4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/fm/fs/zfs.h>
31 #include <sys/spa_impl.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/uberblock_impl.h>
36 #include <sys/metaslab.h>
37 #include <sys/metaslab_impl.h>
38 #include <sys/space_map.h>
41 #include <sys/fs/zfs.h>
44 #include <sys/dsl_scan.h>
47 * Virtual device management.
50 static vdev_ops_t
*vdev_ops_table
[] = {
63 /* maximum scrub/resilver I/O queue per leaf vdev */
64 int zfs_scrub_limit
= 10;
67 * Given a vdev type, return the appropriate ops vector.
70 vdev_getops(const char *type
)
72 vdev_ops_t
*ops
, **opspp
;
74 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
75 if (strcmp(ops
->vdev_op_type
, type
) == 0)
82 * Default asize function: return the MAX of psize with the asize of
83 * all children. This is what's used by anything other than RAID-Z.
86 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
88 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
92 for (c
= 0; c
< vd
->vdev_children
; c
++) {
93 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
94 asize
= MAX(asize
, csize
);
101 * Get the minimum allocatable size. We define the allocatable size as
102 * the vdev's asize rounded to the nearest metaslab. This allows us to
103 * replace or attach devices which don't have the same physical size but
104 * can still satisfy the same number of allocations.
107 vdev_get_min_asize(vdev_t
*vd
)
109 vdev_t
*pvd
= vd
->vdev_parent
;
112 * If our parent is NULL (inactive spare or cache) or is the root,
113 * just return our own asize.
116 return (vd
->vdev_asize
);
119 * The top-level vdev just returns the allocatable size rounded
120 * to the nearest metaslab.
122 if (vd
== vd
->vdev_top
)
123 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
126 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
127 * so each child must provide at least 1/Nth of its asize.
129 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
130 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
132 return (pvd
->vdev_min_asize
);
136 vdev_set_min_asize(vdev_t
*vd
)
139 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
141 for (c
= 0; c
< vd
->vdev_children
; c
++)
142 vdev_set_min_asize(vd
->vdev_child
[c
]);
146 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
148 vdev_t
*rvd
= spa
->spa_root_vdev
;
150 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
152 if (vdev
< rvd
->vdev_children
) {
153 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
154 return (rvd
->vdev_child
[vdev
]);
161 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
166 if (vd
->vdev_guid
== guid
)
169 for (c
= 0; c
< vd
->vdev_children
; c
++)
170 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
178 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
180 size_t oldsize
, newsize
;
181 uint64_t id
= cvd
->vdev_id
;
184 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
185 ASSERT(cvd
->vdev_parent
== NULL
);
187 cvd
->vdev_parent
= pvd
;
192 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
194 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
195 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
196 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
198 newchild
= kmem_zalloc(newsize
, KM_PUSHPAGE
);
199 if (pvd
->vdev_child
!= NULL
) {
200 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
201 kmem_free(pvd
->vdev_child
, oldsize
);
204 pvd
->vdev_child
= newchild
;
205 pvd
->vdev_child
[id
] = cvd
;
207 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
208 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
211 * Walk up all ancestors to update guid sum.
213 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
214 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
218 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
221 uint_t id
= cvd
->vdev_id
;
223 ASSERT(cvd
->vdev_parent
== pvd
);
228 ASSERT(id
< pvd
->vdev_children
);
229 ASSERT(pvd
->vdev_child
[id
] == cvd
);
231 pvd
->vdev_child
[id
] = NULL
;
232 cvd
->vdev_parent
= NULL
;
234 for (c
= 0; c
< pvd
->vdev_children
; c
++)
235 if (pvd
->vdev_child
[c
])
238 if (c
== pvd
->vdev_children
) {
239 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
240 pvd
->vdev_child
= NULL
;
241 pvd
->vdev_children
= 0;
245 * Walk up all ancestors to update guid sum.
247 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
248 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
252 * Remove any holes in the child array.
255 vdev_compact_children(vdev_t
*pvd
)
257 vdev_t
**newchild
, *cvd
;
258 int oldc
= pvd
->vdev_children
;
262 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
264 for (c
= newc
= 0; c
< oldc
; c
++)
265 if (pvd
->vdev_child
[c
])
268 newchild
= kmem_alloc(newc
* sizeof (vdev_t
*), KM_PUSHPAGE
);
270 for (c
= newc
= 0; c
< oldc
; c
++) {
271 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
272 newchild
[newc
] = cvd
;
273 cvd
->vdev_id
= newc
++;
277 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
278 pvd
->vdev_child
= newchild
;
279 pvd
->vdev_children
= newc
;
283 * Allocate and minimally initialize a vdev_t.
286 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
291 vd
= kmem_zalloc(sizeof (vdev_t
), KM_PUSHPAGE
);
293 if (spa
->spa_root_vdev
== NULL
) {
294 ASSERT(ops
== &vdev_root_ops
);
295 spa
->spa_root_vdev
= vd
;
296 spa
->spa_load_guid
= spa_generate_guid(NULL
);
299 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
300 if (spa
->spa_root_vdev
== vd
) {
302 * The root vdev's guid will also be the pool guid,
303 * which must be unique among all pools.
305 guid
= spa_generate_guid(NULL
);
308 * Any other vdev's guid must be unique within the pool.
310 guid
= spa_generate_guid(spa
);
312 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
317 vd
->vdev_guid
= guid
;
318 vd
->vdev_guid_sum
= guid
;
320 vd
->vdev_state
= VDEV_STATE_CLOSED
;
321 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
323 list_link_init(&vd
->vdev_config_dirty_node
);
324 list_link_init(&vd
->vdev_state_dirty_node
);
325 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
326 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
327 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
328 for (t
= 0; t
< DTL_TYPES
; t
++) {
329 space_map_create(&vd
->vdev_dtl
[t
], 0, -1ULL, 0,
332 txg_list_create(&vd
->vdev_ms_list
,
333 offsetof(struct metaslab
, ms_txg_node
));
334 txg_list_create(&vd
->vdev_dtl_list
,
335 offsetof(struct vdev
, vdev_dtl_node
));
336 vd
->vdev_stat
.vs_timestamp
= gethrtime();
344 * Allocate a new vdev. The 'alloctype' is used to control whether we are
345 * creating a new vdev or loading an existing one - the behavior is slightly
346 * different for each case.
349 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
354 uint64_t guid
= 0, islog
, nparity
;
357 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
359 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
362 if ((ops
= vdev_getops(type
)) == NULL
)
366 * If this is a load, get the vdev guid from the nvlist.
367 * Otherwise, vdev_alloc_common() will generate one for us.
369 if (alloctype
== VDEV_ALLOC_LOAD
) {
372 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
376 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
378 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
379 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
381 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
382 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
384 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
385 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
390 * The first allocated vdev must be of type 'root'.
392 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
396 * Determine whether we're a log vdev.
399 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
400 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
403 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
407 * Set the nparity property for RAID-Z vdevs.
410 if (ops
== &vdev_raidz_ops
) {
411 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
413 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
416 * Previous versions could only support 1 or 2 parity
420 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
423 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
427 * We require the parity to be specified for SPAs that
428 * support multiple parity levels.
430 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
433 * Otherwise, we default to 1 parity device for RAID-Z.
440 ASSERT(nparity
!= -1ULL);
442 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
444 vd
->vdev_islog
= islog
;
445 vd
->vdev_nparity
= nparity
;
447 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
448 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
449 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
450 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
451 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
452 &vd
->vdev_physpath
) == 0)
453 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
454 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
455 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
458 * Set the whole_disk property. If it's not specified, leave the value
461 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
462 &vd
->vdev_wholedisk
) != 0)
463 vd
->vdev_wholedisk
= -1ULL;
466 * Look for the 'not present' flag. This will only be set if the device
467 * was not present at the time of import.
469 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
470 &vd
->vdev_not_present
);
473 * Get the alignment requirement.
475 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
478 * Retrieve the vdev creation time.
480 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
484 * If we're a top-level vdev, try to load the allocation parameters.
486 if (parent
&& !parent
->vdev_parent
&&
487 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
488 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
490 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
492 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
494 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
498 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
499 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
500 alloctype
== VDEV_ALLOC_ADD
||
501 alloctype
== VDEV_ALLOC_SPLIT
||
502 alloctype
== VDEV_ALLOC_ROOTPOOL
);
503 vd
->vdev_mg
= metaslab_group_create(islog
?
504 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
508 * If we're a leaf vdev, try to load the DTL object and other state.
510 if (vd
->vdev_ops
->vdev_op_leaf
&&
511 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
512 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
513 if (alloctype
== VDEV_ALLOC_LOAD
) {
514 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
515 &vd
->vdev_dtl_smo
.smo_object
);
516 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
520 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
523 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
524 &spare
) == 0 && spare
)
528 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
531 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVERING
,
532 &vd
->vdev_resilvering
);
535 * When importing a pool, we want to ignore the persistent fault
536 * state, as the diagnosis made on another system may not be
537 * valid in the current context. Local vdevs will
538 * remain in the faulted state.
540 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
541 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
543 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
545 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
548 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
552 VDEV_AUX_ERR_EXCEEDED
;
553 if (nvlist_lookup_string(nv
,
554 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
555 strcmp(aux
, "external") == 0)
556 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
562 * Add ourselves to the parent's list of children.
564 vdev_add_child(parent
, vd
);
572 vdev_free(vdev_t
*vd
)
575 spa_t
*spa
= vd
->vdev_spa
;
578 * vdev_free() implies closing the vdev first. This is simpler than
579 * trying to ensure complicated semantics for all callers.
583 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
584 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
589 for (c
= 0; c
< vd
->vdev_children
; c
++)
590 vdev_free(vd
->vdev_child
[c
]);
592 ASSERT(vd
->vdev_child
== NULL
);
593 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
596 * Discard allocation state.
598 if (vd
->vdev_mg
!= NULL
) {
599 vdev_metaslab_fini(vd
);
600 metaslab_group_destroy(vd
->vdev_mg
);
603 ASSERT3U(vd
->vdev_stat
.vs_space
, ==, 0);
604 ASSERT3U(vd
->vdev_stat
.vs_dspace
, ==, 0);
605 ASSERT3U(vd
->vdev_stat
.vs_alloc
, ==, 0);
608 * Remove this vdev from its parent's child list.
610 vdev_remove_child(vd
->vdev_parent
, vd
);
612 ASSERT(vd
->vdev_parent
== NULL
);
615 * Clean up vdev structure.
621 spa_strfree(vd
->vdev_path
);
623 spa_strfree(vd
->vdev_devid
);
624 if (vd
->vdev_physpath
)
625 spa_strfree(vd
->vdev_physpath
);
627 spa_strfree(vd
->vdev_fru
);
629 if (vd
->vdev_isspare
)
630 spa_spare_remove(vd
);
631 if (vd
->vdev_isl2cache
)
632 spa_l2cache_remove(vd
);
634 txg_list_destroy(&vd
->vdev_ms_list
);
635 txg_list_destroy(&vd
->vdev_dtl_list
);
637 mutex_enter(&vd
->vdev_dtl_lock
);
638 for (t
= 0; t
< DTL_TYPES
; t
++) {
639 space_map_unload(&vd
->vdev_dtl
[t
]);
640 space_map_destroy(&vd
->vdev_dtl
[t
]);
642 mutex_exit(&vd
->vdev_dtl_lock
);
644 mutex_destroy(&vd
->vdev_dtl_lock
);
645 mutex_destroy(&vd
->vdev_stat_lock
);
646 mutex_destroy(&vd
->vdev_probe_lock
);
648 if (vd
== spa
->spa_root_vdev
)
649 spa
->spa_root_vdev
= NULL
;
651 kmem_free(vd
, sizeof (vdev_t
));
655 * Transfer top-level vdev state from svd to tvd.
658 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
660 spa_t
*spa
= svd
->vdev_spa
;
665 ASSERT(tvd
== tvd
->vdev_top
);
667 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
668 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
669 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
671 svd
->vdev_ms_array
= 0;
672 svd
->vdev_ms_shift
= 0;
673 svd
->vdev_ms_count
= 0;
676 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
677 tvd
->vdev_mg
= svd
->vdev_mg
;
678 tvd
->vdev_ms
= svd
->vdev_ms
;
683 if (tvd
->vdev_mg
!= NULL
)
684 tvd
->vdev_mg
->mg_vd
= tvd
;
686 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
687 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
688 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
690 svd
->vdev_stat
.vs_alloc
= 0;
691 svd
->vdev_stat
.vs_space
= 0;
692 svd
->vdev_stat
.vs_dspace
= 0;
694 for (t
= 0; t
< TXG_SIZE
; t
++) {
695 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
696 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
697 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
698 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
699 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
700 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
703 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
704 vdev_config_clean(svd
);
705 vdev_config_dirty(tvd
);
708 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
709 vdev_state_clean(svd
);
710 vdev_state_dirty(tvd
);
713 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
714 svd
->vdev_deflate_ratio
= 0;
716 tvd
->vdev_islog
= svd
->vdev_islog
;
721 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
730 for (c
= 0; c
< vd
->vdev_children
; c
++)
731 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
735 * Add a mirror/replacing vdev above an existing vdev.
738 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
740 spa_t
*spa
= cvd
->vdev_spa
;
741 vdev_t
*pvd
= cvd
->vdev_parent
;
744 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
746 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
748 mvd
->vdev_asize
= cvd
->vdev_asize
;
749 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
750 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
751 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
752 mvd
->vdev_state
= cvd
->vdev_state
;
753 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
755 vdev_remove_child(pvd
, cvd
);
756 vdev_add_child(pvd
, mvd
);
757 cvd
->vdev_id
= mvd
->vdev_children
;
758 vdev_add_child(mvd
, cvd
);
759 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
761 if (mvd
== mvd
->vdev_top
)
762 vdev_top_transfer(cvd
, mvd
);
768 * Remove a 1-way mirror/replacing vdev from the tree.
771 vdev_remove_parent(vdev_t
*cvd
)
773 vdev_t
*mvd
= cvd
->vdev_parent
;
774 vdev_t
*pvd
= mvd
->vdev_parent
;
776 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
778 ASSERT(mvd
->vdev_children
== 1);
779 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
780 mvd
->vdev_ops
== &vdev_replacing_ops
||
781 mvd
->vdev_ops
== &vdev_spare_ops
);
782 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
784 vdev_remove_child(mvd
, cvd
);
785 vdev_remove_child(pvd
, mvd
);
788 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
789 * Otherwise, we could have detached an offline device, and when we
790 * go to import the pool we'll think we have two top-level vdevs,
791 * instead of a different version of the same top-level vdev.
793 if (mvd
->vdev_top
== mvd
) {
794 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
795 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
796 cvd
->vdev_guid
+= guid_delta
;
797 cvd
->vdev_guid_sum
+= guid_delta
;
799 cvd
->vdev_id
= mvd
->vdev_id
;
800 vdev_add_child(pvd
, cvd
);
801 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
803 if (cvd
== cvd
->vdev_top
)
804 vdev_top_transfer(mvd
, cvd
);
806 ASSERT(mvd
->vdev_children
== 0);
811 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
813 spa_t
*spa
= vd
->vdev_spa
;
814 objset_t
*mos
= spa
->spa_meta_objset
;
816 uint64_t oldc
= vd
->vdev_ms_count
;
817 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
821 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
824 * This vdev is not being allocated from yet or is a hole.
826 if (vd
->vdev_ms_shift
== 0)
829 ASSERT(!vd
->vdev_ishole
);
832 * Compute the raidz-deflation ratio. Note, we hard-code
833 * in 128k (1 << 17) because it is the current "typical" blocksize.
834 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
835 * or we will inconsistently account for existing bp's.
837 vd
->vdev_deflate_ratio
= (1 << 17) /
838 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
840 ASSERT(oldc
<= newc
);
842 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_PUSHPAGE
| KM_NODEBUG
);
845 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
846 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
850 vd
->vdev_ms_count
= newc
;
852 for (m
= oldc
; m
< newc
; m
++) {
853 space_map_obj_t smo
= { 0, 0, 0 };
856 error
= dmu_read(mos
, vd
->vdev_ms_array
,
857 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
863 error
= dmu_bonus_hold(mos
, object
, FTAG
, &db
);
866 ASSERT3U(db
->db_size
, >=, sizeof (smo
));
867 bcopy(db
->db_data
, &smo
, sizeof (smo
));
868 ASSERT3U(smo
.smo_object
, ==, object
);
869 dmu_buf_rele(db
, FTAG
);
872 vd
->vdev_ms
[m
] = metaslab_init(vd
->vdev_mg
, &smo
,
873 m
<< vd
->vdev_ms_shift
, 1ULL << vd
->vdev_ms_shift
, txg
);
877 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
880 * If the vdev is being removed we don't activate
881 * the metaslabs since we want to ensure that no new
882 * allocations are performed on this device.
884 if (oldc
== 0 && !vd
->vdev_removing
)
885 metaslab_group_activate(vd
->vdev_mg
);
888 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
894 vdev_metaslab_fini(vdev_t
*vd
)
897 uint64_t count
= vd
->vdev_ms_count
;
899 if (vd
->vdev_ms
!= NULL
) {
900 metaslab_group_passivate(vd
->vdev_mg
);
901 for (m
= 0; m
< count
; m
++)
902 if (vd
->vdev_ms
[m
] != NULL
)
903 metaslab_fini(vd
->vdev_ms
[m
]);
904 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
908 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
911 typedef struct vdev_probe_stats
{
912 boolean_t vps_readable
;
913 boolean_t vps_writeable
;
915 } vdev_probe_stats_t
;
918 vdev_probe_done(zio_t
*zio
)
920 spa_t
*spa
= zio
->io_spa
;
921 vdev_t
*vd
= zio
->io_vd
;
922 vdev_probe_stats_t
*vps
= zio
->io_private
;
924 ASSERT(vd
->vdev_probe_zio
!= NULL
);
926 if (zio
->io_type
== ZIO_TYPE_READ
) {
927 if (zio
->io_error
== 0)
928 vps
->vps_readable
= 1;
929 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
930 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
931 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
932 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
933 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
935 zio_buf_free(zio
->io_data
, zio
->io_size
);
937 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
938 if (zio
->io_error
== 0)
939 vps
->vps_writeable
= 1;
940 zio_buf_free(zio
->io_data
, zio
->io_size
);
941 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
944 vd
->vdev_cant_read
|= !vps
->vps_readable
;
945 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
947 if (vdev_readable(vd
) &&
948 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
951 ASSERT(zio
->io_error
!= 0);
952 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
953 spa
, vd
, NULL
, 0, 0);
954 zio
->io_error
= ENXIO
;
957 mutex_enter(&vd
->vdev_probe_lock
);
958 ASSERT(vd
->vdev_probe_zio
== zio
);
959 vd
->vdev_probe_zio
= NULL
;
960 mutex_exit(&vd
->vdev_probe_lock
);
962 while ((pio
= zio_walk_parents(zio
)) != NULL
)
963 if (!vdev_accessible(vd
, pio
))
964 pio
->io_error
= ENXIO
;
966 kmem_free(vps
, sizeof (*vps
));
971 * Determine whether this device is accessible by reading and writing
972 * to several known locations: the pad regions of each vdev label
973 * but the first (which we leave alone in case it contains a VTOC).
976 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
978 spa_t
*spa
= vd
->vdev_spa
;
979 vdev_probe_stats_t
*vps
= NULL
;
983 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
986 * Don't probe the probe.
988 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
992 * To prevent 'probe storms' when a device fails, we create
993 * just one probe i/o at a time. All zios that want to probe
994 * this vdev will become parents of the probe io.
996 mutex_enter(&vd
->vdev_probe_lock
);
998 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
999 vps
= kmem_zalloc(sizeof (*vps
), KM_PUSHPAGE
);
1001 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1002 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1005 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1007 * vdev_cant_read and vdev_cant_write can only
1008 * transition from TRUE to FALSE when we have the
1009 * SCL_ZIO lock as writer; otherwise they can only
1010 * transition from FALSE to TRUE. This ensures that
1011 * any zio looking at these values can assume that
1012 * failures persist for the life of the I/O. That's
1013 * important because when a device has intermittent
1014 * connectivity problems, we want to ensure that
1015 * they're ascribed to the device (ENXIO) and not
1018 * Since we hold SCL_ZIO as writer here, clear both
1019 * values so the probe can reevaluate from first
1022 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1023 vd
->vdev_cant_read
= B_FALSE
;
1024 vd
->vdev_cant_write
= B_FALSE
;
1027 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1028 vdev_probe_done
, vps
,
1029 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1032 * We can't change the vdev state in this context, so we
1033 * kick off an async task to do it on our behalf.
1036 vd
->vdev_probe_wanted
= B_TRUE
;
1037 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1042 zio_add_child(zio
, pio
);
1044 mutex_exit(&vd
->vdev_probe_lock
);
1047 ASSERT(zio
!= NULL
);
1051 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1052 zio_nowait(zio_read_phys(pio
, vd
,
1053 vdev_label_offset(vd
->vdev_psize
, l
,
1054 offsetof(vdev_label_t
, vl_pad2
)),
1055 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1056 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1057 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1068 vdev_open_child(void *arg
)
1072 vd
->vdev_open_thread
= curthread
;
1073 vd
->vdev_open_error
= vdev_open(vd
);
1074 vd
->vdev_open_thread
= NULL
;
1078 vdev_uses_zvols(vdev_t
*vd
)
1081 * Stacking zpools on top of zvols is unsupported until we implement a method
1082 * for determining if an arbitrary block device is a zvol without using the
1083 * path. Solaris would check the 'zvol' path component but this does not
1084 * exist in the Linux port, so we really should do something like stat the
1085 * file and check the major number. This is complicated by the fact that
1086 * we need to do this portably in user or kernel space.
1091 if (vd
->vdev_path
&& strncmp(vd
->vdev_path
, ZVOL_DIR
,
1092 strlen(ZVOL_DIR
)) == 0)
1094 for (c
= 0; c
< vd
->vdev_children
; c
++)
1095 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1102 vdev_open_children(vdev_t
*vd
)
1105 int children
= vd
->vdev_children
;
1109 * in order to handle pools on top of zvols, do the opens
1110 * in a single thread so that the same thread holds the
1111 * spa_namespace_lock
1113 if (vdev_uses_zvols(vd
)) {
1114 for (c
= 0; c
< children
; c
++)
1115 vd
->vdev_child
[c
]->vdev_open_error
=
1116 vdev_open(vd
->vdev_child
[c
]);
1119 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1120 children
, children
, TASKQ_PREPOPULATE
);
1122 for (c
= 0; c
< children
; c
++)
1123 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1130 * Prepare a virtual device for access.
1133 vdev_open(vdev_t
*vd
)
1135 spa_t
*spa
= vd
->vdev_spa
;
1138 uint64_t max_osize
= 0;
1139 uint64_t asize
, max_asize
, psize
;
1140 uint64_t ashift
= 0;
1143 ASSERT(vd
->vdev_open_thread
== curthread
||
1144 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1145 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1146 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1147 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1149 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1150 vd
->vdev_cant_read
= B_FALSE
;
1151 vd
->vdev_cant_write
= B_FALSE
;
1152 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1155 * If this vdev is not removed, check its fault status. If it's
1156 * faulted, bail out of the open.
1158 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1159 ASSERT(vd
->vdev_children
== 0);
1160 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1161 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1162 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1163 vd
->vdev_label_aux
);
1165 } else if (vd
->vdev_offline
) {
1166 ASSERT(vd
->vdev_children
== 0);
1167 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1171 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1174 * Reset the vdev_reopening flag so that we actually close
1175 * the vdev on error.
1177 vd
->vdev_reopening
= B_FALSE
;
1178 if (zio_injection_enabled
&& error
== 0)
1179 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1182 if (vd
->vdev_removed
&&
1183 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1184 vd
->vdev_removed
= B_FALSE
;
1186 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1187 vd
->vdev_stat
.vs_aux
);
1191 vd
->vdev_removed
= B_FALSE
;
1194 * Recheck the faulted flag now that we have confirmed that
1195 * the vdev is accessible. If we're faulted, bail.
1197 if (vd
->vdev_faulted
) {
1198 ASSERT(vd
->vdev_children
== 0);
1199 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1200 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1201 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1202 vd
->vdev_label_aux
);
1206 if (vd
->vdev_degraded
) {
1207 ASSERT(vd
->vdev_children
== 0);
1208 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1209 VDEV_AUX_ERR_EXCEEDED
);
1211 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1215 * For hole or missing vdevs we just return success.
1217 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1220 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1221 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1222 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1228 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1229 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1231 if (vd
->vdev_children
== 0) {
1232 if (osize
< SPA_MINDEVSIZE
) {
1233 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1234 VDEV_AUX_TOO_SMALL
);
1238 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1239 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1240 VDEV_LABEL_END_SIZE
);
1242 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1243 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1244 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1245 VDEV_AUX_TOO_SMALL
);
1250 max_asize
= max_osize
;
1253 vd
->vdev_psize
= psize
;
1256 * Make sure the allocatable size hasn't shrunk.
1258 if (asize
< vd
->vdev_min_asize
) {
1259 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1260 VDEV_AUX_BAD_LABEL
);
1264 if (vd
->vdev_asize
== 0) {
1266 * This is the first-ever open, so use the computed values.
1267 * For testing purposes, a higher ashift can be requested.
1269 vd
->vdev_asize
= asize
;
1270 vd
->vdev_max_asize
= max_asize
;
1271 vd
->vdev_ashift
= MAX(ashift
, vd
->vdev_ashift
);
1274 * Detect if the alignment requirement has increased.
1275 * We don't want to make the pool unavailable, just
1276 * post an event instead.
1278 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1279 vd
->vdev_ops
->vdev_op_leaf
) {
1280 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1281 spa
, vd
, NULL
, 0, 0);
1284 vd
->vdev_max_asize
= max_asize
;
1288 * If all children are healthy and the asize has increased,
1289 * then we've experienced dynamic LUN growth. If automatic
1290 * expansion is enabled then use the additional space.
1292 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1293 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1294 vd
->vdev_asize
= asize
;
1296 vdev_set_min_asize(vd
);
1299 * Ensure we can issue some IO before declaring the
1300 * vdev open for business.
1302 if (vd
->vdev_ops
->vdev_op_leaf
&&
1303 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1304 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1305 VDEV_AUX_ERR_EXCEEDED
);
1310 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1311 * resilver. But don't do this if we are doing a reopen for a scrub,
1312 * since this would just restart the scrub we are already doing.
1314 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1315 vdev_resilver_needed(vd
, NULL
, NULL
))
1316 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1322 * Called once the vdevs are all opened, this routine validates the label
1323 * contents. This needs to be done before vdev_load() so that we don't
1324 * inadvertently do repair I/Os to the wrong device.
1326 * If 'strict' is false ignore the spa guid check. This is necessary because
1327 * if the machine crashed during a re-guid the new guid might have been written
1328 * to all of the vdev labels, but not the cached config. The strict check
1329 * will be performed when the pool is opened again using the mos config.
1331 * This function will only return failure if one of the vdevs indicates that it
1332 * has since been destroyed or exported. This is only possible if
1333 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1334 * will be updated but the function will return 0.
1337 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1339 spa_t
*spa
= vd
->vdev_spa
;
1341 uint64_t guid
= 0, top_guid
;
1345 for (c
= 0; c
< vd
->vdev_children
; c
++)
1346 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1350 * If the device has already failed, or was marked offline, don't do
1351 * any further validation. Otherwise, label I/O will fail and we will
1352 * overwrite the previous state.
1354 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1355 uint64_t aux_guid
= 0;
1358 if ((label
= vdev_label_read_config(vd
)) == NULL
) {
1359 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1360 VDEV_AUX_BAD_LABEL
);
1365 * Determine if this vdev has been split off into another
1366 * pool. If so, then refuse to open it.
1368 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1369 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1370 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1371 VDEV_AUX_SPLIT_POOL
);
1376 if (strict
&& (nvlist_lookup_uint64(label
,
1377 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1378 guid
!= spa_guid(spa
))) {
1379 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1380 VDEV_AUX_CORRUPT_DATA
);
1385 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1386 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1391 * If this vdev just became a top-level vdev because its
1392 * sibling was detached, it will have adopted the parent's
1393 * vdev guid -- but the label may or may not be on disk yet.
1394 * Fortunately, either version of the label will have the
1395 * same top guid, so if we're a top-level vdev, we can
1396 * safely compare to that instead.
1398 * If we split this vdev off instead, then we also check the
1399 * original pool's guid. We don't want to consider the vdev
1400 * corrupt if it is partway through a split operation.
1402 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1404 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1406 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1407 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1408 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1409 VDEV_AUX_CORRUPT_DATA
);
1414 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1416 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1417 VDEV_AUX_CORRUPT_DATA
);
1425 * If this is a verbatim import, no need to check the
1426 * state of the pool.
1428 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1429 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1430 state
!= POOL_STATE_ACTIVE
)
1434 * If we were able to open and validate a vdev that was
1435 * previously marked permanently unavailable, clear that state
1438 if (vd
->vdev_not_present
)
1439 vd
->vdev_not_present
= 0;
1446 * Close a virtual device.
1449 vdev_close(vdev_t
*vd
)
1451 vdev_t
*pvd
= vd
->vdev_parent
;
1452 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1454 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1457 * If our parent is reopening, then we are as well, unless we are
1460 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1461 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1463 vd
->vdev_ops
->vdev_op_close(vd
);
1465 vdev_cache_purge(vd
);
1468 * We record the previous state before we close it, so that if we are
1469 * doing a reopen(), we don't generate FMA ereports if we notice that
1470 * it's still faulted.
1472 vd
->vdev_prevstate
= vd
->vdev_state
;
1474 if (vd
->vdev_offline
)
1475 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1477 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1478 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1482 vdev_hold(vdev_t
*vd
)
1484 spa_t
*spa
= vd
->vdev_spa
;
1487 ASSERT(spa_is_root(spa
));
1488 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1491 for (c
= 0; c
< vd
->vdev_children
; c
++)
1492 vdev_hold(vd
->vdev_child
[c
]);
1494 if (vd
->vdev_ops
->vdev_op_leaf
)
1495 vd
->vdev_ops
->vdev_op_hold(vd
);
1499 vdev_rele(vdev_t
*vd
)
1503 ASSERT(spa_is_root(vd
->vdev_spa
));
1504 for (c
= 0; c
< vd
->vdev_children
; c
++)
1505 vdev_rele(vd
->vdev_child
[c
]);
1507 if (vd
->vdev_ops
->vdev_op_leaf
)
1508 vd
->vdev_ops
->vdev_op_rele(vd
);
1512 * Reopen all interior vdevs and any unopened leaves. We don't actually
1513 * reopen leaf vdevs which had previously been opened as they might deadlock
1514 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1515 * If the leaf has never been opened then open it, as usual.
1518 vdev_reopen(vdev_t
*vd
)
1520 spa_t
*spa
= vd
->vdev_spa
;
1522 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1524 /* set the reopening flag unless we're taking the vdev offline */
1525 vd
->vdev_reopening
= !vd
->vdev_offline
;
1527 (void) vdev_open(vd
);
1530 * Call vdev_validate() here to make sure we have the same device.
1531 * Otherwise, a device with an invalid label could be successfully
1532 * opened in response to vdev_reopen().
1535 (void) vdev_validate_aux(vd
);
1536 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1537 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1538 !l2arc_vdev_present(vd
))
1539 l2arc_add_vdev(spa
, vd
);
1541 (void) vdev_validate(vd
, B_TRUE
);
1545 * Reassess parent vdev's health.
1547 vdev_propagate_state(vd
);
1551 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1556 * Normally, partial opens (e.g. of a mirror) are allowed.
1557 * For a create, however, we want to fail the request if
1558 * there are any components we can't open.
1560 error
= vdev_open(vd
);
1562 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1564 return (error
? error
: ENXIO
);
1568 * Recursively initialize all labels.
1570 if ((error
= vdev_label_init(vd
, txg
, isreplacing
?
1571 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1580 vdev_metaslab_set_size(vdev_t
*vd
)
1583 * Aim for roughly 200 metaslabs per vdev.
1585 vd
->vdev_ms_shift
= highbit(vd
->vdev_asize
/ 200);
1586 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1590 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1592 ASSERT(vd
== vd
->vdev_top
);
1593 ASSERT(!vd
->vdev_ishole
);
1594 ASSERT(ISP2(flags
));
1595 ASSERT(spa_writeable(vd
->vdev_spa
));
1597 if (flags
& VDD_METASLAB
)
1598 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1600 if (flags
& VDD_DTL
)
1601 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1603 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1609 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1610 * the vdev has less than perfect replication. There are four kinds of DTL:
1612 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1614 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1616 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1617 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1618 * txgs that was scrubbed.
1620 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1621 * persistent errors or just some device being offline.
1622 * Unlike the other three, the DTL_OUTAGE map is not generally
1623 * maintained; it's only computed when needed, typically to
1624 * determine whether a device can be detached.
1626 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1627 * either has the data or it doesn't.
1629 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1630 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1631 * if any child is less than fully replicated, then so is its parent.
1632 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1633 * comprising only those txgs which appear in 'maxfaults' or more children;
1634 * those are the txgs we don't have enough replication to read. For example,
1635 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1636 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1637 * two child DTL_MISSING maps.
1639 * It should be clear from the above that to compute the DTLs and outage maps
1640 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1641 * Therefore, that is all we keep on disk. When loading the pool, or after
1642 * a configuration change, we generate all other DTLs from first principles.
1645 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1647 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1649 ASSERT(t
< DTL_TYPES
);
1650 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1651 ASSERT(spa_writeable(vd
->vdev_spa
));
1653 mutex_enter(sm
->sm_lock
);
1654 if (!space_map_contains(sm
, txg
, size
))
1655 space_map_add(sm
, txg
, size
);
1656 mutex_exit(sm
->sm_lock
);
1660 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1662 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1663 boolean_t dirty
= B_FALSE
;
1665 ASSERT(t
< DTL_TYPES
);
1666 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1668 mutex_enter(sm
->sm_lock
);
1669 if (sm
->sm_space
!= 0)
1670 dirty
= space_map_contains(sm
, txg
, size
);
1671 mutex_exit(sm
->sm_lock
);
1677 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1679 space_map_t
*sm
= &vd
->vdev_dtl
[t
];
1682 mutex_enter(sm
->sm_lock
);
1683 empty
= (sm
->sm_space
== 0);
1684 mutex_exit(sm
->sm_lock
);
1690 * Reassess DTLs after a config change or scrub completion.
1693 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1695 spa_t
*spa
= vd
->vdev_spa
;
1699 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1701 for (c
= 0; c
< vd
->vdev_children
; c
++)
1702 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1703 scrub_txg
, scrub_done
);
1705 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1708 if (vd
->vdev_ops
->vdev_op_leaf
) {
1709 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1711 mutex_enter(&vd
->vdev_dtl_lock
);
1712 if (scrub_txg
!= 0 &&
1713 (spa
->spa_scrub_started
||
1714 (scn
&& scn
->scn_phys
.scn_errors
== 0))) {
1716 * We completed a scrub up to scrub_txg. If we
1717 * did it without rebooting, then the scrub dtl
1718 * will be valid, so excise the old region and
1719 * fold in the scrub dtl. Otherwise, leave the
1720 * dtl as-is if there was an error.
1722 * There's little trick here: to excise the beginning
1723 * of the DTL_MISSING map, we put it into a reference
1724 * tree and then add a segment with refcnt -1 that
1725 * covers the range [0, scrub_txg). This means
1726 * that each txg in that range has refcnt -1 or 0.
1727 * We then add DTL_SCRUB with a refcnt of 2, so that
1728 * entries in the range [0, scrub_txg) will have a
1729 * positive refcnt -- either 1 or 2. We then convert
1730 * the reference tree into the new DTL_MISSING map.
1732 space_map_ref_create(&reftree
);
1733 space_map_ref_add_map(&reftree
,
1734 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1735 space_map_ref_add_seg(&reftree
, 0, scrub_txg
, -1);
1736 space_map_ref_add_map(&reftree
,
1737 &vd
->vdev_dtl
[DTL_SCRUB
], 2);
1738 space_map_ref_generate_map(&reftree
,
1739 &vd
->vdev_dtl
[DTL_MISSING
], 1);
1740 space_map_ref_destroy(&reftree
);
1742 space_map_vacate(&vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1743 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1744 space_map_add
, &vd
->vdev_dtl
[DTL_PARTIAL
]);
1746 space_map_vacate(&vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1747 space_map_vacate(&vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1748 if (!vdev_readable(vd
))
1749 space_map_add(&vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1751 space_map_walk(&vd
->vdev_dtl
[DTL_MISSING
],
1752 space_map_add
, &vd
->vdev_dtl
[DTL_OUTAGE
]);
1753 mutex_exit(&vd
->vdev_dtl_lock
);
1756 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1760 mutex_enter(&vd
->vdev_dtl_lock
);
1761 for (t
= 0; t
< DTL_TYPES
; t
++) {
1762 /* account for child's outage in parent's missing map */
1763 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1765 continue; /* leaf vdevs only */
1766 if (t
== DTL_PARTIAL
)
1767 minref
= 1; /* i.e. non-zero */
1768 else if (vd
->vdev_nparity
!= 0)
1769 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1771 minref
= vd
->vdev_children
; /* any kind of mirror */
1772 space_map_ref_create(&reftree
);
1773 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1774 vdev_t
*cvd
= vd
->vdev_child
[c
];
1775 mutex_enter(&cvd
->vdev_dtl_lock
);
1776 space_map_ref_add_map(&reftree
, &cvd
->vdev_dtl
[s
], 1);
1777 mutex_exit(&cvd
->vdev_dtl_lock
);
1779 space_map_ref_generate_map(&reftree
, &vd
->vdev_dtl
[t
], minref
);
1780 space_map_ref_destroy(&reftree
);
1782 mutex_exit(&vd
->vdev_dtl_lock
);
1786 vdev_dtl_load(vdev_t
*vd
)
1788 spa_t
*spa
= vd
->vdev_spa
;
1789 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1790 objset_t
*mos
= spa
->spa_meta_objset
;
1794 ASSERT(vd
->vdev_children
== 0);
1796 if (smo
->smo_object
== 0)
1799 ASSERT(!vd
->vdev_ishole
);
1801 if ((error
= dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
)) != 0)
1804 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1805 bcopy(db
->db_data
, smo
, sizeof (*smo
));
1806 dmu_buf_rele(db
, FTAG
);
1808 mutex_enter(&vd
->vdev_dtl_lock
);
1809 error
= space_map_load(&vd
->vdev_dtl
[DTL_MISSING
],
1810 NULL
, SM_ALLOC
, smo
, mos
);
1811 mutex_exit(&vd
->vdev_dtl_lock
);
1817 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1819 spa_t
*spa
= vd
->vdev_spa
;
1820 space_map_obj_t
*smo
= &vd
->vdev_dtl_smo
;
1821 space_map_t
*sm
= &vd
->vdev_dtl
[DTL_MISSING
];
1822 objset_t
*mos
= spa
->spa_meta_objset
;
1828 ASSERT(!vd
->vdev_ishole
);
1830 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
1832 if (vd
->vdev_detached
) {
1833 if (smo
->smo_object
!= 0) {
1834 VERIFY(0 == dmu_object_free(mos
, smo
->smo_object
, tx
));
1835 smo
->smo_object
= 0;
1841 if (smo
->smo_object
== 0) {
1842 ASSERT(smo
->smo_objsize
== 0);
1843 ASSERT(smo
->smo_alloc
== 0);
1844 smo
->smo_object
= dmu_object_alloc(mos
,
1845 DMU_OT_SPACE_MAP
, 1 << SPACE_MAP_BLOCKSHIFT
,
1846 DMU_OT_SPACE_MAP_HEADER
, sizeof (*smo
), tx
);
1847 ASSERT(smo
->smo_object
!= 0);
1848 vdev_config_dirty(vd
->vdev_top
);
1851 mutex_init(&smlock
, NULL
, MUTEX_DEFAULT
, NULL
);
1853 space_map_create(&smsync
, sm
->sm_start
, sm
->sm_size
, sm
->sm_shift
,
1856 mutex_enter(&smlock
);
1858 mutex_enter(&vd
->vdev_dtl_lock
);
1859 space_map_walk(sm
, space_map_add
, &smsync
);
1860 mutex_exit(&vd
->vdev_dtl_lock
);
1862 space_map_truncate(smo
, mos
, tx
);
1863 space_map_sync(&smsync
, SM_ALLOC
, smo
, mos
, tx
);
1865 space_map_destroy(&smsync
);
1867 mutex_exit(&smlock
);
1868 mutex_destroy(&smlock
);
1870 VERIFY(0 == dmu_bonus_hold(mos
, smo
->smo_object
, FTAG
, &db
));
1871 dmu_buf_will_dirty(db
, tx
);
1872 ASSERT3U(db
->db_size
, >=, sizeof (*smo
));
1873 bcopy(smo
, db
->db_data
, sizeof (*smo
));
1874 dmu_buf_rele(db
, FTAG
);
1880 * Determine whether the specified vdev can be offlined/detached/removed
1881 * without losing data.
1884 vdev_dtl_required(vdev_t
*vd
)
1886 spa_t
*spa
= vd
->vdev_spa
;
1887 vdev_t
*tvd
= vd
->vdev_top
;
1888 uint8_t cant_read
= vd
->vdev_cant_read
;
1891 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1893 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
1897 * Temporarily mark the device as unreadable, and then determine
1898 * whether this results in any DTL outages in the top-level vdev.
1899 * If not, we can safely offline/detach/remove the device.
1901 vd
->vdev_cant_read
= B_TRUE
;
1902 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1903 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
1904 vd
->vdev_cant_read
= cant_read
;
1905 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
1907 if (!required
&& zio_injection_enabled
)
1908 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
1914 * Determine if resilver is needed, and if so the txg range.
1917 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
1919 boolean_t needed
= B_FALSE
;
1920 uint64_t thismin
= UINT64_MAX
;
1921 uint64_t thismax
= 0;
1924 if (vd
->vdev_children
== 0) {
1925 mutex_enter(&vd
->vdev_dtl_lock
);
1926 if (vd
->vdev_dtl
[DTL_MISSING
].sm_space
!= 0 &&
1927 vdev_writeable(vd
)) {
1930 ss
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1931 thismin
= ss
->ss_start
- 1;
1932 ss
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
].sm_root
);
1933 thismax
= ss
->ss_end
;
1936 mutex_exit(&vd
->vdev_dtl_lock
);
1938 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1939 vdev_t
*cvd
= vd
->vdev_child
[c
];
1940 uint64_t cmin
, cmax
;
1942 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
1943 thismin
= MIN(thismin
, cmin
);
1944 thismax
= MAX(thismax
, cmax
);
1950 if (needed
&& minp
) {
1958 vdev_load(vdev_t
*vd
)
1963 * Recursively load all children.
1965 for (c
= 0; c
< vd
->vdev_children
; c
++)
1966 vdev_load(vd
->vdev_child
[c
]);
1969 * If this is a top-level vdev, initialize its metaslabs.
1971 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
1972 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
1973 vdev_metaslab_init(vd
, 0) != 0))
1974 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1975 VDEV_AUX_CORRUPT_DATA
);
1978 * If this is a leaf vdev, load its DTL.
1980 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
1981 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1982 VDEV_AUX_CORRUPT_DATA
);
1986 * The special vdev case is used for hot spares and l2cache devices. Its
1987 * sole purpose it to set the vdev state for the associated vdev. To do this,
1988 * we make sure that we can open the underlying device, then try to read the
1989 * label, and make sure that the label is sane and that it hasn't been
1990 * repurposed to another pool.
1993 vdev_validate_aux(vdev_t
*vd
)
1996 uint64_t guid
, version
;
1999 if (!vdev_readable(vd
))
2002 if ((label
= vdev_label_read_config(vd
)) == NULL
) {
2003 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2004 VDEV_AUX_CORRUPT_DATA
);
2008 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2009 version
> SPA_VERSION
||
2010 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2011 guid
!= vd
->vdev_guid
||
2012 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2013 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2014 VDEV_AUX_CORRUPT_DATA
);
2020 * We don't actually check the pool state here. If it's in fact in
2021 * use by another pool, we update this fact on the fly when requested.
2028 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2030 spa_t
*spa
= vd
->vdev_spa
;
2031 objset_t
*mos
= spa
->spa_meta_objset
;
2035 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2037 if (vd
->vdev_dtl_smo
.smo_object
) {
2038 ASSERT3U(vd
->vdev_dtl_smo
.smo_alloc
, ==, 0);
2039 (void) dmu_object_free(mos
, vd
->vdev_dtl_smo
.smo_object
, tx
);
2040 vd
->vdev_dtl_smo
.smo_object
= 0;
2043 if (vd
->vdev_ms
!= NULL
) {
2044 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2045 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2047 if (msp
== NULL
|| msp
->ms_smo
.smo_object
== 0)
2050 ASSERT3U(msp
->ms_smo
.smo_alloc
, ==, 0);
2051 (void) dmu_object_free(mos
, msp
->ms_smo
.smo_object
, tx
);
2052 msp
->ms_smo
.smo_object
= 0;
2056 if (vd
->vdev_ms_array
) {
2057 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2058 vd
->vdev_ms_array
= 0;
2059 vd
->vdev_ms_shift
= 0;
2065 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2068 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2070 ASSERT(!vd
->vdev_ishole
);
2072 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2073 metaslab_sync_done(msp
, txg
);
2076 metaslab_sync_reassess(vd
->vdev_mg
);
2080 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2082 spa_t
*spa
= vd
->vdev_spa
;
2087 ASSERT(!vd
->vdev_ishole
);
2089 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2090 ASSERT(vd
== vd
->vdev_top
);
2091 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2092 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2093 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2094 ASSERT(vd
->vdev_ms_array
!= 0);
2095 vdev_config_dirty(vd
);
2100 * Remove the metadata associated with this vdev once it's empty.
2102 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2103 vdev_remove(vd
, txg
);
2105 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2106 metaslab_sync(msp
, txg
);
2107 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2110 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2111 vdev_dtl_sync(lvd
, txg
);
2113 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2117 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2119 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2123 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2124 * not be opened, and no I/O is attempted.
2127 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2131 spa_vdev_state_enter(spa
, SCL_NONE
);
2133 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2134 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2136 if (!vd
->vdev_ops
->vdev_op_leaf
)
2137 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2142 * We don't directly use the aux state here, but if we do a
2143 * vdev_reopen(), we need this value to be present to remember why we
2146 vd
->vdev_label_aux
= aux
;
2149 * Faulted state takes precedence over degraded.
2151 vd
->vdev_delayed_close
= B_FALSE
;
2152 vd
->vdev_faulted
= 1ULL;
2153 vd
->vdev_degraded
= 0ULL;
2154 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2157 * If this device has the only valid copy of the data, then
2158 * back off and simply mark the vdev as degraded instead.
2160 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2161 vd
->vdev_degraded
= 1ULL;
2162 vd
->vdev_faulted
= 0ULL;
2165 * If we reopen the device and it's not dead, only then do we
2170 if (vdev_readable(vd
))
2171 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2174 return (spa_vdev_state_exit(spa
, vd
, 0));
2178 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2179 * user that something is wrong. The vdev continues to operate as normal as far
2180 * as I/O is concerned.
2183 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2187 spa_vdev_state_enter(spa
, SCL_NONE
);
2189 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2190 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2192 if (!vd
->vdev_ops
->vdev_op_leaf
)
2193 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2196 * If the vdev is already faulted, then don't do anything.
2198 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2199 return (spa_vdev_state_exit(spa
, NULL
, 0));
2201 vd
->vdev_degraded
= 1ULL;
2202 if (!vdev_is_dead(vd
))
2203 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2206 return (spa_vdev_state_exit(spa
, vd
, 0));
2210 * Online the given vdev. If 'unspare' is set, it implies two things. First,
2211 * any attached spare device should be detached when the device finishes
2212 * resilvering. Second, the online should be treated like a 'test' online case,
2213 * so no FMA events are generated if the device fails to open.
2216 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2218 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2220 spa_vdev_state_enter(spa
, SCL_NONE
);
2222 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2223 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2225 if (!vd
->vdev_ops
->vdev_op_leaf
)
2226 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2229 vd
->vdev_offline
= B_FALSE
;
2230 vd
->vdev_tmpoffline
= B_FALSE
;
2231 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2232 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2234 /* XXX - L2ARC 1.0 does not support expansion */
2235 if (!vd
->vdev_aux
) {
2236 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2237 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2241 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2243 if (!vd
->vdev_aux
) {
2244 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2245 pvd
->vdev_expanding
= B_FALSE
;
2249 *newstate
= vd
->vdev_state
;
2250 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2251 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2252 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2253 vd
->vdev_parent
->vdev_child
[0] == vd
)
2254 vd
->vdev_unspare
= B_TRUE
;
2256 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2258 /* XXX - L2ARC 1.0 does not support expansion */
2260 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2261 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2263 return (spa_vdev_state_exit(spa
, vd
, 0));
2267 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2271 uint64_t generation
;
2272 metaslab_group_t
*mg
;
2275 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2277 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2278 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2280 if (!vd
->vdev_ops
->vdev_op_leaf
)
2281 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2285 generation
= spa
->spa_config_generation
+ 1;
2288 * If the device isn't already offline, try to offline it.
2290 if (!vd
->vdev_offline
) {
2292 * If this device has the only valid copy of some data,
2293 * don't allow it to be offlined. Log devices are always
2296 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2297 vdev_dtl_required(vd
))
2298 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2301 * If the top-level is a slog and it has had allocations
2302 * then proceed. We check that the vdev's metaslab group
2303 * is not NULL since it's possible that we may have just
2304 * added this vdev but not yet initialized its metaslabs.
2306 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2308 * Prevent any future allocations.
2310 metaslab_group_passivate(mg
);
2311 (void) spa_vdev_state_exit(spa
, vd
, 0);
2313 error
= spa_offline_log(spa
);
2315 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2318 * Check to see if the config has changed.
2320 if (error
|| generation
!= spa
->spa_config_generation
) {
2321 metaslab_group_activate(mg
);
2323 return (spa_vdev_state_exit(spa
,
2325 (void) spa_vdev_state_exit(spa
, vd
, 0);
2328 ASSERT3U(tvd
->vdev_stat
.vs_alloc
, ==, 0);
2332 * Offline this device and reopen its top-level vdev.
2333 * If the top-level vdev is a log device then just offline
2334 * it. Otherwise, if this action results in the top-level
2335 * vdev becoming unusable, undo it and fail the request.
2337 vd
->vdev_offline
= B_TRUE
;
2340 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2341 vdev_is_dead(tvd
)) {
2342 vd
->vdev_offline
= B_FALSE
;
2344 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2348 * Add the device back into the metaslab rotor so that
2349 * once we online the device it's open for business.
2351 if (tvd
->vdev_islog
&& mg
!= NULL
)
2352 metaslab_group_activate(mg
);
2355 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2357 return (spa_vdev_state_exit(spa
, vd
, 0));
2361 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2365 mutex_enter(&spa
->spa_vdev_top_lock
);
2366 error
= vdev_offline_locked(spa
, guid
, flags
);
2367 mutex_exit(&spa
->spa_vdev_top_lock
);
2373 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2374 * vdev_offline(), we assume the spa config is locked. We also clear all
2375 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2378 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2380 vdev_t
*rvd
= spa
->spa_root_vdev
;
2383 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2388 vd
->vdev_stat
.vs_read_errors
= 0;
2389 vd
->vdev_stat
.vs_write_errors
= 0;
2390 vd
->vdev_stat
.vs_checksum_errors
= 0;
2392 for (c
= 0; c
< vd
->vdev_children
; c
++)
2393 vdev_clear(spa
, vd
->vdev_child
[c
]);
2396 * If we're in the FAULTED state or have experienced failed I/O, then
2397 * clear the persistent state and attempt to reopen the device. We
2398 * also mark the vdev config dirty, so that the new faulted state is
2399 * written out to disk.
2401 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2402 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2405 * When reopening in reponse to a clear event, it may be due to
2406 * a fmadm repair request. In this case, if the device is
2407 * still broken, we want to still post the ereport again.
2409 vd
->vdev_forcefault
= B_TRUE
;
2411 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2412 vd
->vdev_cant_read
= B_FALSE
;
2413 vd
->vdev_cant_write
= B_FALSE
;
2415 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2417 vd
->vdev_forcefault
= B_FALSE
;
2419 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2420 vdev_state_dirty(vd
->vdev_top
);
2422 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2423 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2425 spa_event_notify(spa
, vd
, FM_EREPORT_ZFS_DEVICE_CLEAR
);
2429 * When clearing a FMA-diagnosed fault, we always want to
2430 * unspare the device, as we assume that the original spare was
2431 * done in response to the FMA fault.
2433 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2434 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2435 vd
->vdev_parent
->vdev_child
[0] == vd
)
2436 vd
->vdev_unspare
= B_TRUE
;
2440 vdev_is_dead(vdev_t
*vd
)
2443 * Holes and missing devices are always considered "dead".
2444 * This simplifies the code since we don't have to check for
2445 * these types of devices in the various code paths.
2446 * Instead we rely on the fact that we skip over dead devices
2447 * before issuing I/O to them.
2449 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2450 vd
->vdev_ops
== &vdev_missing_ops
);
2454 vdev_readable(vdev_t
*vd
)
2456 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2460 vdev_writeable(vdev_t
*vd
)
2462 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2466 vdev_allocatable(vdev_t
*vd
)
2468 uint64_t state
= vd
->vdev_state
;
2471 * We currently allow allocations from vdevs which may be in the
2472 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2473 * fails to reopen then we'll catch it later when we're holding
2474 * the proper locks. Note that we have to get the vdev state
2475 * in a local variable because although it changes atomically,
2476 * we're asking two separate questions about it.
2478 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2479 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2483 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2485 ASSERT(zio
->io_vd
== vd
);
2487 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2490 if (zio
->io_type
== ZIO_TYPE_READ
)
2491 return (!vd
->vdev_cant_read
);
2493 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2494 return (!vd
->vdev_cant_write
);
2500 * Get statistics for the given vdev.
2503 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2505 vdev_t
*rvd
= vd
->vdev_spa
->spa_root_vdev
;
2508 mutex_enter(&vd
->vdev_stat_lock
);
2509 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2510 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2511 vs
->vs_state
= vd
->vdev_state
;
2512 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2513 if (vd
->vdev_ops
->vdev_op_leaf
)
2514 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2515 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2516 mutex_exit(&vd
->vdev_stat_lock
);
2519 * If we're getting stats on the root vdev, aggregate the I/O counts
2520 * over all top-level vdevs (i.e. the direct children of the root).
2523 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2524 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2525 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2527 mutex_enter(&vd
->vdev_stat_lock
);
2528 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2529 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2530 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2532 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2533 mutex_exit(&vd
->vdev_stat_lock
);
2539 vdev_clear_stats(vdev_t
*vd
)
2541 mutex_enter(&vd
->vdev_stat_lock
);
2542 vd
->vdev_stat
.vs_space
= 0;
2543 vd
->vdev_stat
.vs_dspace
= 0;
2544 vd
->vdev_stat
.vs_alloc
= 0;
2545 mutex_exit(&vd
->vdev_stat_lock
);
2549 vdev_scan_stat_init(vdev_t
*vd
)
2551 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2554 for (c
= 0; c
< vd
->vdev_children
; c
++)
2555 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2557 mutex_enter(&vd
->vdev_stat_lock
);
2558 vs
->vs_scan_processed
= 0;
2559 mutex_exit(&vd
->vdev_stat_lock
);
2563 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2565 spa_t
*spa
= zio
->io_spa
;
2566 vdev_t
*rvd
= spa
->spa_root_vdev
;
2567 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2569 uint64_t txg
= zio
->io_txg
;
2570 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2571 zio_type_t type
= zio
->io_type
;
2572 int flags
= zio
->io_flags
;
2575 * If this i/o is a gang leader, it didn't do any actual work.
2577 if (zio
->io_gang_tree
)
2580 if (zio
->io_error
== 0) {
2582 * If this is a root i/o, don't count it -- we've already
2583 * counted the top-level vdevs, and vdev_get_stats() will
2584 * aggregate them when asked. This reduces contention on
2585 * the root vdev_stat_lock and implicitly handles blocks
2586 * that compress away to holes, for which there is no i/o.
2587 * (Holes never create vdev children, so all the counters
2588 * remain zero, which is what we want.)
2590 * Note: this only applies to successful i/o (io_error == 0)
2591 * because unlike i/o counts, errors are not additive.
2592 * When reading a ditto block, for example, failure of
2593 * one top-level vdev does not imply a root-level error.
2598 ASSERT(vd
== zio
->io_vd
);
2600 if (flags
& ZIO_FLAG_IO_BYPASS
)
2603 mutex_enter(&vd
->vdev_stat_lock
);
2605 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2606 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2607 dsl_scan_phys_t
*scn_phys
=
2608 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2609 uint64_t *processed
= &scn_phys
->scn_processed
;
2612 if (vd
->vdev_ops
->vdev_op_leaf
)
2613 atomic_add_64(processed
, psize
);
2614 vs
->vs_scan_processed
+= psize
;
2617 if (flags
& ZIO_FLAG_SELF_HEAL
)
2618 vs
->vs_self_healed
+= psize
;
2622 vs
->vs_bytes
[type
] += psize
;
2624 mutex_exit(&vd
->vdev_stat_lock
);
2628 if (flags
& ZIO_FLAG_SPECULATIVE
)
2632 * If this is an I/O error that is going to be retried, then ignore the
2633 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2634 * hard errors, when in reality they can happen for any number of
2635 * innocuous reasons (bus resets, MPxIO link failure, etc).
2637 if (zio
->io_error
== EIO
&&
2638 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2642 * Intent logs writes won't propagate their error to the root
2643 * I/O so don't mark these types of failures as pool-level
2646 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2649 mutex_enter(&vd
->vdev_stat_lock
);
2650 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2651 if (zio
->io_error
== ECKSUM
)
2652 vs
->vs_checksum_errors
++;
2654 vs
->vs_read_errors
++;
2656 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2657 vs
->vs_write_errors
++;
2658 mutex_exit(&vd
->vdev_stat_lock
);
2660 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2661 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2662 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2663 spa
->spa_claiming
)) {
2665 * This is either a normal write (not a repair), or it's
2666 * a repair induced by the scrub thread, or it's a repair
2667 * made by zil_claim() during spa_load() in the first txg.
2668 * In the normal case, we commit the DTL change in the same
2669 * txg as the block was born. In the scrub-induced repair
2670 * case, we know that scrubs run in first-pass syncing context,
2671 * so we commit the DTL change in spa_syncing_txg(spa).
2672 * In the zil_claim() case, we commit in spa_first_txg(spa).
2674 * We currently do not make DTL entries for failed spontaneous
2675 * self-healing writes triggered by normal (non-scrubbing)
2676 * reads, because we have no transactional context in which to
2677 * do so -- and it's not clear that it'd be desirable anyway.
2679 if (vd
->vdev_ops
->vdev_op_leaf
) {
2680 uint64_t commit_txg
= txg
;
2681 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2682 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2683 ASSERT(spa_sync_pass(spa
) == 1);
2684 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2685 commit_txg
= spa_syncing_txg(spa
);
2686 } else if (spa
->spa_claiming
) {
2687 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2688 commit_txg
= spa_first_txg(spa
);
2690 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2691 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2693 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2694 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2695 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2698 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2703 * Update the in-core space usage stats for this vdev, its metaslab class,
2704 * and the root vdev.
2707 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2708 int64_t space_delta
)
2710 int64_t dspace_delta
= space_delta
;
2711 spa_t
*spa
= vd
->vdev_spa
;
2712 vdev_t
*rvd
= spa
->spa_root_vdev
;
2713 metaslab_group_t
*mg
= vd
->vdev_mg
;
2714 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2716 ASSERT(vd
== vd
->vdev_top
);
2719 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2720 * factor. We must calculate this here and not at the root vdev
2721 * because the root vdev's psize-to-asize is simply the max of its
2722 * childrens', thus not accurate enough for us.
2724 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2725 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2726 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2727 vd
->vdev_deflate_ratio
;
2729 mutex_enter(&vd
->vdev_stat_lock
);
2730 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2731 vd
->vdev_stat
.vs_space
+= space_delta
;
2732 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2733 mutex_exit(&vd
->vdev_stat_lock
);
2735 if (mc
== spa_normal_class(spa
)) {
2736 mutex_enter(&rvd
->vdev_stat_lock
);
2737 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2738 rvd
->vdev_stat
.vs_space
+= space_delta
;
2739 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2740 mutex_exit(&rvd
->vdev_stat_lock
);
2744 ASSERT(rvd
== vd
->vdev_parent
);
2745 ASSERT(vd
->vdev_ms_count
!= 0);
2747 metaslab_class_space_update(mc
,
2748 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2753 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2754 * so that it will be written out next time the vdev configuration is synced.
2755 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2758 vdev_config_dirty(vdev_t
*vd
)
2760 spa_t
*spa
= vd
->vdev_spa
;
2761 vdev_t
*rvd
= spa
->spa_root_vdev
;
2764 ASSERT(spa_writeable(spa
));
2767 * If this is an aux vdev (as with l2cache and spare devices), then we
2768 * update the vdev config manually and set the sync flag.
2770 if (vd
->vdev_aux
!= NULL
) {
2771 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
2775 for (c
= 0; c
< sav
->sav_count
; c
++) {
2776 if (sav
->sav_vdevs
[c
] == vd
)
2780 if (c
== sav
->sav_count
) {
2782 * We're being removed. There's nothing more to do.
2784 ASSERT(sav
->sav_sync
== B_TRUE
);
2788 sav
->sav_sync
= B_TRUE
;
2790 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
2791 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
2792 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
2793 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
2799 * Setting the nvlist in the middle if the array is a little
2800 * sketchy, but it will work.
2802 nvlist_free(aux
[c
]);
2803 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
2809 * The dirty list is protected by the SCL_CONFIG lock. The caller
2810 * must either hold SCL_CONFIG as writer, or must be the sync thread
2811 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2812 * so this is sufficient to ensure mutual exclusion.
2814 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2815 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2816 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2819 for (c
= 0; c
< rvd
->vdev_children
; c
++)
2820 vdev_config_dirty(rvd
->vdev_child
[c
]);
2822 ASSERT(vd
== vd
->vdev_top
);
2824 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
2826 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
2831 vdev_config_clean(vdev_t
*vd
)
2833 spa_t
*spa
= vd
->vdev_spa
;
2835 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
2836 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2837 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
2839 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
2840 list_remove(&spa
->spa_config_dirty_list
, vd
);
2844 * Mark a top-level vdev's state as dirty, so that the next pass of
2845 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2846 * the state changes from larger config changes because they require
2847 * much less locking, and are often needed for administrative actions.
2850 vdev_state_dirty(vdev_t
*vd
)
2852 spa_t
*spa
= vd
->vdev_spa
;
2854 ASSERT(spa_writeable(spa
));
2855 ASSERT(vd
== vd
->vdev_top
);
2858 * The state list is protected by the SCL_STATE lock. The caller
2859 * must either hold SCL_STATE as writer, or must be the sync thread
2860 * (which holds SCL_STATE as reader). There's only one sync thread,
2861 * so this is sufficient to ensure mutual exclusion.
2863 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2864 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2865 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2867 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
2868 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
2872 vdev_state_clean(vdev_t
*vd
)
2874 spa_t
*spa
= vd
->vdev_spa
;
2876 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
2877 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
2878 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
2880 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
2881 list_remove(&spa
->spa_state_dirty_list
, vd
);
2885 * Propagate vdev state up from children to parent.
2888 vdev_propagate_state(vdev_t
*vd
)
2890 spa_t
*spa
= vd
->vdev_spa
;
2891 vdev_t
*rvd
= spa
->spa_root_vdev
;
2892 int degraded
= 0, faulted
= 0;
2897 if (vd
->vdev_children
> 0) {
2898 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2899 child
= vd
->vdev_child
[c
];
2902 * Don't factor holes into the decision.
2904 if (child
->vdev_ishole
)
2907 if (!vdev_readable(child
) ||
2908 (!vdev_writeable(child
) && spa_writeable(spa
))) {
2910 * Root special: if there is a top-level log
2911 * device, treat the root vdev as if it were
2914 if (child
->vdev_islog
&& vd
== rvd
)
2918 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
2922 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
2926 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
2929 * Root special: if there is a top-level vdev that cannot be
2930 * opened due to corrupted metadata, then propagate the root
2931 * vdev's aux state as 'corrupt' rather than 'insufficient
2934 if (corrupted
&& vd
== rvd
&&
2935 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
2936 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2937 VDEV_AUX_CORRUPT_DATA
);
2940 if (vd
->vdev_parent
)
2941 vdev_propagate_state(vd
->vdev_parent
);
2945 * Set a vdev's state. If this is during an open, we don't update the parent
2946 * state, because we're in the process of opening children depth-first.
2947 * Otherwise, we propagate the change to the parent.
2949 * If this routine places a device in a faulted state, an appropriate ereport is
2953 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
2955 uint64_t save_state
;
2956 spa_t
*spa
= vd
->vdev_spa
;
2958 if (state
== vd
->vdev_state
) {
2959 vd
->vdev_stat
.vs_aux
= aux
;
2963 save_state
= vd
->vdev_state
;
2965 vd
->vdev_state
= state
;
2966 vd
->vdev_stat
.vs_aux
= aux
;
2969 * If we are setting the vdev state to anything but an open state, then
2970 * always close the underlying device unless the device has requested
2971 * a delayed close (i.e. we're about to remove or fault the device).
2972 * Otherwise, we keep accessible but invalid devices open forever.
2973 * We don't call vdev_close() itself, because that implies some extra
2974 * checks (offline, etc) that we don't want here. This is limited to
2975 * leaf devices, because otherwise closing the device will affect other
2978 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
2979 vd
->vdev_ops
->vdev_op_leaf
)
2980 vd
->vdev_ops
->vdev_op_close(vd
);
2983 * If we have brought this vdev back into service, we need
2984 * to notify fmd so that it can gracefully repair any outstanding
2985 * cases due to a missing device. We do this in all cases, even those
2986 * that probably don't correlate to a repaired fault. This is sure to
2987 * catch all cases, and we let the zfs-retire agent sort it out. If
2988 * this is a transient state it's OK, as the retire agent will
2989 * double-check the state of the vdev before repairing it.
2991 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
2992 vd
->vdev_prevstate
!= state
)
2993 zfs_post_state_change(spa
, vd
);
2995 if (vd
->vdev_removed
&&
2996 state
== VDEV_STATE_CANT_OPEN
&&
2997 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
2999 * If the previous state is set to VDEV_STATE_REMOVED, then this
3000 * device was previously marked removed and someone attempted to
3001 * reopen it. If this failed due to a nonexistent device, then
3002 * keep the device in the REMOVED state. We also let this be if
3003 * it is one of our special test online cases, which is only
3004 * attempting to online the device and shouldn't generate an FMA
3007 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3008 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3009 } else if (state
== VDEV_STATE_REMOVED
) {
3010 vd
->vdev_removed
= B_TRUE
;
3011 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3013 * If we fail to open a vdev during an import or recovery, we
3014 * mark it as "not available", which signifies that it was
3015 * never there to begin with. Failure to open such a device
3016 * is not considered an error.
3018 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3019 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3020 vd
->vdev_ops
->vdev_op_leaf
)
3021 vd
->vdev_not_present
= 1;
3024 * Post the appropriate ereport. If the 'prevstate' field is
3025 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3026 * that this is part of a vdev_reopen(). In this case, we don't
3027 * want to post the ereport if the device was already in the
3028 * CANT_OPEN state beforehand.
3030 * If the 'checkremove' flag is set, then this is an attempt to
3031 * online the device in response to an insertion event. If we
3032 * hit this case, then we have detected an insertion event for a
3033 * faulted or offline device that wasn't in the removed state.
3034 * In this scenario, we don't post an ereport because we are
3035 * about to replace the device, or attempt an online with
3036 * vdev_forcefault, which will generate the fault for us.
3038 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3039 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3040 vd
!= spa
->spa_root_vdev
) {
3044 case VDEV_AUX_OPEN_FAILED
:
3045 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3047 case VDEV_AUX_CORRUPT_DATA
:
3048 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3050 case VDEV_AUX_NO_REPLICAS
:
3051 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3053 case VDEV_AUX_BAD_GUID_SUM
:
3054 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3056 case VDEV_AUX_TOO_SMALL
:
3057 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3059 case VDEV_AUX_BAD_LABEL
:
3060 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3063 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3066 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3069 /* Erase any notion of persistent removed state */
3070 vd
->vdev_removed
= B_FALSE
;
3072 vd
->vdev_removed
= B_FALSE
;
3075 if (!isopen
&& vd
->vdev_parent
)
3076 vdev_propagate_state(vd
->vdev_parent
);
3080 * Check the vdev configuration to ensure that it's capable of supporting
3084 vdev_is_bootable(vdev_t
*vd
)
3086 #if defined(__sun__) || defined(__sun)
3088 * Currently, we do not support RAID-Z or partial configuration.
3089 * In addition, only a single top-level vdev is allowed and none of the
3090 * leaves can be wholedisks.
3094 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3095 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3097 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3098 vd
->vdev_children
> 1) {
3100 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3101 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3104 } else if (vd
->vdev_wholedisk
== 1) {
3108 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3109 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3112 #endif /* __sun__ || __sun */
3117 * Load the state from the original vdev tree (ovd) which
3118 * we've retrieved from the MOS config object. If the original
3119 * vdev was offline or faulted then we transfer that state to the
3120 * device in the current vdev tree (nvd).
3123 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3127 ASSERT(nvd
->vdev_top
->vdev_islog
);
3128 ASSERT(spa_config_held(nvd
->vdev_spa
,
3129 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3130 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3132 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3133 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3135 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3137 * Restore the persistent vdev state
3139 nvd
->vdev_offline
= ovd
->vdev_offline
;
3140 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3141 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3142 nvd
->vdev_removed
= ovd
->vdev_removed
;
3147 * Determine if a log device has valid content. If the vdev was
3148 * removed or faulted in the MOS config then we know that
3149 * the content on the log device has already been written to the pool.
3152 vdev_log_state_valid(vdev_t
*vd
)
3156 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3160 for (c
= 0; c
< vd
->vdev_children
; c
++)
3161 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3168 * Expand a vdev if possible.
3171 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3173 ASSERT(vd
->vdev_top
== vd
);
3174 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3176 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3177 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3178 vdev_config_dirty(vd
);
3186 vdev_split(vdev_t
*vd
)
3188 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3190 vdev_remove_child(pvd
, vd
);
3191 vdev_compact_children(pvd
);
3193 cvd
= pvd
->vdev_child
[0];
3194 if (pvd
->vdev_children
== 1) {
3195 vdev_remove_parent(cvd
);
3196 cvd
->vdev_splitting
= B_TRUE
;
3198 vdev_propagate_state(cvd
);
3201 #if defined(_KERNEL) && defined(HAVE_SPL)
3202 EXPORT_SYMBOL(vdev_fault
);
3203 EXPORT_SYMBOL(vdev_degrade
);
3204 EXPORT_SYMBOL(vdev_online
);
3205 EXPORT_SYMBOL(vdev_offline
);
3206 EXPORT_SYMBOL(vdev_clear
);
3208 module_param(zfs_scrub_limit
, int, 0644);
3209 MODULE_PARM_DESC(zfs_scrub_limit
, "Max scrub/resilver I/O per leaf vdev");