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
; /* use detected value */
1351 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1352 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1353 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1354 VDEV_AUX_BAD_ASHIFT
);
1355 return (SET_ERROR(EDOM
));
1359 * Detect if the alignment requirement has increased.
1360 * We don't want to make the pool unavailable, just
1361 * post an event instead.
1363 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1364 vd
->vdev_ops
->vdev_op_leaf
) {
1365 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1366 spa
, vd
, NULL
, 0, 0);
1369 vd
->vdev_max_asize
= max_asize
;
1373 * If all children are healthy and the asize has increased,
1374 * then we've experienced dynamic LUN growth. If automatic
1375 * expansion is enabled then use the additional space.
1377 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1378 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1379 vd
->vdev_asize
= asize
;
1381 vdev_set_min_asize(vd
);
1384 * Ensure we can issue some IO before declaring the
1385 * vdev open for business.
1387 if (vd
->vdev_ops
->vdev_op_leaf
&&
1388 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1389 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1390 VDEV_AUX_ERR_EXCEEDED
);
1395 * Track the min and max ashift values for normal data devices.
1397 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1398 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1399 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1400 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1401 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1402 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1406 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1407 * resilver. But don't do this if we are doing a reopen for a scrub,
1408 * since this would just restart the scrub we are already doing.
1410 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1411 vdev_resilver_needed(vd
, NULL
, NULL
))
1412 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1418 * Called once the vdevs are all opened, this routine validates the label
1419 * contents. This needs to be done before vdev_load() so that we don't
1420 * inadvertently do repair I/Os to the wrong device.
1422 * If 'strict' is false ignore the spa guid check. This is necessary because
1423 * if the machine crashed during a re-guid the new guid might have been written
1424 * to all of the vdev labels, but not the cached config. The strict check
1425 * will be performed when the pool is opened again using the mos config.
1427 * This function will only return failure if one of the vdevs indicates that it
1428 * has since been destroyed or exported. This is only possible if
1429 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1430 * will be updated but the function will return 0.
1433 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1435 spa_t
*spa
= vd
->vdev_spa
;
1437 uint64_t guid
= 0, top_guid
;
1441 for (c
= 0; c
< vd
->vdev_children
; c
++)
1442 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1443 return (SET_ERROR(EBADF
));
1446 * If the device has already failed, or was marked offline, don't do
1447 * any further validation. Otherwise, label I/O will fail and we will
1448 * overwrite the previous state.
1450 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1451 uint64_t aux_guid
= 0;
1453 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1454 spa_last_synced_txg(spa
) : -1ULL;
1456 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1457 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1458 VDEV_AUX_BAD_LABEL
);
1463 * Determine if this vdev has been split off into another
1464 * pool. If so, then refuse to open it.
1466 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1467 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1468 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1469 VDEV_AUX_SPLIT_POOL
);
1474 if (strict
&& (nvlist_lookup_uint64(label
,
1475 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1476 guid
!= spa_guid(spa
))) {
1477 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1478 VDEV_AUX_CORRUPT_DATA
);
1483 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1484 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1489 * If this vdev just became a top-level vdev because its
1490 * sibling was detached, it will have adopted the parent's
1491 * vdev guid -- but the label may or may not be on disk yet.
1492 * Fortunately, either version of the label will have the
1493 * same top guid, so if we're a top-level vdev, we can
1494 * safely compare to that instead.
1496 * If we split this vdev off instead, then we also check the
1497 * original pool's guid. We don't want to consider the vdev
1498 * corrupt if it is partway through a split operation.
1500 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1502 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1504 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1505 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1506 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1507 VDEV_AUX_CORRUPT_DATA
);
1512 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1514 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1515 VDEV_AUX_CORRUPT_DATA
);
1523 * If this is a verbatim import, no need to check the
1524 * state of the pool.
1526 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1527 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1528 state
!= POOL_STATE_ACTIVE
)
1529 return (SET_ERROR(EBADF
));
1532 * If we were able to open and validate a vdev that was
1533 * previously marked permanently unavailable, clear that state
1536 if (vd
->vdev_not_present
)
1537 vd
->vdev_not_present
= 0;
1544 * Close a virtual device.
1547 vdev_close(vdev_t
*vd
)
1549 vdev_t
*pvd
= vd
->vdev_parent
;
1550 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1552 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1555 * If our parent is reopening, then we are as well, unless we are
1558 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1559 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1561 vd
->vdev_ops
->vdev_op_close(vd
);
1563 vdev_cache_purge(vd
);
1566 * We record the previous state before we close it, so that if we are
1567 * doing a reopen(), we don't generate FMA ereports if we notice that
1568 * it's still faulted.
1570 vd
->vdev_prevstate
= vd
->vdev_state
;
1572 if (vd
->vdev_offline
)
1573 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1575 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1576 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1580 vdev_hold(vdev_t
*vd
)
1582 spa_t
*spa
= vd
->vdev_spa
;
1585 ASSERT(spa_is_root(spa
));
1586 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1589 for (c
= 0; c
< vd
->vdev_children
; c
++)
1590 vdev_hold(vd
->vdev_child
[c
]);
1592 if (vd
->vdev_ops
->vdev_op_leaf
)
1593 vd
->vdev_ops
->vdev_op_hold(vd
);
1597 vdev_rele(vdev_t
*vd
)
1601 ASSERT(spa_is_root(vd
->vdev_spa
));
1602 for (c
= 0; c
< vd
->vdev_children
; c
++)
1603 vdev_rele(vd
->vdev_child
[c
]);
1605 if (vd
->vdev_ops
->vdev_op_leaf
)
1606 vd
->vdev_ops
->vdev_op_rele(vd
);
1610 * Reopen all interior vdevs and any unopened leaves. We don't actually
1611 * reopen leaf vdevs which had previously been opened as they might deadlock
1612 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1613 * If the leaf has never been opened then open it, as usual.
1616 vdev_reopen(vdev_t
*vd
)
1618 spa_t
*spa
= vd
->vdev_spa
;
1620 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1622 /* set the reopening flag unless we're taking the vdev offline */
1623 vd
->vdev_reopening
= !vd
->vdev_offline
;
1625 (void) vdev_open(vd
);
1628 * Call vdev_validate() here to make sure we have the same device.
1629 * Otherwise, a device with an invalid label could be successfully
1630 * opened in response to vdev_reopen().
1633 (void) vdev_validate_aux(vd
);
1634 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1635 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1636 !l2arc_vdev_present(vd
))
1637 l2arc_add_vdev(spa
, vd
);
1639 (void) vdev_validate(vd
, B_TRUE
);
1643 * Reassess parent vdev's health.
1645 vdev_propagate_state(vd
);
1649 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1654 * Normally, partial opens (e.g. of a mirror) are allowed.
1655 * For a create, however, we want to fail the request if
1656 * there are any components we can't open.
1658 error
= vdev_open(vd
);
1660 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1662 return (error
? error
: ENXIO
);
1666 * Recursively load DTLs and initialize all labels.
1668 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1669 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1670 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1679 vdev_metaslab_set_size(vdev_t
*vd
)
1682 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1684 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1685 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1689 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1691 ASSERT(vd
== vd
->vdev_top
);
1692 ASSERT(!vd
->vdev_ishole
);
1693 ASSERT(ISP2(flags
));
1694 ASSERT(spa_writeable(vd
->vdev_spa
));
1696 if (flags
& VDD_METASLAB
)
1697 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1699 if (flags
& VDD_DTL
)
1700 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1702 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1706 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1710 for (c
= 0; c
< vd
->vdev_children
; c
++)
1711 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1713 if (vd
->vdev_ops
->vdev_op_leaf
)
1714 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1720 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1721 * the vdev has less than perfect replication. There are four kinds of DTL:
1723 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1725 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1727 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1728 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1729 * txgs that was scrubbed.
1731 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1732 * persistent errors or just some device being offline.
1733 * Unlike the other three, the DTL_OUTAGE map is not generally
1734 * maintained; it's only computed when needed, typically to
1735 * determine whether a device can be detached.
1737 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1738 * either has the data or it doesn't.
1740 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1741 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1742 * if any child is less than fully replicated, then so is its parent.
1743 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1744 * comprising only those txgs which appear in 'maxfaults' or more children;
1745 * those are the txgs we don't have enough replication to read. For example,
1746 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1747 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1748 * two child DTL_MISSING maps.
1750 * It should be clear from the above that to compute the DTLs and outage maps
1751 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1752 * Therefore, that is all we keep on disk. When loading the pool, or after
1753 * a configuration change, we generate all other DTLs from first principles.
1756 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1758 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1760 ASSERT(t
< DTL_TYPES
);
1761 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1762 ASSERT(spa_writeable(vd
->vdev_spa
));
1764 mutex_enter(rt
->rt_lock
);
1765 if (!range_tree_contains(rt
, txg
, size
))
1766 range_tree_add(rt
, txg
, size
);
1767 mutex_exit(rt
->rt_lock
);
1771 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1773 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1774 boolean_t dirty
= B_FALSE
;
1776 ASSERT(t
< DTL_TYPES
);
1777 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1779 mutex_enter(rt
->rt_lock
);
1780 if (range_tree_space(rt
) != 0)
1781 dirty
= range_tree_contains(rt
, txg
, size
);
1782 mutex_exit(rt
->rt_lock
);
1788 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1790 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1793 mutex_enter(rt
->rt_lock
);
1794 empty
= (range_tree_space(rt
) == 0);
1795 mutex_exit(rt
->rt_lock
);
1801 * Returns the lowest txg in the DTL range.
1804 vdev_dtl_min(vdev_t
*vd
)
1808 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1809 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1810 ASSERT0(vd
->vdev_children
);
1812 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1813 return (rs
->rs_start
- 1);
1817 * Returns the highest txg in the DTL.
1820 vdev_dtl_max(vdev_t
*vd
)
1824 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1825 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1826 ASSERT0(vd
->vdev_children
);
1828 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1829 return (rs
->rs_end
);
1833 * Determine if a resilvering vdev should remove any DTL entries from
1834 * its range. If the vdev was resilvering for the entire duration of the
1835 * scan then it should excise that range from its DTLs. Otherwise, this
1836 * vdev is considered partially resilvered and should leave its DTL
1837 * entries intact. The comment in vdev_dtl_reassess() describes how we
1841 vdev_dtl_should_excise(vdev_t
*vd
)
1843 spa_t
*spa
= vd
->vdev_spa
;
1844 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1846 ASSERT0(scn
->scn_phys
.scn_errors
);
1847 ASSERT0(vd
->vdev_children
);
1849 if (vd
->vdev_resilver_txg
== 0 ||
1850 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1854 * When a resilver is initiated the scan will assign the scn_max_txg
1855 * value to the highest txg value that exists in all DTLs. If this
1856 * device's max DTL is not part of this scan (i.e. it is not in
1857 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1860 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1861 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1862 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1863 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1870 * Reassess DTLs after a config change or scrub completion.
1873 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1875 spa_t
*spa
= vd
->vdev_spa
;
1879 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1881 for (c
= 0; c
< vd
->vdev_children
; c
++)
1882 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1883 scrub_txg
, scrub_done
);
1885 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1888 if (vd
->vdev_ops
->vdev_op_leaf
) {
1889 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1891 mutex_enter(&vd
->vdev_dtl_lock
);
1894 * If we've completed a scan cleanly then determine
1895 * if this vdev should remove any DTLs. We only want to
1896 * excise regions on vdevs that were available during
1897 * the entire duration of this scan.
1899 if (scrub_txg
!= 0 &&
1900 (spa
->spa_scrub_started
||
1901 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1902 vdev_dtl_should_excise(vd
)) {
1904 * We completed a scrub up to scrub_txg. If we
1905 * did it without rebooting, then the scrub dtl
1906 * will be valid, so excise the old region and
1907 * fold in the scrub dtl. Otherwise, leave the
1908 * dtl as-is if there was an error.
1910 * There's little trick here: to excise the beginning
1911 * of the DTL_MISSING map, we put it into a reference
1912 * tree and then add a segment with refcnt -1 that
1913 * covers the range [0, scrub_txg). This means
1914 * that each txg in that range has refcnt -1 or 0.
1915 * We then add DTL_SCRUB with a refcnt of 2, so that
1916 * entries in the range [0, scrub_txg) will have a
1917 * positive refcnt -- either 1 or 2. We then convert
1918 * the reference tree into the new DTL_MISSING map.
1920 space_reftree_create(&reftree
);
1921 space_reftree_add_map(&reftree
,
1922 vd
->vdev_dtl
[DTL_MISSING
], 1);
1923 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1924 space_reftree_add_map(&reftree
,
1925 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1926 space_reftree_generate_map(&reftree
,
1927 vd
->vdev_dtl
[DTL_MISSING
], 1);
1928 space_reftree_destroy(&reftree
);
1930 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1931 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1932 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1934 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1935 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1936 if (!vdev_readable(vd
))
1937 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1939 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1940 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1943 * If the vdev was resilvering and no longer has any
1944 * DTLs then reset its resilvering flag and dirty
1945 * the top level so that we persist the change.
1947 if (vd
->vdev_resilver_txg
!= 0 &&
1948 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1949 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
1950 vd
->vdev_resilver_txg
= 0;
1951 vdev_config_dirty(vd
->vdev_top
);
1954 mutex_exit(&vd
->vdev_dtl_lock
);
1957 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1961 mutex_enter(&vd
->vdev_dtl_lock
);
1962 for (t
= 0; t
< DTL_TYPES
; t
++) {
1965 /* account for child's outage in parent's missing map */
1966 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1968 continue; /* leaf vdevs only */
1969 if (t
== DTL_PARTIAL
)
1970 minref
= 1; /* i.e. non-zero */
1971 else if (vd
->vdev_nparity
!= 0)
1972 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1974 minref
= vd
->vdev_children
; /* any kind of mirror */
1975 space_reftree_create(&reftree
);
1976 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1977 vdev_t
*cvd
= vd
->vdev_child
[c
];
1978 mutex_enter(&cvd
->vdev_dtl_lock
);
1979 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1980 mutex_exit(&cvd
->vdev_dtl_lock
);
1982 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1983 space_reftree_destroy(&reftree
);
1985 mutex_exit(&vd
->vdev_dtl_lock
);
1989 vdev_dtl_load(vdev_t
*vd
)
1991 spa_t
*spa
= vd
->vdev_spa
;
1992 objset_t
*mos
= spa
->spa_meta_objset
;
1996 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
1997 ASSERT(!vd
->vdev_ishole
);
1999 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2000 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
2003 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2005 mutex_enter(&vd
->vdev_dtl_lock
);
2008 * Now that we've opened the space_map we need to update
2011 space_map_update(vd
->vdev_dtl_sm
);
2013 error
= space_map_load(vd
->vdev_dtl_sm
,
2014 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2015 mutex_exit(&vd
->vdev_dtl_lock
);
2020 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2021 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2030 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2032 spa_t
*spa
= vd
->vdev_spa
;
2034 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2035 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2040 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2042 spa_t
*spa
= vd
->vdev_spa
;
2043 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2044 DMU_OT_NONE
, 0, tx
);
2047 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2054 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2058 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2059 vd
->vdev_ops
!= &vdev_missing_ops
&&
2060 vd
->vdev_ops
!= &vdev_root_ops
&&
2061 !vd
->vdev_top
->vdev_removing
) {
2062 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2063 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2065 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2066 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2069 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2070 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2075 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2077 spa_t
*spa
= vd
->vdev_spa
;
2078 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2079 objset_t
*mos
= spa
->spa_meta_objset
;
2080 range_tree_t
*rtsync
;
2083 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2085 ASSERT(!vd
->vdev_ishole
);
2086 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2088 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2090 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2091 mutex_enter(&vd
->vdev_dtl_lock
);
2092 space_map_free(vd
->vdev_dtl_sm
, tx
);
2093 space_map_close(vd
->vdev_dtl_sm
);
2094 vd
->vdev_dtl_sm
= NULL
;
2095 mutex_exit(&vd
->vdev_dtl_lock
);
2098 * We only destroy the leaf ZAP for detached leaves or for
2099 * removed log devices. Removed data devices handle leaf ZAP
2100 * cleanup later, once cancellation is no longer possible.
2102 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2103 vd
->vdev_top
->vdev_islog
)) {
2104 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2105 vd
->vdev_leaf_zap
= 0;
2112 if (vd
->vdev_dtl_sm
== NULL
) {
2113 uint64_t new_object
;
2115 new_object
= space_map_alloc(mos
, tx
);
2116 VERIFY3U(new_object
, !=, 0);
2118 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2119 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2120 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2123 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2125 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2127 mutex_enter(&rtlock
);
2129 mutex_enter(&vd
->vdev_dtl_lock
);
2130 range_tree_walk(rt
, range_tree_add
, rtsync
);
2131 mutex_exit(&vd
->vdev_dtl_lock
);
2133 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2134 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2135 range_tree_vacate(rtsync
, NULL
, NULL
);
2137 range_tree_destroy(rtsync
);
2139 mutex_exit(&rtlock
);
2140 mutex_destroy(&rtlock
);
2143 * If the object for the space map has changed then dirty
2144 * the top level so that we update the config.
2146 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2147 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2148 "new object %llu", txg
, spa_name(spa
), object
,
2149 space_map_object(vd
->vdev_dtl_sm
));
2150 vdev_config_dirty(vd
->vdev_top
);
2155 mutex_enter(&vd
->vdev_dtl_lock
);
2156 space_map_update(vd
->vdev_dtl_sm
);
2157 mutex_exit(&vd
->vdev_dtl_lock
);
2161 * Determine whether the specified vdev can be offlined/detached/removed
2162 * without losing data.
2165 vdev_dtl_required(vdev_t
*vd
)
2167 spa_t
*spa
= vd
->vdev_spa
;
2168 vdev_t
*tvd
= vd
->vdev_top
;
2169 uint8_t cant_read
= vd
->vdev_cant_read
;
2172 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2174 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2178 * Temporarily mark the device as unreadable, and then determine
2179 * whether this results in any DTL outages in the top-level vdev.
2180 * If not, we can safely offline/detach/remove the device.
2182 vd
->vdev_cant_read
= B_TRUE
;
2183 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2184 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2185 vd
->vdev_cant_read
= cant_read
;
2186 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2188 if (!required
&& zio_injection_enabled
)
2189 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2195 * Determine if resilver is needed, and if so the txg range.
2198 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2200 boolean_t needed
= B_FALSE
;
2201 uint64_t thismin
= UINT64_MAX
;
2202 uint64_t thismax
= 0;
2205 if (vd
->vdev_children
== 0) {
2206 mutex_enter(&vd
->vdev_dtl_lock
);
2207 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2208 vdev_writeable(vd
)) {
2210 thismin
= vdev_dtl_min(vd
);
2211 thismax
= vdev_dtl_max(vd
);
2214 mutex_exit(&vd
->vdev_dtl_lock
);
2216 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2217 vdev_t
*cvd
= vd
->vdev_child
[c
];
2218 uint64_t cmin
, cmax
;
2220 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2221 thismin
= MIN(thismin
, cmin
);
2222 thismax
= MAX(thismax
, cmax
);
2228 if (needed
&& minp
) {
2236 vdev_load(vdev_t
*vd
)
2241 * Recursively load all children.
2243 for (c
= 0; c
< vd
->vdev_children
; c
++)
2244 vdev_load(vd
->vdev_child
[c
]);
2247 * If this is a top-level vdev, initialize its metaslabs.
2249 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2250 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2251 vdev_metaslab_init(vd
, 0) != 0))
2252 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2253 VDEV_AUX_CORRUPT_DATA
);
2255 * If this is a leaf vdev, load its DTL.
2257 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2258 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2259 VDEV_AUX_CORRUPT_DATA
);
2263 * The special vdev case is used for hot spares and l2cache devices. Its
2264 * sole purpose it to set the vdev state for the associated vdev. To do this,
2265 * we make sure that we can open the underlying device, then try to read the
2266 * label, and make sure that the label is sane and that it hasn't been
2267 * repurposed to another pool.
2270 vdev_validate_aux(vdev_t
*vd
)
2273 uint64_t guid
, version
;
2276 if (!vdev_readable(vd
))
2279 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2280 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2281 VDEV_AUX_CORRUPT_DATA
);
2285 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2286 !SPA_VERSION_IS_SUPPORTED(version
) ||
2287 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2288 guid
!= vd
->vdev_guid
||
2289 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2290 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2291 VDEV_AUX_CORRUPT_DATA
);
2297 * We don't actually check the pool state here. If it's in fact in
2298 * use by another pool, we update this fact on the fly when requested.
2305 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2307 spa_t
*spa
= vd
->vdev_spa
;
2308 objset_t
*mos
= spa
->spa_meta_objset
;
2312 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2313 ASSERT(vd
== vd
->vdev_top
);
2314 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2316 if (vd
->vdev_ms
!= NULL
) {
2317 metaslab_group_t
*mg
= vd
->vdev_mg
;
2319 metaslab_group_histogram_verify(mg
);
2320 metaslab_class_histogram_verify(mg
->mg_class
);
2322 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2323 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2325 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2328 mutex_enter(&msp
->ms_lock
);
2330 * If the metaslab was not loaded when the vdev
2331 * was removed then the histogram accounting may
2332 * not be accurate. Update the histogram information
2333 * here so that we ensure that the metaslab group
2334 * and metaslab class are up-to-date.
2336 metaslab_group_histogram_remove(mg
, msp
);
2338 VERIFY0(space_map_allocated(msp
->ms_sm
));
2339 space_map_free(msp
->ms_sm
, tx
);
2340 space_map_close(msp
->ms_sm
);
2342 mutex_exit(&msp
->ms_lock
);
2345 metaslab_group_histogram_verify(mg
);
2346 metaslab_class_histogram_verify(mg
->mg_class
);
2347 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2348 ASSERT0(mg
->mg_histogram
[i
]);
2352 if (vd
->vdev_ms_array
) {
2353 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2354 vd
->vdev_ms_array
= 0;
2357 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2358 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2359 vd
->vdev_top_zap
= 0;
2365 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2368 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2370 ASSERT(!vd
->vdev_ishole
);
2372 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2373 metaslab_sync_done(msp
, txg
);
2376 metaslab_sync_reassess(vd
->vdev_mg
);
2380 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2382 spa_t
*spa
= vd
->vdev_spa
;
2387 ASSERT(!vd
->vdev_ishole
);
2389 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2390 ASSERT(vd
== vd
->vdev_top
);
2391 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2392 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2393 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2394 ASSERT(vd
->vdev_ms_array
!= 0);
2395 vdev_config_dirty(vd
);
2400 * Remove the metadata associated with this vdev once it's empty.
2402 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2403 vdev_remove(vd
, txg
);
2405 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2406 metaslab_sync(msp
, txg
);
2407 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2410 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2411 vdev_dtl_sync(lvd
, txg
);
2413 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2417 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2419 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2423 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2424 * not be opened, and no I/O is attempted.
2427 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2431 spa_vdev_state_enter(spa
, SCL_NONE
);
2433 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2434 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2436 if (!vd
->vdev_ops
->vdev_op_leaf
)
2437 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2442 * We don't directly use the aux state here, but if we do a
2443 * vdev_reopen(), we need this value to be present to remember why we
2446 vd
->vdev_label_aux
= aux
;
2449 * Faulted state takes precedence over degraded.
2451 vd
->vdev_delayed_close
= B_FALSE
;
2452 vd
->vdev_faulted
= 1ULL;
2453 vd
->vdev_degraded
= 0ULL;
2454 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2457 * If this device has the only valid copy of the data, then
2458 * back off and simply mark the vdev as degraded instead.
2460 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2461 vd
->vdev_degraded
= 1ULL;
2462 vd
->vdev_faulted
= 0ULL;
2465 * If we reopen the device and it's not dead, only then do we
2470 if (vdev_readable(vd
))
2471 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2474 return (spa_vdev_state_exit(spa
, vd
, 0));
2478 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2479 * user that something is wrong. The vdev continues to operate as normal as far
2480 * as I/O is concerned.
2483 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2487 spa_vdev_state_enter(spa
, SCL_NONE
);
2489 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2490 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2492 if (!vd
->vdev_ops
->vdev_op_leaf
)
2493 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2496 * If the vdev is already faulted, then don't do anything.
2498 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2499 return (spa_vdev_state_exit(spa
, NULL
, 0));
2501 vd
->vdev_degraded
= 1ULL;
2502 if (!vdev_is_dead(vd
))
2503 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2506 return (spa_vdev_state_exit(spa
, vd
, 0));
2510 * Online the given vdev.
2512 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2513 * spare device should be detached when the device finishes resilvering.
2514 * Second, the online should be treated like a 'test' online case, so no FMA
2515 * events are generated if the device fails to open.
2518 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2520 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2521 boolean_t postevent
= B_FALSE
;
2523 spa_vdev_state_enter(spa
, SCL_NONE
);
2525 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2526 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2528 if (!vd
->vdev_ops
->vdev_op_leaf
)
2529 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2532 (vd
->vdev_offline
== B_TRUE
|| vd
->vdev_tmpoffline
== B_TRUE
) ?
2536 vd
->vdev_offline
= B_FALSE
;
2537 vd
->vdev_tmpoffline
= B_FALSE
;
2538 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2539 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2541 /* XXX - L2ARC 1.0 does not support expansion */
2542 if (!vd
->vdev_aux
) {
2543 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2544 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2548 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2550 if (!vd
->vdev_aux
) {
2551 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2552 pvd
->vdev_expanding
= B_FALSE
;
2556 *newstate
= vd
->vdev_state
;
2557 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2558 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2559 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2560 vd
->vdev_parent
->vdev_child
[0] == vd
)
2561 vd
->vdev_unspare
= B_TRUE
;
2563 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2565 /* XXX - L2ARC 1.0 does not support expansion */
2567 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2568 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2572 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2574 return (spa_vdev_state_exit(spa
, vd
, 0));
2578 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2582 uint64_t generation
;
2583 metaslab_group_t
*mg
;
2586 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2588 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2589 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2591 if (!vd
->vdev_ops
->vdev_op_leaf
)
2592 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2596 generation
= spa
->spa_config_generation
+ 1;
2599 * If the device isn't already offline, try to offline it.
2601 if (!vd
->vdev_offline
) {
2603 * If this device has the only valid copy of some data,
2604 * don't allow it to be offlined. Log devices are always
2607 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2608 vdev_dtl_required(vd
))
2609 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2612 * If the top-level is a slog and it has had allocations
2613 * then proceed. We check that the vdev's metaslab group
2614 * is not NULL since it's possible that we may have just
2615 * added this vdev but not yet initialized its metaslabs.
2617 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2619 * Prevent any future allocations.
2621 metaslab_group_passivate(mg
);
2622 (void) spa_vdev_state_exit(spa
, vd
, 0);
2624 error
= spa_offline_log(spa
);
2626 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2629 * Check to see if the config has changed.
2631 if (error
|| generation
!= spa
->spa_config_generation
) {
2632 metaslab_group_activate(mg
);
2634 return (spa_vdev_state_exit(spa
,
2636 (void) spa_vdev_state_exit(spa
, vd
, 0);
2639 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2643 * Offline this device and reopen its top-level vdev.
2644 * If the top-level vdev is a log device then just offline
2645 * it. Otherwise, if this action results in the top-level
2646 * vdev becoming unusable, undo it and fail the request.
2648 vd
->vdev_offline
= B_TRUE
;
2651 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2652 vdev_is_dead(tvd
)) {
2653 vd
->vdev_offline
= B_FALSE
;
2655 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2659 * Add the device back into the metaslab rotor so that
2660 * once we online the device it's open for business.
2662 if (tvd
->vdev_islog
&& mg
!= NULL
)
2663 metaslab_group_activate(mg
);
2666 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2668 return (spa_vdev_state_exit(spa
, vd
, 0));
2672 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2676 mutex_enter(&spa
->spa_vdev_top_lock
);
2677 error
= vdev_offline_locked(spa
, guid
, flags
);
2678 mutex_exit(&spa
->spa_vdev_top_lock
);
2684 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2685 * vdev_offline(), we assume the spa config is locked. We also clear all
2686 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2689 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2691 vdev_t
*rvd
= spa
->spa_root_vdev
;
2694 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2699 vd
->vdev_stat
.vs_read_errors
= 0;
2700 vd
->vdev_stat
.vs_write_errors
= 0;
2701 vd
->vdev_stat
.vs_checksum_errors
= 0;
2703 for (c
= 0; c
< vd
->vdev_children
; c
++)
2704 vdev_clear(spa
, vd
->vdev_child
[c
]);
2707 * If we're in the FAULTED state or have experienced failed I/O, then
2708 * clear the persistent state and attempt to reopen the device. We
2709 * also mark the vdev config dirty, so that the new faulted state is
2710 * written out to disk.
2712 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2713 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2716 * When reopening in response to a clear event, it may be due to
2717 * a fmadm repair request. In this case, if the device is
2718 * still broken, we want to still post the ereport again.
2720 vd
->vdev_forcefault
= B_TRUE
;
2722 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2723 vd
->vdev_cant_read
= B_FALSE
;
2724 vd
->vdev_cant_write
= B_FALSE
;
2726 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2728 vd
->vdev_forcefault
= B_FALSE
;
2730 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2731 vdev_state_dirty(vd
->vdev_top
);
2733 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2734 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2736 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2740 * When clearing a FMA-diagnosed fault, we always want to
2741 * unspare the device, as we assume that the original spare was
2742 * done in response to the FMA fault.
2744 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2745 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2746 vd
->vdev_parent
->vdev_child
[0] == vd
)
2747 vd
->vdev_unspare
= B_TRUE
;
2751 vdev_is_dead(vdev_t
*vd
)
2754 * Holes and missing devices are always considered "dead".
2755 * This simplifies the code since we don't have to check for
2756 * these types of devices in the various code paths.
2757 * Instead we rely on the fact that we skip over dead devices
2758 * before issuing I/O to them.
2760 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2761 vd
->vdev_ops
== &vdev_missing_ops
);
2765 vdev_readable(vdev_t
*vd
)
2767 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2771 vdev_writeable(vdev_t
*vd
)
2773 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2777 vdev_allocatable(vdev_t
*vd
)
2779 uint64_t state
= vd
->vdev_state
;
2782 * We currently allow allocations from vdevs which may be in the
2783 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2784 * fails to reopen then we'll catch it later when we're holding
2785 * the proper locks. Note that we have to get the vdev state
2786 * in a local variable because although it changes atomically,
2787 * we're asking two separate questions about it.
2789 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2790 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
&&
2791 vd
->vdev_mg
->mg_initialized
);
2795 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2797 ASSERT(zio
->io_vd
== vd
);
2799 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2802 if (zio
->io_type
== ZIO_TYPE_READ
)
2803 return (!vd
->vdev_cant_read
);
2805 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2806 return (!vd
->vdev_cant_write
);
2812 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2815 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2816 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2817 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2820 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2824 * Get extended stats
2827 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2830 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2831 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2832 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2834 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2835 vsx
->vsx_total_histo
[t
][b
] +=
2836 cvsx
->vsx_total_histo
[t
][b
];
2840 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2841 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2842 vsx
->vsx_queue_histo
[t
][b
] +=
2843 cvsx
->vsx_queue_histo
[t
][b
];
2845 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2846 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2848 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2849 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2851 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2852 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2858 * Get statistics for the given vdev.
2861 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2865 * If we're getting stats on the root vdev, aggregate the I/O counts
2866 * over all top-level vdevs (i.e. the direct children of the root).
2868 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2870 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2871 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2874 memset(vsx
, 0, sizeof (*vsx
));
2876 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2877 vdev_t
*cvd
= vd
->vdev_child
[c
];
2878 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2879 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2881 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2883 vdev_get_child_stat(cvd
, vs
, cvs
);
2885 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2890 * We're a leaf. Just copy our ZIO active queue stats in. The
2891 * other leaf stats are updated in vdev_stat_update().
2896 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2898 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2899 vsx
->vsx_active_queue
[t
] =
2900 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2901 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2902 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2908 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2910 vdev_t
*tvd
= vd
->vdev_top
;
2911 mutex_enter(&vd
->vdev_stat_lock
);
2913 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2914 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2915 vs
->vs_state
= vd
->vdev_state
;
2916 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2917 if (vd
->vdev_ops
->vdev_op_leaf
)
2918 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
2919 VDEV_LABEL_END_SIZE
;
2921 * Report expandable space on top-level, non-auxillary devices
2922 * only. The expandable space is reported in terms of metaslab
2923 * sized units since that determines how much space the pool
2926 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
2927 vs
->vs_esize
= P2ALIGN(
2928 vd
->vdev_max_asize
- vd
->vdev_asize
,
2929 1ULL << tvd
->vdev_ms_shift
);
2931 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2932 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
2934 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2938 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
2939 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
2940 mutex_exit(&vd
->vdev_stat_lock
);
2944 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2946 return (vdev_get_stats_ex(vd
, vs
, NULL
));
2950 vdev_clear_stats(vdev_t
*vd
)
2952 mutex_enter(&vd
->vdev_stat_lock
);
2953 vd
->vdev_stat
.vs_space
= 0;
2954 vd
->vdev_stat
.vs_dspace
= 0;
2955 vd
->vdev_stat
.vs_alloc
= 0;
2956 mutex_exit(&vd
->vdev_stat_lock
);
2960 vdev_scan_stat_init(vdev_t
*vd
)
2962 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2965 for (c
= 0; c
< vd
->vdev_children
; c
++)
2966 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2968 mutex_enter(&vd
->vdev_stat_lock
);
2969 vs
->vs_scan_processed
= 0;
2970 mutex_exit(&vd
->vdev_stat_lock
);
2974 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2976 spa_t
*spa
= zio
->io_spa
;
2977 vdev_t
*rvd
= spa
->spa_root_vdev
;
2978 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2980 uint64_t txg
= zio
->io_txg
;
2981 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2982 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
2983 zio_type_t type
= zio
->io_type
;
2984 int flags
= zio
->io_flags
;
2987 * If this i/o is a gang leader, it didn't do any actual work.
2989 if (zio
->io_gang_tree
)
2992 if (zio
->io_error
== 0) {
2994 * If this is a root i/o, don't count it -- we've already
2995 * counted the top-level vdevs, and vdev_get_stats() will
2996 * aggregate them when asked. This reduces contention on
2997 * the root vdev_stat_lock and implicitly handles blocks
2998 * that compress away to holes, for which there is no i/o.
2999 * (Holes never create vdev children, so all the counters
3000 * remain zero, which is what we want.)
3002 * Note: this only applies to successful i/o (io_error == 0)
3003 * because unlike i/o counts, errors are not additive.
3004 * When reading a ditto block, for example, failure of
3005 * one top-level vdev does not imply a root-level error.
3010 ASSERT(vd
== zio
->io_vd
);
3012 if (flags
& ZIO_FLAG_IO_BYPASS
)
3015 mutex_enter(&vd
->vdev_stat_lock
);
3017 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3018 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3019 dsl_scan_phys_t
*scn_phys
=
3020 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3021 uint64_t *processed
= &scn_phys
->scn_processed
;
3024 if (vd
->vdev_ops
->vdev_op_leaf
)
3025 atomic_add_64(processed
, psize
);
3026 vs
->vs_scan_processed
+= psize
;
3029 if (flags
& ZIO_FLAG_SELF_HEAL
)
3030 vs
->vs_self_healed
+= psize
;
3034 * The bytes/ops/histograms are recorded at the leaf level and
3035 * aggregated into the higher level vdevs in vdev_get_stats().
3037 if (vd
->vdev_ops
->vdev_op_leaf
&&
3038 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3041 vs
->vs_bytes
[type
] += psize
;
3043 if (flags
& ZIO_FLAG_DELEGATED
) {
3044 vsx
->vsx_agg_histo
[zio
->io_priority
]
3045 [RQ_HISTO(zio
->io_size
)]++;
3047 vsx
->vsx_ind_histo
[zio
->io_priority
]
3048 [RQ_HISTO(zio
->io_size
)]++;
3051 if (zio
->io_delta
&& zio
->io_delay
) {
3052 vsx
->vsx_queue_histo
[zio
->io_priority
]
3053 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3054 vsx
->vsx_disk_histo
[type
]
3055 [L_HISTO(zio
->io_delay
)]++;
3056 vsx
->vsx_total_histo
[type
]
3057 [L_HISTO(zio
->io_delta
)]++;
3061 mutex_exit(&vd
->vdev_stat_lock
);
3065 if (flags
& ZIO_FLAG_SPECULATIVE
)
3069 * If this is an I/O error that is going to be retried, then ignore the
3070 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3071 * hard errors, when in reality they can happen for any number of
3072 * innocuous reasons (bus resets, MPxIO link failure, etc).
3074 if (zio
->io_error
== EIO
&&
3075 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3079 * Intent logs writes won't propagate their error to the root
3080 * I/O so don't mark these types of failures as pool-level
3083 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3086 mutex_enter(&vd
->vdev_stat_lock
);
3087 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3088 if (zio
->io_error
== ECKSUM
)
3089 vs
->vs_checksum_errors
++;
3091 vs
->vs_read_errors
++;
3093 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3094 vs
->vs_write_errors
++;
3095 mutex_exit(&vd
->vdev_stat_lock
);
3097 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3098 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3099 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3100 spa
->spa_claiming
)) {
3102 * This is either a normal write (not a repair), or it's
3103 * a repair induced by the scrub thread, or it's a repair
3104 * made by zil_claim() during spa_load() in the first txg.
3105 * In the normal case, we commit the DTL change in the same
3106 * txg as the block was born. In the scrub-induced repair
3107 * case, we know that scrubs run in first-pass syncing context,
3108 * so we commit the DTL change in spa_syncing_txg(spa).
3109 * In the zil_claim() case, we commit in spa_first_txg(spa).
3111 * We currently do not make DTL entries for failed spontaneous
3112 * self-healing writes triggered by normal (non-scrubbing)
3113 * reads, because we have no transactional context in which to
3114 * do so -- and it's not clear that it'd be desirable anyway.
3116 if (vd
->vdev_ops
->vdev_op_leaf
) {
3117 uint64_t commit_txg
= txg
;
3118 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3119 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3120 ASSERT(spa_sync_pass(spa
) == 1);
3121 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3122 commit_txg
= spa_syncing_txg(spa
);
3123 } else if (spa
->spa_claiming
) {
3124 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3125 commit_txg
= spa_first_txg(spa
);
3127 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3128 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3130 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3131 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3132 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3135 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3140 * Update the in-core space usage stats for this vdev, its metaslab class,
3141 * and the root vdev.
3144 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3145 int64_t space_delta
)
3147 int64_t dspace_delta
= space_delta
;
3148 spa_t
*spa
= vd
->vdev_spa
;
3149 vdev_t
*rvd
= spa
->spa_root_vdev
;
3150 metaslab_group_t
*mg
= vd
->vdev_mg
;
3151 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3153 ASSERT(vd
== vd
->vdev_top
);
3156 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3157 * factor. We must calculate this here and not at the root vdev
3158 * because the root vdev's psize-to-asize is simply the max of its
3159 * childrens', thus not accurate enough for us.
3161 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3162 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3163 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3164 vd
->vdev_deflate_ratio
;
3166 mutex_enter(&vd
->vdev_stat_lock
);
3167 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3168 vd
->vdev_stat
.vs_space
+= space_delta
;
3169 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3170 mutex_exit(&vd
->vdev_stat_lock
);
3172 if (mc
== spa_normal_class(spa
)) {
3173 mutex_enter(&rvd
->vdev_stat_lock
);
3174 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3175 rvd
->vdev_stat
.vs_space
+= space_delta
;
3176 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3177 mutex_exit(&rvd
->vdev_stat_lock
);
3181 ASSERT(rvd
== vd
->vdev_parent
);
3182 ASSERT(vd
->vdev_ms_count
!= 0);
3184 metaslab_class_space_update(mc
,
3185 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3190 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3191 * so that it will be written out next time the vdev configuration is synced.
3192 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3195 vdev_config_dirty(vdev_t
*vd
)
3197 spa_t
*spa
= vd
->vdev_spa
;
3198 vdev_t
*rvd
= spa
->spa_root_vdev
;
3201 ASSERT(spa_writeable(spa
));
3204 * If this is an aux vdev (as with l2cache and spare devices), then we
3205 * update the vdev config manually and set the sync flag.
3207 if (vd
->vdev_aux
!= NULL
) {
3208 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3212 for (c
= 0; c
< sav
->sav_count
; c
++) {
3213 if (sav
->sav_vdevs
[c
] == vd
)
3217 if (c
== sav
->sav_count
) {
3219 * We're being removed. There's nothing more to do.
3221 ASSERT(sav
->sav_sync
== B_TRUE
);
3225 sav
->sav_sync
= B_TRUE
;
3227 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3228 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3229 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3230 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3236 * Setting the nvlist in the middle if the array is a little
3237 * sketchy, but it will work.
3239 nvlist_free(aux
[c
]);
3240 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3246 * The dirty list is protected by the SCL_CONFIG lock. The caller
3247 * must either hold SCL_CONFIG as writer, or must be the sync thread
3248 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3249 * so this is sufficient to ensure mutual exclusion.
3251 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3252 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3253 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3256 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3257 vdev_config_dirty(rvd
->vdev_child
[c
]);
3259 ASSERT(vd
== vd
->vdev_top
);
3261 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3263 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3268 vdev_config_clean(vdev_t
*vd
)
3270 spa_t
*spa
= vd
->vdev_spa
;
3272 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3273 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3274 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3276 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3277 list_remove(&spa
->spa_config_dirty_list
, vd
);
3281 * Mark a top-level vdev's state as dirty, so that the next pass of
3282 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3283 * the state changes from larger config changes because they require
3284 * much less locking, and are often needed for administrative actions.
3287 vdev_state_dirty(vdev_t
*vd
)
3289 spa_t
*spa
= vd
->vdev_spa
;
3291 ASSERT(spa_writeable(spa
));
3292 ASSERT(vd
== vd
->vdev_top
);
3295 * The state list is protected by the SCL_STATE lock. The caller
3296 * must either hold SCL_STATE as writer, or must be the sync thread
3297 * (which holds SCL_STATE as reader). There's only one sync thread,
3298 * so this is sufficient to ensure mutual exclusion.
3300 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3301 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3302 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3304 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3305 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3309 vdev_state_clean(vdev_t
*vd
)
3311 spa_t
*spa
= vd
->vdev_spa
;
3313 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3314 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3315 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3317 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3318 list_remove(&spa
->spa_state_dirty_list
, vd
);
3322 * Propagate vdev state up from children to parent.
3325 vdev_propagate_state(vdev_t
*vd
)
3327 spa_t
*spa
= vd
->vdev_spa
;
3328 vdev_t
*rvd
= spa
->spa_root_vdev
;
3329 int degraded
= 0, faulted
= 0;
3334 if (vd
->vdev_children
> 0) {
3335 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3336 child
= vd
->vdev_child
[c
];
3339 * Don't factor holes into the decision.
3341 if (child
->vdev_ishole
)
3344 if (!vdev_readable(child
) ||
3345 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3347 * Root special: if there is a top-level log
3348 * device, treat the root vdev as if it were
3351 if (child
->vdev_islog
&& vd
== rvd
)
3355 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3359 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3363 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3366 * Root special: if there is a top-level vdev that cannot be
3367 * opened due to corrupted metadata, then propagate the root
3368 * vdev's aux state as 'corrupt' rather than 'insufficient
3371 if (corrupted
&& vd
== rvd
&&
3372 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3373 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3374 VDEV_AUX_CORRUPT_DATA
);
3377 if (vd
->vdev_parent
)
3378 vdev_propagate_state(vd
->vdev_parent
);
3382 * Set a vdev's state. If this is during an open, we don't update the parent
3383 * state, because we're in the process of opening children depth-first.
3384 * Otherwise, we propagate the change to the parent.
3386 * If this routine places a device in a faulted state, an appropriate ereport is
3390 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3392 uint64_t save_state
;
3393 spa_t
*spa
= vd
->vdev_spa
;
3395 if (state
== vd
->vdev_state
) {
3397 * Since vdev_offline() code path is already in an offline
3398 * state we can miss a statechange event to OFFLINE. Check
3399 * the previous state to catch this condition.
3401 if (vd
->vdev_ops
->vdev_op_leaf
&&
3402 (state
== VDEV_STATE_OFFLINE
) &&
3403 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
3404 /* post an offline state change */
3405 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
3407 vd
->vdev_stat
.vs_aux
= aux
;
3411 save_state
= vd
->vdev_state
;
3413 vd
->vdev_state
= state
;
3414 vd
->vdev_stat
.vs_aux
= aux
;
3417 * If we are setting the vdev state to anything but an open state, then
3418 * always close the underlying device unless the device has requested
3419 * a delayed close (i.e. we're about to remove or fault the device).
3420 * Otherwise, we keep accessible but invalid devices open forever.
3421 * We don't call vdev_close() itself, because that implies some extra
3422 * checks (offline, etc) that we don't want here. This is limited to
3423 * leaf devices, because otherwise closing the device will affect other
3426 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3427 vd
->vdev_ops
->vdev_op_leaf
)
3428 vd
->vdev_ops
->vdev_op_close(vd
);
3430 if (vd
->vdev_removed
&&
3431 state
== VDEV_STATE_CANT_OPEN
&&
3432 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3434 * If the previous state is set to VDEV_STATE_REMOVED, then this
3435 * device was previously marked removed and someone attempted to
3436 * reopen it. If this failed due to a nonexistent device, then
3437 * keep the device in the REMOVED state. We also let this be if
3438 * it is one of our special test online cases, which is only
3439 * attempting to online the device and shouldn't generate an FMA
3442 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3443 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3444 } else if (state
== VDEV_STATE_REMOVED
) {
3445 vd
->vdev_removed
= B_TRUE
;
3446 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3448 * If we fail to open a vdev during an import or recovery, we
3449 * mark it as "not available", which signifies that it was
3450 * never there to begin with. Failure to open such a device
3451 * is not considered an error.
3453 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3454 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3455 vd
->vdev_ops
->vdev_op_leaf
)
3456 vd
->vdev_not_present
= 1;
3459 * Post the appropriate ereport. If the 'prevstate' field is
3460 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3461 * that this is part of a vdev_reopen(). In this case, we don't
3462 * want to post the ereport if the device was already in the
3463 * CANT_OPEN state beforehand.
3465 * If the 'checkremove' flag is set, then this is an attempt to
3466 * online the device in response to an insertion event. If we
3467 * hit this case, then we have detected an insertion event for a
3468 * faulted or offline device that wasn't in the removed state.
3469 * In this scenario, we don't post an ereport because we are
3470 * about to replace the device, or attempt an online with
3471 * vdev_forcefault, which will generate the fault for us.
3473 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3474 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3475 vd
!= spa
->spa_root_vdev
) {
3479 case VDEV_AUX_OPEN_FAILED
:
3480 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3482 case VDEV_AUX_CORRUPT_DATA
:
3483 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3485 case VDEV_AUX_NO_REPLICAS
:
3486 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3488 case VDEV_AUX_BAD_GUID_SUM
:
3489 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3491 case VDEV_AUX_TOO_SMALL
:
3492 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3494 case VDEV_AUX_BAD_LABEL
:
3495 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3497 case VDEV_AUX_BAD_ASHIFT
:
3498 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
3501 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3504 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3507 /* Erase any notion of persistent removed state */
3508 vd
->vdev_removed
= B_FALSE
;
3510 vd
->vdev_removed
= B_FALSE
;
3514 * Notify ZED of any significant state-change on a leaf vdev.
3517 if (vd
->vdev_ops
->vdev_op_leaf
) {
3518 /* preserve original state from a vdev_reopen() */
3519 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
3520 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
3521 (save_state
<= VDEV_STATE_CLOSED
))
3522 save_state
= vd
->vdev_prevstate
;
3524 /* filter out state change due to initial vdev_open */
3525 if (save_state
> VDEV_STATE_CLOSED
)
3526 zfs_post_state_change(spa
, vd
, save_state
);
3529 if (!isopen
&& vd
->vdev_parent
)
3530 vdev_propagate_state(vd
->vdev_parent
);
3534 * Check the vdev configuration to ensure that it's capable of supporting
3538 vdev_is_bootable(vdev_t
*vd
)
3540 #if defined(__sun__) || defined(__sun)
3542 * Currently, we do not support RAID-Z or partial configuration.
3543 * In addition, only a single top-level vdev is allowed and none of the
3544 * leaves can be wholedisks.
3548 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3549 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3551 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3552 vd
->vdev_children
> 1) {
3554 } else if (strcmp(vdev_type
, VDEV_TYPE_RAIDZ
) == 0 ||
3555 strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3560 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3561 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3564 #endif /* __sun__ || __sun */
3569 * Load the state from the original vdev tree (ovd) which
3570 * we've retrieved from the MOS config object. If the original
3571 * vdev was offline or faulted then we transfer that state to the
3572 * device in the current vdev tree (nvd).
3575 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3579 ASSERT(nvd
->vdev_top
->vdev_islog
);
3580 ASSERT(spa_config_held(nvd
->vdev_spa
,
3581 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3582 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3584 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3585 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3587 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3589 * Restore the persistent vdev state
3591 nvd
->vdev_offline
= ovd
->vdev_offline
;
3592 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3593 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3594 nvd
->vdev_removed
= ovd
->vdev_removed
;
3599 * Determine if a log device has valid content. If the vdev was
3600 * removed or faulted in the MOS config then we know that
3601 * the content on the log device has already been written to the pool.
3604 vdev_log_state_valid(vdev_t
*vd
)
3608 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3612 for (c
= 0; c
< vd
->vdev_children
; c
++)
3613 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3620 * Expand a vdev if possible.
3623 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3625 ASSERT(vd
->vdev_top
== vd
);
3626 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3628 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3629 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3630 vdev_config_dirty(vd
);
3638 vdev_split(vdev_t
*vd
)
3640 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3642 vdev_remove_child(pvd
, vd
);
3643 vdev_compact_children(pvd
);
3645 cvd
= pvd
->vdev_child
[0];
3646 if (pvd
->vdev_children
== 1) {
3647 vdev_remove_parent(cvd
);
3648 cvd
->vdev_splitting
= B_TRUE
;
3650 vdev_propagate_state(cvd
);
3654 vdev_deadman(vdev_t
*vd
)
3658 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3659 vdev_t
*cvd
= vd
->vdev_child
[c
];
3664 if (vd
->vdev_ops
->vdev_op_leaf
) {
3665 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3667 mutex_enter(&vq
->vq_lock
);
3668 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3669 spa_t
*spa
= vd
->vdev_spa
;
3674 * Look at the head of all the pending queues,
3675 * if any I/O has been outstanding for longer than
3676 * the spa_deadman_synctime we log a zevent.
3678 fio
= avl_first(&vq
->vq_active_tree
);
3679 delta
= gethrtime() - fio
->io_timestamp
;
3680 if (delta
> spa_deadman_synctime(spa
)) {
3681 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3682 "delta %lluns, last io %lluns",
3683 fio
->io_timestamp
, delta
,
3684 vq
->vq_io_complete_ts
);
3685 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3686 spa
, vd
, fio
, 0, 0);
3689 mutex_exit(&vq
->vq_lock
);
3693 #if defined(_KERNEL) && defined(HAVE_SPL)
3694 EXPORT_SYMBOL(vdev_fault
);
3695 EXPORT_SYMBOL(vdev_degrade
);
3696 EXPORT_SYMBOL(vdev_online
);
3697 EXPORT_SYMBOL(vdev_offline
);
3698 EXPORT_SYMBOL(vdev_clear
);
3700 module_param(metaslabs_per_vdev
, int, 0644);
3701 MODULE_PARM_DESC(metaslabs_per_vdev
,
3702 "Divide added vdev into approximately (but no more than) this number "