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_RESILVERING
,
530 &vd
->vdev_resilvering
);
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 by reading and writing
970 * to several known locations: the pad regions of each vdev label
971 * but the first (which we leave alone in case it contains a VTOC).
974 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
976 spa_t
*spa
= vd
->vdev_spa
;
977 vdev_probe_stats_t
*vps
= NULL
;
981 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
984 * Don't probe the probe.
986 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
990 * To prevent 'probe storms' when a device fails, we create
991 * just one probe i/o at a time. All zios that want to probe
992 * this vdev will become parents of the probe io.
994 mutex_enter(&vd
->vdev_probe_lock
);
996 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
997 vps
= kmem_zalloc(sizeof (*vps
), KM_PUSHPAGE
);
999 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1000 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1003 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1005 * vdev_cant_read and vdev_cant_write can only
1006 * transition from TRUE to FALSE when we have the
1007 * SCL_ZIO lock as writer; otherwise they can only
1008 * transition from FALSE to TRUE. This ensures that
1009 * any zio looking at these values can assume that
1010 * failures persist for the life of the I/O. That's
1011 * important because when a device has intermittent
1012 * connectivity problems, we want to ensure that
1013 * they're ascribed to the device (ENXIO) and not
1016 * Since we hold SCL_ZIO as writer here, clear both
1017 * values so the probe can reevaluate from first
1020 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1021 vd
->vdev_cant_read
= B_FALSE
;
1022 vd
->vdev_cant_write
= B_FALSE
;
1025 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1026 vdev_probe_done
, vps
,
1027 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1030 * We can't change the vdev state in this context, so we
1031 * kick off an async task to do it on our behalf.
1034 vd
->vdev_probe_wanted
= B_TRUE
;
1035 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1040 zio_add_child(zio
, pio
);
1042 mutex_exit(&vd
->vdev_probe_lock
);
1045 ASSERT(zio
!= NULL
);
1049 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1050 zio_nowait(zio_read_phys(pio
, vd
,
1051 vdev_label_offset(vd
->vdev_psize
, l
,
1052 offsetof(vdev_label_t
, vl_pad2
)),
1053 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1054 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1055 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1066 vdev_open_child(void *arg
)
1070 vd
->vdev_open_thread
= curthread
;
1071 vd
->vdev_open_error
= vdev_open(vd
);
1072 vd
->vdev_open_thread
= NULL
;
1076 vdev_uses_zvols(vdev_t
*vd
)
1081 if (zvol_is_zvol(vd
->vdev_path
))
1085 for (c
= 0; c
< vd
->vdev_children
; c
++)
1086 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1093 vdev_open_children(vdev_t
*vd
)
1096 int children
= vd
->vdev_children
;
1100 * in order to handle pools on top of zvols, do the opens
1101 * in a single thread so that the same thread holds the
1102 * spa_namespace_lock
1104 if (vdev_uses_zvols(vd
)) {
1105 for (c
= 0; c
< children
; c
++)
1106 vd
->vdev_child
[c
]->vdev_open_error
=
1107 vdev_open(vd
->vdev_child
[c
]);
1110 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1111 children
, children
, TASKQ_PREPOPULATE
);
1113 for (c
= 0; c
< children
; c
++)
1114 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1121 * Prepare a virtual device for access.
1124 vdev_open(vdev_t
*vd
)
1126 spa_t
*spa
= vd
->vdev_spa
;
1129 uint64_t max_osize
= 0;
1130 uint64_t asize
, max_asize
, psize
;
1131 uint64_t ashift
= 0;
1134 ASSERT(vd
->vdev_open_thread
== curthread
||
1135 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1136 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1137 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1138 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1140 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1141 vd
->vdev_cant_read
= B_FALSE
;
1142 vd
->vdev_cant_write
= B_FALSE
;
1143 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1146 * If this vdev is not removed, check its fault status. If it's
1147 * faulted, bail out of the open.
1149 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1150 ASSERT(vd
->vdev_children
== 0);
1151 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1152 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1153 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1154 vd
->vdev_label_aux
);
1155 return (SET_ERROR(ENXIO
));
1156 } else if (vd
->vdev_offline
) {
1157 ASSERT(vd
->vdev_children
== 0);
1158 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1159 return (SET_ERROR(ENXIO
));
1162 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1165 * Reset the vdev_reopening flag so that we actually close
1166 * the vdev on error.
1168 vd
->vdev_reopening
= B_FALSE
;
1169 if (zio_injection_enabled
&& error
== 0)
1170 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1173 if (vd
->vdev_removed
&&
1174 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1175 vd
->vdev_removed
= B_FALSE
;
1177 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1178 vd
->vdev_stat
.vs_aux
);
1182 vd
->vdev_removed
= B_FALSE
;
1185 * Recheck the faulted flag now that we have confirmed that
1186 * the vdev is accessible. If we're faulted, bail.
1188 if (vd
->vdev_faulted
) {
1189 ASSERT(vd
->vdev_children
== 0);
1190 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1191 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1192 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1193 vd
->vdev_label_aux
);
1194 return (SET_ERROR(ENXIO
));
1197 if (vd
->vdev_degraded
) {
1198 ASSERT(vd
->vdev_children
== 0);
1199 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1200 VDEV_AUX_ERR_EXCEEDED
);
1202 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1206 * For hole or missing vdevs we just return success.
1208 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1211 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1212 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1213 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1219 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1220 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1222 if (vd
->vdev_children
== 0) {
1223 if (osize
< SPA_MINDEVSIZE
) {
1224 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1225 VDEV_AUX_TOO_SMALL
);
1226 return (SET_ERROR(EOVERFLOW
));
1229 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1230 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1231 VDEV_LABEL_END_SIZE
);
1233 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1234 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1235 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1236 VDEV_AUX_TOO_SMALL
);
1237 return (SET_ERROR(EOVERFLOW
));
1241 max_asize
= max_osize
;
1244 vd
->vdev_psize
= psize
;
1247 * Make sure the allocatable size hasn't shrunk.
1249 if (asize
< vd
->vdev_min_asize
) {
1250 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1251 VDEV_AUX_BAD_LABEL
);
1252 return (SET_ERROR(EINVAL
));
1255 if (vd
->vdev_asize
== 0) {
1257 * This is the first-ever open, so use the computed values.
1258 * For compatibility, a different ashift can be requested.
1260 vd
->vdev_asize
= asize
;
1261 vd
->vdev_max_asize
= max_asize
;
1262 if (vd
->vdev_ashift
== 0)
1263 vd
->vdev_ashift
= ashift
;
1266 * Detect if the alignment requirement has increased.
1267 * We don't want to make the pool unavailable, just
1268 * post an event instead.
1270 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1271 vd
->vdev_ops
->vdev_op_leaf
) {
1272 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1273 spa
, vd
, NULL
, 0, 0);
1276 vd
->vdev_max_asize
= max_asize
;
1280 * If all children are healthy and the asize has increased,
1281 * then we've experienced dynamic LUN growth. If automatic
1282 * expansion is enabled then use the additional space.
1284 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1285 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1286 vd
->vdev_asize
= asize
;
1288 vdev_set_min_asize(vd
);
1291 * Ensure we can issue some IO before declaring the
1292 * vdev open for business.
1294 if (vd
->vdev_ops
->vdev_op_leaf
&&
1295 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1296 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1297 VDEV_AUX_ERR_EXCEEDED
);
1302 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1303 * resilver. But don't do this if we are doing a reopen for a scrub,
1304 * since this would just restart the scrub we are already doing.
1306 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1307 vdev_resilver_needed(vd
, NULL
, NULL
))
1308 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1314 * Called once the vdevs are all opened, this routine validates the label
1315 * contents. This needs to be done before vdev_load() so that we don't
1316 * inadvertently do repair I/Os to the wrong device.
1318 * If 'strict' is false ignore the spa guid check. This is necessary because
1319 * if the machine crashed during a re-guid the new guid might have been written
1320 * to all of the vdev labels, but not the cached config. The strict check
1321 * will be performed when the pool is opened again using the mos config.
1323 * This function will only return failure if one of the vdevs indicates that it
1324 * has since been destroyed or exported. This is only possible if
1325 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1326 * will be updated but the function will return 0.
1329 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1331 spa_t
*spa
= vd
->vdev_spa
;
1333 uint64_t guid
= 0, top_guid
;
1337 for (c
= 0; c
< vd
->vdev_children
; c
++)
1338 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1339 return (SET_ERROR(EBADF
));
1342 * If the device has already failed, or was marked offline, don't do
1343 * any further validation. Otherwise, label I/O will fail and we will
1344 * overwrite the previous state.
1346 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1347 uint64_t aux_guid
= 0;
1349 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1350 spa_last_synced_txg(spa
) : -1ULL;
1352 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1353 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1354 VDEV_AUX_BAD_LABEL
);
1359 * Determine if this vdev has been split off into another
1360 * pool. If so, then refuse to open it.
1362 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1363 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1364 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1365 VDEV_AUX_SPLIT_POOL
);
1370 if (strict
&& (nvlist_lookup_uint64(label
,
1371 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1372 guid
!= spa_guid(spa
))) {
1373 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1374 VDEV_AUX_CORRUPT_DATA
);
1379 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1380 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1385 * If this vdev just became a top-level vdev because its
1386 * sibling was detached, it will have adopted the parent's
1387 * vdev guid -- but the label may or may not be on disk yet.
1388 * Fortunately, either version of the label will have the
1389 * same top guid, so if we're a top-level vdev, we can
1390 * safely compare to that instead.
1392 * If we split this vdev off instead, then we also check the
1393 * original pool's guid. We don't want to consider the vdev
1394 * corrupt if it is partway through a split operation.
1396 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1398 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1400 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1401 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1402 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1403 VDEV_AUX_CORRUPT_DATA
);
1408 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1410 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1411 VDEV_AUX_CORRUPT_DATA
);
1419 * If this is a verbatim import, no need to check the
1420 * state of the pool.
1422 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1423 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1424 state
!= POOL_STATE_ACTIVE
)
1425 return (SET_ERROR(EBADF
));
1428 * If we were able to open and validate a vdev that was
1429 * previously marked permanently unavailable, clear that state
1432 if (vd
->vdev_not_present
)
1433 vd
->vdev_not_present
= 0;
1440 * Close a virtual device.
1443 vdev_close(vdev_t
*vd
)
1445 vdev_t
*pvd
= vd
->vdev_parent
;
1446 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1448 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1451 * If our parent is reopening, then we are as well, unless we are
1454 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1455 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1457 vd
->vdev_ops
->vdev_op_close(vd
);
1459 vdev_cache_purge(vd
);
1462 * We record the previous state before we close it, so that if we are
1463 * doing a reopen(), we don't generate FMA ereports if we notice that
1464 * it's still faulted.
1466 vd
->vdev_prevstate
= vd
->vdev_state
;
1468 if (vd
->vdev_offline
)
1469 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1471 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1472 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1476 vdev_hold(vdev_t
*vd
)
1478 spa_t
*spa
= vd
->vdev_spa
;
1481 ASSERT(spa_is_root(spa
));
1482 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1485 for (c
= 0; c
< vd
->vdev_children
; c
++)
1486 vdev_hold(vd
->vdev_child
[c
]);
1488 if (vd
->vdev_ops
->vdev_op_leaf
)
1489 vd
->vdev_ops
->vdev_op_hold(vd
);
1493 vdev_rele(vdev_t
*vd
)
1497 ASSERT(spa_is_root(vd
->vdev_spa
));
1498 for (c
= 0; c
< vd
->vdev_children
; c
++)
1499 vdev_rele(vd
->vdev_child
[c
]);
1501 if (vd
->vdev_ops
->vdev_op_leaf
)
1502 vd
->vdev_ops
->vdev_op_rele(vd
);
1506 * Reopen all interior vdevs and any unopened leaves. We don't actually
1507 * reopen leaf vdevs which had previously been opened as they might deadlock
1508 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1509 * If the leaf has never been opened then open it, as usual.
1512 vdev_reopen(vdev_t
*vd
)
1514 spa_t
*spa
= vd
->vdev_spa
;
1516 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1518 /* set the reopening flag unless we're taking the vdev offline */
1519 vd
->vdev_reopening
= !vd
->vdev_offline
;
1521 (void) vdev_open(vd
);
1524 * Call vdev_validate() here to make sure we have the same device.
1525 * Otherwise, a device with an invalid label could be successfully
1526 * opened in response to vdev_reopen().
1529 (void) vdev_validate_aux(vd
);
1530 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1531 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1532 !l2arc_vdev_present(vd
))
1533 l2arc_add_vdev(spa
, vd
);
1535 (void) vdev_validate(vd
, B_TRUE
);
1539 * Reassess parent vdev's health.
1541 vdev_propagate_state(vd
);
1545 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1550 * Normally, partial opens (e.g. of a mirror) are allowed.
1551 * For a create, however, we want to fail the request if
1552 * there are any components we can't open.
1554 error
= vdev_open(vd
);
1556 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1558 return (error
? error
: ENXIO
);
1562 * Recursively initialize all labels.
1564 if ((error
= vdev_label_init(vd
, txg
, isreplacing
?
1565 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1574 vdev_metaslab_set_size(vdev_t
*vd
)
1577 * Aim for roughly 200 metaslabs per vdev.
1579 vd
->vdev_ms_shift
= highbit(vd
->vdev_asize
/ 200);
1580 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1584 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1586 ASSERT(vd
== vd
->vdev_top
);
1587 ASSERT(!vd
->vdev_ishole
);
1588 ASSERT(ISP2(flags
));
1589 ASSERT(spa_writeable(vd
->vdev_spa
));
1591 if (flags
& VDD_METASLAB
)
1592 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1594 if (flags
& VDD_DTL
)
1595 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1597 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1603 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1604 * the vdev has less than perfect replication. There are four kinds of DTL:
1606 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1608 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1610 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1611 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1612 * txgs that was scrubbed.
1614 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1615 * persistent errors or just some device being offline.
1616 * Unlike the other three, the DTL_OUTAGE map is not generally
1617 * maintained; it's only computed when needed, typically to
1618 * determine whether a device can be detached.
1620 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1621 * either has the data or it doesn't.
1623 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1624 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1625 * if any child is less than fully replicated, then so is its parent.
1626 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1627 * comprising only those txgs which appear in 'maxfaults' or more children;
1628 * those are the txgs we don't have enough replication to read. For example,
1629 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1630 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1631 * two child DTL_MISSING maps.
1633 * It should be clear from the above that to compute the DTLs and outage maps
1634 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1635 * Therefore, that is all we keep on disk. When loading the pool, or after
1636 * a configuration change, we generate all other DTLs from first principles.
1639 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1641 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1643 ASSERT(t
< DTL_TYPES
);
1644 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1645 ASSERT(spa_writeable(vd
->vdev_spa
));
1647 mutex_enter(sm
->sm_lock
);
1648 if (!space_map_contains(sm
, txg
, size
))
1649 space_map_add(sm
, txg
, size
);
1650 mutex_exit(sm
->sm_lock
);
1654 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1656 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1657 boolean_t dirty
= B_FALSE
;
1659 ASSERT(t
< DTL_TYPES
);
1660 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1662 mutex_enter(sm
->sm_lock
);
1663 if (sm
->sm_space
!= 0)
1664 dirty
= space_map_contains(sm
, txg
, size
);
1665 mutex_exit(sm
->sm_lock
);
1671 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1673 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1676 mutex_enter(sm
->sm_lock
);
1677 empty
= (sm
->sm_space
== 0);
1678 mutex_exit(sm
->sm_lock
);
1684 * Reassess DTLs after a config change or scrub completion.
1687 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1689 spa_t
*spa
= vd
->vdev_spa
;
1693 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1695 for (c
= 0; c
< vd
->vdev_children
; c
++)
1696 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1697 scrub_txg
, scrub_done
);
1699 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1702 if (vd
->vdev_ops
->vdev_op_leaf
) {
1703 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1705 mutex_enter(&vd
->vdev_dtl_lock
);
1706 if (scrub_txg
!= 0 &&
1707 (spa
->spa_scrub_started
||
1708 (scn
&& scn
->scn_phys
.scn_errors
== 0))) {
1710 * We completed a scrub up to scrub_txg. If we
1711 * did it without rebooting, then the scrub dtl
1712 * will be valid, so excise the old region and
1713 * fold in the scrub dtl. Otherwise, leave the
1714 * dtl as-is if there was an error.
1716 * There's little trick here: to excise the beginning
1717 * of the DTL_MISSING map, we put it into a reference
1718 * tree and then add a segment with refcnt -1 that
1719 * covers the range [0, scrub_txg). This means
1720 * that each txg in that range has refcnt -1 or 0.
1721 * We then add DTL_SCRUB with a refcnt of 2, so that
1722 * entries in the range [0, scrub_txg) will have a
1723 * positive refcnt -- either 1 or 2. We then convert
1724 * the reference tree into the new DTL_MISSING map.
1726 space_map_ref_create(&reftree
);
1727 space_map_ref_add_map(&reftree
,
1728 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1729 space_map_ref_add_seg(&reftree
, 0, scrub_txg
, -1);
1730 space_map_ref_add_map(&reftree
,
1731 &vd
->vdev_dtl
[DTL_SCRUB
], 2);
1732 space_map_ref_generate_map(&reftree
,
1733 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1734 space_map_ref_destroy(&reftree
);
1736 space_map_vacate(&vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1737 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1738 space_map_add
, &vd
->vdev_dtl
[DTL_PARTIAL
]);
1740 space_map_vacate(&vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1741 space_map_vacate(&vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1742 if (!vdev_readable(vd
))
1743 space_map_add(&vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1745 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1746 space_map_add
, &vd
->vdev_dtl
[DTL_OUTAGE
]);
1747 mutex_exit(&vd
->vdev_dtl_lock
);
1750 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1754 mutex_enter(&vd
->vdev_dtl_lock
);
1755 for (t
= 0; t
< DTL_TYPES
; t
++) {
1756 /* account for child's outage in parent's missing map */
1757 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1759 continue; /* leaf vdevs only */
1760 if (t
== DTL_PARTIAL
)
1761 minref
= 1; /* i.e. non-zero */
1762 else if (vd
->vdev_nparity
!= 0)
1763 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1765 minref
= vd
->vdev_children
; /* any kind of mirror */
1766 space_map_ref_create(&reftree
);
1767 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1768 vdev_t
*cvd
= vd
->vdev_child
[c
];
1769 mutex_enter(&cvd
->vdev_dtl_lock
);
1770 space_map_ref_add_map(&reftree
, &cvd
->vdev_dtl
[s
], 1);
1771 mutex_exit(&cvd
->vdev_dtl_lock
);
1773 space_map_ref_generate_map(&reftree
, &vd
->vdev_dtl
[t
], minref
);
1774 space_map_ref_destroy(&reftree
);
1776 mutex_exit(&vd
->vdev_dtl_lock
);
1780 vdev_dtl_load(vdev_t
*vd
)
1782 spa_t
*spa
= vd
->vdev_spa
;
1783 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1784 objset_t
*mos
= spa
->spa_meta_objset
;
1788 ASSERT(vd
->vdev_children
== 0);
1790 if (smo
->smo_object
== 0)
1793 ASSERT(!vd
->vdev_ishole
);
1795 if ((error
= dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
)) != 0)
1798 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1799 bcopy(db
->db_data
, smo
, sizeof (*smo
));
1800 dmu_buf_rele(db
, FTAG
);
1802 mutex_enter(&vd
->vdev_dtl_lock
);
1803 error
= space_map_load(&vd
->vdev_dtl
[DTL_MISSING
],
1804 NULL
, SM_ALLOC
, smo
, mos
);
1805 mutex_exit(&vd
->vdev_dtl_lock
);
1811 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1813 spa_t
*spa
= vd
->vdev_spa
;
1814 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1815 space_map_t
*sm
= &vd
->vdev_dtl
[DTL_MISSING
];
1816 objset_t
*mos
= spa
->spa_meta_objset
;
1822 ASSERT(!vd
->vdev_ishole
);
1824 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1826 if (vd
->vdev_detached
) {
1827 if (smo
->smo_object
!= 0) {
1828 VERIFY0(dmu_object_free(mos
, smo
->smo_object
, tx
));
1829 smo
->smo_object
= 0;
1835 if (smo
->smo_object
== 0) {
1836 ASSERT(smo
->smo_objsize
== 0);
1837 ASSERT(smo
->smo_alloc
== 0);
1838 smo
->smo_object
= dmu_object_alloc(mos
,
1839 DMU_OT_SPACE_MAP
, 1 << SPACE_MAP_BLOCKSHIFT
,
1840 DMU_OT_SPACE_MAP_HEADER
, sizeof (*smo
), tx
);
1841 ASSERT(smo
->smo_object
!= 0);
1842 vdev_config_dirty(vd
->vdev_top
);
1845 mutex_init(&smlock
, NULL
, MUTEX_DEFAULT
, NULL
);
1847 space_map_create(&smsync
, sm
->sm_start
, sm
->sm_size
, sm
->sm_shift
,
1850 mutex_enter(&smlock
);
1852 mutex_enter(&vd
->vdev_dtl_lock
);
1853 space_map_walk(sm
, space_map_add
, &smsync
);
1854 mutex_exit(&vd
->vdev_dtl_lock
);
1856 space_map_truncate(smo
, mos
, tx
);
1857 space_map_sync(&smsync
, SM_ALLOC
, smo
, mos
, tx
);
1858 space_map_vacate(&smsync
, NULL
, NULL
);
1860 space_map_destroy(&smsync
);
1862 mutex_exit(&smlock
);
1863 mutex_destroy(&smlock
);
1865 VERIFY(0 == dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
));
1866 dmu_buf_will_dirty(db
, tx
);
1867 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1868 bcopy(smo
, db
->db_data
, sizeof (*smo
));
1869 dmu_buf_rele(db
, FTAG
);
1875 * Determine whether the specified vdev can be offlined/detached/removed
1876 * without losing data.
1879 vdev_dtl_required(vdev_t
*vd
)
1881 spa_t
*spa
= vd
->vdev_spa
;
1882 vdev_t
*tvd
= vd
->vdev_top
;
1883 uint8_t cant_read
= vd
->vdev_cant_read
;
1886 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1888 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
1892 * Temporarily mark the device as unreadable, and then determine
1893 * whether this results in any DTL outages in the top-level vdev.
1894 * If not, we can safely offline/detach/remove the device.
1896 vd
->vdev_cant_read
= B_TRUE
;
1897 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1898 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
1899 vd
->vdev_cant_read
= cant_read
;
1900 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1902 if (!required
&& zio_injection_enabled
)
1903 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
1909 * Determine if resilver is needed, and if so the txg range.
1912 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
1914 boolean_t needed
= B_FALSE
;
1915 uint64_t thismin
= UINT64_MAX
;
1916 uint64_t thismax
= 0;
1919 if (vd
->vdev_children
== 0) {
1920 mutex_enter(&vd
->vdev_dtl_lock
);
1921 if (vd
->vdev_dtl
[DTL_MISSING
].sm_space
!= 0 &&
1922 vdev_writeable(vd
)) {
1925 ss
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1926 thismin
= ss
->ss_start
- 1;
1927 ss
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1928 thismax
= ss
->ss_end
;
1931 mutex_exit(&vd
->vdev_dtl_lock
);
1933 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1934 vdev_t
*cvd
= vd
->vdev_child
[c
];
1935 uint64_t cmin
, cmax
;
1937 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
1938 thismin
= MIN(thismin
, cmin
);
1939 thismax
= MAX(thismax
, cmax
);
1945 if (needed
&& minp
) {
1953 vdev_load(vdev_t
*vd
)
1958 * Recursively load all children.
1960 for (c
= 0; c
< vd
->vdev_children
; c
++)
1961 vdev_load(vd
->vdev_child
[c
]);
1964 * If this is a top-level vdev, initialize its metaslabs.
1966 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
1967 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
1968 vdev_metaslab_init(vd
, 0) != 0))
1969 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1970 VDEV_AUX_CORRUPT_DATA
);
1973 * If this is a leaf vdev, load its DTL.
1975 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
1976 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1977 VDEV_AUX_CORRUPT_DATA
);
1981 * The special vdev case is used for hot spares and l2cache devices. Its
1982 * sole purpose it to set the vdev state for the associated vdev. To do this,
1983 * we make sure that we can open the underlying device, then try to read the
1984 * label, and make sure that the label is sane and that it hasn't been
1985 * repurposed to another pool.
1988 vdev_validate_aux(vdev_t
*vd
)
1991 uint64_t guid
, version
;
1994 if (!vdev_readable(vd
))
1997 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
1998 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1999 VDEV_AUX_CORRUPT_DATA
);
2003 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2004 !SPA_VERSION_IS_SUPPORTED(version
) ||
2005 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2006 guid
!= vd
->vdev_guid
||
2007 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2008 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2009 VDEV_AUX_CORRUPT_DATA
);
2015 * We don't actually check the pool state here. If it's in fact in
2016 * use by another pool, we update this fact on the fly when requested.
2023 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2025 spa_t
*spa
= vd
->vdev_spa
;
2026 objset_t
*mos
= spa
->spa_meta_objset
;
2030 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2032 if (vd
->vdev_dtl_smo
.smo_object
) {
2033 ASSERT0(vd
->vdev_dtl_smo
.smo_alloc
);
2034 (void) dmu_object_free(mos
, vd
->vdev_dtl_smo
.smo_object
, tx
);
2035 vd
->vdev_dtl_smo
.smo_object
= 0;
2038 if (vd
->vdev_ms
!= NULL
) {
2039 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2040 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2042 if (msp
== NULL
|| msp
->ms_smo
.smo_object
== 0)
2045 ASSERT0(msp
->ms_smo
.smo_alloc
);
2046 (void) dmu_object_free(mos
, msp
->ms_smo
.smo_object
, tx
);
2047 msp
->ms_smo
.smo_object
= 0;
2051 if (vd
->vdev_ms_array
) {
2052 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2053 vd
->vdev_ms_array
= 0;
2054 vd
->vdev_ms_shift
= 0;
2060 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2063 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2065 ASSERT(!vd
->vdev_ishole
);
2067 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2068 metaslab_sync_done(msp
, txg
);
2071 metaslab_sync_reassess(vd
->vdev_mg
);
2075 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2077 spa_t
*spa
= vd
->vdev_spa
;
2082 ASSERT(!vd
->vdev_ishole
);
2084 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2085 ASSERT(vd
== vd
->vdev_top
);
2086 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2087 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2088 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2089 ASSERT(vd
->vdev_ms_array
!= 0);
2090 vdev_config_dirty(vd
);
2095 * Remove the metadata associated with this vdev once it's empty.
2097 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2098 vdev_remove(vd
, txg
);
2100 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2101 metaslab_sync(msp
, txg
);
2102 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2105 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2106 vdev_dtl_sync(lvd
, txg
);
2108 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2112 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2114 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2118 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2119 * not be opened, and no I/O is attempted.
2122 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2126 spa_vdev_state_enter(spa
, SCL_NONE
);
2128 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2129 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2131 if (!vd
->vdev_ops
->vdev_op_leaf
)
2132 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2137 * We don't directly use the aux state here, but if we do a
2138 * vdev_reopen(), we need this value to be present to remember why we
2141 vd
->vdev_label_aux
= aux
;
2144 * Faulted state takes precedence over degraded.
2146 vd
->vdev_delayed_close
= B_FALSE
;
2147 vd
->vdev_faulted
= 1ULL;
2148 vd
->vdev_degraded
= 0ULL;
2149 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2152 * If this device has the only valid copy of the data, then
2153 * back off and simply mark the vdev as degraded instead.
2155 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2156 vd
->vdev_degraded
= 1ULL;
2157 vd
->vdev_faulted
= 0ULL;
2160 * If we reopen the device and it's not dead, only then do we
2165 if (vdev_readable(vd
))
2166 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2169 return (spa_vdev_state_exit(spa
, vd
, 0));
2173 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2174 * user that something is wrong. The vdev continues to operate as normal as far
2175 * as I/O is concerned.
2178 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2182 spa_vdev_state_enter(spa
, SCL_NONE
);
2184 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2185 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2187 if (!vd
->vdev_ops
->vdev_op_leaf
)
2188 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2191 * If the vdev is already faulted, then don't do anything.
2193 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2194 return (spa_vdev_state_exit(spa
, NULL
, 0));
2196 vd
->vdev_degraded
= 1ULL;
2197 if (!vdev_is_dead(vd
))
2198 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2201 return (spa_vdev_state_exit(spa
, vd
, 0));
2205 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2206 * any attached spare device should be detached when the device finishes
2207 * resilvering. Second, the online should be treated like a 'test' online case,
2208 * so no FMA events are generated if the device fails to open.
2211 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2213 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2215 spa_vdev_state_enter(spa
, SCL_NONE
);
2217 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2218 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2220 if (!vd
->vdev_ops
->vdev_op_leaf
)
2221 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2224 vd
->vdev_offline
= B_FALSE
;
2225 vd
->vdev_tmpoffline
= B_FALSE
;
2226 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2227 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2229 /* XXX - L2ARC 1.0 does not support expansion */
2230 if (!vd
->vdev_aux
) {
2231 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2232 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2236 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2238 if (!vd
->vdev_aux
) {
2239 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2240 pvd
->vdev_expanding
= B_FALSE
;
2244 *newstate
= vd
->vdev_state
;
2245 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2246 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2247 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2248 vd
->vdev_parent
->vdev_child
[0] == vd
)
2249 vd
->vdev_unspare
= B_TRUE
;
2251 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2253 /* XXX - L2ARC 1.0 does not support expansion */
2255 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2256 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2258 return (spa_vdev_state_exit(spa
, vd
, 0));
2262 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2266 uint64_t generation
;
2267 metaslab_group_t
*mg
;
2270 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2272 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2273 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2275 if (!vd
->vdev_ops
->vdev_op_leaf
)
2276 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2280 generation
= spa
->spa_config_generation
+ 1;
2283 * If the device isn't already offline, try to offline it.
2285 if (!vd
->vdev_offline
) {
2287 * If this device has the only valid copy of some data,
2288 * don't allow it to be offlined. Log devices are always
2291 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2292 vdev_dtl_required(vd
))
2293 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2296 * If the top-level is a slog and it has had allocations
2297 * then proceed. We check that the vdev's metaslab group
2298 * is not NULL since it's possible that we may have just
2299 * added this vdev but not yet initialized its metaslabs.
2301 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2303 * Prevent any future allocations.
2305 metaslab_group_passivate(mg
);
2306 (void) spa_vdev_state_exit(spa
, vd
, 0);
2308 error
= spa_offline_log(spa
);
2310 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2313 * Check to see if the config has changed.
2315 if (error
|| generation
!= spa
->spa_config_generation
) {
2316 metaslab_group_activate(mg
);
2318 return (spa_vdev_state_exit(spa
,
2320 (void) spa_vdev_state_exit(spa
, vd
, 0);
2323 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2327 * Offline this device and reopen its top-level vdev.
2328 * If the top-level vdev is a log device then just offline
2329 * it. Otherwise, if this action results in the top-level
2330 * vdev becoming unusable, undo it and fail the request.
2332 vd
->vdev_offline
= B_TRUE
;
2335 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2336 vdev_is_dead(tvd
)) {
2337 vd
->vdev_offline
= B_FALSE
;
2339 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2343 * Add the device back into the metaslab rotor so that
2344 * once we online the device it's open for business.
2346 if (tvd
->vdev_islog
&& mg
!= NULL
)
2347 metaslab_group_activate(mg
);
2350 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2352 return (spa_vdev_state_exit(spa
, vd
, 0));
2356 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2360 mutex_enter(&spa
->spa_vdev_top_lock
);
2361 error
= vdev_offline_locked(spa
, guid
, flags
);
2362 mutex_exit(&spa
->spa_vdev_top_lock
);
2368 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2369 * vdev_offline(), we assume the spa config is locked. We also clear all
2370 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2373 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2375 vdev_t
*rvd
= spa
->spa_root_vdev
;
2378 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2383 vd
->vdev_stat
.vs_read_errors
= 0;
2384 vd
->vdev_stat
.vs_write_errors
= 0;
2385 vd
->vdev_stat
.vs_checksum_errors
= 0;
2387 for (c
= 0; c
< vd
->vdev_children
; c
++)
2388 vdev_clear(spa
, vd
->vdev_child
[c
]);
2391 * If we're in the FAULTED state or have experienced failed I/O, then
2392 * clear the persistent state and attempt to reopen the device. We
2393 * also mark the vdev config dirty, so that the new faulted state is
2394 * written out to disk.
2396 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2397 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2400 * When reopening in reponse to a clear event, it may be due to
2401 * a fmadm repair request. In this case, if the device is
2402 * still broken, we want to still post the ereport again.
2404 vd
->vdev_forcefault
= B_TRUE
;
2406 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2407 vd
->vdev_cant_read
= B_FALSE
;
2408 vd
->vdev_cant_write
= B_FALSE
;
2410 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2412 vd
->vdev_forcefault
= B_FALSE
;
2414 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2415 vdev_state_dirty(vd
->vdev_top
);
2417 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2418 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2420 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_CLEAR
);
2424 * When clearing a FMA-diagnosed fault, we always want to
2425 * unspare the device, as we assume that the original spare was
2426 * done in response to the FMA fault.
2428 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2429 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2430 vd
->vdev_parent
->vdev_child
[0] == vd
)
2431 vd
->vdev_unspare
= B_TRUE
;
2435 vdev_is_dead(vdev_t
*vd
)
2438 * Holes and missing devices are always considered "dead".
2439 * This simplifies the code since we don't have to check for
2440 * these types of devices in the various code paths.
2441 * Instead we rely on the fact that we skip over dead devices
2442 * before issuing I/O to them.
2444 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2445 vd
->vdev_ops
== &vdev_missing_ops
);
2449 vdev_readable(vdev_t
*vd
)
2451 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2455 vdev_writeable(vdev_t
*vd
)
2457 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2461 vdev_allocatable(vdev_t
*vd
)
2463 uint64_t state
= vd
->vdev_state
;
2466 * We currently allow allocations from vdevs which may be in the
2467 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2468 * fails to reopen then we'll catch it later when we're holding
2469 * the proper locks. Note that we have to get the vdev state
2470 * in a local variable because although it changes atomically,
2471 * we're asking two separate questions about it.
2473 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2474 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2478 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2480 ASSERT(zio
->io_vd
== vd
);
2482 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2485 if (zio
->io_type
== ZIO_TYPE_READ
)
2486 return (!vd
->vdev_cant_read
);
2488 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2489 return (!vd
->vdev_cant_write
);
2495 * Get statistics for the given vdev.
2498 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2500 vdev_t
*rvd
= vd
->vdev_spa
->spa_root_vdev
;
2503 mutex_enter(&vd
->vdev_stat_lock
);
2504 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2505 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2506 vs
->vs_state
= vd
->vdev_state
;
2507 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2508 if (vd
->vdev_ops
->vdev_op_leaf
)
2509 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2510 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2511 mutex_exit(&vd
->vdev_stat_lock
);
2514 * If we're getting stats on the root vdev, aggregate the I/O counts
2515 * over all top-level vdevs (i.e. the direct children of the root).
2518 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2519 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2520 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2522 mutex_enter(&vd
->vdev_stat_lock
);
2523 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2524 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2525 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2527 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2528 mutex_exit(&vd
->vdev_stat_lock
);
2534 vdev_clear_stats(vdev_t
*vd
)
2536 mutex_enter(&vd
->vdev_stat_lock
);
2537 vd
->vdev_stat
.vs_space
= 0;
2538 vd
->vdev_stat
.vs_dspace
= 0;
2539 vd
->vdev_stat
.vs_alloc
= 0;
2540 mutex_exit(&vd
->vdev_stat_lock
);
2544 vdev_scan_stat_init(vdev_t
*vd
)
2546 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2549 for (c
= 0; c
< vd
->vdev_children
; c
++)
2550 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2552 mutex_enter(&vd
->vdev_stat_lock
);
2553 vs
->vs_scan_processed
= 0;
2554 mutex_exit(&vd
->vdev_stat_lock
);
2558 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2560 spa_t
*spa
= zio
->io_spa
;
2561 vdev_t
*rvd
= spa
->spa_root_vdev
;
2562 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2564 uint64_t txg
= zio
->io_txg
;
2565 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2566 zio_type_t type
= zio
->io_type
;
2567 int flags
= zio
->io_flags
;
2570 * If this i/o is a gang leader, it didn't do any actual work.
2572 if (zio
->io_gang_tree
)
2575 if (zio
->io_error
== 0) {
2577 * If this is a root i/o, don't count it -- we've already
2578 * counted the top-level vdevs, and vdev_get_stats() will
2579 * aggregate them when asked. This reduces contention on
2580 * the root vdev_stat_lock and implicitly handles blocks
2581 * that compress away to holes, for which there is no i/o.
2582 * (Holes never create vdev children, so all the counters
2583 * remain zero, which is what we want.)
2585 * Note: this only applies to successful i/o (io_error == 0)
2586 * because unlike i/o counts, errors are not additive.
2587 * When reading a ditto block, for example, failure of
2588 * one top-level vdev does not imply a root-level error.
2593 ASSERT(vd
== zio
->io_vd
);
2595 if (flags
& ZIO_FLAG_IO_BYPASS
)
2598 mutex_enter(&vd
->vdev_stat_lock
);
2600 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2601 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2602 dsl_scan_phys_t
*scn_phys
=
2603 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2604 uint64_t *processed
= &scn_phys
->scn_processed
;
2607 if (vd
->vdev_ops
->vdev_op_leaf
)
2608 atomic_add_64(processed
, psize
);
2609 vs
->vs_scan_processed
+= psize
;
2612 if (flags
& ZIO_FLAG_SELF_HEAL
)
2613 vs
->vs_self_healed
+= psize
;
2617 vs
->vs_bytes
[type
] += psize
;
2619 mutex_exit(&vd
->vdev_stat_lock
);
2623 if (flags
& ZIO_FLAG_SPECULATIVE
)
2627 * If this is an I/O error that is going to be retried, then ignore the
2628 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2629 * hard errors, when in reality they can happen for any number of
2630 * innocuous reasons (bus resets, MPxIO link failure, etc).
2632 if (zio
->io_error
== EIO
&&
2633 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2637 * Intent logs writes won't propagate their error to the root
2638 * I/O so don't mark these types of failures as pool-level
2641 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2644 mutex_enter(&vd
->vdev_stat_lock
);
2645 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2646 if (zio
->io_error
== ECKSUM
)
2647 vs
->vs_checksum_errors
++;
2649 vs
->vs_read_errors
++;
2651 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2652 vs
->vs_write_errors
++;
2653 mutex_exit(&vd
->vdev_stat_lock
);
2655 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2656 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2657 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2658 spa
->spa_claiming
)) {
2660 * This is either a normal write (not a repair), or it's
2661 * a repair induced by the scrub thread, or it's a repair
2662 * made by zil_claim() during spa_load() in the first txg.
2663 * In the normal case, we commit the DTL change in the same
2664 * txg as the block was born. In the scrub-induced repair
2665 * case, we know that scrubs run in first-pass syncing context,
2666 * so we commit the DTL change in spa_syncing_txg(spa).
2667 * In the zil_claim() case, we commit in spa_first_txg(spa).
2669 * We currently do not make DTL entries for failed spontaneous
2670 * self-healing writes triggered by normal (non-scrubbing)
2671 * reads, because we have no transactional context in which to
2672 * do so -- and it's not clear that it'd be desirable anyway.
2674 if (vd
->vdev_ops
->vdev_op_leaf
) {
2675 uint64_t commit_txg
= txg
;
2676 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2677 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2678 ASSERT(spa_sync_pass(spa
) == 1);
2679 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2680 commit_txg
= spa_syncing_txg(spa
);
2681 } else if (spa
->spa_claiming
) {
2682 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2683 commit_txg
= spa_first_txg(spa
);
2685 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2686 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2688 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2689 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2690 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2693 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2698 * Update the in-core space usage stats for this vdev, its metaslab class,
2699 * and the root vdev.
2702 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2703 int64_t space_delta
)
2705 int64_t dspace_delta
= space_delta
;
2706 spa_t
*spa
= vd
->vdev_spa
;
2707 vdev_t
*rvd
= spa
->spa_root_vdev
;
2708 metaslab_group_t
*mg
= vd
->vdev_mg
;
2709 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2711 ASSERT(vd
== vd
->vdev_top
);
2714 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2715 * factor. We must calculate this here and not at the root vdev
2716 * because the root vdev's psize-to-asize is simply the max of its
2717 * childrens', thus not accurate enough for us.
2719 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2720 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2721 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2722 vd
->vdev_deflate_ratio
;
2724 mutex_enter(&vd
->vdev_stat_lock
);
2725 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2726 vd
->vdev_stat
.vs_space
+= space_delta
;
2727 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2728 mutex_exit(&vd
->vdev_stat_lock
);
2730 if (mc
== spa_normal_class(spa
)) {
2731 mutex_enter(&rvd
->vdev_stat_lock
);
2732 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2733 rvd
->vdev_stat
.vs_space
+= space_delta
;
2734 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2735 mutex_exit(&rvd
->vdev_stat_lock
);
2739 ASSERT(rvd
== vd
->vdev_parent
);
2740 ASSERT(vd
->vdev_ms_count
!= 0);
2742 metaslab_class_space_update(mc
,
2743 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2748 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2749 * so that it will be written out next time the vdev configuration is synced.
2750 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2753 vdev_config_dirty(vdev_t
*vd
)
2755 spa_t
*spa
= vd
->vdev_spa
;
2756 vdev_t
*rvd
= spa
->spa_root_vdev
;
2759 ASSERT(spa_writeable(spa
));
2762 * If this is an aux vdev (as with l2cache and spare devices), then we
2763 * update the vdev config manually and set the sync flag.
2765 if (vd
->vdev_aux
!= NULL
) {
2766 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2770 for (c
= 0; c
< sav
->sav_count
; c
++) {
2771 if (sav
->sav_vdevs
[c
] == vd
)
2775 if (c
== sav
->sav_count
) {
2777 * We're being removed. There's nothing more to do.
2779 ASSERT(sav
->sav_sync
== B_TRUE
);
2783 sav
->sav_sync
= B_TRUE
;
2785 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2786 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2787 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2788 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2794 * Setting the nvlist in the middle if the array is a little
2795 * sketchy, but it will work.
2797 nvlist_free(aux
[c
]);
2798 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
2804 * The dirty list is protected by the SCL_CONFIG lock. The caller
2805 * must either hold SCL_CONFIG as writer, or must be the sync thread
2806 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2807 * so this is sufficient to ensure mutual exclusion.
2809 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2810 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2811 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2814 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2815 vdev_config_dirty(rvd
->vdev_child
[c
]);
2817 ASSERT(vd
== vd
->vdev_top
);
2819 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
2821 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
2826 vdev_config_clean(vdev_t
*vd
)
2828 spa_t
*spa
= vd
->vdev_spa
;
2830 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2831 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2832 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2834 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
2835 list_remove(&spa
->spa_config_dirty_list
, vd
);
2839 * Mark a top-level vdev's state as dirty, so that the next pass of
2840 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2841 * the state changes from larger config changes because they require
2842 * much less locking, and are often needed for administrative actions.
2845 vdev_state_dirty(vdev_t
*vd
)
2847 spa_t
*spa
= vd
->vdev_spa
;
2849 ASSERT(spa_writeable(spa
));
2850 ASSERT(vd
== vd
->vdev_top
);
2853 * The state list is protected by the SCL_STATE lock. The caller
2854 * must either hold SCL_STATE as writer, or must be the sync thread
2855 * (which holds SCL_STATE as reader). There's only one sync thread,
2856 * so this is sufficient to ensure mutual exclusion.
2858 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2859 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2860 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2862 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
2863 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
2867 vdev_state_clean(vdev_t
*vd
)
2869 spa_t
*spa
= vd
->vdev_spa
;
2871 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2872 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2873 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2875 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
2876 list_remove(&spa
->spa_state_dirty_list
, vd
);
2880 * Propagate vdev state up from children to parent.
2883 vdev_propagate_state(vdev_t
*vd
)
2885 spa_t
*spa
= vd
->vdev_spa
;
2886 vdev_t
*rvd
= spa
->spa_root_vdev
;
2887 int degraded
= 0, faulted
= 0;
2892 if (vd
->vdev_children
> 0) {
2893 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2894 child
= vd
->vdev_child
[c
];
2897 * Don't factor holes into the decision.
2899 if (child
->vdev_ishole
)
2902 if (!vdev_readable(child
) ||
2903 (!vdev_writeable(child
) && spa_writeable(spa
))) {
2905 * Root special: if there is a top-level log
2906 * device, treat the root vdev as if it were
2909 if (child
->vdev_islog
&& vd
== rvd
)
2913 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
2917 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
2921 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
2924 * Root special: if there is a top-level vdev that cannot be
2925 * opened due to corrupted metadata, then propagate the root
2926 * vdev's aux state as 'corrupt' rather than 'insufficient
2929 if (corrupted
&& vd
== rvd
&&
2930 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
2931 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2932 VDEV_AUX_CORRUPT_DATA
);
2935 if (vd
->vdev_parent
)
2936 vdev_propagate_state(vd
->vdev_parent
);
2940 * Set a vdev's state. If this is during an open, we don't update the parent
2941 * state, because we're in the process of opening children depth-first.
2942 * Otherwise, we propagate the change to the parent.
2944 * If this routine places a device in a faulted state, an appropriate ereport is
2948 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
2950 uint64_t save_state
;
2951 spa_t
*spa
= vd
->vdev_spa
;
2953 if (state
== vd
->vdev_state
) {
2954 vd
->vdev_stat
.vs_aux
= aux
;
2958 save_state
= vd
->vdev_state
;
2960 vd
->vdev_state
= state
;
2961 vd
->vdev_stat
.vs_aux
= aux
;
2964 * If we are setting the vdev state to anything but an open state, then
2965 * always close the underlying device unless the device has requested
2966 * a delayed close (i.e. we're about to remove or fault the device).
2967 * Otherwise, we keep accessible but invalid devices open forever.
2968 * We don't call vdev_close() itself, because that implies some extra
2969 * checks (offline, etc) that we don't want here. This is limited to
2970 * leaf devices, because otherwise closing the device will affect other
2973 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
2974 vd
->vdev_ops
->vdev_op_leaf
)
2975 vd
->vdev_ops
->vdev_op_close(vd
);
2978 * If we have brought this vdev back into service, we need
2979 * to notify fmd so that it can gracefully repair any outstanding
2980 * cases due to a missing device. We do this in all cases, even those
2981 * that probably don't correlate to a repaired fault. This is sure to
2982 * catch all cases, and we let the zfs-retire agent sort it out. If
2983 * this is a transient state it's OK, as the retire agent will
2984 * double-check the state of the vdev before repairing it.
2986 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
2987 vd
->vdev_prevstate
!= state
)
2988 zfs_post_state_change(spa
, vd
);
2990 if (vd
->vdev_removed
&&
2991 state
== VDEV_STATE_CANT_OPEN
&&
2992 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
2994 * If the previous state is set to VDEV_STATE_REMOVED, then this
2995 * device was previously marked removed and someone attempted to
2996 * reopen it. If this failed due to a nonexistent device, then
2997 * keep the device in the REMOVED state. We also let this be if
2998 * it is one of our special test online cases, which is only
2999 * attempting to online the device and shouldn't generate an FMA
3002 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3003 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3004 } else if (state
== VDEV_STATE_REMOVED
) {
3005 vd
->vdev_removed
= B_TRUE
;
3006 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3008 * If we fail to open a vdev during an import or recovery, we
3009 * mark it as "not available", which signifies that it was
3010 * never there to begin with. Failure to open such a device
3011 * is not considered an error.
3013 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3014 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3015 vd
->vdev_ops
->vdev_op_leaf
)
3016 vd
->vdev_not_present
= 1;
3019 * Post the appropriate ereport. If the 'prevstate' field is
3020 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3021 * that this is part of a vdev_reopen(). In this case, we don't
3022 * want to post the ereport if the device was already in the
3023 * CANT_OPEN state beforehand.
3025 * If the 'checkremove' flag is set, then this is an attempt to
3026 * online the device in response to an insertion event. If we
3027 * hit this case, then we have detected an insertion event for a
3028 * faulted or offline device that wasn't in the removed state.
3029 * In this scenario, we don't post an ereport because we are
3030 * about to replace the device, or attempt an online with
3031 * vdev_forcefault, which will generate the fault for us.
3033 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3034 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3035 vd
!= spa
->spa_root_vdev
) {
3039 case VDEV_AUX_OPEN_FAILED
:
3040 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3042 case VDEV_AUX_CORRUPT_DATA
:
3043 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3045 case VDEV_AUX_NO_REPLICAS
:
3046 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3048 case VDEV_AUX_BAD_GUID_SUM
:
3049 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3051 case VDEV_AUX_TOO_SMALL
:
3052 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3054 case VDEV_AUX_BAD_LABEL
:
3055 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3058 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3061 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3064 /* Erase any notion of persistent removed state */
3065 vd
->vdev_removed
= B_FALSE
;
3067 vd
->vdev_removed
= B_FALSE
;
3070 if (!isopen
&& vd
->vdev_parent
)
3071 vdev_propagate_state(vd
->vdev_parent
);
3075 * Check the vdev configuration to ensure that it's capable of supporting
3079 vdev_is_bootable(vdev_t
*vd
)
3081 #if defined(__sun__) || defined(__sun)
3083 * Currently, we do not support RAID-Z or partial configuration.
3084 * In addition, only a single top-level vdev is allowed and none of the
3085 * leaves can be wholedisks.
3089 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3090 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3092 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3093 vd
->vdev_children
> 1) {
3095 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3096 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3099 } else if (vd
->vdev_wholedisk
== 1) {
3103 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3104 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3107 #endif /* __sun__ || __sun */
3112 * Load the state from the original vdev tree (ovd) which
3113 * we've retrieved from the MOS config object. If the original
3114 * vdev was offline or faulted then we transfer that state to the
3115 * device in the current vdev tree (nvd).
3118 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3122 ASSERT(nvd
->vdev_top
->vdev_islog
);
3123 ASSERT(spa_config_held(nvd
->vdev_spa
,
3124 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3125 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3127 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3128 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3130 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3132 * Restore the persistent vdev state
3134 nvd
->vdev_offline
= ovd
->vdev_offline
;
3135 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3136 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3137 nvd
->vdev_removed
= ovd
->vdev_removed
;
3142 * Determine if a log device has valid content. If the vdev was
3143 * removed or faulted in the MOS config then we know that
3144 * the content on the log device has already been written to the pool.
3147 vdev_log_state_valid(vdev_t
*vd
)
3151 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3155 for (c
= 0; c
< vd
->vdev_children
; c
++)
3156 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3163 * Expand a vdev if possible.
3166 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3168 ASSERT(vd
->vdev_top
== vd
);
3169 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3171 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3172 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3173 vdev_config_dirty(vd
);
3181 vdev_split(vdev_t
*vd
)
3183 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3185 vdev_remove_child(pvd
, vd
);
3186 vdev_compact_children(pvd
);
3188 cvd
= pvd
->vdev_child
[0];
3189 if (pvd
->vdev_children
== 1) {
3190 vdev_remove_parent(cvd
);
3191 cvd
->vdev_splitting
= B_TRUE
;
3193 vdev_propagate_state(cvd
);
3197 vdev_deadman(vdev_t
*vd
)
3201 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3202 vdev_t
*cvd
= vd
->vdev_child
[c
];
3207 if (vd
->vdev_ops
->vdev_op_leaf
) {
3208 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3210 mutex_enter(&vq
->vq_lock
);
3211 if (avl_numnodes(&vq
->vq_pending_tree
) > 0) {
3212 spa_t
*spa
= vd
->vdev_spa
;
3217 * Look at the head of all the pending queues,
3218 * if any I/O has been outstanding for longer than
3219 * the spa_deadman_synctime we log a zevent.
3221 fio
= avl_first(&vq
->vq_pending_tree
);
3222 delta
= gethrtime() - fio
->io_timestamp
;
3223 if (delta
> spa_deadman_synctime(spa
)) {
3224 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3225 "delta %lluns, last io %lluns",
3226 fio
->io_timestamp
, delta
,
3227 vq
->vq_io_complete_ts
);
3228 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3229 spa
, vd
, fio
, 0, 0);
3232 mutex_exit(&vq
->vq_lock
);
3236 #if defined(_KERNEL) && defined(HAVE_SPL)
3237 EXPORT_SYMBOL(vdev_fault
);
3238 EXPORT_SYMBOL(vdev_degrade
);
3239 EXPORT_SYMBOL(vdev_online
);
3240 EXPORT_SYMBOL(vdev_offline
);
3241 EXPORT_SYMBOL(vdev_clear
);