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
);
90 for (c
= 0; c
< vd
->vdev_children
; c
++) {
91 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
92 asize
= MAX(asize
, csize
);
99 * Get the minimum allocatable size. We define the allocatable size as
100 * the vdev's asize rounded to the nearest metaslab. This allows us to
101 * replace or attach devices which don't have the same physical size but
102 * can still satisfy the same number of allocations.
105 vdev_get_min_asize(vdev_t
*vd
)
107 vdev_t
*pvd
= vd
->vdev_parent
;
110 * The our parent is NULL (inactive spare or cache) or is the root,
111 * just return our own asize.
114 return (vd
->vdev_asize
);
117 * The top-level vdev just returns the allocatable size rounded
118 * to the nearest metaslab.
120 if (vd
== vd
->vdev_top
)
121 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
124 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
125 * so each child must provide at least 1/Nth of its asize.
127 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
128 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
130 return (pvd
->vdev_min_asize
);
134 vdev_set_min_asize(vdev_t
*vd
)
137 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
139 for (c
= 0; c
< vd
->vdev_children
; c
++)
140 vdev_set_min_asize(vd
->vdev_child
[c
]);
144 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
146 vdev_t
*rvd
= spa
->spa_root_vdev
;
148 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
150 if (vdev
< rvd
->vdev_children
) {
151 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
152 return (rvd
->vdev_child
[vdev
]);
159 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
164 if (vd
->vdev_guid
== guid
)
167 for (c
= 0; c
< vd
->vdev_children
; c
++)
168 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
176 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
178 size_t oldsize
, newsize
;
179 uint64_t id
= cvd
->vdev_id
;
182 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
183 ASSERT(cvd
->vdev_parent
== NULL
);
185 cvd
->vdev_parent
= pvd
;
190 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
192 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
193 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
194 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
196 newchild
= kmem_zalloc(newsize
, KM_SLEEP
);
197 if (pvd
->vdev_child
!= NULL
) {
198 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
199 kmem_free(pvd
->vdev_child
, oldsize
);
202 pvd
->vdev_child
= newchild
;
203 pvd
->vdev_child
[id
] = cvd
;
205 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
206 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
209 * Walk up all ancestors to update guid sum.
211 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
212 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
216 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
219 uint_t id
= cvd
->vdev_id
;
221 ASSERT(cvd
->vdev_parent
== pvd
);
226 ASSERT(id
< pvd
->vdev_children
);
227 ASSERT(pvd
->vdev_child
[id
] == cvd
);
229 pvd
->vdev_child
[id
] = NULL
;
230 cvd
->vdev_parent
= NULL
;
232 for (c
= 0; c
< pvd
->vdev_children
; c
++)
233 if (pvd
->vdev_child
[c
])
236 if (c
== pvd
->vdev_children
) {
237 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
238 pvd
->vdev_child
= NULL
;
239 pvd
->vdev_children
= 0;
243 * Walk up all ancestors to update guid sum.
245 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
246 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
250 * Remove any holes in the child array.
253 vdev_compact_children(vdev_t
*pvd
)
255 vdev_t
**newchild
, *cvd
;
256 int oldc
= pvd
->vdev_children
;
260 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
262 for (c
= newc
= 0; c
< oldc
; c
++)
263 if (pvd
->vdev_child
[c
])
266 newchild
= kmem_alloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
268 for (c
= newc
= 0; c
< oldc
; c
++) {
269 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
270 newchild
[newc
] = cvd
;
271 cvd
->vdev_id
= newc
++;
275 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
276 pvd
->vdev_child
= newchild
;
277 pvd
->vdev_children
= newc
;
281 * Allocate and minimally initialize a vdev_t.
284 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
289 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
291 if (spa
->spa_root_vdev
== NULL
) {
292 ASSERT(ops
== &vdev_root_ops
);
293 spa
->spa_root_vdev
= vd
;
296 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
297 if (spa
->spa_root_vdev
== vd
) {
299 * The root vdev's guid will also be the pool guid,
300 * which must be unique among all pools.
302 guid
= spa_generate_guid(NULL
);
305 * Any other vdev's guid must be unique within the pool.
307 guid
= spa_generate_guid(spa
);
309 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
314 vd
->vdev_guid
= guid
;
315 vd
->vdev_guid_sum
= guid
;
317 vd
->vdev_state
= VDEV_STATE_CLOSED
;
318 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
320 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
321 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
322 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
323 for (t
= 0; t
< DTL_TYPES
; t
++) {
324 space_map_create(&vd
->vdev_dtl
[t
], 0, -1ULL, 0,
327 txg_list_create(&vd
->vdev_ms_list
,
328 offsetof(struct metaslab
, ms_txg_node
));
329 txg_list_create(&vd
->vdev_dtl_list
,
330 offsetof(struct vdev
, vdev_dtl_node
));
331 vd
->vdev_stat
.vs_timestamp
= gethrtime();
339 * Allocate a new vdev. The 'alloctype' is used to control whether we are
340 * creating a new vdev or loading an existing one - the behavior is slightly
341 * different for each case.
344 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
349 uint64_t guid
= 0, islog
, nparity
;
352 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
354 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
357 if ((ops
= vdev_getops(type
)) == NULL
)
361 * If this is a load, get the vdev guid from the nvlist.
362 * Otherwise, vdev_alloc_common() will generate one for us.
364 if (alloctype
== VDEV_ALLOC_LOAD
) {
367 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
371 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
373 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
374 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
376 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
377 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
379 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
380 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
385 * The first allocated vdev must be of type 'root'.
387 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
391 * Determine whether we're a log vdev.
394 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
395 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
398 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
402 * Set the nparity property for RAID-Z vdevs.
405 if (ops
== &vdev_raidz_ops
) {
406 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
408 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
411 * Previous versions could only support 1 or 2 parity
415 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
418 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
422 * We require the parity to be specified for SPAs that
423 * support multiple parity levels.
425 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
428 * Otherwise, we default to 1 parity device for RAID-Z.
435 ASSERT(nparity
!= -1ULL);
437 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
439 vd
->vdev_islog
= islog
;
440 vd
->vdev_nparity
= nparity
;
442 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
443 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
444 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
445 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
446 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
447 &vd
->vdev_physpath
) == 0)
448 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
449 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
450 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
453 * Set the whole_disk property. If it's not specified, leave the value
456 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
457 &vd
->vdev_wholedisk
) != 0)
458 vd
->vdev_wholedisk
= -1ULL;
461 * Look for the 'not present' flag. This will only be set if the device
462 * was not present at the time of import.
464 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
465 &vd
->vdev_not_present
);
468 * Get the alignment requirement.
470 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
473 * Retrieve the vdev creation time.
475 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
479 * If we're a top-level vdev, try to load the allocation parameters.
481 if (parent
&& !parent
->vdev_parent
&&
482 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
483 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
485 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
487 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
489 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
493 if (parent
&& !parent
->vdev_parent
) {
494 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
495 alloctype
== VDEV_ALLOC_ADD
||
496 alloctype
== VDEV_ALLOC_SPLIT
||
497 alloctype
== VDEV_ALLOC_ROOTPOOL
);
498 vd
->vdev_mg
= metaslab_group_create(islog
?
499 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
503 * If we're a leaf vdev, try to load the DTL object and other state.
505 if (vd
->vdev_ops
->vdev_op_leaf
&&
506 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
507 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
508 if (alloctype
== VDEV_ALLOC_LOAD
) {
509 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
510 &vd
->vdev_dtl_smo
.smo_object
);
511 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
515 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
518 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
519 &spare
) == 0 && spare
)
523 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
526 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVERING
,
527 &vd
->vdev_resilvering
);
530 * When importing a pool, we want to ignore the persistent fault
531 * state, as the diagnosis made on another system may not be
532 * valid in the current context. Local vdevs will
533 * remain in the faulted state.
535 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
536 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
538 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
540 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
543 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
547 VDEV_AUX_ERR_EXCEEDED
;
548 if (nvlist_lookup_string(nv
,
549 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
550 strcmp(aux
, "external") == 0)
551 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
557 * Add ourselves to the parent's list of children.
559 vdev_add_child(parent
, vd
);
567 vdev_free(vdev_t
*vd
)
570 spa_t
*spa
= vd
->vdev_spa
;
573 * vdev_free() implies closing the vdev first. This is simpler than
574 * trying to ensure complicated semantics for all callers.
578 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
579 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
584 for (c
= 0; c
< vd
->vdev_children
; c
++)
585 vdev_free(vd
->vdev_child
[c
]);
587 ASSERT(vd
->vdev_child
== NULL
);
588 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
591 * Discard allocation state.
593 if (vd
->vdev_mg
!= NULL
) {
594 vdev_metaslab_fini(vd
);
595 metaslab_group_destroy(vd
->vdev_mg
);
598 ASSERT3U(vd
->vdev_stat
.vs_space
, ==, 0);
599 ASSERT3U(vd
->vdev_stat
.vs_dspace
, ==, 0);
600 ASSERT3U(vd
->vdev_stat
.vs_alloc
, ==, 0);
603 * Remove this vdev from its parent's child list.
605 vdev_remove_child(vd
->vdev_parent
, vd
);
607 ASSERT(vd
->vdev_parent
== NULL
);
610 * Clean up vdev structure.
616 spa_strfree(vd
->vdev_path
);
618 spa_strfree(vd
->vdev_devid
);
619 if (vd
->vdev_physpath
)
620 spa_strfree(vd
->vdev_physpath
);
622 spa_strfree(vd
->vdev_fru
);
624 if (vd
->vdev_isspare
)
625 spa_spare_remove(vd
);
626 if (vd
->vdev_isl2cache
)
627 spa_l2cache_remove(vd
);
629 txg_list_destroy(&vd
->vdev_ms_list
);
630 txg_list_destroy(&vd
->vdev_dtl_list
);
632 mutex_enter(&vd
->vdev_dtl_lock
);
633 for (t
= 0; t
< DTL_TYPES
; t
++) {
634 space_map_unload(&vd
->vdev_dtl
[t
]);
635 space_map_destroy(&vd
->vdev_dtl
[t
]);
637 mutex_exit(&vd
->vdev_dtl_lock
);
639 mutex_destroy(&vd
->vdev_dtl_lock
);
640 mutex_destroy(&vd
->vdev_stat_lock
);
641 mutex_destroy(&vd
->vdev_probe_lock
);
643 if (vd
== spa
->spa_root_vdev
)
644 spa
->spa_root_vdev
= NULL
;
646 kmem_free(vd
, sizeof (vdev_t
));
650 * Transfer top-level vdev state from svd to tvd.
653 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
655 spa_t
*spa
= svd
->vdev_spa
;
660 ASSERT(tvd
== tvd
->vdev_top
);
662 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
663 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
664 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
666 svd
->vdev_ms_array
= 0;
667 svd
->vdev_ms_shift
= 0;
668 svd
->vdev_ms_count
= 0;
670 tvd
->vdev_mg
= svd
->vdev_mg
;
671 tvd
->vdev_ms
= svd
->vdev_ms
;
676 if (tvd
->vdev_mg
!= NULL
)
677 tvd
->vdev_mg
->mg_vd
= tvd
;
679 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
680 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
681 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
683 svd
->vdev_stat
.vs_alloc
= 0;
684 svd
->vdev_stat
.vs_space
= 0;
685 svd
->vdev_stat
.vs_dspace
= 0;
687 for (t
= 0; t
< TXG_SIZE
; t
++) {
688 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
689 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
690 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
691 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
692 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
693 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
696 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
697 vdev_config_clean(svd
);
698 vdev_config_dirty(tvd
);
701 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
702 vdev_state_clean(svd
);
703 vdev_state_dirty(tvd
);
706 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
707 svd
->vdev_deflate_ratio
= 0;
709 tvd
->vdev_islog
= svd
->vdev_islog
;
714 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
723 for (c
= 0; c
< vd
->vdev_children
; c
++)
724 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
728 * Add a mirror/replacing vdev above an existing vdev.
731 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
733 spa_t
*spa
= cvd
->vdev_spa
;
734 vdev_t
*pvd
= cvd
->vdev_parent
;
737 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
739 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
741 mvd
->vdev_asize
= cvd
->vdev_asize
;
742 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
743 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
744 mvd
->vdev_state
= cvd
->vdev_state
;
745 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
747 vdev_remove_child(pvd
, cvd
);
748 vdev_add_child(pvd
, mvd
);
749 cvd
->vdev_id
= mvd
->vdev_children
;
750 vdev_add_child(mvd
, cvd
);
751 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
753 if (mvd
== mvd
->vdev_top
)
754 vdev_top_transfer(cvd
, mvd
);
760 * Remove a 1-way mirror/replacing vdev from the tree.
763 vdev_remove_parent(vdev_t
*cvd
)
765 vdev_t
*mvd
= cvd
->vdev_parent
;
766 vdev_t
*pvd
= mvd
->vdev_parent
;
768 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
770 ASSERT(mvd
->vdev_children
== 1);
771 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
772 mvd
->vdev_ops
== &vdev_replacing_ops
||
773 mvd
->vdev_ops
== &vdev_spare_ops
);
774 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
776 vdev_remove_child(mvd
, cvd
);
777 vdev_remove_child(pvd
, mvd
);
780 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
781 * Otherwise, we could have detached an offline device, and when we
782 * go to import the pool we'll think we have two top-level vdevs,
783 * instead of a different version of the same top-level vdev.
785 if (mvd
->vdev_top
== mvd
) {
786 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
787 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
788 cvd
->vdev_guid
+= guid_delta
;
789 cvd
->vdev_guid_sum
+= guid_delta
;
791 cvd
->vdev_id
= mvd
->vdev_id
;
792 vdev_add_child(pvd
, cvd
);
793 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
795 if (cvd
== cvd
->vdev_top
)
796 vdev_top_transfer(mvd
, cvd
);
798 ASSERT(mvd
->vdev_children
== 0);
803 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
805 spa_t
*spa
= vd
->vdev_spa
;
806 objset_t
*mos
= spa
->spa_meta_objset
;
808 uint64_t oldc
= vd
->vdev_ms_count
;
809 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
813 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
816 * This vdev is not being allocated from yet or is a hole.
818 if (vd
->vdev_ms_shift
== 0)
821 ASSERT(!vd
->vdev_ishole
);
824 * Compute the raidz-deflation ratio. Note, we hard-code
825 * in 128k (1 << 17) because it is the current "typical" blocksize.
826 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
827 * or we will inconsistently account for existing bp's.
829 vd
->vdev_deflate_ratio
= (1 << 17) /
830 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
832 ASSERT(oldc
<= newc
);
834 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
837 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
838 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
842 vd
->vdev_ms_count
= newc
;
844 for (m
= oldc
; m
< newc
; m
++) {
845 space_map_obj_t smo
= { 0, 0, 0 };
848 error
= dmu_read(mos
, vd
->vdev_ms_array
,
849 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
855 error
= dmu_bonus_hold(mos
, object
, FTAG
, &db
);
858 ASSERT3U(db
->db_size
, >=, sizeof (smo
));
859 bcopy(db
->db_data
, &smo
, sizeof (smo
));
860 ASSERT3U(smo
.smo_object
, ==, object
);
861 dmu_buf_rele(db
, FTAG
);
864 vd
->vdev_ms
[m
] = metaslab_init(vd
->vdev_mg
, &smo
,
865 m
<< vd
->vdev_ms_shift
, 1ULL << vd
->vdev_ms_shift
, txg
);
869 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
872 * If the vdev is being removed we don't activate
873 * the metaslabs since we want to ensure that no new
874 * allocations are performed on this device.
876 if (oldc
== 0 && !vd
->vdev_removing
)
877 metaslab_group_activate(vd
->vdev_mg
);
880 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
886 vdev_metaslab_fini(vdev_t
*vd
)
889 uint64_t count
= vd
->vdev_ms_count
;
891 if (vd
->vdev_ms
!= NULL
) {
892 metaslab_group_passivate(vd
->vdev_mg
);
893 for (m
= 0; m
< count
; m
++)
894 if (vd
->vdev_ms
[m
] != NULL
)
895 metaslab_fini(vd
->vdev_ms
[m
]);
896 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
901 typedef struct vdev_probe_stats
{
902 boolean_t vps_readable
;
903 boolean_t vps_writeable
;
905 } vdev_probe_stats_t
;
908 vdev_probe_done(zio_t
*zio
)
910 spa_t
*spa
= zio
->io_spa
;
911 vdev_t
*vd
= zio
->io_vd
;
912 vdev_probe_stats_t
*vps
= zio
->io_private
;
914 ASSERT(vd
->vdev_probe_zio
!= NULL
);
916 if (zio
->io_type
== ZIO_TYPE_READ
) {
917 if (zio
->io_error
== 0)
918 vps
->vps_readable
= 1;
919 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
920 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
921 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
922 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
923 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
925 zio_buf_free(zio
->io_data
, zio
->io_size
);
927 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
928 if (zio
->io_error
== 0)
929 vps
->vps_writeable
= 1;
930 zio_buf_free(zio
->io_data
, zio
->io_size
);
931 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
934 vd
->vdev_cant_read
|= !vps
->vps_readable
;
935 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
937 if (vdev_readable(vd
) &&
938 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
941 ASSERT(zio
->io_error
!= 0);
942 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
943 spa
, vd
, NULL
, 0, 0);
944 zio
->io_error
= ENXIO
;
947 mutex_enter(&vd
->vdev_probe_lock
);
948 ASSERT(vd
->vdev_probe_zio
== zio
);
949 vd
->vdev_probe_zio
= NULL
;
950 mutex_exit(&vd
->vdev_probe_lock
);
952 while ((pio
= zio_walk_parents(zio
)) != NULL
)
953 if (!vdev_accessible(vd
, pio
))
954 pio
->io_error
= ENXIO
;
956 kmem_free(vps
, sizeof (*vps
));
961 * Determine whether this device is accessible by reading and writing
962 * to several known locations: the pad regions of each vdev label
963 * but the first (which we leave alone in case it contains a VTOC).
966 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
968 spa_t
*spa
= vd
->vdev_spa
;
969 vdev_probe_stats_t
*vps
= NULL
;
973 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
976 * Don't probe the probe.
978 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
982 * To prevent 'probe storms' when a device fails, we create
983 * just one probe i/o at a time. All zios that want to probe
984 * this vdev will become parents of the probe io.
986 mutex_enter(&vd
->vdev_probe_lock
);
988 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
989 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
991 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
992 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
995 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
997 * vdev_cant_read and vdev_cant_write can only
998 * transition from TRUE to FALSE when we have the
999 * SCL_ZIO lock as writer; otherwise they can only
1000 * transition from FALSE to TRUE. This ensures that
1001 * any zio looking at these values can assume that
1002 * failures persist for the life of the I/O. That's
1003 * important because when a device has intermittent
1004 * connectivity problems, we want to ensure that
1005 * they're ascribed to the device (ENXIO) and not
1008 * Since we hold SCL_ZIO as writer here, clear both
1009 * values so the probe can reevaluate from first
1012 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1013 vd
->vdev_cant_read
= B_FALSE
;
1014 vd
->vdev_cant_write
= B_FALSE
;
1017 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1018 vdev_probe_done
, vps
,
1019 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1022 * We can't change the vdev state in this context, so we
1023 * kick off an async task to do it on our behalf.
1026 vd
->vdev_probe_wanted
= B_TRUE
;
1027 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1032 zio_add_child(zio
, pio
);
1034 mutex_exit(&vd
->vdev_probe_lock
);
1037 ASSERT(zio
!= NULL
);
1041 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1042 zio_nowait(zio_read_phys(pio
, vd
,
1043 vdev_label_offset(vd
->vdev_psize
, l
,
1044 offsetof(vdev_label_t
, vl_pad2
)),
1045 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1046 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1047 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1058 vdev_open_child(void *arg
)
1062 vd
->vdev_open_thread
= curthread
;
1063 vd
->vdev_open_error
= vdev_open(vd
);
1064 vd
->vdev_open_thread
= NULL
;
1068 vdev_uses_zvols(vdev_t
*vd
)
1072 if (vd
->vdev_path
&& strncmp(vd
->vdev_path
, ZVOL_DIR
,
1073 strlen(ZVOL_DIR
)) == 0)
1075 for (c
= 0; c
< vd
->vdev_children
; c
++)
1076 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1082 vdev_open_children(vdev_t
*vd
)
1085 int children
= vd
->vdev_children
;
1089 * in order to handle pools on top of zvols, do the opens
1090 * in a single thread so that the same thread holds the
1091 * spa_namespace_lock
1093 if (vdev_uses_zvols(vd
)) {
1094 for (c
= 0; c
< children
; c
++)
1095 vd
->vdev_child
[c
]->vdev_open_error
=
1096 vdev_open(vd
->vdev_child
[c
]);
1099 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1100 children
, children
, TASKQ_PREPOPULATE
);
1102 for (c
= 0; c
< children
; c
++)
1103 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1110 * Prepare a virtual device for access.
1113 vdev_open(vdev_t
*vd
)
1115 spa_t
*spa
= vd
->vdev_spa
;
1118 uint64_t asize
, psize
;
1119 uint64_t ashift
= 0;
1122 ASSERT(vd
->vdev_open_thread
== curthread
||
1123 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1124 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1125 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1126 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1128 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1129 vd
->vdev_cant_read
= B_FALSE
;
1130 vd
->vdev_cant_write
= B_FALSE
;
1131 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1134 * If this vdev is not removed, check its fault status. If it's
1135 * faulted, bail out of the open.
1137 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1138 ASSERT(vd
->vdev_children
== 0);
1139 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1140 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1141 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1142 vd
->vdev_label_aux
);
1144 } else if (vd
->vdev_offline
) {
1145 ASSERT(vd
->vdev_children
== 0);
1146 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1150 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &ashift
);
1153 * Reset the vdev_reopening flag so that we actually close
1154 * the vdev on error.
1156 vd
->vdev_reopening
= B_FALSE
;
1157 if (zio_injection_enabled
&& error
== 0)
1158 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1161 if (vd
->vdev_removed
&&
1162 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1163 vd
->vdev_removed
= B_FALSE
;
1165 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1166 vd
->vdev_stat
.vs_aux
);
1170 vd
->vdev_removed
= B_FALSE
;
1173 * Recheck the faulted flag now that we have confirmed that
1174 * the vdev is accessible. If we're faulted, bail.
1176 if (vd
->vdev_faulted
) {
1177 ASSERT(vd
->vdev_children
== 0);
1178 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1179 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1180 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1181 vd
->vdev_label_aux
);
1185 if (vd
->vdev_degraded
) {
1186 ASSERT(vd
->vdev_children
== 0);
1187 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1188 VDEV_AUX_ERR_EXCEEDED
);
1190 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1194 * For hole or missing vdevs we just return success.
1196 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1199 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1200 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1201 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1207 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1209 if (vd
->vdev_children
== 0) {
1210 if (osize
< SPA_MINDEVSIZE
) {
1211 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1212 VDEV_AUX_TOO_SMALL
);
1216 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1218 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1219 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1220 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1221 VDEV_AUX_TOO_SMALL
);
1228 vd
->vdev_psize
= psize
;
1231 * Make sure the allocatable size hasn't shrunk.
1233 if (asize
< vd
->vdev_min_asize
) {
1234 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1235 VDEV_AUX_BAD_LABEL
);
1239 if (vd
->vdev_asize
== 0) {
1241 * This is the first-ever open, so use the computed values.
1242 * For testing purposes, a higher ashift can be requested.
1244 vd
->vdev_asize
= asize
;
1245 vd
->vdev_ashift
= MAX(ashift
, vd
->vdev_ashift
);
1248 * Make sure the alignment requirement hasn't increased.
1250 if (ashift
> vd
->vdev_top
->vdev_ashift
) {
1251 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1252 VDEV_AUX_BAD_LABEL
);
1258 * If all children are healthy and the asize has increased,
1259 * then we've experienced dynamic LUN growth. If automatic
1260 * expansion is enabled then use the additional space.
1262 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1263 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1264 vd
->vdev_asize
= asize
;
1266 vdev_set_min_asize(vd
);
1269 * Ensure we can issue some IO before declaring the
1270 * vdev open for business.
1272 if (vd
->vdev_ops
->vdev_op_leaf
&&
1273 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1274 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1275 VDEV_AUX_ERR_EXCEEDED
);
1280 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1281 * resilver. But don't do this if we are doing a reopen for a scrub,
1282 * since this would just restart the scrub we are already doing.
1284 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1285 vdev_resilver_needed(vd
, NULL
, NULL
))
1286 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1292 * Called once the vdevs are all opened, this routine validates the label
1293 * contents. This needs to be done before vdev_load() so that we don't
1294 * inadvertently do repair I/Os to the wrong device.
1296 * This function will only return failure if one of the vdevs indicates that it
1297 * has since been destroyed or exported. This is only possible if
1298 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1299 * will be updated but the function will return 0.
1302 vdev_validate(vdev_t
*vd
)
1304 spa_t
*spa
= vd
->vdev_spa
;
1306 uint64_t guid
= 0, top_guid
;
1310 for (c
= 0; c
< vd
->vdev_children
; c
++)
1311 if (vdev_validate(vd
->vdev_child
[c
]) != 0)
1315 * If the device has already failed, or was marked offline, don't do
1316 * any further validation. Otherwise, label I/O will fail and we will
1317 * overwrite the previous state.
1319 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1320 uint64_t aux_guid
= 0;
1323 if ((label
= vdev_label_read_config(vd
)) == NULL
) {
1324 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1325 VDEV_AUX_BAD_LABEL
);
1330 * Determine if this vdev has been split off into another
1331 * pool. If so, then refuse to open it.
1333 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1334 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1335 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1336 VDEV_AUX_SPLIT_POOL
);
1341 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
,
1342 &guid
) != 0 || guid
!= spa_guid(spa
)) {
1343 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1344 VDEV_AUX_CORRUPT_DATA
);
1349 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1350 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1355 * If this vdev just became a top-level vdev because its
1356 * sibling was detached, it will have adopted the parent's
1357 * vdev guid -- but the label may or may not be on disk yet.
1358 * Fortunately, either version of the label will have the
1359 * same top guid, so if we're a top-level vdev, we can
1360 * safely compare to that instead.
1362 * If we split this vdev off instead, then we also check the
1363 * original pool's guid. We don't want to consider the vdev
1364 * corrupt if it is partway through a split operation.
1366 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1368 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1370 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1371 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1372 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1373 VDEV_AUX_CORRUPT_DATA
);
1378 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1380 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1381 VDEV_AUX_CORRUPT_DATA
);
1389 * If this is a verbatim import, no need to check the
1390 * state of the pool.
1392 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1393 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1394 state
!= POOL_STATE_ACTIVE
)
1398 * If we were able to open and validate a vdev that was
1399 * previously marked permanently unavailable, clear that state
1402 if (vd
->vdev_not_present
)
1403 vd
->vdev_not_present
= 0;
1410 * Close a virtual device.
1413 vdev_close(vdev_t
*vd
)
1415 spa_t
*spa
= vd
->vdev_spa
;
1416 vdev_t
*pvd
= vd
->vdev_parent
;
1418 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1421 * If our parent is reopening, then we are as well, unless we are
1424 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1425 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1427 vd
->vdev_ops
->vdev_op_close(vd
);
1429 vdev_cache_purge(vd
);
1432 * We record the previous state before we close it, so that if we are
1433 * doing a reopen(), we don't generate FMA ereports if we notice that
1434 * it's still faulted.
1436 vd
->vdev_prevstate
= vd
->vdev_state
;
1438 if (vd
->vdev_offline
)
1439 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1441 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1442 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1446 vdev_hold(vdev_t
*vd
)
1448 spa_t
*spa
= vd
->vdev_spa
;
1451 ASSERT(spa_is_root(spa
));
1452 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1455 for (c
= 0; c
< vd
->vdev_children
; c
++)
1456 vdev_hold(vd
->vdev_child
[c
]);
1458 if (vd
->vdev_ops
->vdev_op_leaf
)
1459 vd
->vdev_ops
->vdev_op_hold(vd
);
1463 vdev_rele(vdev_t
*vd
)
1467 ASSERT(spa_is_root(vd
->vdev_spa
));
1468 for (c
= 0; c
< vd
->vdev_children
; c
++)
1469 vdev_rele(vd
->vdev_child
[c
]);
1471 if (vd
->vdev_ops
->vdev_op_leaf
)
1472 vd
->vdev_ops
->vdev_op_rele(vd
);
1476 * Reopen all interior vdevs and any unopened leaves. We don't actually
1477 * reopen leaf vdevs which had previously been opened as they might deadlock
1478 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1479 * If the leaf has never been opened then open it, as usual.
1482 vdev_reopen(vdev_t
*vd
)
1484 spa_t
*spa
= vd
->vdev_spa
;
1486 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1488 /* set the reopening flag unless we're taking the vdev offline */
1489 vd
->vdev_reopening
= !vd
->vdev_offline
;
1491 (void) vdev_open(vd
);
1494 * Call vdev_validate() here to make sure we have the same device.
1495 * Otherwise, a device with an invalid label could be successfully
1496 * opened in response to vdev_reopen().
1499 (void) vdev_validate_aux(vd
);
1500 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1501 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1502 !l2arc_vdev_present(vd
))
1503 l2arc_add_vdev(spa
, vd
);
1505 (void) vdev_validate(vd
);
1509 * Reassess parent vdev's health.
1511 vdev_propagate_state(vd
);
1515 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1520 * Normally, partial opens (e.g. of a mirror) are allowed.
1521 * For a create, however, we want to fail the request if
1522 * there are any components we can't open.
1524 error
= vdev_open(vd
);
1526 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1528 return (error
? error
: ENXIO
);
1532 * Recursively initialize all labels.
1534 if ((error
= vdev_label_init(vd
, txg
, isreplacing
?
1535 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1544 vdev_metaslab_set_size(vdev_t
*vd
)
1547 * Aim for roughly 200 metaslabs per vdev.
1549 vd
->vdev_ms_shift
= highbit(vd
->vdev_asize
/ 200);
1550 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1554 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1556 ASSERT(vd
== vd
->vdev_top
);
1557 ASSERT(!vd
->vdev_ishole
);
1558 ASSERT(ISP2(flags
));
1559 ASSERT(spa_writeable(vd
->vdev_spa
));
1561 if (flags
& VDD_METASLAB
)
1562 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1564 if (flags
& VDD_DTL
)
1565 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1567 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1573 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1574 * the vdev has less than perfect replication. There are four kinds of DTL:
1576 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1578 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1580 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1581 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1582 * txgs that was scrubbed.
1584 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1585 * persistent errors or just some device being offline.
1586 * Unlike the other three, the DTL_OUTAGE map is not generally
1587 * maintained; it's only computed when needed, typically to
1588 * determine whether a device can be detached.
1590 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1591 * either has the data or it doesn't.
1593 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1594 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1595 * if any child is less than fully replicated, then so is its parent.
1596 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1597 * comprising only those txgs which appear in 'maxfaults' or more children;
1598 * those are the txgs we don't have enough replication to read. For example,
1599 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1600 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1601 * two child DTL_MISSING maps.
1603 * It should be clear from the above that to compute the DTLs and outage maps
1604 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1605 * Therefore, that is all we keep on disk. When loading the pool, or after
1606 * a configuration change, we generate all other DTLs from first principles.
1609 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1611 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1613 ASSERT(t
< DTL_TYPES
);
1614 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1615 ASSERT(spa_writeable(vd
->vdev_spa
));
1617 mutex_enter(sm
->sm_lock
);
1618 if (!space_map_contains(sm
, txg
, size
))
1619 space_map_add(sm
, txg
, size
);
1620 mutex_exit(sm
->sm_lock
);
1624 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1626 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1627 boolean_t dirty
= B_FALSE
;
1629 ASSERT(t
< DTL_TYPES
);
1630 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1632 mutex_enter(sm
->sm_lock
);
1633 if (sm
->sm_space
!= 0)
1634 dirty
= space_map_contains(sm
, txg
, size
);
1635 mutex_exit(sm
->sm_lock
);
1641 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1643 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1646 mutex_enter(sm
->sm_lock
);
1647 empty
= (sm
->sm_space
== 0);
1648 mutex_exit(sm
->sm_lock
);
1654 * Reassess DTLs after a config change or scrub completion.
1657 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1659 spa_t
*spa
= vd
->vdev_spa
;
1663 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1665 for (c
= 0; c
< vd
->vdev_children
; c
++)
1666 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1667 scrub_txg
, scrub_done
);
1669 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1672 if (vd
->vdev_ops
->vdev_op_leaf
) {
1673 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1675 mutex_enter(&vd
->vdev_dtl_lock
);
1676 if (scrub_txg
!= 0 &&
1677 (spa
->spa_scrub_started
||
1678 (scn
&& scn
->scn_phys
.scn_errors
== 0))) {
1680 * We completed a scrub up to scrub_txg. If we
1681 * did it without rebooting, then the scrub dtl
1682 * will be valid, so excise the old region and
1683 * fold in the scrub dtl. Otherwise, leave the
1684 * dtl as-is if there was an error.
1686 * There's little trick here: to excise the beginning
1687 * of the DTL_MISSING map, we put it into a reference
1688 * tree and then add a segment with refcnt -1 that
1689 * covers the range [0, scrub_txg). This means
1690 * that each txg in that range has refcnt -1 or 0.
1691 * We then add DTL_SCRUB with a refcnt of 2, so that
1692 * entries in the range [0, scrub_txg) will have a
1693 * positive refcnt -- either 1 or 2. We then convert
1694 * the reference tree into the new DTL_MISSING map.
1696 space_map_ref_create(&reftree
);
1697 space_map_ref_add_map(&reftree
,
1698 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1699 space_map_ref_add_seg(&reftree
, 0, scrub_txg
, -1);
1700 space_map_ref_add_map(&reftree
,
1701 &vd
->vdev_dtl
[DTL_SCRUB
], 2);
1702 space_map_ref_generate_map(&reftree
,
1703 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1704 space_map_ref_destroy(&reftree
);
1706 space_map_vacate(&vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1707 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1708 space_map_add
, &vd
->vdev_dtl
[DTL_PARTIAL
]);
1710 space_map_vacate(&vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1711 space_map_vacate(&vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1712 if (!vdev_readable(vd
))
1713 space_map_add(&vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1715 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1716 space_map_add
, &vd
->vdev_dtl
[DTL_OUTAGE
]);
1717 mutex_exit(&vd
->vdev_dtl_lock
);
1720 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1724 mutex_enter(&vd
->vdev_dtl_lock
);
1725 for (t
= 0; t
< DTL_TYPES
; t
++) {
1726 /* account for child's outage in parent's missing map */
1727 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1729 continue; /* leaf vdevs only */
1730 if (t
== DTL_PARTIAL
)
1731 minref
= 1; /* i.e. non-zero */
1732 else if (vd
->vdev_nparity
!= 0)
1733 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1735 minref
= vd
->vdev_children
; /* any kind of mirror */
1736 space_map_ref_create(&reftree
);
1737 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1738 vdev_t
*cvd
= vd
->vdev_child
[c
];
1739 mutex_enter(&cvd
->vdev_dtl_lock
);
1740 space_map_ref_add_map(&reftree
, &cvd
->vdev_dtl
[s
], 1);
1741 mutex_exit(&cvd
->vdev_dtl_lock
);
1743 space_map_ref_generate_map(&reftree
, &vd
->vdev_dtl
[t
], minref
);
1744 space_map_ref_destroy(&reftree
);
1746 mutex_exit(&vd
->vdev_dtl_lock
);
1750 vdev_dtl_load(vdev_t
*vd
)
1752 spa_t
*spa
= vd
->vdev_spa
;
1753 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1754 objset_t
*mos
= spa
->spa_meta_objset
;
1758 ASSERT(vd
->vdev_children
== 0);
1760 if (smo
->smo_object
== 0)
1763 ASSERT(!vd
->vdev_ishole
);
1765 if ((error
= dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
)) != 0)
1768 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1769 bcopy(db
->db_data
, smo
, sizeof (*smo
));
1770 dmu_buf_rele(db
, FTAG
);
1772 mutex_enter(&vd
->vdev_dtl_lock
);
1773 error
= space_map_load(&vd
->vdev_dtl
[DTL_MISSING
],
1774 NULL
, SM_ALLOC
, smo
, mos
);
1775 mutex_exit(&vd
->vdev_dtl_lock
);
1781 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1783 spa_t
*spa
= vd
->vdev_spa
;
1784 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1785 space_map_t
*sm
= &vd
->vdev_dtl
[DTL_MISSING
];
1786 objset_t
*mos
= spa
->spa_meta_objset
;
1792 ASSERT(!vd
->vdev_ishole
);
1794 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1796 if (vd
->vdev_detached
) {
1797 if (smo
->smo_object
!= 0) {
1798 int err
= dmu_object_free(mos
, smo
->smo_object
, tx
);
1799 ASSERT3U(err
, ==, 0);
1800 smo
->smo_object
= 0;
1806 if (smo
->smo_object
== 0) {
1807 ASSERT(smo
->smo_objsize
== 0);
1808 ASSERT(smo
->smo_alloc
== 0);
1809 smo
->smo_object
= dmu_object_alloc(mos
,
1810 DMU_OT_SPACE_MAP
, 1 << SPACE_MAP_BLOCKSHIFT
,
1811 DMU_OT_SPACE_MAP_HEADER
, sizeof (*smo
), tx
);
1812 ASSERT(smo
->smo_object
!= 0);
1813 vdev_config_dirty(vd
->vdev_top
);
1816 mutex_init(&smlock
, NULL
, MUTEX_DEFAULT
, NULL
);
1818 space_map_create(&smsync
, sm
->sm_start
, sm
->sm_size
, sm
->sm_shift
,
1821 mutex_enter(&smlock
);
1823 mutex_enter(&vd
->vdev_dtl_lock
);
1824 space_map_walk(sm
, space_map_add
, &smsync
);
1825 mutex_exit(&vd
->vdev_dtl_lock
);
1827 space_map_truncate(smo
, mos
, tx
);
1828 space_map_sync(&smsync
, SM_ALLOC
, smo
, mos
, tx
);
1830 space_map_destroy(&smsync
);
1832 mutex_exit(&smlock
);
1833 mutex_destroy(&smlock
);
1835 VERIFY(0 == dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
));
1836 dmu_buf_will_dirty(db
, tx
);
1837 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1838 bcopy(smo
, db
->db_data
, sizeof (*smo
));
1839 dmu_buf_rele(db
, FTAG
);
1845 * Determine whether the specified vdev can be offlined/detached/removed
1846 * without losing data.
1849 vdev_dtl_required(vdev_t
*vd
)
1851 spa_t
*spa
= vd
->vdev_spa
;
1852 vdev_t
*tvd
= vd
->vdev_top
;
1853 uint8_t cant_read
= vd
->vdev_cant_read
;
1856 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1858 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
1862 * Temporarily mark the device as unreadable, and then determine
1863 * whether this results in any DTL outages in the top-level vdev.
1864 * If not, we can safely offline/detach/remove the device.
1866 vd
->vdev_cant_read
= B_TRUE
;
1867 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1868 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
1869 vd
->vdev_cant_read
= cant_read
;
1870 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1872 if (!required
&& zio_injection_enabled
)
1873 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
1879 * Determine if resilver is needed, and if so the txg range.
1882 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
1884 boolean_t needed
= B_FALSE
;
1885 uint64_t thismin
= UINT64_MAX
;
1886 uint64_t thismax
= 0;
1889 if (vd
->vdev_children
== 0) {
1890 mutex_enter(&vd
->vdev_dtl_lock
);
1891 if (vd
->vdev_dtl
[DTL_MISSING
].sm_space
!= 0 &&
1892 vdev_writeable(vd
)) {
1895 ss
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1896 thismin
= ss
->ss_start
- 1;
1897 ss
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1898 thismax
= ss
->ss_end
;
1901 mutex_exit(&vd
->vdev_dtl_lock
);
1903 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1904 vdev_t
*cvd
= vd
->vdev_child
[c
];
1905 uint64_t cmin
, cmax
;
1907 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
1908 thismin
= MIN(thismin
, cmin
);
1909 thismax
= MAX(thismax
, cmax
);
1915 if (needed
&& minp
) {
1923 vdev_load(vdev_t
*vd
)
1928 * Recursively load all children.
1930 for (c
= 0; c
< vd
->vdev_children
; c
++)
1931 vdev_load(vd
->vdev_child
[c
]);
1934 * If this is a top-level vdev, initialize its metaslabs.
1936 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
1937 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
1938 vdev_metaslab_init(vd
, 0) != 0))
1939 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1940 VDEV_AUX_CORRUPT_DATA
);
1943 * If this is a leaf vdev, load its DTL.
1945 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
1946 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1947 VDEV_AUX_CORRUPT_DATA
);
1951 * The special vdev case is used for hot spares and l2cache devices. Its
1952 * sole purpose it to set the vdev state for the associated vdev. To do this,
1953 * we make sure that we can open the underlying device, then try to read the
1954 * label, and make sure that the label is sane and that it hasn't been
1955 * repurposed to another pool.
1958 vdev_validate_aux(vdev_t
*vd
)
1961 uint64_t guid
, version
;
1964 if (!vdev_readable(vd
))
1967 if ((label
= vdev_label_read_config(vd
)) == NULL
) {
1968 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1969 VDEV_AUX_CORRUPT_DATA
);
1973 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
1974 version
> SPA_VERSION
||
1975 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
1976 guid
!= vd
->vdev_guid
||
1977 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
1978 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1979 VDEV_AUX_CORRUPT_DATA
);
1985 * We don't actually check the pool state here. If it's in fact in
1986 * use by another pool, we update this fact on the fly when requested.
1993 vdev_remove(vdev_t
*vd
, uint64_t txg
)
1995 spa_t
*spa
= vd
->vdev_spa
;
1996 objset_t
*mos
= spa
->spa_meta_objset
;
2000 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2002 if (vd
->vdev_dtl_smo
.smo_object
) {
2003 ASSERT3U(vd
->vdev_dtl_smo
.smo_alloc
, ==, 0);
2004 (void) dmu_object_free(mos
, vd
->vdev_dtl_smo
.smo_object
, tx
);
2005 vd
->vdev_dtl_smo
.smo_object
= 0;
2008 if (vd
->vdev_ms
!= NULL
) {
2009 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2010 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2012 if (msp
== NULL
|| msp
->ms_smo
.smo_object
== 0)
2015 ASSERT3U(msp
->ms_smo
.smo_alloc
, ==, 0);
2016 (void) dmu_object_free(mos
, msp
->ms_smo
.smo_object
, tx
);
2017 msp
->ms_smo
.smo_object
= 0;
2021 if (vd
->vdev_ms_array
) {
2022 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2023 vd
->vdev_ms_array
= 0;
2024 vd
->vdev_ms_shift
= 0;
2030 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2033 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2035 ASSERT(!vd
->vdev_ishole
);
2037 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2038 metaslab_sync_done(msp
, txg
);
2041 metaslab_sync_reassess(vd
->vdev_mg
);
2045 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2047 spa_t
*spa
= vd
->vdev_spa
;
2052 ASSERT(!vd
->vdev_ishole
);
2054 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2055 ASSERT(vd
== vd
->vdev_top
);
2056 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2057 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2058 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2059 ASSERT(vd
->vdev_ms_array
!= 0);
2060 vdev_config_dirty(vd
);
2065 * Remove the metadata associated with this vdev once it's empty.
2067 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2068 vdev_remove(vd
, txg
);
2070 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2071 metaslab_sync(msp
, txg
);
2072 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2075 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2076 vdev_dtl_sync(lvd
, txg
);
2078 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2082 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2084 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2088 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2089 * not be opened, and no I/O is attempted.
2092 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2096 spa_vdev_state_enter(spa
, SCL_NONE
);
2098 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2099 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2101 if (!vd
->vdev_ops
->vdev_op_leaf
)
2102 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2107 * We don't directly use the aux state here, but if we do a
2108 * vdev_reopen(), we need this value to be present to remember why we
2111 vd
->vdev_label_aux
= aux
;
2114 * Faulted state takes precedence over degraded.
2116 vd
->vdev_delayed_close
= B_FALSE
;
2117 vd
->vdev_faulted
= 1ULL;
2118 vd
->vdev_degraded
= 0ULL;
2119 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2122 * If this device has the only valid copy of the data, then
2123 * back off and simply mark the vdev as degraded instead.
2125 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2126 vd
->vdev_degraded
= 1ULL;
2127 vd
->vdev_faulted
= 0ULL;
2130 * If we reopen the device and it's not dead, only then do we
2135 if (vdev_readable(vd
))
2136 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2139 return (spa_vdev_state_exit(spa
, vd
, 0));
2143 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2144 * user that something is wrong. The vdev continues to operate as normal as far
2145 * as I/O is concerned.
2148 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2152 spa_vdev_state_enter(spa
, SCL_NONE
);
2154 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2155 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2157 if (!vd
->vdev_ops
->vdev_op_leaf
)
2158 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2161 * If the vdev is already faulted, then don't do anything.
2163 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2164 return (spa_vdev_state_exit(spa
, NULL
, 0));
2166 vd
->vdev_degraded
= 1ULL;
2167 if (!vdev_is_dead(vd
))
2168 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2171 return (spa_vdev_state_exit(spa
, vd
, 0));
2175 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2176 * any attached spare device should be detached when the device finishes
2177 * resilvering. Second, the online should be treated like a 'test' online case,
2178 * so no FMA events are generated if the device fails to open.
2181 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2183 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2185 spa_vdev_state_enter(spa
, SCL_NONE
);
2187 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2188 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2190 if (!vd
->vdev_ops
->vdev_op_leaf
)
2191 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2194 vd
->vdev_offline
= B_FALSE
;
2195 vd
->vdev_tmpoffline
= B_FALSE
;
2196 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2197 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2199 /* XXX - L2ARC 1.0 does not support expansion */
2200 if (!vd
->vdev_aux
) {
2201 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2202 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2206 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2208 if (!vd
->vdev_aux
) {
2209 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2210 pvd
->vdev_expanding
= B_FALSE
;
2214 *newstate
= vd
->vdev_state
;
2215 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2216 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2217 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2218 vd
->vdev_parent
->vdev_child
[0] == vd
)
2219 vd
->vdev_unspare
= B_TRUE
;
2221 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2223 /* XXX - L2ARC 1.0 does not support expansion */
2225 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2226 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2228 return (spa_vdev_state_exit(spa
, vd
, 0));
2232 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2236 uint64_t generation
;
2237 metaslab_group_t
*mg
;
2240 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2242 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2243 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2245 if (!vd
->vdev_ops
->vdev_op_leaf
)
2246 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2250 generation
= spa
->spa_config_generation
+ 1;
2253 * If the device isn't already offline, try to offline it.
2255 if (!vd
->vdev_offline
) {
2257 * If this device has the only valid copy of some data,
2258 * don't allow it to be offlined. Log devices are always
2261 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2262 vdev_dtl_required(vd
))
2263 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2266 * If the top-level is a slog and it has had allocations
2267 * then proceed. We check that the vdev's metaslab group
2268 * is not NULL since it's possible that we may have just
2269 * added this vdev but not yet initialized its metaslabs.
2271 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2273 * Prevent any future allocations.
2275 metaslab_group_passivate(mg
);
2276 (void) spa_vdev_state_exit(spa
, vd
, 0);
2278 error
= spa_offline_log(spa
);
2280 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2283 * Check to see if the config has changed.
2285 if (error
|| generation
!= spa
->spa_config_generation
) {
2286 metaslab_group_activate(mg
);
2288 return (spa_vdev_state_exit(spa
,
2290 (void) spa_vdev_state_exit(spa
, vd
, 0);
2293 ASSERT3U(tvd
->vdev_stat
.vs_alloc
, ==, 0);
2297 * Offline this device and reopen its top-level vdev.
2298 * If the top-level vdev is a log device then just offline
2299 * it. Otherwise, if this action results in the top-level
2300 * vdev becoming unusable, undo it and fail the request.
2302 vd
->vdev_offline
= B_TRUE
;
2305 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2306 vdev_is_dead(tvd
)) {
2307 vd
->vdev_offline
= B_FALSE
;
2309 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2313 * Add the device back into the metaslab rotor so that
2314 * once we online the device it's open for business.
2316 if (tvd
->vdev_islog
&& mg
!= NULL
)
2317 metaslab_group_activate(mg
);
2320 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2322 return (spa_vdev_state_exit(spa
, vd
, 0));
2326 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2330 mutex_enter(&spa
->spa_vdev_top_lock
);
2331 error
= vdev_offline_locked(spa
, guid
, flags
);
2332 mutex_exit(&spa
->spa_vdev_top_lock
);
2338 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2339 * vdev_offline(), we assume the spa config is locked. We also clear all
2340 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2343 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2345 vdev_t
*rvd
= spa
->spa_root_vdev
;
2348 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2353 vd
->vdev_stat
.vs_read_errors
= 0;
2354 vd
->vdev_stat
.vs_write_errors
= 0;
2355 vd
->vdev_stat
.vs_checksum_errors
= 0;
2357 for (c
= 0; c
< vd
->vdev_children
; c
++)
2358 vdev_clear(spa
, vd
->vdev_child
[c
]);
2361 * If we're in the FAULTED state or have experienced failed I/O, then
2362 * clear the persistent state and attempt to reopen the device. We
2363 * also mark the vdev config dirty, so that the new faulted state is
2364 * written out to disk.
2366 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2367 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2370 * When reopening in reponse to a clear event, it may be due to
2371 * a fmadm repair request. In this case, if the device is
2372 * still broken, we want to still post the ereport again.
2374 vd
->vdev_forcefault
= B_TRUE
;
2376 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2377 vd
->vdev_cant_read
= B_FALSE
;
2378 vd
->vdev_cant_write
= B_FALSE
;
2380 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2382 vd
->vdev_forcefault
= B_FALSE
;
2384 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2385 vdev_state_dirty(vd
->vdev_top
);
2387 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2388 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2390 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2394 * When clearing a FMA-diagnosed fault, we always want to
2395 * unspare the device, as we assume that the original spare was
2396 * done in response to the FMA fault.
2398 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2399 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2400 vd
->vdev_parent
->vdev_child
[0] == vd
)
2401 vd
->vdev_unspare
= B_TRUE
;
2405 vdev_is_dead(vdev_t
*vd
)
2408 * Holes and missing devices are always considered "dead".
2409 * This simplifies the code since we don't have to check for
2410 * these types of devices in the various code paths.
2411 * Instead we rely on the fact that we skip over dead devices
2412 * before issuing I/O to them.
2414 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2415 vd
->vdev_ops
== &vdev_missing_ops
);
2419 vdev_readable(vdev_t
*vd
)
2421 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2425 vdev_writeable(vdev_t
*vd
)
2427 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2431 vdev_allocatable(vdev_t
*vd
)
2433 uint64_t state
= vd
->vdev_state
;
2436 * We currently allow allocations from vdevs which may be in the
2437 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2438 * fails to reopen then we'll catch it later when we're holding
2439 * the proper locks. Note that we have to get the vdev state
2440 * in a local variable because although it changes atomically,
2441 * we're asking two separate questions about it.
2443 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2444 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2448 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2450 ASSERT(zio
->io_vd
== vd
);
2452 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2455 if (zio
->io_type
== ZIO_TYPE_READ
)
2456 return (!vd
->vdev_cant_read
);
2458 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2459 return (!vd
->vdev_cant_write
);
2465 * Get statistics for the given vdev.
2468 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2470 vdev_t
*rvd
= vd
->vdev_spa
->spa_root_vdev
;
2473 mutex_enter(&vd
->vdev_stat_lock
);
2474 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2475 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2476 vs
->vs_state
= vd
->vdev_state
;
2477 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2478 if (vd
->vdev_ops
->vdev_op_leaf
)
2479 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2480 mutex_exit(&vd
->vdev_stat_lock
);
2483 * If we're getting stats on the root vdev, aggregate the I/O counts
2484 * over all top-level vdevs (i.e. the direct children of the root).
2487 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2488 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2489 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2491 mutex_enter(&vd
->vdev_stat_lock
);
2492 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2493 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2494 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2496 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2497 mutex_exit(&vd
->vdev_stat_lock
);
2503 vdev_clear_stats(vdev_t
*vd
)
2505 mutex_enter(&vd
->vdev_stat_lock
);
2506 vd
->vdev_stat
.vs_space
= 0;
2507 vd
->vdev_stat
.vs_dspace
= 0;
2508 vd
->vdev_stat
.vs_alloc
= 0;
2509 mutex_exit(&vd
->vdev_stat_lock
);
2513 vdev_scan_stat_init(vdev_t
*vd
)
2515 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2518 for (c
= 0; c
< vd
->vdev_children
; c
++)
2519 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2521 mutex_enter(&vd
->vdev_stat_lock
);
2522 vs
->vs_scan_processed
= 0;
2523 mutex_exit(&vd
->vdev_stat_lock
);
2527 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2529 spa_t
*spa
= zio
->io_spa
;
2530 vdev_t
*rvd
= spa
->spa_root_vdev
;
2531 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2533 uint64_t txg
= zio
->io_txg
;
2534 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2535 zio_type_t type
= zio
->io_type
;
2536 int flags
= zio
->io_flags
;
2539 * If this i/o is a gang leader, it didn't do any actual work.
2541 if (zio
->io_gang_tree
)
2544 if (zio
->io_error
== 0) {
2546 * If this is a root i/o, don't count it -- we've already
2547 * counted the top-level vdevs, and vdev_get_stats() will
2548 * aggregate them when asked. This reduces contention on
2549 * the root vdev_stat_lock and implicitly handles blocks
2550 * that compress away to holes, for which there is no i/o.
2551 * (Holes never create vdev children, so all the counters
2552 * remain zero, which is what we want.)
2554 * Note: this only applies to successful i/o (io_error == 0)
2555 * because unlike i/o counts, errors are not additive.
2556 * When reading a ditto block, for example, failure of
2557 * one top-level vdev does not imply a root-level error.
2562 ASSERT(vd
== zio
->io_vd
);
2564 if (flags
& ZIO_FLAG_IO_BYPASS
)
2567 mutex_enter(&vd
->vdev_stat_lock
);
2569 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2570 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2571 dsl_scan_phys_t
*scn_phys
=
2572 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2573 uint64_t *processed
= &scn_phys
->scn_processed
;
2576 if (vd
->vdev_ops
->vdev_op_leaf
)
2577 atomic_add_64(processed
, psize
);
2578 vs
->vs_scan_processed
+= psize
;
2581 if (flags
& ZIO_FLAG_SELF_HEAL
)
2582 vs
->vs_self_healed
+= psize
;
2586 vs
->vs_bytes
[type
] += psize
;
2588 mutex_exit(&vd
->vdev_stat_lock
);
2592 if (flags
& ZIO_FLAG_SPECULATIVE
)
2596 * If this is an I/O error that is going to be retried, then ignore the
2597 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2598 * hard errors, when in reality they can happen for any number of
2599 * innocuous reasons (bus resets, MPxIO link failure, etc).
2601 if (zio
->io_error
== EIO
&&
2602 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2606 * Intent logs writes won't propagate their error to the root
2607 * I/O so don't mark these types of failures as pool-level
2610 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2613 mutex_enter(&vd
->vdev_stat_lock
);
2614 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2615 if (zio
->io_error
== ECKSUM
)
2616 vs
->vs_checksum_errors
++;
2618 vs
->vs_read_errors
++;
2620 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2621 vs
->vs_write_errors
++;
2622 mutex_exit(&vd
->vdev_stat_lock
);
2624 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2625 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2626 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2627 spa
->spa_claiming
)) {
2629 * This is either a normal write (not a repair), or it's
2630 * a repair induced by the scrub thread, or it's a repair
2631 * made by zil_claim() during spa_load() in the first txg.
2632 * In the normal case, we commit the DTL change in the same
2633 * txg as the block was born. In the scrub-induced repair
2634 * case, we know that scrubs run in first-pass syncing context,
2635 * so we commit the DTL change in spa_syncing_txg(spa).
2636 * In the zil_claim() case, we commit in spa_first_txg(spa).
2638 * We currently do not make DTL entries for failed spontaneous
2639 * self-healing writes triggered by normal (non-scrubbing)
2640 * reads, because we have no transactional context in which to
2641 * do so -- and it's not clear that it'd be desirable anyway.
2643 if (vd
->vdev_ops
->vdev_op_leaf
) {
2644 uint64_t commit_txg
= txg
;
2645 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2646 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2647 ASSERT(spa_sync_pass(spa
) == 1);
2648 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2649 commit_txg
= spa_syncing_txg(spa
);
2650 } else if (spa
->spa_claiming
) {
2651 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2652 commit_txg
= spa_first_txg(spa
);
2654 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2655 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2657 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2658 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2659 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2662 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2667 * Update the in-core space usage stats for this vdev, its metaslab class,
2668 * and the root vdev.
2671 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2672 int64_t space_delta
)
2674 int64_t dspace_delta
= space_delta
;
2675 spa_t
*spa
= vd
->vdev_spa
;
2676 vdev_t
*rvd
= spa
->spa_root_vdev
;
2677 metaslab_group_t
*mg
= vd
->vdev_mg
;
2678 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2680 ASSERT(vd
== vd
->vdev_top
);
2683 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2684 * factor. We must calculate this here and not at the root vdev
2685 * because the root vdev's psize-to-asize is simply the max of its
2686 * childrens', thus not accurate enough for us.
2688 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2689 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2690 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2691 vd
->vdev_deflate_ratio
;
2693 mutex_enter(&vd
->vdev_stat_lock
);
2694 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2695 vd
->vdev_stat
.vs_space
+= space_delta
;
2696 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2697 mutex_exit(&vd
->vdev_stat_lock
);
2699 if (mc
== spa_normal_class(spa
)) {
2700 mutex_enter(&rvd
->vdev_stat_lock
);
2701 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2702 rvd
->vdev_stat
.vs_space
+= space_delta
;
2703 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2704 mutex_exit(&rvd
->vdev_stat_lock
);
2708 ASSERT(rvd
== vd
->vdev_parent
);
2709 ASSERT(vd
->vdev_ms_count
!= 0);
2711 metaslab_class_space_update(mc
,
2712 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2717 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2718 * so that it will be written out next time the vdev configuration is synced.
2719 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2722 vdev_config_dirty(vdev_t
*vd
)
2724 spa_t
*spa
= vd
->vdev_spa
;
2725 vdev_t
*rvd
= spa
->spa_root_vdev
;
2728 ASSERT(spa_writeable(spa
));
2731 * If this is an aux vdev (as with l2cache and spare devices), then we
2732 * update the vdev config manually and set the sync flag.
2734 if (vd
->vdev_aux
!= NULL
) {
2735 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2739 for (c
= 0; c
< sav
->sav_count
; c
++) {
2740 if (sav
->sav_vdevs
[c
] == vd
)
2744 if (c
== sav
->sav_count
) {
2746 * We're being removed. There's nothing more to do.
2748 ASSERT(sav
->sav_sync
== B_TRUE
);
2752 sav
->sav_sync
= B_TRUE
;
2754 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2755 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2756 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2757 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2763 * Setting the nvlist in the middle if the array is a little
2764 * sketchy, but it will work.
2766 nvlist_free(aux
[c
]);
2767 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
2773 * The dirty list is protected by the SCL_CONFIG lock. The caller
2774 * must either hold SCL_CONFIG as writer, or must be the sync thread
2775 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2776 * so this is sufficient to ensure mutual exclusion.
2778 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2779 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2780 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2783 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2784 vdev_config_dirty(rvd
->vdev_child
[c
]);
2786 ASSERT(vd
== vd
->vdev_top
);
2788 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
2790 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
2795 vdev_config_clean(vdev_t
*vd
)
2797 spa_t
*spa
= vd
->vdev_spa
;
2799 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2800 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2801 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2803 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
2804 list_remove(&spa
->spa_config_dirty_list
, vd
);
2808 * Mark a top-level vdev's state as dirty, so that the next pass of
2809 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2810 * the state changes from larger config changes because they require
2811 * much less locking, and are often needed for administrative actions.
2814 vdev_state_dirty(vdev_t
*vd
)
2816 spa_t
*spa
= vd
->vdev_spa
;
2818 ASSERT(spa_writeable(spa
));
2819 ASSERT(vd
== vd
->vdev_top
);
2822 * The state list is protected by the SCL_STATE lock. The caller
2823 * must either hold SCL_STATE as writer, or must be the sync thread
2824 * (which holds SCL_STATE as reader). There's only one sync thread,
2825 * so this is sufficient to ensure mutual exclusion.
2827 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2828 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2829 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2831 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
2832 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
2836 vdev_state_clean(vdev_t
*vd
)
2838 spa_t
*spa
= vd
->vdev_spa
;
2840 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2841 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2842 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2844 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
2845 list_remove(&spa
->spa_state_dirty_list
, vd
);
2849 * Propagate vdev state up from children to parent.
2852 vdev_propagate_state(vdev_t
*vd
)
2854 spa_t
*spa
= vd
->vdev_spa
;
2855 vdev_t
*rvd
= spa
->spa_root_vdev
;
2856 int degraded
= 0, faulted
= 0;
2861 if (vd
->vdev_children
> 0) {
2862 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2863 child
= vd
->vdev_child
[c
];
2866 * Don't factor holes into the decision.
2868 if (child
->vdev_ishole
)
2871 if (!vdev_readable(child
) ||
2872 (!vdev_writeable(child
) && spa_writeable(spa
))) {
2874 * Root special: if there is a top-level log
2875 * device, treat the root vdev as if it were
2878 if (child
->vdev_islog
&& vd
== rvd
)
2882 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
2886 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
2890 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
2893 * Root special: if there is a top-level vdev that cannot be
2894 * opened due to corrupted metadata, then propagate the root
2895 * vdev's aux state as 'corrupt' rather than 'insufficient
2898 if (corrupted
&& vd
== rvd
&&
2899 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
2900 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2901 VDEV_AUX_CORRUPT_DATA
);
2904 if (vd
->vdev_parent
)
2905 vdev_propagate_state(vd
->vdev_parent
);
2909 * Set a vdev's state. If this is during an open, we don't update the parent
2910 * state, because we're in the process of opening children depth-first.
2911 * Otherwise, we propagate the change to the parent.
2913 * If this routine places a device in a faulted state, an appropriate ereport is
2917 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
2919 uint64_t save_state
;
2920 spa_t
*spa
= vd
->vdev_spa
;
2922 if (state
== vd
->vdev_state
) {
2923 vd
->vdev_stat
.vs_aux
= aux
;
2927 save_state
= vd
->vdev_state
;
2929 vd
->vdev_state
= state
;
2930 vd
->vdev_stat
.vs_aux
= aux
;
2933 * If we are setting the vdev state to anything but an open state, then
2934 * always close the underlying device unless the device has requested
2935 * a delayed close (i.e. we're about to remove or fault the device).
2936 * Otherwise, we keep accessible but invalid devices open forever.
2937 * We don't call vdev_close() itself, because that implies some extra
2938 * checks (offline, etc) that we don't want here. This is limited to
2939 * leaf devices, because otherwise closing the device will affect other
2942 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
2943 vd
->vdev_ops
->vdev_op_leaf
)
2944 vd
->vdev_ops
->vdev_op_close(vd
);
2947 * If we have brought this vdev back into service, we need
2948 * to notify fmd so that it can gracefully repair any outstanding
2949 * cases due to a missing device. We do this in all cases, even those
2950 * that probably don't correlate to a repaired fault. This is sure to
2951 * catch all cases, and we let the zfs-retire agent sort it out. If
2952 * this is a transient state it's OK, as the retire agent will
2953 * double-check the state of the vdev before repairing it.
2955 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
2956 vd
->vdev_prevstate
!= state
)
2957 zfs_post_state_change(spa
, vd
);
2959 if (vd
->vdev_removed
&&
2960 state
== VDEV_STATE_CANT_OPEN
&&
2961 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
2963 * If the previous state is set to VDEV_STATE_REMOVED, then this
2964 * device was previously marked removed and someone attempted to
2965 * reopen it. If this failed due to a nonexistent device, then
2966 * keep the device in the REMOVED state. We also let this be if
2967 * it is one of our special test online cases, which is only
2968 * attempting to online the device and shouldn't generate an FMA
2971 vd
->vdev_state
= VDEV_STATE_REMOVED
;
2972 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2973 } else if (state
== VDEV_STATE_REMOVED
) {
2974 vd
->vdev_removed
= B_TRUE
;
2975 } else if (state
== VDEV_STATE_CANT_OPEN
) {
2977 * If we fail to open a vdev during an import or recovery, we
2978 * mark it as "not available", which signifies that it was
2979 * never there to begin with. Failure to open such a device
2980 * is not considered an error.
2982 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
2983 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
2984 vd
->vdev_ops
->vdev_op_leaf
)
2985 vd
->vdev_not_present
= 1;
2988 * Post the appropriate ereport. If the 'prevstate' field is
2989 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2990 * that this is part of a vdev_reopen(). In this case, we don't
2991 * want to post the ereport if the device was already in the
2992 * CANT_OPEN state beforehand.
2994 * If the 'checkremove' flag is set, then this is an attempt to
2995 * online the device in response to an insertion event. If we
2996 * hit this case, then we have detected an insertion event for a
2997 * faulted or offline device that wasn't in the removed state.
2998 * In this scenario, we don't post an ereport because we are
2999 * about to replace the device, or attempt an online with
3000 * vdev_forcefault, which will generate the fault for us.
3002 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3003 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3004 vd
!= spa
->spa_root_vdev
) {
3008 case VDEV_AUX_OPEN_FAILED
:
3009 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3011 case VDEV_AUX_CORRUPT_DATA
:
3012 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3014 case VDEV_AUX_NO_REPLICAS
:
3015 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3017 case VDEV_AUX_BAD_GUID_SUM
:
3018 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3020 case VDEV_AUX_TOO_SMALL
:
3021 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3023 case VDEV_AUX_BAD_LABEL
:
3024 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3027 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3030 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3033 /* Erase any notion of persistent removed state */
3034 vd
->vdev_removed
= B_FALSE
;
3036 vd
->vdev_removed
= B_FALSE
;
3039 if (!isopen
&& vd
->vdev_parent
)
3040 vdev_propagate_state(vd
->vdev_parent
);
3044 * Check the vdev configuration to ensure that it's capable of supporting
3045 * a root pool. Currently, we do not support RAID-Z or partial configuration.
3046 * In addition, only a single top-level vdev is allowed and none of the leaves
3047 * can be wholedisks.
3050 vdev_is_bootable(vdev_t
*vd
)
3054 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3055 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3057 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3058 vd
->vdev_children
> 1) {
3060 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3061 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3064 } else if (vd
->vdev_wholedisk
== 1) {
3068 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3069 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3076 * Load the state from the original vdev tree (ovd) which
3077 * we've retrieved from the MOS config object. If the original
3078 * vdev was offline or faulted then we transfer that state to the
3079 * device in the current vdev tree (nvd).
3082 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3086 ASSERT(nvd
->vdev_top
->vdev_islog
);
3087 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3088 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3090 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3091 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3093 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3095 * Restore the persistent vdev state
3097 nvd
->vdev_offline
= ovd
->vdev_offline
;
3098 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3099 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3100 nvd
->vdev_removed
= ovd
->vdev_removed
;
3105 * Determine if a log device has valid content. If the vdev was
3106 * removed or faulted in the MOS config then we know that
3107 * the content on the log device has already been written to the pool.
3110 vdev_log_state_valid(vdev_t
*vd
)
3114 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3118 for (c
= 0; c
< vd
->vdev_children
; c
++)
3119 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3126 * Expand a vdev if possible.
3129 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3131 ASSERT(vd
->vdev_top
== vd
);
3132 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3134 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3135 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3136 vdev_config_dirty(vd
);
3144 vdev_split(vdev_t
*vd
)
3146 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3148 vdev_remove_child(pvd
, vd
);
3149 vdev_compact_children(pvd
);
3151 cvd
= pvd
->vdev_child
[0];
3152 if (pvd
->vdev_children
== 1) {
3153 vdev_remove_parent(cvd
);
3154 cvd
->vdev_splitting
= B_TRUE
;
3156 vdev_propagate_state(cvd
);