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) 2013 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>
41 #include <sys/fs/zfs.h>
44 #include <sys/dsl_scan.h>
48 * Virtual device management.
51 static vdev_ops_t
*vdev_ops_table
[] = {
65 * Given a vdev type, return the appropriate ops vector.
68 vdev_getops(const char *type
)
70 vdev_ops_t
*ops
, **opspp
;
72 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
73 if (strcmp(ops
->vdev_op_type
, type
) == 0)
80 * Default asize function: return the MAX of psize with the asize of
81 * all children. This is what's used by anything other than RAID-Z.
84 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
86 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
90 for (c
= 0; c
< vd
->vdev_children
; c
++) {
91 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
92 asize
= MAX(asize
, csize
);
99 * Get the minimum allocatable size. We define the allocatable size as
100 * the vdev's asize rounded to the nearest metaslab. This allows us to
101 * replace or attach devices which don't have the same physical size but
102 * can still satisfy the same number of allocations.
105 vdev_get_min_asize(vdev_t
*vd
)
107 vdev_t
*pvd
= vd
->vdev_parent
;
110 * If our parent is NULL (inactive spare or cache) or is the root,
111 * just return our own asize.
114 return (vd
->vdev_asize
);
117 * The top-level vdev just returns the allocatable size rounded
118 * to the nearest metaslab.
120 if (vd
== vd
->vdev_top
)
121 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
124 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
125 * so each child must provide at least 1/Nth of its asize.
127 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
128 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
130 return (pvd
->vdev_min_asize
);
134 vdev_set_min_asize(vdev_t
*vd
)
137 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
139 for (c
= 0; c
< vd
->vdev_children
; c
++)
140 vdev_set_min_asize(vd
->vdev_child
[c
]);
144 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
146 vdev_t
*rvd
= spa
->spa_root_vdev
;
148 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
150 if (vdev
< rvd
->vdev_children
) {
151 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
152 return (rvd
->vdev_child
[vdev
]);
159 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
164 if (vd
->vdev_guid
== guid
)
167 for (c
= 0; c
< vd
->vdev_children
; c
++)
168 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
176 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
178 size_t oldsize
, newsize
;
179 uint64_t id
= cvd
->vdev_id
;
182 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
183 ASSERT(cvd
->vdev_parent
== NULL
);
185 cvd
->vdev_parent
= pvd
;
190 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
192 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
193 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
194 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
196 newchild
= kmem_zalloc(newsize
, KM_PUSHPAGE
);
197 if (pvd
->vdev_child
!= NULL
) {
198 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
199 kmem_free(pvd
->vdev_child
, oldsize
);
202 pvd
->vdev_child
= newchild
;
203 pvd
->vdev_child
[id
] = cvd
;
205 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
206 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
209 * Walk up all ancestors to update guid sum.
211 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
212 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
216 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
219 uint_t id
= cvd
->vdev_id
;
221 ASSERT(cvd
->vdev_parent
== pvd
);
226 ASSERT(id
< pvd
->vdev_children
);
227 ASSERT(pvd
->vdev_child
[id
] == cvd
);
229 pvd
->vdev_child
[id
] = NULL
;
230 cvd
->vdev_parent
= NULL
;
232 for (c
= 0; c
< pvd
->vdev_children
; c
++)
233 if (pvd
->vdev_child
[c
])
236 if (c
== pvd
->vdev_children
) {
237 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
238 pvd
->vdev_child
= NULL
;
239 pvd
->vdev_children
= 0;
243 * Walk up all ancestors to update guid sum.
245 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
246 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
250 * Remove any holes in the child array.
253 vdev_compact_children(vdev_t
*pvd
)
255 vdev_t
**newchild
, *cvd
;
256 int oldc
= pvd
->vdev_children
;
260 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
262 for (c
= newc
= 0; c
< oldc
; c
++)
263 if (pvd
->vdev_child
[c
])
266 newchild
= kmem_alloc(newc
* sizeof (vdev_t
*), KM_PUSHPAGE
);
268 for (c
= newc
= 0; c
< oldc
; c
++) {
269 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
270 newchild
[newc
] = cvd
;
271 cvd
->vdev_id
= newc
++;
275 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
276 pvd
->vdev_child
= newchild
;
277 pvd
->vdev_children
= newc
;
281 * Allocate and minimally initialize a vdev_t.
284 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
289 vd
= kmem_zalloc(sizeof (vdev_t
), KM_PUSHPAGE
);
291 if (spa
->spa_root_vdev
== NULL
) {
292 ASSERT(ops
== &vdev_root_ops
);
293 spa
->spa_root_vdev
= vd
;
294 spa
->spa_load_guid
= spa_generate_guid(NULL
);
297 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
298 if (spa
->spa_root_vdev
== vd
) {
300 * The root vdev's guid will also be the pool guid,
301 * which must be unique among all pools.
303 guid
= spa_generate_guid(NULL
);
306 * Any other vdev's guid must be unique within the pool.
308 guid
= spa_generate_guid(spa
);
310 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
315 vd
->vdev_guid
= guid
;
316 vd
->vdev_guid_sum
= guid
;
318 vd
->vdev_state
= VDEV_STATE_CLOSED
;
319 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
321 list_link_init(&vd
->vdev_config_dirty_node
);
322 list_link_init(&vd
->vdev_state_dirty_node
);
323 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
324 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
325 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
326 for (t
= 0; t
< DTL_TYPES
; t
++) {
327 space_map_create(&vd
->vdev_dtl
[t
], 0, -1ULL, 0,
330 txg_list_create(&vd
->vdev_ms_list
,
331 offsetof(struct metaslab
, ms_txg_node
));
332 txg_list_create(&vd
->vdev_dtl_list
,
333 offsetof(struct vdev
, vdev_dtl_node
));
334 vd
->vdev_stat
.vs_timestamp
= gethrtime();
342 * Allocate a new vdev. The 'alloctype' is used to control whether we are
343 * creating a new vdev or loading an existing one - the behavior is slightly
344 * different for each case.
347 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
352 uint64_t guid
= 0, islog
, nparity
;
355 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
357 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
358 return (SET_ERROR(EINVAL
));
360 if ((ops
= vdev_getops(type
)) == NULL
)
361 return (SET_ERROR(EINVAL
));
364 * If this is a load, get the vdev guid from the nvlist.
365 * Otherwise, vdev_alloc_common() will generate one for us.
367 if (alloctype
== VDEV_ALLOC_LOAD
) {
370 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
372 return (SET_ERROR(EINVAL
));
374 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
375 return (SET_ERROR(EINVAL
));
376 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
377 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
378 return (SET_ERROR(EINVAL
));
379 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
380 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
381 return (SET_ERROR(EINVAL
));
382 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
383 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
384 return (SET_ERROR(EINVAL
));
388 * The first allocated vdev must be of type 'root'.
390 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
391 return (SET_ERROR(EINVAL
));
394 * Determine whether we're a log vdev.
397 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
398 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
399 return (SET_ERROR(ENOTSUP
));
401 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
402 return (SET_ERROR(ENOTSUP
));
405 * Set the nparity property for RAID-Z vdevs.
408 if (ops
== &vdev_raidz_ops
) {
409 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
411 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
412 return (SET_ERROR(EINVAL
));
414 * Previous versions could only support 1 or 2 parity
418 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
419 return (SET_ERROR(ENOTSUP
));
421 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
422 return (SET_ERROR(ENOTSUP
));
425 * We require the parity to be specified for SPAs that
426 * support multiple parity levels.
428 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
429 return (SET_ERROR(EINVAL
));
431 * Otherwise, we default to 1 parity device for RAID-Z.
438 ASSERT(nparity
!= -1ULL);
440 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
442 vd
->vdev_islog
= islog
;
443 vd
->vdev_nparity
= nparity
;
445 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
446 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
447 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
448 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
449 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
450 &vd
->vdev_physpath
) == 0)
451 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
452 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
453 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
456 * Set the whole_disk property. If it's not specified, leave the value
459 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
460 &vd
->vdev_wholedisk
) != 0)
461 vd
->vdev_wholedisk
= -1ULL;
464 * Look for the 'not present' flag. This will only be set if the device
465 * was not present at the time of import.
467 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
468 &vd
->vdev_not_present
);
471 * Get the alignment requirement.
473 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
476 * Retrieve the vdev creation time.
478 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
482 * If we're a top-level vdev, try to load the allocation parameters.
484 if (parent
&& !parent
->vdev_parent
&&
485 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
486 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
488 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
490 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
492 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
496 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
497 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
498 alloctype
== VDEV_ALLOC_ADD
||
499 alloctype
== VDEV_ALLOC_SPLIT
||
500 alloctype
== VDEV_ALLOC_ROOTPOOL
);
501 vd
->vdev_mg
= metaslab_group_create(islog
?
502 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
506 * If we're a leaf vdev, try to load the DTL object and other state.
508 if (vd
->vdev_ops
->vdev_op_leaf
&&
509 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
510 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
511 if (alloctype
== VDEV_ALLOC_LOAD
) {
512 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
513 &vd
->vdev_dtl_smo
.smo_object
);
514 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
518 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
521 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
522 &spare
) == 0 && spare
)
526 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
529 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
530 &vd
->vdev_resilver_txg
);
533 * When importing a pool, we want to ignore the persistent fault
534 * state, as the diagnosis made on another system may not be
535 * valid in the current context. Local vdevs will
536 * remain in the faulted state.
538 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
539 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
541 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
543 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
546 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
550 VDEV_AUX_ERR_EXCEEDED
;
551 if (nvlist_lookup_string(nv
,
552 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
553 strcmp(aux
, "external") == 0)
554 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
560 * Add ourselves to the parent's list of children.
562 vdev_add_child(parent
, vd
);
570 vdev_free(vdev_t
*vd
)
573 spa_t
*spa
= vd
->vdev_spa
;
576 * vdev_free() implies closing the vdev first. This is simpler than
577 * trying to ensure complicated semantics for all callers.
581 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
582 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
587 for (c
= 0; c
< vd
->vdev_children
; c
++)
588 vdev_free(vd
->vdev_child
[c
]);
590 ASSERT(vd
->vdev_child
== NULL
);
591 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
594 * Discard allocation state.
596 if (vd
->vdev_mg
!= NULL
) {
597 vdev_metaslab_fini(vd
);
598 metaslab_group_destroy(vd
->vdev_mg
);
601 ASSERT0(vd
->vdev_stat
.vs_space
);
602 ASSERT0(vd
->vdev_stat
.vs_dspace
);
603 ASSERT0(vd
->vdev_stat
.vs_alloc
);
606 * Remove this vdev from its parent's child list.
608 vdev_remove_child(vd
->vdev_parent
, vd
);
610 ASSERT(vd
->vdev_parent
== NULL
);
613 * Clean up vdev structure.
619 spa_strfree(vd
->vdev_path
);
621 spa_strfree(vd
->vdev_devid
);
622 if (vd
->vdev_physpath
)
623 spa_strfree(vd
->vdev_physpath
);
625 spa_strfree(vd
->vdev_fru
);
627 if (vd
->vdev_isspare
)
628 spa_spare_remove(vd
);
629 if (vd
->vdev_isl2cache
)
630 spa_l2cache_remove(vd
);
632 txg_list_destroy(&vd
->vdev_ms_list
);
633 txg_list_destroy(&vd
->vdev_dtl_list
);
635 mutex_enter(&vd
->vdev_dtl_lock
);
636 for (t
= 0; t
< DTL_TYPES
; t
++) {
637 space_map_unload(&vd
->vdev_dtl
[t
]);
638 space_map_destroy(&vd
->vdev_dtl
[t
]);
640 mutex_exit(&vd
->vdev_dtl_lock
);
642 mutex_destroy(&vd
->vdev_dtl_lock
);
643 mutex_destroy(&vd
->vdev_stat_lock
);
644 mutex_destroy(&vd
->vdev_probe_lock
);
646 if (vd
== spa
->spa_root_vdev
)
647 spa
->spa_root_vdev
= NULL
;
649 kmem_free(vd
, sizeof (vdev_t
));
653 * Transfer top-level vdev state from svd to tvd.
656 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
658 spa_t
*spa
= svd
->vdev_spa
;
663 ASSERT(tvd
== tvd
->vdev_top
);
665 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
666 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
667 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
669 svd
->vdev_ms_array
= 0;
670 svd
->vdev_ms_shift
= 0;
671 svd
->vdev_ms_count
= 0;
674 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
675 tvd
->vdev_mg
= svd
->vdev_mg
;
676 tvd
->vdev_ms
= svd
->vdev_ms
;
681 if (tvd
->vdev_mg
!= NULL
)
682 tvd
->vdev_mg
->mg_vd
= tvd
;
684 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
685 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
686 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
688 svd
->vdev_stat
.vs_alloc
= 0;
689 svd
->vdev_stat
.vs_space
= 0;
690 svd
->vdev_stat
.vs_dspace
= 0;
692 for (t
= 0; t
< TXG_SIZE
; t
++) {
693 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
694 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
695 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
696 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
697 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
698 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
701 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
702 vdev_config_clean(svd
);
703 vdev_config_dirty(tvd
);
706 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
707 vdev_state_clean(svd
);
708 vdev_state_dirty(tvd
);
711 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
712 svd
->vdev_deflate_ratio
= 0;
714 tvd
->vdev_islog
= svd
->vdev_islog
;
719 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
728 for (c
= 0; c
< vd
->vdev_children
; c
++)
729 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
733 * Add a mirror/replacing vdev above an existing vdev.
736 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
738 spa_t
*spa
= cvd
->vdev_spa
;
739 vdev_t
*pvd
= cvd
->vdev_parent
;
742 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
744 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
746 mvd
->vdev_asize
= cvd
->vdev_asize
;
747 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
748 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
749 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
750 mvd
->vdev_state
= cvd
->vdev_state
;
751 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
753 vdev_remove_child(pvd
, cvd
);
754 vdev_add_child(pvd
, mvd
);
755 cvd
->vdev_id
= mvd
->vdev_children
;
756 vdev_add_child(mvd
, cvd
);
757 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
759 if (mvd
== mvd
->vdev_top
)
760 vdev_top_transfer(cvd
, mvd
);
766 * Remove a 1-way mirror/replacing vdev from the tree.
769 vdev_remove_parent(vdev_t
*cvd
)
771 vdev_t
*mvd
= cvd
->vdev_parent
;
772 vdev_t
*pvd
= mvd
->vdev_parent
;
774 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
776 ASSERT(mvd
->vdev_children
== 1);
777 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
778 mvd
->vdev_ops
== &vdev_replacing_ops
||
779 mvd
->vdev_ops
== &vdev_spare_ops
);
780 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
782 vdev_remove_child(mvd
, cvd
);
783 vdev_remove_child(pvd
, mvd
);
786 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
787 * Otherwise, we could have detached an offline device, and when we
788 * go to import the pool we'll think we have two top-level vdevs,
789 * instead of a different version of the same top-level vdev.
791 if (mvd
->vdev_top
== mvd
) {
792 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
793 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
794 cvd
->vdev_guid
+= guid_delta
;
795 cvd
->vdev_guid_sum
+= guid_delta
;
797 cvd
->vdev_id
= mvd
->vdev_id
;
798 vdev_add_child(pvd
, cvd
);
799 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
801 if (cvd
== cvd
->vdev_top
)
802 vdev_top_transfer(mvd
, cvd
);
804 ASSERT(mvd
->vdev_children
== 0);
809 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
811 spa_t
*spa
= vd
->vdev_spa
;
812 objset_t
*mos
= spa
->spa_meta_objset
;
814 uint64_t oldc
= vd
->vdev_ms_count
;
815 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
819 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
822 * This vdev is not being allocated from yet or is a hole.
824 if (vd
->vdev_ms_shift
== 0)
827 ASSERT(!vd
->vdev_ishole
);
830 * Compute the raidz-deflation ratio. Note, we hard-code
831 * in 128k (1 << 17) because it is the current "typical" blocksize.
832 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
833 * or we will inconsistently account for existing bp's.
835 vd
->vdev_deflate_ratio
= (1 << 17) /
836 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
838 ASSERT(oldc
<= newc
);
840 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_PUSHPAGE
| KM_NODEBUG
);
843 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
844 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
848 vd
->vdev_ms_count
= newc
;
850 for (m
= oldc
; m
< newc
; m
++) {
851 space_map_obj_t smo
= { 0, 0, 0 };
854 error
= dmu_read(mos
, vd
->vdev_ms_array
,
855 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
861 error
= dmu_bonus_hold(mos
, object
, FTAG
, &db
);
864 ASSERT3U(db
->db_size
, >=, sizeof (smo
));
865 bcopy(db
->db_data
, &smo
, sizeof (smo
));
866 ASSERT3U(smo
.smo_object
, ==, object
);
867 dmu_buf_rele(db
, FTAG
);
870 vd
->vdev_ms
[m
] = metaslab_init(vd
->vdev_mg
, &smo
,
871 m
<< vd
->vdev_ms_shift
, 1ULL << vd
->vdev_ms_shift
, txg
);
875 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
878 * If the vdev is being removed we don't activate
879 * the metaslabs since we want to ensure that no new
880 * allocations are performed on this device.
882 if (oldc
== 0 && !vd
->vdev_removing
)
883 metaslab_group_activate(vd
->vdev_mg
);
886 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
892 vdev_metaslab_fini(vdev_t
*vd
)
895 uint64_t count
= vd
->vdev_ms_count
;
897 if (vd
->vdev_ms
!= NULL
) {
898 metaslab_group_passivate(vd
->vdev_mg
);
899 for (m
= 0; m
< count
; m
++)
900 if (vd
->vdev_ms
[m
] != NULL
)
901 metaslab_fini(vd
->vdev_ms
[m
]);
902 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
906 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
909 typedef struct vdev_probe_stats
{
910 boolean_t vps_readable
;
911 boolean_t vps_writeable
;
913 } vdev_probe_stats_t
;
916 vdev_probe_done(zio_t
*zio
)
918 spa_t
*spa
= zio
->io_spa
;
919 vdev_t
*vd
= zio
->io_vd
;
920 vdev_probe_stats_t
*vps
= zio
->io_private
;
922 ASSERT(vd
->vdev_probe_zio
!= NULL
);
924 if (zio
->io_type
== ZIO_TYPE_READ
) {
925 if (zio
->io_error
== 0)
926 vps
->vps_readable
= 1;
927 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
928 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
929 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
930 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
931 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
933 zio_buf_free(zio
->io_data
, zio
->io_size
);
935 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
936 if (zio
->io_error
== 0)
937 vps
->vps_writeable
= 1;
938 zio_buf_free(zio
->io_data
, zio
->io_size
);
939 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
942 vd
->vdev_cant_read
|= !vps
->vps_readable
;
943 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
945 if (vdev_readable(vd
) &&
946 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
949 ASSERT(zio
->io_error
!= 0);
950 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
951 spa
, vd
, NULL
, 0, 0);
952 zio
->io_error
= SET_ERROR(ENXIO
);
955 mutex_enter(&vd
->vdev_probe_lock
);
956 ASSERT(vd
->vdev_probe_zio
== zio
);
957 vd
->vdev_probe_zio
= NULL
;
958 mutex_exit(&vd
->vdev_probe_lock
);
960 while ((pio
= zio_walk_parents(zio
)) != NULL
)
961 if (!vdev_accessible(vd
, pio
))
962 pio
->io_error
= SET_ERROR(ENXIO
);
964 kmem_free(vps
, sizeof (*vps
));
969 * Determine whether this device is accessible.
971 * Read and write to several known locations: the pad regions of each
972 * vdev label but the first, which we leave alone in case it contains
976 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
978 spa_t
*spa
= vd
->vdev_spa
;
979 vdev_probe_stats_t
*vps
= NULL
;
983 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
986 * Don't probe the probe.
988 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
992 * To prevent 'probe storms' when a device fails, we create
993 * just one probe i/o at a time. All zios that want to probe
994 * this vdev will become parents of the probe io.
996 mutex_enter(&vd
->vdev_probe_lock
);
998 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
999 vps
= kmem_zalloc(sizeof (*vps
), KM_PUSHPAGE
);
1001 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1002 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1005 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1007 * vdev_cant_read and vdev_cant_write can only
1008 * transition from TRUE to FALSE when we have the
1009 * SCL_ZIO lock as writer; otherwise they can only
1010 * transition from FALSE to TRUE. This ensures that
1011 * any zio looking at these values can assume that
1012 * failures persist for the life of the I/O. That's
1013 * important because when a device has intermittent
1014 * connectivity problems, we want to ensure that
1015 * they're ascribed to the device (ENXIO) and not
1018 * Since we hold SCL_ZIO as writer here, clear both
1019 * values so the probe can reevaluate from first
1022 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1023 vd
->vdev_cant_read
= B_FALSE
;
1024 vd
->vdev_cant_write
= B_FALSE
;
1027 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1028 vdev_probe_done
, vps
,
1029 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1032 * We can't change the vdev state in this context, so we
1033 * kick off an async task to do it on our behalf.
1036 vd
->vdev_probe_wanted
= B_TRUE
;
1037 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1042 zio_add_child(zio
, pio
);
1044 mutex_exit(&vd
->vdev_probe_lock
);
1047 ASSERT(zio
!= NULL
);
1051 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1052 zio_nowait(zio_read_phys(pio
, vd
,
1053 vdev_label_offset(vd
->vdev_psize
, l
,
1054 offsetof(vdev_label_t
, vl_pad2
)),
1055 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1056 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1057 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1068 vdev_open_child(void *arg
)
1072 vd
->vdev_open_thread
= curthread
;
1073 vd
->vdev_open_error
= vdev_open(vd
);
1074 vd
->vdev_open_thread
= NULL
;
1078 vdev_uses_zvols(vdev_t
*vd
)
1083 if (zvol_is_zvol(vd
->vdev_path
))
1087 for (c
= 0; c
< vd
->vdev_children
; c
++)
1088 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1095 vdev_open_children(vdev_t
*vd
)
1098 int children
= vd
->vdev_children
;
1102 * in order to handle pools on top of zvols, do the opens
1103 * in a single thread so that the same thread holds the
1104 * spa_namespace_lock
1106 if (vdev_uses_zvols(vd
)) {
1107 for (c
= 0; c
< children
; c
++)
1108 vd
->vdev_child
[c
]->vdev_open_error
=
1109 vdev_open(vd
->vdev_child
[c
]);
1112 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1113 children
, children
, TASKQ_PREPOPULATE
);
1115 for (c
= 0; c
< children
; c
++)
1116 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1123 * Prepare a virtual device for access.
1126 vdev_open(vdev_t
*vd
)
1128 spa_t
*spa
= vd
->vdev_spa
;
1131 uint64_t max_osize
= 0;
1132 uint64_t asize
, max_asize
, psize
;
1133 uint64_t ashift
= 0;
1136 ASSERT(vd
->vdev_open_thread
== curthread
||
1137 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1138 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1139 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1140 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1142 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1143 vd
->vdev_cant_read
= B_FALSE
;
1144 vd
->vdev_cant_write
= B_FALSE
;
1145 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1148 * If this vdev is not removed, check its fault status. If it's
1149 * faulted, bail out of the open.
1151 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1152 ASSERT(vd
->vdev_children
== 0);
1153 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1154 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1155 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1156 vd
->vdev_label_aux
);
1157 return (SET_ERROR(ENXIO
));
1158 } else if (vd
->vdev_offline
) {
1159 ASSERT(vd
->vdev_children
== 0);
1160 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1161 return (SET_ERROR(ENXIO
));
1164 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1167 * Reset the vdev_reopening flag so that we actually close
1168 * the vdev on error.
1170 vd
->vdev_reopening
= B_FALSE
;
1171 if (zio_injection_enabled
&& error
== 0)
1172 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1175 if (vd
->vdev_removed
&&
1176 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1177 vd
->vdev_removed
= B_FALSE
;
1179 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1180 vd
->vdev_stat
.vs_aux
);
1184 vd
->vdev_removed
= B_FALSE
;
1187 * Recheck the faulted flag now that we have confirmed that
1188 * the vdev is accessible. If we're faulted, bail.
1190 if (vd
->vdev_faulted
) {
1191 ASSERT(vd
->vdev_children
== 0);
1192 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1193 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1194 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1195 vd
->vdev_label_aux
);
1196 return (SET_ERROR(ENXIO
));
1199 if (vd
->vdev_degraded
) {
1200 ASSERT(vd
->vdev_children
== 0);
1201 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1202 VDEV_AUX_ERR_EXCEEDED
);
1204 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1208 * For hole or missing vdevs we just return success.
1210 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1213 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1214 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1215 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1221 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1222 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1224 if (vd
->vdev_children
== 0) {
1225 if (osize
< SPA_MINDEVSIZE
) {
1226 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1227 VDEV_AUX_TOO_SMALL
);
1228 return (SET_ERROR(EOVERFLOW
));
1231 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1232 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1233 VDEV_LABEL_END_SIZE
);
1235 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1236 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1237 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1238 VDEV_AUX_TOO_SMALL
);
1239 return (SET_ERROR(EOVERFLOW
));
1243 max_asize
= max_osize
;
1246 vd
->vdev_psize
= psize
;
1249 * Make sure the allocatable size hasn't shrunk.
1251 if (asize
< vd
->vdev_min_asize
) {
1252 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1253 VDEV_AUX_BAD_LABEL
);
1254 return (SET_ERROR(EINVAL
));
1257 if (vd
->vdev_asize
== 0) {
1259 * This is the first-ever open, so use the computed values.
1260 * For compatibility, a different ashift can be requested.
1262 vd
->vdev_asize
= asize
;
1263 vd
->vdev_max_asize
= max_asize
;
1264 if (vd
->vdev_ashift
== 0)
1265 vd
->vdev_ashift
= ashift
;
1268 * Detect if the alignment requirement has increased.
1269 * We don't want to make the pool unavailable, just
1270 * post an event instead.
1272 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1273 vd
->vdev_ops
->vdev_op_leaf
) {
1274 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1275 spa
, vd
, NULL
, 0, 0);
1278 vd
->vdev_max_asize
= max_asize
;
1282 * If all children are healthy and the asize has increased,
1283 * then we've experienced dynamic LUN growth. If automatic
1284 * expansion is enabled then use the additional space.
1286 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1287 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1288 vd
->vdev_asize
= asize
;
1290 vdev_set_min_asize(vd
);
1293 * Ensure we can issue some IO before declaring the
1294 * vdev open for business.
1296 if (vd
->vdev_ops
->vdev_op_leaf
&&
1297 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1298 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1299 VDEV_AUX_ERR_EXCEEDED
);
1304 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1305 * resilver. But don't do this if we are doing a reopen for a scrub,
1306 * since this would just restart the scrub we are already doing.
1308 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1309 vdev_resilver_needed(vd
, NULL
, NULL
))
1310 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1316 * Called once the vdevs are all opened, this routine validates the label
1317 * contents. This needs to be done before vdev_load() so that we don't
1318 * inadvertently do repair I/Os to the wrong device.
1320 * If 'strict' is false ignore the spa guid check. This is necessary because
1321 * if the machine crashed during a re-guid the new guid might have been written
1322 * to all of the vdev labels, but not the cached config. The strict check
1323 * will be performed when the pool is opened again using the mos config.
1325 * This function will only return failure if one of the vdevs indicates that it
1326 * has since been destroyed or exported. This is only possible if
1327 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1328 * will be updated but the function will return 0.
1331 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1333 spa_t
*spa
= vd
->vdev_spa
;
1335 uint64_t guid
= 0, top_guid
;
1339 for (c
= 0; c
< vd
->vdev_children
; c
++)
1340 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1341 return (SET_ERROR(EBADF
));
1344 * If the device has already failed, or was marked offline, don't do
1345 * any further validation. Otherwise, label I/O will fail and we will
1346 * overwrite the previous state.
1348 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1349 uint64_t aux_guid
= 0;
1351 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1352 spa_last_synced_txg(spa
) : -1ULL;
1354 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1355 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1356 VDEV_AUX_BAD_LABEL
);
1361 * Determine if this vdev has been split off into another
1362 * pool. If so, then refuse to open it.
1364 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1365 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1366 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1367 VDEV_AUX_SPLIT_POOL
);
1372 if (strict
&& (nvlist_lookup_uint64(label
,
1373 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1374 guid
!= spa_guid(spa
))) {
1375 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1376 VDEV_AUX_CORRUPT_DATA
);
1381 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1382 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1387 * If this vdev just became a top-level vdev because its
1388 * sibling was detached, it will have adopted the parent's
1389 * vdev guid -- but the label may or may not be on disk yet.
1390 * Fortunately, either version of the label will have the
1391 * same top guid, so if we're a top-level vdev, we can
1392 * safely compare to that instead.
1394 * If we split this vdev off instead, then we also check the
1395 * original pool's guid. We don't want to consider the vdev
1396 * corrupt if it is partway through a split operation.
1398 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1400 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1402 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1403 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1404 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1405 VDEV_AUX_CORRUPT_DATA
);
1410 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1412 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1413 VDEV_AUX_CORRUPT_DATA
);
1421 * If this is a verbatim import, no need to check the
1422 * state of the pool.
1424 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1425 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1426 state
!= POOL_STATE_ACTIVE
)
1427 return (SET_ERROR(EBADF
));
1430 * If we were able to open and validate a vdev that was
1431 * previously marked permanently unavailable, clear that state
1434 if (vd
->vdev_not_present
)
1435 vd
->vdev_not_present
= 0;
1442 * Close a virtual device.
1445 vdev_close(vdev_t
*vd
)
1447 vdev_t
*pvd
= vd
->vdev_parent
;
1448 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1450 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1453 * If our parent is reopening, then we are as well, unless we are
1456 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1457 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1459 vd
->vdev_ops
->vdev_op_close(vd
);
1461 vdev_cache_purge(vd
);
1464 * We record the previous state before we close it, so that if we are
1465 * doing a reopen(), we don't generate FMA ereports if we notice that
1466 * it's still faulted.
1468 vd
->vdev_prevstate
= vd
->vdev_state
;
1470 if (vd
->vdev_offline
)
1471 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1473 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1474 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1478 vdev_hold(vdev_t
*vd
)
1480 spa_t
*spa
= vd
->vdev_spa
;
1483 ASSERT(spa_is_root(spa
));
1484 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1487 for (c
= 0; c
< vd
->vdev_children
; c
++)
1488 vdev_hold(vd
->vdev_child
[c
]);
1490 if (vd
->vdev_ops
->vdev_op_leaf
)
1491 vd
->vdev_ops
->vdev_op_hold(vd
);
1495 vdev_rele(vdev_t
*vd
)
1499 ASSERT(spa_is_root(vd
->vdev_spa
));
1500 for (c
= 0; c
< vd
->vdev_children
; c
++)
1501 vdev_rele(vd
->vdev_child
[c
]);
1503 if (vd
->vdev_ops
->vdev_op_leaf
)
1504 vd
->vdev_ops
->vdev_op_rele(vd
);
1508 * Reopen all interior vdevs and any unopened leaves. We don't actually
1509 * reopen leaf vdevs which had previously been opened as they might deadlock
1510 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1511 * If the leaf has never been opened then open it, as usual.
1514 vdev_reopen(vdev_t
*vd
)
1516 spa_t
*spa
= vd
->vdev_spa
;
1518 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1520 /* set the reopening flag unless we're taking the vdev offline */
1521 vd
->vdev_reopening
= !vd
->vdev_offline
;
1523 (void) vdev_open(vd
);
1526 * Call vdev_validate() here to make sure we have the same device.
1527 * Otherwise, a device with an invalid label could be successfully
1528 * opened in response to vdev_reopen().
1531 (void) vdev_validate_aux(vd
);
1532 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1533 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1534 !l2arc_vdev_present(vd
))
1535 l2arc_add_vdev(spa
, vd
);
1537 (void) vdev_validate(vd
, B_TRUE
);
1541 * Reassess parent vdev's health.
1543 vdev_propagate_state(vd
);
1547 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1552 * Normally, partial opens (e.g. of a mirror) are allowed.
1553 * For a create, however, we want to fail the request if
1554 * there are any components we can't open.
1556 error
= vdev_open(vd
);
1558 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1560 return (error
? error
: ENXIO
);
1564 * Recursively initialize all labels.
1566 if ((error
= vdev_label_init(vd
, txg
, isreplacing
?
1567 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1576 vdev_metaslab_set_size(vdev_t
*vd
)
1579 * Aim for roughly 200 metaslabs per vdev.
1581 vd
->vdev_ms_shift
= highbit(vd
->vdev_asize
/ 200);
1582 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1586 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1588 ASSERT(vd
== vd
->vdev_top
);
1589 ASSERT(!vd
->vdev_ishole
);
1590 ASSERT(ISP2(flags
));
1591 ASSERT(spa_writeable(vd
->vdev_spa
));
1593 if (flags
& VDD_METASLAB
)
1594 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1596 if (flags
& VDD_DTL
)
1597 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1599 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1605 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1606 * the vdev has less than perfect replication. There are four kinds of DTL:
1608 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1610 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1612 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1613 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1614 * txgs that was scrubbed.
1616 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1617 * persistent errors or just some device being offline.
1618 * Unlike the other three, the DTL_OUTAGE map is not generally
1619 * maintained; it's only computed when needed, typically to
1620 * determine whether a device can be detached.
1622 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1623 * either has the data or it doesn't.
1625 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1626 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1627 * if any child is less than fully replicated, then so is its parent.
1628 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1629 * comprising only those txgs which appear in 'maxfaults' or more children;
1630 * those are the txgs we don't have enough replication to read. For example,
1631 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1632 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1633 * two child DTL_MISSING maps.
1635 * It should be clear from the above that to compute the DTLs and outage maps
1636 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1637 * Therefore, that is all we keep on disk. When loading the pool, or after
1638 * a configuration change, we generate all other DTLs from first principles.
1641 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1643 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1645 ASSERT(t
< DTL_TYPES
);
1646 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1647 ASSERT(spa_writeable(vd
->vdev_spa
));
1649 mutex_enter(sm
->sm_lock
);
1650 if (!space_map_contains(sm
, txg
, size
))
1651 space_map_add(sm
, txg
, size
);
1652 mutex_exit(sm
->sm_lock
);
1656 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1658 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1659 boolean_t dirty
= B_FALSE
;
1661 ASSERT(t
< DTL_TYPES
);
1662 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1664 mutex_enter(sm
->sm_lock
);
1665 if (sm
->sm_space
!= 0)
1666 dirty
= space_map_contains(sm
, txg
, size
);
1667 mutex_exit(sm
->sm_lock
);
1673 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1675 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1678 mutex_enter(sm
->sm_lock
);
1679 empty
= (sm
->sm_space
== 0);
1680 mutex_exit(sm
->sm_lock
);
1686 * Returns the lowest txg in the DTL range.
1689 vdev_dtl_min(vdev_t
*vd
)
1693 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1694 ASSERT3U(vd
->vdev_dtl
[DTL_MISSING
].sm_space
, !=, 0);
1695 ASSERT0(vd
->vdev_children
);
1697 ss
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1698 return (ss
->ss_start
- 1);
1702 * Returns the highest txg in the DTL.
1705 vdev_dtl_max(vdev_t
*vd
)
1709 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1710 ASSERT3U(vd
->vdev_dtl
[DTL_MISSING
].sm_space
, !=, 0);
1711 ASSERT0(vd
->vdev_children
);
1713 ss
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1714 return (ss
->ss_end
);
1718 * Determine if a resilvering vdev should remove any DTL entries from
1719 * its range. If the vdev was resilvering for the entire duration of the
1720 * scan then it should excise that range from its DTLs. Otherwise, this
1721 * vdev is considered partially resilvered and should leave its DTL
1722 * entries intact. The comment in vdev_dtl_reassess() describes how we
1726 vdev_dtl_should_excise(vdev_t
*vd
)
1728 spa_t
*spa
= vd
->vdev_spa
;
1729 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1731 ASSERT0(scn
->scn_phys
.scn_errors
);
1732 ASSERT0(vd
->vdev_children
);
1734 if (vd
->vdev_resilver_txg
== 0 ||
1735 vd
->vdev_dtl
[DTL_MISSING
].sm_space
== 0)
1739 * When a resilver is initiated the scan will assign the scn_max_txg
1740 * value to the highest txg value that exists in all DTLs. If this
1741 * device's max DTL is not part of this scan (i.e. it is not in
1742 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1745 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1746 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1747 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1748 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1755 * Reassess DTLs after a config change or scrub completion.
1758 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1760 spa_t
*spa
= vd
->vdev_spa
;
1764 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1766 for (c
= 0; c
< vd
->vdev_children
; c
++)
1767 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1768 scrub_txg
, scrub_done
);
1770 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1773 if (vd
->vdev_ops
->vdev_op_leaf
) {
1774 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1776 mutex_enter(&vd
->vdev_dtl_lock
);
1779 * If we've completed a scan cleanly then determine
1780 * if this vdev should remove any DTLs. We only want to
1781 * excise regions on vdevs that were available during
1782 * the entire duration of this scan.
1784 if (scrub_txg
!= 0 &&
1785 (spa
->spa_scrub_started
||
1786 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1787 vdev_dtl_should_excise(vd
)) {
1789 * We completed a scrub up to scrub_txg. If we
1790 * did it without rebooting, then the scrub dtl
1791 * will be valid, so excise the old region and
1792 * fold in the scrub dtl. Otherwise, leave the
1793 * dtl as-is if there was an error.
1795 * There's little trick here: to excise the beginning
1796 * of the DTL_MISSING map, we put it into a reference
1797 * tree and then add a segment with refcnt -1 that
1798 * covers the range [0, scrub_txg). This means
1799 * that each txg in that range has refcnt -1 or 0.
1800 * We then add DTL_SCRUB with a refcnt of 2, so that
1801 * entries in the range [0, scrub_txg) will have a
1802 * positive refcnt -- either 1 or 2. We then convert
1803 * the reference tree into the new DTL_MISSING map.
1805 space_map_ref_create(&reftree
);
1806 space_map_ref_add_map(&reftree
,
1807 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1808 space_map_ref_add_seg(&reftree
, 0, scrub_txg
, -1);
1809 space_map_ref_add_map(&reftree
,
1810 &vd
->vdev_dtl
[DTL_SCRUB
], 2);
1811 space_map_ref_generate_map(&reftree
,
1812 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1813 space_map_ref_destroy(&reftree
);
1815 space_map_vacate(&vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1816 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1817 space_map_add
, &vd
->vdev_dtl
[DTL_PARTIAL
]);
1819 space_map_vacate(&vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1820 space_map_vacate(&vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1821 if (!vdev_readable(vd
))
1822 space_map_add(&vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1824 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1825 space_map_add
, &vd
->vdev_dtl
[DTL_OUTAGE
]);
1828 * If the vdev was resilvering and no longer has any
1829 * DTLs then reset its resilvering flag.
1831 if (vd
->vdev_resilver_txg
!= 0 &&
1832 vd
->vdev_dtl
[DTL_MISSING
].sm_space
== 0 &&
1833 vd
->vdev_dtl
[DTL_OUTAGE
].sm_space
== 0)
1834 vd
->vdev_resilver_txg
= 0;
1836 mutex_exit(&vd
->vdev_dtl_lock
);
1839 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1843 mutex_enter(&vd
->vdev_dtl_lock
);
1844 for (t
= 0; t
< DTL_TYPES
; t
++) {
1845 /* account for child's outage in parent's missing map */
1846 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1848 continue; /* leaf vdevs only */
1849 if (t
== DTL_PARTIAL
)
1850 minref
= 1; /* i.e. non-zero */
1851 else if (vd
->vdev_nparity
!= 0)
1852 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1854 minref
= vd
->vdev_children
; /* any kind of mirror */
1855 space_map_ref_create(&reftree
);
1856 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1857 vdev_t
*cvd
= vd
->vdev_child
[c
];
1858 mutex_enter(&cvd
->vdev_dtl_lock
);
1859 space_map_ref_add_map(&reftree
, &cvd
->vdev_dtl
[s
], 1);
1860 mutex_exit(&cvd
->vdev_dtl_lock
);
1862 space_map_ref_generate_map(&reftree
, &vd
->vdev_dtl
[t
], minref
);
1863 space_map_ref_destroy(&reftree
);
1865 mutex_exit(&vd
->vdev_dtl_lock
);
1869 vdev_dtl_load(vdev_t
*vd
)
1871 spa_t
*spa
= vd
->vdev_spa
;
1872 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1873 objset_t
*mos
= spa
->spa_meta_objset
;
1877 ASSERT(vd
->vdev_children
== 0);
1879 if (smo
->smo_object
== 0)
1882 ASSERT(!vd
->vdev_ishole
);
1884 if ((error
= dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
)) != 0)
1887 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1888 bcopy(db
->db_data
, smo
, sizeof (*smo
));
1889 dmu_buf_rele(db
, FTAG
);
1891 mutex_enter(&vd
->vdev_dtl_lock
);
1892 error
= space_map_load(&vd
->vdev_dtl
[DTL_MISSING
],
1893 NULL
, SM_ALLOC
, smo
, mos
);
1894 mutex_exit(&vd
->vdev_dtl_lock
);
1900 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1902 spa_t
*spa
= vd
->vdev_spa
;
1903 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1904 space_map_t
*sm
= &vd
->vdev_dtl
[DTL_MISSING
];
1905 objset_t
*mos
= spa
->spa_meta_objset
;
1911 ASSERT(!vd
->vdev_ishole
);
1913 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1915 if (vd
->vdev_detached
) {
1916 if (smo
->smo_object
!= 0) {
1917 VERIFY0(dmu_object_free(mos
, smo
->smo_object
, tx
));
1918 smo
->smo_object
= 0;
1924 if (smo
->smo_object
== 0) {
1925 ASSERT(smo
->smo_objsize
== 0);
1926 ASSERT(smo
->smo_alloc
== 0);
1927 smo
->smo_object
= dmu_object_alloc(mos
,
1928 DMU_OT_SPACE_MAP
, 1 << SPACE_MAP_BLOCKSHIFT
,
1929 DMU_OT_SPACE_MAP_HEADER
, sizeof (*smo
), tx
);
1930 ASSERT(smo
->smo_object
!= 0);
1931 vdev_config_dirty(vd
->vdev_top
);
1934 mutex_init(&smlock
, NULL
, MUTEX_DEFAULT
, NULL
);
1936 space_map_create(&smsync
, sm
->sm_start
, sm
->sm_size
, sm
->sm_shift
,
1939 mutex_enter(&smlock
);
1941 mutex_enter(&vd
->vdev_dtl_lock
);
1942 space_map_walk(sm
, space_map_add
, &smsync
);
1943 mutex_exit(&vd
->vdev_dtl_lock
);
1945 space_map_truncate(smo
, mos
, tx
);
1946 space_map_sync(&smsync
, SM_ALLOC
, smo
, mos
, tx
);
1947 space_map_vacate(&smsync
, NULL
, NULL
);
1949 space_map_destroy(&smsync
);
1951 mutex_exit(&smlock
);
1952 mutex_destroy(&smlock
);
1954 VERIFY(0 == dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
));
1955 dmu_buf_will_dirty(db
, tx
);
1956 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1957 bcopy(smo
, db
->db_data
, sizeof (*smo
));
1958 dmu_buf_rele(db
, FTAG
);
1964 * Determine whether the specified vdev can be offlined/detached/removed
1965 * without losing data.
1968 vdev_dtl_required(vdev_t
*vd
)
1970 spa_t
*spa
= vd
->vdev_spa
;
1971 vdev_t
*tvd
= vd
->vdev_top
;
1972 uint8_t cant_read
= vd
->vdev_cant_read
;
1975 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1977 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
1981 * Temporarily mark the device as unreadable, and then determine
1982 * whether this results in any DTL outages in the top-level vdev.
1983 * If not, we can safely offline/detach/remove the device.
1985 vd
->vdev_cant_read
= B_TRUE
;
1986 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1987 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
1988 vd
->vdev_cant_read
= cant_read
;
1989 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1991 if (!required
&& zio_injection_enabled
)
1992 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
1998 * Determine if resilver is needed, and if so the txg range.
2001 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2003 boolean_t needed
= B_FALSE
;
2004 uint64_t thismin
= UINT64_MAX
;
2005 uint64_t thismax
= 0;
2008 if (vd
->vdev_children
== 0) {
2009 mutex_enter(&vd
->vdev_dtl_lock
);
2010 if (vd
->vdev_dtl
[DTL_MISSING
].sm_space
!= 0 &&
2011 vdev_writeable(vd
)) {
2013 thismin
= vdev_dtl_min(vd
);
2014 thismax
= vdev_dtl_max(vd
);
2017 mutex_exit(&vd
->vdev_dtl_lock
);
2019 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2020 vdev_t
*cvd
= vd
->vdev_child
[c
];
2021 uint64_t cmin
, cmax
;
2023 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2024 thismin
= MIN(thismin
, cmin
);
2025 thismax
= MAX(thismax
, cmax
);
2031 if (needed
&& minp
) {
2039 vdev_load(vdev_t
*vd
)
2044 * Recursively load all children.
2046 for (c
= 0; c
< vd
->vdev_children
; c
++)
2047 vdev_load(vd
->vdev_child
[c
]);
2050 * If this is a top-level vdev, initialize its metaslabs.
2052 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2053 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2054 vdev_metaslab_init(vd
, 0) != 0))
2055 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2056 VDEV_AUX_CORRUPT_DATA
);
2059 * If this is a leaf vdev, load its DTL.
2061 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2062 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2063 VDEV_AUX_CORRUPT_DATA
);
2067 * The special vdev case is used for hot spares and l2cache devices. Its
2068 * sole purpose it to set the vdev state for the associated vdev. To do this,
2069 * we make sure that we can open the underlying device, then try to read the
2070 * label, and make sure that the label is sane and that it hasn't been
2071 * repurposed to another pool.
2074 vdev_validate_aux(vdev_t
*vd
)
2077 uint64_t guid
, version
;
2080 if (!vdev_readable(vd
))
2083 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2084 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2085 VDEV_AUX_CORRUPT_DATA
);
2089 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2090 !SPA_VERSION_IS_SUPPORTED(version
) ||
2091 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2092 guid
!= vd
->vdev_guid
||
2093 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2094 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2095 VDEV_AUX_CORRUPT_DATA
);
2101 * We don't actually check the pool state here. If it's in fact in
2102 * use by another pool, we update this fact on the fly when requested.
2109 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2111 spa_t
*spa
= vd
->vdev_spa
;
2112 objset_t
*mos
= spa
->spa_meta_objset
;
2116 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2118 if (vd
->vdev_dtl_smo
.smo_object
) {
2119 ASSERT0(vd
->vdev_dtl_smo
.smo_alloc
);
2120 (void) dmu_object_free(mos
, vd
->vdev_dtl_smo
.smo_object
, tx
);
2121 vd
->vdev_dtl_smo
.smo_object
= 0;
2124 if (vd
->vdev_ms
!= NULL
) {
2125 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2126 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2128 if (msp
== NULL
|| msp
->ms_smo
.smo_object
== 0)
2131 ASSERT0(msp
->ms_smo
.smo_alloc
);
2132 (void) dmu_object_free(mos
, msp
->ms_smo
.smo_object
, tx
);
2133 msp
->ms_smo
.smo_object
= 0;
2137 if (vd
->vdev_ms_array
) {
2138 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2139 vd
->vdev_ms_array
= 0;
2140 vd
->vdev_ms_shift
= 0;
2146 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2149 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2151 ASSERT(!vd
->vdev_ishole
);
2153 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2154 metaslab_sync_done(msp
, txg
);
2157 metaslab_sync_reassess(vd
->vdev_mg
);
2161 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2163 spa_t
*spa
= vd
->vdev_spa
;
2168 ASSERT(!vd
->vdev_ishole
);
2170 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2171 ASSERT(vd
== vd
->vdev_top
);
2172 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2173 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2174 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2175 ASSERT(vd
->vdev_ms_array
!= 0);
2176 vdev_config_dirty(vd
);
2181 * Remove the metadata associated with this vdev once it's empty.
2183 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2184 vdev_remove(vd
, txg
);
2186 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2187 metaslab_sync(msp
, txg
);
2188 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2191 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2192 vdev_dtl_sync(lvd
, txg
);
2194 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2198 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2200 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2204 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2205 * not be opened, and no I/O is attempted.
2208 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2212 spa_vdev_state_enter(spa
, SCL_NONE
);
2214 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2215 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2217 if (!vd
->vdev_ops
->vdev_op_leaf
)
2218 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2223 * We don't directly use the aux state here, but if we do a
2224 * vdev_reopen(), we need this value to be present to remember why we
2227 vd
->vdev_label_aux
= aux
;
2230 * Faulted state takes precedence over degraded.
2232 vd
->vdev_delayed_close
= B_FALSE
;
2233 vd
->vdev_faulted
= 1ULL;
2234 vd
->vdev_degraded
= 0ULL;
2235 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2238 * If this device has the only valid copy of the data, then
2239 * back off and simply mark the vdev as degraded instead.
2241 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2242 vd
->vdev_degraded
= 1ULL;
2243 vd
->vdev_faulted
= 0ULL;
2246 * If we reopen the device and it's not dead, only then do we
2251 if (vdev_readable(vd
))
2252 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2255 return (spa_vdev_state_exit(spa
, vd
, 0));
2259 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2260 * user that something is wrong. The vdev continues to operate as normal as far
2261 * as I/O is concerned.
2264 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2268 spa_vdev_state_enter(spa
, SCL_NONE
);
2270 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2271 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2273 if (!vd
->vdev_ops
->vdev_op_leaf
)
2274 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2277 * If the vdev is already faulted, then don't do anything.
2279 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2280 return (spa_vdev_state_exit(spa
, NULL
, 0));
2282 vd
->vdev_degraded
= 1ULL;
2283 if (!vdev_is_dead(vd
))
2284 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2287 return (spa_vdev_state_exit(spa
, vd
, 0));
2291 * Online the given vdev.
2293 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2294 * spare device should be detached when the device finishes resilvering.
2295 * Second, the online should be treated like a 'test' online case, so no FMA
2296 * events are generated if the device fails to open.
2299 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2301 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2303 spa_vdev_state_enter(spa
, SCL_NONE
);
2305 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2306 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2308 if (!vd
->vdev_ops
->vdev_op_leaf
)
2309 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2312 vd
->vdev_offline
= B_FALSE
;
2313 vd
->vdev_tmpoffline
= B_FALSE
;
2314 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2315 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2317 /* XXX - L2ARC 1.0 does not support expansion */
2318 if (!vd
->vdev_aux
) {
2319 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2320 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2324 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2326 if (!vd
->vdev_aux
) {
2327 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2328 pvd
->vdev_expanding
= B_FALSE
;
2332 *newstate
= vd
->vdev_state
;
2333 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2334 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2335 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2336 vd
->vdev_parent
->vdev_child
[0] == vd
)
2337 vd
->vdev_unspare
= B_TRUE
;
2339 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2341 /* XXX - L2ARC 1.0 does not support expansion */
2343 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2344 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2346 return (spa_vdev_state_exit(spa
, vd
, 0));
2350 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2354 uint64_t generation
;
2355 metaslab_group_t
*mg
;
2358 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2360 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2361 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2363 if (!vd
->vdev_ops
->vdev_op_leaf
)
2364 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2368 generation
= spa
->spa_config_generation
+ 1;
2371 * If the device isn't already offline, try to offline it.
2373 if (!vd
->vdev_offline
) {
2375 * If this device has the only valid copy of some data,
2376 * don't allow it to be offlined. Log devices are always
2379 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2380 vdev_dtl_required(vd
))
2381 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2384 * If the top-level is a slog and it has had allocations
2385 * then proceed. We check that the vdev's metaslab group
2386 * is not NULL since it's possible that we may have just
2387 * added this vdev but not yet initialized its metaslabs.
2389 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2391 * Prevent any future allocations.
2393 metaslab_group_passivate(mg
);
2394 (void) spa_vdev_state_exit(spa
, vd
, 0);
2396 error
= spa_offline_log(spa
);
2398 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2401 * Check to see if the config has changed.
2403 if (error
|| generation
!= spa
->spa_config_generation
) {
2404 metaslab_group_activate(mg
);
2406 return (spa_vdev_state_exit(spa
,
2408 (void) spa_vdev_state_exit(spa
, vd
, 0);
2411 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2415 * Offline this device and reopen its top-level vdev.
2416 * If the top-level vdev is a log device then just offline
2417 * it. Otherwise, if this action results in the top-level
2418 * vdev becoming unusable, undo it and fail the request.
2420 vd
->vdev_offline
= B_TRUE
;
2423 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2424 vdev_is_dead(tvd
)) {
2425 vd
->vdev_offline
= B_FALSE
;
2427 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2431 * Add the device back into the metaslab rotor so that
2432 * once we online the device it's open for business.
2434 if (tvd
->vdev_islog
&& mg
!= NULL
)
2435 metaslab_group_activate(mg
);
2438 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2440 return (spa_vdev_state_exit(spa
, vd
, 0));
2444 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2448 mutex_enter(&spa
->spa_vdev_top_lock
);
2449 error
= vdev_offline_locked(spa
, guid
, flags
);
2450 mutex_exit(&spa
->spa_vdev_top_lock
);
2456 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2457 * vdev_offline(), we assume the spa config is locked. We also clear all
2458 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2461 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2463 vdev_t
*rvd
= spa
->spa_root_vdev
;
2466 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2471 vd
->vdev_stat
.vs_read_errors
= 0;
2472 vd
->vdev_stat
.vs_write_errors
= 0;
2473 vd
->vdev_stat
.vs_checksum_errors
= 0;
2475 for (c
= 0; c
< vd
->vdev_children
; c
++)
2476 vdev_clear(spa
, vd
->vdev_child
[c
]);
2479 * If we're in the FAULTED state or have experienced failed I/O, then
2480 * clear the persistent state and attempt to reopen the device. We
2481 * also mark the vdev config dirty, so that the new faulted state is
2482 * written out to disk.
2484 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2485 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2488 * When reopening in reponse to a clear event, it may be due to
2489 * a fmadm repair request. In this case, if the device is
2490 * still broken, we want to still post the ereport again.
2492 vd
->vdev_forcefault
= B_TRUE
;
2494 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2495 vd
->vdev_cant_read
= B_FALSE
;
2496 vd
->vdev_cant_write
= B_FALSE
;
2498 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2500 vd
->vdev_forcefault
= B_FALSE
;
2502 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2503 vdev_state_dirty(vd
->vdev_top
);
2505 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2506 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2508 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_CLEAR
);
2512 * When clearing a FMA-diagnosed fault, we always want to
2513 * unspare the device, as we assume that the original spare was
2514 * done in response to the FMA fault.
2516 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2517 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2518 vd
->vdev_parent
->vdev_child
[0] == vd
)
2519 vd
->vdev_unspare
= B_TRUE
;
2523 vdev_is_dead(vdev_t
*vd
)
2526 * Holes and missing devices are always considered "dead".
2527 * This simplifies the code since we don't have to check for
2528 * these types of devices in the various code paths.
2529 * Instead we rely on the fact that we skip over dead devices
2530 * before issuing I/O to them.
2532 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2533 vd
->vdev_ops
== &vdev_missing_ops
);
2537 vdev_readable(vdev_t
*vd
)
2539 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2543 vdev_writeable(vdev_t
*vd
)
2545 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2549 vdev_allocatable(vdev_t
*vd
)
2551 uint64_t state
= vd
->vdev_state
;
2554 * We currently allow allocations from vdevs which may be in the
2555 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2556 * fails to reopen then we'll catch it later when we're holding
2557 * the proper locks. Note that we have to get the vdev state
2558 * in a local variable because although it changes atomically,
2559 * we're asking two separate questions about it.
2561 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2562 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2566 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2568 ASSERT(zio
->io_vd
== vd
);
2570 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2573 if (zio
->io_type
== ZIO_TYPE_READ
)
2574 return (!vd
->vdev_cant_read
);
2576 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2577 return (!vd
->vdev_cant_write
);
2583 * Get statistics for the given vdev.
2586 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2588 vdev_t
*rvd
= vd
->vdev_spa
->spa_root_vdev
;
2591 mutex_enter(&vd
->vdev_stat_lock
);
2592 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2593 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2594 vs
->vs_state
= vd
->vdev_state
;
2595 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2596 if (vd
->vdev_ops
->vdev_op_leaf
)
2597 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2598 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2599 mutex_exit(&vd
->vdev_stat_lock
);
2602 * If we're getting stats on the root vdev, aggregate the I/O counts
2603 * over all top-level vdevs (i.e. the direct children of the root).
2606 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2607 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2608 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2610 mutex_enter(&vd
->vdev_stat_lock
);
2611 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2612 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2613 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2615 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2616 mutex_exit(&vd
->vdev_stat_lock
);
2622 vdev_clear_stats(vdev_t
*vd
)
2624 mutex_enter(&vd
->vdev_stat_lock
);
2625 vd
->vdev_stat
.vs_space
= 0;
2626 vd
->vdev_stat
.vs_dspace
= 0;
2627 vd
->vdev_stat
.vs_alloc
= 0;
2628 mutex_exit(&vd
->vdev_stat_lock
);
2632 vdev_scan_stat_init(vdev_t
*vd
)
2634 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2637 for (c
= 0; c
< vd
->vdev_children
; c
++)
2638 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2640 mutex_enter(&vd
->vdev_stat_lock
);
2641 vs
->vs_scan_processed
= 0;
2642 mutex_exit(&vd
->vdev_stat_lock
);
2646 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2648 spa_t
*spa
= zio
->io_spa
;
2649 vdev_t
*rvd
= spa
->spa_root_vdev
;
2650 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2652 uint64_t txg
= zio
->io_txg
;
2653 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2654 zio_type_t type
= zio
->io_type
;
2655 int flags
= zio
->io_flags
;
2658 * If this i/o is a gang leader, it didn't do any actual work.
2660 if (zio
->io_gang_tree
)
2663 if (zio
->io_error
== 0) {
2665 * If this is a root i/o, don't count it -- we've already
2666 * counted the top-level vdevs, and vdev_get_stats() will
2667 * aggregate them when asked. This reduces contention on
2668 * the root vdev_stat_lock and implicitly handles blocks
2669 * that compress away to holes, for which there is no i/o.
2670 * (Holes never create vdev children, so all the counters
2671 * remain zero, which is what we want.)
2673 * Note: this only applies to successful i/o (io_error == 0)
2674 * because unlike i/o counts, errors are not additive.
2675 * When reading a ditto block, for example, failure of
2676 * one top-level vdev does not imply a root-level error.
2681 ASSERT(vd
== zio
->io_vd
);
2683 if (flags
& ZIO_FLAG_IO_BYPASS
)
2686 mutex_enter(&vd
->vdev_stat_lock
);
2688 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2689 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2690 dsl_scan_phys_t
*scn_phys
=
2691 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2692 uint64_t *processed
= &scn_phys
->scn_processed
;
2695 if (vd
->vdev_ops
->vdev_op_leaf
)
2696 atomic_add_64(processed
, psize
);
2697 vs
->vs_scan_processed
+= psize
;
2700 if (flags
& ZIO_FLAG_SELF_HEAL
)
2701 vs
->vs_self_healed
+= psize
;
2705 vs
->vs_bytes
[type
] += psize
;
2707 mutex_exit(&vd
->vdev_stat_lock
);
2711 if (flags
& ZIO_FLAG_SPECULATIVE
)
2715 * If this is an I/O error that is going to be retried, then ignore the
2716 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2717 * hard errors, when in reality they can happen for any number of
2718 * innocuous reasons (bus resets, MPxIO link failure, etc).
2720 if (zio
->io_error
== EIO
&&
2721 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2725 * Intent logs writes won't propagate their error to the root
2726 * I/O so don't mark these types of failures as pool-level
2729 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2732 mutex_enter(&vd
->vdev_stat_lock
);
2733 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2734 if (zio
->io_error
== ECKSUM
)
2735 vs
->vs_checksum_errors
++;
2737 vs
->vs_read_errors
++;
2739 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2740 vs
->vs_write_errors
++;
2741 mutex_exit(&vd
->vdev_stat_lock
);
2743 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2744 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2745 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2746 spa
->spa_claiming
)) {
2748 * This is either a normal write (not a repair), or it's
2749 * a repair induced by the scrub thread, or it's a repair
2750 * made by zil_claim() during spa_load() in the first txg.
2751 * In the normal case, we commit the DTL change in the same
2752 * txg as the block was born. In the scrub-induced repair
2753 * case, we know that scrubs run in first-pass syncing context,
2754 * so we commit the DTL change in spa_syncing_txg(spa).
2755 * In the zil_claim() case, we commit in spa_first_txg(spa).
2757 * We currently do not make DTL entries for failed spontaneous
2758 * self-healing writes triggered by normal (non-scrubbing)
2759 * reads, because we have no transactional context in which to
2760 * do so -- and it's not clear that it'd be desirable anyway.
2762 if (vd
->vdev_ops
->vdev_op_leaf
) {
2763 uint64_t commit_txg
= txg
;
2764 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2765 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2766 ASSERT(spa_sync_pass(spa
) == 1);
2767 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2768 commit_txg
= spa_syncing_txg(spa
);
2769 } else if (spa
->spa_claiming
) {
2770 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2771 commit_txg
= spa_first_txg(spa
);
2773 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2774 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2776 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2777 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2778 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2781 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2786 * Update the in-core space usage stats for this vdev, its metaslab class,
2787 * and the root vdev.
2790 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2791 int64_t space_delta
)
2793 int64_t dspace_delta
= space_delta
;
2794 spa_t
*spa
= vd
->vdev_spa
;
2795 vdev_t
*rvd
= spa
->spa_root_vdev
;
2796 metaslab_group_t
*mg
= vd
->vdev_mg
;
2797 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2799 ASSERT(vd
== vd
->vdev_top
);
2802 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2803 * factor. We must calculate this here and not at the root vdev
2804 * because the root vdev's psize-to-asize is simply the max of its
2805 * childrens', thus not accurate enough for us.
2807 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2808 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2809 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2810 vd
->vdev_deflate_ratio
;
2812 mutex_enter(&vd
->vdev_stat_lock
);
2813 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2814 vd
->vdev_stat
.vs_space
+= space_delta
;
2815 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2816 mutex_exit(&vd
->vdev_stat_lock
);
2818 if (mc
== spa_normal_class(spa
)) {
2819 mutex_enter(&rvd
->vdev_stat_lock
);
2820 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2821 rvd
->vdev_stat
.vs_space
+= space_delta
;
2822 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2823 mutex_exit(&rvd
->vdev_stat_lock
);
2827 ASSERT(rvd
== vd
->vdev_parent
);
2828 ASSERT(vd
->vdev_ms_count
!= 0);
2830 metaslab_class_space_update(mc
,
2831 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2836 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2837 * so that it will be written out next time the vdev configuration is synced.
2838 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2841 vdev_config_dirty(vdev_t
*vd
)
2843 spa_t
*spa
= vd
->vdev_spa
;
2844 vdev_t
*rvd
= spa
->spa_root_vdev
;
2847 ASSERT(spa_writeable(spa
));
2850 * If this is an aux vdev (as with l2cache and spare devices), then we
2851 * update the vdev config manually and set the sync flag.
2853 if (vd
->vdev_aux
!= NULL
) {
2854 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2858 for (c
= 0; c
< sav
->sav_count
; c
++) {
2859 if (sav
->sav_vdevs
[c
] == vd
)
2863 if (c
== sav
->sav_count
) {
2865 * We're being removed. There's nothing more to do.
2867 ASSERT(sav
->sav_sync
== B_TRUE
);
2871 sav
->sav_sync
= B_TRUE
;
2873 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2874 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2875 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2876 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2882 * Setting the nvlist in the middle if the array is a little
2883 * sketchy, but it will work.
2885 nvlist_free(aux
[c
]);
2886 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
2892 * The dirty list is protected by the SCL_CONFIG lock. The caller
2893 * must either hold SCL_CONFIG as writer, or must be the sync thread
2894 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2895 * so this is sufficient to ensure mutual exclusion.
2897 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2898 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2899 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2902 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2903 vdev_config_dirty(rvd
->vdev_child
[c
]);
2905 ASSERT(vd
== vd
->vdev_top
);
2907 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
2909 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
2914 vdev_config_clean(vdev_t
*vd
)
2916 spa_t
*spa
= vd
->vdev_spa
;
2918 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2919 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2920 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2922 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
2923 list_remove(&spa
->spa_config_dirty_list
, vd
);
2927 * Mark a top-level vdev's state as dirty, so that the next pass of
2928 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2929 * the state changes from larger config changes because they require
2930 * much less locking, and are often needed for administrative actions.
2933 vdev_state_dirty(vdev_t
*vd
)
2935 spa_t
*spa
= vd
->vdev_spa
;
2937 ASSERT(spa_writeable(spa
));
2938 ASSERT(vd
== vd
->vdev_top
);
2941 * The state list is protected by the SCL_STATE lock. The caller
2942 * must either hold SCL_STATE as writer, or must be the sync thread
2943 * (which holds SCL_STATE as reader). There's only one sync thread,
2944 * so this is sufficient to ensure mutual exclusion.
2946 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2947 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2948 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2950 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
2951 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
2955 vdev_state_clean(vdev_t
*vd
)
2957 spa_t
*spa
= vd
->vdev_spa
;
2959 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2960 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2961 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2963 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
2964 list_remove(&spa
->spa_state_dirty_list
, vd
);
2968 * Propagate vdev state up from children to parent.
2971 vdev_propagate_state(vdev_t
*vd
)
2973 spa_t
*spa
= vd
->vdev_spa
;
2974 vdev_t
*rvd
= spa
->spa_root_vdev
;
2975 int degraded
= 0, faulted
= 0;
2980 if (vd
->vdev_children
> 0) {
2981 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2982 child
= vd
->vdev_child
[c
];
2985 * Don't factor holes into the decision.
2987 if (child
->vdev_ishole
)
2990 if (!vdev_readable(child
) ||
2991 (!vdev_writeable(child
) && spa_writeable(spa
))) {
2993 * Root special: if there is a top-level log
2994 * device, treat the root vdev as if it were
2997 if (child
->vdev_islog
&& vd
== rvd
)
3001 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3005 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3009 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3012 * Root special: if there is a top-level vdev that cannot be
3013 * opened due to corrupted metadata, then propagate the root
3014 * vdev's aux state as 'corrupt' rather than 'insufficient
3017 if (corrupted
&& vd
== rvd
&&
3018 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3019 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3020 VDEV_AUX_CORRUPT_DATA
);
3023 if (vd
->vdev_parent
)
3024 vdev_propagate_state(vd
->vdev_parent
);
3028 * Set a vdev's state. If this is during an open, we don't update the parent
3029 * state, because we're in the process of opening children depth-first.
3030 * Otherwise, we propagate the change to the parent.
3032 * If this routine places a device in a faulted state, an appropriate ereport is
3036 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3038 uint64_t save_state
;
3039 spa_t
*spa
= vd
->vdev_spa
;
3041 if (state
== vd
->vdev_state
) {
3042 vd
->vdev_stat
.vs_aux
= aux
;
3046 save_state
= vd
->vdev_state
;
3048 vd
->vdev_state
= state
;
3049 vd
->vdev_stat
.vs_aux
= aux
;
3052 * If we are setting the vdev state to anything but an open state, then
3053 * always close the underlying device unless the device has requested
3054 * a delayed close (i.e. we're about to remove or fault the device).
3055 * Otherwise, we keep accessible but invalid devices open forever.
3056 * We don't call vdev_close() itself, because that implies some extra
3057 * checks (offline, etc) that we don't want here. This is limited to
3058 * leaf devices, because otherwise closing the device will affect other
3061 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3062 vd
->vdev_ops
->vdev_op_leaf
)
3063 vd
->vdev_ops
->vdev_op_close(vd
);
3066 * If we have brought this vdev back into service, we need
3067 * to notify fmd so that it can gracefully repair any outstanding
3068 * cases due to a missing device. We do this in all cases, even those
3069 * that probably don't correlate to a repaired fault. This is sure to
3070 * catch all cases, and we let the zfs-retire agent sort it out. If
3071 * this is a transient state it's OK, as the retire agent will
3072 * double-check the state of the vdev before repairing it.
3074 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
3075 vd
->vdev_prevstate
!= state
)
3076 zfs_post_state_change(spa
, vd
);
3078 if (vd
->vdev_removed
&&
3079 state
== VDEV_STATE_CANT_OPEN
&&
3080 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3082 * If the previous state is set to VDEV_STATE_REMOVED, then this
3083 * device was previously marked removed and someone attempted to
3084 * reopen it. If this failed due to a nonexistent device, then
3085 * keep the device in the REMOVED state. We also let this be if
3086 * it is one of our special test online cases, which is only
3087 * attempting to online the device and shouldn't generate an FMA
3090 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3091 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3092 } else if (state
== VDEV_STATE_REMOVED
) {
3093 vd
->vdev_removed
= B_TRUE
;
3094 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3096 * If we fail to open a vdev during an import or recovery, we
3097 * mark it as "not available", which signifies that it was
3098 * never there to begin with. Failure to open such a device
3099 * is not considered an error.
3101 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3102 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3103 vd
->vdev_ops
->vdev_op_leaf
)
3104 vd
->vdev_not_present
= 1;
3107 * Post the appropriate ereport. If the 'prevstate' field is
3108 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3109 * that this is part of a vdev_reopen(). In this case, we don't
3110 * want to post the ereport if the device was already in the
3111 * CANT_OPEN state beforehand.
3113 * If the 'checkremove' flag is set, then this is an attempt to
3114 * online the device in response to an insertion event. If we
3115 * hit this case, then we have detected an insertion event for a
3116 * faulted or offline device that wasn't in the removed state.
3117 * In this scenario, we don't post an ereport because we are
3118 * about to replace the device, or attempt an online with
3119 * vdev_forcefault, which will generate the fault for us.
3121 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3122 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3123 vd
!= spa
->spa_root_vdev
) {
3127 case VDEV_AUX_OPEN_FAILED
:
3128 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3130 case VDEV_AUX_CORRUPT_DATA
:
3131 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3133 case VDEV_AUX_NO_REPLICAS
:
3134 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3136 case VDEV_AUX_BAD_GUID_SUM
:
3137 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3139 case VDEV_AUX_TOO_SMALL
:
3140 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3142 case VDEV_AUX_BAD_LABEL
:
3143 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3146 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3149 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3152 /* Erase any notion of persistent removed state */
3153 vd
->vdev_removed
= B_FALSE
;
3155 vd
->vdev_removed
= B_FALSE
;
3158 if (!isopen
&& vd
->vdev_parent
)
3159 vdev_propagate_state(vd
->vdev_parent
);
3163 * Check the vdev configuration to ensure that it's capable of supporting
3167 vdev_is_bootable(vdev_t
*vd
)
3169 #if defined(__sun__) || defined(__sun)
3171 * Currently, we do not support RAID-Z or partial configuration.
3172 * In addition, only a single top-level vdev is allowed and none of the
3173 * leaves can be wholedisks.
3177 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3178 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3180 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3181 vd
->vdev_children
> 1) {
3183 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3184 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3187 } else if (vd
->vdev_wholedisk
== 1) {
3191 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3192 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3195 #endif /* __sun__ || __sun */
3200 * Load the state from the original vdev tree (ovd) which
3201 * we've retrieved from the MOS config object. If the original
3202 * vdev was offline or faulted then we transfer that state to the
3203 * device in the current vdev tree (nvd).
3206 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3210 ASSERT(nvd
->vdev_top
->vdev_islog
);
3211 ASSERT(spa_config_held(nvd
->vdev_spa
,
3212 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3213 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3215 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3216 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3218 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3220 * Restore the persistent vdev state
3222 nvd
->vdev_offline
= ovd
->vdev_offline
;
3223 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3224 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3225 nvd
->vdev_removed
= ovd
->vdev_removed
;
3230 * Determine if a log device has valid content. If the vdev was
3231 * removed or faulted in the MOS config then we know that
3232 * the content on the log device has already been written to the pool.
3235 vdev_log_state_valid(vdev_t
*vd
)
3239 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3243 for (c
= 0; c
< vd
->vdev_children
; c
++)
3244 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3251 * Expand a vdev if possible.
3254 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3256 ASSERT(vd
->vdev_top
== vd
);
3257 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3259 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3260 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3261 vdev_config_dirty(vd
);
3269 vdev_split(vdev_t
*vd
)
3271 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3273 vdev_remove_child(pvd
, vd
);
3274 vdev_compact_children(pvd
);
3276 cvd
= pvd
->vdev_child
[0];
3277 if (pvd
->vdev_children
== 1) {
3278 vdev_remove_parent(cvd
);
3279 cvd
->vdev_splitting
= B_TRUE
;
3281 vdev_propagate_state(cvd
);
3285 vdev_deadman(vdev_t
*vd
)
3289 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3290 vdev_t
*cvd
= vd
->vdev_child
[c
];
3295 if (vd
->vdev_ops
->vdev_op_leaf
) {
3296 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3298 mutex_enter(&vq
->vq_lock
);
3299 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3300 spa_t
*spa
= vd
->vdev_spa
;
3305 * Look at the head of all the pending queues,
3306 * if any I/O has been outstanding for longer than
3307 * the spa_deadman_synctime we log a zevent.
3309 fio
= avl_first(&vq
->vq_active_tree
);
3310 delta
= gethrtime() - fio
->io_timestamp
;
3311 if (delta
> spa_deadman_synctime(spa
)) {
3312 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3313 "delta %lluns, last io %lluns",
3314 fio
->io_timestamp
, delta
,
3315 vq
->vq_io_complete_ts
);
3316 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3317 spa
, vd
, fio
, 0, 0);
3320 mutex_exit(&vq
->vq_lock
);
3324 #if defined(_KERNEL) && defined(HAVE_SPL)
3325 EXPORT_SYMBOL(vdev_fault
);
3326 EXPORT_SYMBOL(vdev_degrade
);
3327 EXPORT_SYMBOL(vdev_online
);
3328 EXPORT_SYMBOL(vdev_offline
);
3329 EXPORT_SYMBOL(vdev_clear
);