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) 2011, 2015 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>
39 #include <sys/space_reftree.h>
42 #include <sys/fs/zfs.h>
45 #include <sys/dsl_scan.h>
47 #include <sys/zfs_ratelimit.h>
50 * When a vdev is added, it will be divided into approximately (but no
51 * more than) this number of metaslabs.
53 int metaslabs_per_vdev
= 200;
56 * Virtual device management.
59 static vdev_ops_t
*vdev_ops_table
[] = {
73 * Given a vdev type, return the appropriate ops vector.
76 vdev_getops(const char *type
)
78 vdev_ops_t
*ops
, **opspp
;
80 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
81 if (strcmp(ops
->vdev_op_type
, type
) == 0)
88 * Default asize function: return the MAX of psize with the asize of
89 * all children. This is what's used by anything other than RAID-Z.
92 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
94 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
98 for (c
= 0; c
< vd
->vdev_children
; c
++) {
99 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
100 asize
= MAX(asize
, csize
);
107 * Get the minimum allocatable size. We define the allocatable size as
108 * the vdev's asize rounded to the nearest metaslab. This allows us to
109 * replace or attach devices which don't have the same physical size but
110 * can still satisfy the same number of allocations.
113 vdev_get_min_asize(vdev_t
*vd
)
115 vdev_t
*pvd
= vd
->vdev_parent
;
118 * If our parent is NULL (inactive spare or cache) or is the root,
119 * just return our own asize.
122 return (vd
->vdev_asize
);
125 * The top-level vdev just returns the allocatable size rounded
126 * to the nearest metaslab.
128 if (vd
== vd
->vdev_top
)
129 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
132 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
133 * so each child must provide at least 1/Nth of its asize.
135 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
136 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
138 return (pvd
->vdev_min_asize
);
142 vdev_set_min_asize(vdev_t
*vd
)
145 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
147 for (c
= 0; c
< vd
->vdev_children
; c
++)
148 vdev_set_min_asize(vd
->vdev_child
[c
]);
152 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
154 vdev_t
*rvd
= spa
->spa_root_vdev
;
156 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
158 if (vdev
< rvd
->vdev_children
) {
159 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
160 return (rvd
->vdev_child
[vdev
]);
167 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
172 if (vd
->vdev_guid
== guid
)
175 for (c
= 0; c
< vd
->vdev_children
; c
++)
176 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
184 vdev_count_leaves_impl(vdev_t
*vd
)
189 if (vd
->vdev_ops
->vdev_op_leaf
)
192 for (c
= 0; c
< vd
->vdev_children
; c
++)
193 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
199 vdev_count_leaves(spa_t
*spa
)
201 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
205 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
207 size_t oldsize
, newsize
;
208 uint64_t id
= cvd
->vdev_id
;
211 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
212 ASSERT(cvd
->vdev_parent
== NULL
);
214 cvd
->vdev_parent
= pvd
;
219 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
221 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
222 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
223 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
225 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
226 if (pvd
->vdev_child
!= NULL
) {
227 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
228 kmem_free(pvd
->vdev_child
, oldsize
);
231 pvd
->vdev_child
= newchild
;
232 pvd
->vdev_child
[id
] = cvd
;
234 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
235 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
238 * Walk up all ancestors to update guid sum.
240 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
241 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
245 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
248 uint_t id
= cvd
->vdev_id
;
250 ASSERT(cvd
->vdev_parent
== pvd
);
255 ASSERT(id
< pvd
->vdev_children
);
256 ASSERT(pvd
->vdev_child
[id
] == cvd
);
258 pvd
->vdev_child
[id
] = NULL
;
259 cvd
->vdev_parent
= NULL
;
261 for (c
= 0; c
< pvd
->vdev_children
; c
++)
262 if (pvd
->vdev_child
[c
])
265 if (c
== pvd
->vdev_children
) {
266 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
267 pvd
->vdev_child
= NULL
;
268 pvd
->vdev_children
= 0;
272 * Walk up all ancestors to update guid sum.
274 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
275 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
279 * Remove any holes in the child array.
282 vdev_compact_children(vdev_t
*pvd
)
284 vdev_t
**newchild
, *cvd
;
285 int oldc
= pvd
->vdev_children
;
289 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
291 for (c
= newc
= 0; c
< oldc
; c
++)
292 if (pvd
->vdev_child
[c
])
295 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
297 for (c
= newc
= 0; c
< oldc
; c
++) {
298 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
299 newchild
[newc
] = cvd
;
300 cvd
->vdev_id
= newc
++;
304 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
305 pvd
->vdev_child
= newchild
;
306 pvd
->vdev_children
= newc
;
310 * Allocate and minimally initialize a vdev_t.
313 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
318 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
320 if (spa
->spa_root_vdev
== NULL
) {
321 ASSERT(ops
== &vdev_root_ops
);
322 spa
->spa_root_vdev
= vd
;
323 spa
->spa_load_guid
= spa_generate_guid(NULL
);
326 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
327 if (spa
->spa_root_vdev
== vd
) {
329 * The root vdev's guid will also be the pool guid,
330 * which must be unique among all pools.
332 guid
= spa_generate_guid(NULL
);
335 * Any other vdev's guid must be unique within the pool.
337 guid
= spa_generate_guid(spa
);
339 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
344 vd
->vdev_guid
= guid
;
345 vd
->vdev_guid_sum
= guid
;
347 vd
->vdev_state
= VDEV_STATE_CLOSED
;
348 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
351 * Initialize rate limit structs for events. We rate limit ZIO delay
352 * and checksum events so that we don't overwhelm ZED with thousands
353 * of events when a disk is acting up.
355 zfs_ratelimit_init(&vd
->vdev_delay_rl
, DELAYS_PER_SECOND
, 1);
356 zfs_ratelimit_init(&vd
->vdev_checksum_rl
, CHECKSUMS_PER_SECOND
, 1);
358 list_link_init(&vd
->vdev_config_dirty_node
);
359 list_link_init(&vd
->vdev_state_dirty_node
);
360 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
361 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
362 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
363 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
365 for (t
= 0; t
< DTL_TYPES
; t
++) {
366 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
369 txg_list_create(&vd
->vdev_ms_list
,
370 offsetof(struct metaslab
, ms_txg_node
));
371 txg_list_create(&vd
->vdev_dtl_list
,
372 offsetof(struct vdev
, vdev_dtl_node
));
373 vd
->vdev_stat
.vs_timestamp
= gethrtime();
381 * Allocate a new vdev. The 'alloctype' is used to control whether we are
382 * creating a new vdev or loading an existing one - the behavior is slightly
383 * different for each case.
386 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
391 uint64_t guid
= 0, islog
, nparity
;
394 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
396 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
397 return (SET_ERROR(EINVAL
));
399 if ((ops
= vdev_getops(type
)) == NULL
)
400 return (SET_ERROR(EINVAL
));
403 * If this is a load, get the vdev guid from the nvlist.
404 * Otherwise, vdev_alloc_common() will generate one for us.
406 if (alloctype
== VDEV_ALLOC_LOAD
) {
409 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
411 return (SET_ERROR(EINVAL
));
413 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
414 return (SET_ERROR(EINVAL
));
415 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
416 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
417 return (SET_ERROR(EINVAL
));
418 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
419 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
420 return (SET_ERROR(EINVAL
));
421 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
422 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
423 return (SET_ERROR(EINVAL
));
427 * The first allocated vdev must be of type 'root'.
429 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
430 return (SET_ERROR(EINVAL
));
433 * Determine whether we're a log vdev.
436 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
437 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
438 return (SET_ERROR(ENOTSUP
));
440 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
441 return (SET_ERROR(ENOTSUP
));
444 * Set the nparity property for RAID-Z vdevs.
447 if (ops
== &vdev_raidz_ops
) {
448 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
450 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
451 return (SET_ERROR(EINVAL
));
453 * Previous versions could only support 1 or 2 parity
457 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
458 return (SET_ERROR(ENOTSUP
));
460 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
461 return (SET_ERROR(ENOTSUP
));
464 * We require the parity to be specified for SPAs that
465 * support multiple parity levels.
467 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
468 return (SET_ERROR(EINVAL
));
470 * Otherwise, we default to 1 parity device for RAID-Z.
477 ASSERT(nparity
!= -1ULL);
479 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
481 vd
->vdev_islog
= islog
;
482 vd
->vdev_nparity
= nparity
;
484 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
485 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
486 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
487 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
488 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
489 &vd
->vdev_physpath
) == 0)
490 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
492 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
493 &vd
->vdev_enc_sysfs_path
) == 0)
494 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
496 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
497 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
500 * Set the whole_disk property. If it's not specified, leave the value
503 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
504 &vd
->vdev_wholedisk
) != 0)
505 vd
->vdev_wholedisk
= -1ULL;
508 * Look for the 'not present' flag. This will only be set if the device
509 * was not present at the time of import.
511 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
512 &vd
->vdev_not_present
);
515 * Get the alignment requirement.
517 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
520 * Retrieve the vdev creation time.
522 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
526 * If we're a top-level vdev, try to load the allocation parameters.
528 if (parent
&& !parent
->vdev_parent
&&
529 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
530 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
532 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
534 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
536 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
538 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
541 ASSERT0(vd
->vdev_top_zap
);
544 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
545 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
546 alloctype
== VDEV_ALLOC_ADD
||
547 alloctype
== VDEV_ALLOC_SPLIT
||
548 alloctype
== VDEV_ALLOC_ROOTPOOL
);
549 vd
->vdev_mg
= metaslab_group_create(islog
?
550 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
553 if (vd
->vdev_ops
->vdev_op_leaf
&&
554 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
555 (void) nvlist_lookup_uint64(nv
,
556 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
558 ASSERT0(vd
->vdev_leaf_zap
);
562 * If we're a leaf vdev, try to load the DTL object and other state.
565 if (vd
->vdev_ops
->vdev_op_leaf
&&
566 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
567 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
568 if (alloctype
== VDEV_ALLOC_LOAD
) {
569 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
570 &vd
->vdev_dtl_object
);
571 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
575 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
578 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
579 &spare
) == 0 && spare
)
583 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
586 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
587 &vd
->vdev_resilver_txg
);
590 * When importing a pool, we want to ignore the persistent fault
591 * state, as the diagnosis made on another system may not be
592 * valid in the current context. Local vdevs will
593 * remain in the faulted state.
595 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
596 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
598 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
600 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
603 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
607 VDEV_AUX_ERR_EXCEEDED
;
608 if (nvlist_lookup_string(nv
,
609 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
610 strcmp(aux
, "external") == 0)
611 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
617 * Add ourselves to the parent's list of children.
619 vdev_add_child(parent
, vd
);
627 vdev_free(vdev_t
*vd
)
630 spa_t
*spa
= vd
->vdev_spa
;
633 * vdev_free() implies closing the vdev first. This is simpler than
634 * trying to ensure complicated semantics for all callers.
638 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
639 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
644 for (c
= 0; c
< vd
->vdev_children
; c
++)
645 vdev_free(vd
->vdev_child
[c
]);
647 ASSERT(vd
->vdev_child
== NULL
);
648 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
651 * Discard allocation state.
653 if (vd
->vdev_mg
!= NULL
) {
654 vdev_metaslab_fini(vd
);
655 metaslab_group_destroy(vd
->vdev_mg
);
658 ASSERT0(vd
->vdev_stat
.vs_space
);
659 ASSERT0(vd
->vdev_stat
.vs_dspace
);
660 ASSERT0(vd
->vdev_stat
.vs_alloc
);
663 * Remove this vdev from its parent's child list.
665 vdev_remove_child(vd
->vdev_parent
, vd
);
667 ASSERT(vd
->vdev_parent
== NULL
);
670 * Clean up vdev structure.
676 spa_strfree(vd
->vdev_path
);
678 spa_strfree(vd
->vdev_devid
);
679 if (vd
->vdev_physpath
)
680 spa_strfree(vd
->vdev_physpath
);
682 if (vd
->vdev_enc_sysfs_path
)
683 spa_strfree(vd
->vdev_enc_sysfs_path
);
686 spa_strfree(vd
->vdev_fru
);
688 if (vd
->vdev_isspare
)
689 spa_spare_remove(vd
);
690 if (vd
->vdev_isl2cache
)
691 spa_l2cache_remove(vd
);
693 txg_list_destroy(&vd
->vdev_ms_list
);
694 txg_list_destroy(&vd
->vdev_dtl_list
);
696 mutex_enter(&vd
->vdev_dtl_lock
);
697 space_map_close(vd
->vdev_dtl_sm
);
698 for (t
= 0; t
< DTL_TYPES
; t
++) {
699 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
700 range_tree_destroy(vd
->vdev_dtl
[t
]);
702 mutex_exit(&vd
->vdev_dtl_lock
);
704 mutex_destroy(&vd
->vdev_queue_lock
);
705 mutex_destroy(&vd
->vdev_dtl_lock
);
706 mutex_destroy(&vd
->vdev_stat_lock
);
707 mutex_destroy(&vd
->vdev_probe_lock
);
709 if (vd
== spa
->spa_root_vdev
)
710 spa
->spa_root_vdev
= NULL
;
712 kmem_free(vd
, sizeof (vdev_t
));
716 * Transfer top-level vdev state from svd to tvd.
719 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
721 spa_t
*spa
= svd
->vdev_spa
;
726 ASSERT(tvd
== tvd
->vdev_top
);
728 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
729 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
730 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
731 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
732 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
734 svd
->vdev_ms_array
= 0;
735 svd
->vdev_ms_shift
= 0;
736 svd
->vdev_ms_count
= 0;
737 svd
->vdev_top_zap
= 0;
740 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
741 tvd
->vdev_mg
= svd
->vdev_mg
;
742 tvd
->vdev_ms
= svd
->vdev_ms
;
747 if (tvd
->vdev_mg
!= NULL
)
748 tvd
->vdev_mg
->mg_vd
= tvd
;
750 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
751 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
752 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
754 svd
->vdev_stat
.vs_alloc
= 0;
755 svd
->vdev_stat
.vs_space
= 0;
756 svd
->vdev_stat
.vs_dspace
= 0;
758 for (t
= 0; t
< TXG_SIZE
; t
++) {
759 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
760 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
761 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
762 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
763 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
764 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
767 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
768 vdev_config_clean(svd
);
769 vdev_config_dirty(tvd
);
772 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
773 vdev_state_clean(svd
);
774 vdev_state_dirty(tvd
);
777 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
778 svd
->vdev_deflate_ratio
= 0;
780 tvd
->vdev_islog
= svd
->vdev_islog
;
785 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
794 for (c
= 0; c
< vd
->vdev_children
; c
++)
795 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
799 * Add a mirror/replacing vdev above an existing vdev.
802 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
804 spa_t
*spa
= cvd
->vdev_spa
;
805 vdev_t
*pvd
= cvd
->vdev_parent
;
808 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
810 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
812 mvd
->vdev_asize
= cvd
->vdev_asize
;
813 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
814 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
815 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
816 mvd
->vdev_state
= cvd
->vdev_state
;
817 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
819 vdev_remove_child(pvd
, cvd
);
820 vdev_add_child(pvd
, mvd
);
821 cvd
->vdev_id
= mvd
->vdev_children
;
822 vdev_add_child(mvd
, cvd
);
823 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
825 if (mvd
== mvd
->vdev_top
)
826 vdev_top_transfer(cvd
, mvd
);
832 * Remove a 1-way mirror/replacing vdev from the tree.
835 vdev_remove_parent(vdev_t
*cvd
)
837 vdev_t
*mvd
= cvd
->vdev_parent
;
838 vdev_t
*pvd
= mvd
->vdev_parent
;
840 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
842 ASSERT(mvd
->vdev_children
== 1);
843 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
844 mvd
->vdev_ops
== &vdev_replacing_ops
||
845 mvd
->vdev_ops
== &vdev_spare_ops
);
846 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
848 vdev_remove_child(mvd
, cvd
);
849 vdev_remove_child(pvd
, mvd
);
852 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
853 * Otherwise, we could have detached an offline device, and when we
854 * go to import the pool we'll think we have two top-level vdevs,
855 * instead of a different version of the same top-level vdev.
857 if (mvd
->vdev_top
== mvd
) {
858 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
859 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
860 cvd
->vdev_guid
+= guid_delta
;
861 cvd
->vdev_guid_sum
+= guid_delta
;
864 * If pool not set for autoexpand, we need to also preserve
865 * mvd's asize to prevent automatic expansion of cvd.
866 * Otherwise if we are adjusting the mirror by attaching and
867 * detaching children of non-uniform sizes, the mirror could
868 * autoexpand, unexpectedly requiring larger devices to
869 * re-establish the mirror.
871 if (!cvd
->vdev_spa
->spa_autoexpand
)
872 cvd
->vdev_asize
= mvd
->vdev_asize
;
874 cvd
->vdev_id
= mvd
->vdev_id
;
875 vdev_add_child(pvd
, cvd
);
876 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
878 if (cvd
== cvd
->vdev_top
)
879 vdev_top_transfer(mvd
, cvd
);
881 ASSERT(mvd
->vdev_children
== 0);
886 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
888 spa_t
*spa
= vd
->vdev_spa
;
889 objset_t
*mos
= spa
->spa_meta_objset
;
891 uint64_t oldc
= vd
->vdev_ms_count
;
892 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
896 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
899 * This vdev is not being allocated from yet or is a hole.
901 if (vd
->vdev_ms_shift
== 0)
904 ASSERT(!vd
->vdev_ishole
);
907 * Compute the raidz-deflation ratio. Note, we hard-code
908 * in 128k (1 << 17) because it is the "typical" blocksize.
909 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
910 * otherwise it would inconsistently account for existing bp's.
912 vd
->vdev_deflate_ratio
= (1 << 17) /
913 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
915 ASSERT(oldc
<= newc
);
917 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
920 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
921 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
925 vd
->vdev_ms_count
= newc
;
927 for (m
= oldc
; m
< newc
; m
++) {
931 error
= dmu_read(mos
, vd
->vdev_ms_array
,
932 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
938 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
945 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
948 * If the vdev is being removed we don't activate
949 * the metaslabs since we want to ensure that no new
950 * allocations are performed on this device.
952 if (oldc
== 0 && !vd
->vdev_removing
)
953 metaslab_group_activate(vd
->vdev_mg
);
956 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
962 vdev_metaslab_fini(vdev_t
*vd
)
965 uint64_t count
= vd
->vdev_ms_count
;
967 if (vd
->vdev_ms
!= NULL
) {
968 metaslab_group_passivate(vd
->vdev_mg
);
969 for (m
= 0; m
< count
; m
++) {
970 metaslab_t
*msp
= vd
->vdev_ms
[m
];
975 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
979 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
982 typedef struct vdev_probe_stats
{
983 boolean_t vps_readable
;
984 boolean_t vps_writeable
;
986 } vdev_probe_stats_t
;
989 vdev_probe_done(zio_t
*zio
)
991 spa_t
*spa
= zio
->io_spa
;
992 vdev_t
*vd
= zio
->io_vd
;
993 vdev_probe_stats_t
*vps
= zio
->io_private
;
995 ASSERT(vd
->vdev_probe_zio
!= NULL
);
997 if (zio
->io_type
== ZIO_TYPE_READ
) {
998 if (zio
->io_error
== 0)
999 vps
->vps_readable
= 1;
1000 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1001 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1002 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
1003 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1004 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1006 zio_buf_free(zio
->io_data
, zio
->io_size
);
1008 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1009 if (zio
->io_error
== 0)
1010 vps
->vps_writeable
= 1;
1011 zio_buf_free(zio
->io_data
, zio
->io_size
);
1012 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1016 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1017 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1019 if (vdev_readable(vd
) &&
1020 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1023 ASSERT(zio
->io_error
!= 0);
1024 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1025 spa
, vd
, NULL
, 0, 0);
1026 zio
->io_error
= SET_ERROR(ENXIO
);
1029 mutex_enter(&vd
->vdev_probe_lock
);
1030 ASSERT(vd
->vdev_probe_zio
== zio
);
1031 vd
->vdev_probe_zio
= NULL
;
1032 mutex_exit(&vd
->vdev_probe_lock
);
1035 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1036 if (!vdev_accessible(vd
, pio
))
1037 pio
->io_error
= SET_ERROR(ENXIO
);
1039 kmem_free(vps
, sizeof (*vps
));
1044 * Determine whether this device is accessible.
1046 * Read and write to several known locations: the pad regions of each
1047 * vdev label but the first, which we leave alone in case it contains
1051 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1053 spa_t
*spa
= vd
->vdev_spa
;
1054 vdev_probe_stats_t
*vps
= NULL
;
1058 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1061 * Don't probe the probe.
1063 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1067 * To prevent 'probe storms' when a device fails, we create
1068 * just one probe i/o at a time. All zios that want to probe
1069 * this vdev will become parents of the probe io.
1071 mutex_enter(&vd
->vdev_probe_lock
);
1073 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1074 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1076 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1077 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1080 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1082 * vdev_cant_read and vdev_cant_write can only
1083 * transition from TRUE to FALSE when we have the
1084 * SCL_ZIO lock as writer; otherwise they can only
1085 * transition from FALSE to TRUE. This ensures that
1086 * any zio looking at these values can assume that
1087 * failures persist for the life of the I/O. That's
1088 * important because when a device has intermittent
1089 * connectivity problems, we want to ensure that
1090 * they're ascribed to the device (ENXIO) and not
1093 * Since we hold SCL_ZIO as writer here, clear both
1094 * values so the probe can reevaluate from first
1097 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1098 vd
->vdev_cant_read
= B_FALSE
;
1099 vd
->vdev_cant_write
= B_FALSE
;
1102 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1103 vdev_probe_done
, vps
,
1104 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1107 * We can't change the vdev state in this context, so we
1108 * kick off an async task to do it on our behalf.
1111 vd
->vdev_probe_wanted
= B_TRUE
;
1112 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1117 zio_add_child(zio
, pio
);
1119 mutex_exit(&vd
->vdev_probe_lock
);
1122 ASSERT(zio
!= NULL
);
1126 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1127 zio_nowait(zio_read_phys(pio
, vd
,
1128 vdev_label_offset(vd
->vdev_psize
, l
,
1129 offsetof(vdev_label_t
, vl_pad2
)),
1130 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1131 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1132 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1143 vdev_open_child(void *arg
)
1147 vd
->vdev_open_thread
= curthread
;
1148 vd
->vdev_open_error
= vdev_open(vd
);
1149 vd
->vdev_open_thread
= NULL
;
1153 vdev_uses_zvols(vdev_t
*vd
)
1158 if (zvol_is_zvol(vd
->vdev_path
))
1162 for (c
= 0; c
< vd
->vdev_children
; c
++)
1163 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1170 vdev_open_children(vdev_t
*vd
)
1173 int children
= vd
->vdev_children
;
1177 * in order to handle pools on top of zvols, do the opens
1178 * in a single thread so that the same thread holds the
1179 * spa_namespace_lock
1181 if (vdev_uses_zvols(vd
)) {
1182 for (c
= 0; c
< children
; c
++)
1183 vd
->vdev_child
[c
]->vdev_open_error
=
1184 vdev_open(vd
->vdev_child
[c
]);
1186 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1187 children
, children
, TASKQ_PREPOPULATE
);
1189 for (c
= 0; c
< children
; c
++)
1190 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1191 vd
->vdev_child
[c
], TQ_SLEEP
) != 0);
1196 vd
->vdev_nonrot
= B_TRUE
;
1198 for (c
= 0; c
< children
; c
++)
1199 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1203 * Prepare a virtual device for access.
1206 vdev_open(vdev_t
*vd
)
1208 spa_t
*spa
= vd
->vdev_spa
;
1211 uint64_t max_osize
= 0;
1212 uint64_t asize
, max_asize
, psize
;
1213 uint64_t ashift
= 0;
1216 ASSERT(vd
->vdev_open_thread
== curthread
||
1217 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1218 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1219 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1220 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1222 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1223 vd
->vdev_cant_read
= B_FALSE
;
1224 vd
->vdev_cant_write
= B_FALSE
;
1225 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1228 * If this vdev is not removed, check its fault status. If it's
1229 * faulted, bail out of the open.
1231 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1232 ASSERT(vd
->vdev_children
== 0);
1233 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1234 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1235 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1236 vd
->vdev_label_aux
);
1237 return (SET_ERROR(ENXIO
));
1238 } else if (vd
->vdev_offline
) {
1239 ASSERT(vd
->vdev_children
== 0);
1240 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1241 return (SET_ERROR(ENXIO
));
1244 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1247 * Reset the vdev_reopening flag so that we actually close
1248 * the vdev on error.
1250 vd
->vdev_reopening
= B_FALSE
;
1251 if (zio_injection_enabled
&& error
== 0)
1252 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1255 if (vd
->vdev_removed
&&
1256 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1257 vd
->vdev_removed
= B_FALSE
;
1259 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1260 vd
->vdev_stat
.vs_aux
);
1264 vd
->vdev_removed
= B_FALSE
;
1267 * Recheck the faulted flag now that we have confirmed that
1268 * the vdev is accessible. If we're faulted, bail.
1270 if (vd
->vdev_faulted
) {
1271 ASSERT(vd
->vdev_children
== 0);
1272 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1273 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1274 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1275 vd
->vdev_label_aux
);
1276 return (SET_ERROR(ENXIO
));
1279 if (vd
->vdev_degraded
) {
1280 ASSERT(vd
->vdev_children
== 0);
1281 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1282 VDEV_AUX_ERR_EXCEEDED
);
1284 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1288 * For hole or missing vdevs we just return success.
1290 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1293 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1294 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1295 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1301 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1302 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1304 if (vd
->vdev_children
== 0) {
1305 if (osize
< SPA_MINDEVSIZE
) {
1306 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1307 VDEV_AUX_TOO_SMALL
);
1308 return (SET_ERROR(EOVERFLOW
));
1311 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1312 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1313 VDEV_LABEL_END_SIZE
);
1315 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1316 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1317 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1318 VDEV_AUX_TOO_SMALL
);
1319 return (SET_ERROR(EOVERFLOW
));
1323 max_asize
= max_osize
;
1326 vd
->vdev_psize
= psize
;
1329 * Make sure the allocatable size hasn't shrunk.
1331 if (asize
< vd
->vdev_min_asize
) {
1332 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1333 VDEV_AUX_BAD_LABEL
);
1334 return (SET_ERROR(EINVAL
));
1337 if (vd
->vdev_asize
== 0) {
1339 * This is the first-ever open, so use the computed values.
1340 * For compatibility, a different ashift can be requested.
1342 vd
->vdev_asize
= asize
;
1343 vd
->vdev_max_asize
= max_asize
;
1344 if (vd
->vdev_ashift
== 0)
1345 vd
->vdev_ashift
= ashift
;
1348 * Detect if the alignment requirement has increased.
1349 * We don't want to make the pool unavailable, just
1350 * post an event instead.
1352 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1353 vd
->vdev_ops
->vdev_op_leaf
) {
1354 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1355 spa
, vd
, NULL
, 0, 0);
1358 vd
->vdev_max_asize
= max_asize
;
1362 * If all children are healthy and the asize has increased,
1363 * then we've experienced dynamic LUN growth. If automatic
1364 * expansion is enabled then use the additional space.
1366 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1367 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1368 vd
->vdev_asize
= asize
;
1370 vdev_set_min_asize(vd
);
1373 * Ensure we can issue some IO before declaring the
1374 * vdev open for business.
1376 if (vd
->vdev_ops
->vdev_op_leaf
&&
1377 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1378 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1379 VDEV_AUX_ERR_EXCEEDED
);
1384 * Track the min and max ashift values for normal data devices.
1386 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1387 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1388 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1389 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1390 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1391 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1395 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1396 * resilver. But don't do this if we are doing a reopen for a scrub,
1397 * since this would just restart the scrub we are already doing.
1399 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1400 vdev_resilver_needed(vd
, NULL
, NULL
))
1401 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1407 * Called once the vdevs are all opened, this routine validates the label
1408 * contents. This needs to be done before vdev_load() so that we don't
1409 * inadvertently do repair I/Os to the wrong device.
1411 * If 'strict' is false ignore the spa guid check. This is necessary because
1412 * if the machine crashed during a re-guid the new guid might have been written
1413 * to all of the vdev labels, but not the cached config. The strict check
1414 * will be performed when the pool is opened again using the mos config.
1416 * This function will only return failure if one of the vdevs indicates that it
1417 * has since been destroyed or exported. This is only possible if
1418 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1419 * will be updated but the function will return 0.
1422 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1424 spa_t
*spa
= vd
->vdev_spa
;
1426 uint64_t guid
= 0, top_guid
;
1430 for (c
= 0; c
< vd
->vdev_children
; c
++)
1431 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1432 return (SET_ERROR(EBADF
));
1435 * If the device has already failed, or was marked offline, don't do
1436 * any further validation. Otherwise, label I/O will fail and we will
1437 * overwrite the previous state.
1439 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1440 uint64_t aux_guid
= 0;
1442 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1443 spa_last_synced_txg(spa
) : -1ULL;
1445 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1446 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1447 VDEV_AUX_BAD_LABEL
);
1452 * Determine if this vdev has been split off into another
1453 * pool. If so, then refuse to open it.
1455 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1456 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1457 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1458 VDEV_AUX_SPLIT_POOL
);
1463 if (strict
&& (nvlist_lookup_uint64(label
,
1464 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1465 guid
!= spa_guid(spa
))) {
1466 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1467 VDEV_AUX_CORRUPT_DATA
);
1472 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1473 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1478 * If this vdev just became a top-level vdev because its
1479 * sibling was detached, it will have adopted the parent's
1480 * vdev guid -- but the label may or may not be on disk yet.
1481 * Fortunately, either version of the label will have the
1482 * same top guid, so if we're a top-level vdev, we can
1483 * safely compare to that instead.
1485 * If we split this vdev off instead, then we also check the
1486 * original pool's guid. We don't want to consider the vdev
1487 * corrupt if it is partway through a split operation.
1489 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1491 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1493 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1494 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1495 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1496 VDEV_AUX_CORRUPT_DATA
);
1501 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1503 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1504 VDEV_AUX_CORRUPT_DATA
);
1512 * If this is a verbatim import, no need to check the
1513 * state of the pool.
1515 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1516 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1517 state
!= POOL_STATE_ACTIVE
)
1518 return (SET_ERROR(EBADF
));
1521 * If we were able to open and validate a vdev that was
1522 * previously marked permanently unavailable, clear that state
1525 if (vd
->vdev_not_present
)
1526 vd
->vdev_not_present
= 0;
1533 * Close a virtual device.
1536 vdev_close(vdev_t
*vd
)
1538 vdev_t
*pvd
= vd
->vdev_parent
;
1539 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1541 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1544 * If our parent is reopening, then we are as well, unless we are
1547 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1548 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1550 vd
->vdev_ops
->vdev_op_close(vd
);
1552 vdev_cache_purge(vd
);
1555 * We record the previous state before we close it, so that if we are
1556 * doing a reopen(), we don't generate FMA ereports if we notice that
1557 * it's still faulted.
1559 vd
->vdev_prevstate
= vd
->vdev_state
;
1561 if (vd
->vdev_offline
)
1562 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1564 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1565 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1569 vdev_hold(vdev_t
*vd
)
1571 spa_t
*spa
= vd
->vdev_spa
;
1574 ASSERT(spa_is_root(spa
));
1575 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1578 for (c
= 0; c
< vd
->vdev_children
; c
++)
1579 vdev_hold(vd
->vdev_child
[c
]);
1581 if (vd
->vdev_ops
->vdev_op_leaf
)
1582 vd
->vdev_ops
->vdev_op_hold(vd
);
1586 vdev_rele(vdev_t
*vd
)
1590 ASSERT(spa_is_root(vd
->vdev_spa
));
1591 for (c
= 0; c
< vd
->vdev_children
; c
++)
1592 vdev_rele(vd
->vdev_child
[c
]);
1594 if (vd
->vdev_ops
->vdev_op_leaf
)
1595 vd
->vdev_ops
->vdev_op_rele(vd
);
1599 * Reopen all interior vdevs and any unopened leaves. We don't actually
1600 * reopen leaf vdevs which had previously been opened as they might deadlock
1601 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1602 * If the leaf has never been opened then open it, as usual.
1605 vdev_reopen(vdev_t
*vd
)
1607 spa_t
*spa
= vd
->vdev_spa
;
1609 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1611 /* set the reopening flag unless we're taking the vdev offline */
1612 vd
->vdev_reopening
= !vd
->vdev_offline
;
1614 (void) vdev_open(vd
);
1617 * Call vdev_validate() here to make sure we have the same device.
1618 * Otherwise, a device with an invalid label could be successfully
1619 * opened in response to vdev_reopen().
1622 (void) vdev_validate_aux(vd
);
1623 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1624 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1625 !l2arc_vdev_present(vd
))
1626 l2arc_add_vdev(spa
, vd
);
1628 (void) vdev_validate(vd
, B_TRUE
);
1632 * Reassess parent vdev's health.
1634 vdev_propagate_state(vd
);
1638 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1643 * Normally, partial opens (e.g. of a mirror) are allowed.
1644 * For a create, however, we want to fail the request if
1645 * there are any components we can't open.
1647 error
= vdev_open(vd
);
1649 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1651 return (error
? error
: ENXIO
);
1655 * Recursively load DTLs and initialize all labels.
1657 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1658 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1659 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1668 vdev_metaslab_set_size(vdev_t
*vd
)
1671 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1673 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1674 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1678 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1680 ASSERT(vd
== vd
->vdev_top
);
1681 ASSERT(!vd
->vdev_ishole
);
1682 ASSERT(ISP2(flags
));
1683 ASSERT(spa_writeable(vd
->vdev_spa
));
1685 if (flags
& VDD_METASLAB
)
1686 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1688 if (flags
& VDD_DTL
)
1689 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1691 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1695 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1699 for (c
= 0; c
< vd
->vdev_children
; c
++)
1700 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1702 if (vd
->vdev_ops
->vdev_op_leaf
)
1703 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1709 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1710 * the vdev has less than perfect replication. There are four kinds of DTL:
1712 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1714 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1716 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1717 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1718 * txgs that was scrubbed.
1720 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1721 * persistent errors or just some device being offline.
1722 * Unlike the other three, the DTL_OUTAGE map is not generally
1723 * maintained; it's only computed when needed, typically to
1724 * determine whether a device can be detached.
1726 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1727 * either has the data or it doesn't.
1729 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1730 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1731 * if any child is less than fully replicated, then so is its parent.
1732 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1733 * comprising only those txgs which appear in 'maxfaults' or more children;
1734 * those are the txgs we don't have enough replication to read. For example,
1735 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1736 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1737 * two child DTL_MISSING maps.
1739 * It should be clear from the above that to compute the DTLs and outage maps
1740 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1741 * Therefore, that is all we keep on disk. When loading the pool, or after
1742 * a configuration change, we generate all other DTLs from first principles.
1745 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1747 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1749 ASSERT(t
< DTL_TYPES
);
1750 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1751 ASSERT(spa_writeable(vd
->vdev_spa
));
1753 mutex_enter(rt
->rt_lock
);
1754 if (!range_tree_contains(rt
, txg
, size
))
1755 range_tree_add(rt
, txg
, size
);
1756 mutex_exit(rt
->rt_lock
);
1760 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1762 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1763 boolean_t dirty
= B_FALSE
;
1765 ASSERT(t
< DTL_TYPES
);
1766 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1768 mutex_enter(rt
->rt_lock
);
1769 if (range_tree_space(rt
) != 0)
1770 dirty
= range_tree_contains(rt
, txg
, size
);
1771 mutex_exit(rt
->rt_lock
);
1777 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1779 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1782 mutex_enter(rt
->rt_lock
);
1783 empty
= (range_tree_space(rt
) == 0);
1784 mutex_exit(rt
->rt_lock
);
1790 * Returns the lowest txg in the DTL range.
1793 vdev_dtl_min(vdev_t
*vd
)
1797 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1798 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1799 ASSERT0(vd
->vdev_children
);
1801 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1802 return (rs
->rs_start
- 1);
1806 * Returns the highest txg in the DTL.
1809 vdev_dtl_max(vdev_t
*vd
)
1813 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1814 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1815 ASSERT0(vd
->vdev_children
);
1817 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1818 return (rs
->rs_end
);
1822 * Determine if a resilvering vdev should remove any DTL entries from
1823 * its range. If the vdev was resilvering for the entire duration of the
1824 * scan then it should excise that range from its DTLs. Otherwise, this
1825 * vdev is considered partially resilvered and should leave its DTL
1826 * entries intact. The comment in vdev_dtl_reassess() describes how we
1830 vdev_dtl_should_excise(vdev_t
*vd
)
1832 spa_t
*spa
= vd
->vdev_spa
;
1833 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1835 ASSERT0(scn
->scn_phys
.scn_errors
);
1836 ASSERT0(vd
->vdev_children
);
1838 if (vd
->vdev_resilver_txg
== 0 ||
1839 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1843 * When a resilver is initiated the scan will assign the scn_max_txg
1844 * value to the highest txg value that exists in all DTLs. If this
1845 * device's max DTL is not part of this scan (i.e. it is not in
1846 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1849 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1850 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1851 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1852 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1859 * Reassess DTLs after a config change or scrub completion.
1862 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1864 spa_t
*spa
= vd
->vdev_spa
;
1868 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1870 for (c
= 0; c
< vd
->vdev_children
; c
++)
1871 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1872 scrub_txg
, scrub_done
);
1874 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1877 if (vd
->vdev_ops
->vdev_op_leaf
) {
1878 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1880 mutex_enter(&vd
->vdev_dtl_lock
);
1883 * If we've completed a scan cleanly then determine
1884 * if this vdev should remove any DTLs. We only want to
1885 * excise regions on vdevs that were available during
1886 * the entire duration of this scan.
1888 if (scrub_txg
!= 0 &&
1889 (spa
->spa_scrub_started
||
1890 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1891 vdev_dtl_should_excise(vd
)) {
1893 * We completed a scrub up to scrub_txg. If we
1894 * did it without rebooting, then the scrub dtl
1895 * will be valid, so excise the old region and
1896 * fold in the scrub dtl. Otherwise, leave the
1897 * dtl as-is if there was an error.
1899 * There's little trick here: to excise the beginning
1900 * of the DTL_MISSING map, we put it into a reference
1901 * tree and then add a segment with refcnt -1 that
1902 * covers the range [0, scrub_txg). This means
1903 * that each txg in that range has refcnt -1 or 0.
1904 * We then add DTL_SCRUB with a refcnt of 2, so that
1905 * entries in the range [0, scrub_txg) will have a
1906 * positive refcnt -- either 1 or 2. We then convert
1907 * the reference tree into the new DTL_MISSING map.
1909 space_reftree_create(&reftree
);
1910 space_reftree_add_map(&reftree
,
1911 vd
->vdev_dtl
[DTL_MISSING
], 1);
1912 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1913 space_reftree_add_map(&reftree
,
1914 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1915 space_reftree_generate_map(&reftree
,
1916 vd
->vdev_dtl
[DTL_MISSING
], 1);
1917 space_reftree_destroy(&reftree
);
1919 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1920 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1921 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1923 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1924 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1925 if (!vdev_readable(vd
))
1926 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1928 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1929 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1932 * If the vdev was resilvering and no longer has any
1933 * DTLs then reset its resilvering flag and dirty
1934 * the top level so that we persist the change.
1936 if (vd
->vdev_resilver_txg
!= 0 &&
1937 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1938 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
1939 vd
->vdev_resilver_txg
= 0;
1940 vdev_config_dirty(vd
->vdev_top
);
1943 mutex_exit(&vd
->vdev_dtl_lock
);
1946 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1950 mutex_enter(&vd
->vdev_dtl_lock
);
1951 for (t
= 0; t
< DTL_TYPES
; t
++) {
1954 /* account for child's outage in parent's missing map */
1955 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1957 continue; /* leaf vdevs only */
1958 if (t
== DTL_PARTIAL
)
1959 minref
= 1; /* i.e. non-zero */
1960 else if (vd
->vdev_nparity
!= 0)
1961 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1963 minref
= vd
->vdev_children
; /* any kind of mirror */
1964 space_reftree_create(&reftree
);
1965 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1966 vdev_t
*cvd
= vd
->vdev_child
[c
];
1967 mutex_enter(&cvd
->vdev_dtl_lock
);
1968 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1969 mutex_exit(&cvd
->vdev_dtl_lock
);
1971 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1972 space_reftree_destroy(&reftree
);
1974 mutex_exit(&vd
->vdev_dtl_lock
);
1978 vdev_dtl_load(vdev_t
*vd
)
1980 spa_t
*spa
= vd
->vdev_spa
;
1981 objset_t
*mos
= spa
->spa_meta_objset
;
1985 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
1986 ASSERT(!vd
->vdev_ishole
);
1988 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
1989 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
1992 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
1994 mutex_enter(&vd
->vdev_dtl_lock
);
1997 * Now that we've opened the space_map we need to update
2000 space_map_update(vd
->vdev_dtl_sm
);
2002 error
= space_map_load(vd
->vdev_dtl_sm
,
2003 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2004 mutex_exit(&vd
->vdev_dtl_lock
);
2009 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2010 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2019 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2021 spa_t
*spa
= vd
->vdev_spa
;
2023 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2024 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2029 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2031 spa_t
*spa
= vd
->vdev_spa
;
2032 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2033 DMU_OT_NONE
, 0, tx
);
2036 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2043 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2047 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2048 vd
->vdev_ops
!= &vdev_missing_ops
&&
2049 vd
->vdev_ops
!= &vdev_root_ops
&&
2050 !vd
->vdev_top
->vdev_removing
) {
2051 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2052 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2054 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2055 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2058 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2059 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2064 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2066 spa_t
*spa
= vd
->vdev_spa
;
2067 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2068 objset_t
*mos
= spa
->spa_meta_objset
;
2069 range_tree_t
*rtsync
;
2072 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2074 ASSERT(!vd
->vdev_ishole
);
2075 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2077 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2079 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2080 mutex_enter(&vd
->vdev_dtl_lock
);
2081 space_map_free(vd
->vdev_dtl_sm
, tx
);
2082 space_map_close(vd
->vdev_dtl_sm
);
2083 vd
->vdev_dtl_sm
= NULL
;
2084 mutex_exit(&vd
->vdev_dtl_lock
);
2087 * We only destroy the leaf ZAP for detached leaves or for
2088 * removed log devices. Removed data devices handle leaf ZAP
2089 * cleanup later, once cancellation is no longer possible.
2091 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2092 vd
->vdev_top
->vdev_islog
)) {
2093 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2094 vd
->vdev_leaf_zap
= 0;
2101 if (vd
->vdev_dtl_sm
== NULL
) {
2102 uint64_t new_object
;
2104 new_object
= space_map_alloc(mos
, tx
);
2105 VERIFY3U(new_object
, !=, 0);
2107 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2108 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2109 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2112 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2114 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2116 mutex_enter(&rtlock
);
2118 mutex_enter(&vd
->vdev_dtl_lock
);
2119 range_tree_walk(rt
, range_tree_add
, rtsync
);
2120 mutex_exit(&vd
->vdev_dtl_lock
);
2122 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2123 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2124 range_tree_vacate(rtsync
, NULL
, NULL
);
2126 range_tree_destroy(rtsync
);
2128 mutex_exit(&rtlock
);
2129 mutex_destroy(&rtlock
);
2132 * If the object for the space map has changed then dirty
2133 * the top level so that we update the config.
2135 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2136 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2137 "new object %llu", txg
, spa_name(spa
), object
,
2138 space_map_object(vd
->vdev_dtl_sm
));
2139 vdev_config_dirty(vd
->vdev_top
);
2144 mutex_enter(&vd
->vdev_dtl_lock
);
2145 space_map_update(vd
->vdev_dtl_sm
);
2146 mutex_exit(&vd
->vdev_dtl_lock
);
2150 * Determine whether the specified vdev can be offlined/detached/removed
2151 * without losing data.
2154 vdev_dtl_required(vdev_t
*vd
)
2156 spa_t
*spa
= vd
->vdev_spa
;
2157 vdev_t
*tvd
= vd
->vdev_top
;
2158 uint8_t cant_read
= vd
->vdev_cant_read
;
2161 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2163 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2167 * Temporarily mark the device as unreadable, and then determine
2168 * whether this results in any DTL outages in the top-level vdev.
2169 * If not, we can safely offline/detach/remove the device.
2171 vd
->vdev_cant_read
= B_TRUE
;
2172 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2173 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2174 vd
->vdev_cant_read
= cant_read
;
2175 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2177 if (!required
&& zio_injection_enabled
)
2178 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2184 * Determine if resilver is needed, and if so the txg range.
2187 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2189 boolean_t needed
= B_FALSE
;
2190 uint64_t thismin
= UINT64_MAX
;
2191 uint64_t thismax
= 0;
2194 if (vd
->vdev_children
== 0) {
2195 mutex_enter(&vd
->vdev_dtl_lock
);
2196 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2197 vdev_writeable(vd
)) {
2199 thismin
= vdev_dtl_min(vd
);
2200 thismax
= vdev_dtl_max(vd
);
2203 mutex_exit(&vd
->vdev_dtl_lock
);
2205 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2206 vdev_t
*cvd
= vd
->vdev_child
[c
];
2207 uint64_t cmin
, cmax
;
2209 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2210 thismin
= MIN(thismin
, cmin
);
2211 thismax
= MAX(thismax
, cmax
);
2217 if (needed
&& minp
) {
2225 vdev_load(vdev_t
*vd
)
2230 * Recursively load all children.
2232 for (c
= 0; c
< vd
->vdev_children
; c
++)
2233 vdev_load(vd
->vdev_child
[c
]);
2236 * If this is a top-level vdev, initialize its metaslabs.
2238 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2239 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2240 vdev_metaslab_init(vd
, 0) != 0))
2241 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2242 VDEV_AUX_CORRUPT_DATA
);
2244 * If this is a leaf vdev, load its DTL.
2246 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2247 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2248 VDEV_AUX_CORRUPT_DATA
);
2252 * The special vdev case is used for hot spares and l2cache devices. Its
2253 * sole purpose it to set the vdev state for the associated vdev. To do this,
2254 * we make sure that we can open the underlying device, then try to read the
2255 * label, and make sure that the label is sane and that it hasn't been
2256 * repurposed to another pool.
2259 vdev_validate_aux(vdev_t
*vd
)
2262 uint64_t guid
, version
;
2265 if (!vdev_readable(vd
))
2268 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2269 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2270 VDEV_AUX_CORRUPT_DATA
);
2274 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2275 !SPA_VERSION_IS_SUPPORTED(version
) ||
2276 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2277 guid
!= vd
->vdev_guid
||
2278 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2279 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2280 VDEV_AUX_CORRUPT_DATA
);
2286 * We don't actually check the pool state here. If it's in fact in
2287 * use by another pool, we update this fact on the fly when requested.
2294 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2296 spa_t
*spa
= vd
->vdev_spa
;
2297 objset_t
*mos
= spa
->spa_meta_objset
;
2301 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2302 ASSERT(vd
== vd
->vdev_top
);
2303 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2305 if (vd
->vdev_ms
!= NULL
) {
2306 metaslab_group_t
*mg
= vd
->vdev_mg
;
2308 metaslab_group_histogram_verify(mg
);
2309 metaslab_class_histogram_verify(mg
->mg_class
);
2311 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2312 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2314 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2317 mutex_enter(&msp
->ms_lock
);
2319 * If the metaslab was not loaded when the vdev
2320 * was removed then the histogram accounting may
2321 * not be accurate. Update the histogram information
2322 * here so that we ensure that the metaslab group
2323 * and metaslab class are up-to-date.
2325 metaslab_group_histogram_remove(mg
, msp
);
2327 VERIFY0(space_map_allocated(msp
->ms_sm
));
2328 space_map_free(msp
->ms_sm
, tx
);
2329 space_map_close(msp
->ms_sm
);
2331 mutex_exit(&msp
->ms_lock
);
2334 metaslab_group_histogram_verify(mg
);
2335 metaslab_class_histogram_verify(mg
->mg_class
);
2336 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2337 ASSERT0(mg
->mg_histogram
[i
]);
2341 if (vd
->vdev_ms_array
) {
2342 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2343 vd
->vdev_ms_array
= 0;
2346 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2347 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2348 vd
->vdev_top_zap
= 0;
2354 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2357 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2359 ASSERT(!vd
->vdev_ishole
);
2361 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2362 metaslab_sync_done(msp
, txg
);
2365 metaslab_sync_reassess(vd
->vdev_mg
);
2369 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2371 spa_t
*spa
= vd
->vdev_spa
;
2376 ASSERT(!vd
->vdev_ishole
);
2378 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2379 ASSERT(vd
== vd
->vdev_top
);
2380 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2381 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2382 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2383 ASSERT(vd
->vdev_ms_array
!= 0);
2384 vdev_config_dirty(vd
);
2389 * Remove the metadata associated with this vdev once it's empty.
2391 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2392 vdev_remove(vd
, txg
);
2394 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2395 metaslab_sync(msp
, txg
);
2396 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2399 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2400 vdev_dtl_sync(lvd
, txg
);
2402 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2406 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2408 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2412 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2413 * not be opened, and no I/O is attempted.
2416 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2420 spa_vdev_state_enter(spa
, SCL_NONE
);
2422 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2423 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2425 if (!vd
->vdev_ops
->vdev_op_leaf
)
2426 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2431 * We don't directly use the aux state here, but if we do a
2432 * vdev_reopen(), we need this value to be present to remember why we
2435 vd
->vdev_label_aux
= aux
;
2438 * Faulted state takes precedence over degraded.
2440 vd
->vdev_delayed_close
= B_FALSE
;
2441 vd
->vdev_faulted
= 1ULL;
2442 vd
->vdev_degraded
= 0ULL;
2443 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2446 * If this device has the only valid copy of the data, then
2447 * back off and simply mark the vdev as degraded instead.
2449 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2450 vd
->vdev_degraded
= 1ULL;
2451 vd
->vdev_faulted
= 0ULL;
2454 * If we reopen the device and it's not dead, only then do we
2459 if (vdev_readable(vd
))
2460 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2463 return (spa_vdev_state_exit(spa
, vd
, 0));
2467 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2468 * user that something is wrong. The vdev continues to operate as normal as far
2469 * as I/O is concerned.
2472 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2476 spa_vdev_state_enter(spa
, SCL_NONE
);
2478 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2479 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2481 if (!vd
->vdev_ops
->vdev_op_leaf
)
2482 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2485 * If the vdev is already faulted, then don't do anything.
2487 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2488 return (spa_vdev_state_exit(spa
, NULL
, 0));
2490 vd
->vdev_degraded
= 1ULL;
2491 if (!vdev_is_dead(vd
))
2492 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2495 return (spa_vdev_state_exit(spa
, vd
, 0));
2499 * Online the given vdev.
2501 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2502 * spare device should be detached when the device finishes resilvering.
2503 * Second, the online should be treated like a 'test' online case, so no FMA
2504 * events are generated if the device fails to open.
2507 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2509 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2510 boolean_t postevent
= B_FALSE
;
2512 spa_vdev_state_enter(spa
, SCL_NONE
);
2514 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2515 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2517 if (!vd
->vdev_ops
->vdev_op_leaf
)
2518 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2521 (vd
->vdev_offline
== B_TRUE
|| vd
->vdev_tmpoffline
== B_TRUE
) ?
2525 vd
->vdev_offline
= B_FALSE
;
2526 vd
->vdev_tmpoffline
= B_FALSE
;
2527 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2528 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2530 /* XXX - L2ARC 1.0 does not support expansion */
2531 if (!vd
->vdev_aux
) {
2532 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2533 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2537 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2539 if (!vd
->vdev_aux
) {
2540 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2541 pvd
->vdev_expanding
= B_FALSE
;
2545 *newstate
= vd
->vdev_state
;
2546 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2547 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2548 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2549 vd
->vdev_parent
->vdev_child
[0] == vd
)
2550 vd
->vdev_unspare
= B_TRUE
;
2552 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2554 /* XXX - L2ARC 1.0 does not support expansion */
2556 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2557 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2561 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2563 return (spa_vdev_state_exit(spa
, vd
, 0));
2567 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2571 uint64_t generation
;
2572 metaslab_group_t
*mg
;
2575 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2577 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2578 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2580 if (!vd
->vdev_ops
->vdev_op_leaf
)
2581 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2585 generation
= spa
->spa_config_generation
+ 1;
2588 * If the device isn't already offline, try to offline it.
2590 if (!vd
->vdev_offline
) {
2592 * If this device has the only valid copy of some data,
2593 * don't allow it to be offlined. Log devices are always
2596 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2597 vdev_dtl_required(vd
))
2598 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2601 * If the top-level is a slog and it has had allocations
2602 * then proceed. We check that the vdev's metaslab group
2603 * is not NULL since it's possible that we may have just
2604 * added this vdev but not yet initialized its metaslabs.
2606 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2608 * Prevent any future allocations.
2610 metaslab_group_passivate(mg
);
2611 (void) spa_vdev_state_exit(spa
, vd
, 0);
2613 error
= spa_offline_log(spa
);
2615 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2618 * Check to see if the config has changed.
2620 if (error
|| generation
!= spa
->spa_config_generation
) {
2621 metaslab_group_activate(mg
);
2623 return (spa_vdev_state_exit(spa
,
2625 (void) spa_vdev_state_exit(spa
, vd
, 0);
2628 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2632 * Offline this device and reopen its top-level vdev.
2633 * If the top-level vdev is a log device then just offline
2634 * it. Otherwise, if this action results in the top-level
2635 * vdev becoming unusable, undo it and fail the request.
2637 vd
->vdev_offline
= B_TRUE
;
2640 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2641 vdev_is_dead(tvd
)) {
2642 vd
->vdev_offline
= B_FALSE
;
2644 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2648 * Add the device back into the metaslab rotor so that
2649 * once we online the device it's open for business.
2651 if (tvd
->vdev_islog
&& mg
!= NULL
)
2652 metaslab_group_activate(mg
);
2655 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2657 return (spa_vdev_state_exit(spa
, vd
, 0));
2661 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2665 mutex_enter(&spa
->spa_vdev_top_lock
);
2666 error
= vdev_offline_locked(spa
, guid
, flags
);
2667 mutex_exit(&spa
->spa_vdev_top_lock
);
2673 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2674 * vdev_offline(), we assume the spa config is locked. We also clear all
2675 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2678 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2680 vdev_t
*rvd
= spa
->spa_root_vdev
;
2683 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2688 vd
->vdev_stat
.vs_read_errors
= 0;
2689 vd
->vdev_stat
.vs_write_errors
= 0;
2690 vd
->vdev_stat
.vs_checksum_errors
= 0;
2692 for (c
= 0; c
< vd
->vdev_children
; c
++)
2693 vdev_clear(spa
, vd
->vdev_child
[c
]);
2696 * If we're in the FAULTED state or have experienced failed I/O, then
2697 * clear the persistent state and attempt to reopen the device. We
2698 * also mark the vdev config dirty, so that the new faulted state is
2699 * written out to disk.
2701 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2702 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2705 * When reopening in reponse to a clear event, it may be due to
2706 * a fmadm repair request. In this case, if the device is
2707 * still broken, we want to still post the ereport again.
2709 vd
->vdev_forcefault
= B_TRUE
;
2711 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2712 vd
->vdev_cant_read
= B_FALSE
;
2713 vd
->vdev_cant_write
= B_FALSE
;
2715 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2717 vd
->vdev_forcefault
= B_FALSE
;
2719 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2720 vdev_state_dirty(vd
->vdev_top
);
2722 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2723 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2725 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2729 * When clearing a FMA-diagnosed fault, we always want to
2730 * unspare the device, as we assume that the original spare was
2731 * done in response to the FMA fault.
2733 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2734 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2735 vd
->vdev_parent
->vdev_child
[0] == vd
)
2736 vd
->vdev_unspare
= B_TRUE
;
2740 vdev_is_dead(vdev_t
*vd
)
2743 * Holes and missing devices are always considered "dead".
2744 * This simplifies the code since we don't have to check for
2745 * these types of devices in the various code paths.
2746 * Instead we rely on the fact that we skip over dead devices
2747 * before issuing I/O to them.
2749 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2750 vd
->vdev_ops
== &vdev_missing_ops
);
2754 vdev_readable(vdev_t
*vd
)
2756 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2760 vdev_writeable(vdev_t
*vd
)
2762 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2766 vdev_allocatable(vdev_t
*vd
)
2768 uint64_t state
= vd
->vdev_state
;
2771 * We currently allow allocations from vdevs which may be in the
2772 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2773 * fails to reopen then we'll catch it later when we're holding
2774 * the proper locks. Note that we have to get the vdev state
2775 * in a local variable because although it changes atomically,
2776 * we're asking two separate questions about it.
2778 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2779 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
&&
2780 vd
->vdev_mg
->mg_initialized
);
2784 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2786 ASSERT(zio
->io_vd
== vd
);
2788 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2791 if (zio
->io_type
== ZIO_TYPE_READ
)
2792 return (!vd
->vdev_cant_read
);
2794 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2795 return (!vd
->vdev_cant_write
);
2801 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2804 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2805 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2806 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2809 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2813 * Get extended stats
2816 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2819 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2820 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2821 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2823 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2824 vsx
->vsx_total_histo
[t
][b
] +=
2825 cvsx
->vsx_total_histo
[t
][b
];
2829 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2830 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2831 vsx
->vsx_queue_histo
[t
][b
] +=
2832 cvsx
->vsx_queue_histo
[t
][b
];
2834 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2835 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2837 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2838 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2840 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2841 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2847 * Get statistics for the given vdev.
2850 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2854 * If we're getting stats on the root vdev, aggregate the I/O counts
2855 * over all top-level vdevs (i.e. the direct children of the root).
2857 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2859 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2860 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2863 memset(vsx
, 0, sizeof (*vsx
));
2865 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2866 vdev_t
*cvd
= vd
->vdev_child
[c
];
2867 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2868 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2870 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2872 vdev_get_child_stat(cvd
, vs
, cvs
);
2874 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2879 * We're a leaf. Just copy our ZIO active queue stats in. The
2880 * other leaf stats are updated in vdev_stat_update().
2885 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2887 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2888 vsx
->vsx_active_queue
[t
] =
2889 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2890 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2891 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2897 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2899 mutex_enter(&vd
->vdev_stat_lock
);
2901 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2902 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2903 vs
->vs_state
= vd
->vdev_state
;
2904 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2905 if (vd
->vdev_ops
->vdev_op_leaf
)
2906 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
2907 VDEV_LABEL_END_SIZE
;
2908 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2909 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
2911 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2915 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
2916 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
2917 mutex_exit(&vd
->vdev_stat_lock
);
2921 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2923 return (vdev_get_stats_ex(vd
, vs
, NULL
));
2927 vdev_clear_stats(vdev_t
*vd
)
2929 mutex_enter(&vd
->vdev_stat_lock
);
2930 vd
->vdev_stat
.vs_space
= 0;
2931 vd
->vdev_stat
.vs_dspace
= 0;
2932 vd
->vdev_stat
.vs_alloc
= 0;
2933 mutex_exit(&vd
->vdev_stat_lock
);
2937 vdev_scan_stat_init(vdev_t
*vd
)
2939 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2942 for (c
= 0; c
< vd
->vdev_children
; c
++)
2943 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2945 mutex_enter(&vd
->vdev_stat_lock
);
2946 vs
->vs_scan_processed
= 0;
2947 mutex_exit(&vd
->vdev_stat_lock
);
2951 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2953 spa_t
*spa
= zio
->io_spa
;
2954 vdev_t
*rvd
= spa
->spa_root_vdev
;
2955 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2957 uint64_t txg
= zio
->io_txg
;
2958 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2959 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
2960 zio_type_t type
= zio
->io_type
;
2961 int flags
= zio
->io_flags
;
2964 * If this i/o is a gang leader, it didn't do any actual work.
2966 if (zio
->io_gang_tree
)
2969 if (zio
->io_error
== 0) {
2971 * If this is a root i/o, don't count it -- we've already
2972 * counted the top-level vdevs, and vdev_get_stats() will
2973 * aggregate them when asked. This reduces contention on
2974 * the root vdev_stat_lock and implicitly handles blocks
2975 * that compress away to holes, for which there is no i/o.
2976 * (Holes never create vdev children, so all the counters
2977 * remain zero, which is what we want.)
2979 * Note: this only applies to successful i/o (io_error == 0)
2980 * because unlike i/o counts, errors are not additive.
2981 * When reading a ditto block, for example, failure of
2982 * one top-level vdev does not imply a root-level error.
2987 ASSERT(vd
== zio
->io_vd
);
2989 if (flags
& ZIO_FLAG_IO_BYPASS
)
2992 mutex_enter(&vd
->vdev_stat_lock
);
2994 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2995 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2996 dsl_scan_phys_t
*scn_phys
=
2997 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2998 uint64_t *processed
= &scn_phys
->scn_processed
;
3001 if (vd
->vdev_ops
->vdev_op_leaf
)
3002 atomic_add_64(processed
, psize
);
3003 vs
->vs_scan_processed
+= psize
;
3006 if (flags
& ZIO_FLAG_SELF_HEAL
)
3007 vs
->vs_self_healed
+= psize
;
3011 * The bytes/ops/histograms are recorded at the leaf level and
3012 * aggregated into the higher level vdevs in vdev_get_stats().
3014 if (vd
->vdev_ops
->vdev_op_leaf
&&
3015 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3018 vs
->vs_bytes
[type
] += psize
;
3020 if (flags
& ZIO_FLAG_DELEGATED
) {
3021 vsx
->vsx_agg_histo
[zio
->io_priority
]
3022 [RQ_HISTO(zio
->io_size
)]++;
3024 vsx
->vsx_ind_histo
[zio
->io_priority
]
3025 [RQ_HISTO(zio
->io_size
)]++;
3028 if (zio
->io_delta
&& zio
->io_delay
) {
3029 vsx
->vsx_queue_histo
[zio
->io_priority
]
3030 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3031 vsx
->vsx_disk_histo
[type
]
3032 [L_HISTO(zio
->io_delay
)]++;
3033 vsx
->vsx_total_histo
[type
]
3034 [L_HISTO(zio
->io_delta
)]++;
3038 mutex_exit(&vd
->vdev_stat_lock
);
3042 if (flags
& ZIO_FLAG_SPECULATIVE
)
3046 * If this is an I/O error that is going to be retried, then ignore the
3047 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3048 * hard errors, when in reality they can happen for any number of
3049 * innocuous reasons (bus resets, MPxIO link failure, etc).
3051 if (zio
->io_error
== EIO
&&
3052 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3056 * Intent logs writes won't propagate their error to the root
3057 * I/O so don't mark these types of failures as pool-level
3060 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3063 mutex_enter(&vd
->vdev_stat_lock
);
3064 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3065 if (zio
->io_error
== ECKSUM
)
3066 vs
->vs_checksum_errors
++;
3068 vs
->vs_read_errors
++;
3070 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3071 vs
->vs_write_errors
++;
3072 mutex_exit(&vd
->vdev_stat_lock
);
3074 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3075 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3076 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3077 spa
->spa_claiming
)) {
3079 * This is either a normal write (not a repair), or it's
3080 * a repair induced by the scrub thread, or it's a repair
3081 * made by zil_claim() during spa_load() in the first txg.
3082 * In the normal case, we commit the DTL change in the same
3083 * txg as the block was born. In the scrub-induced repair
3084 * case, we know that scrubs run in first-pass syncing context,
3085 * so we commit the DTL change in spa_syncing_txg(spa).
3086 * In the zil_claim() case, we commit in spa_first_txg(spa).
3088 * We currently do not make DTL entries for failed spontaneous
3089 * self-healing writes triggered by normal (non-scrubbing)
3090 * reads, because we have no transactional context in which to
3091 * do so -- and it's not clear that it'd be desirable anyway.
3093 if (vd
->vdev_ops
->vdev_op_leaf
) {
3094 uint64_t commit_txg
= txg
;
3095 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3096 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3097 ASSERT(spa_sync_pass(spa
) == 1);
3098 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3099 commit_txg
= spa_syncing_txg(spa
);
3100 } else if (spa
->spa_claiming
) {
3101 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3102 commit_txg
= spa_first_txg(spa
);
3104 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3105 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3107 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3108 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3109 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3112 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3117 * Update the in-core space usage stats for this vdev, its metaslab class,
3118 * and the root vdev.
3121 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3122 int64_t space_delta
)
3124 int64_t dspace_delta
= space_delta
;
3125 spa_t
*spa
= vd
->vdev_spa
;
3126 vdev_t
*rvd
= spa
->spa_root_vdev
;
3127 metaslab_group_t
*mg
= vd
->vdev_mg
;
3128 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3130 ASSERT(vd
== vd
->vdev_top
);
3133 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3134 * factor. We must calculate this here and not at the root vdev
3135 * because the root vdev's psize-to-asize is simply the max of its
3136 * childrens', thus not accurate enough for us.
3138 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3139 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3140 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3141 vd
->vdev_deflate_ratio
;
3143 mutex_enter(&vd
->vdev_stat_lock
);
3144 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3145 vd
->vdev_stat
.vs_space
+= space_delta
;
3146 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3147 mutex_exit(&vd
->vdev_stat_lock
);
3149 if (mc
== spa_normal_class(spa
)) {
3150 mutex_enter(&rvd
->vdev_stat_lock
);
3151 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3152 rvd
->vdev_stat
.vs_space
+= space_delta
;
3153 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3154 mutex_exit(&rvd
->vdev_stat_lock
);
3158 ASSERT(rvd
== vd
->vdev_parent
);
3159 ASSERT(vd
->vdev_ms_count
!= 0);
3161 metaslab_class_space_update(mc
,
3162 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3167 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3168 * so that it will be written out next time the vdev configuration is synced.
3169 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3172 vdev_config_dirty(vdev_t
*vd
)
3174 spa_t
*spa
= vd
->vdev_spa
;
3175 vdev_t
*rvd
= spa
->spa_root_vdev
;
3178 ASSERT(spa_writeable(spa
));
3181 * If this is an aux vdev (as with l2cache and spare devices), then we
3182 * update the vdev config manually and set the sync flag.
3184 if (vd
->vdev_aux
!= NULL
) {
3185 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3189 for (c
= 0; c
< sav
->sav_count
; c
++) {
3190 if (sav
->sav_vdevs
[c
] == vd
)
3194 if (c
== sav
->sav_count
) {
3196 * We're being removed. There's nothing more to do.
3198 ASSERT(sav
->sav_sync
== B_TRUE
);
3202 sav
->sav_sync
= B_TRUE
;
3204 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3205 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3206 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3207 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3213 * Setting the nvlist in the middle if the array is a little
3214 * sketchy, but it will work.
3216 nvlist_free(aux
[c
]);
3217 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3223 * The dirty list is protected by the SCL_CONFIG lock. The caller
3224 * must either hold SCL_CONFIG as writer, or must be the sync thread
3225 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3226 * so this is sufficient to ensure mutual exclusion.
3228 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3229 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3230 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3233 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3234 vdev_config_dirty(rvd
->vdev_child
[c
]);
3236 ASSERT(vd
== vd
->vdev_top
);
3238 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3240 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3245 vdev_config_clean(vdev_t
*vd
)
3247 spa_t
*spa
= vd
->vdev_spa
;
3249 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3250 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3251 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3253 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3254 list_remove(&spa
->spa_config_dirty_list
, vd
);
3258 * Mark a top-level vdev's state as dirty, so that the next pass of
3259 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3260 * the state changes from larger config changes because they require
3261 * much less locking, and are often needed for administrative actions.
3264 vdev_state_dirty(vdev_t
*vd
)
3266 spa_t
*spa
= vd
->vdev_spa
;
3268 ASSERT(spa_writeable(spa
));
3269 ASSERT(vd
== vd
->vdev_top
);
3272 * The state list is protected by the SCL_STATE lock. The caller
3273 * must either hold SCL_STATE as writer, or must be the sync thread
3274 * (which holds SCL_STATE as reader). There's only one sync thread,
3275 * so this is sufficient to ensure mutual exclusion.
3277 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3278 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3279 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3281 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3282 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3286 vdev_state_clean(vdev_t
*vd
)
3288 spa_t
*spa
= vd
->vdev_spa
;
3290 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3291 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3292 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3294 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3295 list_remove(&spa
->spa_state_dirty_list
, vd
);
3299 * Propagate vdev state up from children to parent.
3302 vdev_propagate_state(vdev_t
*vd
)
3304 spa_t
*spa
= vd
->vdev_spa
;
3305 vdev_t
*rvd
= spa
->spa_root_vdev
;
3306 int degraded
= 0, faulted
= 0;
3311 if (vd
->vdev_children
> 0) {
3312 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3313 child
= vd
->vdev_child
[c
];
3316 * Don't factor holes into the decision.
3318 if (child
->vdev_ishole
)
3321 if (!vdev_readable(child
) ||
3322 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3324 * Root special: if there is a top-level log
3325 * device, treat the root vdev as if it were
3328 if (child
->vdev_islog
&& vd
== rvd
)
3332 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3336 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3340 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3343 * Root special: if there is a top-level vdev that cannot be
3344 * opened due to corrupted metadata, then propagate the root
3345 * vdev's aux state as 'corrupt' rather than 'insufficient
3348 if (corrupted
&& vd
== rvd
&&
3349 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3350 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3351 VDEV_AUX_CORRUPT_DATA
);
3354 if (vd
->vdev_parent
)
3355 vdev_propagate_state(vd
->vdev_parent
);
3359 * Set a vdev's state. If this is during an open, we don't update the parent
3360 * state, because we're in the process of opening children depth-first.
3361 * Otherwise, we propagate the change to the parent.
3363 * If this routine places a device in a faulted state, an appropriate ereport is
3367 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3369 uint64_t save_state
;
3370 spa_t
*spa
= vd
->vdev_spa
;
3372 if (state
== vd
->vdev_state
) {
3373 vd
->vdev_stat
.vs_aux
= aux
;
3377 save_state
= vd
->vdev_state
;
3379 vd
->vdev_state
= state
;
3380 vd
->vdev_stat
.vs_aux
= aux
;
3383 * If we are setting the vdev state to anything but an open state, then
3384 * always close the underlying device unless the device has requested
3385 * a delayed close (i.e. we're about to remove or fault the device).
3386 * Otherwise, we keep accessible but invalid devices open forever.
3387 * We don't call vdev_close() itself, because that implies some extra
3388 * checks (offline, etc) that we don't want here. This is limited to
3389 * leaf devices, because otherwise closing the device will affect other
3392 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3393 vd
->vdev_ops
->vdev_op_leaf
)
3394 vd
->vdev_ops
->vdev_op_close(vd
);
3396 if (vd
->vdev_removed
&&
3397 state
== VDEV_STATE_CANT_OPEN
&&
3398 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3400 * If the previous state is set to VDEV_STATE_REMOVED, then this
3401 * device was previously marked removed and someone attempted to
3402 * reopen it. If this failed due to a nonexistent device, then
3403 * keep the device in the REMOVED state. We also let this be if
3404 * it is one of our special test online cases, which is only
3405 * attempting to online the device and shouldn't generate an FMA
3408 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3409 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3410 } else if (state
== VDEV_STATE_REMOVED
) {
3411 vd
->vdev_removed
= B_TRUE
;
3412 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3414 * If we fail to open a vdev during an import or recovery, we
3415 * mark it as "not available", which signifies that it was
3416 * never there to begin with. Failure to open such a device
3417 * is not considered an error.
3419 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3420 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3421 vd
->vdev_ops
->vdev_op_leaf
)
3422 vd
->vdev_not_present
= 1;
3425 * Post the appropriate ereport. If the 'prevstate' field is
3426 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3427 * that this is part of a vdev_reopen(). In this case, we don't
3428 * want to post the ereport if the device was already in the
3429 * CANT_OPEN state beforehand.
3431 * If the 'checkremove' flag is set, then this is an attempt to
3432 * online the device in response to an insertion event. If we
3433 * hit this case, then we have detected an insertion event for a
3434 * faulted or offline device that wasn't in the removed state.
3435 * In this scenario, we don't post an ereport because we are
3436 * about to replace the device, or attempt an online with
3437 * vdev_forcefault, which will generate the fault for us.
3439 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3440 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3441 vd
!= spa
->spa_root_vdev
) {
3445 case VDEV_AUX_OPEN_FAILED
:
3446 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3448 case VDEV_AUX_CORRUPT_DATA
:
3449 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3451 case VDEV_AUX_NO_REPLICAS
:
3452 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3454 case VDEV_AUX_BAD_GUID_SUM
:
3455 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3457 case VDEV_AUX_TOO_SMALL
:
3458 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3460 case VDEV_AUX_BAD_LABEL
:
3461 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3464 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3467 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3470 /* Erase any notion of persistent removed state */
3471 vd
->vdev_removed
= B_FALSE
;
3473 vd
->vdev_removed
= B_FALSE
;
3477 * Notify ZED of any significant state-change on a leaf vdev.
3480 if (vd
->vdev_ops
->vdev_op_leaf
) {
3481 /* preserve original state from a vdev_reopen() */
3482 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
3483 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
3484 (save_state
<= VDEV_STATE_CLOSED
))
3485 save_state
= vd
->vdev_prevstate
;
3487 /* filter out state change due to initial vdev_open */
3488 if (save_state
> VDEV_STATE_CLOSED
)
3489 zfs_post_state_change(spa
, vd
, save_state
);
3492 if (!isopen
&& vd
->vdev_parent
)
3493 vdev_propagate_state(vd
->vdev_parent
);
3497 * Check the vdev configuration to ensure that it's capable of supporting
3501 vdev_is_bootable(vdev_t
*vd
)
3503 #if defined(__sun__) || defined(__sun)
3505 * Currently, we do not support RAID-Z or partial configuration.
3506 * In addition, only a single top-level vdev is allowed and none of the
3507 * leaves can be wholedisks.
3511 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3512 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3514 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3515 vd
->vdev_children
> 1) {
3517 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3518 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3521 } else if (vd
->vdev_wholedisk
== 1) {
3525 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3526 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3529 #endif /* __sun__ || __sun */
3534 * Load the state from the original vdev tree (ovd) which
3535 * we've retrieved from the MOS config object. If the original
3536 * vdev was offline or faulted then we transfer that state to the
3537 * device in the current vdev tree (nvd).
3540 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3544 ASSERT(nvd
->vdev_top
->vdev_islog
);
3545 ASSERT(spa_config_held(nvd
->vdev_spa
,
3546 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3547 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3549 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3550 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3552 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3554 * Restore the persistent vdev state
3556 nvd
->vdev_offline
= ovd
->vdev_offline
;
3557 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3558 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3559 nvd
->vdev_removed
= ovd
->vdev_removed
;
3564 * Determine if a log device has valid content. If the vdev was
3565 * removed or faulted in the MOS config then we know that
3566 * the content on the log device has already been written to the pool.
3569 vdev_log_state_valid(vdev_t
*vd
)
3573 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3577 for (c
= 0; c
< vd
->vdev_children
; c
++)
3578 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3585 * Expand a vdev if possible.
3588 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3590 ASSERT(vd
->vdev_top
== vd
);
3591 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3593 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3594 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3595 vdev_config_dirty(vd
);
3603 vdev_split(vdev_t
*vd
)
3605 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3607 vdev_remove_child(pvd
, vd
);
3608 vdev_compact_children(pvd
);
3610 cvd
= pvd
->vdev_child
[0];
3611 if (pvd
->vdev_children
== 1) {
3612 vdev_remove_parent(cvd
);
3613 cvd
->vdev_splitting
= B_TRUE
;
3615 vdev_propagate_state(cvd
);
3619 vdev_deadman(vdev_t
*vd
)
3623 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3624 vdev_t
*cvd
= vd
->vdev_child
[c
];
3629 if (vd
->vdev_ops
->vdev_op_leaf
) {
3630 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3632 mutex_enter(&vq
->vq_lock
);
3633 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3634 spa_t
*spa
= vd
->vdev_spa
;
3639 * Look at the head of all the pending queues,
3640 * if any I/O has been outstanding for longer than
3641 * the spa_deadman_synctime we log a zevent.
3643 fio
= avl_first(&vq
->vq_active_tree
);
3644 delta
= gethrtime() - fio
->io_timestamp
;
3645 if (delta
> spa_deadman_synctime(spa
)) {
3646 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3647 "delta %lluns, last io %lluns",
3648 fio
->io_timestamp
, delta
,
3649 vq
->vq_io_complete_ts
);
3650 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3651 spa
, vd
, fio
, 0, 0);
3654 mutex_exit(&vq
->vq_lock
);
3658 #if defined(_KERNEL) && defined(HAVE_SPL)
3659 EXPORT_SYMBOL(vdev_fault
);
3660 EXPORT_SYMBOL(vdev_degrade
);
3661 EXPORT_SYMBOL(vdev_online
);
3662 EXPORT_SYMBOL(vdev_offline
);
3663 EXPORT_SYMBOL(vdev_clear
);
3665 module_param(metaslabs_per_vdev
, int, 0644);
3666 MODULE_PARM_DESC(metaslabs_per_vdev
,
3667 "Divide added vdev into approximately (but no more than) this number "