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 2017 Nexenta Systems, Inc.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Toomas Soome <tsoome@me.com>
28 * Copyright 2017 Joyent, Inc.
31 #include <sys/zfs_context.h>
32 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
36 #include <sys/dmu_tx.h>
37 #include <sys/vdev_impl.h>
38 #include <sys/uberblock_impl.h>
39 #include <sys/metaslab.h>
40 #include <sys/metaslab_impl.h>
41 #include <sys/space_map.h>
42 #include <sys/space_reftree.h>
45 #include <sys/fs/zfs.h>
48 #include <sys/dsl_scan.h>
51 #include <sys/zfs_ratelimit.h>
54 * When a vdev is added, it will be divided into approximately (but no
55 * more than) this number of metaslabs.
57 int metaslabs_per_vdev
= 200;
60 * Virtual device management.
63 static vdev_ops_t
*vdev_ops_table
[] = {
77 * Given a vdev type, return the appropriate ops vector.
80 vdev_getops(const char *type
)
82 vdev_ops_t
*ops
, **opspp
;
84 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
85 if (strcmp(ops
->vdev_op_type
, type
) == 0)
92 * Default asize function: return the MAX of psize with the asize of
93 * all children. This is what's used by anything other than RAID-Z.
96 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
98 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
101 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
102 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
103 asize
= MAX(asize
, csize
);
110 * Get the minimum allocatable size. We define the allocatable size as
111 * the vdev's asize rounded to the nearest metaslab. This allows us to
112 * replace or attach devices which don't have the same physical size but
113 * can still satisfy the same number of allocations.
116 vdev_get_min_asize(vdev_t
*vd
)
118 vdev_t
*pvd
= vd
->vdev_parent
;
121 * If our parent is NULL (inactive spare or cache) or is the root,
122 * just return our own asize.
125 return (vd
->vdev_asize
);
128 * The top-level vdev just returns the allocatable size rounded
129 * to the nearest metaslab.
131 if (vd
== vd
->vdev_top
)
132 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
135 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
136 * so each child must provide at least 1/Nth of its asize.
138 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
139 return ((pvd
->vdev_min_asize
+ pvd
->vdev_children
- 1) /
142 return (pvd
->vdev_min_asize
);
146 vdev_set_min_asize(vdev_t
*vd
)
148 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
150 for (int c
= 0; c
< vd
->vdev_children
; c
++)
151 vdev_set_min_asize(vd
->vdev_child
[c
]);
155 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
157 vdev_t
*rvd
= spa
->spa_root_vdev
;
159 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
161 if (vdev
< rvd
->vdev_children
) {
162 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
163 return (rvd
->vdev_child
[vdev
]);
170 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
174 if (vd
->vdev_guid
== guid
)
177 for (int c
= 0; c
< vd
->vdev_children
; c
++)
178 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
186 vdev_count_leaves_impl(vdev_t
*vd
)
190 if (vd
->vdev_ops
->vdev_op_leaf
)
193 for (int 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
;
289 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
291 for (int c
= newc
= 0; c
< oldc
; c
++)
292 if (pvd
->vdev_child
[c
])
295 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
297 for (int c
= newc
= 0; c
< oldc
; c
++) {
298 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
299 newchild
[newc
] = cvd
;
300 cvd
->vdev_id
= newc
++;
304 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
305 pvd
->vdev_child
= newchild
;
306 pvd
->vdev_children
= newc
;
310 * Allocate and minimally initialize a vdev_t.
313 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
317 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
319 if (spa
->spa_root_vdev
== NULL
) {
320 ASSERT(ops
== &vdev_root_ops
);
321 spa
->spa_root_vdev
= vd
;
322 spa
->spa_load_guid
= spa_generate_guid(NULL
);
325 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
326 if (spa
->spa_root_vdev
== vd
) {
328 * The root vdev's guid will also be the pool guid,
329 * which must be unique among all pools.
331 guid
= spa_generate_guid(NULL
);
334 * Any other vdev's guid must be unique within the pool.
336 guid
= spa_generate_guid(spa
);
338 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
343 vd
->vdev_guid
= guid
;
344 vd
->vdev_guid_sum
= guid
;
346 vd
->vdev_state
= VDEV_STATE_CLOSED
;
347 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
350 * Initialize rate limit structs for events. We rate limit ZIO delay
351 * and checksum events so that we don't overwhelm ZED with thousands
352 * of events when a disk is acting up.
354 zfs_ratelimit_init(&vd
->vdev_delay_rl
, DELAYS_PER_SECOND
, 1);
355 zfs_ratelimit_init(&vd
->vdev_checksum_rl
, CHECKSUMS_PER_SECOND
, 1);
357 list_link_init(&vd
->vdev_config_dirty_node
);
358 list_link_init(&vd
->vdev_state_dirty_node
);
359 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
360 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
361 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
362 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
363 mutex_init(&vd
->vdev_scan_io_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
365 for (int t
= 0; t
< DTL_TYPES
; t
++) {
366 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
369 txg_list_create(&vd
->vdev_ms_list
, spa
,
370 offsetof(struct metaslab
, ms_txg_node
));
371 txg_list_create(&vd
->vdev_dtl_list
, spa
,
372 offsetof(struct vdev
, vdev_dtl_node
));
373 vd
->vdev_stat
.vs_timestamp
= gethrtime();
381 * Allocate a new vdev. The 'alloctype' is used to control whether we are
382 * creating a new vdev or loading an existing one - the behavior is slightly
383 * different for each case.
386 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
391 uint64_t guid
= 0, islog
, nparity
;
396 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
398 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
399 return (SET_ERROR(EINVAL
));
401 if ((ops
= vdev_getops(type
)) == NULL
)
402 return (SET_ERROR(EINVAL
));
405 * If this is a load, get the vdev guid from the nvlist.
406 * Otherwise, vdev_alloc_common() will generate one for us.
408 if (alloctype
== VDEV_ALLOC_LOAD
) {
411 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
413 return (SET_ERROR(EINVAL
));
415 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
416 return (SET_ERROR(EINVAL
));
417 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
418 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
419 return (SET_ERROR(EINVAL
));
420 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
421 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
422 return (SET_ERROR(EINVAL
));
423 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
424 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
425 return (SET_ERROR(EINVAL
));
429 * The first allocated vdev must be of type 'root'.
431 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
432 return (SET_ERROR(EINVAL
));
435 * Determine whether we're a log vdev.
438 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
439 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
440 return (SET_ERROR(ENOTSUP
));
442 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
443 return (SET_ERROR(ENOTSUP
));
446 * Set the nparity property for RAID-Z vdevs.
449 if (ops
== &vdev_raidz_ops
) {
450 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
452 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
453 return (SET_ERROR(EINVAL
));
455 * Previous versions could only support 1 or 2 parity
459 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
460 return (SET_ERROR(ENOTSUP
));
462 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
463 return (SET_ERROR(ENOTSUP
));
466 * We require the parity to be specified for SPAs that
467 * support multiple parity levels.
469 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
470 return (SET_ERROR(EINVAL
));
472 * Otherwise, we default to 1 parity device for RAID-Z.
479 ASSERT(nparity
!= -1ULL);
481 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
483 vd
->vdev_islog
= islog
;
484 vd
->vdev_nparity
= nparity
;
486 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
487 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
490 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
491 * fault on a vdev and want it to persist across imports (like with
494 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
495 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
496 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
497 vd
->vdev_faulted
= 1;
498 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
501 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
502 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
503 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
504 &vd
->vdev_physpath
) == 0)
505 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
507 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
508 &vd
->vdev_enc_sysfs_path
) == 0)
509 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
511 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
512 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
515 * Set the whole_disk property. If it's not specified, leave the value
518 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
519 &vd
->vdev_wholedisk
) != 0)
520 vd
->vdev_wholedisk
= -1ULL;
523 * Look for the 'not present' flag. This will only be set if the device
524 * was not present at the time of import.
526 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
527 &vd
->vdev_not_present
);
530 * Get the alignment requirement.
532 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
535 * Retrieve the vdev creation time.
537 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
541 * If we're a top-level vdev, try to load the allocation parameters.
543 if (parent
&& !parent
->vdev_parent
&&
544 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
545 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
547 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
549 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
551 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
553 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
556 ASSERT0(vd
->vdev_top_zap
);
559 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
560 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
561 alloctype
== VDEV_ALLOC_ADD
||
562 alloctype
== VDEV_ALLOC_SPLIT
||
563 alloctype
== VDEV_ALLOC_ROOTPOOL
);
564 vd
->vdev_mg
= metaslab_group_create(islog
?
565 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
568 if (vd
->vdev_ops
->vdev_op_leaf
&&
569 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
570 (void) nvlist_lookup_uint64(nv
,
571 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
573 ASSERT0(vd
->vdev_leaf_zap
);
577 * If we're a leaf vdev, try to load the DTL object and other state.
580 if (vd
->vdev_ops
->vdev_op_leaf
&&
581 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
582 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
583 if (alloctype
== VDEV_ALLOC_LOAD
) {
584 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
585 &vd
->vdev_dtl_object
);
586 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
590 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
593 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
594 &spare
) == 0 && spare
)
598 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
601 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
602 &vd
->vdev_resilver_txg
);
605 * In general, when importing a pool we want to ignore the
606 * persistent fault state, as the diagnosis made on another
607 * system may not be valid in the current context. The only
608 * exception is if we forced a vdev to a persistently faulted
609 * state with 'zpool offline -f'. The persistent fault will
610 * remain across imports until cleared.
612 * Local vdevs will remain in the faulted state.
614 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
615 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
616 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
618 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
620 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
623 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
627 VDEV_AUX_ERR_EXCEEDED
;
628 if (nvlist_lookup_string(nv
,
629 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
630 strcmp(aux
, "external") == 0)
631 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
637 * Add ourselves to the parent's list of children.
639 vdev_add_child(parent
, vd
);
647 vdev_free(vdev_t
*vd
)
649 spa_t
*spa
= vd
->vdev_spa
;
652 * Scan queues are normally destroyed at the end of a scan. If the
653 * queue exists here, that implies the vdev is being removed while
654 * the scan is still running.
656 if (vd
->vdev_scan_io_queue
!= NULL
) {
657 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
658 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
659 vd
->vdev_scan_io_queue
= NULL
;
660 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
664 * vdev_free() implies closing the vdev first. This is simpler than
665 * trying to ensure complicated semantics for all callers.
669 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
670 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
675 for (int c
= 0; c
< vd
->vdev_children
; c
++)
676 vdev_free(vd
->vdev_child
[c
]);
678 ASSERT(vd
->vdev_child
== NULL
);
679 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
682 * Discard allocation state.
684 if (vd
->vdev_mg
!= NULL
) {
685 vdev_metaslab_fini(vd
);
686 metaslab_group_destroy(vd
->vdev_mg
);
689 ASSERT0(vd
->vdev_stat
.vs_space
);
690 ASSERT0(vd
->vdev_stat
.vs_dspace
);
691 ASSERT0(vd
->vdev_stat
.vs_alloc
);
694 * Remove this vdev from its parent's child list.
696 vdev_remove_child(vd
->vdev_parent
, vd
);
698 ASSERT(vd
->vdev_parent
== NULL
);
701 * Clean up vdev structure.
707 spa_strfree(vd
->vdev_path
);
709 spa_strfree(vd
->vdev_devid
);
710 if (vd
->vdev_physpath
)
711 spa_strfree(vd
->vdev_physpath
);
713 if (vd
->vdev_enc_sysfs_path
)
714 spa_strfree(vd
->vdev_enc_sysfs_path
);
717 spa_strfree(vd
->vdev_fru
);
719 if (vd
->vdev_isspare
)
720 spa_spare_remove(vd
);
721 if (vd
->vdev_isl2cache
)
722 spa_l2cache_remove(vd
);
724 txg_list_destroy(&vd
->vdev_ms_list
);
725 txg_list_destroy(&vd
->vdev_dtl_list
);
727 mutex_enter(&vd
->vdev_dtl_lock
);
728 space_map_close(vd
->vdev_dtl_sm
);
729 for (int t
= 0; t
< DTL_TYPES
; t
++) {
730 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
731 range_tree_destroy(vd
->vdev_dtl
[t
]);
733 mutex_exit(&vd
->vdev_dtl_lock
);
735 mutex_destroy(&vd
->vdev_queue_lock
);
736 mutex_destroy(&vd
->vdev_dtl_lock
);
737 mutex_destroy(&vd
->vdev_stat_lock
);
738 mutex_destroy(&vd
->vdev_probe_lock
);
739 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
741 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
742 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
744 if (vd
== spa
->spa_root_vdev
)
745 spa
->spa_root_vdev
= NULL
;
747 kmem_free(vd
, sizeof (vdev_t
));
751 * Transfer top-level vdev state from svd to tvd.
754 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
756 spa_t
*spa
= svd
->vdev_spa
;
761 ASSERT(tvd
== tvd
->vdev_top
);
763 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
764 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
765 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
766 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
767 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
769 svd
->vdev_ms_array
= 0;
770 svd
->vdev_ms_shift
= 0;
771 svd
->vdev_ms_count
= 0;
772 svd
->vdev_top_zap
= 0;
775 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
776 tvd
->vdev_mg
= svd
->vdev_mg
;
777 tvd
->vdev_ms
= svd
->vdev_ms
;
782 if (tvd
->vdev_mg
!= NULL
)
783 tvd
->vdev_mg
->mg_vd
= tvd
;
785 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
786 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
787 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
789 svd
->vdev_stat
.vs_alloc
= 0;
790 svd
->vdev_stat
.vs_space
= 0;
791 svd
->vdev_stat
.vs_dspace
= 0;
793 for (t
= 0; t
< TXG_SIZE
; t
++) {
794 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
795 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
796 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
797 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
798 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
799 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
802 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
803 vdev_config_clean(svd
);
804 vdev_config_dirty(tvd
);
807 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
808 vdev_state_clean(svd
);
809 vdev_state_dirty(tvd
);
812 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
813 svd
->vdev_deflate_ratio
= 0;
815 tvd
->vdev_islog
= svd
->vdev_islog
;
818 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
822 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
829 for (int c
= 0; c
< vd
->vdev_children
; c
++)
830 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
834 * Add a mirror/replacing vdev above an existing vdev.
837 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
839 spa_t
*spa
= cvd
->vdev_spa
;
840 vdev_t
*pvd
= cvd
->vdev_parent
;
843 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
845 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
847 mvd
->vdev_asize
= cvd
->vdev_asize
;
848 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
849 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
850 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
851 mvd
->vdev_state
= cvd
->vdev_state
;
852 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
854 vdev_remove_child(pvd
, cvd
);
855 vdev_add_child(pvd
, mvd
);
856 cvd
->vdev_id
= mvd
->vdev_children
;
857 vdev_add_child(mvd
, cvd
);
858 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
860 if (mvd
== mvd
->vdev_top
)
861 vdev_top_transfer(cvd
, mvd
);
867 * Remove a 1-way mirror/replacing vdev from the tree.
870 vdev_remove_parent(vdev_t
*cvd
)
872 vdev_t
*mvd
= cvd
->vdev_parent
;
873 vdev_t
*pvd
= mvd
->vdev_parent
;
875 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
877 ASSERT(mvd
->vdev_children
== 1);
878 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
879 mvd
->vdev_ops
== &vdev_replacing_ops
||
880 mvd
->vdev_ops
== &vdev_spare_ops
);
881 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
883 vdev_remove_child(mvd
, cvd
);
884 vdev_remove_child(pvd
, mvd
);
887 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
888 * Otherwise, we could have detached an offline device, and when we
889 * go to import the pool we'll think we have two top-level vdevs,
890 * instead of a different version of the same top-level vdev.
892 if (mvd
->vdev_top
== mvd
) {
893 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
894 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
895 cvd
->vdev_guid
+= guid_delta
;
896 cvd
->vdev_guid_sum
+= guid_delta
;
899 * If pool not set for autoexpand, we need to also preserve
900 * mvd's asize to prevent automatic expansion of cvd.
901 * Otherwise if we are adjusting the mirror by attaching and
902 * detaching children of non-uniform sizes, the mirror could
903 * autoexpand, unexpectedly requiring larger devices to
904 * re-establish the mirror.
906 if (!cvd
->vdev_spa
->spa_autoexpand
)
907 cvd
->vdev_asize
= mvd
->vdev_asize
;
909 cvd
->vdev_id
= mvd
->vdev_id
;
910 vdev_add_child(pvd
, cvd
);
911 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
913 if (cvd
== cvd
->vdev_top
)
914 vdev_top_transfer(mvd
, cvd
);
916 ASSERT(mvd
->vdev_children
== 0);
921 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
923 spa_t
*spa
= vd
->vdev_spa
;
924 objset_t
*mos
= spa
->spa_meta_objset
;
926 uint64_t oldc
= vd
->vdev_ms_count
;
927 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
931 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
934 * This vdev is not being allocated from yet or is a hole.
936 if (vd
->vdev_ms_shift
== 0)
939 ASSERT(!vd
->vdev_ishole
);
942 * Compute the raidz-deflation ratio. Note, we hard-code
943 * in 128k (1 << 17) because it is the "typical" blocksize.
944 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
945 * otherwise it would inconsistently account for existing bp's.
947 vd
->vdev_deflate_ratio
= (1 << 17) /
948 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
950 ASSERT(oldc
<= newc
);
952 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
955 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
956 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
960 vd
->vdev_ms_count
= newc
;
962 for (m
= oldc
; m
< newc
; m
++) {
966 error
= dmu_read(mos
, vd
->vdev_ms_array
,
967 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
973 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
980 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
983 * If the vdev is being removed we don't activate
984 * the metaslabs since we want to ensure that no new
985 * allocations are performed on this device.
987 if (oldc
== 0 && !vd
->vdev_removing
)
988 metaslab_group_activate(vd
->vdev_mg
);
991 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
997 vdev_metaslab_fini(vdev_t
*vd
)
1000 uint64_t count
= vd
->vdev_ms_count
;
1002 if (vd
->vdev_ms
!= NULL
) {
1003 metaslab_group_passivate(vd
->vdev_mg
);
1004 for (m
= 0; m
< count
; m
++) {
1005 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1010 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1014 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1017 typedef struct vdev_probe_stats
{
1018 boolean_t vps_readable
;
1019 boolean_t vps_writeable
;
1021 } vdev_probe_stats_t
;
1024 vdev_probe_done(zio_t
*zio
)
1026 spa_t
*spa
= zio
->io_spa
;
1027 vdev_t
*vd
= zio
->io_vd
;
1028 vdev_probe_stats_t
*vps
= zio
->io_private
;
1030 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1032 if (zio
->io_type
== ZIO_TYPE_READ
) {
1033 if (zio
->io_error
== 0)
1034 vps
->vps_readable
= 1;
1035 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1036 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1037 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1038 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1039 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1041 abd_free(zio
->io_abd
);
1043 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1044 if (zio
->io_error
== 0)
1045 vps
->vps_writeable
= 1;
1046 abd_free(zio
->io_abd
);
1047 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1051 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1052 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1054 if (vdev_readable(vd
) &&
1055 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1058 ASSERT(zio
->io_error
!= 0);
1059 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1060 spa
, vd
, NULL
, NULL
, 0, 0);
1061 zio
->io_error
= SET_ERROR(ENXIO
);
1064 mutex_enter(&vd
->vdev_probe_lock
);
1065 ASSERT(vd
->vdev_probe_zio
== zio
);
1066 vd
->vdev_probe_zio
= NULL
;
1067 mutex_exit(&vd
->vdev_probe_lock
);
1070 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1071 if (!vdev_accessible(vd
, pio
))
1072 pio
->io_error
= SET_ERROR(ENXIO
);
1074 kmem_free(vps
, sizeof (*vps
));
1079 * Determine whether this device is accessible.
1081 * Read and write to several known locations: the pad regions of each
1082 * vdev label but the first, which we leave alone in case it contains
1086 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1088 spa_t
*spa
= vd
->vdev_spa
;
1089 vdev_probe_stats_t
*vps
= NULL
;
1092 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1095 * Don't probe the probe.
1097 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1101 * To prevent 'probe storms' when a device fails, we create
1102 * just one probe i/o at a time. All zios that want to probe
1103 * this vdev will become parents of the probe io.
1105 mutex_enter(&vd
->vdev_probe_lock
);
1107 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1108 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1110 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1111 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1114 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1116 * vdev_cant_read and vdev_cant_write can only
1117 * transition from TRUE to FALSE when we have the
1118 * SCL_ZIO lock as writer; otherwise they can only
1119 * transition from FALSE to TRUE. This ensures that
1120 * any zio looking at these values can assume that
1121 * failures persist for the life of the I/O. That's
1122 * important because when a device has intermittent
1123 * connectivity problems, we want to ensure that
1124 * they're ascribed to the device (ENXIO) and not
1127 * Since we hold SCL_ZIO as writer here, clear both
1128 * values so the probe can reevaluate from first
1131 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1132 vd
->vdev_cant_read
= B_FALSE
;
1133 vd
->vdev_cant_write
= B_FALSE
;
1136 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1137 vdev_probe_done
, vps
,
1138 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1141 * We can't change the vdev state in this context, so we
1142 * kick off an async task to do it on our behalf.
1145 vd
->vdev_probe_wanted
= B_TRUE
;
1146 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1151 zio_add_child(zio
, pio
);
1153 mutex_exit(&vd
->vdev_probe_lock
);
1156 ASSERT(zio
!= NULL
);
1160 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1161 zio_nowait(zio_read_phys(pio
, vd
,
1162 vdev_label_offset(vd
->vdev_psize
, l
,
1163 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1164 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1165 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1166 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1177 vdev_open_child(void *arg
)
1181 vd
->vdev_open_thread
= curthread
;
1182 vd
->vdev_open_error
= vdev_open(vd
);
1183 vd
->vdev_open_thread
= NULL
;
1187 vdev_uses_zvols(vdev_t
*vd
)
1190 if (zvol_is_zvol(vd
->vdev_path
))
1194 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1195 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1202 vdev_open_children(vdev_t
*vd
)
1205 int children
= vd
->vdev_children
;
1208 * in order to handle pools on top of zvols, do the opens
1209 * in a single thread so that the same thread holds the
1210 * spa_namespace_lock
1212 if (vdev_uses_zvols(vd
)) {
1214 for (int c
= 0; c
< children
; c
++)
1215 vd
->vdev_child
[c
]->vdev_open_error
=
1216 vdev_open(vd
->vdev_child
[c
]);
1218 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1219 children
, children
, TASKQ_PREPOPULATE
);
1223 for (int c
= 0; c
< children
; c
++)
1224 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1225 vd
->vdev_child
[c
], TQ_SLEEP
) != TASKQID_INVALID
);
1230 vd
->vdev_nonrot
= B_TRUE
;
1232 for (int c
= 0; c
< children
; c
++)
1233 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1237 * Prepare a virtual device for access.
1240 vdev_open(vdev_t
*vd
)
1242 spa_t
*spa
= vd
->vdev_spa
;
1245 uint64_t max_osize
= 0;
1246 uint64_t asize
, max_asize
, psize
;
1247 uint64_t ashift
= 0;
1249 ASSERT(vd
->vdev_open_thread
== curthread
||
1250 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1251 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1252 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1253 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1255 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1256 vd
->vdev_cant_read
= B_FALSE
;
1257 vd
->vdev_cant_write
= B_FALSE
;
1258 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1261 * If this vdev is not removed, check its fault status. If it's
1262 * faulted, bail out of the open.
1264 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1265 ASSERT(vd
->vdev_children
== 0);
1266 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1267 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1268 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1269 vd
->vdev_label_aux
);
1270 return (SET_ERROR(ENXIO
));
1271 } else if (vd
->vdev_offline
) {
1272 ASSERT(vd
->vdev_children
== 0);
1273 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1274 return (SET_ERROR(ENXIO
));
1277 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1280 * Reset the vdev_reopening flag so that we actually close
1281 * the vdev on error.
1283 vd
->vdev_reopening
= B_FALSE
;
1284 if (zio_injection_enabled
&& error
== 0)
1285 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1288 if (vd
->vdev_removed
&&
1289 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1290 vd
->vdev_removed
= B_FALSE
;
1292 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1293 vd
->vdev_stat
.vs_aux
);
1297 vd
->vdev_removed
= B_FALSE
;
1300 * Recheck the faulted flag now that we have confirmed that
1301 * the vdev is accessible. If we're faulted, bail.
1303 if (vd
->vdev_faulted
) {
1304 ASSERT(vd
->vdev_children
== 0);
1305 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1306 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1307 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1308 vd
->vdev_label_aux
);
1309 return (SET_ERROR(ENXIO
));
1312 if (vd
->vdev_degraded
) {
1313 ASSERT(vd
->vdev_children
== 0);
1314 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1315 VDEV_AUX_ERR_EXCEEDED
);
1317 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1321 * For hole or missing vdevs we just return success.
1323 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1326 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1327 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1328 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1334 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1335 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1337 if (vd
->vdev_children
== 0) {
1338 if (osize
< SPA_MINDEVSIZE
) {
1339 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1340 VDEV_AUX_TOO_SMALL
);
1341 return (SET_ERROR(EOVERFLOW
));
1344 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1345 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1346 VDEV_LABEL_END_SIZE
);
1348 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1349 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1350 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1351 VDEV_AUX_TOO_SMALL
);
1352 return (SET_ERROR(EOVERFLOW
));
1356 max_asize
= max_osize
;
1360 * If the vdev was expanded, record this so that we can re-create the
1361 * uberblock rings in labels {2,3}, during the next sync.
1363 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
1364 vd
->vdev_copy_uberblocks
= B_TRUE
;
1366 vd
->vdev_psize
= psize
;
1369 * Make sure the allocatable size hasn't shrunk too much.
1371 if (asize
< vd
->vdev_min_asize
) {
1372 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1373 VDEV_AUX_BAD_LABEL
);
1374 return (SET_ERROR(EINVAL
));
1377 if (vd
->vdev_asize
== 0) {
1379 * This is the first-ever open, so use the computed values.
1380 * For compatibility, a different ashift can be requested.
1382 vd
->vdev_asize
= asize
;
1383 vd
->vdev_max_asize
= max_asize
;
1384 if (vd
->vdev_ashift
== 0) {
1385 vd
->vdev_ashift
= ashift
; /* use detected value */
1387 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1388 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1389 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1390 VDEV_AUX_BAD_ASHIFT
);
1391 return (SET_ERROR(EDOM
));
1395 * Detect if the alignment requirement has increased.
1396 * We don't want to make the pool unavailable, just
1397 * post an event instead.
1399 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1400 vd
->vdev_ops
->vdev_op_leaf
) {
1401 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1402 spa
, vd
, NULL
, NULL
, 0, 0);
1405 vd
->vdev_max_asize
= max_asize
;
1409 * If all children are healthy we update asize if either:
1410 * The asize has increased, due to a device expansion caused by dynamic
1411 * LUN growth or vdev replacement, and automatic expansion is enabled;
1412 * making the additional space available.
1414 * The asize has decreased, due to a device shrink usually caused by a
1415 * vdev replace with a smaller device. This ensures that calculations
1416 * based of max_asize and asize e.g. esize are always valid. It's safe
1417 * to do this as we've already validated that asize is greater than
1420 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1421 ((asize
> vd
->vdev_asize
&&
1422 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1423 (asize
< vd
->vdev_asize
)))
1424 vd
->vdev_asize
= asize
;
1426 vdev_set_min_asize(vd
);
1429 * Ensure we can issue some IO before declaring the
1430 * vdev open for business.
1432 if (vd
->vdev_ops
->vdev_op_leaf
&&
1433 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1434 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1435 VDEV_AUX_ERR_EXCEEDED
);
1440 * Track the min and max ashift values for normal data devices.
1442 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1443 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1444 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1445 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1446 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1447 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1451 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1452 * resilver. But don't do this if we are doing a reopen for a scrub,
1453 * since this would just restart the scrub we are already doing.
1455 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1456 vdev_resilver_needed(vd
, NULL
, NULL
))
1457 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1463 * Called once the vdevs are all opened, this routine validates the label
1464 * contents. This needs to be done before vdev_load() so that we don't
1465 * inadvertently do repair I/Os to the wrong device.
1467 * If 'strict' is false ignore the spa guid check. This is necessary because
1468 * if the machine crashed during a re-guid the new guid might have been written
1469 * to all of the vdev labels, but not the cached config. The strict check
1470 * will be performed when the pool is opened again using the mos config.
1472 * This function will only return failure if one of the vdevs indicates that it
1473 * has since been destroyed or exported. This is only possible if
1474 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1475 * will be updated but the function will return 0.
1478 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1480 spa_t
*spa
= vd
->vdev_spa
;
1482 uint64_t guid
= 0, top_guid
;
1485 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1486 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1487 return (SET_ERROR(EBADF
));
1490 * If the device has already failed, or was marked offline, don't do
1491 * any further validation. Otherwise, label I/O will fail and we will
1492 * overwrite the previous state.
1494 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1495 uint64_t aux_guid
= 0;
1497 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1498 spa_last_synced_txg(spa
) : -1ULL;
1500 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1501 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1502 VDEV_AUX_BAD_LABEL
);
1507 * Determine if this vdev has been split off into another
1508 * pool. If so, then refuse to open it.
1510 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1511 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1512 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1513 VDEV_AUX_SPLIT_POOL
);
1518 if (strict
&& (nvlist_lookup_uint64(label
,
1519 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1520 guid
!= spa_guid(spa
))) {
1521 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1522 VDEV_AUX_CORRUPT_DATA
);
1527 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1528 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1533 * If this vdev just became a top-level vdev because its
1534 * sibling was detached, it will have adopted the parent's
1535 * vdev guid -- but the label may or may not be on disk yet.
1536 * Fortunately, either version of the label will have the
1537 * same top guid, so if we're a top-level vdev, we can
1538 * safely compare to that instead.
1540 * If we split this vdev off instead, then we also check the
1541 * original pool's guid. We don't want to consider the vdev
1542 * corrupt if it is partway through a split operation.
1544 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1546 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1548 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1549 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1550 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1551 VDEV_AUX_CORRUPT_DATA
);
1556 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1558 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1559 VDEV_AUX_CORRUPT_DATA
);
1567 * If this is a verbatim import, no need to check the
1568 * state of the pool.
1570 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1571 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1572 state
!= POOL_STATE_ACTIVE
)
1573 return (SET_ERROR(EBADF
));
1576 * If we were able to open and validate a vdev that was
1577 * previously marked permanently unavailable, clear that state
1580 if (vd
->vdev_not_present
)
1581 vd
->vdev_not_present
= 0;
1588 * Close a virtual device.
1591 vdev_close(vdev_t
*vd
)
1593 vdev_t
*pvd
= vd
->vdev_parent
;
1594 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1596 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1599 * If our parent is reopening, then we are as well, unless we are
1602 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1603 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1605 vd
->vdev_ops
->vdev_op_close(vd
);
1607 vdev_cache_purge(vd
);
1610 * We record the previous state before we close it, so that if we are
1611 * doing a reopen(), we don't generate FMA ereports if we notice that
1612 * it's still faulted.
1614 vd
->vdev_prevstate
= vd
->vdev_state
;
1616 if (vd
->vdev_offline
)
1617 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1619 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1620 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1624 vdev_hold(vdev_t
*vd
)
1626 spa_t
*spa
= vd
->vdev_spa
;
1628 ASSERT(spa_is_root(spa
));
1629 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1632 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1633 vdev_hold(vd
->vdev_child
[c
]);
1635 if (vd
->vdev_ops
->vdev_op_leaf
)
1636 vd
->vdev_ops
->vdev_op_hold(vd
);
1640 vdev_rele(vdev_t
*vd
)
1642 ASSERT(spa_is_root(vd
->vdev_spa
));
1643 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1644 vdev_rele(vd
->vdev_child
[c
]);
1646 if (vd
->vdev_ops
->vdev_op_leaf
)
1647 vd
->vdev_ops
->vdev_op_rele(vd
);
1651 * Reopen all interior vdevs and any unopened leaves. We don't actually
1652 * reopen leaf vdevs which had previously been opened as they might deadlock
1653 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1654 * If the leaf has never been opened then open it, as usual.
1657 vdev_reopen(vdev_t
*vd
)
1659 spa_t
*spa
= vd
->vdev_spa
;
1661 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1663 /* set the reopening flag unless we're taking the vdev offline */
1664 vd
->vdev_reopening
= !vd
->vdev_offline
;
1666 (void) vdev_open(vd
);
1669 * Call vdev_validate() here to make sure we have the same device.
1670 * Otherwise, a device with an invalid label could be successfully
1671 * opened in response to vdev_reopen().
1674 (void) vdev_validate_aux(vd
);
1675 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1676 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1677 !l2arc_vdev_present(vd
))
1678 l2arc_add_vdev(spa
, vd
);
1680 (void) vdev_validate(vd
, B_TRUE
);
1684 * Reassess parent vdev's health.
1686 vdev_propagate_state(vd
);
1690 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1695 * Normally, partial opens (e.g. of a mirror) are allowed.
1696 * For a create, however, we want to fail the request if
1697 * there are any components we can't open.
1699 error
= vdev_open(vd
);
1701 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1703 return (error
? error
: ENXIO
);
1707 * Recursively load DTLs and initialize all labels.
1709 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1710 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1711 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1720 vdev_metaslab_set_size(vdev_t
*vd
)
1723 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1725 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1726 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1730 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1732 ASSERT(vd
== vd
->vdev_top
);
1733 ASSERT(!vd
->vdev_ishole
);
1734 ASSERT(ISP2(flags
));
1735 ASSERT(spa_writeable(vd
->vdev_spa
));
1737 if (flags
& VDD_METASLAB
)
1738 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1740 if (flags
& VDD_DTL
)
1741 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1743 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1747 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1749 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1750 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1752 if (vd
->vdev_ops
->vdev_op_leaf
)
1753 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1759 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1760 * the vdev has less than perfect replication. There are four kinds of DTL:
1762 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1764 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1766 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1767 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1768 * txgs that was scrubbed.
1770 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1771 * persistent errors or just some device being offline.
1772 * Unlike the other three, the DTL_OUTAGE map is not generally
1773 * maintained; it's only computed when needed, typically to
1774 * determine whether a device can be detached.
1776 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1777 * either has the data or it doesn't.
1779 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1780 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1781 * if any child is less than fully replicated, then so is its parent.
1782 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1783 * comprising only those txgs which appear in 'maxfaults' or more children;
1784 * those are the txgs we don't have enough replication to read. For example,
1785 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1786 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1787 * two child DTL_MISSING maps.
1789 * It should be clear from the above that to compute the DTLs and outage maps
1790 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1791 * Therefore, that is all we keep on disk. When loading the pool, or after
1792 * a configuration change, we generate all other DTLs from first principles.
1795 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1797 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1799 ASSERT(t
< DTL_TYPES
);
1800 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1801 ASSERT(spa_writeable(vd
->vdev_spa
));
1803 mutex_enter(rt
->rt_lock
);
1804 if (!range_tree_contains(rt
, txg
, size
))
1805 range_tree_add(rt
, txg
, size
);
1806 mutex_exit(rt
->rt_lock
);
1810 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1812 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1813 boolean_t dirty
= B_FALSE
;
1815 ASSERT(t
< DTL_TYPES
);
1816 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1818 mutex_enter(rt
->rt_lock
);
1819 if (range_tree_space(rt
) != 0)
1820 dirty
= range_tree_contains(rt
, txg
, size
);
1821 mutex_exit(rt
->rt_lock
);
1827 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1829 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1832 mutex_enter(rt
->rt_lock
);
1833 empty
= (range_tree_space(rt
) == 0);
1834 mutex_exit(rt
->rt_lock
);
1840 * Returns B_TRUE if vdev determines offset needs to be resilvered.
1843 vdev_dtl_need_resilver(vdev_t
*vd
, uint64_t offset
, size_t psize
)
1845 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1847 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
1848 vd
->vdev_ops
->vdev_op_leaf
)
1851 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, offset
, psize
));
1855 * Returns the lowest txg in the DTL range.
1858 vdev_dtl_min(vdev_t
*vd
)
1862 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1863 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1864 ASSERT0(vd
->vdev_children
);
1866 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1867 return (rs
->rs_start
- 1);
1871 * Returns the highest txg in the DTL.
1874 vdev_dtl_max(vdev_t
*vd
)
1878 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1879 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1880 ASSERT0(vd
->vdev_children
);
1882 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1883 return (rs
->rs_end
);
1887 * Determine if a resilvering vdev should remove any DTL entries from
1888 * its range. If the vdev was resilvering for the entire duration of the
1889 * scan then it should excise that range from its DTLs. Otherwise, this
1890 * vdev is considered partially resilvered and should leave its DTL
1891 * entries intact. The comment in vdev_dtl_reassess() describes how we
1895 vdev_dtl_should_excise(vdev_t
*vd
)
1897 spa_t
*spa
= vd
->vdev_spa
;
1898 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1900 ASSERT0(scn
->scn_phys
.scn_errors
);
1901 ASSERT0(vd
->vdev_children
);
1903 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
1906 if (vd
->vdev_resilver_txg
== 0 ||
1907 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1911 * When a resilver is initiated the scan will assign the scn_max_txg
1912 * value to the highest txg value that exists in all DTLs. If this
1913 * device's max DTL is not part of this scan (i.e. it is not in
1914 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1917 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1918 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1919 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1920 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1927 * Reassess DTLs after a config change or scrub completion.
1930 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1932 spa_t
*spa
= vd
->vdev_spa
;
1936 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1938 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1939 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1940 scrub_txg
, scrub_done
);
1942 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1945 if (vd
->vdev_ops
->vdev_op_leaf
) {
1946 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1948 mutex_enter(&vd
->vdev_dtl_lock
);
1951 * If we've completed a scan cleanly then determine
1952 * if this vdev should remove any DTLs. We only want to
1953 * excise regions on vdevs that were available during
1954 * the entire duration of this scan.
1956 if (scrub_txg
!= 0 &&
1957 (spa
->spa_scrub_started
||
1958 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1959 vdev_dtl_should_excise(vd
)) {
1961 * We completed a scrub up to scrub_txg. If we
1962 * did it without rebooting, then the scrub dtl
1963 * will be valid, so excise the old region and
1964 * fold in the scrub dtl. Otherwise, leave the
1965 * dtl as-is if there was an error.
1967 * There's little trick here: to excise the beginning
1968 * of the DTL_MISSING map, we put it into a reference
1969 * tree and then add a segment with refcnt -1 that
1970 * covers the range [0, scrub_txg). This means
1971 * that each txg in that range has refcnt -1 or 0.
1972 * We then add DTL_SCRUB with a refcnt of 2, so that
1973 * entries in the range [0, scrub_txg) will have a
1974 * positive refcnt -- either 1 or 2. We then convert
1975 * the reference tree into the new DTL_MISSING map.
1977 space_reftree_create(&reftree
);
1978 space_reftree_add_map(&reftree
,
1979 vd
->vdev_dtl
[DTL_MISSING
], 1);
1980 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1981 space_reftree_add_map(&reftree
,
1982 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1983 space_reftree_generate_map(&reftree
,
1984 vd
->vdev_dtl
[DTL_MISSING
], 1);
1985 space_reftree_destroy(&reftree
);
1987 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1988 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1989 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1991 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1992 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1993 if (!vdev_readable(vd
))
1994 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1996 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1997 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
2000 * If the vdev was resilvering and no longer has any
2001 * DTLs then reset its resilvering flag and dirty
2002 * the top level so that we persist the change.
2004 if (vd
->vdev_resilver_txg
!= 0 &&
2005 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
2006 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
2007 vd
->vdev_resilver_txg
= 0;
2008 vdev_config_dirty(vd
->vdev_top
);
2011 mutex_exit(&vd
->vdev_dtl_lock
);
2014 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
2018 mutex_enter(&vd
->vdev_dtl_lock
);
2019 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2020 /* account for child's outage in parent's missing map */
2021 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
2023 continue; /* leaf vdevs only */
2024 if (t
== DTL_PARTIAL
)
2025 minref
= 1; /* i.e. non-zero */
2026 else if (vd
->vdev_nparity
!= 0)
2027 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
2029 minref
= vd
->vdev_children
; /* any kind of mirror */
2030 space_reftree_create(&reftree
);
2031 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2032 vdev_t
*cvd
= vd
->vdev_child
[c
];
2033 mutex_enter(&cvd
->vdev_dtl_lock
);
2034 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
2035 mutex_exit(&cvd
->vdev_dtl_lock
);
2037 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
2038 space_reftree_destroy(&reftree
);
2040 mutex_exit(&vd
->vdev_dtl_lock
);
2044 vdev_dtl_load(vdev_t
*vd
)
2046 spa_t
*spa
= vd
->vdev_spa
;
2047 objset_t
*mos
= spa
->spa_meta_objset
;
2050 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2051 ASSERT(!vd
->vdev_ishole
);
2053 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2054 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
2057 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2059 mutex_enter(&vd
->vdev_dtl_lock
);
2062 * Now that we've opened the space_map we need to update
2065 space_map_update(vd
->vdev_dtl_sm
);
2067 error
= space_map_load(vd
->vdev_dtl_sm
,
2068 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2069 mutex_exit(&vd
->vdev_dtl_lock
);
2074 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2075 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2084 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2086 spa_t
*spa
= vd
->vdev_spa
;
2088 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2089 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2094 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2096 spa_t
*spa
= vd
->vdev_spa
;
2097 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2098 DMU_OT_NONE
, 0, tx
);
2101 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2108 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2110 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2111 vd
->vdev_ops
!= &vdev_missing_ops
&&
2112 vd
->vdev_ops
!= &vdev_root_ops
&&
2113 !vd
->vdev_top
->vdev_removing
) {
2114 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2115 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2117 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2118 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2121 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2122 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2127 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2129 spa_t
*spa
= vd
->vdev_spa
;
2130 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2131 objset_t
*mos
= spa
->spa_meta_objset
;
2132 range_tree_t
*rtsync
;
2135 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2137 ASSERT(!vd
->vdev_ishole
);
2138 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2140 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2142 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2143 mutex_enter(&vd
->vdev_dtl_lock
);
2144 space_map_free(vd
->vdev_dtl_sm
, tx
);
2145 space_map_close(vd
->vdev_dtl_sm
);
2146 vd
->vdev_dtl_sm
= NULL
;
2147 mutex_exit(&vd
->vdev_dtl_lock
);
2150 * We only destroy the leaf ZAP for detached leaves or for
2151 * removed log devices. Removed data devices handle leaf ZAP
2152 * cleanup later, once cancellation is no longer possible.
2154 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2155 vd
->vdev_top
->vdev_islog
)) {
2156 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2157 vd
->vdev_leaf_zap
= 0;
2164 if (vd
->vdev_dtl_sm
== NULL
) {
2165 uint64_t new_object
;
2167 new_object
= space_map_alloc(mos
, tx
);
2168 VERIFY3U(new_object
, !=, 0);
2170 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2171 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2172 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2175 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2177 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2179 mutex_enter(&rtlock
);
2181 mutex_enter(&vd
->vdev_dtl_lock
);
2182 range_tree_walk(rt
, range_tree_add
, rtsync
);
2183 mutex_exit(&vd
->vdev_dtl_lock
);
2185 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2186 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2187 range_tree_vacate(rtsync
, NULL
, NULL
);
2189 range_tree_destroy(rtsync
);
2191 mutex_exit(&rtlock
);
2192 mutex_destroy(&rtlock
);
2195 * If the object for the space map has changed then dirty
2196 * the top level so that we update the config.
2198 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2199 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2200 "new object %llu", txg
, spa_name(spa
), object
,
2201 space_map_object(vd
->vdev_dtl_sm
));
2202 vdev_config_dirty(vd
->vdev_top
);
2207 mutex_enter(&vd
->vdev_dtl_lock
);
2208 space_map_update(vd
->vdev_dtl_sm
);
2209 mutex_exit(&vd
->vdev_dtl_lock
);
2213 * Determine whether the specified vdev can be offlined/detached/removed
2214 * without losing data.
2217 vdev_dtl_required(vdev_t
*vd
)
2219 spa_t
*spa
= vd
->vdev_spa
;
2220 vdev_t
*tvd
= vd
->vdev_top
;
2221 uint8_t cant_read
= vd
->vdev_cant_read
;
2224 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2226 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2230 * Temporarily mark the device as unreadable, and then determine
2231 * whether this results in any DTL outages in the top-level vdev.
2232 * If not, we can safely offline/detach/remove the device.
2234 vd
->vdev_cant_read
= B_TRUE
;
2235 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2236 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2237 vd
->vdev_cant_read
= cant_read
;
2238 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2240 if (!required
&& zio_injection_enabled
)
2241 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2247 * Determine if resilver is needed, and if so the txg range.
2250 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2252 boolean_t needed
= B_FALSE
;
2253 uint64_t thismin
= UINT64_MAX
;
2254 uint64_t thismax
= 0;
2256 if (vd
->vdev_children
== 0) {
2257 mutex_enter(&vd
->vdev_dtl_lock
);
2258 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2259 vdev_writeable(vd
)) {
2261 thismin
= vdev_dtl_min(vd
);
2262 thismax
= vdev_dtl_max(vd
);
2265 mutex_exit(&vd
->vdev_dtl_lock
);
2267 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2268 vdev_t
*cvd
= vd
->vdev_child
[c
];
2269 uint64_t cmin
, cmax
;
2271 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2272 thismin
= MIN(thismin
, cmin
);
2273 thismax
= MAX(thismax
, cmax
);
2279 if (needed
&& minp
) {
2287 vdev_load(vdev_t
*vd
)
2290 * Recursively load all children.
2292 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2293 vdev_load(vd
->vdev_child
[c
]);
2296 * If this is a top-level vdev, initialize its metaslabs.
2298 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2299 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2300 vdev_metaslab_init(vd
, 0) != 0))
2301 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2302 VDEV_AUX_CORRUPT_DATA
);
2304 * If this is a leaf vdev, load its DTL.
2306 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2307 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2308 VDEV_AUX_CORRUPT_DATA
);
2312 * The special vdev case is used for hot spares and l2cache devices. Its
2313 * sole purpose it to set the vdev state for the associated vdev. To do this,
2314 * we make sure that we can open the underlying device, then try to read the
2315 * label, and make sure that the label is sane and that it hasn't been
2316 * repurposed to another pool.
2319 vdev_validate_aux(vdev_t
*vd
)
2322 uint64_t guid
, version
;
2325 if (!vdev_readable(vd
))
2328 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2329 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2330 VDEV_AUX_CORRUPT_DATA
);
2334 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2335 !SPA_VERSION_IS_SUPPORTED(version
) ||
2336 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2337 guid
!= vd
->vdev_guid
||
2338 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2339 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2340 VDEV_AUX_CORRUPT_DATA
);
2346 * We don't actually check the pool state here. If it's in fact in
2347 * use by another pool, we update this fact on the fly when requested.
2354 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2356 spa_t
*spa
= vd
->vdev_spa
;
2357 objset_t
*mos
= spa
->spa_meta_objset
;
2360 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2361 ASSERT(vd
== vd
->vdev_top
);
2362 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2364 if (vd
->vdev_ms
!= NULL
) {
2365 metaslab_group_t
*mg
= vd
->vdev_mg
;
2367 metaslab_group_histogram_verify(mg
);
2368 metaslab_class_histogram_verify(mg
->mg_class
);
2370 for (int m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2371 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2373 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2376 mutex_enter(&msp
->ms_lock
);
2378 * If the metaslab was not loaded when the vdev
2379 * was removed then the histogram accounting may
2380 * not be accurate. Update the histogram information
2381 * here so that we ensure that the metaslab group
2382 * and metaslab class are up-to-date.
2384 metaslab_group_histogram_remove(mg
, msp
);
2386 VERIFY0(space_map_allocated(msp
->ms_sm
));
2387 space_map_free(msp
->ms_sm
, tx
);
2388 space_map_close(msp
->ms_sm
);
2390 mutex_exit(&msp
->ms_lock
);
2393 metaslab_group_histogram_verify(mg
);
2394 metaslab_class_histogram_verify(mg
->mg_class
);
2395 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2396 ASSERT0(mg
->mg_histogram
[i
]);
2400 if (vd
->vdev_ms_array
) {
2401 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2402 vd
->vdev_ms_array
= 0;
2405 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2406 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2407 vd
->vdev_top_zap
= 0;
2413 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2416 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2418 ASSERT(!vd
->vdev_ishole
);
2420 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2421 metaslab_sync_done(msp
, txg
);
2424 metaslab_sync_reassess(vd
->vdev_mg
);
2428 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2430 spa_t
*spa
= vd
->vdev_spa
;
2435 ASSERT(!vd
->vdev_ishole
);
2437 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2438 ASSERT(vd
== vd
->vdev_top
);
2439 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2440 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2441 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2442 ASSERT(vd
->vdev_ms_array
!= 0);
2443 vdev_config_dirty(vd
);
2448 * Remove the metadata associated with this vdev once it's empty.
2450 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2451 vdev_remove(vd
, txg
);
2453 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2454 metaslab_sync(msp
, txg
);
2455 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2458 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2459 vdev_dtl_sync(lvd
, txg
);
2461 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2465 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2467 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2471 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2472 * not be opened, and no I/O is attempted.
2475 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2479 spa_vdev_state_enter(spa
, SCL_NONE
);
2481 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2482 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2484 if (!vd
->vdev_ops
->vdev_op_leaf
)
2485 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2490 * If user did a 'zpool offline -f' then make the fault persist across
2493 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
2495 * There are two kinds of forced faults: temporary and
2496 * persistent. Temporary faults go away at pool import, while
2497 * persistent faults stay set. Both types of faults can be
2498 * cleared with a zpool clear.
2500 * We tell if a vdev is persistently faulted by looking at the
2501 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
2502 * import then it's a persistent fault. Otherwise, it's
2503 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
2504 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
2505 * tells vdev_config_generate() (which gets run later) to set
2506 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
2508 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
2509 vd
->vdev_tmpoffline
= B_FALSE
;
2510 aux
= VDEV_AUX_EXTERNAL
;
2512 vd
->vdev_tmpoffline
= B_TRUE
;
2516 * We don't directly use the aux state here, but if we do a
2517 * vdev_reopen(), we need this value to be present to remember why we
2520 vd
->vdev_label_aux
= aux
;
2523 * Faulted state takes precedence over degraded.
2525 vd
->vdev_delayed_close
= B_FALSE
;
2526 vd
->vdev_faulted
= 1ULL;
2527 vd
->vdev_degraded
= 0ULL;
2528 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2531 * If this device has the only valid copy of the data, then
2532 * back off and simply mark the vdev as degraded instead.
2534 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2535 vd
->vdev_degraded
= 1ULL;
2536 vd
->vdev_faulted
= 0ULL;
2539 * If we reopen the device and it's not dead, only then do we
2544 if (vdev_readable(vd
))
2545 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2548 return (spa_vdev_state_exit(spa
, vd
, 0));
2552 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2553 * user that something is wrong. The vdev continues to operate as normal as far
2554 * as I/O is concerned.
2557 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2561 spa_vdev_state_enter(spa
, SCL_NONE
);
2563 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2564 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2566 if (!vd
->vdev_ops
->vdev_op_leaf
)
2567 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2570 * If the vdev is already faulted, then don't do anything.
2572 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2573 return (spa_vdev_state_exit(spa
, NULL
, 0));
2575 vd
->vdev_degraded
= 1ULL;
2576 if (!vdev_is_dead(vd
))
2577 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2580 return (spa_vdev_state_exit(spa
, vd
, 0));
2584 * Online the given vdev.
2586 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2587 * spare device should be detached when the device finishes resilvering.
2588 * Second, the online should be treated like a 'test' online case, so no FMA
2589 * events are generated if the device fails to open.
2592 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2594 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2595 boolean_t wasoffline
;
2596 vdev_state_t oldstate
;
2598 spa_vdev_state_enter(spa
, SCL_NONE
);
2600 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2601 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2603 if (!vd
->vdev_ops
->vdev_op_leaf
)
2604 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2606 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
2607 oldstate
= vd
->vdev_state
;
2610 vd
->vdev_offline
= B_FALSE
;
2611 vd
->vdev_tmpoffline
= B_FALSE
;
2612 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2613 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2615 /* XXX - L2ARC 1.0 does not support expansion */
2616 if (!vd
->vdev_aux
) {
2617 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2618 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2622 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2624 if (!vd
->vdev_aux
) {
2625 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2626 pvd
->vdev_expanding
= B_FALSE
;
2630 *newstate
= vd
->vdev_state
;
2631 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2632 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2633 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2634 vd
->vdev_parent
->vdev_child
[0] == vd
)
2635 vd
->vdev_unspare
= B_TRUE
;
2637 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2639 /* XXX - L2ARC 1.0 does not support expansion */
2641 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2642 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2646 (oldstate
< VDEV_STATE_DEGRADED
&&
2647 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
2648 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
2650 return (spa_vdev_state_exit(spa
, vd
, 0));
2654 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2658 uint64_t generation
;
2659 metaslab_group_t
*mg
;
2662 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2664 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2665 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2667 if (!vd
->vdev_ops
->vdev_op_leaf
)
2668 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2672 generation
= spa
->spa_config_generation
+ 1;
2675 * If the device isn't already offline, try to offline it.
2677 if (!vd
->vdev_offline
) {
2679 * If this device has the only valid copy of some data,
2680 * don't allow it to be offlined. Log devices are always
2683 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2684 vdev_dtl_required(vd
))
2685 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2688 * If the top-level is a slog and it has had allocations
2689 * then proceed. We check that the vdev's metaslab group
2690 * is not NULL since it's possible that we may have just
2691 * added this vdev but not yet initialized its metaslabs.
2693 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2695 * Prevent any future allocations.
2697 metaslab_group_passivate(mg
);
2698 (void) spa_vdev_state_exit(spa
, vd
, 0);
2700 error
= spa_offline_log(spa
);
2702 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2705 * Check to see if the config has changed.
2707 if (error
|| generation
!= spa
->spa_config_generation
) {
2708 metaslab_group_activate(mg
);
2710 return (spa_vdev_state_exit(spa
,
2712 (void) spa_vdev_state_exit(spa
, vd
, 0);
2715 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2719 * Offline this device and reopen its top-level vdev.
2720 * If the top-level vdev is a log device then just offline
2721 * it. Otherwise, if this action results in the top-level
2722 * vdev becoming unusable, undo it and fail the request.
2724 vd
->vdev_offline
= B_TRUE
;
2727 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2728 vdev_is_dead(tvd
)) {
2729 vd
->vdev_offline
= B_FALSE
;
2731 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2735 * Add the device back into the metaslab rotor so that
2736 * once we online the device it's open for business.
2738 if (tvd
->vdev_islog
&& mg
!= NULL
)
2739 metaslab_group_activate(mg
);
2742 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2744 return (spa_vdev_state_exit(spa
, vd
, 0));
2748 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2752 mutex_enter(&spa
->spa_vdev_top_lock
);
2753 error
= vdev_offline_locked(spa
, guid
, flags
);
2754 mutex_exit(&spa
->spa_vdev_top_lock
);
2760 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2761 * vdev_offline(), we assume the spa config is locked. We also clear all
2762 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2765 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2767 vdev_t
*rvd
= spa
->spa_root_vdev
;
2769 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2774 vd
->vdev_stat
.vs_read_errors
= 0;
2775 vd
->vdev_stat
.vs_write_errors
= 0;
2776 vd
->vdev_stat
.vs_checksum_errors
= 0;
2778 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2779 vdev_clear(spa
, vd
->vdev_child
[c
]);
2782 * If we're in the FAULTED state or have experienced failed I/O, then
2783 * clear the persistent state and attempt to reopen the device. We
2784 * also mark the vdev config dirty, so that the new faulted state is
2785 * written out to disk.
2787 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2788 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2790 * When reopening in response to a clear event, it may be due to
2791 * a fmadm repair request. In this case, if the device is
2792 * still broken, we want to still post the ereport again.
2794 vd
->vdev_forcefault
= B_TRUE
;
2796 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2797 vd
->vdev_cant_read
= B_FALSE
;
2798 vd
->vdev_cant_write
= B_FALSE
;
2799 vd
->vdev_stat
.vs_aux
= 0;
2801 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2803 vd
->vdev_forcefault
= B_FALSE
;
2805 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2806 vdev_state_dirty(vd
->vdev_top
);
2808 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2809 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2811 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
2815 * When clearing a FMA-diagnosed fault, we always want to
2816 * unspare the device, as we assume that the original spare was
2817 * done in response to the FMA fault.
2819 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2820 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2821 vd
->vdev_parent
->vdev_child
[0] == vd
)
2822 vd
->vdev_unspare
= B_TRUE
;
2826 vdev_is_dead(vdev_t
*vd
)
2829 * Holes and missing devices are always considered "dead".
2830 * This simplifies the code since we don't have to check for
2831 * these types of devices in the various code paths.
2832 * Instead we rely on the fact that we skip over dead devices
2833 * before issuing I/O to them.
2835 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2836 vd
->vdev_ops
== &vdev_missing_ops
);
2840 vdev_readable(vdev_t
*vd
)
2842 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2846 vdev_writeable(vdev_t
*vd
)
2848 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2852 vdev_allocatable(vdev_t
*vd
)
2854 uint64_t state
= vd
->vdev_state
;
2857 * We currently allow allocations from vdevs which may be in the
2858 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2859 * fails to reopen then we'll catch it later when we're holding
2860 * the proper locks. Note that we have to get the vdev state
2861 * in a local variable because although it changes atomically,
2862 * we're asking two separate questions about it.
2864 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2865 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
&&
2866 vd
->vdev_mg
->mg_initialized
);
2870 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2872 ASSERT(zio
->io_vd
== vd
);
2874 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2877 if (zio
->io_type
== ZIO_TYPE_READ
)
2878 return (!vd
->vdev_cant_read
);
2880 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2881 return (!vd
->vdev_cant_write
);
2887 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2890 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2891 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2892 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2895 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2899 * Get extended stats
2902 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2905 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2906 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2907 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2909 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2910 vsx
->vsx_total_histo
[t
][b
] +=
2911 cvsx
->vsx_total_histo
[t
][b
];
2915 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2916 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2917 vsx
->vsx_queue_histo
[t
][b
] +=
2918 cvsx
->vsx_queue_histo
[t
][b
];
2920 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2921 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2923 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2924 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2926 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2927 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2933 * Get statistics for the given vdev.
2936 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2940 * If we're getting stats on the root vdev, aggregate the I/O counts
2941 * over all top-level vdevs (i.e. the direct children of the root).
2943 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2945 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2946 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2949 memset(vsx
, 0, sizeof (*vsx
));
2951 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2952 vdev_t
*cvd
= vd
->vdev_child
[c
];
2953 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2954 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2956 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2958 vdev_get_child_stat(cvd
, vs
, cvs
);
2960 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2965 * We're a leaf. Just copy our ZIO active queue stats in. The
2966 * other leaf stats are updated in vdev_stat_update().
2971 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2973 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2974 vsx
->vsx_active_queue
[t
] =
2975 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2976 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2977 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2983 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2985 vdev_t
*tvd
= vd
->vdev_top
;
2986 mutex_enter(&vd
->vdev_stat_lock
);
2988 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2989 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2990 vs
->vs_state
= vd
->vdev_state
;
2991 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2992 if (vd
->vdev_ops
->vdev_op_leaf
)
2993 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
2994 VDEV_LABEL_END_SIZE
;
2996 * Report expandable space on top-level, non-auxillary devices
2997 * only. The expandable space is reported in terms of metaslab
2998 * sized units since that determines how much space the pool
3001 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
3002 vs
->vs_esize
= P2ALIGN(
3003 vd
->vdev_max_asize
- vd
->vdev_asize
,
3004 1ULL << tvd
->vdev_ms_shift
);
3006 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
3007 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
3009 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
3013 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
3014 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
3015 mutex_exit(&vd
->vdev_stat_lock
);
3019 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
3021 return (vdev_get_stats_ex(vd
, vs
, NULL
));
3025 vdev_clear_stats(vdev_t
*vd
)
3027 mutex_enter(&vd
->vdev_stat_lock
);
3028 vd
->vdev_stat
.vs_space
= 0;
3029 vd
->vdev_stat
.vs_dspace
= 0;
3030 vd
->vdev_stat
.vs_alloc
= 0;
3031 mutex_exit(&vd
->vdev_stat_lock
);
3035 vdev_scan_stat_init(vdev_t
*vd
)
3037 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3039 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3040 vdev_scan_stat_init(vd
->vdev_child
[c
]);
3042 mutex_enter(&vd
->vdev_stat_lock
);
3043 vs
->vs_scan_processed
= 0;
3044 mutex_exit(&vd
->vdev_stat_lock
);
3048 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
3050 spa_t
*spa
= zio
->io_spa
;
3051 vdev_t
*rvd
= spa
->spa_root_vdev
;
3052 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
3054 uint64_t txg
= zio
->io_txg
;
3055 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3056 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
3057 zio_type_t type
= zio
->io_type
;
3058 int flags
= zio
->io_flags
;
3061 * If this i/o is a gang leader, it didn't do any actual work.
3063 if (zio
->io_gang_tree
)
3066 if (zio
->io_error
== 0) {
3068 * If this is a root i/o, don't count it -- we've already
3069 * counted the top-level vdevs, and vdev_get_stats() will
3070 * aggregate them when asked. This reduces contention on
3071 * the root vdev_stat_lock and implicitly handles blocks
3072 * that compress away to holes, for which there is no i/o.
3073 * (Holes never create vdev children, so all the counters
3074 * remain zero, which is what we want.)
3076 * Note: this only applies to successful i/o (io_error == 0)
3077 * because unlike i/o counts, errors are not additive.
3078 * When reading a ditto block, for example, failure of
3079 * one top-level vdev does not imply a root-level error.
3084 ASSERT(vd
== zio
->io_vd
);
3086 if (flags
& ZIO_FLAG_IO_BYPASS
)
3089 mutex_enter(&vd
->vdev_stat_lock
);
3091 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3092 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3093 dsl_scan_phys_t
*scn_phys
=
3094 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3095 uint64_t *processed
= &scn_phys
->scn_processed
;
3098 if (vd
->vdev_ops
->vdev_op_leaf
)
3099 atomic_add_64(processed
, psize
);
3100 vs
->vs_scan_processed
+= psize
;
3103 if (flags
& ZIO_FLAG_SELF_HEAL
)
3104 vs
->vs_self_healed
+= psize
;
3108 * The bytes/ops/histograms are recorded at the leaf level and
3109 * aggregated into the higher level vdevs in vdev_get_stats().
3111 if (vd
->vdev_ops
->vdev_op_leaf
&&
3112 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3115 vs
->vs_bytes
[type
] += psize
;
3117 if (flags
& ZIO_FLAG_DELEGATED
) {
3118 vsx
->vsx_agg_histo
[zio
->io_priority
]
3119 [RQ_HISTO(zio
->io_size
)]++;
3121 vsx
->vsx_ind_histo
[zio
->io_priority
]
3122 [RQ_HISTO(zio
->io_size
)]++;
3125 if (zio
->io_delta
&& zio
->io_delay
) {
3126 vsx
->vsx_queue_histo
[zio
->io_priority
]
3127 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3128 vsx
->vsx_disk_histo
[type
]
3129 [L_HISTO(zio
->io_delay
)]++;
3130 vsx
->vsx_total_histo
[type
]
3131 [L_HISTO(zio
->io_delta
)]++;
3135 mutex_exit(&vd
->vdev_stat_lock
);
3139 if (flags
& ZIO_FLAG_SPECULATIVE
)
3143 * If this is an I/O error that is going to be retried, then ignore the
3144 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3145 * hard errors, when in reality they can happen for any number of
3146 * innocuous reasons (bus resets, MPxIO link failure, etc).
3148 if (zio
->io_error
== EIO
&&
3149 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3153 * Intent logs writes won't propagate their error to the root
3154 * I/O so don't mark these types of failures as pool-level
3157 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3160 mutex_enter(&vd
->vdev_stat_lock
);
3161 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3162 if (zio
->io_error
== ECKSUM
)
3163 vs
->vs_checksum_errors
++;
3165 vs
->vs_read_errors
++;
3167 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3168 vs
->vs_write_errors
++;
3169 mutex_exit(&vd
->vdev_stat_lock
);
3171 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3172 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3173 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3174 spa
->spa_claiming
)) {
3176 * This is either a normal write (not a repair), or it's
3177 * a repair induced by the scrub thread, or it's a repair
3178 * made by zil_claim() during spa_load() in the first txg.
3179 * In the normal case, we commit the DTL change in the same
3180 * txg as the block was born. In the scrub-induced repair
3181 * case, we know that scrubs run in first-pass syncing context,
3182 * so we commit the DTL change in spa_syncing_txg(spa).
3183 * In the zil_claim() case, we commit in spa_first_txg(spa).
3185 * We currently do not make DTL entries for failed spontaneous
3186 * self-healing writes triggered by normal (non-scrubbing)
3187 * reads, because we have no transactional context in which to
3188 * do so -- and it's not clear that it'd be desirable anyway.
3190 if (vd
->vdev_ops
->vdev_op_leaf
) {
3191 uint64_t commit_txg
= txg
;
3192 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3193 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3194 ASSERT(spa_sync_pass(spa
) == 1);
3195 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3196 commit_txg
= spa_syncing_txg(spa
);
3197 } else if (spa
->spa_claiming
) {
3198 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3199 commit_txg
= spa_first_txg(spa
);
3201 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3202 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3204 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3205 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3206 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3209 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3214 * Update the in-core space usage stats for this vdev, its metaslab class,
3215 * and the root vdev.
3218 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3219 int64_t space_delta
)
3221 int64_t dspace_delta
= space_delta
;
3222 spa_t
*spa
= vd
->vdev_spa
;
3223 vdev_t
*rvd
= spa
->spa_root_vdev
;
3224 metaslab_group_t
*mg
= vd
->vdev_mg
;
3225 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3227 ASSERT(vd
== vd
->vdev_top
);
3230 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3231 * factor. We must calculate this here and not at the root vdev
3232 * because the root vdev's psize-to-asize is simply the max of its
3233 * childrens', thus not accurate enough for us.
3235 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3236 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3237 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3238 vd
->vdev_deflate_ratio
;
3240 mutex_enter(&vd
->vdev_stat_lock
);
3241 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3242 vd
->vdev_stat
.vs_space
+= space_delta
;
3243 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3244 mutex_exit(&vd
->vdev_stat_lock
);
3246 if (mc
== spa_normal_class(spa
)) {
3247 mutex_enter(&rvd
->vdev_stat_lock
);
3248 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3249 rvd
->vdev_stat
.vs_space
+= space_delta
;
3250 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3251 mutex_exit(&rvd
->vdev_stat_lock
);
3255 ASSERT(rvd
== vd
->vdev_parent
);
3256 ASSERT(vd
->vdev_ms_count
!= 0);
3258 metaslab_class_space_update(mc
,
3259 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3264 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3265 * so that it will be written out next time the vdev configuration is synced.
3266 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3269 vdev_config_dirty(vdev_t
*vd
)
3271 spa_t
*spa
= vd
->vdev_spa
;
3272 vdev_t
*rvd
= spa
->spa_root_vdev
;
3275 ASSERT(spa_writeable(spa
));
3278 * If this is an aux vdev (as with l2cache and spare devices), then we
3279 * update the vdev config manually and set the sync flag.
3281 if (vd
->vdev_aux
!= NULL
) {
3282 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3286 for (c
= 0; c
< sav
->sav_count
; c
++) {
3287 if (sav
->sav_vdevs
[c
] == vd
)
3291 if (c
== sav
->sav_count
) {
3293 * We're being removed. There's nothing more to do.
3295 ASSERT(sav
->sav_sync
== B_TRUE
);
3299 sav
->sav_sync
= B_TRUE
;
3301 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3302 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3303 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3304 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3310 * Setting the nvlist in the middle if the array is a little
3311 * sketchy, but it will work.
3313 nvlist_free(aux
[c
]);
3314 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3320 * The dirty list is protected by the SCL_CONFIG lock. The caller
3321 * must either hold SCL_CONFIG as writer, or must be the sync thread
3322 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3323 * so this is sufficient to ensure mutual exclusion.
3325 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3326 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3327 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3330 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3331 vdev_config_dirty(rvd
->vdev_child
[c
]);
3333 ASSERT(vd
== vd
->vdev_top
);
3335 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3337 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3342 vdev_config_clean(vdev_t
*vd
)
3344 spa_t
*spa
= vd
->vdev_spa
;
3346 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3347 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3348 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3350 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3351 list_remove(&spa
->spa_config_dirty_list
, vd
);
3355 * Mark a top-level vdev's state as dirty, so that the next pass of
3356 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3357 * the state changes from larger config changes because they require
3358 * much less locking, and are often needed for administrative actions.
3361 vdev_state_dirty(vdev_t
*vd
)
3363 spa_t
*spa
= vd
->vdev_spa
;
3365 ASSERT(spa_writeable(spa
));
3366 ASSERT(vd
== vd
->vdev_top
);
3369 * The state list is protected by the SCL_STATE lock. The caller
3370 * must either hold SCL_STATE as writer, or must be the sync thread
3371 * (which holds SCL_STATE as reader). There's only one sync thread,
3372 * so this is sufficient to ensure mutual exclusion.
3374 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3375 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3376 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3378 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3379 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3383 vdev_state_clean(vdev_t
*vd
)
3385 spa_t
*spa
= vd
->vdev_spa
;
3387 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3388 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3389 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3391 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3392 list_remove(&spa
->spa_state_dirty_list
, vd
);
3396 * Propagate vdev state up from children to parent.
3399 vdev_propagate_state(vdev_t
*vd
)
3401 spa_t
*spa
= vd
->vdev_spa
;
3402 vdev_t
*rvd
= spa
->spa_root_vdev
;
3403 int degraded
= 0, faulted
= 0;
3407 if (vd
->vdev_children
> 0) {
3408 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3409 child
= vd
->vdev_child
[c
];
3412 * Don't factor holes into the decision.
3414 if (child
->vdev_ishole
)
3417 if (!vdev_readable(child
) ||
3418 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3420 * Root special: if there is a top-level log
3421 * device, treat the root vdev as if it were
3424 if (child
->vdev_islog
&& vd
== rvd
)
3428 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3432 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3436 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3439 * Root special: if there is a top-level vdev that cannot be
3440 * opened due to corrupted metadata, then propagate the root
3441 * vdev's aux state as 'corrupt' rather than 'insufficient
3444 if (corrupted
&& vd
== rvd
&&
3445 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3446 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3447 VDEV_AUX_CORRUPT_DATA
);
3450 if (vd
->vdev_parent
)
3451 vdev_propagate_state(vd
->vdev_parent
);
3455 * Set a vdev's state. If this is during an open, we don't update the parent
3456 * state, because we're in the process of opening children depth-first.
3457 * Otherwise, we propagate the change to the parent.
3459 * If this routine places a device in a faulted state, an appropriate ereport is
3463 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3465 uint64_t save_state
;
3466 spa_t
*spa
= vd
->vdev_spa
;
3468 if (state
== vd
->vdev_state
) {
3470 * Since vdev_offline() code path is already in an offline
3471 * state we can miss a statechange event to OFFLINE. Check
3472 * the previous state to catch this condition.
3474 if (vd
->vdev_ops
->vdev_op_leaf
&&
3475 (state
== VDEV_STATE_OFFLINE
) &&
3476 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
3477 /* post an offline state change */
3478 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
3480 vd
->vdev_stat
.vs_aux
= aux
;
3484 save_state
= vd
->vdev_state
;
3486 vd
->vdev_state
= state
;
3487 vd
->vdev_stat
.vs_aux
= aux
;
3490 * If we are setting the vdev state to anything but an open state, then
3491 * always close the underlying device unless the device has requested
3492 * a delayed close (i.e. we're about to remove or fault the device).
3493 * Otherwise, we keep accessible but invalid devices open forever.
3494 * We don't call vdev_close() itself, because that implies some extra
3495 * checks (offline, etc) that we don't want here. This is limited to
3496 * leaf devices, because otherwise closing the device will affect other
3499 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3500 vd
->vdev_ops
->vdev_op_leaf
)
3501 vd
->vdev_ops
->vdev_op_close(vd
);
3503 if (vd
->vdev_removed
&&
3504 state
== VDEV_STATE_CANT_OPEN
&&
3505 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3507 * If the previous state is set to VDEV_STATE_REMOVED, then this
3508 * device was previously marked removed and someone attempted to
3509 * reopen it. If this failed due to a nonexistent device, then
3510 * keep the device in the REMOVED state. We also let this be if
3511 * it is one of our special test online cases, which is only
3512 * attempting to online the device and shouldn't generate an FMA
3515 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3516 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3517 } else if (state
== VDEV_STATE_REMOVED
) {
3518 vd
->vdev_removed
= B_TRUE
;
3519 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3521 * If we fail to open a vdev during an import or recovery, we
3522 * mark it as "not available", which signifies that it was
3523 * never there to begin with. Failure to open such a device
3524 * is not considered an error.
3526 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3527 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3528 vd
->vdev_ops
->vdev_op_leaf
)
3529 vd
->vdev_not_present
= 1;
3532 * Post the appropriate ereport. If the 'prevstate' field is
3533 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3534 * that this is part of a vdev_reopen(). In this case, we don't
3535 * want to post the ereport if the device was already in the
3536 * CANT_OPEN state beforehand.
3538 * If the 'checkremove' flag is set, then this is an attempt to
3539 * online the device in response to an insertion event. If we
3540 * hit this case, then we have detected an insertion event for a
3541 * faulted or offline device that wasn't in the removed state.
3542 * In this scenario, we don't post an ereport because we are
3543 * about to replace the device, or attempt an online with
3544 * vdev_forcefault, which will generate the fault for us.
3546 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3547 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3548 vd
!= spa
->spa_root_vdev
) {
3552 case VDEV_AUX_OPEN_FAILED
:
3553 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3555 case VDEV_AUX_CORRUPT_DATA
:
3556 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3558 case VDEV_AUX_NO_REPLICAS
:
3559 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3561 case VDEV_AUX_BAD_GUID_SUM
:
3562 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3564 case VDEV_AUX_TOO_SMALL
:
3565 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3567 case VDEV_AUX_BAD_LABEL
:
3568 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3570 case VDEV_AUX_BAD_ASHIFT
:
3571 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
3574 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3577 zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
3581 /* Erase any notion of persistent removed state */
3582 vd
->vdev_removed
= B_FALSE
;
3584 vd
->vdev_removed
= B_FALSE
;
3588 * Notify ZED of any significant state-change on a leaf vdev.
3591 if (vd
->vdev_ops
->vdev_op_leaf
) {
3592 /* preserve original state from a vdev_reopen() */
3593 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
3594 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
3595 (save_state
<= VDEV_STATE_CLOSED
))
3596 save_state
= vd
->vdev_prevstate
;
3598 /* filter out state change due to initial vdev_open */
3599 if (save_state
> VDEV_STATE_CLOSED
)
3600 zfs_post_state_change(spa
, vd
, save_state
);
3603 if (!isopen
&& vd
->vdev_parent
)
3604 vdev_propagate_state(vd
->vdev_parent
);
3608 * Check the vdev configuration to ensure that it's capable of supporting
3609 * a root pool. We do not support partial configuration.
3612 vdev_is_bootable(vdev_t
*vd
)
3614 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3615 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3617 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0)
3621 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3622 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3629 * Load the state from the original vdev tree (ovd) which
3630 * we've retrieved from the MOS config object. If the original
3631 * vdev was offline or faulted then we transfer that state to the
3632 * device in the current vdev tree (nvd).
3635 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3637 ASSERT(nvd
->vdev_top
->vdev_islog
);
3638 ASSERT(spa_config_held(nvd
->vdev_spa
,
3639 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3640 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3642 for (int c
= 0; c
< nvd
->vdev_children
; c
++)
3643 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3645 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3647 * Restore the persistent vdev state
3649 nvd
->vdev_offline
= ovd
->vdev_offline
;
3650 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3651 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3652 nvd
->vdev_removed
= ovd
->vdev_removed
;
3657 * Determine if a log device has valid content. If the vdev was
3658 * removed or faulted in the MOS config then we know that
3659 * the content on the log device has already been written to the pool.
3662 vdev_log_state_valid(vdev_t
*vd
)
3664 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3668 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3669 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3676 * Expand a vdev if possible.
3679 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3681 ASSERT(vd
->vdev_top
== vd
);
3682 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3684 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3685 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3686 vdev_config_dirty(vd
);
3694 vdev_split(vdev_t
*vd
)
3696 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3698 vdev_remove_child(pvd
, vd
);
3699 vdev_compact_children(pvd
);
3701 cvd
= pvd
->vdev_child
[0];
3702 if (pvd
->vdev_children
== 1) {
3703 vdev_remove_parent(cvd
);
3704 cvd
->vdev_splitting
= B_TRUE
;
3706 vdev_propagate_state(cvd
);
3710 vdev_deadman(vdev_t
*vd
)
3712 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3713 vdev_t
*cvd
= vd
->vdev_child
[c
];
3718 if (vd
->vdev_ops
->vdev_op_leaf
) {
3719 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3721 mutex_enter(&vq
->vq_lock
);
3722 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3723 spa_t
*spa
= vd
->vdev_spa
;
3728 * Look at the head of all the pending queues,
3729 * if any I/O has been outstanding for longer than
3730 * the spa_deadman_synctime we log a zevent.
3732 fio
= avl_first(&vq
->vq_active_tree
);
3733 delta
= gethrtime() - fio
->io_timestamp
;
3734 if (delta
> spa_deadman_synctime(spa
)) {
3735 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3736 "delta %lluns, last io %lluns",
3737 fio
->io_timestamp
, delta
,
3738 vq
->vq_io_complete_ts
);
3739 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3740 spa
, vd
, &fio
->io_bookmark
, fio
, 0, 0);
3743 mutex_exit(&vq
->vq_lock
);
3747 #if defined(_KERNEL) && defined(HAVE_SPL)
3748 EXPORT_SYMBOL(vdev_fault
);
3749 EXPORT_SYMBOL(vdev_degrade
);
3750 EXPORT_SYMBOL(vdev_online
);
3751 EXPORT_SYMBOL(vdev_offline
);
3752 EXPORT_SYMBOL(vdev_clear
);
3754 module_param(metaslabs_per_vdev
, int, 0644);
3755 MODULE_PARM_DESC(metaslabs_per_vdev
,
3756 "Divide added vdev into approximately (but no more than) this number "