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>
48 * Virtual device management.
51 static vdev_ops_t
*vdev_ops_table
[] = {
64 /* maximum scrub/resilver I/O queue per leaf vdev */
65 int zfs_scrub_limit
= 10;
68 * Given a vdev type, return the appropriate ops vector.
71 vdev_getops(const char *type
)
73 vdev_ops_t
*ops
, **opspp
;
75 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
76 if (strcmp(ops
->vdev_op_type
, type
) == 0)
83 * Default asize function: return the MAX of psize with the asize of
84 * all children. This is what's used by anything other than RAID-Z.
87 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
89 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
93 for (c
= 0; c
< vd
->vdev_children
; c
++) {
94 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
95 asize
= MAX(asize
, csize
);
102 * Get the minimum allocatable size. We define the allocatable size as
103 * the vdev's asize rounded to the nearest metaslab. This allows us to
104 * replace or attach devices which don't have the same physical size but
105 * can still satisfy the same number of allocations.
108 vdev_get_min_asize(vdev_t
*vd
)
110 vdev_t
*pvd
= vd
->vdev_parent
;
113 * If our parent is NULL (inactive spare or cache) or is the root,
114 * just return our own asize.
117 return (vd
->vdev_asize
);
120 * The top-level vdev just returns the allocatable size rounded
121 * to the nearest metaslab.
123 if (vd
== vd
->vdev_top
)
124 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
127 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
128 * so each child must provide at least 1/Nth of its asize.
130 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
131 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
133 return (pvd
->vdev_min_asize
);
137 vdev_set_min_asize(vdev_t
*vd
)
140 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
142 for (c
= 0; c
< vd
->vdev_children
; c
++)
143 vdev_set_min_asize(vd
->vdev_child
[c
]);
147 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
149 vdev_t
*rvd
= spa
->spa_root_vdev
;
151 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
153 if (vdev
< rvd
->vdev_children
) {
154 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
155 return (rvd
->vdev_child
[vdev
]);
162 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
167 if (vd
->vdev_guid
== guid
)
170 for (c
= 0; c
< vd
->vdev_children
; c
++)
171 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
179 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
181 size_t oldsize
, newsize
;
182 uint64_t id
= cvd
->vdev_id
;
185 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
186 ASSERT(cvd
->vdev_parent
== NULL
);
188 cvd
->vdev_parent
= pvd
;
193 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
195 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
196 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
197 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
199 newchild
= kmem_zalloc(newsize
, KM_PUSHPAGE
);
200 if (pvd
->vdev_child
!= NULL
) {
201 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
202 kmem_free(pvd
->vdev_child
, oldsize
);
205 pvd
->vdev_child
= newchild
;
206 pvd
->vdev_child
[id
] = cvd
;
208 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
209 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
212 * Walk up all ancestors to update guid sum.
214 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
215 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
219 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
222 uint_t id
= cvd
->vdev_id
;
224 ASSERT(cvd
->vdev_parent
== pvd
);
229 ASSERT(id
< pvd
->vdev_children
);
230 ASSERT(pvd
->vdev_child
[id
] == cvd
);
232 pvd
->vdev_child
[id
] = NULL
;
233 cvd
->vdev_parent
= NULL
;
235 for (c
= 0; c
< pvd
->vdev_children
; c
++)
236 if (pvd
->vdev_child
[c
])
239 if (c
== pvd
->vdev_children
) {
240 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
241 pvd
->vdev_child
= NULL
;
242 pvd
->vdev_children
= 0;
246 * Walk up all ancestors to update guid sum.
248 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
249 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
253 * Remove any holes in the child array.
256 vdev_compact_children(vdev_t
*pvd
)
258 vdev_t
**newchild
, *cvd
;
259 int oldc
= pvd
->vdev_children
;
263 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
265 for (c
= newc
= 0; c
< oldc
; c
++)
266 if (pvd
->vdev_child
[c
])
269 newchild
= kmem_alloc(newc
* sizeof (vdev_t
*), KM_PUSHPAGE
);
271 for (c
= newc
= 0; c
< oldc
; c
++) {
272 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
273 newchild
[newc
] = cvd
;
274 cvd
->vdev_id
= newc
++;
278 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
279 pvd
->vdev_child
= newchild
;
280 pvd
->vdev_children
= newc
;
284 * Allocate and minimally initialize a vdev_t.
287 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
292 vd
= kmem_zalloc(sizeof (vdev_t
), KM_PUSHPAGE
);
294 if (spa
->spa_root_vdev
== NULL
) {
295 ASSERT(ops
== &vdev_root_ops
);
296 spa
->spa_root_vdev
= vd
;
297 spa
->spa_load_guid
= spa_generate_guid(NULL
);
300 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
301 if (spa
->spa_root_vdev
== vd
) {
303 * The root vdev's guid will also be the pool guid,
304 * which must be unique among all pools.
306 guid
= spa_generate_guid(NULL
);
309 * Any other vdev's guid must be unique within the pool.
311 guid
= spa_generate_guid(spa
);
313 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
318 vd
->vdev_guid
= guid
;
319 vd
->vdev_guid_sum
= guid
;
321 vd
->vdev_state
= VDEV_STATE_CLOSED
;
322 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
324 list_link_init(&vd
->vdev_config_dirty_node
);
325 list_link_init(&vd
->vdev_state_dirty_node
);
326 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
327 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
328 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
329 for (t
= 0; t
< DTL_TYPES
; t
++) {
330 space_map_create(&vd
->vdev_dtl
[t
], 0, -1ULL, 0,
333 txg_list_create(&vd
->vdev_ms_list
,
334 offsetof(struct metaslab
, ms_txg_node
));
335 txg_list_create(&vd
->vdev_dtl_list
,
336 offsetof(struct vdev
, vdev_dtl_node
));
337 vd
->vdev_stat
.vs_timestamp
= gethrtime();
345 * Allocate a new vdev. The 'alloctype' is used to control whether we are
346 * creating a new vdev or loading an existing one - the behavior is slightly
347 * different for each case.
350 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
355 uint64_t guid
= 0, islog
, nparity
;
358 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
360 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
363 if ((ops
= vdev_getops(type
)) == NULL
)
367 * If this is a load, get the vdev guid from the nvlist.
368 * Otherwise, vdev_alloc_common() will generate one for us.
370 if (alloctype
== VDEV_ALLOC_LOAD
) {
373 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
377 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
379 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
380 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
382 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
383 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
385 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
386 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
391 * The first allocated vdev must be of type 'root'.
393 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
397 * Determine whether we're a log vdev.
400 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
401 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
404 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
408 * Set the nparity property for RAID-Z vdevs.
411 if (ops
== &vdev_raidz_ops
) {
412 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
414 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
417 * Previous versions could only support 1 or 2 parity
421 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
424 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
428 * We require the parity to be specified for SPAs that
429 * support multiple parity levels.
431 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
434 * Otherwise, we default to 1 parity device for RAID-Z.
441 ASSERT(nparity
!= -1ULL);
443 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
445 vd
->vdev_islog
= islog
;
446 vd
->vdev_nparity
= nparity
;
448 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
449 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
450 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
451 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
452 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
453 &vd
->vdev_physpath
) == 0)
454 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
455 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
456 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
459 * Set the whole_disk property. If it's not specified, leave the value
462 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
463 &vd
->vdev_wholedisk
) != 0)
464 vd
->vdev_wholedisk
= -1ULL;
467 * Look for the 'not present' flag. This will only be set if the device
468 * was not present at the time of import.
470 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
471 &vd
->vdev_not_present
);
474 * Get the alignment requirement.
476 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
479 * Retrieve the vdev creation time.
481 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
485 * If we're a top-level vdev, try to load the allocation parameters.
487 if (parent
&& !parent
->vdev_parent
&&
488 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
489 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
491 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
493 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
495 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
499 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
500 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
501 alloctype
== VDEV_ALLOC_ADD
||
502 alloctype
== VDEV_ALLOC_SPLIT
||
503 alloctype
== VDEV_ALLOC_ROOTPOOL
);
504 vd
->vdev_mg
= metaslab_group_create(islog
?
505 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
509 * If we're a leaf vdev, try to load the DTL object and other state.
511 if (vd
->vdev_ops
->vdev_op_leaf
&&
512 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
513 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
514 if (alloctype
== VDEV_ALLOC_LOAD
) {
515 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
516 &vd
->vdev_dtl_smo
.smo_object
);
517 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
521 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
524 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
525 &spare
) == 0 && spare
)
529 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
532 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVERING
,
533 &vd
->vdev_resilvering
);
536 * When importing a pool, we want to ignore the persistent fault
537 * state, as the diagnosis made on another system may not be
538 * valid in the current context. Local vdevs will
539 * remain in the faulted state.
541 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
542 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
544 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
546 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
549 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
553 VDEV_AUX_ERR_EXCEEDED
;
554 if (nvlist_lookup_string(nv
,
555 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
556 strcmp(aux
, "external") == 0)
557 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
563 * Add ourselves to the parent's list of children.
565 vdev_add_child(parent
, vd
);
573 vdev_free(vdev_t
*vd
)
576 spa_t
*spa
= vd
->vdev_spa
;
579 * vdev_free() implies closing the vdev first. This is simpler than
580 * trying to ensure complicated semantics for all callers.
584 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
585 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
590 for (c
= 0; c
< vd
->vdev_children
; c
++)
591 vdev_free(vd
->vdev_child
[c
]);
593 ASSERT(vd
->vdev_child
== NULL
);
594 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
597 * Discard allocation state.
599 if (vd
->vdev_mg
!= NULL
) {
600 vdev_metaslab_fini(vd
);
601 metaslab_group_destroy(vd
->vdev_mg
);
604 ASSERT3U(vd
->vdev_stat
.vs_space
, ==, 0);
605 ASSERT3U(vd
->vdev_stat
.vs_dspace
, ==, 0);
606 ASSERT3U(vd
->vdev_stat
.vs_alloc
, ==, 0);
609 * Remove this vdev from its parent's child list.
611 vdev_remove_child(vd
->vdev_parent
, vd
);
613 ASSERT(vd
->vdev_parent
== NULL
);
616 * Clean up vdev structure.
622 spa_strfree(vd
->vdev_path
);
624 spa_strfree(vd
->vdev_devid
);
625 if (vd
->vdev_physpath
)
626 spa_strfree(vd
->vdev_physpath
);
628 spa_strfree(vd
->vdev_fru
);
630 if (vd
->vdev_isspare
)
631 spa_spare_remove(vd
);
632 if (vd
->vdev_isl2cache
)
633 spa_l2cache_remove(vd
);
635 txg_list_destroy(&vd
->vdev_ms_list
);
636 txg_list_destroy(&vd
->vdev_dtl_list
);
638 mutex_enter(&vd
->vdev_dtl_lock
);
639 for (t
= 0; t
< DTL_TYPES
; t
++) {
640 space_map_unload(&vd
->vdev_dtl
[t
]);
641 space_map_destroy(&vd
->vdev_dtl
[t
]);
643 mutex_exit(&vd
->vdev_dtl_lock
);
645 mutex_destroy(&vd
->vdev_dtl_lock
);
646 mutex_destroy(&vd
->vdev_stat_lock
);
647 mutex_destroy(&vd
->vdev_probe_lock
);
649 if (vd
== spa
->spa_root_vdev
)
650 spa
->spa_root_vdev
= NULL
;
652 kmem_free(vd
, sizeof (vdev_t
));
656 * Transfer top-level vdev state from svd to tvd.
659 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
661 spa_t
*spa
= svd
->vdev_spa
;
666 ASSERT(tvd
== tvd
->vdev_top
);
668 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
669 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
670 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
672 svd
->vdev_ms_array
= 0;
673 svd
->vdev_ms_shift
= 0;
674 svd
->vdev_ms_count
= 0;
677 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
678 tvd
->vdev_mg
= svd
->vdev_mg
;
679 tvd
->vdev_ms
= svd
->vdev_ms
;
684 if (tvd
->vdev_mg
!= NULL
)
685 tvd
->vdev_mg
->mg_vd
= tvd
;
687 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
688 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
689 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
691 svd
->vdev_stat
.vs_alloc
= 0;
692 svd
->vdev_stat
.vs_space
= 0;
693 svd
->vdev_stat
.vs_dspace
= 0;
695 for (t
= 0; t
< TXG_SIZE
; t
++) {
696 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
697 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
698 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
699 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
700 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
701 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
704 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
705 vdev_config_clean(svd
);
706 vdev_config_dirty(tvd
);
709 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
710 vdev_state_clean(svd
);
711 vdev_state_dirty(tvd
);
714 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
715 svd
->vdev_deflate_ratio
= 0;
717 tvd
->vdev_islog
= svd
->vdev_islog
;
722 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
731 for (c
= 0; c
< vd
->vdev_children
; c
++)
732 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
736 * Add a mirror/replacing vdev above an existing vdev.
739 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
741 spa_t
*spa
= cvd
->vdev_spa
;
742 vdev_t
*pvd
= cvd
->vdev_parent
;
745 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
747 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
749 mvd
->vdev_asize
= cvd
->vdev_asize
;
750 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
751 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
752 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
753 mvd
->vdev_state
= cvd
->vdev_state
;
754 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
756 vdev_remove_child(pvd
, cvd
);
757 vdev_add_child(pvd
, mvd
);
758 cvd
->vdev_id
= mvd
->vdev_children
;
759 vdev_add_child(mvd
, cvd
);
760 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
762 if (mvd
== mvd
->vdev_top
)
763 vdev_top_transfer(cvd
, mvd
);
769 * Remove a 1-way mirror/replacing vdev from the tree.
772 vdev_remove_parent(vdev_t
*cvd
)
774 vdev_t
*mvd
= cvd
->vdev_parent
;
775 vdev_t
*pvd
= mvd
->vdev_parent
;
777 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
779 ASSERT(mvd
->vdev_children
== 1);
780 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
781 mvd
->vdev_ops
== &vdev_replacing_ops
||
782 mvd
->vdev_ops
== &vdev_spare_ops
);
783 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
785 vdev_remove_child(mvd
, cvd
);
786 vdev_remove_child(pvd
, mvd
);
789 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
790 * Otherwise, we could have detached an offline device, and when we
791 * go to import the pool we'll think we have two top-level vdevs,
792 * instead of a different version of the same top-level vdev.
794 if (mvd
->vdev_top
== mvd
) {
795 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
796 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
797 cvd
->vdev_guid
+= guid_delta
;
798 cvd
->vdev_guid_sum
+= guid_delta
;
800 cvd
->vdev_id
= mvd
->vdev_id
;
801 vdev_add_child(pvd
, cvd
);
802 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
804 if (cvd
== cvd
->vdev_top
)
805 vdev_top_transfer(mvd
, cvd
);
807 ASSERT(mvd
->vdev_children
== 0);
812 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
814 spa_t
*spa
= vd
->vdev_spa
;
815 objset_t
*mos
= spa
->spa_meta_objset
;
817 uint64_t oldc
= vd
->vdev_ms_count
;
818 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
822 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
825 * This vdev is not being allocated from yet or is a hole.
827 if (vd
->vdev_ms_shift
== 0)
830 ASSERT(!vd
->vdev_ishole
);
833 * Compute the raidz-deflation ratio. Note, we hard-code
834 * in 128k (1 << 17) because it is the current "typical" blocksize.
835 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
836 * or we will inconsistently account for existing bp's.
838 vd
->vdev_deflate_ratio
= (1 << 17) /
839 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
841 ASSERT(oldc
<= newc
);
843 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_PUSHPAGE
| KM_NODEBUG
);
846 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
847 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
851 vd
->vdev_ms_count
= newc
;
853 for (m
= oldc
; m
< newc
; m
++) {
854 space_map_obj_t smo
= { 0, 0, 0 };
857 error
= dmu_read(mos
, vd
->vdev_ms_array
,
858 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
864 error
= dmu_bonus_hold(mos
, object
, FTAG
, &db
);
867 ASSERT3U(db
->db_size
, >=, sizeof (smo
));
868 bcopy(db
->db_data
, &smo
, sizeof (smo
));
869 ASSERT3U(smo
.smo_object
, ==, object
);
870 dmu_buf_rele(db
, FTAG
);
873 vd
->vdev_ms
[m
] = metaslab_init(vd
->vdev_mg
, &smo
,
874 m
<< vd
->vdev_ms_shift
, 1ULL << vd
->vdev_ms_shift
, txg
);
878 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
881 * If the vdev is being removed we don't activate
882 * the metaslabs since we want to ensure that no new
883 * allocations are performed on this device.
885 if (oldc
== 0 && !vd
->vdev_removing
)
886 metaslab_group_activate(vd
->vdev_mg
);
889 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
895 vdev_metaslab_fini(vdev_t
*vd
)
898 uint64_t count
= vd
->vdev_ms_count
;
900 if (vd
->vdev_ms
!= NULL
) {
901 metaslab_group_passivate(vd
->vdev_mg
);
902 for (m
= 0; m
< count
; m
++)
903 if (vd
->vdev_ms
[m
] != NULL
)
904 metaslab_fini(vd
->vdev_ms
[m
]);
905 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
909 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
912 typedef struct vdev_probe_stats
{
913 boolean_t vps_readable
;
914 boolean_t vps_writeable
;
916 } vdev_probe_stats_t
;
919 vdev_probe_done(zio_t
*zio
)
921 spa_t
*spa
= zio
->io_spa
;
922 vdev_t
*vd
= zio
->io_vd
;
923 vdev_probe_stats_t
*vps
= zio
->io_private
;
925 ASSERT(vd
->vdev_probe_zio
!= NULL
);
927 if (zio
->io_type
== ZIO_TYPE_READ
) {
928 if (zio
->io_error
== 0)
929 vps
->vps_readable
= 1;
930 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
931 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
932 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
933 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
934 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
936 zio_buf_free(zio
->io_data
, zio
->io_size
);
938 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
939 if (zio
->io_error
== 0)
940 vps
->vps_writeable
= 1;
941 zio_buf_free(zio
->io_data
, zio
->io_size
);
942 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
945 vd
->vdev_cant_read
|= !vps
->vps_readable
;
946 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
948 if (vdev_readable(vd
) &&
949 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
952 ASSERT(zio
->io_error
!= 0);
953 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
954 spa
, vd
, NULL
, 0, 0);
955 zio
->io_error
= ENXIO
;
958 mutex_enter(&vd
->vdev_probe_lock
);
959 ASSERT(vd
->vdev_probe_zio
== zio
);
960 vd
->vdev_probe_zio
= NULL
;
961 mutex_exit(&vd
->vdev_probe_lock
);
963 while ((pio
= zio_walk_parents(zio
)) != NULL
)
964 if (!vdev_accessible(vd
, pio
))
965 pio
->io_error
= ENXIO
;
967 kmem_free(vps
, sizeof (*vps
));
972 * Determine whether this device is accessible by reading and writing
973 * to several known locations: the pad regions of each vdev label
974 * but the first (which we leave alone in case it contains a VTOC).
977 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
979 spa_t
*spa
= vd
->vdev_spa
;
980 vdev_probe_stats_t
*vps
= NULL
;
984 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
987 * Don't probe the probe.
989 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
993 * To prevent 'probe storms' when a device fails, we create
994 * just one probe i/o at a time. All zios that want to probe
995 * this vdev will become parents of the probe io.
997 mutex_enter(&vd
->vdev_probe_lock
);
999 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1000 vps
= kmem_zalloc(sizeof (*vps
), KM_PUSHPAGE
);
1002 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1003 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1006 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1008 * vdev_cant_read and vdev_cant_write can only
1009 * transition from TRUE to FALSE when we have the
1010 * SCL_ZIO lock as writer; otherwise they can only
1011 * transition from FALSE to TRUE. This ensures that
1012 * any zio looking at these values can assume that
1013 * failures persist for the life of the I/O. That's
1014 * important because when a device has intermittent
1015 * connectivity problems, we want to ensure that
1016 * they're ascribed to the device (ENXIO) and not
1019 * Since we hold SCL_ZIO as writer here, clear both
1020 * values so the probe can reevaluate from first
1023 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1024 vd
->vdev_cant_read
= B_FALSE
;
1025 vd
->vdev_cant_write
= B_FALSE
;
1028 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1029 vdev_probe_done
, vps
,
1030 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1033 * We can't change the vdev state in this context, so we
1034 * kick off an async task to do it on our behalf.
1037 vd
->vdev_probe_wanted
= B_TRUE
;
1038 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1043 zio_add_child(zio
, pio
);
1045 mutex_exit(&vd
->vdev_probe_lock
);
1048 ASSERT(zio
!= NULL
);
1052 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1053 zio_nowait(zio_read_phys(pio
, vd
,
1054 vdev_label_offset(vd
->vdev_psize
, l
,
1055 offsetof(vdev_label_t
, vl_pad2
)),
1056 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1057 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1058 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1069 vdev_open_child(void *arg
)
1073 vd
->vdev_open_thread
= curthread
;
1074 vd
->vdev_open_error
= vdev_open(vd
);
1075 vd
->vdev_open_thread
= NULL
;
1079 vdev_uses_zvols(vdev_t
*vd
)
1084 if (zvol_is_zvol(vd
->vdev_path
))
1088 for (c
= 0; c
< vd
->vdev_children
; c
++)
1089 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1096 vdev_open_children(vdev_t
*vd
)
1099 int children
= vd
->vdev_children
;
1103 * in order to handle pools on top of zvols, do the opens
1104 * in a single thread so that the same thread holds the
1105 * spa_namespace_lock
1107 if (vdev_uses_zvols(vd
)) {
1108 for (c
= 0; c
< children
; c
++)
1109 vd
->vdev_child
[c
]->vdev_open_error
=
1110 vdev_open(vd
->vdev_child
[c
]);
1113 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1114 children
, children
, TASKQ_PREPOPULATE
);
1116 for (c
= 0; c
< children
; c
++)
1117 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1124 * Prepare a virtual device for access.
1127 vdev_open(vdev_t
*vd
)
1129 spa_t
*spa
= vd
->vdev_spa
;
1132 uint64_t max_osize
= 0;
1133 uint64_t asize
, max_asize
, psize
;
1134 uint64_t ashift
= 0;
1137 ASSERT(vd
->vdev_open_thread
== curthread
||
1138 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1139 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1140 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1141 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1143 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1144 vd
->vdev_cant_read
= B_FALSE
;
1145 vd
->vdev_cant_write
= B_FALSE
;
1146 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1149 * If this vdev is not removed, check its fault status. If it's
1150 * faulted, bail out of the open.
1152 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1153 ASSERT(vd
->vdev_children
== 0);
1154 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1155 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1156 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1157 vd
->vdev_label_aux
);
1159 } else if (vd
->vdev_offline
) {
1160 ASSERT(vd
->vdev_children
== 0);
1161 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1165 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1168 * Reset the vdev_reopening flag so that we actually close
1169 * the vdev on error.
1171 vd
->vdev_reopening
= B_FALSE
;
1172 if (zio_injection_enabled
&& error
== 0)
1173 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1176 if (vd
->vdev_removed
&&
1177 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1178 vd
->vdev_removed
= B_FALSE
;
1180 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1181 vd
->vdev_stat
.vs_aux
);
1185 vd
->vdev_removed
= B_FALSE
;
1188 * Recheck the faulted flag now that we have confirmed that
1189 * the vdev is accessible. If we're faulted, bail.
1191 if (vd
->vdev_faulted
) {
1192 ASSERT(vd
->vdev_children
== 0);
1193 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1194 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1195 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1196 vd
->vdev_label_aux
);
1200 if (vd
->vdev_degraded
) {
1201 ASSERT(vd
->vdev_children
== 0);
1202 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1203 VDEV_AUX_ERR_EXCEEDED
);
1205 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1209 * For hole or missing vdevs we just return success.
1211 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1214 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1215 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1216 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1222 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1223 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1225 if (vd
->vdev_children
== 0) {
1226 if (osize
< SPA_MINDEVSIZE
) {
1227 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1228 VDEV_AUX_TOO_SMALL
);
1232 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1233 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1234 VDEV_LABEL_END_SIZE
);
1236 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1237 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1238 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1239 VDEV_AUX_TOO_SMALL
);
1244 max_asize
= max_osize
;
1247 vd
->vdev_psize
= psize
;
1250 * Make sure the allocatable size hasn't shrunk.
1252 if (asize
< vd
->vdev_min_asize
) {
1253 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1254 VDEV_AUX_BAD_LABEL
);
1258 if (vd
->vdev_asize
== 0) {
1260 * This is the first-ever open, so use the computed values.
1261 * For compatibility, a different ashift can be requested.
1263 vd
->vdev_asize
= asize
;
1264 vd
->vdev_max_asize
= max_asize
;
1265 if (vd
->vdev_ashift
== 0)
1266 vd
->vdev_ashift
= ashift
;
1269 * Detect if the alignment requirement has increased.
1270 * We don't want to make the pool unavailable, just
1271 * post an event instead.
1273 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1274 vd
->vdev_ops
->vdev_op_leaf
) {
1275 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1276 spa
, vd
, NULL
, 0, 0);
1279 vd
->vdev_max_asize
= max_asize
;
1283 * If all children are healthy and the asize has increased,
1284 * then we've experienced dynamic LUN growth. If automatic
1285 * expansion is enabled then use the additional space.
1287 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1288 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1289 vd
->vdev_asize
= asize
;
1291 vdev_set_min_asize(vd
);
1294 * Ensure we can issue some IO before declaring the
1295 * vdev open for business.
1297 if (vd
->vdev_ops
->vdev_op_leaf
&&
1298 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1299 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1300 VDEV_AUX_ERR_EXCEEDED
);
1305 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1306 * resilver. But don't do this if we are doing a reopen for a scrub,
1307 * since this would just restart the scrub we are already doing.
1309 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1310 vdev_resilver_needed(vd
, NULL
, NULL
))
1311 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1317 * Called once the vdevs are all opened, this routine validates the label
1318 * contents. This needs to be done before vdev_load() so that we don't
1319 * inadvertently do repair I/Os to the wrong device.
1321 * If 'strict' is false ignore the spa guid check. This is necessary because
1322 * if the machine crashed during a re-guid the new guid might have been written
1323 * to all of the vdev labels, but not the cached config. The strict check
1324 * will be performed when the pool is opened again using the mos config.
1326 * This function will only return failure if one of the vdevs indicates that it
1327 * has since been destroyed or exported. This is only possible if
1328 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1329 * will be updated but the function will return 0.
1332 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1334 spa_t
*spa
= vd
->vdev_spa
;
1336 uint64_t guid
= 0, top_guid
;
1340 for (c
= 0; c
< vd
->vdev_children
; c
++)
1341 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1345 * If the device has already failed, or was marked offline, don't do
1346 * any further validation. Otherwise, label I/O will fail and we will
1347 * overwrite the previous state.
1349 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1350 uint64_t aux_guid
= 0;
1352 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1353 spa_last_synced_txg(spa
) : -1ULL;
1355 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1356 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1357 VDEV_AUX_BAD_LABEL
);
1362 * Determine if this vdev has been split off into another
1363 * pool. If so, then refuse to open it.
1365 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1366 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1367 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1368 VDEV_AUX_SPLIT_POOL
);
1373 if (strict
&& (nvlist_lookup_uint64(label
,
1374 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1375 guid
!= spa_guid(spa
))) {
1376 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1377 VDEV_AUX_CORRUPT_DATA
);
1382 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1383 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1388 * If this vdev just became a top-level vdev because its
1389 * sibling was detached, it will have adopted the parent's
1390 * vdev guid -- but the label may or may not be on disk yet.
1391 * Fortunately, either version of the label will have the
1392 * same top guid, so if we're a top-level vdev, we can
1393 * safely compare to that instead.
1395 * If we split this vdev off instead, then we also check the
1396 * original pool's guid. We don't want to consider the vdev
1397 * corrupt if it is partway through a split operation.
1399 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1401 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1403 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1404 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1405 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1406 VDEV_AUX_CORRUPT_DATA
);
1411 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1413 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1414 VDEV_AUX_CORRUPT_DATA
);
1422 * If this is a verbatim import, no need to check the
1423 * state of the pool.
1425 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1426 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1427 state
!= POOL_STATE_ACTIVE
)
1431 * If we were able to open and validate a vdev that was
1432 * previously marked permanently unavailable, clear that state
1435 if (vd
->vdev_not_present
)
1436 vd
->vdev_not_present
= 0;
1443 * Close a virtual device.
1446 vdev_close(vdev_t
*vd
)
1448 vdev_t
*pvd
= vd
->vdev_parent
;
1449 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1451 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1454 * If our parent is reopening, then we are as well, unless we are
1457 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1458 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1460 vd
->vdev_ops
->vdev_op_close(vd
);
1462 vdev_cache_purge(vd
);
1465 * We record the previous state before we close it, so that if we are
1466 * doing a reopen(), we don't generate FMA ereports if we notice that
1467 * it's still faulted.
1469 vd
->vdev_prevstate
= vd
->vdev_state
;
1471 if (vd
->vdev_offline
)
1472 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1474 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1475 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1479 vdev_hold(vdev_t
*vd
)
1481 spa_t
*spa
= vd
->vdev_spa
;
1484 ASSERT(spa_is_root(spa
));
1485 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1488 for (c
= 0; c
< vd
->vdev_children
; c
++)
1489 vdev_hold(vd
->vdev_child
[c
]);
1491 if (vd
->vdev_ops
->vdev_op_leaf
)
1492 vd
->vdev_ops
->vdev_op_hold(vd
);
1496 vdev_rele(vdev_t
*vd
)
1500 ASSERT(spa_is_root(vd
->vdev_spa
));
1501 for (c
= 0; c
< vd
->vdev_children
; c
++)
1502 vdev_rele(vd
->vdev_child
[c
]);
1504 if (vd
->vdev_ops
->vdev_op_leaf
)
1505 vd
->vdev_ops
->vdev_op_rele(vd
);
1509 * Reopen all interior vdevs and any unopened leaves. We don't actually
1510 * reopen leaf vdevs which had previously been opened as they might deadlock
1511 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1512 * If the leaf has never been opened then open it, as usual.
1515 vdev_reopen(vdev_t
*vd
)
1517 spa_t
*spa
= vd
->vdev_spa
;
1519 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1521 /* set the reopening flag unless we're taking the vdev offline */
1522 vd
->vdev_reopening
= !vd
->vdev_offline
;
1524 (void) vdev_open(vd
);
1527 * Call vdev_validate() here to make sure we have the same device.
1528 * Otherwise, a device with an invalid label could be successfully
1529 * opened in response to vdev_reopen().
1532 (void) vdev_validate_aux(vd
);
1533 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1534 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1535 !l2arc_vdev_present(vd
))
1536 l2arc_add_vdev(spa
, vd
);
1538 (void) vdev_validate(vd
, B_TRUE
);
1542 * Reassess parent vdev's health.
1544 vdev_propagate_state(vd
);
1548 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1553 * Normally, partial opens (e.g. of a mirror) are allowed.
1554 * For a create, however, we want to fail the request if
1555 * there are any components we can't open.
1557 error
= vdev_open(vd
);
1559 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1561 return (error
? error
: ENXIO
);
1565 * Recursively initialize all labels.
1567 if ((error
= vdev_label_init(vd
, txg
, isreplacing
?
1568 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1577 vdev_metaslab_set_size(vdev_t
*vd
)
1580 * Aim for roughly 200 metaslabs per vdev.
1582 vd
->vdev_ms_shift
= highbit(vd
->vdev_asize
/ 200);
1583 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1587 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1589 ASSERT(vd
== vd
->vdev_top
);
1590 ASSERT(!vd
->vdev_ishole
);
1591 ASSERT(ISP2(flags
));
1592 ASSERT(spa_writeable(vd
->vdev_spa
));
1594 if (flags
& VDD_METASLAB
)
1595 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1597 if (flags
& VDD_DTL
)
1598 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1600 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1606 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1607 * the vdev has less than perfect replication. There are four kinds of DTL:
1609 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1611 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1613 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1614 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1615 * txgs that was scrubbed.
1617 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1618 * persistent errors or just some device being offline.
1619 * Unlike the other three, the DTL_OUTAGE map is not generally
1620 * maintained; it's only computed when needed, typically to
1621 * determine whether a device can be detached.
1623 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1624 * either has the data or it doesn't.
1626 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1627 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1628 * if any child is less than fully replicated, then so is its parent.
1629 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1630 * comprising only those txgs which appear in 'maxfaults' or more children;
1631 * those are the txgs we don't have enough replication to read. For example,
1632 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1633 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1634 * two child DTL_MISSING maps.
1636 * It should be clear from the above that to compute the DTLs and outage maps
1637 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1638 * Therefore, that is all we keep on disk. When loading the pool, or after
1639 * a configuration change, we generate all other DTLs from first principles.
1642 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1644 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1646 ASSERT(t
< DTL_TYPES
);
1647 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1648 ASSERT(spa_writeable(vd
->vdev_spa
));
1650 mutex_enter(sm
->sm_lock
);
1651 if (!space_map_contains(sm
, txg
, size
))
1652 space_map_add(sm
, txg
, size
);
1653 mutex_exit(sm
->sm_lock
);
1657 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1659 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1660 boolean_t dirty
= B_FALSE
;
1662 ASSERT(t
< DTL_TYPES
);
1663 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1665 mutex_enter(sm
->sm_lock
);
1666 if (sm
->sm_space
!= 0)
1667 dirty
= space_map_contains(sm
, txg
, size
);
1668 mutex_exit(sm
->sm_lock
);
1674 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1676 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1679 mutex_enter(sm
->sm_lock
);
1680 empty
= (sm
->sm_space
== 0);
1681 mutex_exit(sm
->sm_lock
);
1687 * Reassess DTLs after a config change or scrub completion.
1690 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1692 spa_t
*spa
= vd
->vdev_spa
;
1696 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1698 for (c
= 0; c
< vd
->vdev_children
; c
++)
1699 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1700 scrub_txg
, scrub_done
);
1702 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1705 if (vd
->vdev_ops
->vdev_op_leaf
) {
1706 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1708 mutex_enter(&vd
->vdev_dtl_lock
);
1709 if (scrub_txg
!= 0 &&
1710 (spa
->spa_scrub_started
||
1711 (scn
&& scn
->scn_phys
.scn_errors
== 0))) {
1713 * We completed a scrub up to scrub_txg. If we
1714 * did it without rebooting, then the scrub dtl
1715 * will be valid, so excise the old region and
1716 * fold in the scrub dtl. Otherwise, leave the
1717 * dtl as-is if there was an error.
1719 * There's little trick here: to excise the beginning
1720 * of the DTL_MISSING map, we put it into a reference
1721 * tree and then add a segment with refcnt -1 that
1722 * covers the range [0, scrub_txg). This means
1723 * that each txg in that range has refcnt -1 or 0.
1724 * We then add DTL_SCRUB with a refcnt of 2, so that
1725 * entries in the range [0, scrub_txg) will have a
1726 * positive refcnt -- either 1 or 2. We then convert
1727 * the reference tree into the new DTL_MISSING map.
1729 space_map_ref_create(&reftree
);
1730 space_map_ref_add_map(&reftree
,
1731 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1732 space_map_ref_add_seg(&reftree
, 0, scrub_txg
, -1);
1733 space_map_ref_add_map(&reftree
,
1734 &vd
->vdev_dtl
[DTL_SCRUB
], 2);
1735 space_map_ref_generate_map(&reftree
,
1736 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1737 space_map_ref_destroy(&reftree
);
1739 space_map_vacate(&vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1740 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1741 space_map_add
, &vd
->vdev_dtl
[DTL_PARTIAL
]);
1743 space_map_vacate(&vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1744 space_map_vacate(&vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1745 if (!vdev_readable(vd
))
1746 space_map_add(&vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1748 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1749 space_map_add
, &vd
->vdev_dtl
[DTL_OUTAGE
]);
1750 mutex_exit(&vd
->vdev_dtl_lock
);
1753 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1757 mutex_enter(&vd
->vdev_dtl_lock
);
1758 for (t
= 0; t
< DTL_TYPES
; t
++) {
1759 /* account for child's outage in parent's missing map */
1760 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1762 continue; /* leaf vdevs only */
1763 if (t
== DTL_PARTIAL
)
1764 minref
= 1; /* i.e. non-zero */
1765 else if (vd
->vdev_nparity
!= 0)
1766 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1768 minref
= vd
->vdev_children
; /* any kind of mirror */
1769 space_map_ref_create(&reftree
);
1770 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1771 vdev_t
*cvd
= vd
->vdev_child
[c
];
1772 mutex_enter(&cvd
->vdev_dtl_lock
);
1773 space_map_ref_add_map(&reftree
, &cvd
->vdev_dtl
[s
], 1);
1774 mutex_exit(&cvd
->vdev_dtl_lock
);
1776 space_map_ref_generate_map(&reftree
, &vd
->vdev_dtl
[t
], minref
);
1777 space_map_ref_destroy(&reftree
);
1779 mutex_exit(&vd
->vdev_dtl_lock
);
1783 vdev_dtl_load(vdev_t
*vd
)
1785 spa_t
*spa
= vd
->vdev_spa
;
1786 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1787 objset_t
*mos
= spa
->spa_meta_objset
;
1791 ASSERT(vd
->vdev_children
== 0);
1793 if (smo
->smo_object
== 0)
1796 ASSERT(!vd
->vdev_ishole
);
1798 if ((error
= dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
)) != 0)
1801 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1802 bcopy(db
->db_data
, smo
, sizeof (*smo
));
1803 dmu_buf_rele(db
, FTAG
);
1805 mutex_enter(&vd
->vdev_dtl_lock
);
1806 error
= space_map_load(&vd
->vdev_dtl
[DTL_MISSING
],
1807 NULL
, SM_ALLOC
, smo
, mos
);
1808 mutex_exit(&vd
->vdev_dtl_lock
);
1814 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1816 spa_t
*spa
= vd
->vdev_spa
;
1817 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1818 space_map_t
*sm
= &vd
->vdev_dtl
[DTL_MISSING
];
1819 objset_t
*mos
= spa
->spa_meta_objset
;
1825 ASSERT(!vd
->vdev_ishole
);
1827 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1829 if (vd
->vdev_detached
) {
1830 if (smo
->smo_object
!= 0) {
1831 VERIFY(0 == dmu_object_free(mos
, smo
->smo_object
, tx
));
1832 smo
->smo_object
= 0;
1838 if (smo
->smo_object
== 0) {
1839 ASSERT(smo
->smo_objsize
== 0);
1840 ASSERT(smo
->smo_alloc
== 0);
1841 smo
->smo_object
= dmu_object_alloc(mos
,
1842 DMU_OT_SPACE_MAP
, 1 << SPACE_MAP_BLOCKSHIFT
,
1843 DMU_OT_SPACE_MAP_HEADER
, sizeof (*smo
), tx
);
1844 ASSERT(smo
->smo_object
!= 0);
1845 vdev_config_dirty(vd
->vdev_top
);
1848 mutex_init(&smlock
, NULL
, MUTEX_DEFAULT
, NULL
);
1850 space_map_create(&smsync
, sm
->sm_start
, sm
->sm_size
, sm
->sm_shift
,
1853 mutex_enter(&smlock
);
1855 mutex_enter(&vd
->vdev_dtl_lock
);
1856 space_map_walk(sm
, space_map_add
, &smsync
);
1857 mutex_exit(&vd
->vdev_dtl_lock
);
1859 space_map_truncate(smo
, mos
, tx
);
1860 space_map_sync(&smsync
, SM_ALLOC
, smo
, mos
, tx
);
1862 space_map_destroy(&smsync
);
1864 mutex_exit(&smlock
);
1865 mutex_destroy(&smlock
);
1867 VERIFY(0 == dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
));
1868 dmu_buf_will_dirty(db
, tx
);
1869 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1870 bcopy(smo
, db
->db_data
, sizeof (*smo
));
1871 dmu_buf_rele(db
, FTAG
);
1877 * Determine whether the specified vdev can be offlined/detached/removed
1878 * without losing data.
1881 vdev_dtl_required(vdev_t
*vd
)
1883 spa_t
*spa
= vd
->vdev_spa
;
1884 vdev_t
*tvd
= vd
->vdev_top
;
1885 uint8_t cant_read
= vd
->vdev_cant_read
;
1888 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1890 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
1894 * Temporarily mark the device as unreadable, and then determine
1895 * whether this results in any DTL outages in the top-level vdev.
1896 * If not, we can safely offline/detach/remove the device.
1898 vd
->vdev_cant_read
= B_TRUE
;
1899 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1900 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
1901 vd
->vdev_cant_read
= cant_read
;
1902 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1904 if (!required
&& zio_injection_enabled
)
1905 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
1911 * Determine if resilver is needed, and if so the txg range.
1914 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
1916 boolean_t needed
= B_FALSE
;
1917 uint64_t thismin
= UINT64_MAX
;
1918 uint64_t thismax
= 0;
1921 if (vd
->vdev_children
== 0) {
1922 mutex_enter(&vd
->vdev_dtl_lock
);
1923 if (vd
->vdev_dtl
[DTL_MISSING
].sm_space
!= 0 &&
1924 vdev_writeable(vd
)) {
1927 ss
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1928 thismin
= ss
->ss_start
- 1;
1929 ss
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1930 thismax
= ss
->ss_end
;
1933 mutex_exit(&vd
->vdev_dtl_lock
);
1935 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1936 vdev_t
*cvd
= vd
->vdev_child
[c
];
1937 uint64_t cmin
, cmax
;
1939 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
1940 thismin
= MIN(thismin
, cmin
);
1941 thismax
= MAX(thismax
, cmax
);
1947 if (needed
&& minp
) {
1955 vdev_load(vdev_t
*vd
)
1960 * Recursively load all children.
1962 for (c
= 0; c
< vd
->vdev_children
; c
++)
1963 vdev_load(vd
->vdev_child
[c
]);
1966 * If this is a top-level vdev, initialize its metaslabs.
1968 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
1969 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
1970 vdev_metaslab_init(vd
, 0) != 0))
1971 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1972 VDEV_AUX_CORRUPT_DATA
);
1975 * If this is a leaf vdev, load its DTL.
1977 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
1978 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1979 VDEV_AUX_CORRUPT_DATA
);
1983 * The special vdev case is used for hot spares and l2cache devices. Its
1984 * sole purpose it to set the vdev state for the associated vdev. To do this,
1985 * we make sure that we can open the underlying device, then try to read the
1986 * label, and make sure that the label is sane and that it hasn't been
1987 * repurposed to another pool.
1990 vdev_validate_aux(vdev_t
*vd
)
1993 uint64_t guid
, version
;
1996 if (!vdev_readable(vd
))
1999 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2000 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2001 VDEV_AUX_CORRUPT_DATA
);
2005 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2006 !SPA_VERSION_IS_SUPPORTED(version
) ||
2007 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2008 guid
!= vd
->vdev_guid
||
2009 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2010 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2011 VDEV_AUX_CORRUPT_DATA
);
2017 * We don't actually check the pool state here. If it's in fact in
2018 * use by another pool, we update this fact on the fly when requested.
2025 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2027 spa_t
*spa
= vd
->vdev_spa
;
2028 objset_t
*mos
= spa
->spa_meta_objset
;
2032 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2034 if (vd
->vdev_dtl_smo
.smo_object
) {
2035 ASSERT3U(vd
->vdev_dtl_smo
.smo_alloc
, ==, 0);
2036 (void) dmu_object_free(mos
, vd
->vdev_dtl_smo
.smo_object
, tx
);
2037 vd
->vdev_dtl_smo
.smo_object
= 0;
2040 if (vd
->vdev_ms
!= NULL
) {
2041 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2042 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2044 if (msp
== NULL
|| msp
->ms_smo
.smo_object
== 0)
2047 ASSERT3U(msp
->ms_smo
.smo_alloc
, ==, 0);
2048 (void) dmu_object_free(mos
, msp
->ms_smo
.smo_object
, tx
);
2049 msp
->ms_smo
.smo_object
= 0;
2053 if (vd
->vdev_ms_array
) {
2054 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2055 vd
->vdev_ms_array
= 0;
2056 vd
->vdev_ms_shift
= 0;
2062 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2065 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2067 ASSERT(!vd
->vdev_ishole
);
2069 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2070 metaslab_sync_done(msp
, txg
);
2073 metaslab_sync_reassess(vd
->vdev_mg
);
2077 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2079 spa_t
*spa
= vd
->vdev_spa
;
2084 ASSERT(!vd
->vdev_ishole
);
2086 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2087 ASSERT(vd
== vd
->vdev_top
);
2088 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2089 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2090 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2091 ASSERT(vd
->vdev_ms_array
!= 0);
2092 vdev_config_dirty(vd
);
2097 * Remove the metadata associated with this vdev once it's empty.
2099 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2100 vdev_remove(vd
, txg
);
2102 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2103 metaslab_sync(msp
, txg
);
2104 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2107 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2108 vdev_dtl_sync(lvd
, txg
);
2110 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2114 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2116 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2120 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2121 * not be opened, and no I/O is attempted.
2124 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2128 spa_vdev_state_enter(spa
, SCL_NONE
);
2130 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2131 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2133 if (!vd
->vdev_ops
->vdev_op_leaf
)
2134 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2139 * We don't directly use the aux state here, but if we do a
2140 * vdev_reopen(), we need this value to be present to remember why we
2143 vd
->vdev_label_aux
= aux
;
2146 * Faulted state takes precedence over degraded.
2148 vd
->vdev_delayed_close
= B_FALSE
;
2149 vd
->vdev_faulted
= 1ULL;
2150 vd
->vdev_degraded
= 0ULL;
2151 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2154 * If this device has the only valid copy of the data, then
2155 * back off and simply mark the vdev as degraded instead.
2157 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2158 vd
->vdev_degraded
= 1ULL;
2159 vd
->vdev_faulted
= 0ULL;
2162 * If we reopen the device and it's not dead, only then do we
2167 if (vdev_readable(vd
))
2168 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2171 return (spa_vdev_state_exit(spa
, vd
, 0));
2175 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2176 * user that something is wrong. The vdev continues to operate as normal as far
2177 * as I/O is concerned.
2180 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2184 spa_vdev_state_enter(spa
, SCL_NONE
);
2186 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2187 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2189 if (!vd
->vdev_ops
->vdev_op_leaf
)
2190 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2193 * If the vdev is already faulted, then don't do anything.
2195 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2196 return (spa_vdev_state_exit(spa
, NULL
, 0));
2198 vd
->vdev_degraded
= 1ULL;
2199 if (!vdev_is_dead(vd
))
2200 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2203 return (spa_vdev_state_exit(spa
, vd
, 0));
2207 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2208 * any attached spare device should be detached when the device finishes
2209 * resilvering. Second, the online should be treated like a 'test' online case,
2210 * so no FMA events are generated if the device fails to open.
2213 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2215 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2217 spa_vdev_state_enter(spa
, SCL_NONE
);
2219 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2220 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2222 if (!vd
->vdev_ops
->vdev_op_leaf
)
2223 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2226 vd
->vdev_offline
= B_FALSE
;
2227 vd
->vdev_tmpoffline
= B_FALSE
;
2228 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2229 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2231 /* XXX - L2ARC 1.0 does not support expansion */
2232 if (!vd
->vdev_aux
) {
2233 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2234 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2238 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2240 if (!vd
->vdev_aux
) {
2241 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2242 pvd
->vdev_expanding
= B_FALSE
;
2246 *newstate
= vd
->vdev_state
;
2247 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2248 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2249 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2250 vd
->vdev_parent
->vdev_child
[0] == vd
)
2251 vd
->vdev_unspare
= B_TRUE
;
2253 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2255 /* XXX - L2ARC 1.0 does not support expansion */
2257 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2258 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2260 return (spa_vdev_state_exit(spa
, vd
, 0));
2264 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2268 uint64_t generation
;
2269 metaslab_group_t
*mg
;
2272 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2274 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2275 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2277 if (!vd
->vdev_ops
->vdev_op_leaf
)
2278 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2282 generation
= spa
->spa_config_generation
+ 1;
2285 * If the device isn't already offline, try to offline it.
2287 if (!vd
->vdev_offline
) {
2289 * If this device has the only valid copy of some data,
2290 * don't allow it to be offlined. Log devices are always
2293 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2294 vdev_dtl_required(vd
))
2295 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2298 * If the top-level is a slog and it has had allocations
2299 * then proceed. We check that the vdev's metaslab group
2300 * is not NULL since it's possible that we may have just
2301 * added this vdev but not yet initialized its metaslabs.
2303 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2305 * Prevent any future allocations.
2307 metaslab_group_passivate(mg
);
2308 (void) spa_vdev_state_exit(spa
, vd
, 0);
2310 error
= spa_offline_log(spa
);
2312 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2315 * Check to see if the config has changed.
2317 if (error
|| generation
!= spa
->spa_config_generation
) {
2318 metaslab_group_activate(mg
);
2320 return (spa_vdev_state_exit(spa
,
2322 (void) spa_vdev_state_exit(spa
, vd
, 0);
2325 ASSERT3U(tvd
->vdev_stat
.vs_alloc
, ==, 0);
2329 * Offline this device and reopen its top-level vdev.
2330 * If the top-level vdev is a log device then just offline
2331 * it. Otherwise, if this action results in the top-level
2332 * vdev becoming unusable, undo it and fail the request.
2334 vd
->vdev_offline
= B_TRUE
;
2337 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2338 vdev_is_dead(tvd
)) {
2339 vd
->vdev_offline
= B_FALSE
;
2341 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2345 * Add the device back into the metaslab rotor so that
2346 * once we online the device it's open for business.
2348 if (tvd
->vdev_islog
&& mg
!= NULL
)
2349 metaslab_group_activate(mg
);
2352 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2354 return (spa_vdev_state_exit(spa
, vd
, 0));
2358 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2362 mutex_enter(&spa
->spa_vdev_top_lock
);
2363 error
= vdev_offline_locked(spa
, guid
, flags
);
2364 mutex_exit(&spa
->spa_vdev_top_lock
);
2370 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2371 * vdev_offline(), we assume the spa config is locked. We also clear all
2372 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2375 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2377 vdev_t
*rvd
= spa
->spa_root_vdev
;
2380 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2385 vd
->vdev_stat
.vs_read_errors
= 0;
2386 vd
->vdev_stat
.vs_write_errors
= 0;
2387 vd
->vdev_stat
.vs_checksum_errors
= 0;
2389 for (c
= 0; c
< vd
->vdev_children
; c
++)
2390 vdev_clear(spa
, vd
->vdev_child
[c
]);
2393 * If we're in the FAULTED state or have experienced failed I/O, then
2394 * clear the persistent state and attempt to reopen the device. We
2395 * also mark the vdev config dirty, so that the new faulted state is
2396 * written out to disk.
2398 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2399 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2402 * When reopening in reponse to a clear event, it may be due to
2403 * a fmadm repair request. In this case, if the device is
2404 * still broken, we want to still post the ereport again.
2406 vd
->vdev_forcefault
= B_TRUE
;
2408 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2409 vd
->vdev_cant_read
= B_FALSE
;
2410 vd
->vdev_cant_write
= B_FALSE
;
2412 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2414 vd
->vdev_forcefault
= B_FALSE
;
2416 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2417 vdev_state_dirty(vd
->vdev_top
);
2419 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2420 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2422 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_CLEAR
);
2426 * When clearing a FMA-diagnosed fault, we always want to
2427 * unspare the device, as we assume that the original spare was
2428 * done in response to the FMA fault.
2430 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2431 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2432 vd
->vdev_parent
->vdev_child
[0] == vd
)
2433 vd
->vdev_unspare
= B_TRUE
;
2437 vdev_is_dead(vdev_t
*vd
)
2440 * Holes and missing devices are always considered "dead".
2441 * This simplifies the code since we don't have to check for
2442 * these types of devices in the various code paths.
2443 * Instead we rely on the fact that we skip over dead devices
2444 * before issuing I/O to them.
2446 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2447 vd
->vdev_ops
== &vdev_missing_ops
);
2451 vdev_readable(vdev_t
*vd
)
2453 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2457 vdev_writeable(vdev_t
*vd
)
2459 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2463 vdev_allocatable(vdev_t
*vd
)
2465 uint64_t state
= vd
->vdev_state
;
2468 * We currently allow allocations from vdevs which may be in the
2469 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2470 * fails to reopen then we'll catch it later when we're holding
2471 * the proper locks. Note that we have to get the vdev state
2472 * in a local variable because although it changes atomically,
2473 * we're asking two separate questions about it.
2475 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2476 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2480 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2482 ASSERT(zio
->io_vd
== vd
);
2484 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2487 if (zio
->io_type
== ZIO_TYPE_READ
)
2488 return (!vd
->vdev_cant_read
);
2490 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2491 return (!vd
->vdev_cant_write
);
2497 * Get statistics for the given vdev.
2500 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2502 vdev_t
*rvd
= vd
->vdev_spa
->spa_root_vdev
;
2505 mutex_enter(&vd
->vdev_stat_lock
);
2506 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2507 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2508 vs
->vs_state
= vd
->vdev_state
;
2509 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2510 if (vd
->vdev_ops
->vdev_op_leaf
)
2511 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2512 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2513 mutex_exit(&vd
->vdev_stat_lock
);
2516 * If we're getting stats on the root vdev, aggregate the I/O counts
2517 * over all top-level vdevs (i.e. the direct children of the root).
2520 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2521 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2522 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2524 mutex_enter(&vd
->vdev_stat_lock
);
2525 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2526 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2527 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2529 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2530 mutex_exit(&vd
->vdev_stat_lock
);
2536 vdev_clear_stats(vdev_t
*vd
)
2538 mutex_enter(&vd
->vdev_stat_lock
);
2539 vd
->vdev_stat
.vs_space
= 0;
2540 vd
->vdev_stat
.vs_dspace
= 0;
2541 vd
->vdev_stat
.vs_alloc
= 0;
2542 mutex_exit(&vd
->vdev_stat_lock
);
2546 vdev_scan_stat_init(vdev_t
*vd
)
2548 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2551 for (c
= 0; c
< vd
->vdev_children
; c
++)
2552 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2554 mutex_enter(&vd
->vdev_stat_lock
);
2555 vs
->vs_scan_processed
= 0;
2556 mutex_exit(&vd
->vdev_stat_lock
);
2560 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2562 spa_t
*spa
= zio
->io_spa
;
2563 vdev_t
*rvd
= spa
->spa_root_vdev
;
2564 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2566 uint64_t txg
= zio
->io_txg
;
2567 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2568 zio_type_t type
= zio
->io_type
;
2569 int flags
= zio
->io_flags
;
2572 * If this i/o is a gang leader, it didn't do any actual work.
2574 if (zio
->io_gang_tree
)
2577 if (zio
->io_error
== 0) {
2579 * If this is a root i/o, don't count it -- we've already
2580 * counted the top-level vdevs, and vdev_get_stats() will
2581 * aggregate them when asked. This reduces contention on
2582 * the root vdev_stat_lock and implicitly handles blocks
2583 * that compress away to holes, for which there is no i/o.
2584 * (Holes never create vdev children, so all the counters
2585 * remain zero, which is what we want.)
2587 * Note: this only applies to successful i/o (io_error == 0)
2588 * because unlike i/o counts, errors are not additive.
2589 * When reading a ditto block, for example, failure of
2590 * one top-level vdev does not imply a root-level error.
2595 ASSERT(vd
== zio
->io_vd
);
2597 if (flags
& ZIO_FLAG_IO_BYPASS
)
2600 mutex_enter(&vd
->vdev_stat_lock
);
2602 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2603 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2604 dsl_scan_phys_t
*scn_phys
=
2605 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2606 uint64_t *processed
= &scn_phys
->scn_processed
;
2609 if (vd
->vdev_ops
->vdev_op_leaf
)
2610 atomic_add_64(processed
, psize
);
2611 vs
->vs_scan_processed
+= psize
;
2614 if (flags
& ZIO_FLAG_SELF_HEAL
)
2615 vs
->vs_self_healed
+= psize
;
2619 vs
->vs_bytes
[type
] += psize
;
2621 mutex_exit(&vd
->vdev_stat_lock
);
2625 if (flags
& ZIO_FLAG_SPECULATIVE
)
2629 * If this is an I/O error that is going to be retried, then ignore the
2630 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2631 * hard errors, when in reality they can happen for any number of
2632 * innocuous reasons (bus resets, MPxIO link failure, etc).
2634 if (zio
->io_error
== EIO
&&
2635 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2639 * Intent logs writes won't propagate their error to the root
2640 * I/O so don't mark these types of failures as pool-level
2643 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2646 mutex_enter(&vd
->vdev_stat_lock
);
2647 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2648 if (zio
->io_error
== ECKSUM
)
2649 vs
->vs_checksum_errors
++;
2651 vs
->vs_read_errors
++;
2653 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2654 vs
->vs_write_errors
++;
2655 mutex_exit(&vd
->vdev_stat_lock
);
2657 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2658 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2659 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2660 spa
->spa_claiming
)) {
2662 * This is either a normal write (not a repair), or it's
2663 * a repair induced by the scrub thread, or it's a repair
2664 * made by zil_claim() during spa_load() in the first txg.
2665 * In the normal case, we commit the DTL change in the same
2666 * txg as the block was born. In the scrub-induced repair
2667 * case, we know that scrubs run in first-pass syncing context,
2668 * so we commit the DTL change in spa_syncing_txg(spa).
2669 * In the zil_claim() case, we commit in spa_first_txg(spa).
2671 * We currently do not make DTL entries for failed spontaneous
2672 * self-healing writes triggered by normal (non-scrubbing)
2673 * reads, because we have no transactional context in which to
2674 * do so -- and it's not clear that it'd be desirable anyway.
2676 if (vd
->vdev_ops
->vdev_op_leaf
) {
2677 uint64_t commit_txg
= txg
;
2678 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2679 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2680 ASSERT(spa_sync_pass(spa
) == 1);
2681 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2682 commit_txg
= spa_syncing_txg(spa
);
2683 } else if (spa
->spa_claiming
) {
2684 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2685 commit_txg
= spa_first_txg(spa
);
2687 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2688 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2690 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2691 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2692 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2695 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2700 * Update the in-core space usage stats for this vdev, its metaslab class,
2701 * and the root vdev.
2704 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2705 int64_t space_delta
)
2707 int64_t dspace_delta
= space_delta
;
2708 spa_t
*spa
= vd
->vdev_spa
;
2709 vdev_t
*rvd
= spa
->spa_root_vdev
;
2710 metaslab_group_t
*mg
= vd
->vdev_mg
;
2711 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2713 ASSERT(vd
== vd
->vdev_top
);
2716 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2717 * factor. We must calculate this here and not at the root vdev
2718 * because the root vdev's psize-to-asize is simply the max of its
2719 * childrens', thus not accurate enough for us.
2721 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2722 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2723 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2724 vd
->vdev_deflate_ratio
;
2726 mutex_enter(&vd
->vdev_stat_lock
);
2727 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2728 vd
->vdev_stat
.vs_space
+= space_delta
;
2729 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2730 mutex_exit(&vd
->vdev_stat_lock
);
2732 if (mc
== spa_normal_class(spa
)) {
2733 mutex_enter(&rvd
->vdev_stat_lock
);
2734 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2735 rvd
->vdev_stat
.vs_space
+= space_delta
;
2736 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2737 mutex_exit(&rvd
->vdev_stat_lock
);
2741 ASSERT(rvd
== vd
->vdev_parent
);
2742 ASSERT(vd
->vdev_ms_count
!= 0);
2744 metaslab_class_space_update(mc
,
2745 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2750 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2751 * so that it will be written out next time the vdev configuration is synced.
2752 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2755 vdev_config_dirty(vdev_t
*vd
)
2757 spa_t
*spa
= vd
->vdev_spa
;
2758 vdev_t
*rvd
= spa
->spa_root_vdev
;
2761 ASSERT(spa_writeable(spa
));
2764 * If this is an aux vdev (as with l2cache and spare devices), then we
2765 * update the vdev config manually and set the sync flag.
2767 if (vd
->vdev_aux
!= NULL
) {
2768 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2772 for (c
= 0; c
< sav
->sav_count
; c
++) {
2773 if (sav
->sav_vdevs
[c
] == vd
)
2777 if (c
== sav
->sav_count
) {
2779 * We're being removed. There's nothing more to do.
2781 ASSERT(sav
->sav_sync
== B_TRUE
);
2785 sav
->sav_sync
= B_TRUE
;
2787 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2788 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2789 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2790 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2796 * Setting the nvlist in the middle if the array is a little
2797 * sketchy, but it will work.
2799 nvlist_free(aux
[c
]);
2800 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
2806 * The dirty list is protected by the SCL_CONFIG lock. The caller
2807 * must either hold SCL_CONFIG as writer, or must be the sync thread
2808 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2809 * so this is sufficient to ensure mutual exclusion.
2811 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2812 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2813 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2816 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2817 vdev_config_dirty(rvd
->vdev_child
[c
]);
2819 ASSERT(vd
== vd
->vdev_top
);
2821 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
2823 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
2828 vdev_config_clean(vdev_t
*vd
)
2830 spa_t
*spa
= vd
->vdev_spa
;
2832 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2833 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2834 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2836 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
2837 list_remove(&spa
->spa_config_dirty_list
, vd
);
2841 * Mark a top-level vdev's state as dirty, so that the next pass of
2842 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2843 * the state changes from larger config changes because they require
2844 * much less locking, and are often needed for administrative actions.
2847 vdev_state_dirty(vdev_t
*vd
)
2849 spa_t
*spa
= vd
->vdev_spa
;
2851 ASSERT(spa_writeable(spa
));
2852 ASSERT(vd
== vd
->vdev_top
);
2855 * The state list is protected by the SCL_STATE lock. The caller
2856 * must either hold SCL_STATE as writer, or must be the sync thread
2857 * (which holds SCL_STATE as reader). There's only one sync thread,
2858 * so this is sufficient to ensure mutual exclusion.
2860 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2861 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2862 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2864 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
2865 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
2869 vdev_state_clean(vdev_t
*vd
)
2871 spa_t
*spa
= vd
->vdev_spa
;
2873 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2874 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2875 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2877 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
2878 list_remove(&spa
->spa_state_dirty_list
, vd
);
2882 * Propagate vdev state up from children to parent.
2885 vdev_propagate_state(vdev_t
*vd
)
2887 spa_t
*spa
= vd
->vdev_spa
;
2888 vdev_t
*rvd
= spa
->spa_root_vdev
;
2889 int degraded
= 0, faulted
= 0;
2894 if (vd
->vdev_children
> 0) {
2895 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2896 child
= vd
->vdev_child
[c
];
2899 * Don't factor holes into the decision.
2901 if (child
->vdev_ishole
)
2904 if (!vdev_readable(child
) ||
2905 (!vdev_writeable(child
) && spa_writeable(spa
))) {
2907 * Root special: if there is a top-level log
2908 * device, treat the root vdev as if it were
2911 if (child
->vdev_islog
&& vd
== rvd
)
2915 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
2919 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
2923 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
2926 * Root special: if there is a top-level vdev that cannot be
2927 * opened due to corrupted metadata, then propagate the root
2928 * vdev's aux state as 'corrupt' rather than 'insufficient
2931 if (corrupted
&& vd
== rvd
&&
2932 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
2933 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2934 VDEV_AUX_CORRUPT_DATA
);
2937 if (vd
->vdev_parent
)
2938 vdev_propagate_state(vd
->vdev_parent
);
2942 * Set a vdev's state. If this is during an open, we don't update the parent
2943 * state, because we're in the process of opening children depth-first.
2944 * Otherwise, we propagate the change to the parent.
2946 * If this routine places a device in a faulted state, an appropriate ereport is
2950 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
2952 uint64_t save_state
;
2953 spa_t
*spa
= vd
->vdev_spa
;
2955 if (state
== vd
->vdev_state
) {
2956 vd
->vdev_stat
.vs_aux
= aux
;
2960 save_state
= vd
->vdev_state
;
2962 vd
->vdev_state
= state
;
2963 vd
->vdev_stat
.vs_aux
= aux
;
2966 * If we are setting the vdev state to anything but an open state, then
2967 * always close the underlying device unless the device has requested
2968 * a delayed close (i.e. we're about to remove or fault the device).
2969 * Otherwise, we keep accessible but invalid devices open forever.
2970 * We don't call vdev_close() itself, because that implies some extra
2971 * checks (offline, etc) that we don't want here. This is limited to
2972 * leaf devices, because otherwise closing the device will affect other
2975 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
2976 vd
->vdev_ops
->vdev_op_leaf
)
2977 vd
->vdev_ops
->vdev_op_close(vd
);
2980 * If we have brought this vdev back into service, we need
2981 * to notify fmd so that it can gracefully repair any outstanding
2982 * cases due to a missing device. We do this in all cases, even those
2983 * that probably don't correlate to a repaired fault. This is sure to
2984 * catch all cases, and we let the zfs-retire agent sort it out. If
2985 * this is a transient state it's OK, as the retire agent will
2986 * double-check the state of the vdev before repairing it.
2988 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
2989 vd
->vdev_prevstate
!= state
)
2990 zfs_post_state_change(spa
, vd
);
2992 if (vd
->vdev_removed
&&
2993 state
== VDEV_STATE_CANT_OPEN
&&
2994 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
2996 * If the previous state is set to VDEV_STATE_REMOVED, then this
2997 * device was previously marked removed and someone attempted to
2998 * reopen it. If this failed due to a nonexistent device, then
2999 * keep the device in the REMOVED state. We also let this be if
3000 * it is one of our special test online cases, which is only
3001 * attempting to online the device and shouldn't generate an FMA
3004 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3005 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3006 } else if (state
== VDEV_STATE_REMOVED
) {
3007 vd
->vdev_removed
= B_TRUE
;
3008 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3010 * If we fail to open a vdev during an import or recovery, we
3011 * mark it as "not available", which signifies that it was
3012 * never there to begin with. Failure to open such a device
3013 * is not considered an error.
3015 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3016 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3017 vd
->vdev_ops
->vdev_op_leaf
)
3018 vd
->vdev_not_present
= 1;
3021 * Post the appropriate ereport. If the 'prevstate' field is
3022 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3023 * that this is part of a vdev_reopen(). In this case, we don't
3024 * want to post the ereport if the device was already in the
3025 * CANT_OPEN state beforehand.
3027 * If the 'checkremove' flag is set, then this is an attempt to
3028 * online the device in response to an insertion event. If we
3029 * hit this case, then we have detected an insertion event for a
3030 * faulted or offline device that wasn't in the removed state.
3031 * In this scenario, we don't post an ereport because we are
3032 * about to replace the device, or attempt an online with
3033 * vdev_forcefault, which will generate the fault for us.
3035 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3036 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3037 vd
!= spa
->spa_root_vdev
) {
3041 case VDEV_AUX_OPEN_FAILED
:
3042 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3044 case VDEV_AUX_CORRUPT_DATA
:
3045 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3047 case VDEV_AUX_NO_REPLICAS
:
3048 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3050 case VDEV_AUX_BAD_GUID_SUM
:
3051 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3053 case VDEV_AUX_TOO_SMALL
:
3054 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3056 case VDEV_AUX_BAD_LABEL
:
3057 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3060 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3063 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3066 /* Erase any notion of persistent removed state */
3067 vd
->vdev_removed
= B_FALSE
;
3069 vd
->vdev_removed
= B_FALSE
;
3072 if (!isopen
&& vd
->vdev_parent
)
3073 vdev_propagate_state(vd
->vdev_parent
);
3077 * Check the vdev configuration to ensure that it's capable of supporting
3081 vdev_is_bootable(vdev_t
*vd
)
3083 #if defined(__sun__) || defined(__sun)
3085 * Currently, we do not support RAID-Z or partial configuration.
3086 * In addition, only a single top-level vdev is allowed and none of the
3087 * leaves can be wholedisks.
3091 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3092 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3094 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3095 vd
->vdev_children
> 1) {
3097 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3098 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3101 } else if (vd
->vdev_wholedisk
== 1) {
3105 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3106 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3109 #endif /* __sun__ || __sun */
3114 * Load the state from the original vdev tree (ovd) which
3115 * we've retrieved from the MOS config object. If the original
3116 * vdev was offline or faulted then we transfer that state to the
3117 * device in the current vdev tree (nvd).
3120 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3124 ASSERT(nvd
->vdev_top
->vdev_islog
);
3125 ASSERT(spa_config_held(nvd
->vdev_spa
,
3126 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3127 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3129 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3130 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3132 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3134 * Restore the persistent vdev state
3136 nvd
->vdev_offline
= ovd
->vdev_offline
;
3137 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3138 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3139 nvd
->vdev_removed
= ovd
->vdev_removed
;
3144 * Determine if a log device has valid content. If the vdev was
3145 * removed or faulted in the MOS config then we know that
3146 * the content on the log device has already been written to the pool.
3149 vdev_log_state_valid(vdev_t
*vd
)
3153 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3157 for (c
= 0; c
< vd
->vdev_children
; c
++)
3158 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3165 * Expand a vdev if possible.
3168 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3170 ASSERT(vd
->vdev_top
== vd
);
3171 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3173 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3174 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3175 vdev_config_dirty(vd
);
3183 vdev_split(vdev_t
*vd
)
3185 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3187 vdev_remove_child(pvd
, vd
);
3188 vdev_compact_children(pvd
);
3190 cvd
= pvd
->vdev_child
[0];
3191 if (pvd
->vdev_children
== 1) {
3192 vdev_remove_parent(cvd
);
3193 cvd
->vdev_splitting
= B_TRUE
;
3195 vdev_propagate_state(cvd
);
3199 vdev_deadman(vdev_t
*vd
)
3203 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3204 vdev_t
*cvd
= vd
->vdev_child
[c
];
3209 if (vd
->vdev_ops
->vdev_op_leaf
) {
3210 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3212 mutex_enter(&vq
->vq_lock
);
3213 if (avl_numnodes(&vq
->vq_pending_tree
) > 0) {
3214 spa_t
*spa
= vd
->vdev_spa
;
3219 * Look at the head of all the pending queues,
3220 * if any I/O has been outstanding for longer than
3221 * the spa_deadman_synctime we log a zevent.
3223 fio
= avl_first(&vq
->vq_pending_tree
);
3224 delta
= ddi_get_lbolt64() - fio
->io_timestamp
;
3225 if (delta
> NSEC_TO_TICK(spa_deadman_synctime(spa
))) {
3226 zfs_dbgmsg("SLOW IO: zio timestamp %llu, "
3227 "delta %llu, last io %llu",
3228 fio
->io_timestamp
, delta
,
3229 vq
->vq_io_complete_ts
);
3230 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3231 spa
, vd
, fio
, 0, 0);
3234 mutex_exit(&vq
->vq_lock
);
3238 #if defined(_KERNEL) && defined(HAVE_SPL)
3239 EXPORT_SYMBOL(vdev_fault
);
3240 EXPORT_SYMBOL(vdev_degrade
);
3241 EXPORT_SYMBOL(vdev_online
);
3242 EXPORT_SYMBOL(vdev_offline
);
3243 EXPORT_SYMBOL(vdev_clear
);
3245 module_param(zfs_scrub_limit
, int, 0644);
3246 MODULE_PARM_DESC(zfs_scrub_limit
, "Max scrub/resilver I/O per leaf vdev");