * the spa_deadman_synctime we log a zevent.
3634 fio
= avl_first(&vq
->vq_active_tree
);
3635 delta
= gethrtime() - fio
->io_timestamp
;
3636 if (delta
> spa_deadman_synctime(spa
)) {
3637 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3638 "delta %lluns, last io %lluns",
3639 fio
->io_timestamp
, delta
,
3640 vq
->vq_io_complete_ts
);
3641 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3642 spa
, vd
, fio
, 0, 0);
3645 mutex_exit(&vq
->vq_lock
);
3649 #if defined(_KERNEL) && defined(HAVE_SPL)
3650 EXPORT_SYMBOL(vdev_fault
);
3651 EXPORT_SYMBOL(vdev_degrade
);
3652 EXPORT_SYMBOL(vdev_online
);
3653 EXPORT_SYMBOL(vdev_offline
);
3654 EXPORT_SYMBOL(vdev_clear
);
3656 module_param(metaslabs_per_vdev
, int, 0644);
3657 MODULE_PARM_DESC(metaslabs_per_vdev
,
3658 "Divide added vdev into approximately (but no more than) this number "