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) 2012 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>
47 * Virtual device management.
50 static vdev_ops_t
*vdev_ops_table
[] = {
63 /* maximum scrub/resilver I/O queue per leaf vdev */
64 int zfs_scrub_limit
= 10;
67 * Given a vdev type, return the appropriate ops vector.
70 vdev_getops(const char *type
)
72 vdev_ops_t
*ops
, **opspp
;
74 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
75 if (strcmp(ops
->vdev_op_type
, type
) == 0)
82 * Default asize function: return the MAX of psize with the asize of
83 * all children. This is what's used by anything other than RAID-Z.
86 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
88 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
91 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
92 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
93 asize
= MAX(asize
, csize
);
100 * Get the minimum allocatable size. We define the allocatable size as
101 * the vdev's asize rounded to the nearest metaslab. This allows us to
102 * replace or attach devices which don't have the same physical size but
103 * can still satisfy the same number of allocations.
106 vdev_get_min_asize(vdev_t
*vd
)
108 vdev_t
*pvd
= vd
->vdev_parent
;
111 * If our parent is NULL (inactive spare or cache) or is the root,
112 * just return our own asize.
115 return (vd
->vdev_asize
);
118 * The top-level vdev just returns the allocatable size rounded
119 * to the nearest metaslab.
121 if (vd
== vd
->vdev_top
)
122 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
125 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
126 * so each child must provide at least 1/Nth of its asize.
128 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
129 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
131 return (pvd
->vdev_min_asize
);
135 vdev_set_min_asize(vdev_t
*vd
)
137 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
139 for (int 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
)
163 if (vd
->vdev_guid
== guid
)
166 for (int c
= 0; c
< vd
->vdev_children
; c
++)
167 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
175 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
177 size_t oldsize
, newsize
;
178 uint64_t id
= cvd
->vdev_id
;
181 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
182 ASSERT(cvd
->vdev_parent
== NULL
);
184 cvd
->vdev_parent
= pvd
;
189 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
191 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
192 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
193 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
195 newchild
= kmem_zalloc(newsize
, KM_SLEEP
);
196 if (pvd
->vdev_child
!= NULL
) {
197 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
198 kmem_free(pvd
->vdev_child
, oldsize
);
201 pvd
->vdev_child
= newchild
;
202 pvd
->vdev_child
[id
] = cvd
;
204 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
205 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
208 * Walk up all ancestors to update guid sum.
210 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
211 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
215 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
218 uint_t id
= cvd
->vdev_id
;
220 ASSERT(cvd
->vdev_parent
== pvd
);
225 ASSERT(id
< pvd
->vdev_children
);
226 ASSERT(pvd
->vdev_child
[id
] == cvd
);
228 pvd
->vdev_child
[id
] = NULL
;
229 cvd
->vdev_parent
= NULL
;
231 for (c
= 0; c
< pvd
->vdev_children
; c
++)
232 if (pvd
->vdev_child
[c
])
235 if (c
== pvd
->vdev_children
) {
236 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
237 pvd
->vdev_child
= NULL
;
238 pvd
->vdev_children
= 0;
242 * Walk up all ancestors to update guid sum.
244 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
245 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
249 * Remove any holes in the child array.
252 vdev_compact_children(vdev_t
*pvd
)
254 vdev_t
**newchild
, *cvd
;
255 int oldc
= pvd
->vdev_children
;
258 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
260 for (int c
= newc
= 0; c
< oldc
; c
++)
261 if (pvd
->vdev_child
[c
])
264 newchild
= kmem_alloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
266 for (int c
= newc
= 0; c
< oldc
; c
++) {
267 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
268 newchild
[newc
] = cvd
;
269 cvd
->vdev_id
= newc
++;
273 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
274 pvd
->vdev_child
= newchild
;
275 pvd
->vdev_children
= newc
;
279 * Allocate and minimally initialize a vdev_t.
282 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
286 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
288 if (spa
->spa_root_vdev
== NULL
) {
289 ASSERT(ops
== &vdev_root_ops
);
290 spa
->spa_root_vdev
= vd
;
291 spa
->spa_load_guid
= spa_generate_guid(NULL
);
294 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
295 if (spa
->spa_root_vdev
== vd
) {
297 * The root vdev's guid will also be the pool guid,
298 * which must be unique among all pools.
300 guid
= spa_generate_guid(NULL
);
303 * Any other vdev's guid must be unique within the pool.
305 guid
= spa_generate_guid(spa
);
307 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
312 vd
->vdev_guid
= guid
;
313 vd
->vdev_guid_sum
= guid
;
315 vd
->vdev_state
= VDEV_STATE_CLOSED
;
316 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
318 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
319 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
320 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
321 for (int t
= 0; t
< DTL_TYPES
; t
++) {
322 space_map_create(&vd
->vdev_dtl
[t
], 0, -1ULL, 0,
325 txg_list_create(&vd
->vdev_ms_list
,
326 offsetof(struct metaslab
, ms_txg_node
));
327 txg_list_create(&vd
->vdev_dtl_list
,
328 offsetof(struct vdev
, vdev_dtl_node
));
329 vd
->vdev_stat
.vs_timestamp
= gethrtime();
337 * Allocate a new vdev. The 'alloctype' is used to control whether we are
338 * creating a new vdev or loading an existing one - the behavior is slightly
339 * different for each case.
342 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
347 uint64_t guid
= 0, islog
, nparity
;
350 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
352 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
355 if ((ops
= vdev_getops(type
)) == NULL
)
359 * If this is a load, get the vdev guid from the nvlist.
360 * Otherwise, vdev_alloc_common() will generate one for us.
362 if (alloctype
== VDEV_ALLOC_LOAD
) {
365 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
369 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
371 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
372 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
374 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
375 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
377 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
378 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
383 * The first allocated vdev must be of type 'root'.
385 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
389 * Determine whether we're a log vdev.
392 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
393 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
396 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
400 * Set the nparity property for RAID-Z vdevs.
403 if (ops
== &vdev_raidz_ops
) {
404 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
406 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
409 * Previous versions could only support 1 or 2 parity
413 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
416 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
420 * We require the parity to be specified for SPAs that
421 * support multiple parity levels.
423 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
426 * Otherwise, we default to 1 parity device for RAID-Z.
433 ASSERT(nparity
!= -1ULL);
435 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
437 vd
->vdev_islog
= islog
;
438 vd
->vdev_nparity
= nparity
;
440 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
441 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
442 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
443 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
444 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
445 &vd
->vdev_physpath
) == 0)
446 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
447 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
448 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
451 * Set the whole_disk property. If it's not specified, leave the value
454 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
455 &vd
->vdev_wholedisk
) != 0)
456 vd
->vdev_wholedisk
= -1ULL;
459 * Look for the 'not present' flag. This will only be set if the device
460 * was not present at the time of import.
462 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
463 &vd
->vdev_not_present
);
466 * Get the alignment requirement.
468 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
471 * Retrieve the vdev creation time.
473 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
477 * If we're a top-level vdev, try to load the allocation parameters.
479 if (parent
&& !parent
->vdev_parent
&&
480 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
481 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
483 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
485 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
487 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
491 if (parent
&& !parent
->vdev_parent
) {
492 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
493 alloctype
== VDEV_ALLOC_ADD
||
494 alloctype
== VDEV_ALLOC_SPLIT
||
495 alloctype
== VDEV_ALLOC_ROOTPOOL
);
496 vd
->vdev_mg
= metaslab_group_create(islog
?
497 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
501 * If we're a leaf vdev, try to load the DTL object and other state.
503 if (vd
->vdev_ops
->vdev_op_leaf
&&
504 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
505 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
506 if (alloctype
== VDEV_ALLOC_LOAD
) {
507 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
508 &vd
->vdev_dtl_smo
.smo_object
);
509 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
513 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
516 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
517 &spare
) == 0 && spare
)
521 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
524 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVERING
,
525 &vd
->vdev_resilvering
);
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_max_asize
= cvd
->vdev_max_asize
;
739 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
740 mvd
->vdev_state
= cvd
->vdev_state
;
741 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
743 vdev_remove_child(pvd
, cvd
);
744 vdev_add_child(pvd
, mvd
);
745 cvd
->vdev_id
= mvd
->vdev_children
;
746 vdev_add_child(mvd
, cvd
);
747 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
749 if (mvd
== mvd
->vdev_top
)
750 vdev_top_transfer(cvd
, mvd
);
756 * Remove a 1-way mirror/replacing vdev from the tree.
759 vdev_remove_parent(vdev_t
*cvd
)
761 vdev_t
*mvd
= cvd
->vdev_parent
;
762 vdev_t
*pvd
= mvd
->vdev_parent
;
764 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
766 ASSERT(mvd
->vdev_children
== 1);
767 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
768 mvd
->vdev_ops
== &vdev_replacing_ops
||
769 mvd
->vdev_ops
== &vdev_spare_ops
);
770 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
772 vdev_remove_child(mvd
, cvd
);
773 vdev_remove_child(pvd
, mvd
);
776 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
777 * Otherwise, we could have detached an offline device, and when we
778 * go to import the pool we'll think we have two top-level vdevs,
779 * instead of a different version of the same top-level vdev.
781 if (mvd
->vdev_top
== mvd
) {
782 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
783 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
784 cvd
->vdev_guid
+= guid_delta
;
785 cvd
->vdev_guid_sum
+= guid_delta
;
787 cvd
->vdev_id
= mvd
->vdev_id
;
788 vdev_add_child(pvd
, cvd
);
789 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
791 if (cvd
== cvd
->vdev_top
)
792 vdev_top_transfer(mvd
, cvd
);
794 ASSERT(mvd
->vdev_children
== 0);
799 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
801 spa_t
*spa
= vd
->vdev_spa
;
802 objset_t
*mos
= spa
->spa_meta_objset
;
804 uint64_t oldc
= vd
->vdev_ms_count
;
805 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
809 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
812 * This vdev is not being allocated from yet or is a hole.
814 if (vd
->vdev_ms_shift
== 0)
817 ASSERT(!vd
->vdev_ishole
);
820 * Compute the raidz-deflation ratio. Note, we hard-code
821 * in 128k (1 << 17) because it is the current "typical" blocksize.
822 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
823 * or we will inconsistently account for existing bp's.
825 vd
->vdev_deflate_ratio
= (1 << 17) /
826 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
828 ASSERT(oldc
<= newc
);
830 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
833 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
834 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
838 vd
->vdev_ms_count
= newc
;
840 for (m
= oldc
; m
< newc
; m
++) {
841 space_map_obj_t smo
= { 0, 0, 0 };
844 error
= dmu_read(mos
, vd
->vdev_ms_array
,
845 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
851 error
= dmu_bonus_hold(mos
, object
, FTAG
, &db
);
854 ASSERT3U(db
->db_size
, >=, sizeof (smo
));
855 bcopy(db
->db_data
, &smo
, sizeof (smo
));
856 ASSERT3U(smo
.smo_object
, ==, object
);
857 dmu_buf_rele(db
, FTAG
);
860 vd
->vdev_ms
[m
] = metaslab_init(vd
->vdev_mg
, &smo
,
861 m
<< vd
->vdev_ms_shift
, 1ULL << vd
->vdev_ms_shift
, txg
);
865 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
868 * If the vdev is being removed we don't activate
869 * the metaslabs since we want to ensure that no new
870 * allocations are performed on this device.
872 if (oldc
== 0 && !vd
->vdev_removing
)
873 metaslab_group_activate(vd
->vdev_mg
);
876 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
882 vdev_metaslab_fini(vdev_t
*vd
)
885 uint64_t count
= vd
->vdev_ms_count
;
887 if (vd
->vdev_ms
!= NULL
) {
888 metaslab_group_passivate(vd
->vdev_mg
);
889 for (m
= 0; m
< count
; m
++)
890 if (vd
->vdev_ms
[m
] != NULL
)
891 metaslab_fini(vd
->vdev_ms
[m
]);
892 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
897 typedef struct vdev_probe_stats
{
898 boolean_t vps_readable
;
899 boolean_t vps_writeable
;
901 } vdev_probe_stats_t
;
904 vdev_probe_done(zio_t
*zio
)
906 spa_t
*spa
= zio
->io_spa
;
907 vdev_t
*vd
= zio
->io_vd
;
908 vdev_probe_stats_t
*vps
= zio
->io_private
;
910 ASSERT(vd
->vdev_probe_zio
!= NULL
);
912 if (zio
->io_type
== ZIO_TYPE_READ
) {
913 if (zio
->io_error
== 0)
914 vps
->vps_readable
= 1;
915 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
916 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
917 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
918 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
919 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
921 zio_buf_free(zio
->io_data
, zio
->io_size
);
923 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
924 if (zio
->io_error
== 0)
925 vps
->vps_writeable
= 1;
926 zio_buf_free(zio
->io_data
, zio
->io_size
);
927 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
930 vd
->vdev_cant_read
|= !vps
->vps_readable
;
931 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
933 if (vdev_readable(vd
) &&
934 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
937 ASSERT(zio
->io_error
!= 0);
938 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
939 spa
, vd
, NULL
, 0, 0);
940 zio
->io_error
= ENXIO
;
943 mutex_enter(&vd
->vdev_probe_lock
);
944 ASSERT(vd
->vdev_probe_zio
== zio
);
945 vd
->vdev_probe_zio
= NULL
;
946 mutex_exit(&vd
->vdev_probe_lock
);
948 while ((pio
= zio_walk_parents(zio
)) != NULL
)
949 if (!vdev_accessible(vd
, pio
))
950 pio
->io_error
= ENXIO
;
952 kmem_free(vps
, sizeof (*vps
));
957 * Determine whether this device is accessible by reading and writing
958 * to several known locations: the pad regions of each vdev label
959 * but the first (which we leave alone in case it contains a VTOC).
962 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
964 spa_t
*spa
= vd
->vdev_spa
;
965 vdev_probe_stats_t
*vps
= NULL
;
968 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
971 * Don't probe the probe.
973 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
977 * To prevent 'probe storms' when a device fails, we create
978 * just one probe i/o at a time. All zios that want to probe
979 * this vdev will become parents of the probe io.
981 mutex_enter(&vd
->vdev_probe_lock
);
983 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
984 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
986 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
987 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
990 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
992 * vdev_cant_read and vdev_cant_write can only
993 * transition from TRUE to FALSE when we have the
994 * SCL_ZIO lock as writer; otherwise they can only
995 * transition from FALSE to TRUE. This ensures that
996 * any zio looking at these values can assume that
997 * failures persist for the life of the I/O. That's
998 * important because when a device has intermittent
999 * connectivity problems, we want to ensure that
1000 * they're ascribed to the device (ENXIO) and not
1003 * Since we hold SCL_ZIO as writer here, clear both
1004 * values so the probe can reevaluate from first
1007 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1008 vd
->vdev_cant_read
= B_FALSE
;
1009 vd
->vdev_cant_write
= B_FALSE
;
1012 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1013 vdev_probe_done
, vps
,
1014 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1017 * We can't change the vdev state in this context, so we
1018 * kick off an async task to do it on our behalf.
1021 vd
->vdev_probe_wanted
= B_TRUE
;
1022 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1027 zio_add_child(zio
, pio
);
1029 mutex_exit(&vd
->vdev_probe_lock
);
1032 ASSERT(zio
!= NULL
);
1036 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1037 zio_nowait(zio_read_phys(pio
, vd
,
1038 vdev_label_offset(vd
->vdev_psize
, l
,
1039 offsetof(vdev_label_t
, vl_pad2
)),
1040 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1041 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1042 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1053 vdev_open_child(void *arg
)
1057 vd
->vdev_open_thread
= curthread
;
1058 vd
->vdev_open_error
= vdev_open(vd
);
1059 vd
->vdev_open_thread
= NULL
;
1063 vdev_uses_zvols(vdev_t
*vd
)
1065 if (vd
->vdev_path
&& strncmp(vd
->vdev_path
, ZVOL_DIR
,
1066 strlen(ZVOL_DIR
)) == 0)
1068 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1069 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1075 vdev_open_children(vdev_t
*vd
)
1078 int children
= vd
->vdev_children
;
1081 * in order to handle pools on top of zvols, do the opens
1082 * in a single thread so that the same thread holds the
1083 * spa_namespace_lock
1085 if (vdev_uses_zvols(vd
)) {
1086 for (int c
= 0; c
< children
; c
++)
1087 vd
->vdev_child
[c
]->vdev_open_error
=
1088 vdev_open(vd
->vdev_child
[c
]);
1091 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1092 children
, children
, TASKQ_PREPOPULATE
);
1094 for (int c
= 0; c
< children
; c
++)
1095 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1102 * Prepare a virtual device for access.
1105 vdev_open(vdev_t
*vd
)
1107 spa_t
*spa
= vd
->vdev_spa
;
1110 uint64_t max_osize
= 0;
1111 uint64_t asize
, max_asize
, psize
;
1112 uint64_t ashift
= 0;
1114 ASSERT(vd
->vdev_open_thread
== curthread
||
1115 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1116 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1117 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1118 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1120 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1121 vd
->vdev_cant_read
= B_FALSE
;
1122 vd
->vdev_cant_write
= B_FALSE
;
1123 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1126 * If this vdev is not removed, check its fault status. If it's
1127 * faulted, bail out of the open.
1129 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1130 ASSERT(vd
->vdev_children
== 0);
1131 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1132 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1133 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1134 vd
->vdev_label_aux
);
1136 } else if (vd
->vdev_offline
) {
1137 ASSERT(vd
->vdev_children
== 0);
1138 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1142 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1145 * Reset the vdev_reopening flag so that we actually close
1146 * the vdev on error.
1148 vd
->vdev_reopening
= B_FALSE
;
1149 if (zio_injection_enabled
&& error
== 0)
1150 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1153 if (vd
->vdev_removed
&&
1154 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1155 vd
->vdev_removed
= B_FALSE
;
1157 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1158 vd
->vdev_stat
.vs_aux
);
1162 vd
->vdev_removed
= B_FALSE
;
1165 * Recheck the faulted flag now that we have confirmed that
1166 * the vdev is accessible. If we're faulted, bail.
1168 if (vd
->vdev_faulted
) {
1169 ASSERT(vd
->vdev_children
== 0);
1170 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1171 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1172 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1173 vd
->vdev_label_aux
);
1177 if (vd
->vdev_degraded
) {
1178 ASSERT(vd
->vdev_children
== 0);
1179 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1180 VDEV_AUX_ERR_EXCEEDED
);
1182 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1186 * For hole or missing vdevs we just return success.
1188 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1191 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1192 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1193 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1199 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1200 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1202 if (vd
->vdev_children
== 0) {
1203 if (osize
< SPA_MINDEVSIZE
) {
1204 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1205 VDEV_AUX_TOO_SMALL
);
1209 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1210 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1211 VDEV_LABEL_END_SIZE
);
1213 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1214 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1215 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1216 VDEV_AUX_TOO_SMALL
);
1221 max_asize
= max_osize
;
1224 vd
->vdev_psize
= psize
;
1227 * Make sure the allocatable size hasn't shrunk.
1229 if (asize
< vd
->vdev_min_asize
) {
1230 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1231 VDEV_AUX_BAD_LABEL
);
1235 if (vd
->vdev_asize
== 0) {
1237 * This is the first-ever open, so use the computed values.
1238 * For testing purposes, a higher ashift can be requested.
1240 vd
->vdev_asize
= asize
;
1241 vd
->vdev_max_asize
= max_asize
;
1242 vd
->vdev_ashift
= MAX(ashift
, vd
->vdev_ashift
);
1245 * Make sure the alignment requirement hasn't increased.
1247 if (ashift
> vd
->vdev_top
->vdev_ashift
) {
1248 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1249 VDEV_AUX_BAD_LABEL
);
1252 vd
->vdev_max_asize
= max_asize
;
1256 * If all children are healthy and the asize has increased,
1257 * then we've experienced dynamic LUN growth. If automatic
1258 * expansion is enabled then use the additional space.
1260 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1261 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1262 vd
->vdev_asize
= asize
;
1264 vdev_set_min_asize(vd
);
1267 * Ensure we can issue some IO before declaring the
1268 * vdev open for business.
1270 if (vd
->vdev_ops
->vdev_op_leaf
&&
1271 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1272 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1273 VDEV_AUX_ERR_EXCEEDED
);
1278 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1279 * resilver. But don't do this if we are doing a reopen for a scrub,
1280 * since this would just restart the scrub we are already doing.
1282 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1283 vdev_resilver_needed(vd
, NULL
, NULL
))
1284 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1290 * Called once the vdevs are all opened, this routine validates the label
1291 * contents. This needs to be done before vdev_load() so that we don't
1292 * inadvertently do repair I/Os to the wrong device.
1294 * If 'strict' is false ignore the spa guid check. This is necessary because
1295 * if the machine crashed during a re-guid the new guid might have been written
1296 * to all of the vdev labels, but not the cached config. The strict check
1297 * will be performed when the pool is opened again using the mos config.
1299 * This function will only return failure if one of the vdevs indicates that it
1300 * has since been destroyed or exported. This is only possible if
1301 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1302 * will be updated but the function will return 0.
1305 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1307 spa_t
*spa
= vd
->vdev_spa
;
1309 uint64_t guid
= 0, top_guid
;
1312 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1313 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1317 * If the device has already failed, or was marked offline, don't do
1318 * any further validation. Otherwise, label I/O will fail and we will
1319 * overwrite the previous state.
1321 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1322 uint64_t aux_guid
= 0;
1325 if ((label
= vdev_label_read_config(vd
)) == NULL
) {
1326 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1327 VDEV_AUX_BAD_LABEL
);
1332 * Determine if this vdev has been split off into another
1333 * pool. If so, then refuse to open it.
1335 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1336 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1337 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1338 VDEV_AUX_SPLIT_POOL
);
1343 if (strict
&& (nvlist_lookup_uint64(label
,
1344 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1345 guid
!= spa_guid(spa
))) {
1346 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1347 VDEV_AUX_CORRUPT_DATA
);
1352 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1353 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1358 * If this vdev just became a top-level vdev because its
1359 * sibling was detached, it will have adopted the parent's
1360 * vdev guid -- but the label may or may not be on disk yet.
1361 * Fortunately, either version of the label will have the
1362 * same top guid, so if we're a top-level vdev, we can
1363 * safely compare to that instead.
1365 * If we split this vdev off instead, then we also check the
1366 * original pool's guid. We don't want to consider the vdev
1367 * corrupt if it is partway through a split operation.
1369 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1371 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1373 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1374 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1375 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1376 VDEV_AUX_CORRUPT_DATA
);
1381 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1383 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1384 VDEV_AUX_CORRUPT_DATA
);
1392 * If this is a verbatim import, no need to check the
1393 * state of the pool.
1395 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1396 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1397 state
!= POOL_STATE_ACTIVE
)
1401 * If we were able to open and validate a vdev that was
1402 * previously marked permanently unavailable, clear that state
1405 if (vd
->vdev_not_present
)
1406 vd
->vdev_not_present
= 0;
1413 * Close a virtual device.
1416 vdev_close(vdev_t
*vd
)
1418 spa_t
*spa
= vd
->vdev_spa
;
1419 vdev_t
*pvd
= vd
->vdev_parent
;
1421 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1424 * If our parent is reopening, then we are as well, unless we are
1427 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1428 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1430 vd
->vdev_ops
->vdev_op_close(vd
);
1432 vdev_cache_purge(vd
);
1435 * We record the previous state before we close it, so that if we are
1436 * doing a reopen(), we don't generate FMA ereports if we notice that
1437 * it's still faulted.
1439 vd
->vdev_prevstate
= vd
->vdev_state
;
1441 if (vd
->vdev_offline
)
1442 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1444 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1445 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1449 vdev_hold(vdev_t
*vd
)
1451 spa_t
*spa
= vd
->vdev_spa
;
1453 ASSERT(spa_is_root(spa
));
1454 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1457 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1458 vdev_hold(vd
->vdev_child
[c
]);
1460 if (vd
->vdev_ops
->vdev_op_leaf
)
1461 vd
->vdev_ops
->vdev_op_hold(vd
);
1465 vdev_rele(vdev_t
*vd
)
1467 spa_t
*spa
= vd
->vdev_spa
;
1469 ASSERT(spa_is_root(spa
));
1470 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1471 vdev_rele(vd
->vdev_child
[c
]);
1473 if (vd
->vdev_ops
->vdev_op_leaf
)
1474 vd
->vdev_ops
->vdev_op_rele(vd
);
1478 * Reopen all interior vdevs and any unopened leaves. We don't actually
1479 * reopen leaf vdevs which had previously been opened as they might deadlock
1480 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1481 * If the leaf has never been opened then open it, as usual.
1484 vdev_reopen(vdev_t
*vd
)
1486 spa_t
*spa
= vd
->vdev_spa
;
1488 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1490 /* set the reopening flag unless we're taking the vdev offline */
1491 vd
->vdev_reopening
= !vd
->vdev_offline
;
1493 (void) vdev_open(vd
);
1496 * Call vdev_validate() here to make sure we have the same device.
1497 * Otherwise, a device with an invalid label could be successfully
1498 * opened in response to vdev_reopen().
1501 (void) vdev_validate_aux(vd
);
1502 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1503 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1504 !l2arc_vdev_present(vd
))
1505 l2arc_add_vdev(spa
, vd
);
1507 (void) vdev_validate(vd
, B_TRUE
);
1511 * Reassess parent vdev's health.
1513 vdev_propagate_state(vd
);
1517 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1522 * Normally, partial opens (e.g. of a mirror) are allowed.
1523 * For a create, however, we want to fail the request if
1524 * there are any components we can't open.
1526 error
= vdev_open(vd
);
1528 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1530 return (error
? error
: ENXIO
);
1534 * Recursively initialize all labels.
1536 if ((error
= vdev_label_init(vd
, txg
, isreplacing
?
1537 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1546 vdev_metaslab_set_size(vdev_t
*vd
)
1549 * Aim for roughly 200 metaslabs per vdev.
1551 vd
->vdev_ms_shift
= highbit(vd
->vdev_asize
/ 200);
1552 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1556 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1558 ASSERT(vd
== vd
->vdev_top
);
1559 ASSERT(!vd
->vdev_ishole
);
1560 ASSERT(ISP2(flags
));
1561 ASSERT(spa_writeable(vd
->vdev_spa
));
1563 if (flags
& VDD_METASLAB
)
1564 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1566 if (flags
& VDD_DTL
)
1567 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1569 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1575 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1576 * the vdev has less than perfect replication. There are four kinds of DTL:
1578 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1580 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1582 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1583 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1584 * txgs that was scrubbed.
1586 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1587 * persistent errors or just some device being offline.
1588 * Unlike the other three, the DTL_OUTAGE map is not generally
1589 * maintained; it's only computed when needed, typically to
1590 * determine whether a device can be detached.
1592 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1593 * either has the data or it doesn't.
1595 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1596 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1597 * if any child is less than fully replicated, then so is its parent.
1598 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1599 * comprising only those txgs which appear in 'maxfaults' or more children;
1600 * those are the txgs we don't have enough replication to read. For example,
1601 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1602 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1603 * two child DTL_MISSING maps.
1605 * It should be clear from the above that to compute the DTLs and outage maps
1606 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1607 * Therefore, that is all we keep on disk. When loading the pool, or after
1608 * a configuration change, we generate all other DTLs from first principles.
1611 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1613 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1615 ASSERT(t
< DTL_TYPES
);
1616 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1617 ASSERT(spa_writeable(vd
->vdev_spa
));
1619 mutex_enter(sm
->sm_lock
);
1620 if (!space_map_contains(sm
, txg
, size
))
1621 space_map_add(sm
, txg
, size
);
1622 mutex_exit(sm
->sm_lock
);
1626 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1628 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1629 boolean_t dirty
= B_FALSE
;
1631 ASSERT(t
< DTL_TYPES
);
1632 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1634 mutex_enter(sm
->sm_lock
);
1635 if (sm
->sm_space
!= 0)
1636 dirty
= space_map_contains(sm
, txg
, size
);
1637 mutex_exit(sm
->sm_lock
);
1643 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1645 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1648 mutex_enter(sm
->sm_lock
);
1649 empty
= (sm
->sm_space
== 0);
1650 mutex_exit(sm
->sm_lock
);
1656 * Reassess DTLs after a config change or scrub completion.
1659 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1661 spa_t
*spa
= vd
->vdev_spa
;
1665 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1667 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1668 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1669 scrub_txg
, scrub_done
);
1671 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1674 if (vd
->vdev_ops
->vdev_op_leaf
) {
1675 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1677 mutex_enter(&vd
->vdev_dtl_lock
);
1678 if (scrub_txg
!= 0 &&
1679 (spa
->spa_scrub_started
||
1680 (scn
&& scn
->scn_phys
.scn_errors
== 0))) {
1682 * We completed a scrub up to scrub_txg. If we
1683 * did it without rebooting, then the scrub dtl
1684 * will be valid, so excise the old region and
1685 * fold in the scrub dtl. Otherwise, leave the
1686 * dtl as-is if there was an error.
1688 * There's little trick here: to excise the beginning
1689 * of the DTL_MISSING map, we put it into a reference
1690 * tree and then add a segment with refcnt -1 that
1691 * covers the range [0, scrub_txg). This means
1692 * that each txg in that range has refcnt -1 or 0.
1693 * We then add DTL_SCRUB with a refcnt of 2, so that
1694 * entries in the range [0, scrub_txg) will have a
1695 * positive refcnt -- either 1 or 2. We then convert
1696 * the reference tree into the new DTL_MISSING map.
1698 space_map_ref_create(&reftree
);
1699 space_map_ref_add_map(&reftree
,
1700 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1701 space_map_ref_add_seg(&reftree
, 0, scrub_txg
, -1);
1702 space_map_ref_add_map(&reftree
,
1703 &vd
->vdev_dtl
[DTL_SCRUB
], 2);
1704 space_map_ref_generate_map(&reftree
,
1705 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1706 space_map_ref_destroy(&reftree
);
1708 space_map_vacate(&vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1709 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1710 space_map_add
, &vd
->vdev_dtl
[DTL_PARTIAL
]);
1712 space_map_vacate(&vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1713 space_map_vacate(&vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1714 if (!vdev_readable(vd
))
1715 space_map_add(&vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1717 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1718 space_map_add
, &vd
->vdev_dtl
[DTL_OUTAGE
]);
1719 mutex_exit(&vd
->vdev_dtl_lock
);
1722 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1726 mutex_enter(&vd
->vdev_dtl_lock
);
1727 for (int t
= 0; t
< DTL_TYPES
; t
++) {
1728 /* account for child's outage in parent's missing map */
1729 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1731 continue; /* leaf vdevs only */
1732 if (t
== DTL_PARTIAL
)
1733 minref
= 1; /* i.e. non-zero */
1734 else if (vd
->vdev_nparity
!= 0)
1735 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1737 minref
= vd
->vdev_children
; /* any kind of mirror */
1738 space_map_ref_create(&reftree
);
1739 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1740 vdev_t
*cvd
= vd
->vdev_child
[c
];
1741 mutex_enter(&cvd
->vdev_dtl_lock
);
1742 space_map_ref_add_map(&reftree
, &cvd
->vdev_dtl
[s
], 1);
1743 mutex_exit(&cvd
->vdev_dtl_lock
);
1745 space_map_ref_generate_map(&reftree
, &vd
->vdev_dtl
[t
], minref
);
1746 space_map_ref_destroy(&reftree
);
1748 mutex_exit(&vd
->vdev_dtl_lock
);
1752 vdev_dtl_load(vdev_t
*vd
)
1754 spa_t
*spa
= vd
->vdev_spa
;
1755 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1756 objset_t
*mos
= spa
->spa_meta_objset
;
1760 ASSERT(vd
->vdev_children
== 0);
1762 if (smo
->smo_object
== 0)
1765 ASSERT(!vd
->vdev_ishole
);
1767 if ((error
= dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
)) != 0)
1770 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1771 bcopy(db
->db_data
, smo
, sizeof (*smo
));
1772 dmu_buf_rele(db
, FTAG
);
1774 mutex_enter(&vd
->vdev_dtl_lock
);
1775 error
= space_map_load(&vd
->vdev_dtl
[DTL_MISSING
],
1776 NULL
, SM_ALLOC
, smo
, mos
);
1777 mutex_exit(&vd
->vdev_dtl_lock
);
1783 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1785 spa_t
*spa
= vd
->vdev_spa
;
1786 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1787 space_map_t
*sm
= &vd
->vdev_dtl
[DTL_MISSING
];
1788 objset_t
*mos
= spa
->spa_meta_objset
;
1794 ASSERT(!vd
->vdev_ishole
);
1796 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1798 if (vd
->vdev_detached
) {
1799 if (smo
->smo_object
!= 0) {
1800 int err
= dmu_object_free(mos
, smo
->smo_object
, tx
);
1801 ASSERT3U(err
, ==, 0);
1802 smo
->smo_object
= 0;
1808 if (smo
->smo_object
== 0) {
1809 ASSERT(smo
->smo_objsize
== 0);
1810 ASSERT(smo
->smo_alloc
== 0);
1811 smo
->smo_object
= dmu_object_alloc(mos
,
1812 DMU_OT_SPACE_MAP
, 1 << SPACE_MAP_BLOCKSHIFT
,
1813 DMU_OT_SPACE_MAP_HEADER
, sizeof (*smo
), tx
);
1814 ASSERT(smo
->smo_object
!= 0);
1815 vdev_config_dirty(vd
->vdev_top
);
1818 mutex_init(&smlock
, NULL
, MUTEX_DEFAULT
, NULL
);
1820 space_map_create(&smsync
, sm
->sm_start
, sm
->sm_size
, sm
->sm_shift
,
1823 mutex_enter(&smlock
);
1825 mutex_enter(&vd
->vdev_dtl_lock
);
1826 space_map_walk(sm
, space_map_add
, &smsync
);
1827 mutex_exit(&vd
->vdev_dtl_lock
);
1829 space_map_truncate(smo
, mos
, tx
);
1830 space_map_sync(&smsync
, SM_ALLOC
, smo
, mos
, tx
);
1832 space_map_destroy(&smsync
);
1834 mutex_exit(&smlock
);
1835 mutex_destroy(&smlock
);
1837 VERIFY(0 == dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
));
1838 dmu_buf_will_dirty(db
, tx
);
1839 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1840 bcopy(smo
, db
->db_data
, sizeof (*smo
));
1841 dmu_buf_rele(db
, FTAG
);
1847 * Determine whether the specified vdev can be offlined/detached/removed
1848 * without losing data.
1851 vdev_dtl_required(vdev_t
*vd
)
1853 spa_t
*spa
= vd
->vdev_spa
;
1854 vdev_t
*tvd
= vd
->vdev_top
;
1855 uint8_t cant_read
= vd
->vdev_cant_read
;
1858 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1860 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
1864 * Temporarily mark the device as unreadable, and then determine
1865 * whether this results in any DTL outages in the top-level vdev.
1866 * If not, we can safely offline/detach/remove the device.
1868 vd
->vdev_cant_read
= B_TRUE
;
1869 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1870 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
1871 vd
->vdev_cant_read
= cant_read
;
1872 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1874 if (!required
&& zio_injection_enabled
)
1875 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
1881 * Determine if resilver is needed, and if so the txg range.
1884 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
1886 boolean_t needed
= B_FALSE
;
1887 uint64_t thismin
= UINT64_MAX
;
1888 uint64_t thismax
= 0;
1890 if (vd
->vdev_children
== 0) {
1891 mutex_enter(&vd
->vdev_dtl_lock
);
1892 if (vd
->vdev_dtl
[DTL_MISSING
].sm_space
!= 0 &&
1893 vdev_writeable(vd
)) {
1896 ss
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1897 thismin
= ss
->ss_start
- 1;
1898 ss
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1899 thismax
= ss
->ss_end
;
1902 mutex_exit(&vd
->vdev_dtl_lock
);
1904 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1905 vdev_t
*cvd
= vd
->vdev_child
[c
];
1906 uint64_t cmin
, cmax
;
1908 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
1909 thismin
= MIN(thismin
, cmin
);
1910 thismax
= MAX(thismax
, cmax
);
1916 if (needed
&& minp
) {
1924 vdev_load(vdev_t
*vd
)
1927 * Recursively load all children.
1929 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1930 vdev_load(vd
->vdev_child
[c
]);
1933 * If this is a top-level vdev, initialize its metaslabs.
1935 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
1936 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
1937 vdev_metaslab_init(vd
, 0) != 0))
1938 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1939 VDEV_AUX_CORRUPT_DATA
);
1942 * If this is a leaf vdev, load its DTL.
1944 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
1945 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1946 VDEV_AUX_CORRUPT_DATA
);
1950 * The special vdev case is used for hot spares and l2cache devices. Its
1951 * sole purpose it to set the vdev state for the associated vdev. To do this,
1952 * we make sure that we can open the underlying device, then try to read the
1953 * label, and make sure that the label is sane and that it hasn't been
1954 * repurposed to another pool.
1957 vdev_validate_aux(vdev_t
*vd
)
1960 uint64_t guid
, version
;
1963 if (!vdev_readable(vd
))
1966 if ((label
= vdev_label_read_config(vd
)) == NULL
) {
1967 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1968 VDEV_AUX_CORRUPT_DATA
);
1972 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
1973 version
> SPA_VERSION
||
1974 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
1975 guid
!= vd
->vdev_guid
||
1976 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
1977 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1978 VDEV_AUX_CORRUPT_DATA
);
1984 * We don't actually check the pool state here. If it's in fact in
1985 * use by another pool, we update this fact on the fly when requested.
1992 vdev_remove(vdev_t
*vd
, uint64_t txg
)
1994 spa_t
*spa
= vd
->vdev_spa
;
1995 objset_t
*mos
= spa
->spa_meta_objset
;
1998 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2000 if (vd
->vdev_dtl_smo
.smo_object
) {
2001 ASSERT3U(vd
->vdev_dtl_smo
.smo_alloc
, ==, 0);
2002 (void) dmu_object_free(mos
, vd
->vdev_dtl_smo
.smo_object
, tx
);
2003 vd
->vdev_dtl_smo
.smo_object
= 0;
2006 if (vd
->vdev_ms
!= NULL
) {
2007 for (int m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2008 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2010 if (msp
== NULL
|| msp
->ms_smo
.smo_object
== 0)
2013 ASSERT3U(msp
->ms_smo
.smo_alloc
, ==, 0);
2014 (void) dmu_object_free(mos
, msp
->ms_smo
.smo_object
, tx
);
2015 msp
->ms_smo
.smo_object
= 0;
2019 if (vd
->vdev_ms_array
) {
2020 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2021 vd
->vdev_ms_array
= 0;
2022 vd
->vdev_ms_shift
= 0;
2028 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2031 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2033 ASSERT(!vd
->vdev_ishole
);
2035 while (msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
2036 metaslab_sync_done(msp
, txg
);
2039 metaslab_sync_reassess(vd
->vdev_mg
);
2043 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2045 spa_t
*spa
= vd
->vdev_spa
;
2050 ASSERT(!vd
->vdev_ishole
);
2052 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2053 ASSERT(vd
== vd
->vdev_top
);
2054 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2055 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2056 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2057 ASSERT(vd
->vdev_ms_array
!= 0);
2058 vdev_config_dirty(vd
);
2063 * Remove the metadata associated with this vdev once it's empty.
2065 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2066 vdev_remove(vd
, txg
);
2068 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2069 metaslab_sync(msp
, txg
);
2070 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2073 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2074 vdev_dtl_sync(lvd
, txg
);
2076 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2080 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2082 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2086 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2087 * not be opened, and no I/O is attempted.
2090 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2094 spa_vdev_state_enter(spa
, SCL_NONE
);
2096 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2097 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2099 if (!vd
->vdev_ops
->vdev_op_leaf
)
2100 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2105 * We don't directly use the aux state here, but if we do a
2106 * vdev_reopen(), we need this value to be present to remember why we
2109 vd
->vdev_label_aux
= aux
;
2112 * Faulted state takes precedence over degraded.
2114 vd
->vdev_delayed_close
= B_FALSE
;
2115 vd
->vdev_faulted
= 1ULL;
2116 vd
->vdev_degraded
= 0ULL;
2117 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2120 * If this device has the only valid copy of the data, then
2121 * back off and simply mark the vdev as degraded instead.
2123 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2124 vd
->vdev_degraded
= 1ULL;
2125 vd
->vdev_faulted
= 0ULL;
2128 * If we reopen the device and it's not dead, only then do we
2133 if (vdev_readable(vd
))
2134 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2137 return (spa_vdev_state_exit(spa
, vd
, 0));
2141 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2142 * user that something is wrong. The vdev continues to operate as normal as far
2143 * as I/O is concerned.
2146 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2150 spa_vdev_state_enter(spa
, SCL_NONE
);
2152 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2153 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2155 if (!vd
->vdev_ops
->vdev_op_leaf
)
2156 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2159 * If the vdev is already faulted, then don't do anything.
2161 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2162 return (spa_vdev_state_exit(spa
, NULL
, 0));
2164 vd
->vdev_degraded
= 1ULL;
2165 if (!vdev_is_dead(vd
))
2166 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2169 return (spa_vdev_state_exit(spa
, vd
, 0));
2173 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2174 * any attached spare device should be detached when the device finishes
2175 * resilvering. Second, the online should be treated like a 'test' online case,
2176 * so no FMA events are generated if the device fails to open.
2179 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2181 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2183 spa_vdev_state_enter(spa
, SCL_NONE
);
2185 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2186 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2188 if (!vd
->vdev_ops
->vdev_op_leaf
)
2189 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2192 vd
->vdev_offline
= B_FALSE
;
2193 vd
->vdev_tmpoffline
= B_FALSE
;
2194 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2195 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2197 /* XXX - L2ARC 1.0 does not support expansion */
2198 if (!vd
->vdev_aux
) {
2199 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2200 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2204 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2206 if (!vd
->vdev_aux
) {
2207 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2208 pvd
->vdev_expanding
= B_FALSE
;
2212 *newstate
= vd
->vdev_state
;
2213 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2214 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2215 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2216 vd
->vdev_parent
->vdev_child
[0] == vd
)
2217 vd
->vdev_unspare
= B_TRUE
;
2219 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2221 /* XXX - L2ARC 1.0 does not support expansion */
2223 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2224 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2226 return (spa_vdev_state_exit(spa
, vd
, 0));
2230 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2234 uint64_t generation
;
2235 metaslab_group_t
*mg
;
2238 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2240 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2241 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2243 if (!vd
->vdev_ops
->vdev_op_leaf
)
2244 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2248 generation
= spa
->spa_config_generation
+ 1;
2251 * If the device isn't already offline, try to offline it.
2253 if (!vd
->vdev_offline
) {
2255 * If this device has the only valid copy of some data,
2256 * don't allow it to be offlined. Log devices are always
2259 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2260 vdev_dtl_required(vd
))
2261 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2264 * If the top-level is a slog and it has had allocations
2265 * then proceed. We check that the vdev's metaslab group
2266 * is not NULL since it's possible that we may have just
2267 * added this vdev but not yet initialized its metaslabs.
2269 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2271 * Prevent any future allocations.
2273 metaslab_group_passivate(mg
);
2274 (void) spa_vdev_state_exit(spa
, vd
, 0);
2276 error
= spa_offline_log(spa
);
2278 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2281 * Check to see if the config has changed.
2283 if (error
|| generation
!= spa
->spa_config_generation
) {
2284 metaslab_group_activate(mg
);
2286 return (spa_vdev_state_exit(spa
,
2288 (void) spa_vdev_state_exit(spa
, vd
, 0);
2291 ASSERT3U(tvd
->vdev_stat
.vs_alloc
, ==, 0);
2295 * Offline this device and reopen its top-level vdev.
2296 * If the top-level vdev is a log device then just offline
2297 * it. Otherwise, if this action results in the top-level
2298 * vdev becoming unusable, undo it and fail the request.
2300 vd
->vdev_offline
= B_TRUE
;
2303 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2304 vdev_is_dead(tvd
)) {
2305 vd
->vdev_offline
= B_FALSE
;
2307 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2311 * Add the device back into the metaslab rotor so that
2312 * once we online the device it's open for business.
2314 if (tvd
->vdev_islog
&& mg
!= NULL
)
2315 metaslab_group_activate(mg
);
2318 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2320 return (spa_vdev_state_exit(spa
, vd
, 0));
2324 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2328 mutex_enter(&spa
->spa_vdev_top_lock
);
2329 error
= vdev_offline_locked(spa
, guid
, flags
);
2330 mutex_exit(&spa
->spa_vdev_top_lock
);
2336 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2337 * vdev_offline(), we assume the spa config is locked. We also clear all
2338 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2341 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2343 vdev_t
*rvd
= spa
->spa_root_vdev
;
2345 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2350 vd
->vdev_stat
.vs_read_errors
= 0;
2351 vd
->vdev_stat
.vs_write_errors
= 0;
2352 vd
->vdev_stat
.vs_checksum_errors
= 0;
2354 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2355 vdev_clear(spa
, vd
->vdev_child
[c
]);
2358 * If we're in the FAULTED state or have experienced failed I/O, then
2359 * clear the persistent state and attempt to reopen the device. We
2360 * also mark the vdev config dirty, so that the new faulted state is
2361 * written out to disk.
2363 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2364 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2367 * When reopening in reponse to a clear event, it may be due to
2368 * a fmadm repair request. In this case, if the device is
2369 * still broken, we want to still post the ereport again.
2371 vd
->vdev_forcefault
= B_TRUE
;
2373 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2374 vd
->vdev_cant_read
= B_FALSE
;
2375 vd
->vdev_cant_write
= B_FALSE
;
2377 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2379 vd
->vdev_forcefault
= B_FALSE
;
2381 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2382 vdev_state_dirty(vd
->vdev_top
);
2384 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2385 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2387 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2391 * When clearing a FMA-diagnosed fault, we always want to
2392 * unspare the device, as we assume that the original spare was
2393 * done in response to the FMA fault.
2395 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2396 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2397 vd
->vdev_parent
->vdev_child
[0] == vd
)
2398 vd
->vdev_unspare
= B_TRUE
;
2402 vdev_is_dead(vdev_t
*vd
)
2405 * Holes and missing devices are always considered "dead".
2406 * This simplifies the code since we don't have to check for
2407 * these types of devices in the various code paths.
2408 * Instead we rely on the fact that we skip over dead devices
2409 * before issuing I/O to them.
2411 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2412 vd
->vdev_ops
== &vdev_missing_ops
);
2416 vdev_readable(vdev_t
*vd
)
2418 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2422 vdev_writeable(vdev_t
*vd
)
2424 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2428 vdev_allocatable(vdev_t
*vd
)
2430 uint64_t state
= vd
->vdev_state
;
2433 * We currently allow allocations from vdevs which may be in the
2434 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2435 * fails to reopen then we'll catch it later when we're holding
2436 * the proper locks. Note that we have to get the vdev state
2437 * in a local variable because although it changes atomically,
2438 * we're asking two separate questions about it.
2440 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2441 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2445 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2447 ASSERT(zio
->io_vd
== vd
);
2449 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2452 if (zio
->io_type
== ZIO_TYPE_READ
)
2453 return (!vd
->vdev_cant_read
);
2455 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2456 return (!vd
->vdev_cant_write
);
2462 * Get statistics for the given vdev.
2465 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2467 vdev_t
*rvd
= vd
->vdev_spa
->spa_root_vdev
;
2469 mutex_enter(&vd
->vdev_stat_lock
);
2470 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2471 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2472 vs
->vs_state
= vd
->vdev_state
;
2473 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2474 if (vd
->vdev_ops
->vdev_op_leaf
)
2475 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2476 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2477 mutex_exit(&vd
->vdev_stat_lock
);
2480 * If we're getting stats on the root vdev, aggregate the I/O counts
2481 * over all top-level vdevs (i.e. the direct children of the root).
2484 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2485 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2486 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2488 mutex_enter(&vd
->vdev_stat_lock
);
2489 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
2490 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2491 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2493 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2494 mutex_exit(&vd
->vdev_stat_lock
);
2500 vdev_clear_stats(vdev_t
*vd
)
2502 mutex_enter(&vd
->vdev_stat_lock
);
2503 vd
->vdev_stat
.vs_space
= 0;
2504 vd
->vdev_stat
.vs_dspace
= 0;
2505 vd
->vdev_stat
.vs_alloc
= 0;
2506 mutex_exit(&vd
->vdev_stat_lock
);
2510 vdev_scan_stat_init(vdev_t
*vd
)
2512 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2514 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2515 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2517 mutex_enter(&vd
->vdev_stat_lock
);
2518 vs
->vs_scan_processed
= 0;
2519 mutex_exit(&vd
->vdev_stat_lock
);
2523 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2525 spa_t
*spa
= zio
->io_spa
;
2526 vdev_t
*rvd
= spa
->spa_root_vdev
;
2527 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2529 uint64_t txg
= zio
->io_txg
;
2530 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2531 zio_type_t type
= zio
->io_type
;
2532 int flags
= zio
->io_flags
;
2535 * If this i/o is a gang leader, it didn't do any actual work.
2537 if (zio
->io_gang_tree
)
2540 if (zio
->io_error
== 0) {
2542 * If this is a root i/o, don't count it -- we've already
2543 * counted the top-level vdevs, and vdev_get_stats() will
2544 * aggregate them when asked. This reduces contention on
2545 * the root vdev_stat_lock and implicitly handles blocks
2546 * that compress away to holes, for which there is no i/o.
2547 * (Holes never create vdev children, so all the counters
2548 * remain zero, which is what we want.)
2550 * Note: this only applies to successful i/o (io_error == 0)
2551 * because unlike i/o counts, errors are not additive.
2552 * When reading a ditto block, for example, failure of
2553 * one top-level vdev does not imply a root-level error.
2558 ASSERT(vd
== zio
->io_vd
);
2560 if (flags
& ZIO_FLAG_IO_BYPASS
)
2563 mutex_enter(&vd
->vdev_stat_lock
);
2565 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2566 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2567 dsl_scan_phys_t
*scn_phys
=
2568 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2569 uint64_t *processed
= &scn_phys
->scn_processed
;
2572 if (vd
->vdev_ops
->vdev_op_leaf
)
2573 atomic_add_64(processed
, psize
);
2574 vs
->vs_scan_processed
+= psize
;
2577 if (flags
& ZIO_FLAG_SELF_HEAL
)
2578 vs
->vs_self_healed
+= psize
;
2582 vs
->vs_bytes
[type
] += psize
;
2584 mutex_exit(&vd
->vdev_stat_lock
);
2588 if (flags
& ZIO_FLAG_SPECULATIVE
)
2592 * If this is an I/O error that is going to be retried, then ignore the
2593 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2594 * hard errors, when in reality they can happen for any number of
2595 * innocuous reasons (bus resets, MPxIO link failure, etc).
2597 if (zio
->io_error
== EIO
&&
2598 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2602 * Intent logs writes won't propagate their error to the root
2603 * I/O so don't mark these types of failures as pool-level
2606 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2609 mutex_enter(&vd
->vdev_stat_lock
);
2610 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2611 if (zio
->io_error
== ECKSUM
)
2612 vs
->vs_checksum_errors
++;
2614 vs
->vs_read_errors
++;
2616 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2617 vs
->vs_write_errors
++;
2618 mutex_exit(&vd
->vdev_stat_lock
);
2620 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2621 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2622 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2623 spa
->spa_claiming
)) {
2625 * This is either a normal write (not a repair), or it's
2626 * a repair induced by the scrub thread, or it's a repair
2627 * made by zil_claim() during spa_load() in the first txg.
2628 * In the normal case, we commit the DTL change in the same
2629 * txg as the block was born. In the scrub-induced repair
2630 * case, we know that scrubs run in first-pass syncing context,
2631 * so we commit the DTL change in spa_syncing_txg(spa).
2632 * In the zil_claim() case, we commit in spa_first_txg(spa).
2634 * We currently do not make DTL entries for failed spontaneous
2635 * self-healing writes triggered by normal (non-scrubbing)
2636 * reads, because we have no transactional context in which to
2637 * do so -- and it's not clear that it'd be desirable anyway.
2639 if (vd
->vdev_ops
->vdev_op_leaf
) {
2640 uint64_t commit_txg
= txg
;
2641 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2642 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2643 ASSERT(spa_sync_pass(spa
) == 1);
2644 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2645 commit_txg
= spa_syncing_txg(spa
);
2646 } else if (spa
->spa_claiming
) {
2647 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2648 commit_txg
= spa_first_txg(spa
);
2650 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2651 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2653 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2654 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2655 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2658 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2663 * Update the in-core space usage stats for this vdev, its metaslab class,
2664 * and the root vdev.
2667 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2668 int64_t space_delta
)
2670 int64_t dspace_delta
= space_delta
;
2671 spa_t
*spa
= vd
->vdev_spa
;
2672 vdev_t
*rvd
= spa
->spa_root_vdev
;
2673 metaslab_group_t
*mg
= vd
->vdev_mg
;
2674 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2676 ASSERT(vd
== vd
->vdev_top
);
2679 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2680 * factor. We must calculate this here and not at the root vdev
2681 * because the root vdev's psize-to-asize is simply the max of its
2682 * childrens', thus not accurate enough for us.
2684 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2685 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2686 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2687 vd
->vdev_deflate_ratio
;
2689 mutex_enter(&vd
->vdev_stat_lock
);
2690 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2691 vd
->vdev_stat
.vs_space
+= space_delta
;
2692 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2693 mutex_exit(&vd
->vdev_stat_lock
);
2695 if (mc
== spa_normal_class(spa
)) {
2696 mutex_enter(&rvd
->vdev_stat_lock
);
2697 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2698 rvd
->vdev_stat
.vs_space
+= space_delta
;
2699 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2700 mutex_exit(&rvd
->vdev_stat_lock
);
2704 ASSERT(rvd
== vd
->vdev_parent
);
2705 ASSERT(vd
->vdev_ms_count
!= 0);
2707 metaslab_class_space_update(mc
,
2708 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2713 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2714 * so that it will be written out next time the vdev configuration is synced.
2715 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2718 vdev_config_dirty(vdev_t
*vd
)
2720 spa_t
*spa
= vd
->vdev_spa
;
2721 vdev_t
*rvd
= spa
->spa_root_vdev
;
2724 ASSERT(spa_writeable(spa
));
2727 * If this is an aux vdev (as with l2cache and spare devices), then we
2728 * update the vdev config manually and set the sync flag.
2730 if (vd
->vdev_aux
!= NULL
) {
2731 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2735 for (c
= 0; c
< sav
->sav_count
; c
++) {
2736 if (sav
->sav_vdevs
[c
] == vd
)
2740 if (c
== sav
->sav_count
) {
2742 * We're being removed. There's nothing more to do.
2744 ASSERT(sav
->sav_sync
== B_TRUE
);
2748 sav
->sav_sync
= B_TRUE
;
2750 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2751 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2752 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2753 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2759 * Setting the nvlist in the middle if the array is a little
2760 * sketchy, but it will work.
2762 nvlist_free(aux
[c
]);
2763 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
2769 * The dirty list is protected by the SCL_CONFIG lock. The caller
2770 * must either hold SCL_CONFIG as writer, or must be the sync thread
2771 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2772 * so this is sufficient to ensure mutual exclusion.
2774 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2775 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2776 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2779 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2780 vdev_config_dirty(rvd
->vdev_child
[c
]);
2782 ASSERT(vd
== vd
->vdev_top
);
2784 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
2786 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
2791 vdev_config_clean(vdev_t
*vd
)
2793 spa_t
*spa
= vd
->vdev_spa
;
2795 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2796 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2797 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2799 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
2800 list_remove(&spa
->spa_config_dirty_list
, vd
);
2804 * Mark a top-level vdev's state as dirty, so that the next pass of
2805 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2806 * the state changes from larger config changes because they require
2807 * much less locking, and are often needed for administrative actions.
2810 vdev_state_dirty(vdev_t
*vd
)
2812 spa_t
*spa
= vd
->vdev_spa
;
2814 ASSERT(spa_writeable(spa
));
2815 ASSERT(vd
== vd
->vdev_top
);
2818 * The state list is protected by the SCL_STATE lock. The caller
2819 * must either hold SCL_STATE as writer, or must be the sync thread
2820 * (which holds SCL_STATE as reader). There's only one sync thread,
2821 * so this is sufficient to ensure mutual exclusion.
2823 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2824 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2825 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2827 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
2828 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
2832 vdev_state_clean(vdev_t
*vd
)
2834 spa_t
*spa
= vd
->vdev_spa
;
2836 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2837 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2838 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2840 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
2841 list_remove(&spa
->spa_state_dirty_list
, vd
);
2845 * Propagate vdev state up from children to parent.
2848 vdev_propagate_state(vdev_t
*vd
)
2850 spa_t
*spa
= vd
->vdev_spa
;
2851 vdev_t
*rvd
= spa
->spa_root_vdev
;
2852 int degraded
= 0, faulted
= 0;
2856 if (vd
->vdev_children
> 0) {
2857 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2858 child
= vd
->vdev_child
[c
];
2861 * Don't factor holes into the decision.
2863 if (child
->vdev_ishole
)
2866 if (!vdev_readable(child
) ||
2867 (!vdev_writeable(child
) && spa_writeable(spa
))) {
2869 * Root special: if there is a top-level log
2870 * device, treat the root vdev as if it were
2873 if (child
->vdev_islog
&& vd
== rvd
)
2877 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
2881 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
2885 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
2888 * Root special: if there is a top-level vdev that cannot be
2889 * opened due to corrupted metadata, then propagate the root
2890 * vdev's aux state as 'corrupt' rather than 'insufficient
2893 if (corrupted
&& vd
== rvd
&&
2894 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
2895 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2896 VDEV_AUX_CORRUPT_DATA
);
2899 if (vd
->vdev_parent
)
2900 vdev_propagate_state(vd
->vdev_parent
);
2904 * Set a vdev's state. If this is during an open, we don't update the parent
2905 * state, because we're in the process of opening children depth-first.
2906 * Otherwise, we propagate the change to the parent.
2908 * If this routine places a device in a faulted state, an appropriate ereport is
2912 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
2914 uint64_t save_state
;
2915 spa_t
*spa
= vd
->vdev_spa
;
2917 if (state
== vd
->vdev_state
) {
2918 vd
->vdev_stat
.vs_aux
= aux
;
2922 save_state
= vd
->vdev_state
;
2924 vd
->vdev_state
= state
;
2925 vd
->vdev_stat
.vs_aux
= aux
;
2928 * If we are setting the vdev state to anything but an open state, then
2929 * always close the underlying device unless the device has requested
2930 * a delayed close (i.e. we're about to remove or fault the device).
2931 * Otherwise, we keep accessible but invalid devices open forever.
2932 * We don't call vdev_close() itself, because that implies some extra
2933 * checks (offline, etc) that we don't want here. This is limited to
2934 * leaf devices, because otherwise closing the device will affect other
2937 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
2938 vd
->vdev_ops
->vdev_op_leaf
)
2939 vd
->vdev_ops
->vdev_op_close(vd
);
2942 * If we have brought this vdev back into service, we need
2943 * to notify fmd so that it can gracefully repair any outstanding
2944 * cases due to a missing device. We do this in all cases, even those
2945 * that probably don't correlate to a repaired fault. This is sure to
2946 * catch all cases, and we let the zfs-retire agent sort it out. If
2947 * this is a transient state it's OK, as the retire agent will
2948 * double-check the state of the vdev before repairing it.
2950 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
2951 vd
->vdev_prevstate
!= state
)
2952 zfs_post_state_change(spa
, vd
);
2954 if (vd
->vdev_removed
&&
2955 state
== VDEV_STATE_CANT_OPEN
&&
2956 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
2958 * If the previous state is set to VDEV_STATE_REMOVED, then this
2959 * device was previously marked removed and someone attempted to
2960 * reopen it. If this failed due to a nonexistent device, then
2961 * keep the device in the REMOVED state. We also let this be if
2962 * it is one of our special test online cases, which is only
2963 * attempting to online the device and shouldn't generate an FMA
2966 vd
->vdev_state
= VDEV_STATE_REMOVED
;
2967 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2968 } else if (state
== VDEV_STATE_REMOVED
) {
2969 vd
->vdev_removed
= B_TRUE
;
2970 } else if (state
== VDEV_STATE_CANT_OPEN
) {
2972 * If we fail to open a vdev during an import or recovery, we
2973 * mark it as "not available", which signifies that it was
2974 * never there to begin with. Failure to open such a device
2975 * is not considered an error.
2977 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
2978 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
2979 vd
->vdev_ops
->vdev_op_leaf
)
2980 vd
->vdev_not_present
= 1;
2983 * Post the appropriate ereport. If the 'prevstate' field is
2984 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2985 * that this is part of a vdev_reopen(). In this case, we don't
2986 * want to post the ereport if the device was already in the
2987 * CANT_OPEN state beforehand.
2989 * If the 'checkremove' flag is set, then this is an attempt to
2990 * online the device in response to an insertion event. If we
2991 * hit this case, then we have detected an insertion event for a
2992 * faulted or offline device that wasn't in the removed state.
2993 * In this scenario, we don't post an ereport because we are
2994 * about to replace the device, or attempt an online with
2995 * vdev_forcefault, which will generate the fault for us.
2997 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
2998 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
2999 vd
!= spa
->spa_root_vdev
) {
3003 case VDEV_AUX_OPEN_FAILED
:
3004 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3006 case VDEV_AUX_CORRUPT_DATA
:
3007 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3009 case VDEV_AUX_NO_REPLICAS
:
3010 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3012 case VDEV_AUX_BAD_GUID_SUM
:
3013 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3015 case VDEV_AUX_TOO_SMALL
:
3016 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3018 case VDEV_AUX_BAD_LABEL
:
3019 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3022 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3025 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3028 /* Erase any notion of persistent removed state */
3029 vd
->vdev_removed
= B_FALSE
;
3031 vd
->vdev_removed
= B_FALSE
;
3034 if (!isopen
&& vd
->vdev_parent
)
3035 vdev_propagate_state(vd
->vdev_parent
);
3039 * Check the vdev configuration to ensure that it's capable of supporting
3040 * a root pool. Currently, we do not support RAID-Z or partial configuration.
3041 * In addition, only a single top-level vdev is allowed and none of the leaves
3042 * can be wholedisks.
3045 vdev_is_bootable(vdev_t
*vd
)
3047 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3048 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3050 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3051 vd
->vdev_children
> 1) {
3053 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3054 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3057 } else if (vd
->vdev_wholedisk
== 1) {
3061 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3062 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3069 * Load the state from the original vdev tree (ovd) which
3070 * we've retrieved from the MOS config object. If the original
3071 * vdev was offline or faulted then we transfer that state to the
3072 * device in the current vdev tree (nvd).
3075 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3077 spa_t
*spa
= nvd
->vdev_spa
;
3079 ASSERT(nvd
->vdev_top
->vdev_islog
);
3080 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3081 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3083 for (int c
= 0; c
< nvd
->vdev_children
; c
++)
3084 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3086 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3088 * Restore the persistent vdev state
3090 nvd
->vdev_offline
= ovd
->vdev_offline
;
3091 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3092 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3093 nvd
->vdev_removed
= ovd
->vdev_removed
;
3098 * Determine if a log device has valid content. If the vdev was
3099 * removed or faulted in the MOS config then we know that
3100 * the content on the log device has already been written to the pool.
3103 vdev_log_state_valid(vdev_t
*vd
)
3105 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3109 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3110 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3117 * Expand a vdev if possible.
3120 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3122 ASSERT(vd
->vdev_top
== vd
);
3123 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3125 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3126 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3127 vdev_config_dirty(vd
);
3135 vdev_split(vdev_t
*vd
)
3137 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3139 vdev_remove_child(pvd
, vd
);
3140 vdev_compact_children(pvd
);
3142 cvd
= pvd
->vdev_child
[0];
3143 if (pvd
->vdev_children
== 1) {
3144 vdev_remove_parent(cvd
);
3145 cvd
->vdev_splitting
= B_TRUE
;
3147 vdev_propagate_state(cvd
);