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 (c) 2011, 2015 by Delphix. All rights reserved.
25 * Copyright 2015 Nexenta Systems, Inc. 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>
48 #include <sys/zfs_ratelimit.h>
51 * When a vdev is added, it will be divided into approximately (but no
52 * more than) this number of metaslabs.
54 int metaslabs_per_vdev
= 200;
57 * Virtual device management.
60 static vdev_ops_t
*vdev_ops_table
[] = {
74 * Given a vdev type, return the appropriate ops vector.
77 vdev_getops(const char *type
)
79 vdev_ops_t
*ops
, **opspp
;
81 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
82 if (strcmp(ops
->vdev_op_type
, type
) == 0)
89 * Default asize function: return the MAX of psize with the asize of
90 * all children. This is what's used by anything other than RAID-Z.
93 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
95 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
99 for (c
= 0; c
< vd
->vdev_children
; c
++) {
100 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
101 asize
= MAX(asize
, csize
);
108 * Get the minimum allocatable size. We define the allocatable size as
109 * the vdev's asize rounded to the nearest metaslab. This allows us to
110 * replace or attach devices which don't have the same physical size but
111 * can still satisfy the same number of allocations.
114 vdev_get_min_asize(vdev_t
*vd
)
116 vdev_t
*pvd
= vd
->vdev_parent
;
119 * If our parent is NULL (inactive spare or cache) or is the root,
120 * just return our own asize.
123 return (vd
->vdev_asize
);
126 * The top-level vdev just returns the allocatable size rounded
127 * to the nearest metaslab.
129 if (vd
== vd
->vdev_top
)
130 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
133 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
134 * so each child must provide at least 1/Nth of its asize.
136 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
137 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
139 return (pvd
->vdev_min_asize
);
143 vdev_set_min_asize(vdev_t
*vd
)
146 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
148 for (c
= 0; c
< vd
->vdev_children
; c
++)
149 vdev_set_min_asize(vd
->vdev_child
[c
]);
153 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
155 vdev_t
*rvd
= spa
->spa_root_vdev
;
157 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
159 if (vdev
< rvd
->vdev_children
) {
160 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
161 return (rvd
->vdev_child
[vdev
]);
168 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
173 if (vd
->vdev_guid
== guid
)
176 for (c
= 0; c
< vd
->vdev_children
; c
++)
177 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
185 vdev_count_leaves_impl(vdev_t
*vd
)
190 if (vd
->vdev_ops
->vdev_op_leaf
)
193 for (c
= 0; c
< vd
->vdev_children
; c
++)
194 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
200 vdev_count_leaves(spa_t
*spa
)
202 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
206 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
208 size_t oldsize
, newsize
;
209 uint64_t id
= cvd
->vdev_id
;
212 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
213 ASSERT(cvd
->vdev_parent
== NULL
);
215 cvd
->vdev_parent
= pvd
;
220 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
222 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
223 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
224 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
226 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
227 if (pvd
->vdev_child
!= NULL
) {
228 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
229 kmem_free(pvd
->vdev_child
, oldsize
);
232 pvd
->vdev_child
= newchild
;
233 pvd
->vdev_child
[id
] = cvd
;
235 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
236 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
239 * Walk up all ancestors to update guid sum.
241 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
242 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
246 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
249 uint_t id
= cvd
->vdev_id
;
251 ASSERT(cvd
->vdev_parent
== pvd
);
256 ASSERT(id
< pvd
->vdev_children
);
257 ASSERT(pvd
->vdev_child
[id
] == cvd
);
259 pvd
->vdev_child
[id
] = NULL
;
260 cvd
->vdev_parent
= NULL
;
262 for (c
= 0; c
< pvd
->vdev_children
; c
++)
263 if (pvd
->vdev_child
[c
])
266 if (c
== pvd
->vdev_children
) {
267 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
268 pvd
->vdev_child
= NULL
;
269 pvd
->vdev_children
= 0;
273 * Walk up all ancestors to update guid sum.
275 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
276 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
280 * Remove any holes in the child array.
283 vdev_compact_children(vdev_t
*pvd
)
285 vdev_t
**newchild
, *cvd
;
286 int oldc
= pvd
->vdev_children
;
290 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
292 for (c
= newc
= 0; c
< oldc
; c
++)
293 if (pvd
->vdev_child
[c
])
296 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
298 for (c
= newc
= 0; c
< oldc
; c
++) {
299 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
300 newchild
[newc
] = cvd
;
301 cvd
->vdev_id
= newc
++;
305 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
306 pvd
->vdev_child
= newchild
;
307 pvd
->vdev_children
= newc
;
311 * Allocate and minimally initialize a vdev_t.
314 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
319 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
321 if (spa
->spa_root_vdev
== NULL
) {
322 ASSERT(ops
== &vdev_root_ops
);
323 spa
->spa_root_vdev
= vd
;
324 spa
->spa_load_guid
= spa_generate_guid(NULL
);
327 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
328 if (spa
->spa_root_vdev
== vd
) {
330 * The root vdev's guid will also be the pool guid,
331 * which must be unique among all pools.
333 guid
= spa_generate_guid(NULL
);
336 * Any other vdev's guid must be unique within the pool.
338 guid
= spa_generate_guid(spa
);
340 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
345 vd
->vdev_guid
= guid
;
346 vd
->vdev_guid_sum
= guid
;
348 vd
->vdev_state
= VDEV_STATE_CLOSED
;
349 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
352 * Initialize rate limit structs for events. We rate limit ZIO delay
353 * and checksum events so that we don't overwhelm ZED with thousands
354 * of events when a disk is acting up.
356 zfs_ratelimit_init(&vd
->vdev_delay_rl
, DELAYS_PER_SECOND
, 1);
357 zfs_ratelimit_init(&vd
->vdev_checksum_rl
, CHECKSUMS_PER_SECOND
, 1);
359 list_link_init(&vd
->vdev_config_dirty_node
);
360 list_link_init(&vd
->vdev_state_dirty_node
);
361 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
362 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
363 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
364 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
366 for (t
= 0; t
< DTL_TYPES
; t
++) {
367 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
370 txg_list_create(&vd
->vdev_ms_list
,
371 offsetof(struct metaslab
, ms_txg_node
));
372 txg_list_create(&vd
->vdev_dtl_list
,
373 offsetof(struct vdev
, vdev_dtl_node
));
374 vd
->vdev_stat
.vs_timestamp
= gethrtime();
382 * Allocate a new vdev. The 'alloctype' is used to control whether we are
383 * creating a new vdev or loading an existing one - the behavior is slightly
384 * different for each case.
387 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
392 uint64_t guid
= 0, islog
, nparity
;
395 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
397 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
398 return (SET_ERROR(EINVAL
));
400 if ((ops
= vdev_getops(type
)) == NULL
)
401 return (SET_ERROR(EINVAL
));
404 * If this is a load, get the vdev guid from the nvlist.
405 * Otherwise, vdev_alloc_common() will generate one for us.
407 if (alloctype
== VDEV_ALLOC_LOAD
) {
410 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
412 return (SET_ERROR(EINVAL
));
414 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
415 return (SET_ERROR(EINVAL
));
416 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
417 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
418 return (SET_ERROR(EINVAL
));
419 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
420 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
421 return (SET_ERROR(EINVAL
));
422 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
423 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
424 return (SET_ERROR(EINVAL
));
428 * The first allocated vdev must be of type 'root'.
430 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
431 return (SET_ERROR(EINVAL
));
434 * Determine whether we're a log vdev.
437 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
438 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
439 return (SET_ERROR(ENOTSUP
));
441 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
442 return (SET_ERROR(ENOTSUP
));
445 * Set the nparity property for RAID-Z vdevs.
448 if (ops
== &vdev_raidz_ops
) {
449 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
451 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
452 return (SET_ERROR(EINVAL
));
454 * Previous versions could only support 1 or 2 parity
458 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
459 return (SET_ERROR(ENOTSUP
));
461 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
462 return (SET_ERROR(ENOTSUP
));
465 * We require the parity to be specified for SPAs that
466 * support multiple parity levels.
468 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
469 return (SET_ERROR(EINVAL
));
471 * Otherwise, we default to 1 parity device for RAID-Z.
478 ASSERT(nparity
!= -1ULL);
480 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
482 vd
->vdev_islog
= islog
;
483 vd
->vdev_nparity
= nparity
;
485 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
486 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
487 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
488 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
489 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
490 &vd
->vdev_physpath
) == 0)
491 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
493 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
494 &vd
->vdev_enc_sysfs_path
) == 0)
495 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
497 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
498 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
501 * Set the whole_disk property. If it's not specified, leave the value
504 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
505 &vd
->vdev_wholedisk
) != 0)
506 vd
->vdev_wholedisk
= -1ULL;
509 * Look for the 'not present' flag. This will only be set if the device
510 * was not present at the time of import.
512 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
513 &vd
->vdev_not_present
);
516 * Get the alignment requirement.
518 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
521 * Retrieve the vdev creation time.
523 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
527 * If we're a top-level vdev, try to load the allocation parameters.
529 if (parent
&& !parent
->vdev_parent
&&
530 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
531 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
533 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
535 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
537 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
539 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
542 ASSERT0(vd
->vdev_top_zap
);
545 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
546 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
547 alloctype
== VDEV_ALLOC_ADD
||
548 alloctype
== VDEV_ALLOC_SPLIT
||
549 alloctype
== VDEV_ALLOC_ROOTPOOL
);
550 vd
->vdev_mg
= metaslab_group_create(islog
?
551 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
554 if (vd
->vdev_ops
->vdev_op_leaf
&&
555 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
556 (void) nvlist_lookup_uint64(nv
,
557 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
559 ASSERT0(vd
->vdev_leaf_zap
);
563 * If we're a leaf vdev, try to load the DTL object and other state.
566 if (vd
->vdev_ops
->vdev_op_leaf
&&
567 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
568 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
569 if (alloctype
== VDEV_ALLOC_LOAD
) {
570 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
571 &vd
->vdev_dtl_object
);
572 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
576 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
579 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
580 &spare
) == 0 && spare
)
584 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
587 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
588 &vd
->vdev_resilver_txg
);
591 * When importing a pool, we want to ignore the persistent fault
592 * state, as the diagnosis made on another system may not be
593 * valid in the current context. Local vdevs will
594 * remain in the faulted state.
596 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
597 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
599 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
601 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
604 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
608 VDEV_AUX_ERR_EXCEEDED
;
609 if (nvlist_lookup_string(nv
,
610 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
611 strcmp(aux
, "external") == 0)
612 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
618 * Add ourselves to the parent's list of children.
620 vdev_add_child(parent
, vd
);
628 vdev_free(vdev_t
*vd
)
631 spa_t
*spa
= vd
->vdev_spa
;
634 * vdev_free() implies closing the vdev first. This is simpler than
635 * trying to ensure complicated semantics for all callers.
639 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
640 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
645 for (c
= 0; c
< vd
->vdev_children
; c
++)
646 vdev_free(vd
->vdev_child
[c
]);
648 ASSERT(vd
->vdev_child
== NULL
);
649 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
652 * Discard allocation state.
654 if (vd
->vdev_mg
!= NULL
) {
655 vdev_metaslab_fini(vd
);
656 metaslab_group_destroy(vd
->vdev_mg
);
659 ASSERT0(vd
->vdev_stat
.vs_space
);
660 ASSERT0(vd
->vdev_stat
.vs_dspace
);
661 ASSERT0(vd
->vdev_stat
.vs_alloc
);
664 * Remove this vdev from its parent's child list.
666 vdev_remove_child(vd
->vdev_parent
, vd
);
668 ASSERT(vd
->vdev_parent
== NULL
);
671 * Clean up vdev structure.
677 spa_strfree(vd
->vdev_path
);
679 spa_strfree(vd
->vdev_devid
);
680 if (vd
->vdev_physpath
)
681 spa_strfree(vd
->vdev_physpath
);
683 if (vd
->vdev_enc_sysfs_path
)
684 spa_strfree(vd
->vdev_enc_sysfs_path
);
687 spa_strfree(vd
->vdev_fru
);
689 if (vd
->vdev_isspare
)
690 spa_spare_remove(vd
);
691 if (vd
->vdev_isl2cache
)
692 spa_l2cache_remove(vd
);
694 txg_list_destroy(&vd
->vdev_ms_list
);
695 txg_list_destroy(&vd
->vdev_dtl_list
);
697 mutex_enter(&vd
->vdev_dtl_lock
);
698 space_map_close(vd
->vdev_dtl_sm
);
699 for (t
= 0; t
< DTL_TYPES
; t
++) {
700 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
701 range_tree_destroy(vd
->vdev_dtl
[t
]);
703 mutex_exit(&vd
->vdev_dtl_lock
);
705 mutex_destroy(&vd
->vdev_queue_lock
);
706 mutex_destroy(&vd
->vdev_dtl_lock
);
707 mutex_destroy(&vd
->vdev_stat_lock
);
708 mutex_destroy(&vd
->vdev_probe_lock
);
710 if (vd
== spa
->spa_root_vdev
)
711 spa
->spa_root_vdev
= NULL
;
713 kmem_free(vd
, sizeof (vdev_t
));
717 * Transfer top-level vdev state from svd to tvd.
720 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
722 spa_t
*spa
= svd
->vdev_spa
;
727 ASSERT(tvd
== tvd
->vdev_top
);
729 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
730 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
731 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
732 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
733 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
735 svd
->vdev_ms_array
= 0;
736 svd
->vdev_ms_shift
= 0;
737 svd
->vdev_ms_count
= 0;
738 svd
->vdev_top_zap
= 0;
741 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
742 tvd
->vdev_mg
= svd
->vdev_mg
;
743 tvd
->vdev_ms
= svd
->vdev_ms
;
748 if (tvd
->vdev_mg
!= NULL
)
749 tvd
->vdev_mg
->mg_vd
= tvd
;
751 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
752 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
753 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
755 svd
->vdev_stat
.vs_alloc
= 0;
756 svd
->vdev_stat
.vs_space
= 0;
757 svd
->vdev_stat
.vs_dspace
= 0;
759 for (t
= 0; t
< TXG_SIZE
; t
++) {
760 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
761 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
762 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
763 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
764 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
765 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
768 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
769 vdev_config_clean(svd
);
770 vdev_config_dirty(tvd
);
773 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
774 vdev_state_clean(svd
);
775 vdev_state_dirty(tvd
);
778 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
779 svd
->vdev_deflate_ratio
= 0;
781 tvd
->vdev_islog
= svd
->vdev_islog
;
786 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
795 for (c
= 0; c
< vd
->vdev_children
; c
++)
796 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
800 * Add a mirror/replacing vdev above an existing vdev.
803 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
805 spa_t
*spa
= cvd
->vdev_spa
;
806 vdev_t
*pvd
= cvd
->vdev_parent
;
809 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
811 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
813 mvd
->vdev_asize
= cvd
->vdev_asize
;
814 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
815 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
816 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
817 mvd
->vdev_state
= cvd
->vdev_state
;
818 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
820 vdev_remove_child(pvd
, cvd
);
821 vdev_add_child(pvd
, mvd
);
822 cvd
->vdev_id
= mvd
->vdev_children
;
823 vdev_add_child(mvd
, cvd
);
824 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
826 if (mvd
== mvd
->vdev_top
)
827 vdev_top_transfer(cvd
, mvd
);
833 * Remove a 1-way mirror/replacing vdev from the tree.
836 vdev_remove_parent(vdev_t
*cvd
)
838 vdev_t
*mvd
= cvd
->vdev_parent
;
839 vdev_t
*pvd
= mvd
->vdev_parent
;
841 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
843 ASSERT(mvd
->vdev_children
== 1);
844 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
845 mvd
->vdev_ops
== &vdev_replacing_ops
||
846 mvd
->vdev_ops
== &vdev_spare_ops
);
847 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
849 vdev_remove_child(mvd
, cvd
);
850 vdev_remove_child(pvd
, mvd
);
853 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
854 * Otherwise, we could have detached an offline device, and when we
855 * go to import the pool we'll think we have two top-level vdevs,
856 * instead of a different version of the same top-level vdev.
858 if (mvd
->vdev_top
== mvd
) {
859 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
860 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
861 cvd
->vdev_guid
+= guid_delta
;
862 cvd
->vdev_guid_sum
+= guid_delta
;
865 * If pool not set for autoexpand, we need to also preserve
866 * mvd's asize to prevent automatic expansion of cvd.
867 * Otherwise if we are adjusting the mirror by attaching and
868 * detaching children of non-uniform sizes, the mirror could
869 * autoexpand, unexpectedly requiring larger devices to
870 * re-establish the mirror.
872 if (!cvd
->vdev_spa
->spa_autoexpand
)
873 cvd
->vdev_asize
= mvd
->vdev_asize
;
875 cvd
->vdev_id
= mvd
->vdev_id
;
876 vdev_add_child(pvd
, cvd
);
877 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
879 if (cvd
== cvd
->vdev_top
)
880 vdev_top_transfer(mvd
, cvd
);
882 ASSERT(mvd
->vdev_children
== 0);
887 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
889 spa_t
*spa
= vd
->vdev_spa
;
890 objset_t
*mos
= spa
->spa_meta_objset
;
892 uint64_t oldc
= vd
->vdev_ms_count
;
893 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
897 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
900 * This vdev is not being allocated from yet or is a hole.
902 if (vd
->vdev_ms_shift
== 0)
905 ASSERT(!vd
->vdev_ishole
);
908 * Compute the raidz-deflation ratio. Note, we hard-code
909 * in 128k (1 << 17) because it is the "typical" blocksize.
910 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
911 * otherwise it would inconsistently account for existing bp's.
913 vd
->vdev_deflate_ratio
= (1 << 17) /
914 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
916 ASSERT(oldc
<= newc
);
918 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
921 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
922 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
926 vd
->vdev_ms_count
= newc
;
928 for (m
= oldc
; m
< newc
; m
++) {
932 error
= dmu_read(mos
, vd
->vdev_ms_array
,
933 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
939 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
946 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
949 * If the vdev is being removed we don't activate
950 * the metaslabs since we want to ensure that no new
951 * allocations are performed on this device.
953 if (oldc
== 0 && !vd
->vdev_removing
)
954 metaslab_group_activate(vd
->vdev_mg
);
957 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
963 vdev_metaslab_fini(vdev_t
*vd
)
966 uint64_t count
= vd
->vdev_ms_count
;
968 if (vd
->vdev_ms
!= NULL
) {
969 metaslab_group_passivate(vd
->vdev_mg
);
970 for (m
= 0; m
< count
; m
++) {
971 metaslab_t
*msp
= vd
->vdev_ms
[m
];
976 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
980 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
983 typedef struct vdev_probe_stats
{
984 boolean_t vps_readable
;
985 boolean_t vps_writeable
;
987 } vdev_probe_stats_t
;
990 vdev_probe_done(zio_t
*zio
)
992 spa_t
*spa
= zio
->io_spa
;
993 vdev_t
*vd
= zio
->io_vd
;
994 vdev_probe_stats_t
*vps
= zio
->io_private
;
996 ASSERT(vd
->vdev_probe_zio
!= NULL
);
998 if (zio
->io_type
== ZIO_TYPE_READ
) {
999 if (zio
->io_error
== 0)
1000 vps
->vps_readable
= 1;
1001 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1002 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1003 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1004 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1005 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1007 abd_free(zio
->io_abd
);
1009 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1010 if (zio
->io_error
== 0)
1011 vps
->vps_writeable
= 1;
1012 abd_free(zio
->io_abd
);
1013 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1017 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1018 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1020 if (vdev_readable(vd
) &&
1021 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1024 ASSERT(zio
->io_error
!= 0);
1025 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1026 spa
, vd
, NULL
, 0, 0);
1027 zio
->io_error
= SET_ERROR(ENXIO
);
1030 mutex_enter(&vd
->vdev_probe_lock
);
1031 ASSERT(vd
->vdev_probe_zio
== zio
);
1032 vd
->vdev_probe_zio
= NULL
;
1033 mutex_exit(&vd
->vdev_probe_lock
);
1036 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1037 if (!vdev_accessible(vd
, pio
))
1038 pio
->io_error
= SET_ERROR(ENXIO
);
1040 kmem_free(vps
, sizeof (*vps
));
1045 * Determine whether this device is accessible.
1047 * Read and write to several known locations: the pad regions of each
1048 * vdev label but the first, which we leave alone in case it contains
1052 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1054 spa_t
*spa
= vd
->vdev_spa
;
1055 vdev_probe_stats_t
*vps
= NULL
;
1059 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1062 * Don't probe the probe.
1064 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1068 * To prevent 'probe storms' when a device fails, we create
1069 * just one probe i/o at a time. All zios that want to probe
1070 * this vdev will become parents of the probe io.
1072 mutex_enter(&vd
->vdev_probe_lock
);
1074 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1075 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1077 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1078 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1081 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1083 * vdev_cant_read and vdev_cant_write can only
1084 * transition from TRUE to FALSE when we have the
1085 * SCL_ZIO lock as writer; otherwise they can only
1086 * transition from FALSE to TRUE. This ensures that
1087 * any zio looking at these values can assume that
1088 * failures persist for the life of the I/O. That's
1089 * important because when a device has intermittent
1090 * connectivity problems, we want to ensure that
1091 * they're ascribed to the device (ENXIO) and not
1094 * Since we hold SCL_ZIO as writer here, clear both
1095 * values so the probe can reevaluate from first
1098 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1099 vd
->vdev_cant_read
= B_FALSE
;
1100 vd
->vdev_cant_write
= B_FALSE
;
1103 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1104 vdev_probe_done
, vps
,
1105 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1108 * We can't change the vdev state in this context, so we
1109 * kick off an async task to do it on our behalf.
1112 vd
->vdev_probe_wanted
= B_TRUE
;
1113 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1118 zio_add_child(zio
, pio
);
1120 mutex_exit(&vd
->vdev_probe_lock
);
1123 ASSERT(zio
!= NULL
);
1127 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1128 zio_nowait(zio_read_phys(pio
, vd
,
1129 vdev_label_offset(vd
->vdev_psize
, l
,
1130 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1131 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1132 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1133 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1144 vdev_open_child(void *arg
)
1148 vd
->vdev_open_thread
= curthread
;
1149 vd
->vdev_open_error
= vdev_open(vd
);
1150 vd
->vdev_open_thread
= NULL
;
1154 vdev_uses_zvols(vdev_t
*vd
)
1159 if (zvol_is_zvol(vd
->vdev_path
))
1163 for (c
= 0; c
< vd
->vdev_children
; c
++)
1164 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1171 vdev_open_children(vdev_t
*vd
)
1174 int children
= vd
->vdev_children
;
1178 * in order to handle pools on top of zvols, do the opens
1179 * in a single thread so that the same thread holds the
1180 * spa_namespace_lock
1182 if (vdev_uses_zvols(vd
)) {
1184 for (c
= 0; c
< children
; c
++)
1185 vd
->vdev_child
[c
]->vdev_open_error
=
1186 vdev_open(vd
->vdev_child
[c
]);
1188 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1189 children
, children
, TASKQ_PREPOPULATE
);
1193 for (c
= 0; c
< children
; c
++)
1194 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1195 vd
->vdev_child
[c
], TQ_SLEEP
) != TASKQID_INVALID
);
1200 vd
->vdev_nonrot
= B_TRUE
;
1202 for (c
= 0; c
< children
; c
++)
1203 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1207 * Prepare a virtual device for access.
1210 vdev_open(vdev_t
*vd
)
1212 spa_t
*spa
= vd
->vdev_spa
;
1215 uint64_t max_osize
= 0;
1216 uint64_t asize
, max_asize
, psize
;
1217 uint64_t ashift
= 0;
1220 ASSERT(vd
->vdev_open_thread
== curthread
||
1221 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1222 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1223 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1224 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1226 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1227 vd
->vdev_cant_read
= B_FALSE
;
1228 vd
->vdev_cant_write
= B_FALSE
;
1229 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1232 * If this vdev is not removed, check its fault status. If it's
1233 * faulted, bail out of the open.
1235 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1236 ASSERT(vd
->vdev_children
== 0);
1237 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1238 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1239 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1240 vd
->vdev_label_aux
);
1241 return (SET_ERROR(ENXIO
));
1242 } else if (vd
->vdev_offline
) {
1243 ASSERT(vd
->vdev_children
== 0);
1244 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1245 return (SET_ERROR(ENXIO
));
1248 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1251 * Reset the vdev_reopening flag so that we actually close
1252 * the vdev on error.
1254 vd
->vdev_reopening
= B_FALSE
;
1255 if (zio_injection_enabled
&& error
== 0)
1256 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1259 if (vd
->vdev_removed
&&
1260 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1261 vd
->vdev_removed
= B_FALSE
;
1263 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1264 vd
->vdev_stat
.vs_aux
);
1268 vd
->vdev_removed
= B_FALSE
;
1271 * Recheck the faulted flag now that we have confirmed that
1272 * the vdev is accessible. If we're faulted, bail.
1274 if (vd
->vdev_faulted
) {
1275 ASSERT(vd
->vdev_children
== 0);
1276 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1277 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1278 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1279 vd
->vdev_label_aux
);
1280 return (SET_ERROR(ENXIO
));
1283 if (vd
->vdev_degraded
) {
1284 ASSERT(vd
->vdev_children
== 0);
1285 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1286 VDEV_AUX_ERR_EXCEEDED
);
1288 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1292 * For hole or missing vdevs we just return success.
1294 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1297 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1298 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1299 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1305 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1306 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1308 if (vd
->vdev_children
== 0) {
1309 if (osize
< SPA_MINDEVSIZE
) {
1310 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1311 VDEV_AUX_TOO_SMALL
);
1312 return (SET_ERROR(EOVERFLOW
));
1315 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1316 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1317 VDEV_LABEL_END_SIZE
);
1319 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1320 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1321 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1322 VDEV_AUX_TOO_SMALL
);
1323 return (SET_ERROR(EOVERFLOW
));
1327 max_asize
= max_osize
;
1330 vd
->vdev_psize
= psize
;
1333 * Make sure the allocatable size hasn't shrunk.
1335 if (asize
< vd
->vdev_min_asize
) {
1336 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1337 VDEV_AUX_BAD_LABEL
);
1338 return (SET_ERROR(EINVAL
));
1341 if (vd
->vdev_asize
== 0) {
1343 * This is the first-ever open, so use the computed values.
1344 * For compatibility, a different ashift can be requested.
1346 vd
->vdev_asize
= asize
;
1347 vd
->vdev_max_asize
= max_asize
;
1348 if (vd
->vdev_ashift
== 0)
1349 vd
->vdev_ashift
= ashift
;
1352 * Detect if the alignment requirement has increased.
1353 * We don't want to make the pool unavailable, just
1354 * post an event instead.
1356 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1357 vd
->vdev_ops
->vdev_op_leaf
) {
1358 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1359 spa
, vd
, NULL
, 0, 0);
1362 vd
->vdev_max_asize
= max_asize
;
1366 * If all children are healthy and the asize has increased,
1367 * then we've experienced dynamic LUN growth. If automatic
1368 * expansion is enabled then use the additional space.
1370 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1371 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1372 vd
->vdev_asize
= asize
;
1374 vdev_set_min_asize(vd
);
1377 * Ensure we can issue some IO before declaring the
1378 * vdev open for business.
1380 if (vd
->vdev_ops
->vdev_op_leaf
&&
1381 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1382 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1383 VDEV_AUX_ERR_EXCEEDED
);
1388 * Track the min and max ashift values for normal data devices.
1390 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1391 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1392 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1393 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1394 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1395 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1399 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1400 * resilver. But don't do this if we are doing a reopen for a scrub,
1401 * since this would just restart the scrub we are already doing.
1403 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1404 vdev_resilver_needed(vd
, NULL
, NULL
))
1405 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1411 * Called once the vdevs are all opened, this routine validates the label
1412 * contents. This needs to be done before vdev_load() so that we don't
1413 * inadvertently do repair I/Os to the wrong device.
1415 * If 'strict' is false ignore the spa guid check. This is necessary because
1416 * if the machine crashed during a re-guid the new guid might have been written
1417 * to all of the vdev labels, but not the cached config. The strict check
1418 * will be performed when the pool is opened again using the mos config.
1420 * This function will only return failure if one of the vdevs indicates that it
1421 * has since been destroyed or exported. This is only possible if
1422 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1423 * will be updated but the function will return 0.
1426 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1428 spa_t
*spa
= vd
->vdev_spa
;
1430 uint64_t guid
= 0, top_guid
;
1434 for (c
= 0; c
< vd
->vdev_children
; c
++)
1435 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1436 return (SET_ERROR(EBADF
));
1439 * If the device has already failed, or was marked offline, don't do
1440 * any further validation. Otherwise, label I/O will fail and we will
1441 * overwrite the previous state.
1443 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1444 uint64_t aux_guid
= 0;
1446 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1447 spa_last_synced_txg(spa
) : -1ULL;
1449 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1450 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1451 VDEV_AUX_BAD_LABEL
);
1456 * Determine if this vdev has been split off into another
1457 * pool. If so, then refuse to open it.
1459 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1460 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1461 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1462 VDEV_AUX_SPLIT_POOL
);
1467 if (strict
&& (nvlist_lookup_uint64(label
,
1468 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1469 guid
!= spa_guid(spa
))) {
1470 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1471 VDEV_AUX_CORRUPT_DATA
);
1476 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1477 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1482 * If this vdev just became a top-level vdev because its
1483 * sibling was detached, it will have adopted the parent's
1484 * vdev guid -- but the label may or may not be on disk yet.
1485 * Fortunately, either version of the label will have the
1486 * same top guid, so if we're a top-level vdev, we can
1487 * safely compare to that instead.
1489 * If we split this vdev off instead, then we also check the
1490 * original pool's guid. We don't want to consider the vdev
1491 * corrupt if it is partway through a split operation.
1493 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1495 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1497 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1498 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1499 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1500 VDEV_AUX_CORRUPT_DATA
);
1505 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1507 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1508 VDEV_AUX_CORRUPT_DATA
);
1516 * If this is a verbatim import, no need to check the
1517 * state of the pool.
1519 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1520 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1521 state
!= POOL_STATE_ACTIVE
)
1522 return (SET_ERROR(EBADF
));
1525 * If we were able to open and validate a vdev that was
1526 * previously marked permanently unavailable, clear that state
1529 if (vd
->vdev_not_present
)
1530 vd
->vdev_not_present
= 0;
1537 * Close a virtual device.
1540 vdev_close(vdev_t
*vd
)
1542 vdev_t
*pvd
= vd
->vdev_parent
;
1543 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1545 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1548 * If our parent is reopening, then we are as well, unless we are
1551 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1552 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1554 vd
->vdev_ops
->vdev_op_close(vd
);
1556 vdev_cache_purge(vd
);
1559 * We record the previous state before we close it, so that if we are
1560 * doing a reopen(), we don't generate FMA ereports if we notice that
1561 * it's still faulted.
1563 vd
->vdev_prevstate
= vd
->vdev_state
;
1565 if (vd
->vdev_offline
)
1566 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1568 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1569 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1573 vdev_hold(vdev_t
*vd
)
1575 spa_t
*spa
= vd
->vdev_spa
;
1578 ASSERT(spa_is_root(spa
));
1579 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1582 for (c
= 0; c
< vd
->vdev_children
; c
++)
1583 vdev_hold(vd
->vdev_child
[c
]);
1585 if (vd
->vdev_ops
->vdev_op_leaf
)
1586 vd
->vdev_ops
->vdev_op_hold(vd
);
1590 vdev_rele(vdev_t
*vd
)
1594 ASSERT(spa_is_root(vd
->vdev_spa
));
1595 for (c
= 0; c
< vd
->vdev_children
; c
++)
1596 vdev_rele(vd
->vdev_child
[c
]);
1598 if (vd
->vdev_ops
->vdev_op_leaf
)
1599 vd
->vdev_ops
->vdev_op_rele(vd
);
1603 * Reopen all interior vdevs and any unopened leaves. We don't actually
1604 * reopen leaf vdevs which had previously been opened as they might deadlock
1605 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1606 * If the leaf has never been opened then open it, as usual.
1609 vdev_reopen(vdev_t
*vd
)
1611 spa_t
*spa
= vd
->vdev_spa
;
1613 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1615 /* set the reopening flag unless we're taking the vdev offline */
1616 vd
->vdev_reopening
= !vd
->vdev_offline
;
1618 (void) vdev_open(vd
);
1621 * Call vdev_validate() here to make sure we have the same device.
1622 * Otherwise, a device with an invalid label could be successfully
1623 * opened in response to vdev_reopen().
1626 (void) vdev_validate_aux(vd
);
1627 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1628 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1629 !l2arc_vdev_present(vd
))
1630 l2arc_add_vdev(spa
, vd
);
1632 (void) vdev_validate(vd
, B_TRUE
);
1636 * Reassess parent vdev's health.
1638 vdev_propagate_state(vd
);
1642 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1647 * Normally, partial opens (e.g. of a mirror) are allowed.
1648 * For a create, however, we want to fail the request if
1649 * there are any components we can't open.
1651 error
= vdev_open(vd
);
1653 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1655 return (error
? error
: ENXIO
);
1659 * Recursively load DTLs and initialize all labels.
1661 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1662 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1663 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1672 vdev_metaslab_set_size(vdev_t
*vd
)
1675 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1677 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1678 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1682 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1684 ASSERT(vd
== vd
->vdev_top
);
1685 ASSERT(!vd
->vdev_ishole
);
1686 ASSERT(ISP2(flags
));
1687 ASSERT(spa_writeable(vd
->vdev_spa
));
1689 if (flags
& VDD_METASLAB
)
1690 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1692 if (flags
& VDD_DTL
)
1693 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1695 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1699 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1703 for (c
= 0; c
< vd
->vdev_children
; c
++)
1704 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1706 if (vd
->vdev_ops
->vdev_op_leaf
)
1707 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1713 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1714 * the vdev has less than perfect replication. There are four kinds of DTL:
1716 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1718 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1720 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1721 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1722 * txgs that was scrubbed.
1724 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1725 * persistent errors or just some device being offline.
1726 * Unlike the other three, the DTL_OUTAGE map is not generally
1727 * maintained; it's only computed when needed, typically to
1728 * determine whether a device can be detached.
1730 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1731 * either has the data or it doesn't.
1733 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1734 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1735 * if any child is less than fully replicated, then so is its parent.
1736 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1737 * comprising only those txgs which appear in 'maxfaults' or more children;
1738 * those are the txgs we don't have enough replication to read. For example,
1739 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1740 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1741 * two child DTL_MISSING maps.
1743 * It should be clear from the above that to compute the DTLs and outage maps
1744 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1745 * Therefore, that is all we keep on disk. When loading the pool, or after
1746 * a configuration change, we generate all other DTLs from first principles.
1749 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1751 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1753 ASSERT(t
< DTL_TYPES
);
1754 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1755 ASSERT(spa_writeable(vd
->vdev_spa
));
1757 mutex_enter(rt
->rt_lock
);
1758 if (!range_tree_contains(rt
, txg
, size
))
1759 range_tree_add(rt
, txg
, size
);
1760 mutex_exit(rt
->rt_lock
);
1764 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1766 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1767 boolean_t dirty
= B_FALSE
;
1769 ASSERT(t
< DTL_TYPES
);
1770 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1772 mutex_enter(rt
->rt_lock
);
1773 if (range_tree_space(rt
) != 0)
1774 dirty
= range_tree_contains(rt
, txg
, size
);
1775 mutex_exit(rt
->rt_lock
);
1781 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1783 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1786 mutex_enter(rt
->rt_lock
);
1787 empty
= (range_tree_space(rt
) == 0);
1788 mutex_exit(rt
->rt_lock
);
1794 * Returns the lowest txg in the DTL range.
1797 vdev_dtl_min(vdev_t
*vd
)
1801 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1802 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1803 ASSERT0(vd
->vdev_children
);
1805 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1806 return (rs
->rs_start
- 1);
1810 * Returns the highest txg in the DTL.
1813 vdev_dtl_max(vdev_t
*vd
)
1817 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1818 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1819 ASSERT0(vd
->vdev_children
);
1821 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1822 return (rs
->rs_end
);
1826 * Determine if a resilvering vdev should remove any DTL entries from
1827 * its range. If the vdev was resilvering for the entire duration of the
1828 * scan then it should excise that range from its DTLs. Otherwise, this
1829 * vdev is considered partially resilvered and should leave its DTL
1830 * entries intact. The comment in vdev_dtl_reassess() describes how we
1834 vdev_dtl_should_excise(vdev_t
*vd
)
1836 spa_t
*spa
= vd
->vdev_spa
;
1837 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1839 ASSERT0(scn
->scn_phys
.scn_errors
);
1840 ASSERT0(vd
->vdev_children
);
1842 if (vd
->vdev_resilver_txg
== 0 ||
1843 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1847 * When a resilver is initiated the scan will assign the scn_max_txg
1848 * value to the highest txg value that exists in all DTLs. If this
1849 * device's max DTL is not part of this scan (i.e. it is not in
1850 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1853 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1854 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1855 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1856 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1863 * Reassess DTLs after a config change or scrub completion.
1866 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1868 spa_t
*spa
= vd
->vdev_spa
;
1872 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1874 for (c
= 0; c
< vd
->vdev_children
; c
++)
1875 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1876 scrub_txg
, scrub_done
);
1878 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1881 if (vd
->vdev_ops
->vdev_op_leaf
) {
1882 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1884 mutex_enter(&vd
->vdev_dtl_lock
);
1887 * If we've completed a scan cleanly then determine
1888 * if this vdev should remove any DTLs. We only want to
1889 * excise regions on vdevs that were available during
1890 * the entire duration of this scan.
1892 if (scrub_txg
!= 0 &&
1893 (spa
->spa_scrub_started
||
1894 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1895 vdev_dtl_should_excise(vd
)) {
1897 * We completed a scrub up to scrub_txg. If we
1898 * did it without rebooting, then the scrub dtl
1899 * will be valid, so excise the old region and
1900 * fold in the scrub dtl. Otherwise, leave the
1901 * dtl as-is if there was an error.
1903 * There's little trick here: to excise the beginning
1904 * of the DTL_MISSING map, we put it into a reference
1905 * tree and then add a segment with refcnt -1 that
1906 * covers the range [0, scrub_txg). This means
1907 * that each txg in that range has refcnt -1 or 0.
1908 * We then add DTL_SCRUB with a refcnt of 2, so that
1909 * entries in the range [0, scrub_txg) will have a
1910 * positive refcnt -- either 1 or 2. We then convert
1911 * the reference tree into the new DTL_MISSING map.
1913 space_reftree_create(&reftree
);
1914 space_reftree_add_map(&reftree
,
1915 vd
->vdev_dtl
[DTL_MISSING
], 1);
1916 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1917 space_reftree_add_map(&reftree
,
1918 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1919 space_reftree_generate_map(&reftree
,
1920 vd
->vdev_dtl
[DTL_MISSING
], 1);
1921 space_reftree_destroy(&reftree
);
1923 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1924 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1925 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1927 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1928 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1929 if (!vdev_readable(vd
))
1930 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1932 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1933 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1936 * If the vdev was resilvering and no longer has any
1937 * DTLs then reset its resilvering flag and dirty
1938 * the top level so that we persist the change.
1940 if (vd
->vdev_resilver_txg
!= 0 &&
1941 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1942 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
1943 vd
->vdev_resilver_txg
= 0;
1944 vdev_config_dirty(vd
->vdev_top
);
1947 mutex_exit(&vd
->vdev_dtl_lock
);
1950 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1954 mutex_enter(&vd
->vdev_dtl_lock
);
1955 for (t
= 0; t
< DTL_TYPES
; t
++) {
1958 /* account for child's outage in parent's missing map */
1959 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1961 continue; /* leaf vdevs only */
1962 if (t
== DTL_PARTIAL
)
1963 minref
= 1; /* i.e. non-zero */
1964 else if (vd
->vdev_nparity
!= 0)
1965 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1967 minref
= vd
->vdev_children
; /* any kind of mirror */
1968 space_reftree_create(&reftree
);
1969 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1970 vdev_t
*cvd
= vd
->vdev_child
[c
];
1971 mutex_enter(&cvd
->vdev_dtl_lock
);
1972 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1973 mutex_exit(&cvd
->vdev_dtl_lock
);
1975 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1976 space_reftree_destroy(&reftree
);
1978 mutex_exit(&vd
->vdev_dtl_lock
);
1982 vdev_dtl_load(vdev_t
*vd
)
1984 spa_t
*spa
= vd
->vdev_spa
;
1985 objset_t
*mos
= spa
->spa_meta_objset
;
1989 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
1990 ASSERT(!vd
->vdev_ishole
);
1992 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
1993 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
1996 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
1998 mutex_enter(&vd
->vdev_dtl_lock
);
2001 * Now that we've opened the space_map we need to update
2004 space_map_update(vd
->vdev_dtl_sm
);
2006 error
= space_map_load(vd
->vdev_dtl_sm
,
2007 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2008 mutex_exit(&vd
->vdev_dtl_lock
);
2013 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2014 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2023 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2025 spa_t
*spa
= vd
->vdev_spa
;
2027 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2028 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2033 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2035 spa_t
*spa
= vd
->vdev_spa
;
2036 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2037 DMU_OT_NONE
, 0, tx
);
2040 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2047 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2051 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2052 vd
->vdev_ops
!= &vdev_missing_ops
&&
2053 vd
->vdev_ops
!= &vdev_root_ops
&&
2054 !vd
->vdev_top
->vdev_removing
) {
2055 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2056 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2058 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2059 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2062 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2063 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2068 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2070 spa_t
*spa
= vd
->vdev_spa
;
2071 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2072 objset_t
*mos
= spa
->spa_meta_objset
;
2073 range_tree_t
*rtsync
;
2076 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2078 ASSERT(!vd
->vdev_ishole
);
2079 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2081 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2083 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2084 mutex_enter(&vd
->vdev_dtl_lock
);
2085 space_map_free(vd
->vdev_dtl_sm
, tx
);
2086 space_map_close(vd
->vdev_dtl_sm
);
2087 vd
->vdev_dtl_sm
= NULL
;
2088 mutex_exit(&vd
->vdev_dtl_lock
);
2091 * We only destroy the leaf ZAP for detached leaves or for
2092 * removed log devices. Removed data devices handle leaf ZAP
2093 * cleanup later, once cancellation is no longer possible.
2095 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2096 vd
->vdev_top
->vdev_islog
)) {
2097 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2098 vd
->vdev_leaf_zap
= 0;
2105 if (vd
->vdev_dtl_sm
== NULL
) {
2106 uint64_t new_object
;
2108 new_object
= space_map_alloc(mos
, tx
);
2109 VERIFY3U(new_object
, !=, 0);
2111 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2112 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2113 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2116 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2118 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2120 mutex_enter(&rtlock
);
2122 mutex_enter(&vd
->vdev_dtl_lock
);
2123 range_tree_walk(rt
, range_tree_add
, rtsync
);
2124 mutex_exit(&vd
->vdev_dtl_lock
);
2126 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2127 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2128 range_tree_vacate(rtsync
, NULL
, NULL
);
2130 range_tree_destroy(rtsync
);
2132 mutex_exit(&rtlock
);
2133 mutex_destroy(&rtlock
);
2136 * If the object for the space map has changed then dirty
2137 * the top level so that we update the config.
2139 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2140 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2141 "new object %llu", txg
, spa_name(spa
), object
,
2142 space_map_object(vd
->vdev_dtl_sm
));
2143 vdev_config_dirty(vd
->vdev_top
);
2148 mutex_enter(&vd
->vdev_dtl_lock
);
2149 space_map_update(vd
->vdev_dtl_sm
);
2150 mutex_exit(&vd
->vdev_dtl_lock
);
2154 * Determine whether the specified vdev can be offlined/detached/removed
2155 * without losing data.
2158 vdev_dtl_required(vdev_t
*vd
)
2160 spa_t
*spa
= vd
->vdev_spa
;
2161 vdev_t
*tvd
= vd
->vdev_top
;
2162 uint8_t cant_read
= vd
->vdev_cant_read
;
2165 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2167 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2171 * Temporarily mark the device as unreadable, and then determine
2172 * whether this results in any DTL outages in the top-level vdev.
2173 * If not, we can safely offline/detach/remove the device.
2175 vd
->vdev_cant_read
= B_TRUE
;
2176 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2177 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2178 vd
->vdev_cant_read
= cant_read
;
2179 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2181 if (!required
&& zio_injection_enabled
)
2182 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2188 * Determine if resilver is needed, and if so the txg range.
2191 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2193 boolean_t needed
= B_FALSE
;
2194 uint64_t thismin
= UINT64_MAX
;
2195 uint64_t thismax
= 0;
2198 if (vd
->vdev_children
== 0) {
2199 mutex_enter(&vd
->vdev_dtl_lock
);
2200 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2201 vdev_writeable(vd
)) {
2203 thismin
= vdev_dtl_min(vd
);
2204 thismax
= vdev_dtl_max(vd
);
2207 mutex_exit(&vd
->vdev_dtl_lock
);
2209 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2210 vdev_t
*cvd
= vd
->vdev_child
[c
];
2211 uint64_t cmin
, cmax
;
2213 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2214 thismin
= MIN(thismin
, cmin
);
2215 thismax
= MAX(thismax
, cmax
);
2221 if (needed
&& minp
) {
2229 vdev_load(vdev_t
*vd
)
2234 * Recursively load all children.
2236 for (c
= 0; c
< vd
->vdev_children
; c
++)
2237 vdev_load(vd
->vdev_child
[c
]);
2240 * If this is a top-level vdev, initialize its metaslabs.
2242 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2243 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2244 vdev_metaslab_init(vd
, 0) != 0))
2245 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2246 VDEV_AUX_CORRUPT_DATA
);
2248 * If this is a leaf vdev, load its DTL.
2250 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2251 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2252 VDEV_AUX_CORRUPT_DATA
);
2256 * The special vdev case is used for hot spares and l2cache devices. Its
2257 * sole purpose it to set the vdev state for the associated vdev. To do this,
2258 * we make sure that we can open the underlying device, then try to read the
2259 * label, and make sure that the label is sane and that it hasn't been
2260 * repurposed to another pool.
2263 vdev_validate_aux(vdev_t
*vd
)
2266 uint64_t guid
, version
;
2269 if (!vdev_readable(vd
))
2272 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2273 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2274 VDEV_AUX_CORRUPT_DATA
);
2278 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2279 !SPA_VERSION_IS_SUPPORTED(version
) ||
2280 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2281 guid
!= vd
->vdev_guid
||
2282 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2283 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2284 VDEV_AUX_CORRUPT_DATA
);
2290 * We don't actually check the pool state here. If it's in fact in
2291 * use by another pool, we update this fact on the fly when requested.
2298 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2300 spa_t
*spa
= vd
->vdev_spa
;
2301 objset_t
*mos
= spa
->spa_meta_objset
;
2305 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2306 ASSERT(vd
== vd
->vdev_top
);
2307 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2309 if (vd
->vdev_ms
!= NULL
) {
2310 metaslab_group_t
*mg
= vd
->vdev_mg
;
2312 metaslab_group_histogram_verify(mg
);
2313 metaslab_class_histogram_verify(mg
->mg_class
);
2315 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2316 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2318 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2321 mutex_enter(&msp
->ms_lock
);
2323 * If the metaslab was not loaded when the vdev
2324 * was removed then the histogram accounting may
2325 * not be accurate. Update the histogram information
2326 * here so that we ensure that the metaslab group
2327 * and metaslab class are up-to-date.
2329 metaslab_group_histogram_remove(mg
, msp
);
2331 VERIFY0(space_map_allocated(msp
->ms_sm
));
2332 space_map_free(msp
->ms_sm
, tx
);
2333 space_map_close(msp
->ms_sm
);
2335 mutex_exit(&msp
->ms_lock
);
2338 metaslab_group_histogram_verify(mg
);
2339 metaslab_class_histogram_verify(mg
->mg_class
);
2340 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2341 ASSERT0(mg
->mg_histogram
[i
]);
2345 if (vd
->vdev_ms_array
) {
2346 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2347 vd
->vdev_ms_array
= 0;
2350 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2351 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2352 vd
->vdev_top_zap
= 0;
2358 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2361 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2363 ASSERT(!vd
->vdev_ishole
);
2365 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2366 metaslab_sync_done(msp
, txg
);
2369 metaslab_sync_reassess(vd
->vdev_mg
);
2373 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2375 spa_t
*spa
= vd
->vdev_spa
;
2380 ASSERT(!vd
->vdev_ishole
);
2382 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2383 ASSERT(vd
== vd
->vdev_top
);
2384 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2385 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2386 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2387 ASSERT(vd
->vdev_ms_array
!= 0);
2388 vdev_config_dirty(vd
);
2393 * Remove the metadata associated with this vdev once it's empty.
2395 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2396 vdev_remove(vd
, txg
);
2398 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2399 metaslab_sync(msp
, txg
);
2400 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2403 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2404 vdev_dtl_sync(lvd
, txg
);
2406 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2410 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2412 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2416 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2417 * not be opened, and no I/O is attempted.
2420 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2424 spa_vdev_state_enter(spa
, SCL_NONE
);
2426 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2427 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2429 if (!vd
->vdev_ops
->vdev_op_leaf
)
2430 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2435 * We don't directly use the aux state here, but if we do a
2436 * vdev_reopen(), we need this value to be present to remember why we
2439 vd
->vdev_label_aux
= aux
;
2442 * Faulted state takes precedence over degraded.
2444 vd
->vdev_delayed_close
= B_FALSE
;
2445 vd
->vdev_faulted
= 1ULL;
2446 vd
->vdev_degraded
= 0ULL;
2447 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2450 * If this device has the only valid copy of the data, then
2451 * back off and simply mark the vdev as degraded instead.
2453 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2454 vd
->vdev_degraded
= 1ULL;
2455 vd
->vdev_faulted
= 0ULL;
2458 * If we reopen the device and it's not dead, only then do we
2463 if (vdev_readable(vd
))
2464 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2467 return (spa_vdev_state_exit(spa
, vd
, 0));
2471 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2472 * user that something is wrong. The vdev continues to operate as normal as far
2473 * as I/O is concerned.
2476 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2480 spa_vdev_state_enter(spa
, SCL_NONE
);
2482 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2483 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2485 if (!vd
->vdev_ops
->vdev_op_leaf
)
2486 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2489 * If the vdev is already faulted, then don't do anything.
2491 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2492 return (spa_vdev_state_exit(spa
, NULL
, 0));
2494 vd
->vdev_degraded
= 1ULL;
2495 if (!vdev_is_dead(vd
))
2496 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2499 return (spa_vdev_state_exit(spa
, vd
, 0));
2503 * Online the given vdev.
2505 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2506 * spare device should be detached when the device finishes resilvering.
2507 * Second, the online should be treated like a 'test' online case, so no FMA
2508 * events are generated if the device fails to open.
2511 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2513 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2514 boolean_t postevent
= B_FALSE
;
2516 spa_vdev_state_enter(spa
, SCL_NONE
);
2518 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2519 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2521 if (!vd
->vdev_ops
->vdev_op_leaf
)
2522 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2525 (vd
->vdev_offline
== B_TRUE
|| vd
->vdev_tmpoffline
== B_TRUE
) ?
2529 vd
->vdev_offline
= B_FALSE
;
2530 vd
->vdev_tmpoffline
= B_FALSE
;
2531 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2532 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2534 /* XXX - L2ARC 1.0 does not support expansion */
2535 if (!vd
->vdev_aux
) {
2536 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2537 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2541 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2543 if (!vd
->vdev_aux
) {
2544 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2545 pvd
->vdev_expanding
= B_FALSE
;
2549 *newstate
= vd
->vdev_state
;
2550 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2551 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2552 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2553 vd
->vdev_parent
->vdev_child
[0] == vd
)
2554 vd
->vdev_unspare
= B_TRUE
;
2556 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2558 /* XXX - L2ARC 1.0 does not support expansion */
2560 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2561 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2565 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2567 return (spa_vdev_state_exit(spa
, vd
, 0));
2571 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2575 uint64_t generation
;
2576 metaslab_group_t
*mg
;
2579 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2581 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2582 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2584 if (!vd
->vdev_ops
->vdev_op_leaf
)
2585 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2589 generation
= spa
->spa_config_generation
+ 1;
2592 * If the device isn't already offline, try to offline it.
2594 if (!vd
->vdev_offline
) {
2596 * If this device has the only valid copy of some data,
2597 * don't allow it to be offlined. Log devices are always
2600 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2601 vdev_dtl_required(vd
))
2602 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2605 * If the top-level is a slog and it has had allocations
2606 * then proceed. We check that the vdev's metaslab group
2607 * is not NULL since it's possible that we may have just
2608 * added this vdev but not yet initialized its metaslabs.
2610 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2612 * Prevent any future allocations.
2614 metaslab_group_passivate(mg
);
2615 (void) spa_vdev_state_exit(spa
, vd
, 0);
2617 error
= spa_offline_log(spa
);
2619 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2622 * Check to see if the config has changed.
2624 if (error
|| generation
!= spa
->spa_config_generation
) {
2625 metaslab_group_activate(mg
);
2627 return (spa_vdev_state_exit(spa
,
2629 (void) spa_vdev_state_exit(spa
, vd
, 0);
2632 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2636 * Offline this device and reopen its top-level vdev.
2637 * If the top-level vdev is a log device then just offline
2638 * it. Otherwise, if this action results in the top-level
2639 * vdev becoming unusable, undo it and fail the request.
2641 vd
->vdev_offline
= B_TRUE
;
2644 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2645 vdev_is_dead(tvd
)) {
2646 vd
->vdev_offline
= B_FALSE
;
2648 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2652 * Add the device back into the metaslab rotor so that
2653 * once we online the device it's open for business.
2655 if (tvd
->vdev_islog
&& mg
!= NULL
)
2656 metaslab_group_activate(mg
);
2659 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2661 return (spa_vdev_state_exit(spa
, vd
, 0));
2665 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2669 mutex_enter(&spa
->spa_vdev_top_lock
);
2670 error
= vdev_offline_locked(spa
, guid
, flags
);
2671 mutex_exit(&spa
->spa_vdev_top_lock
);
2677 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2678 * vdev_offline(), we assume the spa config is locked. We also clear all
2679 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2682 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2684 vdev_t
*rvd
= spa
->spa_root_vdev
;
2687 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2692 vd
->vdev_stat
.vs_read_errors
= 0;
2693 vd
->vdev_stat
.vs_write_errors
= 0;
2694 vd
->vdev_stat
.vs_checksum_errors
= 0;
2696 for (c
= 0; c
< vd
->vdev_children
; c
++)
2697 vdev_clear(spa
, vd
->vdev_child
[c
]);
2700 * If we're in the FAULTED state or have experienced failed I/O, then
2701 * clear the persistent state and attempt to reopen the device. We
2702 * also mark the vdev config dirty, so that the new faulted state is
2703 * written out to disk.
2705 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2706 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2709 * When reopening in response to a clear event, it may be due to
2710 * a fmadm repair request. In this case, if the device is
2711 * still broken, we want to still post the ereport again.
2713 vd
->vdev_forcefault
= B_TRUE
;
2715 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2716 vd
->vdev_cant_read
= B_FALSE
;
2717 vd
->vdev_cant_write
= B_FALSE
;
2719 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2721 vd
->vdev_forcefault
= B_FALSE
;
2723 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2724 vdev_state_dirty(vd
->vdev_top
);
2726 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2727 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2729 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2733 * When clearing a FMA-diagnosed fault, we always want to
2734 * unspare the device, as we assume that the original spare was
2735 * done in response to the FMA fault.
2737 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2738 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2739 vd
->vdev_parent
->vdev_child
[0] == vd
)
2740 vd
->vdev_unspare
= B_TRUE
;
2744 vdev_is_dead(vdev_t
*vd
)
2747 * Holes and missing devices are always considered "dead".
2748 * This simplifies the code since we don't have to check for
2749 * these types of devices in the various code paths.
2750 * Instead we rely on the fact that we skip over dead devices
2751 * before issuing I/O to them.
2753 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2754 vd
->vdev_ops
== &vdev_missing_ops
);
2758 vdev_readable(vdev_t
*vd
)
2760 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2764 vdev_writeable(vdev_t
*vd
)
2766 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2770 vdev_allocatable(vdev_t
*vd
)
2772 uint64_t state
= vd
->vdev_state
;
2775 * We currently allow allocations from vdevs which may be in the
2776 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2777 * fails to reopen then we'll catch it later when we're holding
2778 * the proper locks. Note that we have to get the vdev state
2779 * in a local variable because although it changes atomically,
2780 * we're asking two separate questions about it.
2782 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2783 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
&&
2784 vd
->vdev_mg
->mg_initialized
);
2788 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2790 ASSERT(zio
->io_vd
== vd
);
2792 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2795 if (zio
->io_type
== ZIO_TYPE_READ
)
2796 return (!vd
->vdev_cant_read
);
2798 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2799 return (!vd
->vdev_cant_write
);
2805 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2808 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2809 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2810 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2813 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2817 * Get extended stats
2820 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2823 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2824 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2825 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2827 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2828 vsx
->vsx_total_histo
[t
][b
] +=
2829 cvsx
->vsx_total_histo
[t
][b
];
2833 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2834 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2835 vsx
->vsx_queue_histo
[t
][b
] +=
2836 cvsx
->vsx_queue_histo
[t
][b
];
2838 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2839 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2841 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2842 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2844 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2845 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2851 * Get statistics for the given vdev.
2854 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2858 * If we're getting stats on the root vdev, aggregate the I/O counts
2859 * over all top-level vdevs (i.e. the direct children of the root).
2861 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2863 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2864 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2867 memset(vsx
, 0, sizeof (*vsx
));
2869 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2870 vdev_t
*cvd
= vd
->vdev_child
[c
];
2871 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2872 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2874 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2876 vdev_get_child_stat(cvd
, vs
, cvs
);
2878 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2883 * We're a leaf. Just copy our ZIO active queue stats in. The
2884 * other leaf stats are updated in vdev_stat_update().
2889 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2891 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2892 vsx
->vsx_active_queue
[t
] =
2893 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2894 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2895 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2901 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2903 vdev_t
*tvd
= vd
->vdev_top
;
2904 mutex_enter(&vd
->vdev_stat_lock
);
2906 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2907 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2908 vs
->vs_state
= vd
->vdev_state
;
2909 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2910 if (vd
->vdev_ops
->vdev_op_leaf
)
2911 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
2912 VDEV_LABEL_END_SIZE
;
2914 * Report expandable space on top-level, non-auxillary devices
2915 * only. The expandable space is reported in terms of metaslab
2916 * sized units since that determines how much space the pool
2919 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
2920 vs
->vs_esize
= P2ALIGN(
2921 vd
->vdev_max_asize
- vd
->vdev_asize
,
2922 1ULL << tvd
->vdev_ms_shift
);
2924 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2925 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
2927 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2931 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
2932 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
2933 mutex_exit(&vd
->vdev_stat_lock
);
2937 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2939 return (vdev_get_stats_ex(vd
, vs
, NULL
));
2943 vdev_clear_stats(vdev_t
*vd
)
2945 mutex_enter(&vd
->vdev_stat_lock
);
2946 vd
->vdev_stat
.vs_space
= 0;
2947 vd
->vdev_stat
.vs_dspace
= 0;
2948 vd
->vdev_stat
.vs_alloc
= 0;
2949 mutex_exit(&vd
->vdev_stat_lock
);
2953 vdev_scan_stat_init(vdev_t
*vd
)
2955 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2958 for (c
= 0; c
< vd
->vdev_children
; c
++)
2959 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2961 mutex_enter(&vd
->vdev_stat_lock
);
2962 vs
->vs_scan_processed
= 0;
2963 mutex_exit(&vd
->vdev_stat_lock
);
2967 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2969 spa_t
*spa
= zio
->io_spa
;
2970 vdev_t
*rvd
= spa
->spa_root_vdev
;
2971 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2973 uint64_t txg
= zio
->io_txg
;
2974 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2975 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
2976 zio_type_t type
= zio
->io_type
;
2977 int flags
= zio
->io_flags
;
2980 * If this i/o is a gang leader, it didn't do any actual work.
2982 if (zio
->io_gang_tree
)
2985 if (zio
->io_error
== 0) {
2987 * If this is a root i/o, don't count it -- we've already
2988 * counted the top-level vdevs, and vdev_get_stats() will
2989 * aggregate them when asked. This reduces contention on
2990 * the root vdev_stat_lock and implicitly handles blocks
2991 * that compress away to holes, for which there is no i/o.
2992 * (Holes never create vdev children, so all the counters
2993 * remain zero, which is what we want.)
2995 * Note: this only applies to successful i/o (io_error == 0)
2996 * because unlike i/o counts, errors are not additive.
2997 * When reading a ditto block, for example, failure of
2998 * one top-level vdev does not imply a root-level error.
3003 ASSERT(vd
== zio
->io_vd
);
3005 if (flags
& ZIO_FLAG_IO_BYPASS
)
3008 mutex_enter(&vd
->vdev_stat_lock
);
3010 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3011 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3012 dsl_scan_phys_t
*scn_phys
=
3013 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3014 uint64_t *processed
= &scn_phys
->scn_processed
;
3017 if (vd
->vdev_ops
->vdev_op_leaf
)
3018 atomic_add_64(processed
, psize
);
3019 vs
->vs_scan_processed
+= psize
;
3022 if (flags
& ZIO_FLAG_SELF_HEAL
)
3023 vs
->vs_self_healed
+= psize
;
3027 * The bytes/ops/histograms are recorded at the leaf level and
3028 * aggregated into the higher level vdevs in vdev_get_stats().
3030 if (vd
->vdev_ops
->vdev_op_leaf
&&
3031 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3034 vs
->vs_bytes
[type
] += psize
;
3036 if (flags
& ZIO_FLAG_DELEGATED
) {
3037 vsx
->vsx_agg_histo
[zio
->io_priority
]
3038 [RQ_HISTO(zio
->io_size
)]++;
3040 vsx
->vsx_ind_histo
[zio
->io_priority
]
3041 [RQ_HISTO(zio
->io_size
)]++;
3044 if (zio
->io_delta
&& zio
->io_delay
) {
3045 vsx
->vsx_queue_histo
[zio
->io_priority
]
3046 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3047 vsx
->vsx_disk_histo
[type
]
3048 [L_HISTO(zio
->io_delay
)]++;
3049 vsx
->vsx_total_histo
[type
]
3050 [L_HISTO(zio
->io_delta
)]++;
3054 mutex_exit(&vd
->vdev_stat_lock
);
3058 if (flags
& ZIO_FLAG_SPECULATIVE
)
3062 * If this is an I/O error that is going to be retried, then ignore the
3063 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3064 * hard errors, when in reality they can happen for any number of
3065 * innocuous reasons (bus resets, MPxIO link failure, etc).
3067 if (zio
->io_error
== EIO
&&
3068 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3072 * Intent logs writes won't propagate their error to the root
3073 * I/O so don't mark these types of failures as pool-level
3076 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3079 mutex_enter(&vd
->vdev_stat_lock
);
3080 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3081 if (zio
->io_error
== ECKSUM
)
3082 vs
->vs_checksum_errors
++;
3084 vs
->vs_read_errors
++;
3086 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3087 vs
->vs_write_errors
++;
3088 mutex_exit(&vd
->vdev_stat_lock
);
3090 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3091 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3092 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3093 spa
->spa_claiming
)) {
3095 * This is either a normal write (not a repair), or it's
3096 * a repair induced by the scrub thread, or it's a repair
3097 * made by zil_claim() during spa_load() in the first txg.
3098 * In the normal case, we commit the DTL change in the same
3099 * txg as the block was born. In the scrub-induced repair
3100 * case, we know that scrubs run in first-pass syncing context,
3101 * so we commit the DTL change in spa_syncing_txg(spa).
3102 * In the zil_claim() case, we commit in spa_first_txg(spa).
3104 * We currently do not make DTL entries for failed spontaneous
3105 * self-healing writes triggered by normal (non-scrubbing)
3106 * reads, because we have no transactional context in which to
3107 * do so -- and it's not clear that it'd be desirable anyway.
3109 if (vd
->vdev_ops
->vdev_op_leaf
) {
3110 uint64_t commit_txg
= txg
;
3111 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3112 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3113 ASSERT(spa_sync_pass(spa
) == 1);
3114 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3115 commit_txg
= spa_syncing_txg(spa
);
3116 } else if (spa
->spa_claiming
) {
3117 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3118 commit_txg
= spa_first_txg(spa
);
3120 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3121 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3123 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3124 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3125 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3128 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3133 * Update the in-core space usage stats for this vdev, its metaslab class,
3134 * and the root vdev.
3137 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3138 int64_t space_delta
)
3140 int64_t dspace_delta
= space_delta
;
3141 spa_t
*spa
= vd
->vdev_spa
;
3142 vdev_t
*rvd
= spa
->spa_root_vdev
;
3143 metaslab_group_t
*mg
= vd
->vdev_mg
;
3144 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3146 ASSERT(vd
== vd
->vdev_top
);
3149 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3150 * factor. We must calculate this here and not at the root vdev
3151 * because the root vdev's psize-to-asize is simply the max of its
3152 * childrens', thus not accurate enough for us.
3154 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3155 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3156 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3157 vd
->vdev_deflate_ratio
;
3159 mutex_enter(&vd
->vdev_stat_lock
);
3160 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3161 vd
->vdev_stat
.vs_space
+= space_delta
;
3162 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3163 mutex_exit(&vd
->vdev_stat_lock
);
3165 if (mc
== spa_normal_class(spa
)) {
3166 mutex_enter(&rvd
->vdev_stat_lock
);
3167 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3168 rvd
->vdev_stat
.vs_space
+= space_delta
;
3169 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3170 mutex_exit(&rvd
->vdev_stat_lock
);
3174 ASSERT(rvd
== vd
->vdev_parent
);
3175 ASSERT(vd
->vdev_ms_count
!= 0);
3177 metaslab_class_space_update(mc
,
3178 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3183 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3184 * so that it will be written out next time the vdev configuration is synced.
3185 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3188 vdev_config_dirty(vdev_t
*vd
)
3190 spa_t
*spa
= vd
->vdev_spa
;
3191 vdev_t
*rvd
= spa
->spa_root_vdev
;
3194 ASSERT(spa_writeable(spa
));
3197 * If this is an aux vdev (as with l2cache and spare devices), then we
3198 * update the vdev config manually and set the sync flag.
3200 if (vd
->vdev_aux
!= NULL
) {
3201 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3205 for (c
= 0; c
< sav
->sav_count
; c
++) {
3206 if (sav
->sav_vdevs
[c
] == vd
)
3210 if (c
== sav
->sav_count
) {
3212 * We're being removed. There's nothing more to do.
3214 ASSERT(sav
->sav_sync
== B_TRUE
);
3218 sav
->sav_sync
= B_TRUE
;
3220 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3221 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3222 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3223 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3229 * Setting the nvlist in the middle if the array is a little
3230 * sketchy, but it will work.
3232 nvlist_free(aux
[c
]);
3233 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3239 * The dirty list is protected by the SCL_CONFIG lock. The caller
3240 * must either hold SCL_CONFIG as writer, or must be the sync thread
3241 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3242 * so this is sufficient to ensure mutual exclusion.
3244 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3245 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3246 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3249 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3250 vdev_config_dirty(rvd
->vdev_child
[c
]);
3252 ASSERT(vd
== vd
->vdev_top
);
3254 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3256 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3261 vdev_config_clean(vdev_t
*vd
)
3263 spa_t
*spa
= vd
->vdev_spa
;
3265 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3266 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3267 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3269 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3270 list_remove(&spa
->spa_config_dirty_list
, vd
);
3274 * Mark a top-level vdev's state as dirty, so that the next pass of
3275 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3276 * the state changes from larger config changes because they require
3277 * much less locking, and are often needed for administrative actions.
3280 vdev_state_dirty(vdev_t
*vd
)
3282 spa_t
*spa
= vd
->vdev_spa
;
3284 ASSERT(spa_writeable(spa
));
3285 ASSERT(vd
== vd
->vdev_top
);
3288 * The state list is protected by the SCL_STATE lock. The caller
3289 * must either hold SCL_STATE as writer, or must be the sync thread
3290 * (which holds SCL_STATE as reader). There's only one sync thread,
3291 * so this is sufficient to ensure mutual exclusion.
3293 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3294 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3295 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3297 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3298 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3302 vdev_state_clean(vdev_t
*vd
)
3304 spa_t
*spa
= vd
->vdev_spa
;
3306 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3307 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3308 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3310 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3311 list_remove(&spa
->spa_state_dirty_list
, vd
);
3315 * Propagate vdev state up from children to parent.
3318 vdev_propagate_state(vdev_t
*vd
)
3320 spa_t
*spa
= vd
->vdev_spa
;
3321 vdev_t
*rvd
= spa
->spa_root_vdev
;
3322 int degraded
= 0, faulted
= 0;
3327 if (vd
->vdev_children
> 0) {
3328 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3329 child
= vd
->vdev_child
[c
];
3332 * Don't factor holes into the decision.
3334 if (child
->vdev_ishole
)
3337 if (!vdev_readable(child
) ||
3338 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3340 * Root special: if there is a top-level log
3341 * device, treat the root vdev as if it were
3344 if (child
->vdev_islog
&& vd
== rvd
)
3348 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3352 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3356 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3359 * Root special: if there is a top-level vdev that cannot be
3360 * opened due to corrupted metadata, then propagate the root
3361 * vdev's aux state as 'corrupt' rather than 'insufficient
3364 if (corrupted
&& vd
== rvd
&&
3365 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3366 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3367 VDEV_AUX_CORRUPT_DATA
);
3370 if (vd
->vdev_parent
)
3371 vdev_propagate_state(vd
->vdev_parent
);
3375 * Set a vdev's state. If this is during an open, we don't update the parent
3376 * state, because we're in the process of opening children depth-first.
3377 * Otherwise, we propagate the change to the parent.
3379 * If this routine places a device in a faulted state, an appropriate ereport is
3383 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3385 uint64_t save_state
;
3386 spa_t
*spa
= vd
->vdev_spa
;
3388 if (state
== vd
->vdev_state
) {
3390 * Since vdev_offline() code path is already in an offline
3391 * state we can miss a statechange event to OFFLINE. Check
3392 * the previous state to catch this condition.
3394 if (vd
->vdev_ops
->vdev_op_leaf
&&
3395 (state
== VDEV_STATE_OFFLINE
) &&
3396 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
3397 /* post an offline state change */
3398 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
3400 vd
->vdev_stat
.vs_aux
= aux
;
3404 save_state
= vd
->vdev_state
;
3406 vd
->vdev_state
= state
;
3407 vd
->vdev_stat
.vs_aux
= aux
;
3410 * If we are setting the vdev state to anything but an open state, then
3411 * always close the underlying device unless the device has requested
3412 * a delayed close (i.e. we're about to remove or fault the device).
3413 * Otherwise, we keep accessible but invalid devices open forever.
3414 * We don't call vdev_close() itself, because that implies some extra
3415 * checks (offline, etc) that we don't want here. This is limited to
3416 * leaf devices, because otherwise closing the device will affect other
3419 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3420 vd
->vdev_ops
->vdev_op_leaf
)
3421 vd
->vdev_ops
->vdev_op_close(vd
);
3423 if (vd
->vdev_removed
&&
3424 state
== VDEV_STATE_CANT_OPEN
&&
3425 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3427 * If the previous state is set to VDEV_STATE_REMOVED, then this
3428 * device was previously marked removed and someone attempted to
3429 * reopen it. If this failed due to a nonexistent device, then
3430 * keep the device in the REMOVED state. We also let this be if
3431 * it is one of our special test online cases, which is only
3432 * attempting to online the device and shouldn't generate an FMA
3435 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3436 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3437 } else if (state
== VDEV_STATE_REMOVED
) {
3438 vd
->vdev_removed
= B_TRUE
;
3439 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3441 * If we fail to open a vdev during an import or recovery, we
3442 * mark it as "not available", which signifies that it was
3443 * never there to begin with. Failure to open such a device
3444 * is not considered an error.
3446 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3447 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3448 vd
->vdev_ops
->vdev_op_leaf
)
3449 vd
->vdev_not_present
= 1;
3452 * Post the appropriate ereport. If the 'prevstate' field is
3453 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3454 * that this is part of a vdev_reopen(). In this case, we don't
3455 * want to post the ereport if the device was already in the
3456 * CANT_OPEN state beforehand.
3458 * If the 'checkremove' flag is set, then this is an attempt to
3459 * online the device in response to an insertion event. If we
3460 * hit this case, then we have detected an insertion event for a
3461 * faulted or offline device that wasn't in the removed state.
3462 * In this scenario, we don't post an ereport because we are
3463 * about to replace the device, or attempt an online with
3464 * vdev_forcefault, which will generate the fault for us.
3466 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3467 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3468 vd
!= spa
->spa_root_vdev
) {
3472 case VDEV_AUX_OPEN_FAILED
:
3473 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3475 case VDEV_AUX_CORRUPT_DATA
:
3476 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3478 case VDEV_AUX_NO_REPLICAS
:
3479 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3481 case VDEV_AUX_BAD_GUID_SUM
:
3482 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3484 case VDEV_AUX_TOO_SMALL
:
3485 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3487 case VDEV_AUX_BAD_LABEL
:
3488 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3491 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3494 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3497 /* Erase any notion of persistent removed state */
3498 vd
->vdev_removed
= B_FALSE
;
3500 vd
->vdev_removed
= B_FALSE
;
3504 * Notify ZED of any significant state-change on a leaf vdev.
3507 if (vd
->vdev_ops
->vdev_op_leaf
) {
3508 /* preserve original state from a vdev_reopen() */
3509 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
3510 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
3511 (save_state
<= VDEV_STATE_CLOSED
))
3512 save_state
= vd
->vdev_prevstate
;
3514 /* filter out state change due to initial vdev_open */
3515 if (save_state
> VDEV_STATE_CLOSED
)
3516 zfs_post_state_change(spa
, vd
, save_state
);
3519 if (!isopen
&& vd
->vdev_parent
)
3520 vdev_propagate_state(vd
->vdev_parent
);
3524 * Check the vdev configuration to ensure that it's capable of supporting
3528 vdev_is_bootable(vdev_t
*vd
)
3530 #if defined(__sun__) || defined(__sun)
3532 * Currently, we do not support RAID-Z or partial configuration.
3533 * In addition, only a single top-level vdev is allowed and none of the
3534 * leaves can be wholedisks.
3538 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3539 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3541 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3542 vd
->vdev_children
> 1) {
3544 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3545 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3550 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3551 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3554 #endif /* __sun__ || __sun */
3559 * Load the state from the original vdev tree (ovd) which
3560 * we've retrieved from the MOS config object. If the original
3561 * vdev was offline or faulted then we transfer that state to the
3562 * device in the current vdev tree (nvd).
3565 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3569 ASSERT(nvd
->vdev_top
->vdev_islog
);
3570 ASSERT(spa_config_held(nvd
->vdev_spa
,
3571 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3572 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3574 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3575 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3577 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3579 * Restore the persistent vdev state
3581 nvd
->vdev_offline
= ovd
->vdev_offline
;
3582 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3583 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3584 nvd
->vdev_removed
= ovd
->vdev_removed
;
3589 * Determine if a log device has valid content. If the vdev was
3590 * removed or faulted in the MOS config then we know that
3591 * the content on the log device has already been written to the pool.
3594 vdev_log_state_valid(vdev_t
*vd
)
3598 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3602 for (c
= 0; c
< vd
->vdev_children
; c
++)
3603 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3610 * Expand a vdev if possible.
3613 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3615 ASSERT(vd
->vdev_top
== vd
);
3616 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3618 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3619 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3620 vdev_config_dirty(vd
);
3628 vdev_split(vdev_t
*vd
)
3630 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3632 vdev_remove_child(pvd
, vd
);
3633 vdev_compact_children(pvd
);
3635 cvd
= pvd
->vdev_child
[0];
3636 if (pvd
->vdev_children
== 1) {
3637 vdev_remove_parent(cvd
);
3638 cvd
->vdev_splitting
= B_TRUE
;
3640 vdev_propagate_state(cvd
);
3644 vdev_deadman(vdev_t
*vd
)
3648 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3649 vdev_t
*cvd
= vd
->vdev_child
[c
];
3654 if (vd
->vdev_ops
->vdev_op_leaf
) {
3655 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3657 mutex_enter(&vq
->vq_lock
);
3658 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3659 spa_t
*spa
= vd
->vdev_spa
;
3664 * Look at the head of all the pending queues,
3665 * if any I/O has been outstanding for longer than
3666 * the spa_deadman_synctime we log a zevent.
3668 fio
= avl_first(&vq
->vq_active_tree
);
3669 delta
= gethrtime() - fio
->io_timestamp
;
3670 if (delta
> spa_deadman_synctime(spa
)) {
3671 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3672 "delta %lluns, last io %lluns",
3673 fio
->io_timestamp
, delta
,
3674 vq
->vq_io_complete_ts
);
3675 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3676 spa
, vd
, fio
, 0, 0);
3679 mutex_exit(&vq
->vq_lock
);
3683 #if defined(_KERNEL) && defined(HAVE_SPL)
3684 EXPORT_SYMBOL(vdev_fault
);
3685 EXPORT_SYMBOL(vdev_degrade
);
3686 EXPORT_SYMBOL(vdev_online
);
3687 EXPORT_SYMBOL(vdev_offline
);
3688 EXPORT_SYMBOL(vdev_clear
);
3690 module_param(metaslabs_per_vdev
, int, 0644);
3691 MODULE_PARM_DESC(metaslabs_per_vdev
,
3692 "Divide added vdev into approximately (but no more than) this number "