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.
26 #include <sys/zfs_context.h>
27 #include <sys/fm/fs/zfs.h>
29 #include <sys/spa_impl.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/vdev_impl.h>
33 #include <sys/uberblock_impl.h>
34 #include <sys/metaslab.h>
35 #include <sys/metaslab_impl.h>
36 #include <sys/space_map.h>
39 #include <sys/fs/zfs.h>
42 #include <sys/dsl_scan.h>
45 * Virtual device management.
48 static vdev_ops_t
*vdev_ops_table
[] = {
61 /* maximum scrub/resilver I/O queue per leaf vdev */
62 int zfs_scrub_limit
= 10;
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
);
89 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
90 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
91 asize
= MAX(asize
, csize
);
98 * Get the minimum allocatable size. We define the allocatable size as
99 * the vdev's asize rounded to the nearest metaslab. This allows us to
100 * replace or attach devices which don't have the same physical size but
101 * can still satisfy the same number of allocations.
104 vdev_get_min_asize(vdev_t
*vd
)
106 vdev_t
*pvd
= vd
->vdev_parent
;
109 * The our parent is NULL (inactive spare or cache) or is the root,
110 * just return our own asize.
113 return (vd
->vdev_asize
);
116 * The top-level vdev just returns the allocatable size rounded
117 * to the nearest metaslab.
119 if (vd
== vd
->vdev_top
)
120 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
123 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
124 * so each child must provide at least 1/Nth of its asize.
126 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
127 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
129 return (pvd
->vdev_min_asize
);
133 vdev_set_min_asize(vdev_t
*vd
)
135 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
137 for (int c
= 0; c
< vd
->vdev_children
; c
++)
138 vdev_set_min_asize(vd
->vdev_child
[c
]);
142 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
144 vdev_t
*rvd
= spa
->spa_root_vdev
;
146 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
148 if (vdev
< rvd
->vdev_children
) {
149 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
150 return (rvd
->vdev_child
[vdev
]);
157 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
161 if (vd
->vdev_guid
== guid
)
164 for (int c
= 0; c
< vd
->vdev_children
; c
++)
165 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
173 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
175 size_t oldsize
, newsize
;
176 uint64_t id
= cvd
->vdev_id
;
179 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
180 ASSERT(cvd
->vdev_parent
== NULL
);
182 cvd
->vdev_parent
= pvd
;
187 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
189 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
190 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
191 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
193 newchild
= kmem_zalloc(newsize
, KM_SLEEP
);
194 if (pvd
->vdev_child
!= NULL
) {
195 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
196 kmem_free(pvd
->vdev_child
, oldsize
);
199 pvd
->vdev_child
= newchild
;
200 pvd
->vdev_child
[id
] = cvd
;
202 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
203 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
206 * Walk up all ancestors to update guid sum.
208 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
209 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
211 if (cvd
->vdev_ops
->vdev_op_leaf
)
212 cvd
->vdev_spa
->spa_scrub_maxinflight
+= zfs_scrub_limit
;
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
;
248 if (cvd
->vdev_ops
->vdev_op_leaf
)
249 cvd
->vdev_spa
->spa_scrub_maxinflight
-= zfs_scrub_limit
;
253 * Remove any holes in the child array.
256 vdev_compact_children(vdev_t
*pvd
)
258 vdev_t
**newchild
, *cvd
;
259 int oldc
= pvd
->vdev_children
;
262 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
264 for (int c
= newc
= 0; c
< oldc
; c
++)
265 if (pvd
->vdev_child
[c
])
268 newchild
= kmem_alloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
270 for (int c
= newc
= 0; c
< oldc
; c
++) {
271 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
272 newchild
[newc
] = cvd
;
273 cvd
->vdev_id
= newc
++;
277 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
278 pvd
->vdev_child
= newchild
;
279 pvd
->vdev_children
= newc
;
283 * Allocate and minimally initialize a vdev_t.
286 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
290 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
292 if (spa
->spa_root_vdev
== NULL
) {
293 ASSERT(ops
== &vdev_root_ops
);
294 spa
->spa_root_vdev
= vd
;
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 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
322 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
323 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
324 for (int t
= 0; t
< DTL_TYPES
; t
++) {
325 space_map_create(&vd
->vdev_dtl
[t
], 0, -1ULL, 0,
328 txg_list_create(&vd
->vdev_ms_list
,
329 offsetof(struct metaslab
, ms_txg_node
));
330 txg_list_create(&vd
->vdev_dtl_list
,
331 offsetof(struct vdev
, vdev_dtl_node
));
332 vd
->vdev_stat
.vs_timestamp
= gethrtime();
340 * Allocate a new vdev. The 'alloctype' is used to control whether we are
341 * creating a new vdev or loading an existing one - the behavior is slightly
342 * different for each case.
345 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
350 uint64_t guid
= 0, islog
, nparity
;
353 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
355 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
358 if ((ops
= vdev_getops(type
)) == NULL
)
362 * If this is a load, get the vdev guid from the nvlist.
363 * Otherwise, vdev_alloc_common() will generate one for us.
365 if (alloctype
== VDEV_ALLOC_LOAD
) {
368 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
372 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
374 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
375 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
377 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
378 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
380 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
381 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
386 * The first allocated vdev must be of type 'root'.
388 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
392 * Determine whether we're a log vdev.
395 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
396 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
399 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
403 * Set the nparity property for RAID-Z vdevs.
406 if (ops
== &vdev_raidz_ops
) {
407 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
409 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
412 * Previous versions could only support 1 or 2 parity
416 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
419 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
423 * We require the parity to be specified for SPAs that
424 * support multiple parity levels.
426 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
429 * Otherwise, we default to 1 parity device for RAID-Z.
436 ASSERT(nparity
!= -1ULL);
438 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
440 vd
->vdev_islog
= islog
;
441 vd
->vdev_nparity
= nparity
;
443 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
444 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
445 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
446 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
447 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
448 &vd
->vdev_physpath
) == 0)
449 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
450 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
451 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
454 * Set the whole_disk property. If it's not specified, leave the value
457 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
458 &vd
->vdev_wholedisk
) != 0)
459 vd
->vdev_wholedisk
= -1ULL;
462 * Look for the 'not present' flag. This will only be set if the device
463 * was not present at the time of import.
465 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
466 &vd
->vdev_not_present
);
469 * Get the alignment requirement.
471 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
474 * Retrieve the vdev creation time.
476 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
480 * If we're a top-level vdev, try to load the allocation parameters.
482 if (parent
&& !parent
->vdev_parent
&&
483 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
484 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
486 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
488 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
490 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
494 if (parent
&& !parent
->vdev_parent
) {
495 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
496 alloctype
== VDEV_ALLOC_ADD
||
497 alloctype
== VDEV_ALLOC_SPLIT
||
498 alloctype
== VDEV_ALLOC_ROOTPOOL
);
499 vd
->vdev_mg
= metaslab_group_create(islog
?
500 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
504 * If we're a leaf vdev, try to load the DTL object and other state.
506 if (vd
->vdev_ops
->vdev_op_leaf
&&
507 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
508 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
509 if (alloctype
== VDEV_ALLOC_LOAD
) {
510 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
511 &vd
->vdev_dtl_smo
.smo_object
);
512 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
516 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
519 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
520 &spare
) == 0 && spare
)
524 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
528 * When importing a pool, we want to ignore the persistent fault
529 * state, as the diagnosis made on another system may not be
530 * valid in the current context. Local vdevs will
531 * remain in the faulted state.
533 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
534 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
536 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
538 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
541 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
545 VDEV_AUX_ERR_EXCEEDED
;
546 if (nvlist_lookup_string(nv
,
547 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
548 strcmp(aux
, "external") == 0)
549 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
555 * Add ourselves to the parent's list of children.
557 vdev_add_child(parent
, vd
);
565 vdev_free(vdev_t
*vd
)
567 spa_t
*spa
= vd
->vdev_spa
;
570 * vdev_free() implies closing the vdev first. This is simpler than
571 * trying to ensure complicated semantics for all callers.
575 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
576 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
581 for (int c
= 0; c
< vd
->vdev_children
; c
++)
582 vdev_free(vd
->vdev_child
[c
]);
584 ASSERT(vd
->vdev_child
== NULL
);
585 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
588 * Discard allocation state.
590 if (vd
->vdev_mg
!= NULL
) {
591 vdev_metaslab_fini(vd
);
592 metaslab_group_destroy(vd
->vdev_mg
);
595 ASSERT3U(vd
->vdev_stat
.vs_space
, ==, 0);
596 ASSERT3U(vd
->vdev_stat
.vs_dspace
, ==, 0);
597 ASSERT3U(vd
->vdev_stat
.vs_alloc
, ==, 0);
600 * Remove this vdev from its parent's child list.
602 vdev_remove_child(vd
->vdev_parent
, vd
);
604 ASSERT(vd
->vdev_parent
== NULL
);
607 * Clean up vdev structure.
613 spa_strfree(vd
->vdev_path
);
615 spa_strfree(vd
->vdev_devid
);
616 if (vd
->vdev_physpath
)
617 spa_strfree(vd
->vdev_physpath
);
619 spa_strfree(vd
->vdev_fru
);
621 if (vd
->vdev_isspare
)
622 spa_spare_remove(vd
);
623 if (vd
->vdev_isl2cache
)
624 spa_l2cache_remove(vd
);
626 txg_list_destroy(&vd
->vdev_ms_list
);
627 txg_list_destroy(&vd
->vdev_dtl_list
);
629 mutex_enter(&vd
->vdev_dtl_lock
);
630 for (int t
= 0; t
< DTL_TYPES
; t
++) {
631 space_map_unload(&vd
->vdev_dtl
[t
]);
632 space_map_destroy(&vd
->vdev_dtl
[t
]);
634 mutex_exit(&vd
->vdev_dtl_lock
);
636 mutex_destroy(&vd
->vdev_dtl_lock
);
637 mutex_destroy(&vd
->vdev_stat_lock
);
638 mutex_destroy(&vd
->vdev_probe_lock
);
640 if (vd
== spa
->spa_root_vdev
)
641 spa
->spa_root_vdev
= NULL
;
643 kmem_free(vd
, sizeof (vdev_t
));
647 * Transfer top-level vdev state from svd to tvd.
650 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
652 spa_t
*spa
= svd
->vdev_spa
;
657 ASSERT(tvd
== tvd
->vdev_top
);
659 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
660 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
661 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
663 svd
->vdev_ms_array
= 0;
664 svd
->vdev_ms_shift
= 0;
665 svd
->vdev_ms_count
= 0;
667 tvd
->vdev_mg
= svd
->vdev_mg
;
668 tvd
->vdev_ms
= svd
->vdev_ms
;
673 if (tvd
->vdev_mg
!= NULL
)
674 tvd
->vdev_mg
->mg_vd
= tvd
;
676 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
677 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
678 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
680 svd
->vdev_stat
.vs_alloc
= 0;
681 svd
->vdev_stat
.vs_space
= 0;
682 svd
->vdev_stat
.vs_dspace
= 0;
684 for (t
= 0; t
< TXG_SIZE
; t
++) {
685 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
686 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
687 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
688 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
689 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
690 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
693 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
694 vdev_config_clean(svd
);
695 vdev_config_dirty(tvd
);
698 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
699 vdev_state_clean(svd
);
700 vdev_state_dirty(tvd
);
703 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
704 svd
->vdev_deflate_ratio
= 0;
706 tvd
->vdev_islog
= svd
->vdev_islog
;
711 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
718 for (int c
= 0; c
< vd
->vdev_children
; c
++)
719 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
723 * Add a mirror/replacing vdev above an existing vdev.
726 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
728 spa_t
*spa
= cvd
->vdev_spa
;
729 vdev_t
*pvd
= cvd
->vdev_parent
;
732 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
734 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
736 mvd
->vdev_asize
= cvd
->vdev_asize
;
737 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
738 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
739 mvd
->vdev_state
= cvd
->vdev_state
;
740 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
742 vdev_remove_child(pvd
, cvd
);
743 vdev_add_child(pvd
, mvd
);
744 cvd
->vdev_id
= mvd
->vdev_children
;
745 vdev_add_child(mvd
, cvd
);
746 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
748 if (mvd
== mvd
->vdev_top
)
749 vdev_top_transfer(cvd
, mvd
);
755 * Remove a 1-way mirror/replacing vdev from the tree.
758 vdev_remove_parent(vdev_t
*cvd
)
760 vdev_t
*mvd
= cvd
->vdev_parent
;
761 vdev_t
*pvd
= mvd
->vdev_parent
;
763 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
765 ASSERT(mvd
->vdev_children
== 1);
766 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
767 mvd
->vdev_ops
== &vdev_replacing_ops
||
768 mvd
->vdev_ops
== &vdev_spare_ops
);
769 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
771 vdev_remove_child(mvd
, cvd
);
772 vdev_remove_child(pvd
, mvd
);
775 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
776 * Otherwise, we could have detached an offline device, and when we
777 * go to import the pool we'll think we have two top-level vdevs,
778 * instead of a different version of the same top-level vdev.
780 if (mvd
->vdev_top
== mvd
) {
781 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
782 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
783 cvd
->vdev_guid
+= guid_delta
;
784 cvd
->vdev_guid_sum
+= guid_delta
;
786 cvd
->vdev_id
= mvd
->vdev_id
;
787 vdev_add_child(pvd
, cvd
);
788 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
790 if (cvd
== cvd
->vdev_top
)
791 vdev_top_transfer(mvd
, cvd
);
793 ASSERT(mvd
->vdev_children
== 0);
798 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
800 spa_t
*spa
= vd
->vdev_spa
;
801 objset_t
*mos
= spa
->spa_meta_objset
;
803 uint64_t oldc
= vd
->vdev_ms_count
;
804 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
808 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
811 * This vdev is not being allocated from yet or is a hole.
813 if (vd
->vdev_ms_shift
== 0)
816 ASSERT(!vd
->vdev_ishole
);
819 * Compute the raidz-deflation ratio. Note, we hard-code
820 * in 128k (1 << 17) because it is the current "typical" blocksize.
821 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
822 * or we will inconsistently account for existing bp's.
824 vd
->vdev_deflate_ratio
= (1 << 17) /
825 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
827 ASSERT(oldc
<= newc
);
829 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
832 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
833 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
837 vd
->vdev_ms_count
= newc
;
839 for (m
= oldc
; m
< newc
; m
++) {
840 space_map_obj_t smo
= { 0, 0, 0 };
843 error
= dmu_read(mos
, vd
->vdev_ms_array
,
844 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
850 error
= dmu_bonus_hold(mos
, object
, FTAG
, &db
);
853 ASSERT3U(db
->db_size
, >=, sizeof (smo
));
854 bcopy(db
->db_data
, &smo
, sizeof (smo
));
855 ASSERT3U(smo
.smo_object
, ==, object
);
856 dmu_buf_rele(db
, FTAG
);
859 vd
->vdev_ms
[m
] = metaslab_init(vd
->vdev_mg
, &smo
,
860 m
<< vd
->vdev_ms_shift
, 1ULL << vd
->vdev_ms_shift
, txg
);
864 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
867 * If the vdev is being removed we don't activate
868 * the metaslabs since we want to ensure that no new
869 * allocations are performed on this device.
871 if (oldc
== 0 && !vd
->vdev_removing
)
872 metaslab_group_activate(vd
->vdev_mg
);
875 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
881 vdev_metaslab_fini(vdev_t
*vd
)
884 uint64_t count
= vd
->vdev_ms_count
;
886 if (vd
->vdev_ms
!= NULL
) {
887 metaslab_group_passivate(vd
->vdev_mg
);
888 for (m
= 0; m
< count
; m
++)
889 if (vd
->vdev_ms
[m
] != NULL
)
890 metaslab_fini(vd
->vdev_ms
[m
]);
891 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
896 typedef struct vdev_probe_stats
{
897 boolean_t vps_readable
;
898 boolean_t vps_writeable
;
900 } vdev_probe_stats_t
;
903 vdev_probe_done(zio_t
*zio
)
905 spa_t
*spa
= zio
->io_spa
;
906 vdev_t
*vd
= zio
->io_vd
;
907 vdev_probe_stats_t
*vps
= zio
->io_private
;
909 ASSERT(vd
->vdev_probe_zio
!= NULL
);
911 if (zio
->io_type
== ZIO_TYPE_READ
) {
912 if (zio
->io_error
== 0)
913 vps
->vps_readable
= 1;
914 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
915 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
916 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
917 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
918 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
920 zio_buf_free(zio
->io_data
, zio
->io_size
);
922 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
923 if (zio
->io_error
== 0)
924 vps
->vps_writeable
= 1;
925 zio_buf_free(zio
->io_data
, zio
->io_size
);
926 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
929 vd
->vdev_cant_read
|= !vps
->vps_readable
;
930 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
932 if (vdev_readable(vd
) &&
933 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
936 ASSERT(zio
->io_error
!= 0);
937 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
938 spa
, vd
, NULL
, 0, 0);
939 zio
->io_error
= ENXIO
;
942 mutex_enter(&vd
->vdev_probe_lock
);
943 ASSERT(vd
->vdev_probe_zio
== zio
);
944 vd
->vdev_probe_zio
= NULL
;
945 mutex_exit(&vd
->vdev_probe_lock
);
947 while ((pio
= zio_walk_parents(zio
)) != NULL
)
948 if (!vdev_accessible(vd
, pio
))
949 pio
->io_error
= ENXIO
;
951 kmem_free(vps
, sizeof (*vps
));
956 * Determine whether this device is accessible by reading and writing
957 * to several known locations: the pad regions of each vdev label
958 * but the first (which we leave alone in case it contains a VTOC).
961 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
963 spa_t
*spa
= vd
->vdev_spa
;
964 vdev_probe_stats_t
*vps
= NULL
;
967 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
970 * Don't probe the probe.
972 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
976 * To prevent 'probe storms' when a device fails, we create
977 * just one probe i/o at a time. All zios that want to probe
978 * this vdev will become parents of the probe io.
980 mutex_enter(&vd
->vdev_probe_lock
);
982 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
983 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
985 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
986 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
989 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
991 * vdev_cant_read and vdev_cant_write can only
992 * transition from TRUE to FALSE when we have the
993 * SCL_ZIO lock as writer; otherwise they can only
994 * transition from FALSE to TRUE. This ensures that
995 * any zio looking at these values can assume that
996 * failures persist for the life of the I/O. That's
997 * important because when a device has intermittent
998 * connectivity problems, we want to ensure that
999 * they're ascribed to the device (ENXIO) and not
1002 * Since we hold SCL_ZIO as writer here, clear both
1003 * values so the probe can reevaluate from first
1006 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1007 vd
->vdev_cant_read
= B_FALSE
;
1008 vd
->vdev_cant_write
= B_FALSE
;
1011 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1012 vdev_probe_done
, vps
,
1013 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1016 * We can't change the vdev state in this context, so we
1017 * kick off an async task to do it on our behalf.
1020 vd
->vdev_probe_wanted
= B_TRUE
;
1021 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1026 zio_add_child(zio
, pio
);
1028 mutex_exit(&vd
->vdev_probe_lock
);
1031 ASSERT(zio
!= NULL
);
1035 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1036 zio_nowait(zio_read_phys(pio
, vd
,
1037 vdev_label_offset(vd
->vdev_psize
, l
,
1038 offsetof(vdev_label_t
, vl_pad2
)),
1039 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1040 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1041 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1052 vdev_open_child(void *arg
)
1056 vd
->vdev_open_thread
= curthread
;
1057 vd
->vdev_open_error
= vdev_open(vd
);
1058 vd
->vdev_open_thread
= NULL
;
1062 vdev_uses_zvols(vdev_t
*vd
)
1064 if (vd
->vdev_path
&& strncmp(vd
->vdev_path
, ZVOL_DIR
,
1065 strlen(ZVOL_DIR
)) == 0)
1067 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1068 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1074 vdev_open_children(vdev_t
*vd
)
1077 int children
= vd
->vdev_children
;
1080 * in order to handle pools on top of zvols, do the opens
1081 * in a single thread so that the same thread holds the
1082 * spa_namespace_lock
1084 if (vdev_uses_zvols(vd
)) {
1085 for (int c
= 0; c
< children
; c
++)
1086 vd
->vdev_child
[c
]->vdev_open_error
=
1087 vdev_open(vd
->vdev_child
[c
]);
1090 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1091 children
, children
, TASKQ_PREPOPULATE
);
1093 for (int c
= 0; c
< children
; c
++)
1094 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1101 * Prepare a virtual device for access.
1104 vdev_open(vdev_t
*vd
)
1106 spa_t
*spa
= vd
->vdev_spa
;
1109 uint64_t asize
, psize
;
1110 uint64_t ashift
= 0;
1112 ASSERT(vd
->vdev_open_thread
== curthread
||
1113 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1114 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1115 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1116 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1118 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1119 vd
->vdev_cant_read
= B_FALSE
;
1120 vd
->vdev_cant_write
= B_FALSE
;
1121 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1124 * If this vdev is not removed, check its fault status. If it's
1125 * faulted, bail out of the open.
1127 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1128 ASSERT(vd
->vdev_children
== 0);
1129 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1130 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1131 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1132 vd
->vdev_label_aux
);
1134 } else if (vd
->vdev_offline
) {
1135 ASSERT(vd
->vdev_children
== 0);
1136 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1140 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &ashift
);
1143 * Reset the vdev_reopening flag so that we actually close
1144 * the vdev on error.
1146 vd
->vdev_reopening
= B_FALSE
;
1147 if (zio_injection_enabled
&& error
== 0)
1148 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1151 if (vd
->vdev_removed
&&
1152 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1153 vd
->vdev_removed
= B_FALSE
;
1155 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1156 vd
->vdev_stat
.vs_aux
);
1160 vd
->vdev_removed
= B_FALSE
;
1163 * Recheck the faulted flag now that we have confirmed that
1164 * the vdev is accessible. If we're faulted, bail.
1166 if (vd
->vdev_faulted
) {
1167 ASSERT(vd
->vdev_children
== 0);
1168 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1169 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1170 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1171 vd
->vdev_label_aux
);
1175 if (vd
->vdev_degraded
) {
1176 ASSERT(vd
->vdev_children
== 0);
1177 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1178 VDEV_AUX_ERR_EXCEEDED
);
1180 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1184 * For hole or missing vdevs we just return success.
1186 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1189 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1190 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1191 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1197 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1199 if (vd
->vdev_children
== 0) {
1200 if (osize
< SPA_MINDEVSIZE
) {
1201 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1202 VDEV_AUX_TOO_SMALL
);
1206 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1208 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1209 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1210 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1211 VDEV_AUX_TOO_SMALL
);
1218 vd
->vdev_psize
= psize
;
1221 * Make sure the allocatable size hasn't shrunk.
1223 if (asize
< vd
->vdev_min_asize
) {
1224 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1225 VDEV_AUX_BAD_LABEL
);
1229 if (vd
->vdev_asize
== 0) {
1231 * This is the first-ever open, so use the computed values.
1232 * For testing purposes, a higher ashift can be requested.
1234 vd
->vdev_asize
= asize
;
1235 vd
->vdev_ashift
= MAX(ashift
, vd
->vdev_ashift
);
1238 * Make sure the alignment requirement hasn't increased.
1240 if (ashift
> vd
->vdev_top
->vdev_ashift
) {
1241 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1242 VDEV_AUX_BAD_LABEL
);
1248 * If all children are healthy and the asize has increased,
1249 * then we've experienced dynamic LUN growth. If automatic
1250 * expansion is enabled then use the additional space.
1252 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1253 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1254 vd
->vdev_asize
= asize
;
1256 vdev_set_min_asize(vd
);
1259 * Ensure we can issue some IO before declaring the
1260 * vdev open for business.
1262 if (vd
->vdev_ops
->vdev_op_leaf
&&
1263 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1264 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1265 VDEV_AUX_ERR_EXCEEDED
);
1270 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1271 * resilver. But don't do this if we are doing a reopen for a scrub,
1272 * since this would just restart the scrub we are already doing.
1274 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1275 vdev_resilver_needed(vd
, NULL
, NULL
))
1276 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1282 * Called once the vdevs are all opened, this routine validates the label
1283 * contents. This needs to be done before vdev_load() so that we don't
1284 * inadvertently do repair I/Os to the wrong device.
1286 * This function will only return failure if one of the vdevs indicates that it
1287 * has since been destroyed or exported. This is only possible if
1288 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1289 * will be updated but the function will return 0.
1292 vdev_validate(vdev_t
*vd
)
1294 spa_t
*spa
= vd
->vdev_spa
;
1296 uint64_t guid
= 0, top_guid
;
1299 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1300 if (vdev_validate(vd
->vdev_child
[c
]) != 0)
1304 * If the device has already failed, or was marked offline, don't do
1305 * any further validation. Otherwise, label I/O will fail and we will
1306 * overwrite the previous state.
1308 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1309 uint64_t aux_guid
= 0;
1312 if ((label
= vdev_label_read_config(vd
)) == NULL
) {
1313 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1314 VDEV_AUX_BAD_LABEL
);
1319 * Determine if this vdev has been split off into another
1320 * pool. If so, then refuse to open it.
1322 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1323 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1324 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1325 VDEV_AUX_SPLIT_POOL
);
1330 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
,
1331 &guid
) != 0 || guid
!= spa_guid(spa
)) {
1332 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1333 VDEV_AUX_CORRUPT_DATA
);
1338 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1339 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1344 * If this vdev just became a top-level vdev because its
1345 * sibling was detached, it will have adopted the parent's
1346 * vdev guid -- but the label may or may not be on disk yet.
1347 * Fortunately, either version of the label will have the
1348 * same top guid, so if we're a top-level vdev, we can
1349 * safely compare to that instead.
1351 * If we split this vdev off instead, then we also check the
1352 * original pool's guid. We don't want to consider the vdev
1353 * corrupt if it is partway through a split operation.
1355 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1357 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1359 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1360 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1361 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1362 VDEV_AUX_CORRUPT_DATA
);
1367 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1369 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1370 VDEV_AUX_CORRUPT_DATA
);
1378 * If spa->spa_load_verbatim is true, no need to check the
1379 * state of the pool.
1381 if (!spa
->spa_load_verbatim
&&
1382 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1383 state
!= POOL_STATE_ACTIVE
)
1387 * If we were able to open and validate a vdev that was
1388 * previously marked permanently unavailable, clear that state
1391 if (vd
->vdev_not_present
)
1392 vd
->vdev_not_present
= 0;
1399 * Close a virtual device.
1402 vdev_close(vdev_t
*vd
)
1404 spa_t
*spa
= vd
->vdev_spa
;
1405 vdev_t
*pvd
= vd
->vdev_parent
;
1407 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1410 * If our parent is reopening, then we are as well, unless we are
1413 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1414 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1416 vd
->vdev_ops
->vdev_op_close(vd
);
1418 vdev_cache_purge(vd
);
1421 * We record the previous state before we close it, so that if we are
1422 * doing a reopen(), we don't generate FMA ereports if we notice that
1423 * it's still faulted.
1425 vd
->vdev_prevstate
= vd
->vdev_state
;
1427 if (vd
->vdev_offline
)
1428 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1430 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1431 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1435 vdev_hold(vdev_t
*vd
)
1437 spa_t
*spa
= vd
->vdev_spa
;
1439 ASSERT(spa_is_root(spa
));
1440 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1443 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1444 vdev_hold(vd
->vdev_child
[c
]);
1446 if (vd
->vdev_ops
->vdev_op_leaf
)
1447 vd
->vdev_ops
->vdev_op_hold(vd
);
1451 vdev_rele(vdev_t
*vd
)
1453 spa_t
*spa
= vd
->vdev_spa
;
1455 ASSERT(spa_is_root(spa
));
1456 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1457 vdev_rele(vd
->vdev_child
[c
]);
1459 if (vd
->vdev_ops
->vdev_op_leaf
)
1460 vd
->vdev_ops
->vdev_op_rele(vd
);
1464 * Reopen all interior vdevs and any unopened leaves. We don't actually
1465 * reopen leaf vdevs which had previously been opened as they might deadlock
1466 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1467 * If the leaf has never been opened then open it, as usual.
1470 vdev_reopen(vdev_t
*vd
)
1472 spa_t
*spa
= vd
->vdev_spa
;
1474 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1476 /* set the reopening flag unless we're taking the vdev offline */
1477 vd
->vdev_reopening
= !vd
->vdev_offline
;
1479 (void) vdev_open(vd
);
1482 * Call vdev_validate() here to make sure we have the same device.
1483 * Otherwise, a device with an invalid label could be successfully
1484 * opened in response to vdev_reopen().
1487 (void) vdev_validate_aux(vd
);
1488 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1489 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1490 !l2arc_vdev_present(vd
))
1491 l2arc_add_vdev(spa
, vd
);
1493 (void) vdev_validate(vd
);
1497 * Reassess parent vdev's health.
1499 vdev_propagate_state(vd
);
1503 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1508 * Normally, partial opens (e.g. of a mirror) are allowed.
1509 * For a create, however, we want to fail the request if
1510 * there are any components we can't open.
1512 error
= vdev_open(vd
);
1514 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1516 return (error
? error
: ENXIO
);
1520 * Recursively initialize all labels.
1522 if ((error
= vdev_label_init(vd
, txg
, isreplacing
?
1523 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1532 vdev_metaslab_set_size(vdev_t
*vd
)
1535 * Aim for roughly 200 metaslabs per vdev.
1537 vd
->vdev_ms_shift
= highbit(vd
->vdev_asize
/ 200);
1538 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1542 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1544 ASSERT(vd
== vd
->vdev_top
);
1545 ASSERT(!vd
->vdev_ishole
);
1546 ASSERT(ISP2(flags
));
1548 if (flags
& VDD_METASLAB
)
1549 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1551 if (flags
& VDD_DTL
)
1552 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1554 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1560 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1561 * the vdev has less than perfect replication. There are four kinds of DTL:
1563 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1565 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1567 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1568 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1569 * txgs that was scrubbed.
1571 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1572 * persistent errors or just some device being offline.
1573 * Unlike the other three, the DTL_OUTAGE map is not generally
1574 * maintained; it's only computed when needed, typically to
1575 * determine whether a device can be detached.
1577 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1578 * either has the data or it doesn't.
1580 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1581 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1582 * if any child is less than fully replicated, then so is its parent.
1583 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1584 * comprising only those txgs which appear in 'maxfaults' or more children;
1585 * those are the txgs we don't have enough replication to read. For example,
1586 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1587 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1588 * two child DTL_MISSING maps.
1590 * It should be clear from the above that to compute the DTLs and outage maps
1591 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1592 * Therefore, that is all we keep on disk. When loading the pool, or after
1593 * a configuration change, we generate all other DTLs from first principles.
1596 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1598 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1600 ASSERT(t
< DTL_TYPES
);
1601 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1603 mutex_enter(sm
->sm_lock
);
1604 if (!space_map_contains(sm
, txg
, size
))
1605 space_map_add(sm
, txg
, size
);
1606 mutex_exit(sm
->sm_lock
);
1610 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1612 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1613 boolean_t dirty
= B_FALSE
;
1615 ASSERT(t
< DTL_TYPES
);
1616 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1618 mutex_enter(sm
->sm_lock
);
1619 if (sm
->sm_space
!= 0)
1620 dirty
= space_map_contains(sm
, txg
, size
);
1621 mutex_exit(sm
->sm_lock
);
1627 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1629 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1632 mutex_enter(sm
->sm_lock
);
1633 empty
= (sm
->sm_space
== 0);
1634 mutex_exit(sm
->sm_lock
);
1640 * Reassess DTLs after a config change or scrub completion.
1643 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1645 spa_t
*spa
= vd
->vdev_spa
;
1649 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1651 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1652 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1653 scrub_txg
, scrub_done
);
1655 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1658 if (vd
->vdev_ops
->vdev_op_leaf
) {
1659 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1661 mutex_enter(&vd
->vdev_dtl_lock
);
1662 if (scrub_txg
!= 0 &&
1663 (spa
->spa_scrub_started
||
1664 (scn
&& scn
->scn_phys
.scn_errors
== 0))) {
1666 * We completed a scrub up to scrub_txg. If we
1667 * did it without rebooting, then the scrub dtl
1668 * will be valid, so excise the old region and
1669 * fold in the scrub dtl. Otherwise, leave the
1670 * dtl as-is if there was an error.
1672 * There's little trick here: to excise the beginning
1673 * of the DTL_MISSING map, we put it into a reference
1674 * tree and then add a segment with refcnt -1 that
1675 * covers the range [0, scrub_txg). This means
1676 * that each txg in that range has refcnt -1 or 0.
1677 * We then add DTL_SCRUB with a refcnt of 2, so that
1678 * entries in the range [0, scrub_txg) will have a
1679 * positive refcnt -- either 1 or 2. We then convert
1680 * the reference tree into the new DTL_MISSING map.
1682 space_map_ref_create(&reftree
);
1683 space_map_ref_add_map(&reftree
,
1684 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1685 space_map_ref_add_seg(&reftree
, 0, scrub_txg
, -1);
1686 space_map_ref_add_map(&reftree
,
1687 &vd
->vdev_dtl
[DTL_SCRUB
], 2);
1688 space_map_ref_generate_map(&reftree
,
1689 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1690 space_map_ref_destroy(&reftree
);
1692 space_map_vacate(&vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1693 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1694 space_map_add
, &vd
->vdev_dtl
[DTL_PARTIAL
]);
1696 space_map_vacate(&vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1697 space_map_vacate(&vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1698 if (!vdev_readable(vd
))
1699 space_map_add(&vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1701 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1702 space_map_add
, &vd
->vdev_dtl
[DTL_OUTAGE
]);
1703 mutex_exit(&vd
->vdev_dtl_lock
);
1706 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1710 mutex_enter(&vd
->vdev_dtl_lock
);
1711 for (int t
= 0; t
< DTL_TYPES
; t
++) {
1712 /* account for child's outage in parent's missing map */
1713 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1715 continue; /* leaf vdevs only */
1716 if (t
== DTL_PARTIAL
)
1717 minref
= 1; /* i.e. non-zero */
1718 else if (vd
->vdev_nparity
!= 0)
1719 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1721 minref
= vd
->vdev_children
; /* any kind of mirror */
1722 space_map_ref_create(&reftree
);
1723 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1724 vdev_t
*cvd
= vd
->vdev_child
[c
];
1725 mutex_enter(&cvd
->vdev_dtl_lock
);
1726 space_map_ref_add_map(&reftree
, &cvd
->vdev_dtl
[s
], 1);
1727 mutex_exit(&cvd
->vdev_dtl_lock
);
1729 space_map_ref_generate_map(&reftree
, &vd
->vdev_dtl
[t
], minref
);
1730 space_map_ref_destroy(&reftree
);
1732 mutex_exit(&vd
->vdev_dtl_lock
);
1736 vdev_dtl_load(vdev_t
*vd
)
1738 spa_t
*spa
= vd
->vdev_spa
;
1739 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1740 objset_t
*mos
= spa
->spa_meta_objset
;
1744 ASSERT(vd
->vdev_children
== 0);
1746 if (smo
->smo_object
== 0)
1749 ASSERT(!vd
->vdev_ishole
);
1751 if ((error
= dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
)) != 0)
1754 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1755 bcopy(db
->db_data
, smo
, sizeof (*smo
));
1756 dmu_buf_rele(db
, FTAG
);
1758 mutex_enter(&vd
->vdev_dtl_lock
);
1759 error
= space_map_load(&vd
->vdev_dtl
[DTL_MISSING
],
1760 NULL
, SM_ALLOC
, smo
, mos
);
1761 mutex_exit(&vd
->vdev_dtl_lock
);
1767 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1769 spa_t
*spa
= vd
->vdev_spa
;
1770 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1771 space_map_t
*sm
= &vd
->vdev_dtl
[DTL_MISSING
];
1772 objset_t
*mos
= spa
->spa_meta_objset
;
1778 ASSERT(!vd
->vdev_ishole
);
1780 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1782 if (vd
->vdev_detached
) {
1783 if (smo
->smo_object
!= 0) {
1784 int err
= dmu_object_free(mos
, smo
->smo_object
, tx
);
1785 ASSERT3U(err
, ==, 0);
1786 smo
->smo_object
= 0;
1792 if (smo
->smo_object
== 0) {
1793 ASSERT(smo
->smo_objsize
== 0);
1794 ASSERT(smo
->smo_alloc
== 0);
1795 smo
->smo_object
= dmu_object_alloc(mos
,
1796 DMU_OT_SPACE_MAP
, 1 << SPACE_MAP_BLOCKSHIFT
,
1797 DMU_OT_SPACE_MAP_HEADER
, sizeof (*smo
), tx
);
1798 ASSERT(smo
->smo_object
!= 0);
1799 vdev_config_dirty(vd
->vdev_top
);
1802 mutex_init(&smlock
, NULL
, MUTEX_DEFAULT
, NULL
);
1804 space_map_create(&smsync
, sm
->sm_start
, sm
->sm_size
, sm
->sm_shift
,
1807 mutex_enter(&smlock
);
1809 mutex_enter(&vd
->vdev_dtl_lock
);
1810 space_map_walk(sm
, space_map_add
, &smsync
);
1811 mutex_exit(&vd
->vdev_dtl_lock
);
1813 space_map_truncate(smo
, mos
, tx
);
1814 space_map_sync(&smsync
, SM_ALLOC
, smo
, mos
, tx
);
1816 space_map_destroy(&smsync
);
1818 mutex_exit(&smlock
);
1819 mutex_destroy(&smlock
);
1821 VERIFY(0 == dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
));
1822 dmu_buf_will_dirty(db
, tx
);
1823 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1824 bcopy(smo
, db
->db_data
, sizeof (*smo
));
1825 dmu_buf_rele(db
, FTAG
);
1831 * Determine whether the specified vdev can be offlined/detached/removed
1832 * without losing data.
1835 vdev_dtl_required(vdev_t
*vd
)
1837 spa_t
*spa
= vd
->vdev_spa
;
1838 vdev_t
*tvd
= vd
->vdev_top
;
1839 uint8_t cant_read
= vd
->vdev_cant_read
;
1842 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1844 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
1848 * Temporarily mark the device as unreadable, and then determine
1849 * whether this results in any DTL outages in the top-level vdev.
1850 * If not, we can safely offline/detach/remove the device.
1852 vd
->vdev_cant_read
= B_TRUE
;
1853 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1854 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
1855 vd
->vdev_cant_read
= cant_read
;
1856 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1862 * Determine if resilver is needed, and if so the txg range.
1865 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
1867 boolean_t needed
= B_FALSE
;
1868 uint64_t thismin
= UINT64_MAX
;
1869 uint64_t thismax
= 0;
1871 if (vd
->vdev_children
== 0) {
1872 mutex_enter(&vd
->vdev_dtl_lock
);
1873 if (vd
->vdev_dtl
[DTL_MISSING
].sm_space
!= 0 &&
1874 vdev_writeable(vd
)) {
1877 ss
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1878 thismin
= ss
->ss_start
- 1;
1879 ss
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1880 thismax
= ss
->ss_end
;
1883 mutex_exit(&vd
->vdev_dtl_lock
);
1885 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1886 vdev_t
*cvd
= vd
->vdev_child
[c
];
1887 uint64_t cmin
, cmax
;
1889 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
1890 thismin
= MIN(thismin
, cmin
);
1891 thismax
= MAX(thismax
, cmax
);
1897 if (needed
&& minp
) {
1905 vdev_load(vdev_t
*vd
)
1908 * Recursively load all children.
1910 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1911 vdev_load(vd
->vdev_child
[c
]);
1914 * If this is a top-level vdev, initialize its metaslabs.
1916 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
1917 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
1918 vdev_metaslab_init(vd
, 0) != 0))
1919 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1920 VDEV_AUX_CORRUPT_DATA
);
1923 * If this is a leaf vdev, load its DTL.
1925 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
1926 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1927 VDEV_AUX_CORRUPT_DATA
);
1931 * The special vdev case is used for hot spares and l2cache devices. Its
1932 * sole purpose it to set the vdev state for the associated vdev. To do this,
1933 * we make sure that we can open the underlying device, then try to read the
1934 * label, and make sure that the label is sane and that it hasn't been
1935 * repurposed to another pool.
1938 vdev_validate_aux(vdev_t
*vd
)
1941 uint64_t guid
, version
;
1944 if (!vdev_readable(vd
))
1947 if ((label
= vdev_label_read_config(vd
)) == NULL
) {
1948 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1949 VDEV_AUX_CORRUPT_DATA
);
1953 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
1954 version
> SPA_VERSION
||
1955 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
1956 guid
!= vd
->vdev_guid
||
1957 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
1958 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1959 VDEV_AUX_CORRUPT_DATA
);
1965 * We don't actually check the pool state here. If it's in fact in
1966 * use by another pool, we update this fact on the fly when requested.
1973 vdev_remove(vdev_t
*vd
, uint64_t txg
)
1975 spa_t
*spa
= vd
->vdev_spa
;
1976 objset_t
*mos
= spa
->spa_meta_objset
;
1979 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
1981 if (vd
->vdev_dtl_smo
.smo_object
) {
1982 ASSERT3U(vd
->vdev_dtl_smo
.smo_alloc
, ==, 0);
1983 (void) dmu_object_free(mos
, vd
->vdev_dtl_smo
.smo_object
, tx
);
1984 vd
->vdev_dtl_smo
.smo_object
= 0;
1987 if (vd
->vdev_ms
!= NULL
) {
1988 for (int m
= 0; m
< vd
->vdev_ms_count
; m
++) {
1989 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1991 if (msp
== NULL
|| msp
->ms_smo
.smo_object
== 0)
1994 ASSERT3U(msp
->ms_smo
.smo_alloc
, ==, 0);
1995 (void) dmu_object_free(mos
, msp
->ms_smo
.smo_object
, tx
);
1996 msp
->ms_smo
.smo_object
= 0;
2000 if (vd
->vdev_ms_array
) {
2001 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2002 vd
->vdev_ms_array
= 0;
2003 vd
->vdev_ms_shift
= 0;
2009 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2012 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2014 ASSERT(!vd
->vdev_ishole
);
2016 while (msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
2017 metaslab_sync_done(msp
, txg
);
2020 metaslab_sync_reassess(vd
->vdev_mg
);
2024 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2026 spa_t
*spa
= vd
->vdev_spa
;
2031 ASSERT(!vd
->vdev_ishole
);
2033 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2034 ASSERT(vd
== vd
->vdev_top
);
2035 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2036 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2037 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2038 ASSERT(vd
->vdev_ms_array
!= 0);
2039 vdev_config_dirty(vd
);
2044 * Remove the metadata associated with this vdev once it's empty.
2046 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2047 vdev_remove(vd
, txg
);
2049 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2050 metaslab_sync(msp
, txg
);
2051 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2054 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2055 vdev_dtl_sync(lvd
, txg
);
2057 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2061 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2063 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2067 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2068 * not be opened, and no I/O is attempted.
2071 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2075 spa_vdev_state_enter(spa
, SCL_NONE
);
2077 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2078 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2080 if (!vd
->vdev_ops
->vdev_op_leaf
)
2081 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2084 * We don't directly use the aux state here, but if we do a
2085 * vdev_reopen(), we need this value to be present to remember why we
2088 vd
->vdev_label_aux
= aux
;
2091 * Faulted state takes precedence over degraded.
2093 vd
->vdev_delayed_close
= B_FALSE
;
2094 vd
->vdev_faulted
= 1ULL;
2095 vd
->vdev_degraded
= 0ULL;
2096 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2099 * If this device has the only valid copy of the data, then
2100 * back off and simply mark the vdev as degraded instead.
2102 if (!vd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2103 vd
->vdev_degraded
= 1ULL;
2104 vd
->vdev_faulted
= 0ULL;
2107 * If we reopen the device and it's not dead, only then do we
2112 if (vdev_readable(vd
))
2113 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2116 return (spa_vdev_state_exit(spa
, vd
, 0));
2120 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2121 * user that something is wrong. The vdev continues to operate as normal as far
2122 * as I/O is concerned.
2125 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2129 spa_vdev_state_enter(spa
, SCL_NONE
);
2131 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2132 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2134 if (!vd
->vdev_ops
->vdev_op_leaf
)
2135 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2138 * If the vdev is already faulted, then don't do anything.
2140 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2141 return (spa_vdev_state_exit(spa
, NULL
, 0));
2143 vd
->vdev_degraded
= 1ULL;
2144 if (!vdev_is_dead(vd
))
2145 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2148 return (spa_vdev_state_exit(spa
, vd
, 0));
2152 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2153 * any attached spare device should be detached when the device finishes
2154 * resilvering. Second, the online should be treated like a 'test' online case,
2155 * so no FMA events are generated if the device fails to open.
2158 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2160 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2162 spa_vdev_state_enter(spa
, SCL_NONE
);
2164 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2165 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2167 if (!vd
->vdev_ops
->vdev_op_leaf
)
2168 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2171 vd
->vdev_offline
= B_FALSE
;
2172 vd
->vdev_tmpoffline
= B_FALSE
;
2173 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2174 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2176 /* XXX - L2ARC 1.0 does not support expansion */
2177 if (!vd
->vdev_aux
) {
2178 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2179 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2183 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2185 if (!vd
->vdev_aux
) {
2186 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2187 pvd
->vdev_expanding
= B_FALSE
;
2191 *newstate
= vd
->vdev_state
;
2192 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2193 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2194 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2195 vd
->vdev_parent
->vdev_child
[0] == vd
)
2196 vd
->vdev_unspare
= B_TRUE
;
2198 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2200 /* XXX - L2ARC 1.0 does not support expansion */
2202 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2203 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2205 return (spa_vdev_state_exit(spa
, vd
, 0));
2209 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2213 uint64_t generation
;
2214 metaslab_group_t
*mg
;
2217 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2219 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2220 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2222 if (!vd
->vdev_ops
->vdev_op_leaf
)
2223 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2227 generation
= spa
->spa_config_generation
+ 1;
2230 * If the device isn't already offline, try to offline it.
2232 if (!vd
->vdev_offline
) {
2234 * If this device has the only valid copy of some data,
2235 * don't allow it to be offlined. Log devices are always
2238 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2239 vdev_dtl_required(vd
))
2240 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2243 * If the top-level is a slog and it has had allocations
2244 * then proceed. We check that the vdev's metaslab group
2245 * is not NULL since it's possible that we may have just
2246 * added this vdev but not yet initialized its metaslabs.
2248 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2250 * Prevent any future allocations.
2252 metaslab_group_passivate(mg
);
2253 (void) spa_vdev_state_exit(spa
, vd
, 0);
2255 error
= spa_offline_log(spa
);
2257 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2260 * Check to see if the config has changed.
2262 if (error
|| generation
!= spa
->spa_config_generation
) {
2263 metaslab_group_activate(mg
);
2265 return (spa_vdev_state_exit(spa
,
2267 (void) spa_vdev_state_exit(spa
, vd
, 0);
2270 ASSERT3U(tvd
->vdev_stat
.vs_alloc
, ==, 0);
2274 * Offline this device and reopen its top-level vdev.
2275 * If the top-level vdev is a log device then just offline
2276 * it. Otherwise, if this action results in the top-level
2277 * vdev becoming unusable, undo it and fail the request.
2279 vd
->vdev_offline
= B_TRUE
;
2282 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2283 vdev_is_dead(tvd
)) {
2284 vd
->vdev_offline
= B_FALSE
;
2286 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2290 * Add the device back into the metaslab rotor so that
2291 * once we online the device it's open for business.
2293 if (tvd
->vdev_islog
&& mg
!= NULL
)
2294 metaslab_group_activate(mg
);
2297 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2299 return (spa_vdev_state_exit(spa
, vd
, 0));
2303 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2307 mutex_enter(&spa
->spa_vdev_top_lock
);
2308 error
= vdev_offline_locked(spa
, guid
, flags
);
2309 mutex_exit(&spa
->spa_vdev_top_lock
);
2315 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2316 * vdev_offline(), we assume the spa config is locked. We also clear all
2317 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2320 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2322 vdev_t
*rvd
= spa
->spa_root_vdev
;
2324 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2329 vd
->vdev_stat
.vs_read_errors
= 0;
2330 vd
->vdev_stat
.vs_write_errors
= 0;
2331 vd
->vdev_stat
.vs_checksum_errors
= 0;
2333 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2334 vdev_clear(spa
, vd
->vdev_child
[c
]);
2337 * If we're in the FAULTED state or have experienced failed I/O, then
2338 * clear the persistent state and attempt to reopen the device. We
2339 * also mark the vdev config dirty, so that the new faulted state is
2340 * written out to disk.
2342 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2343 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2346 * When reopening in reponse to a clear event, it may be due to
2347 * a fmadm repair request. In this case, if the device is
2348 * still broken, we want to still post the ereport again.
2350 vd
->vdev_forcefault
= B_TRUE
;
2352 vd
->vdev_faulted
= vd
->vdev_degraded
= 0;
2353 vd
->vdev_cant_read
= B_FALSE
;
2354 vd
->vdev_cant_write
= B_FALSE
;
2358 vd
->vdev_forcefault
= B_FALSE
;
2361 vdev_state_dirty(vd
->vdev_top
);
2363 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2364 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2366 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2370 * When clearing a FMA-diagnosed fault, we always want to
2371 * unspare the device, as we assume that the original spare was
2372 * done in response to the FMA fault.
2374 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2375 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2376 vd
->vdev_parent
->vdev_child
[0] == vd
)
2377 vd
->vdev_unspare
= B_TRUE
;
2381 vdev_is_dead(vdev_t
*vd
)
2384 * Holes and missing devices are always considered "dead".
2385 * This simplifies the code since we don't have to check for
2386 * these types of devices in the various code paths.
2387 * Instead we rely on the fact that we skip over dead devices
2388 * before issuing I/O to them.
2390 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2391 vd
->vdev_ops
== &vdev_missing_ops
);
2395 vdev_readable(vdev_t
*vd
)
2397 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2401 vdev_writeable(vdev_t
*vd
)
2403 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2407 vdev_allocatable(vdev_t
*vd
)
2409 uint64_t state
= vd
->vdev_state
;
2412 * We currently allow allocations from vdevs which may be in the
2413 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2414 * fails to reopen then we'll catch it later when we're holding
2415 * the proper locks. Note that we have to get the vdev state
2416 * in a local variable because although it changes atomically,
2417 * we're asking two separate questions about it.
2419 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2420 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2424 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2426 ASSERT(zio
->io_vd
== vd
);
2428 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2431 if (zio
->io_type
== ZIO_TYPE_READ
)
2432 return (!vd
->vdev_cant_read
);
2434 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2435 return (!vd
->vdev_cant_write
);
2441 * Get statistics for the given vdev.
2444 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2446 vdev_t
*rvd
= vd
->vdev_spa
->spa_root_vdev
;
2448 mutex_enter(&vd
->vdev_stat_lock
);
2449 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2450 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2451 vs
->vs_state
= vd
->vdev_state
;
2452 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2453 if (vd
->vdev_ops
->vdev_op_leaf
)
2454 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2455 mutex_exit(&vd
->vdev_stat_lock
);
2458 * If we're getting stats on the root vdev, aggregate the I/O counts
2459 * over all top-level vdevs (i.e. the direct children of the root).
2462 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2463 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2464 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2466 mutex_enter(&vd
->vdev_stat_lock
);
2467 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
2468 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2469 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2471 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2472 mutex_exit(&vd
->vdev_stat_lock
);
2478 vdev_clear_stats(vdev_t
*vd
)
2480 mutex_enter(&vd
->vdev_stat_lock
);
2481 vd
->vdev_stat
.vs_space
= 0;
2482 vd
->vdev_stat
.vs_dspace
= 0;
2483 vd
->vdev_stat
.vs_alloc
= 0;
2484 mutex_exit(&vd
->vdev_stat_lock
);
2488 vdev_scan_stat_init(vdev_t
*vd
)
2490 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2492 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2493 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2495 mutex_enter(&vd
->vdev_stat_lock
);
2496 vs
->vs_scan_processed
= 0;
2497 mutex_exit(&vd
->vdev_stat_lock
);
2501 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2503 spa_t
*spa
= zio
->io_spa
;
2504 vdev_t
*rvd
= spa
->spa_root_vdev
;
2505 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2507 uint64_t txg
= zio
->io_txg
;
2508 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2509 zio_type_t type
= zio
->io_type
;
2510 int flags
= zio
->io_flags
;
2513 * If this i/o is a gang leader, it didn't do any actual work.
2515 if (zio
->io_gang_tree
)
2518 if (zio
->io_error
== 0) {
2520 * If this is a root i/o, don't count it -- we've already
2521 * counted the top-level vdevs, and vdev_get_stats() will
2522 * aggregate them when asked. This reduces contention on
2523 * the root vdev_stat_lock and implicitly handles blocks
2524 * that compress away to holes, for which there is no i/o.
2525 * (Holes never create vdev children, so all the counters
2526 * remain zero, which is what we want.)
2528 * Note: this only applies to successful i/o (io_error == 0)
2529 * because unlike i/o counts, errors are not additive.
2530 * When reading a ditto block, for example, failure of
2531 * one top-level vdev does not imply a root-level error.
2536 ASSERT(vd
== zio
->io_vd
);
2538 if (flags
& ZIO_FLAG_IO_BYPASS
)
2541 mutex_enter(&vd
->vdev_stat_lock
);
2543 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2544 if (flags
& ZIO_FLAG_SCRUB_THREAD
) {
2545 dsl_scan_phys_t
*scn_phys
=
2546 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2547 uint64_t *processed
= &scn_phys
->scn_processed
;
2550 if (vd
->vdev_ops
->vdev_op_leaf
)
2551 atomic_add_64(processed
, psize
);
2552 vs
->vs_scan_processed
+= psize
;
2555 if (flags
& ZIO_FLAG_SELF_HEAL
)
2556 vs
->vs_self_healed
+= psize
;
2560 vs
->vs_bytes
[type
] += psize
;
2562 mutex_exit(&vd
->vdev_stat_lock
);
2566 if (flags
& ZIO_FLAG_SPECULATIVE
)
2570 * If this is an I/O error that is going to be retried, then ignore the
2571 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2572 * hard errors, when in reality they can happen for any number of
2573 * innocuous reasons (bus resets, MPxIO link failure, etc).
2575 if (zio
->io_error
== EIO
&&
2576 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2580 * Intent logs writes won't propagate their error to the root
2581 * I/O so don't mark these types of failures as pool-level
2584 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2587 mutex_enter(&vd
->vdev_stat_lock
);
2588 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2589 if (zio
->io_error
== ECKSUM
)
2590 vs
->vs_checksum_errors
++;
2592 vs
->vs_read_errors
++;
2594 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2595 vs
->vs_write_errors
++;
2596 mutex_exit(&vd
->vdev_stat_lock
);
2598 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2599 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2600 (flags
& ZIO_FLAG_SCRUB_THREAD
) ||
2601 spa
->spa_claiming
)) {
2603 * This is either a normal write (not a repair), or it's
2604 * a repair induced by the scrub thread, or it's a repair
2605 * made by zil_claim() during spa_load() in the first txg.
2606 * In the normal case, we commit the DTL change in the same
2607 * txg as the block was born. In the scrub-induced repair
2608 * case, we know that scrubs run in first-pass syncing context,
2609 * so we commit the DTL change in spa_syncing_txg(spa).
2610 * In the zil_claim() case, we commit in spa_first_txg(spa).
2612 * We currently do not make DTL entries for failed spontaneous
2613 * self-healing writes triggered by normal (non-scrubbing)
2614 * reads, because we have no transactional context in which to
2615 * do so -- and it's not clear that it'd be desirable anyway.
2617 if (vd
->vdev_ops
->vdev_op_leaf
) {
2618 uint64_t commit_txg
= txg
;
2619 if (flags
& ZIO_FLAG_SCRUB_THREAD
) {
2620 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2621 ASSERT(spa_sync_pass(spa
) == 1);
2622 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2623 commit_txg
= spa_syncing_txg(spa
);
2624 } else if (spa
->spa_claiming
) {
2625 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2626 commit_txg
= spa_first_txg(spa
);
2628 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2629 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2631 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2632 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2633 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2636 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2641 * Update the in-core space usage stats for this vdev, its metaslab class,
2642 * and the root vdev.
2645 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2646 int64_t space_delta
)
2648 int64_t dspace_delta
= space_delta
;
2649 spa_t
*spa
= vd
->vdev_spa
;
2650 vdev_t
*rvd
= spa
->spa_root_vdev
;
2651 metaslab_group_t
*mg
= vd
->vdev_mg
;
2652 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2654 ASSERT(vd
== vd
->vdev_top
);
2657 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2658 * factor. We must calculate this here and not at the root vdev
2659 * because the root vdev's psize-to-asize is simply the max of its
2660 * childrens', thus not accurate enough for us.
2662 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2663 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2664 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2665 vd
->vdev_deflate_ratio
;
2667 mutex_enter(&vd
->vdev_stat_lock
);
2668 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2669 vd
->vdev_stat
.vs_space
+= space_delta
;
2670 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2671 mutex_exit(&vd
->vdev_stat_lock
);
2673 if (mc
== spa_normal_class(spa
)) {
2674 mutex_enter(&rvd
->vdev_stat_lock
);
2675 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2676 rvd
->vdev_stat
.vs_space
+= space_delta
;
2677 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2678 mutex_exit(&rvd
->vdev_stat_lock
);
2682 ASSERT(rvd
== vd
->vdev_parent
);
2683 ASSERT(vd
->vdev_ms_count
!= 0);
2685 metaslab_class_space_update(mc
,
2686 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2691 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2692 * so that it will be written out next time the vdev configuration is synced.
2693 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2696 vdev_config_dirty(vdev_t
*vd
)
2698 spa_t
*spa
= vd
->vdev_spa
;
2699 vdev_t
*rvd
= spa
->spa_root_vdev
;
2703 * If this is an aux vdev (as with l2cache and spare devices), then we
2704 * update the vdev config manually and set the sync flag.
2706 if (vd
->vdev_aux
!= NULL
) {
2707 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2711 for (c
= 0; c
< sav
->sav_count
; c
++) {
2712 if (sav
->sav_vdevs
[c
] == vd
)
2716 if (c
== sav
->sav_count
) {
2718 * We're being removed. There's nothing more to do.
2720 ASSERT(sav
->sav_sync
== B_TRUE
);
2724 sav
->sav_sync
= B_TRUE
;
2726 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2727 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2728 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2729 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2735 * Setting the nvlist in the middle if the array is a little
2736 * sketchy, but it will work.
2738 nvlist_free(aux
[c
]);
2739 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
2745 * The dirty list is protected by the SCL_CONFIG lock. The caller
2746 * must either hold SCL_CONFIG as writer, or must be the sync thread
2747 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2748 * so this is sufficient to ensure mutual exclusion.
2750 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2751 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2752 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2755 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2756 vdev_config_dirty(rvd
->vdev_child
[c
]);
2758 ASSERT(vd
== vd
->vdev_top
);
2760 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
2762 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
2767 vdev_config_clean(vdev_t
*vd
)
2769 spa_t
*spa
= vd
->vdev_spa
;
2771 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2772 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2773 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2775 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
2776 list_remove(&spa
->spa_config_dirty_list
, vd
);
2780 * Mark a top-level vdev's state as dirty, so that the next pass of
2781 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2782 * the state changes from larger config changes because they require
2783 * much less locking, and are often needed for administrative actions.
2786 vdev_state_dirty(vdev_t
*vd
)
2788 spa_t
*spa
= vd
->vdev_spa
;
2790 ASSERT(vd
== vd
->vdev_top
);
2793 * The state list is protected by the SCL_STATE lock. The caller
2794 * must either hold SCL_STATE as writer, or must be the sync thread
2795 * (which holds SCL_STATE as reader). There's only one sync thread,
2796 * so this is sufficient to ensure mutual exclusion.
2798 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2799 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2800 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2802 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
2803 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
2807 vdev_state_clean(vdev_t
*vd
)
2809 spa_t
*spa
= vd
->vdev_spa
;
2811 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2812 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2813 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2815 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
2816 list_remove(&spa
->spa_state_dirty_list
, vd
);
2820 * Propagate vdev state up from children to parent.
2823 vdev_propagate_state(vdev_t
*vd
)
2825 spa_t
*spa
= vd
->vdev_spa
;
2826 vdev_t
*rvd
= spa
->spa_root_vdev
;
2827 int degraded
= 0, faulted
= 0;
2831 if (vd
->vdev_children
> 0) {
2832 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2833 child
= vd
->vdev_child
[c
];
2836 * Don't factor holes into the decision.
2838 if (child
->vdev_ishole
)
2841 if (!vdev_readable(child
) ||
2842 (!vdev_writeable(child
) && spa_writeable(spa
))) {
2844 * Root special: if there is a top-level log
2845 * device, treat the root vdev as if it were
2848 if (child
->vdev_islog
&& vd
== rvd
)
2852 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
2856 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
2860 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
2863 * Root special: if there is a top-level vdev that cannot be
2864 * opened due to corrupted metadata, then propagate the root
2865 * vdev's aux state as 'corrupt' rather than 'insufficient
2868 if (corrupted
&& vd
== rvd
&&
2869 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
2870 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2871 VDEV_AUX_CORRUPT_DATA
);
2874 if (vd
->vdev_parent
)
2875 vdev_propagate_state(vd
->vdev_parent
);
2879 * Set a vdev's state. If this is during an open, we don't update the parent
2880 * state, because we're in the process of opening children depth-first.
2881 * Otherwise, we propagate the change to the parent.
2883 * If this routine places a device in a faulted state, an appropriate ereport is
2887 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
2889 uint64_t save_state
;
2890 spa_t
*spa
= vd
->vdev_spa
;
2892 if (state
== vd
->vdev_state
) {
2893 vd
->vdev_stat
.vs_aux
= aux
;
2897 save_state
= vd
->vdev_state
;
2899 vd
->vdev_state
= state
;
2900 vd
->vdev_stat
.vs_aux
= aux
;
2903 * If we are setting the vdev state to anything but an open state, then
2904 * always close the underlying device unless the device has requested
2905 * a delayed close (i.e. we're about to remove or fault the device).
2906 * Otherwise, we keep accessible but invalid devices open forever.
2907 * We don't call vdev_close() itself, because that implies some extra
2908 * checks (offline, etc) that we don't want here. This is limited to
2909 * leaf devices, because otherwise closing the device will affect other
2912 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
2913 vd
->vdev_ops
->vdev_op_leaf
)
2914 vd
->vdev_ops
->vdev_op_close(vd
);
2917 * If we have brought this vdev back into service, we need
2918 * to notify fmd so that it can gracefully repair any outstanding
2919 * cases due to a missing device. We do this in all cases, even those
2920 * that probably don't correlate to a repaired fault. This is sure to
2921 * catch all cases, and we let the zfs-retire agent sort it out. If
2922 * this is a transient state it's OK, as the retire agent will
2923 * double-check the state of the vdev before repairing it.
2925 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
2926 vd
->vdev_prevstate
!= state
)
2927 zfs_post_state_change(spa
, vd
);
2929 if (vd
->vdev_removed
&&
2930 state
== VDEV_STATE_CANT_OPEN
&&
2931 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
2933 * If the previous state is set to VDEV_STATE_REMOVED, then this
2934 * device was previously marked removed and someone attempted to
2935 * reopen it. If this failed due to a nonexistent device, then
2936 * keep the device in the REMOVED state. We also let this be if
2937 * it is one of our special test online cases, which is only
2938 * attempting to online the device and shouldn't generate an FMA
2941 vd
->vdev_state
= VDEV_STATE_REMOVED
;
2942 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2943 } else if (state
== VDEV_STATE_REMOVED
) {
2944 vd
->vdev_removed
= B_TRUE
;
2945 } else if (state
== VDEV_STATE_CANT_OPEN
) {
2947 * If we fail to open a vdev during an import, we mark it as
2948 * "not available", which signifies that it was never there to
2949 * begin with. Failure to open such a device is not considered
2952 if (spa_load_state(spa
) == SPA_LOAD_IMPORT
&&
2953 vd
->vdev_ops
->vdev_op_leaf
)
2954 vd
->vdev_not_present
= 1;
2957 * Post the appropriate ereport. If the 'prevstate' field is
2958 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2959 * that this is part of a vdev_reopen(). In this case, we don't
2960 * want to post the ereport if the device was already in the
2961 * CANT_OPEN state beforehand.
2963 * If the 'checkremove' flag is set, then this is an attempt to
2964 * online the device in response to an insertion event. If we
2965 * hit this case, then we have detected an insertion event for a
2966 * faulted or offline device that wasn't in the removed state.
2967 * In this scenario, we don't post an ereport because we are
2968 * about to replace the device, or attempt an online with
2969 * vdev_forcefault, which will generate the fault for us.
2971 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
2972 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
2973 vd
!= spa
->spa_root_vdev
) {
2977 case VDEV_AUX_OPEN_FAILED
:
2978 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
2980 case VDEV_AUX_CORRUPT_DATA
:
2981 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
2983 case VDEV_AUX_NO_REPLICAS
:
2984 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
2986 case VDEV_AUX_BAD_GUID_SUM
:
2987 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
2989 case VDEV_AUX_TOO_SMALL
:
2990 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
2992 case VDEV_AUX_BAD_LABEL
:
2993 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
2996 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
2999 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3002 /* Erase any notion of persistent removed state */
3003 vd
->vdev_removed
= B_FALSE
;
3005 vd
->vdev_removed
= B_FALSE
;
3008 if (!isopen
&& vd
->vdev_parent
)
3009 vdev_propagate_state(vd
->vdev_parent
);
3013 * Check the vdev configuration to ensure that it's capable of supporting
3014 * a root pool. Currently, we do not support RAID-Z or partial configuration.
3015 * In addition, only a single top-level vdev is allowed and none of the leaves
3016 * can be wholedisks.
3019 vdev_is_bootable(vdev_t
*vd
)
3021 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3022 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3024 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3025 vd
->vdev_children
> 1) {
3027 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3028 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3031 } else if (vd
->vdev_wholedisk
== 1) {
3035 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3036 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3043 * Load the state from the original vdev tree (ovd) which
3044 * we've retrieved from the MOS config object. If the original
3045 * vdev was offline then we transfer that state to the device
3046 * in the current vdev tree (nvd).
3049 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3051 spa_t
*spa
= nvd
->vdev_spa
;
3053 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3054 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3056 for (int c
= 0; c
< nvd
->vdev_children
; c
++)
3057 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3059 if (nvd
->vdev_ops
->vdev_op_leaf
&& ovd
->vdev_offline
) {
3061 * It would be nice to call vdev_offline()
3062 * directly but the pool isn't fully loaded and
3063 * the txg threads have not been started yet.
3065 nvd
->vdev_offline
= ovd
->vdev_offline
;
3066 vdev_reopen(nvd
->vdev_top
);
3071 * Expand a vdev if possible.
3074 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3076 ASSERT(vd
->vdev_top
== vd
);
3077 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3079 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3080 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3081 vdev_config_dirty(vd
);
3089 vdev_split(vdev_t
*vd
)
3091 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3093 vdev_remove_child(pvd
, vd
);
3094 vdev_compact_children(pvd
);
3096 cvd
= pvd
->vdev_child
[0];
3097 if (pvd
->vdev_children
== 1) {
3098 vdev_remove_parent(cvd
);
3099 cvd
->vdev_splitting
= B_TRUE
;
3101 vdev_propagate_state(cvd
);