4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/fm/fs/zfs.h>
31 #include <sys/spa_impl.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/uberblock_impl.h>
36 #include <sys/metaslab.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
39 #include <sys/space_reftree.h>
42 #include <sys/fs/zfs.h>
45 #include <sys/dsl_scan.h>
49 * When a vdev is added, it will be divided into approximately (but no
50 * more than) this number of metaslabs.
52 int metaslabs_per_vdev
= 200;
55 * Virtual device management.
58 static vdev_ops_t
*vdev_ops_table
[] = {
72 * Given a vdev type, return the appropriate ops vector.
75 vdev_getops(const char *type
)
77 vdev_ops_t
*ops
, **opspp
;
79 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
80 if (strcmp(ops
->vdev_op_type
, type
) == 0)
87 * Default asize function: return the MAX of psize with the asize of
88 * all children. This is what's used by anything other than RAID-Z.
91 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
93 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
97 for (c
= 0; c
< vd
->vdev_children
; c
++) {
98 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
99 asize
= MAX(asize
, csize
);
106 * Get the minimum allocatable size. We define the allocatable size as
107 * the vdev's asize rounded to the nearest metaslab. This allows us to
108 * replace or attach devices which don't have the same physical size but
109 * can still satisfy the same number of allocations.
112 vdev_get_min_asize(vdev_t
*vd
)
114 vdev_t
*pvd
= vd
->vdev_parent
;
117 * If our parent is NULL (inactive spare or cache) or is the root,
118 * just return our own asize.
121 return (vd
->vdev_asize
);
124 * The top-level vdev just returns the allocatable size rounded
125 * to the nearest metaslab.
127 if (vd
== vd
->vdev_top
)
128 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
131 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
132 * so each child must provide at least 1/Nth of its asize.
134 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
135 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
137 return (pvd
->vdev_min_asize
);
141 vdev_set_min_asize(vdev_t
*vd
)
144 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
146 for (c
= 0; c
< vd
->vdev_children
; c
++)
147 vdev_set_min_asize(vd
->vdev_child
[c
]);
151 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
153 vdev_t
*rvd
= spa
->spa_root_vdev
;
155 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
157 if (vdev
< rvd
->vdev_children
) {
158 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
159 return (rvd
->vdev_child
[vdev
]);
166 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
171 if (vd
->vdev_guid
== guid
)
174 for (c
= 0; c
< vd
->vdev_children
; c
++)
175 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
183 vdev_count_leaves_impl(vdev_t
*vd
)
188 if (vd
->vdev_ops
->vdev_op_leaf
)
191 for (c
= 0; c
< vd
->vdev_children
; c
++)
192 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
198 vdev_count_leaves(spa_t
*spa
)
200 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
204 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
206 size_t oldsize
, newsize
;
207 uint64_t id
= cvd
->vdev_id
;
210 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
211 ASSERT(cvd
->vdev_parent
== NULL
);
213 cvd
->vdev_parent
= pvd
;
218 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
220 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
221 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
222 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
224 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
225 if (pvd
->vdev_child
!= NULL
) {
226 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
227 kmem_free(pvd
->vdev_child
, oldsize
);
230 pvd
->vdev_child
= newchild
;
231 pvd
->vdev_child
[id
] = cvd
;
233 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
234 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
237 * Walk up all ancestors to update guid sum.
239 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
240 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
244 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
247 uint_t id
= cvd
->vdev_id
;
249 ASSERT(cvd
->vdev_parent
== pvd
);
254 ASSERT(id
< pvd
->vdev_children
);
255 ASSERT(pvd
->vdev_child
[id
] == cvd
);
257 pvd
->vdev_child
[id
] = NULL
;
258 cvd
->vdev_parent
= NULL
;
260 for (c
= 0; c
< pvd
->vdev_children
; c
++)
261 if (pvd
->vdev_child
[c
])
264 if (c
== pvd
->vdev_children
) {
265 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
266 pvd
->vdev_child
= NULL
;
267 pvd
->vdev_children
= 0;
271 * Walk up all ancestors to update guid sum.
273 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
274 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
278 * Remove any holes in the child array.
281 vdev_compact_children(vdev_t
*pvd
)
283 vdev_t
**newchild
, *cvd
;
284 int oldc
= pvd
->vdev_children
;
288 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
290 for (c
= newc
= 0; c
< oldc
; c
++)
291 if (pvd
->vdev_child
[c
])
294 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
296 for (c
= newc
= 0; c
< oldc
; c
++) {
297 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
298 newchild
[newc
] = cvd
;
299 cvd
->vdev_id
= newc
++;
303 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
304 pvd
->vdev_child
= newchild
;
305 pvd
->vdev_children
= newc
;
309 * Allocate and minimally initialize a vdev_t.
312 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
317 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
319 if (spa
->spa_root_vdev
== NULL
) {
320 ASSERT(ops
== &vdev_root_ops
);
321 spa
->spa_root_vdev
= vd
;
322 spa
->spa_load_guid
= spa_generate_guid(NULL
);
325 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
326 if (spa
->spa_root_vdev
== vd
) {
328 * The root vdev's guid will also be the pool guid,
329 * which must be unique among all pools.
331 guid
= spa_generate_guid(NULL
);
334 * Any other vdev's guid must be unique within the pool.
336 guid
= spa_generate_guid(spa
);
338 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
343 vd
->vdev_guid
= guid
;
344 vd
->vdev_guid_sum
= guid
;
346 vd
->vdev_state
= VDEV_STATE_CLOSED
;
347 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
349 list_link_init(&vd
->vdev_config_dirty_node
);
350 list_link_init(&vd
->vdev_state_dirty_node
);
351 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
352 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
353 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
354 for (t
= 0; t
< DTL_TYPES
; t
++) {
355 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
358 txg_list_create(&vd
->vdev_ms_list
,
359 offsetof(struct metaslab
, ms_txg_node
));
360 txg_list_create(&vd
->vdev_dtl_list
,
361 offsetof(struct vdev
, vdev_dtl_node
));
362 vd
->vdev_stat
.vs_timestamp
= gethrtime();
370 * Allocate a new vdev. The 'alloctype' is used to control whether we are
371 * creating a new vdev or loading an existing one - the behavior is slightly
372 * different for each case.
375 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
380 uint64_t guid
= 0, islog
, nparity
;
383 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
385 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
386 return (SET_ERROR(EINVAL
));
388 if ((ops
= vdev_getops(type
)) == NULL
)
389 return (SET_ERROR(EINVAL
));
392 * If this is a load, get the vdev guid from the nvlist.
393 * Otherwise, vdev_alloc_common() will generate one for us.
395 if (alloctype
== VDEV_ALLOC_LOAD
) {
398 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
400 return (SET_ERROR(EINVAL
));
402 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
403 return (SET_ERROR(EINVAL
));
404 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
405 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
406 return (SET_ERROR(EINVAL
));
407 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
408 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
409 return (SET_ERROR(EINVAL
));
410 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
411 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
412 return (SET_ERROR(EINVAL
));
416 * The first allocated vdev must be of type 'root'.
418 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
419 return (SET_ERROR(EINVAL
));
422 * Determine whether we're a log vdev.
425 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
426 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
427 return (SET_ERROR(ENOTSUP
));
429 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
430 return (SET_ERROR(ENOTSUP
));
433 * Set the nparity property for RAID-Z vdevs.
436 if (ops
== &vdev_raidz_ops
) {
437 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
439 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
440 return (SET_ERROR(EINVAL
));
442 * Previous versions could only support 1 or 2 parity
446 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
447 return (SET_ERROR(ENOTSUP
));
449 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
450 return (SET_ERROR(ENOTSUP
));
453 * We require the parity to be specified for SPAs that
454 * support multiple parity levels.
456 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
457 return (SET_ERROR(EINVAL
));
459 * Otherwise, we default to 1 parity device for RAID-Z.
466 ASSERT(nparity
!= -1ULL);
468 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
470 vd
->vdev_islog
= islog
;
471 vd
->vdev_nparity
= nparity
;
473 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
474 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
475 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
476 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
477 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
478 &vd
->vdev_physpath
) == 0)
479 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
480 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
481 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
484 * Set the whole_disk property. If it's not specified, leave the value
487 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
488 &vd
->vdev_wholedisk
) != 0)
489 vd
->vdev_wholedisk
= -1ULL;
492 * Look for the 'not present' flag. This will only be set if the device
493 * was not present at the time of import.
495 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
496 &vd
->vdev_not_present
);
499 * Get the alignment requirement.
501 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
504 * Retrieve the vdev creation time.
506 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
510 * If we're a top-level vdev, try to load the allocation parameters.
512 if (parent
&& !parent
->vdev_parent
&&
513 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
514 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
516 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
518 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
520 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
522 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
525 ASSERT0(vd
->vdev_top_zap
);
528 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
529 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
530 alloctype
== VDEV_ALLOC_ADD
||
531 alloctype
== VDEV_ALLOC_SPLIT
||
532 alloctype
== VDEV_ALLOC_ROOTPOOL
);
533 vd
->vdev_mg
= metaslab_group_create(islog
?
534 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
537 if (vd
->vdev_ops
->vdev_op_leaf
&&
538 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
539 (void) nvlist_lookup_uint64(nv
,
540 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
542 ASSERT0(vd
->vdev_leaf_zap
);
546 * If we're a leaf vdev, try to load the DTL object and other state.
549 if (vd
->vdev_ops
->vdev_op_leaf
&&
550 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
551 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
552 if (alloctype
== VDEV_ALLOC_LOAD
) {
553 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
554 &vd
->vdev_dtl_object
);
555 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
559 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
562 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
563 &spare
) == 0 && spare
)
567 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
570 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
571 &vd
->vdev_resilver_txg
);
574 * When importing a pool, we want to ignore the persistent fault
575 * state, as the diagnosis made on another system may not be
576 * valid in the current context. Local vdevs will
577 * remain in the faulted state.
579 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
580 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
582 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
584 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
587 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
591 VDEV_AUX_ERR_EXCEEDED
;
592 if (nvlist_lookup_string(nv
,
593 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
594 strcmp(aux
, "external") == 0)
595 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
601 * Add ourselves to the parent's list of children.
603 vdev_add_child(parent
, vd
);
611 vdev_free(vdev_t
*vd
)
614 spa_t
*spa
= vd
->vdev_spa
;
617 * vdev_free() implies closing the vdev first. This is simpler than
618 * trying to ensure complicated semantics for all callers.
622 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
623 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
628 for (c
= 0; c
< vd
->vdev_children
; c
++)
629 vdev_free(vd
->vdev_child
[c
]);
631 ASSERT(vd
->vdev_child
== NULL
);
632 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
635 * Discard allocation state.
637 if (vd
->vdev_mg
!= NULL
) {
638 vdev_metaslab_fini(vd
);
639 metaslab_group_destroy(vd
->vdev_mg
);
642 ASSERT0(vd
->vdev_stat
.vs_space
);
643 ASSERT0(vd
->vdev_stat
.vs_dspace
);
644 ASSERT0(vd
->vdev_stat
.vs_alloc
);
647 * Remove this vdev from its parent's child list.
649 vdev_remove_child(vd
->vdev_parent
, vd
);
651 ASSERT(vd
->vdev_parent
== NULL
);
654 * Clean up vdev structure.
660 spa_strfree(vd
->vdev_path
);
662 spa_strfree(vd
->vdev_devid
);
663 if (vd
->vdev_physpath
)
664 spa_strfree(vd
->vdev_physpath
);
666 spa_strfree(vd
->vdev_fru
);
668 if (vd
->vdev_isspare
)
669 spa_spare_remove(vd
);
670 if (vd
->vdev_isl2cache
)
671 spa_l2cache_remove(vd
);
673 txg_list_destroy(&vd
->vdev_ms_list
);
674 txg_list_destroy(&vd
->vdev_dtl_list
);
676 mutex_enter(&vd
->vdev_dtl_lock
);
677 space_map_close(vd
->vdev_dtl_sm
);
678 for (t
= 0; t
< DTL_TYPES
; t
++) {
679 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
680 range_tree_destroy(vd
->vdev_dtl
[t
]);
682 mutex_exit(&vd
->vdev_dtl_lock
);
684 mutex_destroy(&vd
->vdev_dtl_lock
);
685 mutex_destroy(&vd
->vdev_stat_lock
);
686 mutex_destroy(&vd
->vdev_probe_lock
);
688 if (vd
== spa
->spa_root_vdev
)
689 spa
->spa_root_vdev
= NULL
;
691 kmem_free(vd
, sizeof (vdev_t
));
695 * Transfer top-level vdev state from svd to tvd.
698 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
700 spa_t
*spa
= svd
->vdev_spa
;
705 ASSERT(tvd
== tvd
->vdev_top
);
707 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
708 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
709 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
710 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
711 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
713 svd
->vdev_ms_array
= 0;
714 svd
->vdev_ms_shift
= 0;
715 svd
->vdev_ms_count
= 0;
716 svd
->vdev_top_zap
= 0;
719 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
720 tvd
->vdev_mg
= svd
->vdev_mg
;
721 tvd
->vdev_ms
= svd
->vdev_ms
;
726 if (tvd
->vdev_mg
!= NULL
)
727 tvd
->vdev_mg
->mg_vd
= tvd
;
729 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
730 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
731 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
733 svd
->vdev_stat
.vs_alloc
= 0;
734 svd
->vdev_stat
.vs_space
= 0;
735 svd
->vdev_stat
.vs_dspace
= 0;
737 for (t
= 0; t
< TXG_SIZE
; t
++) {
738 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
739 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
740 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
741 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
742 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
743 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
746 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
747 vdev_config_clean(svd
);
748 vdev_config_dirty(tvd
);
751 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
752 vdev_state_clean(svd
);
753 vdev_state_dirty(tvd
);
756 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
757 svd
->vdev_deflate_ratio
= 0;
759 tvd
->vdev_islog
= svd
->vdev_islog
;
764 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
773 for (c
= 0; c
< vd
->vdev_children
; c
++)
774 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
778 * Add a mirror/replacing vdev above an existing vdev.
781 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
783 spa_t
*spa
= cvd
->vdev_spa
;
784 vdev_t
*pvd
= cvd
->vdev_parent
;
787 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
789 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
791 mvd
->vdev_asize
= cvd
->vdev_asize
;
792 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
793 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
794 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
795 mvd
->vdev_state
= cvd
->vdev_state
;
796 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
798 vdev_remove_child(pvd
, cvd
);
799 vdev_add_child(pvd
, mvd
);
800 cvd
->vdev_id
= mvd
->vdev_children
;
801 vdev_add_child(mvd
, cvd
);
802 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
804 if (mvd
== mvd
->vdev_top
)
805 vdev_top_transfer(cvd
, mvd
);
811 * Remove a 1-way mirror/replacing vdev from the tree.
814 vdev_remove_parent(vdev_t
*cvd
)
816 vdev_t
*mvd
= cvd
->vdev_parent
;
817 vdev_t
*pvd
= mvd
->vdev_parent
;
819 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
821 ASSERT(mvd
->vdev_children
== 1);
822 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
823 mvd
->vdev_ops
== &vdev_replacing_ops
||
824 mvd
->vdev_ops
== &vdev_spare_ops
);
825 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
827 vdev_remove_child(mvd
, cvd
);
828 vdev_remove_child(pvd
, mvd
);
831 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
832 * Otherwise, we could have detached an offline device, and when we
833 * go to import the pool we'll think we have two top-level vdevs,
834 * instead of a different version of the same top-level vdev.
836 if (mvd
->vdev_top
== mvd
) {
837 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
838 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
839 cvd
->vdev_guid
+= guid_delta
;
840 cvd
->vdev_guid_sum
+= guid_delta
;
843 * If pool not set for autoexpand, we need to also preserve
844 * mvd's asize to prevent automatic expansion of cvd.
845 * Otherwise if we are adjusting the mirror by attaching and
846 * detaching children of non-uniform sizes, the mirror could
847 * autoexpand, unexpectedly requiring larger devices to
848 * re-establish the mirror.
850 if (!cvd
->vdev_spa
->spa_autoexpand
)
851 cvd
->vdev_asize
= mvd
->vdev_asize
;
853 cvd
->vdev_id
= mvd
->vdev_id
;
854 vdev_add_child(pvd
, cvd
);
855 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
857 if (cvd
== cvd
->vdev_top
)
858 vdev_top_transfer(mvd
, cvd
);
860 ASSERT(mvd
->vdev_children
== 0);
865 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
867 spa_t
*spa
= vd
->vdev_spa
;
868 objset_t
*mos
= spa
->spa_meta_objset
;
870 uint64_t oldc
= vd
->vdev_ms_count
;
871 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
875 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
878 * This vdev is not being allocated from yet or is a hole.
880 if (vd
->vdev_ms_shift
== 0)
883 ASSERT(!vd
->vdev_ishole
);
886 * Compute the raidz-deflation ratio. Note, we hard-code
887 * in 128k (1 << 17) because it is the "typical" blocksize.
888 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
889 * otherwise it would inconsistently account for existing bp's.
891 vd
->vdev_deflate_ratio
= (1 << 17) /
892 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
894 ASSERT(oldc
<= newc
);
896 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
899 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
900 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
904 vd
->vdev_ms_count
= newc
;
906 for (m
= oldc
; m
< newc
; m
++) {
910 error
= dmu_read(mos
, vd
->vdev_ms_array
,
911 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
917 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
924 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
927 * If the vdev is being removed we don't activate
928 * the metaslabs since we want to ensure that no new
929 * allocations are performed on this device.
931 if (oldc
== 0 && !vd
->vdev_removing
)
932 metaslab_group_activate(vd
->vdev_mg
);
935 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
941 vdev_metaslab_fini(vdev_t
*vd
)
944 uint64_t count
= vd
->vdev_ms_count
;
946 if (vd
->vdev_ms
!= NULL
) {
947 metaslab_group_passivate(vd
->vdev_mg
);
948 for (m
= 0; m
< count
; m
++) {
949 metaslab_t
*msp
= vd
->vdev_ms
[m
];
954 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
958 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
961 typedef struct vdev_probe_stats
{
962 boolean_t vps_readable
;
963 boolean_t vps_writeable
;
965 } vdev_probe_stats_t
;
968 vdev_probe_done(zio_t
*zio
)
970 spa_t
*spa
= zio
->io_spa
;
971 vdev_t
*vd
= zio
->io_vd
;
972 vdev_probe_stats_t
*vps
= zio
->io_private
;
974 ASSERT(vd
->vdev_probe_zio
!= NULL
);
976 if (zio
->io_type
== ZIO_TYPE_READ
) {
977 if (zio
->io_error
== 0)
978 vps
->vps_readable
= 1;
979 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
980 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
981 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
982 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
983 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
985 zio_buf_free(zio
->io_data
, zio
->io_size
);
987 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
988 if (zio
->io_error
== 0)
989 vps
->vps_writeable
= 1;
990 zio_buf_free(zio
->io_data
, zio
->io_size
);
991 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
994 vd
->vdev_cant_read
|= !vps
->vps_readable
;
995 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
997 if (vdev_readable(vd
) &&
998 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1001 ASSERT(zio
->io_error
!= 0);
1002 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1003 spa
, vd
, NULL
, 0, 0);
1004 zio
->io_error
= SET_ERROR(ENXIO
);
1007 mutex_enter(&vd
->vdev_probe_lock
);
1008 ASSERT(vd
->vdev_probe_zio
== zio
);
1009 vd
->vdev_probe_zio
= NULL
;
1010 mutex_exit(&vd
->vdev_probe_lock
);
1012 while ((pio
= zio_walk_parents(zio
)) != NULL
)
1013 if (!vdev_accessible(vd
, pio
))
1014 pio
->io_error
= SET_ERROR(ENXIO
);
1016 kmem_free(vps
, sizeof (*vps
));
1021 * Determine whether this device is accessible.
1023 * Read and write to several known locations: the pad regions of each
1024 * vdev label but the first, which we leave alone in case it contains
1028 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1030 spa_t
*spa
= vd
->vdev_spa
;
1031 vdev_probe_stats_t
*vps
= NULL
;
1035 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1038 * Don't probe the probe.
1040 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1044 * To prevent 'probe storms' when a device fails, we create
1045 * just one probe i/o at a time. All zios that want to probe
1046 * this vdev will become parents of the probe io.
1048 mutex_enter(&vd
->vdev_probe_lock
);
1050 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1051 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1053 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1054 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1057 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1059 * vdev_cant_read and vdev_cant_write can only
1060 * transition from TRUE to FALSE when we have the
1061 * SCL_ZIO lock as writer; otherwise they can only
1062 * transition from FALSE to TRUE. This ensures that
1063 * any zio looking at these values can assume that
1064 * failures persist for the life of the I/O. That's
1065 * important because when a device has intermittent
1066 * connectivity problems, we want to ensure that
1067 * they're ascribed to the device (ENXIO) and not
1070 * Since we hold SCL_ZIO as writer here, clear both
1071 * values so the probe can reevaluate from first
1074 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1075 vd
->vdev_cant_read
= B_FALSE
;
1076 vd
->vdev_cant_write
= B_FALSE
;
1079 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1080 vdev_probe_done
, vps
,
1081 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1084 * We can't change the vdev state in this context, so we
1085 * kick off an async task to do it on our behalf.
1088 vd
->vdev_probe_wanted
= B_TRUE
;
1089 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1094 zio_add_child(zio
, pio
);
1096 mutex_exit(&vd
->vdev_probe_lock
);
1099 ASSERT(zio
!= NULL
);
1103 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1104 zio_nowait(zio_read_phys(pio
, vd
,
1105 vdev_label_offset(vd
->vdev_psize
, l
,
1106 offsetof(vdev_label_t
, vl_pad2
)),
1107 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1108 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1109 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1120 vdev_open_child(void *arg
)
1124 vd
->vdev_open_thread
= curthread
;
1125 vd
->vdev_open_error
= vdev_open(vd
);
1126 vd
->vdev_open_thread
= NULL
;
1127 vd
->vdev_parent
->vdev_nonrot
&= vd
->vdev_nonrot
;
1131 vdev_uses_zvols(vdev_t
*vd
)
1136 if (zvol_is_zvol(vd
->vdev_path
))
1140 for (c
= 0; c
< vd
->vdev_children
; c
++)
1141 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1148 vdev_open_children(vdev_t
*vd
)
1151 int children
= vd
->vdev_children
;
1154 vd
->vdev_nonrot
= B_TRUE
;
1157 * in order to handle pools on top of zvols, do the opens
1158 * in a single thread so that the same thread holds the
1159 * spa_namespace_lock
1161 if (vdev_uses_zvols(vd
)) {
1162 for (c
= 0; c
< children
; c
++) {
1163 vd
->vdev_child
[c
]->vdev_open_error
=
1164 vdev_open(vd
->vdev_child
[c
]);
1165 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1169 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1170 children
, children
, TASKQ_PREPOPULATE
);
1172 for (c
= 0; c
< children
; c
++)
1173 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1178 for (c
= 0; c
< children
; c
++)
1179 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1183 * Prepare a virtual device for access.
1186 vdev_open(vdev_t
*vd
)
1188 spa_t
*spa
= vd
->vdev_spa
;
1191 uint64_t max_osize
= 0;
1192 uint64_t asize
, max_asize
, psize
;
1193 uint64_t ashift
= 0;
1196 ASSERT(vd
->vdev_open_thread
== curthread
||
1197 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1198 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1199 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1200 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1202 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1203 vd
->vdev_cant_read
= B_FALSE
;
1204 vd
->vdev_cant_write
= B_FALSE
;
1205 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1208 * If this vdev is not removed, check its fault status. If it's
1209 * faulted, bail out of the open.
1211 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1212 ASSERT(vd
->vdev_children
== 0);
1213 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1214 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1215 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1216 vd
->vdev_label_aux
);
1217 return (SET_ERROR(ENXIO
));
1218 } else if (vd
->vdev_offline
) {
1219 ASSERT(vd
->vdev_children
== 0);
1220 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1221 return (SET_ERROR(ENXIO
));
1224 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1227 * Reset the vdev_reopening flag so that we actually close
1228 * the vdev on error.
1230 vd
->vdev_reopening
= B_FALSE
;
1231 if (zio_injection_enabled
&& error
== 0)
1232 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1235 if (vd
->vdev_removed
&&
1236 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1237 vd
->vdev_removed
= B_FALSE
;
1239 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1240 vd
->vdev_stat
.vs_aux
);
1244 vd
->vdev_removed
= B_FALSE
;
1247 * Recheck the faulted flag now that we have confirmed that
1248 * the vdev is accessible. If we're faulted, bail.
1250 if (vd
->vdev_faulted
) {
1251 ASSERT(vd
->vdev_children
== 0);
1252 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1253 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1254 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1255 vd
->vdev_label_aux
);
1256 return (SET_ERROR(ENXIO
));
1259 if (vd
->vdev_degraded
) {
1260 ASSERT(vd
->vdev_children
== 0);
1261 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1262 VDEV_AUX_ERR_EXCEEDED
);
1264 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1268 * For hole or missing vdevs we just return success.
1270 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1273 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1274 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1275 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1281 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1282 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1284 if (vd
->vdev_children
== 0) {
1285 if (osize
< SPA_MINDEVSIZE
) {
1286 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1287 VDEV_AUX_TOO_SMALL
);
1288 return (SET_ERROR(EOVERFLOW
));
1291 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1292 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1293 VDEV_LABEL_END_SIZE
);
1295 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1296 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1297 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1298 VDEV_AUX_TOO_SMALL
);
1299 return (SET_ERROR(EOVERFLOW
));
1303 max_asize
= max_osize
;
1306 vd
->vdev_psize
= psize
;
1309 * Make sure the allocatable size hasn't shrunk.
1311 if (asize
< vd
->vdev_min_asize
) {
1312 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1313 VDEV_AUX_BAD_LABEL
);
1314 return (SET_ERROR(EINVAL
));
1317 if (vd
->vdev_asize
== 0) {
1319 * This is the first-ever open, so use the computed values.
1320 * For compatibility, a different ashift can be requested.
1322 vd
->vdev_asize
= asize
;
1323 vd
->vdev_max_asize
= max_asize
;
1324 if (vd
->vdev_ashift
== 0)
1325 vd
->vdev_ashift
= ashift
;
1328 * Detect if the alignment requirement has increased.
1329 * We don't want to make the pool unavailable, just
1330 * post an event instead.
1332 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1333 vd
->vdev_ops
->vdev_op_leaf
) {
1334 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1335 spa
, vd
, NULL
, 0, 0);
1338 vd
->vdev_max_asize
= max_asize
;
1342 * If all children are healthy and the asize has increased,
1343 * then we've experienced dynamic LUN growth. If automatic
1344 * expansion is enabled then use the additional space.
1346 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1347 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1348 vd
->vdev_asize
= asize
;
1350 vdev_set_min_asize(vd
);
1353 * Ensure we can issue some IO before declaring the
1354 * vdev open for business.
1356 if (vd
->vdev_ops
->vdev_op_leaf
&&
1357 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1358 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1359 VDEV_AUX_ERR_EXCEEDED
);
1364 * Track the min and max ashift values for normal data devices.
1366 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1367 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1368 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1369 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1370 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1371 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1375 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1376 * resilver. But don't do this if we are doing a reopen for a scrub,
1377 * since this would just restart the scrub we are already doing.
1379 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1380 vdev_resilver_needed(vd
, NULL
, NULL
))
1381 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1387 * Called once the vdevs are all opened, this routine validates the label
1388 * contents. This needs to be done before vdev_load() so that we don't
1389 * inadvertently do repair I/Os to the wrong device.
1391 * If 'strict' is false ignore the spa guid check. This is necessary because
1392 * if the machine crashed during a re-guid the new guid might have been written
1393 * to all of the vdev labels, but not the cached config. The strict check
1394 * will be performed when the pool is opened again using the mos config.
1396 * This function will only return failure if one of the vdevs indicates that it
1397 * has since been destroyed or exported. This is only possible if
1398 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1399 * will be updated but the function will return 0.
1402 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1404 spa_t
*spa
= vd
->vdev_spa
;
1406 uint64_t guid
= 0, top_guid
;
1410 for (c
= 0; c
< vd
->vdev_children
; c
++)
1411 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1412 return (SET_ERROR(EBADF
));
1415 * If the device has already failed, or was marked offline, don't do
1416 * any further validation. Otherwise, label I/O will fail and we will
1417 * overwrite the previous state.
1419 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1420 uint64_t aux_guid
= 0;
1422 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1423 spa_last_synced_txg(spa
) : -1ULL;
1425 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1426 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1427 VDEV_AUX_BAD_LABEL
);
1432 * Determine if this vdev has been split off into another
1433 * pool. If so, then refuse to open it.
1435 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1436 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1437 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1438 VDEV_AUX_SPLIT_POOL
);
1443 if (strict
&& (nvlist_lookup_uint64(label
,
1444 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1445 guid
!= spa_guid(spa
))) {
1446 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1447 VDEV_AUX_CORRUPT_DATA
);
1452 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1453 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1458 * If this vdev just became a top-level vdev because its
1459 * sibling was detached, it will have adopted the parent's
1460 * vdev guid -- but the label may or may not be on disk yet.
1461 * Fortunately, either version of the label will have the
1462 * same top guid, so if we're a top-level vdev, we can
1463 * safely compare to that instead.
1465 * If we split this vdev off instead, then we also check the
1466 * original pool's guid. We don't want to consider the vdev
1467 * corrupt if it is partway through a split operation.
1469 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1471 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1473 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1474 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1475 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1476 VDEV_AUX_CORRUPT_DATA
);
1481 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1483 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1484 VDEV_AUX_CORRUPT_DATA
);
1492 * If this is a verbatim import, no need to check the
1493 * state of the pool.
1495 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1496 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1497 state
!= POOL_STATE_ACTIVE
)
1498 return (SET_ERROR(EBADF
));
1501 * If we were able to open and validate a vdev that was
1502 * previously marked permanently unavailable, clear that state
1505 if (vd
->vdev_not_present
)
1506 vd
->vdev_not_present
= 0;
1513 * Close a virtual device.
1516 vdev_close(vdev_t
*vd
)
1518 vdev_t
*pvd
= vd
->vdev_parent
;
1519 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1521 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1524 * If our parent is reopening, then we are as well, unless we are
1527 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1528 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1530 vd
->vdev_ops
->vdev_op_close(vd
);
1532 vdev_cache_purge(vd
);
1535 * We record the previous state before we close it, so that if we are
1536 * doing a reopen(), we don't generate FMA ereports if we notice that
1537 * it's still faulted.
1539 vd
->vdev_prevstate
= vd
->vdev_state
;
1541 if (vd
->vdev_offline
)
1542 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1544 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1545 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1549 vdev_hold(vdev_t
*vd
)
1551 spa_t
*spa
= vd
->vdev_spa
;
1554 ASSERT(spa_is_root(spa
));
1555 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1558 for (c
= 0; c
< vd
->vdev_children
; c
++)
1559 vdev_hold(vd
->vdev_child
[c
]);
1561 if (vd
->vdev_ops
->vdev_op_leaf
)
1562 vd
->vdev_ops
->vdev_op_hold(vd
);
1566 vdev_rele(vdev_t
*vd
)
1570 ASSERT(spa_is_root(vd
->vdev_spa
));
1571 for (c
= 0; c
< vd
->vdev_children
; c
++)
1572 vdev_rele(vd
->vdev_child
[c
]);
1574 if (vd
->vdev_ops
->vdev_op_leaf
)
1575 vd
->vdev_ops
->vdev_op_rele(vd
);
1579 * Reopen all interior vdevs and any unopened leaves. We don't actually
1580 * reopen leaf vdevs which had previously been opened as they might deadlock
1581 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1582 * If the leaf has never been opened then open it, as usual.
1585 vdev_reopen(vdev_t
*vd
)
1587 spa_t
*spa
= vd
->vdev_spa
;
1589 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1591 /* set the reopening flag unless we're taking the vdev offline */
1592 vd
->vdev_reopening
= !vd
->vdev_offline
;
1594 (void) vdev_open(vd
);
1597 * Call vdev_validate() here to make sure we have the same device.
1598 * Otherwise, a device with an invalid label could be successfully
1599 * opened in response to vdev_reopen().
1602 (void) vdev_validate_aux(vd
);
1603 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1604 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1605 !l2arc_vdev_present(vd
))
1606 l2arc_add_vdev(spa
, vd
);
1608 (void) vdev_validate(vd
, B_TRUE
);
1612 * Reassess parent vdev's health.
1614 vdev_propagate_state(vd
);
1618 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1623 * Normally, partial opens (e.g. of a mirror) are allowed.
1624 * For a create, however, we want to fail the request if
1625 * there are any components we can't open.
1627 error
= vdev_open(vd
);
1629 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1631 return (error
? error
: ENXIO
);
1635 * Recursively load DTLs and initialize all labels.
1637 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1638 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1639 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1648 vdev_metaslab_set_size(vdev_t
*vd
)
1651 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1653 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1654 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1658 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1660 ASSERT(vd
== vd
->vdev_top
);
1661 ASSERT(!vd
->vdev_ishole
);
1662 ASSERT(ISP2(flags
));
1663 ASSERT(spa_writeable(vd
->vdev_spa
));
1665 if (flags
& VDD_METASLAB
)
1666 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1668 if (flags
& VDD_DTL
)
1669 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1671 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1675 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1679 for (c
= 0; c
< vd
->vdev_children
; c
++)
1680 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1682 if (vd
->vdev_ops
->vdev_op_leaf
)
1683 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1689 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1690 * the vdev has less than perfect replication. There are four kinds of DTL:
1692 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1694 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1696 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1697 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1698 * txgs that was scrubbed.
1700 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1701 * persistent errors or just some device being offline.
1702 * Unlike the other three, the DTL_OUTAGE map is not generally
1703 * maintained; it's only computed when needed, typically to
1704 * determine whether a device can be detached.
1706 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1707 * either has the data or it doesn't.
1709 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1710 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1711 * if any child is less than fully replicated, then so is its parent.
1712 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1713 * comprising only those txgs which appear in 'maxfaults' or more children;
1714 * those are the txgs we don't have enough replication to read. For example,
1715 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1716 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1717 * two child DTL_MISSING maps.
1719 * It should be clear from the above that to compute the DTLs and outage maps
1720 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1721 * Therefore, that is all we keep on disk. When loading the pool, or after
1722 * a configuration change, we generate all other DTLs from first principles.
1725 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1727 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1729 ASSERT(t
< DTL_TYPES
);
1730 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1731 ASSERT(spa_writeable(vd
->vdev_spa
));
1733 mutex_enter(rt
->rt_lock
);
1734 if (!range_tree_contains(rt
, txg
, size
))
1735 range_tree_add(rt
, txg
, size
);
1736 mutex_exit(rt
->rt_lock
);
1740 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1742 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1743 boolean_t dirty
= B_FALSE
;
1745 ASSERT(t
< DTL_TYPES
);
1746 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1748 mutex_enter(rt
->rt_lock
);
1749 if (range_tree_space(rt
) != 0)
1750 dirty
= range_tree_contains(rt
, txg
, size
);
1751 mutex_exit(rt
->rt_lock
);
1757 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1759 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1762 mutex_enter(rt
->rt_lock
);
1763 empty
= (range_tree_space(rt
) == 0);
1764 mutex_exit(rt
->rt_lock
);
1770 * Returns the lowest txg in the DTL range.
1773 vdev_dtl_min(vdev_t
*vd
)
1777 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1778 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1779 ASSERT0(vd
->vdev_children
);
1781 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1782 return (rs
->rs_start
- 1);
1786 * Returns the highest txg in the DTL.
1789 vdev_dtl_max(vdev_t
*vd
)
1793 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1794 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1795 ASSERT0(vd
->vdev_children
);
1797 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1798 return (rs
->rs_end
);
1802 * Determine if a resilvering vdev should remove any DTL entries from
1803 * its range. If the vdev was resilvering for the entire duration of the
1804 * scan then it should excise that range from its DTLs. Otherwise, this
1805 * vdev is considered partially resilvered and should leave its DTL
1806 * entries intact. The comment in vdev_dtl_reassess() describes how we
1810 vdev_dtl_should_excise(vdev_t
*vd
)
1812 spa_t
*spa
= vd
->vdev_spa
;
1813 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1815 ASSERT0(scn
->scn_phys
.scn_errors
);
1816 ASSERT0(vd
->vdev_children
);
1818 if (vd
->vdev_resilver_txg
== 0 ||
1819 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1823 * When a resilver is initiated the scan will assign the scn_max_txg
1824 * value to the highest txg value that exists in all DTLs. If this
1825 * device's max DTL is not part of this scan (i.e. it is not in
1826 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1829 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1830 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1831 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1832 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1839 * Reassess DTLs after a config change or scrub completion.
1842 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1844 spa_t
*spa
= vd
->vdev_spa
;
1848 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1850 for (c
= 0; c
< vd
->vdev_children
; c
++)
1851 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1852 scrub_txg
, scrub_done
);
1854 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1857 if (vd
->vdev_ops
->vdev_op_leaf
) {
1858 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1860 mutex_enter(&vd
->vdev_dtl_lock
);
1863 * If we've completed a scan cleanly then determine
1864 * if this vdev should remove any DTLs. We only want to
1865 * excise regions on vdevs that were available during
1866 * the entire duration of this scan.
1868 if (scrub_txg
!= 0 &&
1869 (spa
->spa_scrub_started
||
1870 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1871 vdev_dtl_should_excise(vd
)) {
1873 * We completed a scrub up to scrub_txg. If we
1874 * did it without rebooting, then the scrub dtl
1875 * will be valid, so excise the old region and
1876 * fold in the scrub dtl. Otherwise, leave the
1877 * dtl as-is if there was an error.
1879 * There's little trick here: to excise the beginning
1880 * of the DTL_MISSING map, we put it into a reference
1881 * tree and then add a segment with refcnt -1 that
1882 * covers the range [0, scrub_txg). This means
1883 * that each txg in that range has refcnt -1 or 0.
1884 * We then add DTL_SCRUB with a refcnt of 2, so that
1885 * entries in the range [0, scrub_txg) will have a
1886 * positive refcnt -- either 1 or 2. We then convert
1887 * the reference tree into the new DTL_MISSING map.
1889 space_reftree_create(&reftree
);
1890 space_reftree_add_map(&reftree
,
1891 vd
->vdev_dtl
[DTL_MISSING
], 1);
1892 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1893 space_reftree_add_map(&reftree
,
1894 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1895 space_reftree_generate_map(&reftree
,
1896 vd
->vdev_dtl
[DTL_MISSING
], 1);
1897 space_reftree_destroy(&reftree
);
1899 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1900 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1901 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1903 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1904 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1905 if (!vdev_readable(vd
))
1906 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1908 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1909 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1912 * If the vdev was resilvering and no longer has any
1913 * DTLs then reset its resilvering flag and dirty
1914 * the top level so that we persist the change.
1916 if (vd
->vdev_resilver_txg
!= 0 &&
1917 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1918 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
1919 vd
->vdev_resilver_txg
= 0;
1920 vdev_config_dirty(vd
->vdev_top
);
1923 mutex_exit(&vd
->vdev_dtl_lock
);
1926 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1930 mutex_enter(&vd
->vdev_dtl_lock
);
1931 for (t
= 0; t
< DTL_TYPES
; t
++) {
1934 /* account for child's outage in parent's missing map */
1935 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1937 continue; /* leaf vdevs only */
1938 if (t
== DTL_PARTIAL
)
1939 minref
= 1; /* i.e. non-zero */
1940 else if (vd
->vdev_nparity
!= 0)
1941 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1943 minref
= vd
->vdev_children
; /* any kind of mirror */
1944 space_reftree_create(&reftree
);
1945 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1946 vdev_t
*cvd
= vd
->vdev_child
[c
];
1947 mutex_enter(&cvd
->vdev_dtl_lock
);
1948 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1949 mutex_exit(&cvd
->vdev_dtl_lock
);
1951 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1952 space_reftree_destroy(&reftree
);
1954 mutex_exit(&vd
->vdev_dtl_lock
);
1958 vdev_dtl_load(vdev_t
*vd
)
1960 spa_t
*spa
= vd
->vdev_spa
;
1961 objset_t
*mos
= spa
->spa_meta_objset
;
1965 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
1966 ASSERT(!vd
->vdev_ishole
);
1968 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
1969 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
1972 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
1974 mutex_enter(&vd
->vdev_dtl_lock
);
1977 * Now that we've opened the space_map we need to update
1980 space_map_update(vd
->vdev_dtl_sm
);
1982 error
= space_map_load(vd
->vdev_dtl_sm
,
1983 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
1984 mutex_exit(&vd
->vdev_dtl_lock
);
1989 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1990 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
1999 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2001 spa_t
*spa
= vd
->vdev_spa
;
2003 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2004 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2009 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2011 spa_t
*spa
= vd
->vdev_spa
;
2012 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2013 DMU_OT_NONE
, 0, tx
);
2016 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2023 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2027 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2028 vd
->vdev_ops
!= &vdev_missing_ops
&&
2029 vd
->vdev_ops
!= &vdev_root_ops
&&
2030 !vd
->vdev_top
->vdev_removing
) {
2031 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2032 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2034 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2035 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2038 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2039 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2044 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2046 spa_t
*spa
= vd
->vdev_spa
;
2047 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2048 objset_t
*mos
= spa
->spa_meta_objset
;
2049 range_tree_t
*rtsync
;
2052 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2054 ASSERT(!vd
->vdev_ishole
);
2055 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2057 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2059 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2060 mutex_enter(&vd
->vdev_dtl_lock
);
2061 space_map_free(vd
->vdev_dtl_sm
, tx
);
2062 space_map_close(vd
->vdev_dtl_sm
);
2063 vd
->vdev_dtl_sm
= NULL
;
2064 mutex_exit(&vd
->vdev_dtl_lock
);
2067 * We only destroy the leaf ZAP for detached leaves or for
2068 * removed log devices. Removed data devices handle leaf ZAP
2069 * cleanup later, once cancellation is no longer possible.
2071 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2072 vd
->vdev_top
->vdev_islog
)) {
2073 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2074 vd
->vdev_leaf_zap
= 0;
2081 if (vd
->vdev_dtl_sm
== NULL
) {
2082 uint64_t new_object
;
2084 new_object
= space_map_alloc(mos
, tx
);
2085 VERIFY3U(new_object
, !=, 0);
2087 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2088 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2089 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2092 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2094 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2096 mutex_enter(&rtlock
);
2098 mutex_enter(&vd
->vdev_dtl_lock
);
2099 range_tree_walk(rt
, range_tree_add
, rtsync
);
2100 mutex_exit(&vd
->vdev_dtl_lock
);
2102 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2103 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2104 range_tree_vacate(rtsync
, NULL
, NULL
);
2106 range_tree_destroy(rtsync
);
2108 mutex_exit(&rtlock
);
2109 mutex_destroy(&rtlock
);
2112 * If the object for the space map has changed then dirty
2113 * the top level so that we update the config.
2115 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2116 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2117 "new object %llu", txg
, spa_name(spa
), object
,
2118 space_map_object(vd
->vdev_dtl_sm
));
2119 vdev_config_dirty(vd
->vdev_top
);
2124 mutex_enter(&vd
->vdev_dtl_lock
);
2125 space_map_update(vd
->vdev_dtl_sm
);
2126 mutex_exit(&vd
->vdev_dtl_lock
);
2130 * Determine whether the specified vdev can be offlined/detached/removed
2131 * without losing data.
2134 vdev_dtl_required(vdev_t
*vd
)
2136 spa_t
*spa
= vd
->vdev_spa
;
2137 vdev_t
*tvd
= vd
->vdev_top
;
2138 uint8_t cant_read
= vd
->vdev_cant_read
;
2141 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2143 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2147 * Temporarily mark the device as unreadable, and then determine
2148 * whether this results in any DTL outages in the top-level vdev.
2149 * If not, we can safely offline/detach/remove the device.
2151 vd
->vdev_cant_read
= B_TRUE
;
2152 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2153 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2154 vd
->vdev_cant_read
= cant_read
;
2155 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2157 if (!required
&& zio_injection_enabled
)
2158 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2164 * Determine if resilver is needed, and if so the txg range.
2167 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2169 boolean_t needed
= B_FALSE
;
2170 uint64_t thismin
= UINT64_MAX
;
2171 uint64_t thismax
= 0;
2174 if (vd
->vdev_children
== 0) {
2175 mutex_enter(&vd
->vdev_dtl_lock
);
2176 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2177 vdev_writeable(vd
)) {
2179 thismin
= vdev_dtl_min(vd
);
2180 thismax
= vdev_dtl_max(vd
);
2183 mutex_exit(&vd
->vdev_dtl_lock
);
2185 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2186 vdev_t
*cvd
= vd
->vdev_child
[c
];
2187 uint64_t cmin
, cmax
;
2189 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2190 thismin
= MIN(thismin
, cmin
);
2191 thismax
= MAX(thismax
, cmax
);
2197 if (needed
&& minp
) {
2205 vdev_load(vdev_t
*vd
)
2210 * Recursively load all children.
2212 for (c
= 0; c
< vd
->vdev_children
; c
++)
2213 vdev_load(vd
->vdev_child
[c
]);
2216 * If this is a top-level vdev, initialize its metaslabs.
2218 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2219 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2220 vdev_metaslab_init(vd
, 0) != 0))
2221 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2222 VDEV_AUX_CORRUPT_DATA
);
2225 * If this is a leaf vdev, load its DTL.
2227 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2228 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2229 VDEV_AUX_CORRUPT_DATA
);
2233 * The special vdev case is used for hot spares and l2cache devices. Its
2234 * sole purpose it to set the vdev state for the associated vdev. To do this,
2235 * we make sure that we can open the underlying device, then try to read the
2236 * label, and make sure that the label is sane and that it hasn't been
2237 * repurposed to another pool.
2240 vdev_validate_aux(vdev_t
*vd
)
2243 uint64_t guid
, version
;
2246 if (!vdev_readable(vd
))
2249 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2250 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2251 VDEV_AUX_CORRUPT_DATA
);
2255 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2256 !SPA_VERSION_IS_SUPPORTED(version
) ||
2257 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2258 guid
!= vd
->vdev_guid
||
2259 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2260 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2261 VDEV_AUX_CORRUPT_DATA
);
2267 * We don't actually check the pool state here. If it's in fact in
2268 * use by another pool, we update this fact on the fly when requested.
2275 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2277 spa_t
*spa
= vd
->vdev_spa
;
2278 objset_t
*mos
= spa
->spa_meta_objset
;
2282 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2283 ASSERT(vd
== vd
->vdev_top
);
2284 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2286 if (vd
->vdev_ms
!= NULL
) {
2287 metaslab_group_t
*mg
= vd
->vdev_mg
;
2289 metaslab_group_histogram_verify(mg
);
2290 metaslab_class_histogram_verify(mg
->mg_class
);
2292 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2293 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2295 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2298 mutex_enter(&msp
->ms_lock
);
2300 * If the metaslab was not loaded when the vdev
2301 * was removed then the histogram accounting may
2302 * not be accurate. Update the histogram information
2303 * here so that we ensure that the metaslab group
2304 * and metaslab class are up-to-date.
2306 metaslab_group_histogram_remove(mg
, msp
);
2308 VERIFY0(space_map_allocated(msp
->ms_sm
));
2309 space_map_free(msp
->ms_sm
, tx
);
2310 space_map_close(msp
->ms_sm
);
2312 mutex_exit(&msp
->ms_lock
);
2315 metaslab_group_histogram_verify(mg
);
2316 metaslab_class_histogram_verify(mg
->mg_class
);
2317 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2318 ASSERT0(mg
->mg_histogram
[i
]);
2322 if (vd
->vdev_ms_array
) {
2323 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2324 vd
->vdev_ms_array
= 0;
2327 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2328 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2329 vd
->vdev_top_zap
= 0;
2335 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2338 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2340 ASSERT(!vd
->vdev_ishole
);
2342 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2343 metaslab_sync_done(msp
, txg
);
2346 metaslab_sync_reassess(vd
->vdev_mg
);
2350 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2352 spa_t
*spa
= vd
->vdev_spa
;
2357 ASSERT(!vd
->vdev_ishole
);
2359 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2360 ASSERT(vd
== vd
->vdev_top
);
2361 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2362 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2363 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2364 ASSERT(vd
->vdev_ms_array
!= 0);
2365 vdev_config_dirty(vd
);
2370 * Remove the metadata associated with this vdev once it's empty.
2372 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2373 vdev_remove(vd
, txg
);
2375 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2376 metaslab_sync(msp
, txg
);
2377 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2380 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2381 vdev_dtl_sync(lvd
, txg
);
2383 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2387 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2389 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2393 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2394 * not be opened, and no I/O is attempted.
2397 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2401 spa_vdev_state_enter(spa
, SCL_NONE
);
2403 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2404 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2406 if (!vd
->vdev_ops
->vdev_op_leaf
)
2407 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2412 * We don't directly use the aux state here, but if we do a
2413 * vdev_reopen(), we need this value to be present to remember why we
2416 vd
->vdev_label_aux
= aux
;
2419 * Faulted state takes precedence over degraded.
2421 vd
->vdev_delayed_close
= B_FALSE
;
2422 vd
->vdev_faulted
= 1ULL;
2423 vd
->vdev_degraded
= 0ULL;
2424 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2427 * If this device has the only valid copy of the data, then
2428 * back off and simply mark the vdev as degraded instead.
2430 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2431 vd
->vdev_degraded
= 1ULL;
2432 vd
->vdev_faulted
= 0ULL;
2435 * If we reopen the device and it's not dead, only then do we
2440 if (vdev_readable(vd
))
2441 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2444 return (spa_vdev_state_exit(spa
, vd
, 0));
2448 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2449 * user that something is wrong. The vdev continues to operate as normal as far
2450 * as I/O is concerned.
2453 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2457 spa_vdev_state_enter(spa
, SCL_NONE
);
2459 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2460 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2462 if (!vd
->vdev_ops
->vdev_op_leaf
)
2463 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2466 * If the vdev is already faulted, then don't do anything.
2468 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2469 return (spa_vdev_state_exit(spa
, NULL
, 0));
2471 vd
->vdev_degraded
= 1ULL;
2472 if (!vdev_is_dead(vd
))
2473 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2476 return (spa_vdev_state_exit(spa
, vd
, 0));
2480 * Online the given vdev.
2482 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2483 * spare device should be detached when the device finishes resilvering.
2484 * Second, the online should be treated like a 'test' online case, so no FMA
2485 * events are generated if the device fails to open.
2488 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2490 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2491 boolean_t postevent
= B_FALSE
;
2493 spa_vdev_state_enter(spa
, SCL_NONE
);
2495 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2496 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2498 if (!vd
->vdev_ops
->vdev_op_leaf
)
2499 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2502 (vd
->vdev_offline
== B_TRUE
|| vd
->vdev_tmpoffline
== B_TRUE
) ?
2506 vd
->vdev_offline
= B_FALSE
;
2507 vd
->vdev_tmpoffline
= B_FALSE
;
2508 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2509 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2511 /* XXX - L2ARC 1.0 does not support expansion */
2512 if (!vd
->vdev_aux
) {
2513 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2514 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2518 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2520 if (!vd
->vdev_aux
) {
2521 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2522 pvd
->vdev_expanding
= B_FALSE
;
2526 *newstate
= vd
->vdev_state
;
2527 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2528 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2529 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2530 vd
->vdev_parent
->vdev_child
[0] == vd
)
2531 vd
->vdev_unspare
= B_TRUE
;
2533 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2535 /* XXX - L2ARC 1.0 does not support expansion */
2537 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2538 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2542 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2544 return (spa_vdev_state_exit(spa
, vd
, 0));
2548 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2552 uint64_t generation
;
2553 metaslab_group_t
*mg
;
2556 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2558 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2559 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2561 if (!vd
->vdev_ops
->vdev_op_leaf
)
2562 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2566 generation
= spa
->spa_config_generation
+ 1;
2569 * If the device isn't already offline, try to offline it.
2571 if (!vd
->vdev_offline
) {
2573 * If this device has the only valid copy of some data,
2574 * don't allow it to be offlined. Log devices are always
2577 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2578 vdev_dtl_required(vd
))
2579 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2582 * If the top-level is a slog and it has had allocations
2583 * then proceed. We check that the vdev's metaslab group
2584 * is not NULL since it's possible that we may have just
2585 * added this vdev but not yet initialized its metaslabs.
2587 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2589 * Prevent any future allocations.
2591 metaslab_group_passivate(mg
);
2592 (void) spa_vdev_state_exit(spa
, vd
, 0);
2594 error
= spa_offline_log(spa
);
2596 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2599 * Check to see if the config has changed.
2601 if (error
|| generation
!= spa
->spa_config_generation
) {
2602 metaslab_group_activate(mg
);
2604 return (spa_vdev_state_exit(spa
,
2606 (void) spa_vdev_state_exit(spa
, vd
, 0);
2609 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2613 * Offline this device and reopen its top-level vdev.
2614 * If the top-level vdev is a log device then just offline
2615 * it. Otherwise, if this action results in the top-level
2616 * vdev becoming unusable, undo it and fail the request.
2618 vd
->vdev_offline
= B_TRUE
;
2621 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2622 vdev_is_dead(tvd
)) {
2623 vd
->vdev_offline
= B_FALSE
;
2625 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2629 * Add the device back into the metaslab rotor so that
2630 * once we online the device it's open for business.
2632 if (tvd
->vdev_islog
&& mg
!= NULL
)
2633 metaslab_group_activate(mg
);
2636 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2638 return (spa_vdev_state_exit(spa
, vd
, 0));
2642 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2646 mutex_enter(&spa
->spa_vdev_top_lock
);
2647 error
= vdev_offline_locked(spa
, guid
, flags
);
2648 mutex_exit(&spa
->spa_vdev_top_lock
);
2654 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2655 * vdev_offline(), we assume the spa config is locked. We also clear all
2656 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2659 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2661 vdev_t
*rvd
= spa
->spa_root_vdev
;
2664 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2669 vd
->vdev_stat
.vs_read_errors
= 0;
2670 vd
->vdev_stat
.vs_write_errors
= 0;
2671 vd
->vdev_stat
.vs_checksum_errors
= 0;
2673 for (c
= 0; c
< vd
->vdev_children
; c
++)
2674 vdev_clear(spa
, vd
->vdev_child
[c
]);
2677 * If we're in the FAULTED state or have experienced failed I/O, then
2678 * clear the persistent state and attempt to reopen the device. We
2679 * also mark the vdev config dirty, so that the new faulted state is
2680 * written out to disk.
2682 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2683 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2686 * When reopening in reponse to a clear event, it may be due to
2687 * a fmadm repair request. In this case, if the device is
2688 * still broken, we want to still post the ereport again.
2690 vd
->vdev_forcefault
= B_TRUE
;
2692 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2693 vd
->vdev_cant_read
= B_FALSE
;
2694 vd
->vdev_cant_write
= B_FALSE
;
2696 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2698 vd
->vdev_forcefault
= B_FALSE
;
2700 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2701 vdev_state_dirty(vd
->vdev_top
);
2703 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2704 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2706 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2710 * When clearing a FMA-diagnosed fault, we always want to
2711 * unspare the device, as we assume that the original spare was
2712 * done in response to the FMA fault.
2714 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2715 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2716 vd
->vdev_parent
->vdev_child
[0] == vd
)
2717 vd
->vdev_unspare
= B_TRUE
;
2721 vdev_is_dead(vdev_t
*vd
)
2724 * Holes and missing devices are always considered "dead".
2725 * This simplifies the code since we don't have to check for
2726 * these types of devices in the various code paths.
2727 * Instead we rely on the fact that we skip over dead devices
2728 * before issuing I/O to them.
2730 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2731 vd
->vdev_ops
== &vdev_missing_ops
);
2735 vdev_readable(vdev_t
*vd
)
2737 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2741 vdev_writeable(vdev_t
*vd
)
2743 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2747 vdev_allocatable(vdev_t
*vd
)
2749 uint64_t state
= vd
->vdev_state
;
2752 * We currently allow allocations from vdevs which may be in the
2753 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2754 * fails to reopen then we'll catch it later when we're holding
2755 * the proper locks. Note that we have to get the vdev state
2756 * in a local variable because although it changes atomically,
2757 * we're asking two separate questions about it.
2759 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2760 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2764 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2766 ASSERT(zio
->io_vd
== vd
);
2768 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2771 if (zio
->io_type
== ZIO_TYPE_READ
)
2772 return (!vd
->vdev_cant_read
);
2774 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2775 return (!vd
->vdev_cant_write
);
2781 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2784 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2785 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2786 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2789 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2793 * Get extended stats
2796 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2799 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2800 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2801 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2803 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2804 vsx
->vsx_total_histo
[t
][b
] +=
2805 cvsx
->vsx_total_histo
[t
][b
];
2809 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2810 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2811 vsx
->vsx_queue_histo
[t
][b
] +=
2812 cvsx
->vsx_queue_histo
[t
][b
];
2814 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2815 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2817 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2818 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2820 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2821 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2827 * Get statistics for the given vdev.
2830 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2834 * If we're getting stats on the root vdev, aggregate the I/O counts
2835 * over all top-level vdevs (i.e. the direct children of the root).
2837 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2839 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2840 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2843 memset(vsx
, 0, sizeof (*vsx
));
2845 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2846 vdev_t
*cvd
= vd
->vdev_child
[c
];
2847 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2848 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2850 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2852 vdev_get_child_stat(cvd
, vs
, cvs
);
2854 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2859 * We're a leaf. Just copy our ZIO active queue stats in. The
2860 * other leaf stats are updated in vdev_stat_update().
2865 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2867 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2868 vsx
->vsx_active_queue
[t
] =
2869 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2870 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2871 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2877 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2879 mutex_enter(&vd
->vdev_stat_lock
);
2881 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2882 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2883 vs
->vs_state
= vd
->vdev_state
;
2884 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2885 if (vd
->vdev_ops
->vdev_op_leaf
)
2886 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
2887 VDEV_LABEL_END_SIZE
;
2888 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2889 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
2891 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2895 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
2896 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
2897 mutex_exit(&vd
->vdev_stat_lock
);
2901 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2903 return (vdev_get_stats_ex(vd
, vs
, NULL
));
2907 vdev_clear_stats(vdev_t
*vd
)
2909 mutex_enter(&vd
->vdev_stat_lock
);
2910 vd
->vdev_stat
.vs_space
= 0;
2911 vd
->vdev_stat
.vs_dspace
= 0;
2912 vd
->vdev_stat
.vs_alloc
= 0;
2913 mutex_exit(&vd
->vdev_stat_lock
);
2917 vdev_scan_stat_init(vdev_t
*vd
)
2919 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2922 for (c
= 0; c
< vd
->vdev_children
; c
++)
2923 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2925 mutex_enter(&vd
->vdev_stat_lock
);
2926 vs
->vs_scan_processed
= 0;
2927 mutex_exit(&vd
->vdev_stat_lock
);
2931 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2933 spa_t
*spa
= zio
->io_spa
;
2934 vdev_t
*rvd
= spa
->spa_root_vdev
;
2935 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2937 uint64_t txg
= zio
->io_txg
;
2938 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2939 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
2940 zio_type_t type
= zio
->io_type
;
2941 int flags
= zio
->io_flags
;
2944 * If this i/o is a gang leader, it didn't do any actual work.
2946 if (zio
->io_gang_tree
)
2949 if (zio
->io_error
== 0) {
2951 * If this is a root i/o, don't count it -- we've already
2952 * counted the top-level vdevs, and vdev_get_stats() will
2953 * aggregate them when asked. This reduces contention on
2954 * the root vdev_stat_lock and implicitly handles blocks
2955 * that compress away to holes, for which there is no i/o.
2956 * (Holes never create vdev children, so all the counters
2957 * remain zero, which is what we want.)
2959 * Note: this only applies to successful i/o (io_error == 0)
2960 * because unlike i/o counts, errors are not additive.
2961 * When reading a ditto block, for example, failure of
2962 * one top-level vdev does not imply a root-level error.
2967 ASSERT(vd
== zio
->io_vd
);
2969 if (flags
& ZIO_FLAG_IO_BYPASS
)
2972 mutex_enter(&vd
->vdev_stat_lock
);
2974 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2975 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2976 dsl_scan_phys_t
*scn_phys
=
2977 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2978 uint64_t *processed
= &scn_phys
->scn_processed
;
2981 if (vd
->vdev_ops
->vdev_op_leaf
)
2982 atomic_add_64(processed
, psize
);
2983 vs
->vs_scan_processed
+= psize
;
2986 if (flags
& ZIO_FLAG_SELF_HEAL
)
2987 vs
->vs_self_healed
+= psize
;
2991 * The bytes/ops/histograms are recorded at the leaf level and
2992 * aggregated into the higher level vdevs in vdev_get_stats().
2994 if (vd
->vdev_ops
->vdev_op_leaf
&&
2995 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
2998 vs
->vs_bytes
[type
] += psize
;
3000 if (flags
& ZIO_FLAG_DELEGATED
) {
3001 vsx
->vsx_agg_histo
[zio
->io_priority
]
3002 [RQ_HISTO(zio
->io_size
)]++;
3004 vsx
->vsx_ind_histo
[zio
->io_priority
]
3005 [RQ_HISTO(zio
->io_size
)]++;
3008 if (zio
->io_delta
&& zio
->io_delay
) {
3009 vsx
->vsx_queue_histo
[zio
->io_priority
]
3010 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3011 vsx
->vsx_disk_histo
[type
]
3012 [L_HISTO(zio
->io_delay
)]++;
3013 vsx
->vsx_total_histo
[type
]
3014 [L_HISTO(zio
->io_delta
)]++;
3018 mutex_exit(&vd
->vdev_stat_lock
);
3022 if (flags
& ZIO_FLAG_SPECULATIVE
)
3026 * If this is an I/O error that is going to be retried, then ignore the
3027 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3028 * hard errors, when in reality they can happen for any number of
3029 * innocuous reasons (bus resets, MPxIO link failure, etc).
3031 if (zio
->io_error
== EIO
&&
3032 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3036 * Intent logs writes won't propagate their error to the root
3037 * I/O so don't mark these types of failures as pool-level
3040 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3043 mutex_enter(&vd
->vdev_stat_lock
);
3044 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3045 if (zio
->io_error
== ECKSUM
)
3046 vs
->vs_checksum_errors
++;
3048 vs
->vs_read_errors
++;
3050 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3051 vs
->vs_write_errors
++;
3052 mutex_exit(&vd
->vdev_stat_lock
);
3054 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3055 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3056 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3057 spa
->spa_claiming
)) {
3059 * This is either a normal write (not a repair), or it's
3060 * a repair induced by the scrub thread, or it's a repair
3061 * made by zil_claim() during spa_load() in the first txg.
3062 * In the normal case, we commit the DTL change in the same
3063 * txg as the block was born. In the scrub-induced repair
3064 * case, we know that scrubs run in first-pass syncing context,
3065 * so we commit the DTL change in spa_syncing_txg(spa).
3066 * In the zil_claim() case, we commit in spa_first_txg(spa).
3068 * We currently do not make DTL entries for failed spontaneous
3069 * self-healing writes triggered by normal (non-scrubbing)
3070 * reads, because we have no transactional context in which to
3071 * do so -- and it's not clear that it'd be desirable anyway.
3073 if (vd
->vdev_ops
->vdev_op_leaf
) {
3074 uint64_t commit_txg
= txg
;
3075 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3076 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3077 ASSERT(spa_sync_pass(spa
) == 1);
3078 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3079 commit_txg
= spa_syncing_txg(spa
);
3080 } else if (spa
->spa_claiming
) {
3081 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3082 commit_txg
= spa_first_txg(spa
);
3084 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3085 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3087 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3088 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3089 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3092 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3097 * Update the in-core space usage stats for this vdev, its metaslab class,
3098 * and the root vdev.
3101 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3102 int64_t space_delta
)
3104 int64_t dspace_delta
= space_delta
;
3105 spa_t
*spa
= vd
->vdev_spa
;
3106 vdev_t
*rvd
= spa
->spa_root_vdev
;
3107 metaslab_group_t
*mg
= vd
->vdev_mg
;
3108 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3110 ASSERT(vd
== vd
->vdev_top
);
3113 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3114 * factor. We must calculate this here and not at the root vdev
3115 * because the root vdev's psize-to-asize is simply the max of its
3116 * childrens', thus not accurate enough for us.
3118 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3119 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3120 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3121 vd
->vdev_deflate_ratio
;
3123 mutex_enter(&vd
->vdev_stat_lock
);
3124 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3125 vd
->vdev_stat
.vs_space
+= space_delta
;
3126 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3127 mutex_exit(&vd
->vdev_stat_lock
);
3129 if (mc
== spa_normal_class(spa
)) {
3130 mutex_enter(&rvd
->vdev_stat_lock
);
3131 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3132 rvd
->vdev_stat
.vs_space
+= space_delta
;
3133 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3134 mutex_exit(&rvd
->vdev_stat_lock
);
3138 ASSERT(rvd
== vd
->vdev_parent
);
3139 ASSERT(vd
->vdev_ms_count
!= 0);
3141 metaslab_class_space_update(mc
,
3142 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3147 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3148 * so that it will be written out next time the vdev configuration is synced.
3149 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3152 vdev_config_dirty(vdev_t
*vd
)
3154 spa_t
*spa
= vd
->vdev_spa
;
3155 vdev_t
*rvd
= spa
->spa_root_vdev
;
3158 ASSERT(spa_writeable(spa
));
3161 * If this is an aux vdev (as with l2cache and spare devices), then we
3162 * update the vdev config manually and set the sync flag.
3164 if (vd
->vdev_aux
!= NULL
) {
3165 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3169 for (c
= 0; c
< sav
->sav_count
; c
++) {
3170 if (sav
->sav_vdevs
[c
] == vd
)
3174 if (c
== sav
->sav_count
) {
3176 * We're being removed. There's nothing more to do.
3178 ASSERT(sav
->sav_sync
== B_TRUE
);
3182 sav
->sav_sync
= B_TRUE
;
3184 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3185 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3186 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3187 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3193 * Setting the nvlist in the middle if the array is a little
3194 * sketchy, but it will work.
3196 nvlist_free(aux
[c
]);
3197 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3203 * The dirty list is protected by the SCL_CONFIG lock. The caller
3204 * must either hold SCL_CONFIG as writer, or must be the sync thread
3205 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3206 * so this is sufficient to ensure mutual exclusion.
3208 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3209 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3210 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3213 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3214 vdev_config_dirty(rvd
->vdev_child
[c
]);
3216 ASSERT(vd
== vd
->vdev_top
);
3218 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3220 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3225 vdev_config_clean(vdev_t
*vd
)
3227 spa_t
*spa
= vd
->vdev_spa
;
3229 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3230 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3231 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3233 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3234 list_remove(&spa
->spa_config_dirty_list
, vd
);
3238 * Mark a top-level vdev's state as dirty, so that the next pass of
3239 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3240 * the state changes from larger config changes because they require
3241 * much less locking, and are often needed for administrative actions.
3244 vdev_state_dirty(vdev_t
*vd
)
3246 spa_t
*spa
= vd
->vdev_spa
;
3248 ASSERT(spa_writeable(spa
));
3249 ASSERT(vd
== vd
->vdev_top
);
3252 * The state list is protected by the SCL_STATE lock. The caller
3253 * must either hold SCL_STATE as writer, or must be the sync thread
3254 * (which holds SCL_STATE as reader). There's only one sync thread,
3255 * so this is sufficient to ensure mutual exclusion.
3257 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3258 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3259 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3261 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3262 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3266 vdev_state_clean(vdev_t
*vd
)
3268 spa_t
*spa
= vd
->vdev_spa
;
3270 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3271 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3272 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3274 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3275 list_remove(&spa
->spa_state_dirty_list
, vd
);
3279 * Propagate vdev state up from children to parent.
3282 vdev_propagate_state(vdev_t
*vd
)
3284 spa_t
*spa
= vd
->vdev_spa
;
3285 vdev_t
*rvd
= spa
->spa_root_vdev
;
3286 int degraded
= 0, faulted
= 0;
3291 if (vd
->vdev_children
> 0) {
3292 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3293 child
= vd
->vdev_child
[c
];
3296 * Don't factor holes into the decision.
3298 if (child
->vdev_ishole
)
3301 if (!vdev_readable(child
) ||
3302 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3304 * Root special: if there is a top-level log
3305 * device, treat the root vdev as if it were
3308 if (child
->vdev_islog
&& vd
== rvd
)
3312 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3316 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3320 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3323 * Root special: if there is a top-level vdev that cannot be
3324 * opened due to corrupted metadata, then propagate the root
3325 * vdev's aux state as 'corrupt' rather than 'insufficient
3328 if (corrupted
&& vd
== rvd
&&
3329 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3330 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3331 VDEV_AUX_CORRUPT_DATA
);
3334 if (vd
->vdev_parent
)
3335 vdev_propagate_state(vd
->vdev_parent
);
3339 * Set a vdev's state. If this is during an open, we don't update the parent
3340 * state, because we're in the process of opening children depth-first.
3341 * Otherwise, we propagate the change to the parent.
3343 * If this routine places a device in a faulted state, an appropriate ereport is
3347 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3349 uint64_t save_state
;
3350 spa_t
*spa
= vd
->vdev_spa
;
3352 if (state
== vd
->vdev_state
) {
3353 vd
->vdev_stat
.vs_aux
= aux
;
3357 save_state
= vd
->vdev_state
;
3359 vd
->vdev_state
= state
;
3360 vd
->vdev_stat
.vs_aux
= aux
;
3363 * If we are setting the vdev state to anything but an open state, then
3364 * always close the underlying device unless the device has requested
3365 * a delayed close (i.e. we're about to remove or fault the device).
3366 * Otherwise, we keep accessible but invalid devices open forever.
3367 * We don't call vdev_close() itself, because that implies some extra
3368 * checks (offline, etc) that we don't want here. This is limited to
3369 * leaf devices, because otherwise closing the device will affect other
3372 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3373 vd
->vdev_ops
->vdev_op_leaf
)
3374 vd
->vdev_ops
->vdev_op_close(vd
);
3377 * If we have brought this vdev back into service, we need
3378 * to notify fmd so that it can gracefully repair any outstanding
3379 * cases due to a missing device. We do this in all cases, even those
3380 * that probably don't correlate to a repaired fault. This is sure to
3381 * catch all cases, and we let the zfs-retire agent sort it out. If
3382 * this is a transient state it's OK, as the retire agent will
3383 * double-check the state of the vdev before repairing it.
3385 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
3386 vd
->vdev_prevstate
!= state
)
3387 zfs_post_state_change(spa
, vd
);
3389 if (vd
->vdev_removed
&&
3390 state
== VDEV_STATE_CANT_OPEN
&&
3391 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3393 * If the previous state is set to VDEV_STATE_REMOVED, then this
3394 * device was previously marked removed and someone attempted to
3395 * reopen it. If this failed due to a nonexistent device, then
3396 * keep the device in the REMOVED state. We also let this be if
3397 * it is one of our special test online cases, which is only
3398 * attempting to online the device and shouldn't generate an FMA
3401 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3402 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3403 } else if (state
== VDEV_STATE_REMOVED
) {
3404 vd
->vdev_removed
= B_TRUE
;
3405 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3407 * If we fail to open a vdev during an import or recovery, we
3408 * mark it as "not available", which signifies that it was
3409 * never there to begin with. Failure to open such a device
3410 * is not considered an error.
3412 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3413 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3414 vd
->vdev_ops
->vdev_op_leaf
)
3415 vd
->vdev_not_present
= 1;
3418 * Post the appropriate ereport. If the 'prevstate' field is
3419 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3420 * that this is part of a vdev_reopen(). In this case, we don't
3421 * want to post the ereport if the device was already in the
3422 * CANT_OPEN state beforehand.
3424 * If the 'checkremove' flag is set, then this is an attempt to
3425 * online the device in response to an insertion event. If we
3426 * hit this case, then we have detected an insertion event for a
3427 * faulted or offline device that wasn't in the removed state.
3428 * In this scenario, we don't post an ereport because we are
3429 * about to replace the device, or attempt an online with
3430 * vdev_forcefault, which will generate the fault for us.
3432 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3433 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3434 vd
!= spa
->spa_root_vdev
) {
3438 case VDEV_AUX_OPEN_FAILED
:
3439 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3441 case VDEV_AUX_CORRUPT_DATA
:
3442 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3444 case VDEV_AUX_NO_REPLICAS
:
3445 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3447 case VDEV_AUX_BAD_GUID_SUM
:
3448 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3450 case VDEV_AUX_TOO_SMALL
:
3451 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3453 case VDEV_AUX_BAD_LABEL
:
3454 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3457 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3460 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3463 /* Erase any notion of persistent removed state */
3464 vd
->vdev_removed
= B_FALSE
;
3466 vd
->vdev_removed
= B_FALSE
;
3469 if (!isopen
&& vd
->vdev_parent
)
3470 vdev_propagate_state(vd
->vdev_parent
);
3474 * Check the vdev configuration to ensure that it's capable of supporting
3478 vdev_is_bootable(vdev_t
*vd
)
3480 #if defined(__sun__) || defined(__sun)
3482 * Currently, we do not support RAID-Z or partial configuration.
3483 * In addition, only a single top-level vdev is allowed and none of the
3484 * leaves can be wholedisks.
3488 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3489 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3491 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3492 vd
->vdev_children
> 1) {
3494 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3495 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3498 } else if (vd
->vdev_wholedisk
== 1) {
3502 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3503 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3506 #endif /* __sun__ || __sun */
3511 * Load the state from the original vdev tree (ovd) which
3512 * we've retrieved from the MOS config object. If the original
3513 * vdev was offline or faulted then we transfer that state to the
3514 * device in the current vdev tree (nvd).
3517 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3521 ASSERT(nvd
->vdev_top
->vdev_islog
);
3522 ASSERT(spa_config_held(nvd
->vdev_spa
,
3523 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3524 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3526 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3527 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3529 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3531 * Restore the persistent vdev state
3533 nvd
->vdev_offline
= ovd
->vdev_offline
;
3534 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3535 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3536 nvd
->vdev_removed
= ovd
->vdev_removed
;
3541 * Determine if a log device has valid content. If the vdev was
3542 * removed or faulted in the MOS config then we know that
3543 * the content on the log device has already been written to the pool.
3546 vdev_log_state_valid(vdev_t
*vd
)
3550 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3554 for (c
= 0; c
< vd
->vdev_children
; c
++)
3555 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3562 * Expand a vdev if possible.
3565 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3567 ASSERT(vd
->vdev_top
== vd
);
3568 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3570 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3571 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3572 vdev_config_dirty(vd
);
3580 vdev_split(vdev_t
*vd
)
3582 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3584 vdev_remove_child(pvd
, vd
);
3585 vdev_compact_children(pvd
);
3587 cvd
= pvd
->vdev_child
[0];
3588 if (pvd
->vdev_children
== 1) {
3589 vdev_remove_parent(cvd
);
3590 cvd
->vdev_splitting
= B_TRUE
;
3592 vdev_propagate_state(cvd
);
3596 vdev_deadman(vdev_t
*vd
)
3600 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3601 vdev_t
*cvd
= vd
->vdev_child
[c
];
3606 if (vd
->vdev_ops
->vdev_op_leaf
) {
3607 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3609 mutex_enter(&vq
->vq_lock
);
3610 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3611 spa_t
*spa
= vd
->vdev_spa
;
3616 * Look at the head of all the pending queues,
3617 * if any I/O has been outstanding for longer than
3618 * the spa_deadman_synctime we log a zevent.
3620 fio
= avl_first(&vq
->vq_active_tree
);
3621 delta
= gethrtime() - fio
->io_timestamp
;
3622 if (delta
> spa_deadman_synctime(spa
)) {
3623 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3624 "delta %lluns, last io %lluns",
3625 fio
->io_timestamp
, delta
,
3626 vq
->vq_io_complete_ts
);
3627 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3628 spa
, vd
, fio
, 0, 0);
3631 mutex_exit(&vq
->vq_lock
);
3635 #if defined(_KERNEL) && defined(HAVE_SPL)
3636 EXPORT_SYMBOL(vdev_fault
);
3637 EXPORT_SYMBOL(vdev_degrade
);
3638 EXPORT_SYMBOL(vdev_online
);
3639 EXPORT_SYMBOL(vdev_offline
);
3640 EXPORT_SYMBOL(vdev_clear
);
3642 module_param(metaslabs_per_vdev
, int, 0644);
3643 MODULE_PARM_DESC(metaslabs_per_vdev
,
3644 "Divide added vdev into approximately (but no more than) this number "