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>
30 #include <sys/zfs_context.h>
31 #include <sys/fm/fs/zfs.h>
33 #include <sys/spa_impl.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/vdev_impl.h>
37 #include <sys/uberblock_impl.h>
38 #include <sys/metaslab.h>
39 #include <sys/metaslab_impl.h>
40 #include <sys/space_map.h>
41 #include <sys/space_reftree.h>
44 #include <sys/fs/zfs.h>
47 #include <sys/dsl_scan.h>
50 #include <sys/zfs_ratelimit.h>
53 * When a vdev is added, it will be divided into approximately (but no
54 * more than) this number of metaslabs.
56 int metaslabs_per_vdev
= 200;
59 * Virtual device management.
62 static vdev_ops_t
*vdev_ops_table
[] = {
76 * Given a vdev type, return the appropriate ops vector.
79 vdev_getops(const char *type
)
81 vdev_ops_t
*ops
, **opspp
;
83 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
84 if (strcmp(ops
->vdev_op_type
, type
) == 0)
91 * Default asize function: return the MAX of psize with the asize of
92 * all children. This is what's used by anything other than RAID-Z.
95 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
97 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
101 for (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
)
149 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
151 for (c
= 0; c
< vd
->vdev_children
; c
++)
152 vdev_set_min_asize(vd
->vdev_child
[c
]);
156 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
158 vdev_t
*rvd
= spa
->spa_root_vdev
;
160 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
162 if (vdev
< rvd
->vdev_children
) {
163 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
164 return (rvd
->vdev_child
[vdev
]);
171 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
176 if (vd
->vdev_guid
== guid
)
179 for (c
= 0; c
< vd
->vdev_children
; c
++)
180 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
188 vdev_count_leaves_impl(vdev_t
*vd
)
193 if (vd
->vdev_ops
->vdev_op_leaf
)
196 for (c
= 0; c
< vd
->vdev_children
; c
++)
197 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
203 vdev_count_leaves(spa_t
*spa
)
205 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
209 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
211 size_t oldsize
, newsize
;
212 uint64_t id
= cvd
->vdev_id
;
215 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
216 ASSERT(cvd
->vdev_parent
== NULL
);
218 cvd
->vdev_parent
= pvd
;
223 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
225 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
226 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
227 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
229 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
230 if (pvd
->vdev_child
!= NULL
) {
231 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
232 kmem_free(pvd
->vdev_child
, oldsize
);
235 pvd
->vdev_child
= newchild
;
236 pvd
->vdev_child
[id
] = cvd
;
238 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
239 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
242 * Walk up all ancestors to update guid sum.
244 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
245 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
249 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
252 uint_t id
= cvd
->vdev_id
;
254 ASSERT(cvd
->vdev_parent
== pvd
);
259 ASSERT(id
< pvd
->vdev_children
);
260 ASSERT(pvd
->vdev_child
[id
] == cvd
);
262 pvd
->vdev_child
[id
] = NULL
;
263 cvd
->vdev_parent
= NULL
;
265 for (c
= 0; c
< pvd
->vdev_children
; c
++)
266 if (pvd
->vdev_child
[c
])
269 if (c
== pvd
->vdev_children
) {
270 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
271 pvd
->vdev_child
= NULL
;
272 pvd
->vdev_children
= 0;
276 * Walk up all ancestors to update guid sum.
278 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
279 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
283 * Remove any holes in the child array.
286 vdev_compact_children(vdev_t
*pvd
)
288 vdev_t
**newchild
, *cvd
;
289 int oldc
= pvd
->vdev_children
;
293 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
295 for (c
= newc
= 0; c
< oldc
; c
++)
296 if (pvd
->vdev_child
[c
])
299 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
301 for (c
= newc
= 0; c
< oldc
; c
++) {
302 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
303 newchild
[newc
] = cvd
;
304 cvd
->vdev_id
= newc
++;
308 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
309 pvd
->vdev_child
= newchild
;
310 pvd
->vdev_children
= newc
;
314 * Allocate and minimally initialize a vdev_t.
317 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
322 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
324 if (spa
->spa_root_vdev
== NULL
) {
325 ASSERT(ops
== &vdev_root_ops
);
326 spa
->spa_root_vdev
= vd
;
327 spa
->spa_load_guid
= spa_generate_guid(NULL
);
330 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
331 if (spa
->spa_root_vdev
== vd
) {
333 * The root vdev's guid will also be the pool guid,
334 * which must be unique among all pools.
336 guid
= spa_generate_guid(NULL
);
339 * Any other vdev's guid must be unique within the pool.
341 guid
= spa_generate_guid(spa
);
343 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
348 vd
->vdev_guid
= guid
;
349 vd
->vdev_guid_sum
= guid
;
351 vd
->vdev_state
= VDEV_STATE_CLOSED
;
352 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
355 * Initialize rate limit structs for events. We rate limit ZIO delay
356 * and checksum events so that we don't overwhelm ZED with thousands
357 * of events when a disk is acting up.
359 zfs_ratelimit_init(&vd
->vdev_delay_rl
, DELAYS_PER_SECOND
, 1);
360 zfs_ratelimit_init(&vd
->vdev_checksum_rl
, CHECKSUMS_PER_SECOND
, 1);
362 list_link_init(&vd
->vdev_config_dirty_node
);
363 list_link_init(&vd
->vdev_state_dirty_node
);
364 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
365 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
366 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
367 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
369 for (t
= 0; t
< DTL_TYPES
; t
++) {
370 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
373 txg_list_create(&vd
->vdev_ms_list
,
374 offsetof(struct metaslab
, ms_txg_node
));
375 txg_list_create(&vd
->vdev_dtl_list
,
376 offsetof(struct vdev
, vdev_dtl_node
));
377 vd
->vdev_stat
.vs_timestamp
= gethrtime();
385 * Allocate a new vdev. The 'alloctype' is used to control whether we are
386 * creating a new vdev or loading an existing one - the behavior is slightly
387 * different for each case.
390 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
395 uint64_t guid
= 0, islog
, nparity
;
398 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
400 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
401 return (SET_ERROR(EINVAL
));
403 if ((ops
= vdev_getops(type
)) == NULL
)
404 return (SET_ERROR(EINVAL
));
407 * If this is a load, get the vdev guid from the nvlist.
408 * Otherwise, vdev_alloc_common() will generate one for us.
410 if (alloctype
== VDEV_ALLOC_LOAD
) {
413 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
415 return (SET_ERROR(EINVAL
));
417 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
418 return (SET_ERROR(EINVAL
));
419 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
420 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
421 return (SET_ERROR(EINVAL
));
422 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
423 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
424 return (SET_ERROR(EINVAL
));
425 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
426 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
427 return (SET_ERROR(EINVAL
));
431 * The first allocated vdev must be of type 'root'.
433 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
434 return (SET_ERROR(EINVAL
));
437 * Determine whether we're a log vdev.
440 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
441 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
442 return (SET_ERROR(ENOTSUP
));
444 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
445 return (SET_ERROR(ENOTSUP
));
448 * Set the nparity property for RAID-Z vdevs.
451 if (ops
== &vdev_raidz_ops
) {
452 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
454 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
455 return (SET_ERROR(EINVAL
));
457 * Previous versions could only support 1 or 2 parity
461 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
462 return (SET_ERROR(ENOTSUP
));
464 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
465 return (SET_ERROR(ENOTSUP
));
468 * We require the parity to be specified for SPAs that
469 * support multiple parity levels.
471 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
472 return (SET_ERROR(EINVAL
));
474 * Otherwise, we default to 1 parity device for RAID-Z.
481 ASSERT(nparity
!= -1ULL);
483 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
485 vd
->vdev_islog
= islog
;
486 vd
->vdev_nparity
= nparity
;
488 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
489 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
490 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
491 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
492 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
493 &vd
->vdev_physpath
) == 0)
494 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
496 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
497 &vd
->vdev_enc_sysfs_path
) == 0)
498 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
500 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
501 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
504 * Set the whole_disk property. If it's not specified, leave the value
507 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
508 &vd
->vdev_wholedisk
) != 0)
509 vd
->vdev_wholedisk
= -1ULL;
512 * Look for the 'not present' flag. This will only be set if the device
513 * was not present at the time of import.
515 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
516 &vd
->vdev_not_present
);
519 * Get the alignment requirement.
521 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
524 * Retrieve the vdev creation time.
526 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
530 * If we're a top-level vdev, try to load the allocation parameters.
532 if (parent
&& !parent
->vdev_parent
&&
533 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
534 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
536 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
538 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
540 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
542 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
545 ASSERT0(vd
->vdev_top_zap
);
548 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
549 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
550 alloctype
== VDEV_ALLOC_ADD
||
551 alloctype
== VDEV_ALLOC_SPLIT
||
552 alloctype
== VDEV_ALLOC_ROOTPOOL
);
553 vd
->vdev_mg
= metaslab_group_create(islog
?
554 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
557 if (vd
->vdev_ops
->vdev_op_leaf
&&
558 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
559 (void) nvlist_lookup_uint64(nv
,
560 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
562 ASSERT0(vd
->vdev_leaf_zap
);
566 * If we're a leaf vdev, try to load the DTL object and other state.
569 if (vd
->vdev_ops
->vdev_op_leaf
&&
570 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
571 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
572 if (alloctype
== VDEV_ALLOC_LOAD
) {
573 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
574 &vd
->vdev_dtl_object
);
575 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
579 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
582 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
583 &spare
) == 0 && spare
)
587 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
590 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
591 &vd
->vdev_resilver_txg
);
594 * When importing a pool, we want to ignore the persistent fault
595 * state, as the diagnosis made on another system may not be
596 * valid in the current context. Local vdevs will
597 * remain in the faulted state.
599 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
600 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
602 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
604 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
607 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
611 VDEV_AUX_ERR_EXCEEDED
;
612 if (nvlist_lookup_string(nv
,
613 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
614 strcmp(aux
, "external") == 0)
615 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
621 * Add ourselves to the parent's list of children.
623 vdev_add_child(parent
, vd
);
631 vdev_free(vdev_t
*vd
)
634 spa_t
*spa
= vd
->vdev_spa
;
637 * vdev_free() implies closing the vdev first. This is simpler than
638 * trying to ensure complicated semantics for all callers.
642 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
643 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
648 for (c
= 0; c
< vd
->vdev_children
; c
++)
649 vdev_free(vd
->vdev_child
[c
]);
651 ASSERT(vd
->vdev_child
== NULL
);
652 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
655 * Discard allocation state.
657 if (vd
->vdev_mg
!= NULL
) {
658 vdev_metaslab_fini(vd
);
659 metaslab_group_destroy(vd
->vdev_mg
);
662 ASSERT0(vd
->vdev_stat
.vs_space
);
663 ASSERT0(vd
->vdev_stat
.vs_dspace
);
664 ASSERT0(vd
->vdev_stat
.vs_alloc
);
667 * Remove this vdev from its parent's child list.
669 vdev_remove_child(vd
->vdev_parent
, vd
);
671 ASSERT(vd
->vdev_parent
== NULL
);
674 * Clean up vdev structure.
680 spa_strfree(vd
->vdev_path
);
682 spa_strfree(vd
->vdev_devid
);
683 if (vd
->vdev_physpath
)
684 spa_strfree(vd
->vdev_physpath
);
686 if (vd
->vdev_enc_sysfs_path
)
687 spa_strfree(vd
->vdev_enc_sysfs_path
);
690 spa_strfree(vd
->vdev_fru
);
692 if (vd
->vdev_isspare
)
693 spa_spare_remove(vd
);
694 if (vd
->vdev_isl2cache
)
695 spa_l2cache_remove(vd
);
697 txg_list_destroy(&vd
->vdev_ms_list
);
698 txg_list_destroy(&vd
->vdev_dtl_list
);
700 mutex_enter(&vd
->vdev_dtl_lock
);
701 space_map_close(vd
->vdev_dtl_sm
);
702 for (t
= 0; t
< DTL_TYPES
; t
++) {
703 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
704 range_tree_destroy(vd
->vdev_dtl
[t
]);
706 mutex_exit(&vd
->vdev_dtl_lock
);
708 mutex_destroy(&vd
->vdev_queue_lock
);
709 mutex_destroy(&vd
->vdev_dtl_lock
);
710 mutex_destroy(&vd
->vdev_stat_lock
);
711 mutex_destroy(&vd
->vdev_probe_lock
);
713 if (vd
== spa
->spa_root_vdev
)
714 spa
->spa_root_vdev
= NULL
;
716 kmem_free(vd
, sizeof (vdev_t
));
720 * Transfer top-level vdev state from svd to tvd.
723 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
725 spa_t
*spa
= svd
->vdev_spa
;
730 ASSERT(tvd
== tvd
->vdev_top
);
732 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
733 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
734 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
735 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
736 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
738 svd
->vdev_ms_array
= 0;
739 svd
->vdev_ms_shift
= 0;
740 svd
->vdev_ms_count
= 0;
741 svd
->vdev_top_zap
= 0;
744 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
745 tvd
->vdev_mg
= svd
->vdev_mg
;
746 tvd
->vdev_ms
= svd
->vdev_ms
;
751 if (tvd
->vdev_mg
!= NULL
)
752 tvd
->vdev_mg
->mg_vd
= tvd
;
754 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
755 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
756 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
758 svd
->vdev_stat
.vs_alloc
= 0;
759 svd
->vdev_stat
.vs_space
= 0;
760 svd
->vdev_stat
.vs_dspace
= 0;
762 for (t
= 0; t
< TXG_SIZE
; t
++) {
763 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
764 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
765 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
766 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
767 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
768 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
771 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
772 vdev_config_clean(svd
);
773 vdev_config_dirty(tvd
);
776 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
777 vdev_state_clean(svd
);
778 vdev_state_dirty(tvd
);
781 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
782 svd
->vdev_deflate_ratio
= 0;
784 tvd
->vdev_islog
= svd
->vdev_islog
;
789 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
798 for (c
= 0; c
< vd
->vdev_children
; c
++)
799 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
803 * Add a mirror/replacing vdev above an existing vdev.
806 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
808 spa_t
*spa
= cvd
->vdev_spa
;
809 vdev_t
*pvd
= cvd
->vdev_parent
;
812 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
814 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
816 mvd
->vdev_asize
= cvd
->vdev_asize
;
817 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
818 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
819 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
820 mvd
->vdev_state
= cvd
->vdev_state
;
821 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
823 vdev_remove_child(pvd
, cvd
);
824 vdev_add_child(pvd
, mvd
);
825 cvd
->vdev_id
= mvd
->vdev_children
;
826 vdev_add_child(mvd
, cvd
);
827 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
829 if (mvd
== mvd
->vdev_top
)
830 vdev_top_transfer(cvd
, mvd
);
836 * Remove a 1-way mirror/replacing vdev from the tree.
839 vdev_remove_parent(vdev_t
*cvd
)
841 vdev_t
*mvd
= cvd
->vdev_parent
;
842 vdev_t
*pvd
= mvd
->vdev_parent
;
844 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
846 ASSERT(mvd
->vdev_children
== 1);
847 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
848 mvd
->vdev_ops
== &vdev_replacing_ops
||
849 mvd
->vdev_ops
== &vdev_spare_ops
);
850 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
852 vdev_remove_child(mvd
, cvd
);
853 vdev_remove_child(pvd
, mvd
);
856 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
857 * Otherwise, we could have detached an offline device, and when we
858 * go to import the pool we'll think we have two top-level vdevs,
859 * instead of a different version of the same top-level vdev.
861 if (mvd
->vdev_top
== mvd
) {
862 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
863 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
864 cvd
->vdev_guid
+= guid_delta
;
865 cvd
->vdev_guid_sum
+= guid_delta
;
868 * If pool not set for autoexpand, we need to also preserve
869 * mvd's asize to prevent automatic expansion of cvd.
870 * Otherwise if we are adjusting the mirror by attaching and
871 * detaching children of non-uniform sizes, the mirror could
872 * autoexpand, unexpectedly requiring larger devices to
873 * re-establish the mirror.
875 if (!cvd
->vdev_spa
->spa_autoexpand
)
876 cvd
->vdev_asize
= mvd
->vdev_asize
;
878 cvd
->vdev_id
= mvd
->vdev_id
;
879 vdev_add_child(pvd
, cvd
);
880 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
882 if (cvd
== cvd
->vdev_top
)
883 vdev_top_transfer(mvd
, cvd
);
885 ASSERT(mvd
->vdev_children
== 0);
890 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
892 spa_t
*spa
= vd
->vdev_spa
;
893 objset_t
*mos
= spa
->spa_meta_objset
;
895 uint64_t oldc
= vd
->vdev_ms_count
;
896 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
900 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
903 * This vdev is not being allocated from yet or is a hole.
905 if (vd
->vdev_ms_shift
== 0)
908 ASSERT(!vd
->vdev_ishole
);
911 * Compute the raidz-deflation ratio. Note, we hard-code
912 * in 128k (1 << 17) because it is the "typical" blocksize.
913 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
914 * otherwise it would inconsistently account for existing bp's.
916 vd
->vdev_deflate_ratio
= (1 << 17) /
917 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
919 ASSERT(oldc
<= newc
);
921 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
924 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
925 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
929 vd
->vdev_ms_count
= newc
;
931 for (m
= oldc
; m
< newc
; m
++) {
935 error
= dmu_read(mos
, vd
->vdev_ms_array
,
936 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
942 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
949 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
952 * If the vdev is being removed we don't activate
953 * the metaslabs since we want to ensure that no new
954 * allocations are performed on this device.
956 if (oldc
== 0 && !vd
->vdev_removing
)
957 metaslab_group_activate(vd
->vdev_mg
);
960 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
966 vdev_metaslab_fini(vdev_t
*vd
)
969 uint64_t count
= vd
->vdev_ms_count
;
971 if (vd
->vdev_ms
!= NULL
) {
972 metaslab_group_passivate(vd
->vdev_mg
);
973 for (m
= 0; m
< count
; m
++) {
974 metaslab_t
*msp
= vd
->vdev_ms
[m
];
979 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
983 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
986 typedef struct vdev_probe_stats
{
987 boolean_t vps_readable
;
988 boolean_t vps_writeable
;
990 } vdev_probe_stats_t
;
993 vdev_probe_done(zio_t
*zio
)
995 spa_t
*spa
= zio
->io_spa
;
996 vdev_t
*vd
= zio
->io_vd
;
997 vdev_probe_stats_t
*vps
= zio
->io_private
;
999 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1001 if (zio
->io_type
== ZIO_TYPE_READ
) {
1002 if (zio
->io_error
== 0)
1003 vps
->vps_readable
= 1;
1004 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1005 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1006 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1007 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1008 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1010 abd_free(zio
->io_abd
);
1012 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1013 if (zio
->io_error
== 0)
1014 vps
->vps_writeable
= 1;
1015 abd_free(zio
->io_abd
);
1016 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1020 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1021 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1023 if (vdev_readable(vd
) &&
1024 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1027 ASSERT(zio
->io_error
!= 0);
1028 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1029 spa
, vd
, NULL
, 0, 0);
1030 zio
->io_error
= SET_ERROR(ENXIO
);
1033 mutex_enter(&vd
->vdev_probe_lock
);
1034 ASSERT(vd
->vdev_probe_zio
== zio
);
1035 vd
->vdev_probe_zio
= NULL
;
1036 mutex_exit(&vd
->vdev_probe_lock
);
1039 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1040 if (!vdev_accessible(vd
, pio
))
1041 pio
->io_error
= SET_ERROR(ENXIO
);
1043 kmem_free(vps
, sizeof (*vps
));
1048 * Determine whether this device is accessible.
1050 * Read and write to several known locations: the pad regions of each
1051 * vdev label but the first, which we leave alone in case it contains
1055 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1057 spa_t
*spa
= vd
->vdev_spa
;
1058 vdev_probe_stats_t
*vps
= NULL
;
1062 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1065 * Don't probe the probe.
1067 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1071 * To prevent 'probe storms' when a device fails, we create
1072 * just one probe i/o at a time. All zios that want to probe
1073 * this vdev will become parents of the probe io.
1075 mutex_enter(&vd
->vdev_probe_lock
);
1077 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1078 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1080 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1081 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1084 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1086 * vdev_cant_read and vdev_cant_write can only
1087 * transition from TRUE to FALSE when we have the
1088 * SCL_ZIO lock as writer; otherwise they can only
1089 * transition from FALSE to TRUE. This ensures that
1090 * any zio looking at these values can assume that
1091 * failures persist for the life of the I/O. That's
1092 * important because when a device has intermittent
1093 * connectivity problems, we want to ensure that
1094 * they're ascribed to the device (ENXIO) and not
1097 * Since we hold SCL_ZIO as writer here, clear both
1098 * values so the probe can reevaluate from first
1101 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1102 vd
->vdev_cant_read
= B_FALSE
;
1103 vd
->vdev_cant_write
= B_FALSE
;
1106 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1107 vdev_probe_done
, vps
,
1108 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1111 * We can't change the vdev state in this context, so we
1112 * kick off an async task to do it on our behalf.
1115 vd
->vdev_probe_wanted
= B_TRUE
;
1116 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1121 zio_add_child(zio
, pio
);
1123 mutex_exit(&vd
->vdev_probe_lock
);
1126 ASSERT(zio
!= NULL
);
1130 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1131 zio_nowait(zio_read_phys(pio
, vd
,
1132 vdev_label_offset(vd
->vdev_psize
, l
,
1133 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1134 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1135 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1136 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1147 vdev_open_child(void *arg
)
1151 vd
->vdev_open_thread
= curthread
;
1152 vd
->vdev_open_error
= vdev_open(vd
);
1153 vd
->vdev_open_thread
= NULL
;
1157 vdev_uses_zvols(vdev_t
*vd
)
1162 if (zvol_is_zvol(vd
->vdev_path
))
1166 for (c
= 0; c
< vd
->vdev_children
; c
++)
1167 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1174 vdev_open_children(vdev_t
*vd
)
1177 int children
= vd
->vdev_children
;
1181 * in order to handle pools on top of zvols, do the opens
1182 * in a single thread so that the same thread holds the
1183 * spa_namespace_lock
1185 if (vdev_uses_zvols(vd
)) {
1187 for (c
= 0; c
< children
; c
++)
1188 vd
->vdev_child
[c
]->vdev_open_error
=
1189 vdev_open(vd
->vdev_child
[c
]);
1191 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1192 children
, children
, TASKQ_PREPOPULATE
);
1196 for (c
= 0; c
< children
; c
++)
1197 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1198 vd
->vdev_child
[c
], TQ_SLEEP
) != TASKQID_INVALID
);
1203 vd
->vdev_nonrot
= B_TRUE
;
1205 for (c
= 0; c
< children
; c
++)
1206 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1210 * Prepare a virtual device for access.
1213 vdev_open(vdev_t
*vd
)
1215 spa_t
*spa
= vd
->vdev_spa
;
1218 uint64_t max_osize
= 0;
1219 uint64_t asize
, max_asize
, psize
;
1220 uint64_t ashift
= 0;
1223 ASSERT(vd
->vdev_open_thread
== curthread
||
1224 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1225 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1226 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1227 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1229 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1230 vd
->vdev_cant_read
= B_FALSE
;
1231 vd
->vdev_cant_write
= B_FALSE
;
1232 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1235 * If this vdev is not removed, check its fault status. If it's
1236 * faulted, bail out of the open.
1238 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1239 ASSERT(vd
->vdev_children
== 0);
1240 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1241 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1242 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1243 vd
->vdev_label_aux
);
1244 return (SET_ERROR(ENXIO
));
1245 } else if (vd
->vdev_offline
) {
1246 ASSERT(vd
->vdev_children
== 0);
1247 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1248 return (SET_ERROR(ENXIO
));
1251 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1254 * Reset the vdev_reopening flag so that we actually close
1255 * the vdev on error.
1257 vd
->vdev_reopening
= B_FALSE
;
1258 if (zio_injection_enabled
&& error
== 0)
1259 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1262 if (vd
->vdev_removed
&&
1263 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1264 vd
->vdev_removed
= B_FALSE
;
1266 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1267 vd
->vdev_stat
.vs_aux
);
1271 vd
->vdev_removed
= B_FALSE
;
1274 * Recheck the faulted flag now that we have confirmed that
1275 * the vdev is accessible. If we're faulted, bail.
1277 if (vd
->vdev_faulted
) {
1278 ASSERT(vd
->vdev_children
== 0);
1279 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1280 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1281 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1282 vd
->vdev_label_aux
);
1283 return (SET_ERROR(ENXIO
));
1286 if (vd
->vdev_degraded
) {
1287 ASSERT(vd
->vdev_children
== 0);
1288 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1289 VDEV_AUX_ERR_EXCEEDED
);
1291 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1295 * For hole or missing vdevs we just return success.
1297 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1300 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1301 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1302 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1308 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1309 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1311 if (vd
->vdev_children
== 0) {
1312 if (osize
< SPA_MINDEVSIZE
) {
1313 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1314 VDEV_AUX_TOO_SMALL
);
1315 return (SET_ERROR(EOVERFLOW
));
1318 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1319 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1320 VDEV_LABEL_END_SIZE
);
1322 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1323 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1324 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1325 VDEV_AUX_TOO_SMALL
);
1326 return (SET_ERROR(EOVERFLOW
));
1330 max_asize
= max_osize
;
1333 vd
->vdev_psize
= psize
;
1336 * Make sure the allocatable size hasn't shrunk too much.
1338 if (asize
< vd
->vdev_min_asize
) {
1339 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1340 VDEV_AUX_BAD_LABEL
);
1341 return (SET_ERROR(EINVAL
));
1344 if (vd
->vdev_asize
== 0) {
1346 * This is the first-ever open, so use the computed values.
1347 * For compatibility, a different ashift can be requested.
1349 vd
->vdev_asize
= asize
;
1350 vd
->vdev_max_asize
= max_asize
;
1351 if (vd
->vdev_ashift
== 0) {
1352 vd
->vdev_ashift
= ashift
; /* use detected value */
1354 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1355 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1356 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1357 VDEV_AUX_BAD_ASHIFT
);
1358 return (SET_ERROR(EDOM
));
1362 * Detect if the alignment requirement has increased.
1363 * We don't want to make the pool unavailable, just
1364 * post an event instead.
1366 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1367 vd
->vdev_ops
->vdev_op_leaf
) {
1368 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1369 spa
, vd
, NULL
, 0, 0);
1372 vd
->vdev_max_asize
= max_asize
;
1376 * If all children are healthy we update asize if either:
1377 * The asize has increased, due to a device expansion caused by dynamic
1378 * LUN growth or vdev replacement, and automatic expansion is enabled;
1379 * making the additional space available.
1381 * The asize has decreased, due to a device shrink usually caused by a
1382 * vdev replace with a smaller device. This ensures that calculations
1383 * based of max_asize and asize e.g. esize are always valid. It's safe
1384 * to do this as we've already validated that asize is greater than
1387 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1388 ((asize
> vd
->vdev_asize
&&
1389 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1390 (asize
< vd
->vdev_asize
)))
1391 vd
->vdev_asize
= asize
;
1393 vdev_set_min_asize(vd
);
1396 * Ensure we can issue some IO before declaring the
1397 * vdev open for business.
1399 if (vd
->vdev_ops
->vdev_op_leaf
&&
1400 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1401 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1402 VDEV_AUX_ERR_EXCEEDED
);
1407 * Track the min and max ashift values for normal data devices.
1409 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1410 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1411 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1412 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1413 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1414 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1418 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1419 * resilver. But don't do this if we are doing a reopen for a scrub,
1420 * since this would just restart the scrub we are already doing.
1422 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1423 vdev_resilver_needed(vd
, NULL
, NULL
))
1424 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1430 * Called once the vdevs are all opened, this routine validates the label
1431 * contents. This needs to be done before vdev_load() so that we don't
1432 * inadvertently do repair I/Os to the wrong device.
1434 * If 'strict' is false ignore the spa guid check. This is necessary because
1435 * if the machine crashed during a re-guid the new guid might have been written
1436 * to all of the vdev labels, but not the cached config. The strict check
1437 * will be performed when the pool is opened again using the mos config.
1439 * This function will only return failure if one of the vdevs indicates that it
1440 * has since been destroyed or exported. This is only possible if
1441 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1442 * will be updated but the function will return 0.
1445 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1447 spa_t
*spa
= vd
->vdev_spa
;
1449 uint64_t guid
= 0, top_guid
;
1453 for (c
= 0; c
< vd
->vdev_children
; c
++)
1454 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1455 return (SET_ERROR(EBADF
));
1458 * If the device has already failed, or was marked offline, don't do
1459 * any further validation. Otherwise, label I/O will fail and we will
1460 * overwrite the previous state.
1462 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1463 uint64_t aux_guid
= 0;
1465 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1466 spa_last_synced_txg(spa
) : -1ULL;
1468 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1469 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1470 VDEV_AUX_BAD_LABEL
);
1475 * Determine if this vdev has been split off into another
1476 * pool. If so, then refuse to open it.
1478 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1479 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1480 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1481 VDEV_AUX_SPLIT_POOL
);
1486 if (strict
&& (nvlist_lookup_uint64(label
,
1487 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1488 guid
!= spa_guid(spa
))) {
1489 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1490 VDEV_AUX_CORRUPT_DATA
);
1495 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1496 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1501 * If this vdev just became a top-level vdev because its
1502 * sibling was detached, it will have adopted the parent's
1503 * vdev guid -- but the label may or may not be on disk yet.
1504 * Fortunately, either version of the label will have the
1505 * same top guid, so if we're a top-level vdev, we can
1506 * safely compare to that instead.
1508 * If we split this vdev off instead, then we also check the
1509 * original pool's guid. We don't want to consider the vdev
1510 * corrupt if it is partway through a split operation.
1512 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1514 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1516 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1517 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1518 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1519 VDEV_AUX_CORRUPT_DATA
);
1524 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1526 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1527 VDEV_AUX_CORRUPT_DATA
);
1535 * If this is a verbatim import, no need to check the
1536 * state of the pool.
1538 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1539 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1540 state
!= POOL_STATE_ACTIVE
)
1541 return (SET_ERROR(EBADF
));
1544 * If we were able to open and validate a vdev that was
1545 * previously marked permanently unavailable, clear that state
1548 if (vd
->vdev_not_present
)
1549 vd
->vdev_not_present
= 0;
1556 * Close a virtual device.
1559 vdev_close(vdev_t
*vd
)
1561 vdev_t
*pvd
= vd
->vdev_parent
;
1562 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1564 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1567 * If our parent is reopening, then we are as well, unless we are
1570 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1571 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1573 vd
->vdev_ops
->vdev_op_close(vd
);
1575 vdev_cache_purge(vd
);
1578 * We record the previous state before we close it, so that if we are
1579 * doing a reopen(), we don't generate FMA ereports if we notice that
1580 * it's still faulted.
1582 vd
->vdev_prevstate
= vd
->vdev_state
;
1584 if (vd
->vdev_offline
)
1585 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1587 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1588 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1592 vdev_hold(vdev_t
*vd
)
1594 spa_t
*spa
= vd
->vdev_spa
;
1597 ASSERT(spa_is_root(spa
));
1598 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1601 for (c
= 0; c
< vd
->vdev_children
; c
++)
1602 vdev_hold(vd
->vdev_child
[c
]);
1604 if (vd
->vdev_ops
->vdev_op_leaf
)
1605 vd
->vdev_ops
->vdev_op_hold(vd
);
1609 vdev_rele(vdev_t
*vd
)
1613 ASSERT(spa_is_root(vd
->vdev_spa
));
1614 for (c
= 0; c
< vd
->vdev_children
; c
++)
1615 vdev_rele(vd
->vdev_child
[c
]);
1617 if (vd
->vdev_ops
->vdev_op_leaf
)
1618 vd
->vdev_ops
->vdev_op_rele(vd
);
1622 * Reopen all interior vdevs and any unopened leaves. We don't actually
1623 * reopen leaf vdevs which had previously been opened as they might deadlock
1624 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1625 * If the leaf has never been opened then open it, as usual.
1628 vdev_reopen(vdev_t
*vd
)
1630 spa_t
*spa
= vd
->vdev_spa
;
1632 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1634 /* set the reopening flag unless we're taking the vdev offline */
1635 vd
->vdev_reopening
= !vd
->vdev_offline
;
1637 (void) vdev_open(vd
);
1640 * Call vdev_validate() here to make sure we have the same device.
1641 * Otherwise, a device with an invalid label could be successfully
1642 * opened in response to vdev_reopen().
1645 (void) vdev_validate_aux(vd
);
1646 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1647 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1648 !l2arc_vdev_present(vd
))
1649 l2arc_add_vdev(spa
, vd
);
1651 (void) vdev_validate(vd
, B_TRUE
);
1655 * Reassess parent vdev's health.
1657 vdev_propagate_state(vd
);
1661 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1666 * Normally, partial opens (e.g. of a mirror) are allowed.
1667 * For a create, however, we want to fail the request if
1668 * there are any components we can't open.
1670 error
= vdev_open(vd
);
1672 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1674 return (error
? error
: ENXIO
);
1678 * Recursively load DTLs and initialize all labels.
1680 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1681 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1682 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1691 vdev_metaslab_set_size(vdev_t
*vd
)
1694 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1696 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1697 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1701 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1703 ASSERT(vd
== vd
->vdev_top
);
1704 ASSERT(!vd
->vdev_ishole
);
1705 ASSERT(ISP2(flags
));
1706 ASSERT(spa_writeable(vd
->vdev_spa
));
1708 if (flags
& VDD_METASLAB
)
1709 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1711 if (flags
& VDD_DTL
)
1712 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1714 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1718 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1722 for (c
= 0; c
< vd
->vdev_children
; c
++)
1723 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1725 if (vd
->vdev_ops
->vdev_op_leaf
)
1726 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1732 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1733 * the vdev has less than perfect replication. There are four kinds of DTL:
1735 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1737 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1739 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1740 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1741 * txgs that was scrubbed.
1743 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1744 * persistent errors or just some device being offline.
1745 * Unlike the other three, the DTL_OUTAGE map is not generally
1746 * maintained; it's only computed when needed, typically to
1747 * determine whether a device can be detached.
1749 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1750 * either has the data or it doesn't.
1752 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1753 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1754 * if any child is less than fully replicated, then so is its parent.
1755 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1756 * comprising only those txgs which appear in 'maxfaults' or more children;
1757 * those are the txgs we don't have enough replication to read. For example,
1758 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1759 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1760 * two child DTL_MISSING maps.
1762 * It should be clear from the above that to compute the DTLs and outage maps
1763 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1764 * Therefore, that is all we keep on disk. When loading the pool, or after
1765 * a configuration change, we generate all other DTLs from first principles.
1768 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1770 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1772 ASSERT(t
< DTL_TYPES
);
1773 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1774 ASSERT(spa_writeable(vd
->vdev_spa
));
1776 mutex_enter(rt
->rt_lock
);
1777 if (!range_tree_contains(rt
, txg
, size
))
1778 range_tree_add(rt
, txg
, size
);
1779 mutex_exit(rt
->rt_lock
);
1783 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1785 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1786 boolean_t dirty
= B_FALSE
;
1788 ASSERT(t
< DTL_TYPES
);
1789 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1791 mutex_enter(rt
->rt_lock
);
1792 if (range_tree_space(rt
) != 0)
1793 dirty
= range_tree_contains(rt
, txg
, size
);
1794 mutex_exit(rt
->rt_lock
);
1800 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1802 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1805 mutex_enter(rt
->rt_lock
);
1806 empty
= (range_tree_space(rt
) == 0);
1807 mutex_exit(rt
->rt_lock
);
1813 * Returns the lowest txg in the DTL range.
1816 vdev_dtl_min(vdev_t
*vd
)
1820 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1821 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1822 ASSERT0(vd
->vdev_children
);
1824 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1825 return (rs
->rs_start
- 1);
1829 * Returns the highest txg in the DTL.
1832 vdev_dtl_max(vdev_t
*vd
)
1836 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1837 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1838 ASSERT0(vd
->vdev_children
);
1840 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1841 return (rs
->rs_end
);
1845 * Determine if a resilvering vdev should remove any DTL entries from
1846 * its range. If the vdev was resilvering for the entire duration of the
1847 * scan then it should excise that range from its DTLs. Otherwise, this
1848 * vdev is considered partially resilvered and should leave its DTL
1849 * entries intact. The comment in vdev_dtl_reassess() describes how we
1853 vdev_dtl_should_excise(vdev_t
*vd
)
1855 spa_t
*spa
= vd
->vdev_spa
;
1856 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1858 ASSERT0(scn
->scn_phys
.scn_errors
);
1859 ASSERT0(vd
->vdev_children
);
1861 if (vd
->vdev_resilver_txg
== 0 ||
1862 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1866 * When a resilver is initiated the scan will assign the scn_max_txg
1867 * value to the highest txg value that exists in all DTLs. If this
1868 * device's max DTL is not part of this scan (i.e. it is not in
1869 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1872 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1873 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1874 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1875 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1882 * Reassess DTLs after a config change or scrub completion.
1885 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1887 spa_t
*spa
= vd
->vdev_spa
;
1891 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1893 for (c
= 0; c
< vd
->vdev_children
; c
++)
1894 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1895 scrub_txg
, scrub_done
);
1897 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1900 if (vd
->vdev_ops
->vdev_op_leaf
) {
1901 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1903 mutex_enter(&vd
->vdev_dtl_lock
);
1906 * If we've completed a scan cleanly then determine
1907 * if this vdev should remove any DTLs. We only want to
1908 * excise regions on vdevs that were available during
1909 * the entire duration of this scan.
1911 if (scrub_txg
!= 0 &&
1912 (spa
->spa_scrub_started
||
1913 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1914 vdev_dtl_should_excise(vd
)) {
1916 * We completed a scrub up to scrub_txg. If we
1917 * did it without rebooting, then the scrub dtl
1918 * will be valid, so excise the old region and
1919 * fold in the scrub dtl. Otherwise, leave the
1920 * dtl as-is if there was an error.
1922 * There's little trick here: to excise the beginning
1923 * of the DTL_MISSING map, we put it into a reference
1924 * tree and then add a segment with refcnt -1 that
1925 * covers the range [0, scrub_txg). This means
1926 * that each txg in that range has refcnt -1 or 0.
1927 * We then add DTL_SCRUB with a refcnt of 2, so that
1928 * entries in the range [0, scrub_txg) will have a
1929 * positive refcnt -- either 1 or 2. We then convert
1930 * the reference tree into the new DTL_MISSING map.
1932 space_reftree_create(&reftree
);
1933 space_reftree_add_map(&reftree
,
1934 vd
->vdev_dtl
[DTL_MISSING
], 1);
1935 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1936 space_reftree_add_map(&reftree
,
1937 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1938 space_reftree_generate_map(&reftree
,
1939 vd
->vdev_dtl
[DTL_MISSING
], 1);
1940 space_reftree_destroy(&reftree
);
1942 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1943 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1944 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1946 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1947 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1948 if (!vdev_readable(vd
))
1949 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1951 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1952 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1955 * If the vdev was resilvering and no longer has any
1956 * DTLs then reset its resilvering flag and dirty
1957 * the top level so that we persist the change.
1959 if (vd
->vdev_resilver_txg
!= 0 &&
1960 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1961 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
1962 vd
->vdev_resilver_txg
= 0;
1963 vdev_config_dirty(vd
->vdev_top
);
1966 mutex_exit(&vd
->vdev_dtl_lock
);
1969 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1973 mutex_enter(&vd
->vdev_dtl_lock
);
1974 for (t
= 0; t
< DTL_TYPES
; t
++) {
1977 /* account for child's outage in parent's missing map */
1978 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1980 continue; /* leaf vdevs only */
1981 if (t
== DTL_PARTIAL
)
1982 minref
= 1; /* i.e. non-zero */
1983 else if (vd
->vdev_nparity
!= 0)
1984 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1986 minref
= vd
->vdev_children
; /* any kind of mirror */
1987 space_reftree_create(&reftree
);
1988 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1989 vdev_t
*cvd
= vd
->vdev_child
[c
];
1990 mutex_enter(&cvd
->vdev_dtl_lock
);
1991 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1992 mutex_exit(&cvd
->vdev_dtl_lock
);
1994 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1995 space_reftree_destroy(&reftree
);
1997 mutex_exit(&vd
->vdev_dtl_lock
);
2001 vdev_dtl_load(vdev_t
*vd
)
2003 spa_t
*spa
= vd
->vdev_spa
;
2004 objset_t
*mos
= spa
->spa_meta_objset
;
2008 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2009 ASSERT(!vd
->vdev_ishole
);
2011 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2012 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
2015 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2017 mutex_enter(&vd
->vdev_dtl_lock
);
2020 * Now that we've opened the space_map we need to update
2023 space_map_update(vd
->vdev_dtl_sm
);
2025 error
= space_map_load(vd
->vdev_dtl_sm
,
2026 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2027 mutex_exit(&vd
->vdev_dtl_lock
);
2032 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2033 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2042 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2044 spa_t
*spa
= vd
->vdev_spa
;
2046 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2047 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2052 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2054 spa_t
*spa
= vd
->vdev_spa
;
2055 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2056 DMU_OT_NONE
, 0, tx
);
2059 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2066 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2070 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2071 vd
->vdev_ops
!= &vdev_missing_ops
&&
2072 vd
->vdev_ops
!= &vdev_root_ops
&&
2073 !vd
->vdev_top
->vdev_removing
) {
2074 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2075 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2077 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2078 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2081 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2082 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2087 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2089 spa_t
*spa
= vd
->vdev_spa
;
2090 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2091 objset_t
*mos
= spa
->spa_meta_objset
;
2092 range_tree_t
*rtsync
;
2095 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2097 ASSERT(!vd
->vdev_ishole
);
2098 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2100 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2102 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2103 mutex_enter(&vd
->vdev_dtl_lock
);
2104 space_map_free(vd
->vdev_dtl_sm
, tx
);
2105 space_map_close(vd
->vdev_dtl_sm
);
2106 vd
->vdev_dtl_sm
= NULL
;
2107 mutex_exit(&vd
->vdev_dtl_lock
);
2110 * We only destroy the leaf ZAP for detached leaves or for
2111 * removed log devices. Removed data devices handle leaf ZAP
2112 * cleanup later, once cancellation is no longer possible.
2114 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2115 vd
->vdev_top
->vdev_islog
)) {
2116 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2117 vd
->vdev_leaf_zap
= 0;
2124 if (vd
->vdev_dtl_sm
== NULL
) {
2125 uint64_t new_object
;
2127 new_object
= space_map_alloc(mos
, tx
);
2128 VERIFY3U(new_object
, !=, 0);
2130 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2131 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2132 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2135 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2137 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2139 mutex_enter(&rtlock
);
2141 mutex_enter(&vd
->vdev_dtl_lock
);
2142 range_tree_walk(rt
, range_tree_add
, rtsync
);
2143 mutex_exit(&vd
->vdev_dtl_lock
);
2145 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2146 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2147 range_tree_vacate(rtsync
, NULL
, NULL
);
2149 range_tree_destroy(rtsync
);
2151 mutex_exit(&rtlock
);
2152 mutex_destroy(&rtlock
);
2155 * If the object for the space map has changed then dirty
2156 * the top level so that we update the config.
2158 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2159 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2160 "new object %llu", txg
, spa_name(spa
), object
,
2161 space_map_object(vd
->vdev_dtl_sm
));
2162 vdev_config_dirty(vd
->vdev_top
);
2167 mutex_enter(&vd
->vdev_dtl_lock
);
2168 space_map_update(vd
->vdev_dtl_sm
);
2169 mutex_exit(&vd
->vdev_dtl_lock
);
2173 * Determine whether the specified vdev can be offlined/detached/removed
2174 * without losing data.
2177 vdev_dtl_required(vdev_t
*vd
)
2179 spa_t
*spa
= vd
->vdev_spa
;
2180 vdev_t
*tvd
= vd
->vdev_top
;
2181 uint8_t cant_read
= vd
->vdev_cant_read
;
2184 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2186 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2190 * Temporarily mark the device as unreadable, and then determine
2191 * whether this results in any DTL outages in the top-level vdev.
2192 * If not, we can safely offline/detach/remove the device.
2194 vd
->vdev_cant_read
= B_TRUE
;
2195 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2196 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2197 vd
->vdev_cant_read
= cant_read
;
2198 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2200 if (!required
&& zio_injection_enabled
)
2201 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2207 * Determine if resilver is needed, and if so the txg range.
2210 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2212 boolean_t needed
= B_FALSE
;
2213 uint64_t thismin
= UINT64_MAX
;
2214 uint64_t thismax
= 0;
2217 if (vd
->vdev_children
== 0) {
2218 mutex_enter(&vd
->vdev_dtl_lock
);
2219 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2220 vdev_writeable(vd
)) {
2222 thismin
= vdev_dtl_min(vd
);
2223 thismax
= vdev_dtl_max(vd
);
2226 mutex_exit(&vd
->vdev_dtl_lock
);
2228 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2229 vdev_t
*cvd
= vd
->vdev_child
[c
];
2230 uint64_t cmin
, cmax
;
2232 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2233 thismin
= MIN(thismin
, cmin
);
2234 thismax
= MAX(thismax
, cmax
);
2240 if (needed
&& minp
) {
2248 vdev_load(vdev_t
*vd
)
2253 * Recursively load all children.
2255 for (c
= 0; c
< vd
->vdev_children
; c
++)
2256 vdev_load(vd
->vdev_child
[c
]);
2259 * If this is a top-level vdev, initialize its metaslabs.
2261 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2262 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2263 vdev_metaslab_init(vd
, 0) != 0))
2264 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2265 VDEV_AUX_CORRUPT_DATA
);
2267 * If this is a leaf vdev, load its DTL.
2269 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2270 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2271 VDEV_AUX_CORRUPT_DATA
);
2275 * The special vdev case is used for hot spares and l2cache devices. Its
2276 * sole purpose it to set the vdev state for the associated vdev. To do this,
2277 * we make sure that we can open the underlying device, then try to read the
2278 * label, and make sure that the label is sane and that it hasn't been
2279 * repurposed to another pool.
2282 vdev_validate_aux(vdev_t
*vd
)
2285 uint64_t guid
, version
;
2288 if (!vdev_readable(vd
))
2291 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2292 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2293 VDEV_AUX_CORRUPT_DATA
);
2297 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2298 !SPA_VERSION_IS_SUPPORTED(version
) ||
2299 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2300 guid
!= vd
->vdev_guid
||
2301 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2302 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2303 VDEV_AUX_CORRUPT_DATA
);
2309 * We don't actually check the pool state here. If it's in fact in
2310 * use by another pool, we update this fact on the fly when requested.
2317 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2319 spa_t
*spa
= vd
->vdev_spa
;
2320 objset_t
*mos
= spa
->spa_meta_objset
;
2324 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2325 ASSERT(vd
== vd
->vdev_top
);
2326 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2328 if (vd
->vdev_ms
!= NULL
) {
2329 metaslab_group_t
*mg
= vd
->vdev_mg
;
2331 metaslab_group_histogram_verify(mg
);
2332 metaslab_class_histogram_verify(mg
->mg_class
);
2334 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2335 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2337 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2340 mutex_enter(&msp
->ms_lock
);
2342 * If the metaslab was not loaded when the vdev
2343 * was removed then the histogram accounting may
2344 * not be accurate. Update the histogram information
2345 * here so that we ensure that the metaslab group
2346 * and metaslab class are up-to-date.
2348 metaslab_group_histogram_remove(mg
, msp
);
2350 VERIFY0(space_map_allocated(msp
->ms_sm
));
2351 space_map_free(msp
->ms_sm
, tx
);
2352 space_map_close(msp
->ms_sm
);
2354 mutex_exit(&msp
->ms_lock
);
2357 metaslab_group_histogram_verify(mg
);
2358 metaslab_class_histogram_verify(mg
->mg_class
);
2359 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2360 ASSERT0(mg
->mg_histogram
[i
]);
2364 if (vd
->vdev_ms_array
) {
2365 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2366 vd
->vdev_ms_array
= 0;
2369 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2370 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2371 vd
->vdev_top_zap
= 0;
2377 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2380 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2382 ASSERT(!vd
->vdev_ishole
);
2384 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2385 metaslab_sync_done(msp
, txg
);
2388 metaslab_sync_reassess(vd
->vdev_mg
);
2392 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2394 spa_t
*spa
= vd
->vdev_spa
;
2399 ASSERT(!vd
->vdev_ishole
);
2401 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2402 ASSERT(vd
== vd
->vdev_top
);
2403 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2404 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2405 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2406 ASSERT(vd
->vdev_ms_array
!= 0);
2407 vdev_config_dirty(vd
);
2412 * Remove the metadata associated with this vdev once it's empty.
2414 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2415 vdev_remove(vd
, txg
);
2417 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2418 metaslab_sync(msp
, txg
);
2419 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2422 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2423 vdev_dtl_sync(lvd
, txg
);
2425 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2429 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2431 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2435 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2436 * not be opened, and no I/O is attempted.
2439 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2443 spa_vdev_state_enter(spa
, SCL_NONE
);
2445 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2446 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2448 if (!vd
->vdev_ops
->vdev_op_leaf
)
2449 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2454 * We don't directly use the aux state here, but if we do a
2455 * vdev_reopen(), we need this value to be present to remember why we
2458 vd
->vdev_label_aux
= aux
;
2461 * Faulted state takes precedence over degraded.
2463 vd
->vdev_delayed_close
= B_FALSE
;
2464 vd
->vdev_faulted
= 1ULL;
2465 vd
->vdev_degraded
= 0ULL;
2466 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2469 * If this device has the only valid copy of the data, then
2470 * back off and simply mark the vdev as degraded instead.
2472 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2473 vd
->vdev_degraded
= 1ULL;
2474 vd
->vdev_faulted
= 0ULL;
2477 * If we reopen the device and it's not dead, only then do we
2482 if (vdev_readable(vd
))
2483 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2486 return (spa_vdev_state_exit(spa
, vd
, 0));
2490 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2491 * user that something is wrong. The vdev continues to operate as normal as far
2492 * as I/O is concerned.
2495 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2499 spa_vdev_state_enter(spa
, SCL_NONE
);
2501 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2502 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2504 if (!vd
->vdev_ops
->vdev_op_leaf
)
2505 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2508 * If the vdev is already faulted, then don't do anything.
2510 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2511 return (spa_vdev_state_exit(spa
, NULL
, 0));
2513 vd
->vdev_degraded
= 1ULL;
2514 if (!vdev_is_dead(vd
))
2515 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2518 return (spa_vdev_state_exit(spa
, vd
, 0));
2522 * Online the given vdev.
2524 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2525 * spare device should be detached when the device finishes resilvering.
2526 * Second, the online should be treated like a 'test' online case, so no FMA
2527 * events are generated if the device fails to open.
2530 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2532 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2533 boolean_t wasoffline
;
2534 vdev_state_t oldstate
;
2536 spa_vdev_state_enter(spa
, SCL_NONE
);
2538 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2539 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2541 if (!vd
->vdev_ops
->vdev_op_leaf
)
2542 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2544 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
2545 oldstate
= vd
->vdev_state
;
2548 vd
->vdev_offline
= B_FALSE
;
2549 vd
->vdev_tmpoffline
= B_FALSE
;
2550 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2551 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2553 /* XXX - L2ARC 1.0 does not support expansion */
2554 if (!vd
->vdev_aux
) {
2555 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2556 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2560 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2562 if (!vd
->vdev_aux
) {
2563 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2564 pvd
->vdev_expanding
= B_FALSE
;
2568 *newstate
= vd
->vdev_state
;
2569 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2570 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2571 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2572 vd
->vdev_parent
->vdev_child
[0] == vd
)
2573 vd
->vdev_unspare
= B_TRUE
;
2575 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2577 /* XXX - L2ARC 1.0 does not support expansion */
2579 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2580 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2584 (oldstate
< VDEV_STATE_DEGRADED
&&
2585 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
2586 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2588 return (spa_vdev_state_exit(spa
, vd
, 0));
2592 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2596 uint64_t generation
;
2597 metaslab_group_t
*mg
;
2600 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2602 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2603 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2605 if (!vd
->vdev_ops
->vdev_op_leaf
)
2606 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2610 generation
= spa
->spa_config_generation
+ 1;
2613 * If the device isn't already offline, try to offline it.
2615 if (!vd
->vdev_offline
) {
2617 * If this device has the only valid copy of some data,
2618 * don't allow it to be offlined. Log devices are always
2621 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2622 vdev_dtl_required(vd
))
2623 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2626 * If the top-level is a slog and it has had allocations
2627 * then proceed. We check that the vdev's metaslab group
2628 * is not NULL since it's possible that we may have just
2629 * added this vdev but not yet initialized its metaslabs.
2631 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2633 * Prevent any future allocations.
2635 metaslab_group_passivate(mg
);
2636 (void) spa_vdev_state_exit(spa
, vd
, 0);
2638 error
= spa_offline_log(spa
);
2640 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2643 * Check to see if the config has changed.
2645 if (error
|| generation
!= spa
->spa_config_generation
) {
2646 metaslab_group_activate(mg
);
2648 return (spa_vdev_state_exit(spa
,
2650 (void) spa_vdev_state_exit(spa
, vd
, 0);
2653 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2657 * Offline this device and reopen its top-level vdev.
2658 * If the top-level vdev is a log device then just offline
2659 * it. Otherwise, if this action results in the top-level
2660 * vdev becoming unusable, undo it and fail the request.
2662 vd
->vdev_offline
= B_TRUE
;
2665 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2666 vdev_is_dead(tvd
)) {
2667 vd
->vdev_offline
= B_FALSE
;
2669 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2673 * Add the device back into the metaslab rotor so that
2674 * once we online the device it's open for business.
2676 if (tvd
->vdev_islog
&& mg
!= NULL
)
2677 metaslab_group_activate(mg
);
2680 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2682 return (spa_vdev_state_exit(spa
, vd
, 0));
2686 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2690 mutex_enter(&spa
->spa_vdev_top_lock
);
2691 error
= vdev_offline_locked(spa
, guid
, flags
);
2692 mutex_exit(&spa
->spa_vdev_top_lock
);
2698 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2699 * vdev_offline(), we assume the spa config is locked. We also clear all
2700 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2703 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2705 vdev_t
*rvd
= spa
->spa_root_vdev
;
2708 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2713 vd
->vdev_stat
.vs_read_errors
= 0;
2714 vd
->vdev_stat
.vs_write_errors
= 0;
2715 vd
->vdev_stat
.vs_checksum_errors
= 0;
2717 for (c
= 0; c
< vd
->vdev_children
; c
++)
2718 vdev_clear(spa
, vd
->vdev_child
[c
]);
2721 * If we're in the FAULTED state or have experienced failed I/O, then
2722 * clear the persistent state and attempt to reopen the device. We
2723 * also mark the vdev config dirty, so that the new faulted state is
2724 * written out to disk.
2726 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2727 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2730 * When reopening in response to a clear event, it may be due to
2731 * a fmadm repair request. In this case, if the device is
2732 * still broken, we want to still post the ereport again.
2734 vd
->vdev_forcefault
= B_TRUE
;
2736 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2737 vd
->vdev_cant_read
= B_FALSE
;
2738 vd
->vdev_cant_write
= B_FALSE
;
2740 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2742 vd
->vdev_forcefault
= B_FALSE
;
2744 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2745 vdev_state_dirty(vd
->vdev_top
);
2747 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2748 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2750 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2754 * When clearing a FMA-diagnosed fault, we always want to
2755 * unspare the device, as we assume that the original spare was
2756 * done in response to the FMA fault.
2758 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2759 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2760 vd
->vdev_parent
->vdev_child
[0] == vd
)
2761 vd
->vdev_unspare
= B_TRUE
;
2765 vdev_is_dead(vdev_t
*vd
)
2768 * Holes and missing devices are always considered "dead".
2769 * This simplifies the code since we don't have to check for
2770 * these types of devices in the various code paths.
2771 * Instead we rely on the fact that we skip over dead devices
2772 * before issuing I/O to them.
2774 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2775 vd
->vdev_ops
== &vdev_missing_ops
);
2779 vdev_readable(vdev_t
*vd
)
2781 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2785 vdev_writeable(vdev_t
*vd
)
2787 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2791 vdev_allocatable(vdev_t
*vd
)
2793 uint64_t state
= vd
->vdev_state
;
2796 * We currently allow allocations from vdevs which may be in the
2797 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2798 * fails to reopen then we'll catch it later when we're holding
2799 * the proper locks. Note that we have to get the vdev state
2800 * in a local variable because although it changes atomically,
2801 * we're asking two separate questions about it.
2803 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2804 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
&&
2805 vd
->vdev_mg
->mg_initialized
);
2809 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2811 ASSERT(zio
->io_vd
== vd
);
2813 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2816 if (zio
->io_type
== ZIO_TYPE_READ
)
2817 return (!vd
->vdev_cant_read
);
2819 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2820 return (!vd
->vdev_cant_write
);
2826 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2829 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2830 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2831 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2834 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2838 * Get extended stats
2841 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2844 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2845 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2846 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2848 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2849 vsx
->vsx_total_histo
[t
][b
] +=
2850 cvsx
->vsx_total_histo
[t
][b
];
2854 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2855 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2856 vsx
->vsx_queue_histo
[t
][b
] +=
2857 cvsx
->vsx_queue_histo
[t
][b
];
2859 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2860 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2862 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2863 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2865 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2866 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2872 * Get statistics for the given vdev.
2875 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2879 * If we're getting stats on the root vdev, aggregate the I/O counts
2880 * over all top-level vdevs (i.e. the direct children of the root).
2882 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2884 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2885 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2888 memset(vsx
, 0, sizeof (*vsx
));
2890 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2891 vdev_t
*cvd
= vd
->vdev_child
[c
];
2892 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2893 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2895 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2897 vdev_get_child_stat(cvd
, vs
, cvs
);
2899 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2904 * We're a leaf. Just copy our ZIO active queue stats in. The
2905 * other leaf stats are updated in vdev_stat_update().
2910 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2912 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2913 vsx
->vsx_active_queue
[t
] =
2914 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2915 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2916 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2922 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2924 vdev_t
*tvd
= vd
->vdev_top
;
2925 mutex_enter(&vd
->vdev_stat_lock
);
2927 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2928 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2929 vs
->vs_state
= vd
->vdev_state
;
2930 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2931 if (vd
->vdev_ops
->vdev_op_leaf
)
2932 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
2933 VDEV_LABEL_END_SIZE
;
2935 * Report expandable space on top-level, non-auxillary devices
2936 * only. The expandable space is reported in terms of metaslab
2937 * sized units since that determines how much space the pool
2940 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
2941 vs
->vs_esize
= P2ALIGN(
2942 vd
->vdev_max_asize
- vd
->vdev_asize
,
2943 1ULL << tvd
->vdev_ms_shift
);
2945 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2946 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
2948 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2952 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
2953 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
2954 mutex_exit(&vd
->vdev_stat_lock
);
2958 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2960 return (vdev_get_stats_ex(vd
, vs
, NULL
));
2964 vdev_clear_stats(vdev_t
*vd
)
2966 mutex_enter(&vd
->vdev_stat_lock
);
2967 vd
->vdev_stat
.vs_space
= 0;
2968 vd
->vdev_stat
.vs_dspace
= 0;
2969 vd
->vdev_stat
.vs_alloc
= 0;
2970 mutex_exit(&vd
->vdev_stat_lock
);
2974 vdev_scan_stat_init(vdev_t
*vd
)
2976 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2979 for (c
= 0; c
< vd
->vdev_children
; c
++)
2980 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2982 mutex_enter(&vd
->vdev_stat_lock
);
2983 vs
->vs_scan_processed
= 0;
2984 mutex_exit(&vd
->vdev_stat_lock
);
2988 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2990 spa_t
*spa
= zio
->io_spa
;
2991 vdev_t
*rvd
= spa
->spa_root_vdev
;
2992 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2994 uint64_t txg
= zio
->io_txg
;
2995 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2996 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
2997 zio_type_t type
= zio
->io_type
;
2998 int flags
= zio
->io_flags
;
3001 * If this i/o is a gang leader, it didn't do any actual work.
3003 if (zio
->io_gang_tree
)
3006 if (zio
->io_error
== 0) {
3008 * If this is a root i/o, don't count it -- we've already
3009 * counted the top-level vdevs, and vdev_get_stats() will
3010 * aggregate them when asked. This reduces contention on
3011 * the root vdev_stat_lock and implicitly handles blocks
3012 * that compress away to holes, for which there is no i/o.
3013 * (Holes never create vdev children, so all the counters
3014 * remain zero, which is what we want.)
3016 * Note: this only applies to successful i/o (io_error == 0)
3017 * because unlike i/o counts, errors are not additive.
3018 * When reading a ditto block, for example, failure of
3019 * one top-level vdev does not imply a root-level error.
3024 ASSERT(vd
== zio
->io_vd
);
3026 if (flags
& ZIO_FLAG_IO_BYPASS
)
3029 mutex_enter(&vd
->vdev_stat_lock
);
3031 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3032 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3033 dsl_scan_phys_t
*scn_phys
=
3034 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3035 uint64_t *processed
= &scn_phys
->scn_processed
;
3038 if (vd
->vdev_ops
->vdev_op_leaf
)
3039 atomic_add_64(processed
, psize
);
3040 vs
->vs_scan_processed
+= psize
;
3043 if (flags
& ZIO_FLAG_SELF_HEAL
)
3044 vs
->vs_self_healed
+= psize
;
3048 * The bytes/ops/histograms are recorded at the leaf level and
3049 * aggregated into the higher level vdevs in vdev_get_stats().
3051 if (vd
->vdev_ops
->vdev_op_leaf
&&
3052 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3055 vs
->vs_bytes
[type
] += psize
;
3057 if (flags
& ZIO_FLAG_DELEGATED
) {
3058 vsx
->vsx_agg_histo
[zio
->io_priority
]
3059 [RQ_HISTO(zio
->io_size
)]++;
3061 vsx
->vsx_ind_histo
[zio
->io_priority
]
3062 [RQ_HISTO(zio
->io_size
)]++;
3065 if (zio
->io_delta
&& zio
->io_delay
) {
3066 vsx
->vsx_queue_histo
[zio
->io_priority
]
3067 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3068 vsx
->vsx_disk_histo
[type
]
3069 [L_HISTO(zio
->io_delay
)]++;
3070 vsx
->vsx_total_histo
[type
]
3071 [L_HISTO(zio
->io_delta
)]++;
3075 mutex_exit(&vd
->vdev_stat_lock
);
3079 if (flags
& ZIO_FLAG_SPECULATIVE
)
3083 * If this is an I/O error that is going to be retried, then ignore the
3084 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3085 * hard errors, when in reality they can happen for any number of
3086 * innocuous reasons (bus resets, MPxIO link failure, etc).
3088 if (zio
->io_error
== EIO
&&
3089 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3093 * Intent logs writes won't propagate their error to the root
3094 * I/O so don't mark these types of failures as pool-level
3097 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3100 mutex_enter(&vd
->vdev_stat_lock
);
3101 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3102 if (zio
->io_error
== ECKSUM
)
3103 vs
->vs_checksum_errors
++;
3105 vs
->vs_read_errors
++;
3107 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3108 vs
->vs_write_errors
++;
3109 mutex_exit(&vd
->vdev_stat_lock
);
3111 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3112 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3113 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3114 spa
->spa_claiming
)) {
3116 * This is either a normal write (not a repair), or it's
3117 * a repair induced by the scrub thread, or it's a repair
3118 * made by zil_claim() during spa_load() in the first txg.
3119 * In the normal case, we commit the DTL change in the same
3120 * txg as the block was born. In the scrub-induced repair
3121 * case, we know that scrubs run in first-pass syncing context,
3122 * so we commit the DTL change in spa_syncing_txg(spa).
3123 * In the zil_claim() case, we commit in spa_first_txg(spa).
3125 * We currently do not make DTL entries for failed spontaneous
3126 * self-healing writes triggered by normal (non-scrubbing)
3127 * reads, because we have no transactional context in which to
3128 * do so -- and it's not clear that it'd be desirable anyway.
3130 if (vd
->vdev_ops
->vdev_op_leaf
) {
3131 uint64_t commit_txg
= txg
;
3132 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3133 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3134 ASSERT(spa_sync_pass(spa
) == 1);
3135 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3136 commit_txg
= spa_syncing_txg(spa
);
3137 } else if (spa
->spa_claiming
) {
3138 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3139 commit_txg
= spa_first_txg(spa
);
3141 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3142 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3144 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3145 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3146 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3149 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3154 * Update the in-core space usage stats for this vdev, its metaslab class,
3155 * and the root vdev.
3158 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3159 int64_t space_delta
)
3161 int64_t dspace_delta
= space_delta
;
3162 spa_t
*spa
= vd
->vdev_spa
;
3163 vdev_t
*rvd
= spa
->spa_root_vdev
;
3164 metaslab_group_t
*mg
= vd
->vdev_mg
;
3165 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3167 ASSERT(vd
== vd
->vdev_top
);
3170 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3171 * factor. We must calculate this here and not at the root vdev
3172 * because the root vdev's psize-to-asize is simply the max of its
3173 * childrens', thus not accurate enough for us.
3175 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3176 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3177 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3178 vd
->vdev_deflate_ratio
;
3180 mutex_enter(&vd
->vdev_stat_lock
);
3181 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3182 vd
->vdev_stat
.vs_space
+= space_delta
;
3183 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3184 mutex_exit(&vd
->vdev_stat_lock
);
3186 if (mc
== spa_normal_class(spa
)) {
3187 mutex_enter(&rvd
->vdev_stat_lock
);
3188 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3189 rvd
->vdev_stat
.vs_space
+= space_delta
;
3190 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3191 mutex_exit(&rvd
->vdev_stat_lock
);
3195 ASSERT(rvd
== vd
->vdev_parent
);
3196 ASSERT(vd
->vdev_ms_count
!= 0);
3198 metaslab_class_space_update(mc
,
3199 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3204 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3205 * so that it will be written out next time the vdev configuration is synced.
3206 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3209 vdev_config_dirty(vdev_t
*vd
)
3211 spa_t
*spa
= vd
->vdev_spa
;
3212 vdev_t
*rvd
= spa
->spa_root_vdev
;
3215 ASSERT(spa_writeable(spa
));
3218 * If this is an aux vdev (as with l2cache and spare devices), then we
3219 * update the vdev config manually and set the sync flag.
3221 if (vd
->vdev_aux
!= NULL
) {
3222 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3226 for (c
= 0; c
< sav
->sav_count
; c
++) {
3227 if (sav
->sav_vdevs
[c
] == vd
)
3231 if (c
== sav
->sav_count
) {
3233 * We're being removed. There's nothing more to do.
3235 ASSERT(sav
->sav_sync
== B_TRUE
);
3239 sav
->sav_sync
= B_TRUE
;
3241 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3242 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3243 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3244 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3250 * Setting the nvlist in the middle if the array is a little
3251 * sketchy, but it will work.
3253 nvlist_free(aux
[c
]);
3254 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3260 * The dirty list is protected by the SCL_CONFIG lock. The caller
3261 * must either hold SCL_CONFIG as writer, or must be the sync thread
3262 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3263 * so this is sufficient to ensure mutual exclusion.
3265 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3266 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3267 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3270 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3271 vdev_config_dirty(rvd
->vdev_child
[c
]);
3273 ASSERT(vd
== vd
->vdev_top
);
3275 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3277 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3282 vdev_config_clean(vdev_t
*vd
)
3284 spa_t
*spa
= vd
->vdev_spa
;
3286 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3287 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3288 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3290 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3291 list_remove(&spa
->spa_config_dirty_list
, vd
);
3295 * Mark a top-level vdev's state as dirty, so that the next pass of
3296 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3297 * the state changes from larger config changes because they require
3298 * much less locking, and are often needed for administrative actions.
3301 vdev_state_dirty(vdev_t
*vd
)
3303 spa_t
*spa
= vd
->vdev_spa
;
3305 ASSERT(spa_writeable(spa
));
3306 ASSERT(vd
== vd
->vdev_top
);
3309 * The state list is protected by the SCL_STATE lock. The caller
3310 * must either hold SCL_STATE as writer, or must be the sync thread
3311 * (which holds SCL_STATE as reader). There's only one sync thread,
3312 * so this is sufficient to ensure mutual exclusion.
3314 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3315 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3316 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3318 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3319 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3323 vdev_state_clean(vdev_t
*vd
)
3325 spa_t
*spa
= vd
->vdev_spa
;
3327 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3328 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3329 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3331 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3332 list_remove(&spa
->spa_state_dirty_list
, vd
);
3336 * Propagate vdev state up from children to parent.
3339 vdev_propagate_state(vdev_t
*vd
)
3341 spa_t
*spa
= vd
->vdev_spa
;
3342 vdev_t
*rvd
= spa
->spa_root_vdev
;
3343 int degraded
= 0, faulted
= 0;
3348 if (vd
->vdev_children
> 0) {
3349 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3350 child
= vd
->vdev_child
[c
];
3353 * Don't factor holes into the decision.
3355 if (child
->vdev_ishole
)
3358 if (!vdev_readable(child
) ||
3359 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3361 * Root special: if there is a top-level log
3362 * device, treat the root vdev as if it were
3365 if (child
->vdev_islog
&& vd
== rvd
)
3369 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3373 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3377 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3380 * Root special: if there is a top-level vdev that cannot be
3381 * opened due to corrupted metadata, then propagate the root
3382 * vdev's aux state as 'corrupt' rather than 'insufficient
3385 if (corrupted
&& vd
== rvd
&&
3386 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3387 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3388 VDEV_AUX_CORRUPT_DATA
);
3391 if (vd
->vdev_parent
)
3392 vdev_propagate_state(vd
->vdev_parent
);
3396 * Set a vdev's state. If this is during an open, we don't update the parent
3397 * state, because we're in the process of opening children depth-first.
3398 * Otherwise, we propagate the change to the parent.
3400 * If this routine places a device in a faulted state, an appropriate ereport is
3404 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3406 uint64_t save_state
;
3407 spa_t
*spa
= vd
->vdev_spa
;
3409 if (state
== vd
->vdev_state
) {
3411 * Since vdev_offline() code path is already in an offline
3412 * state we can miss a statechange event to OFFLINE. Check
3413 * the previous state to catch this condition.
3415 if (vd
->vdev_ops
->vdev_op_leaf
&&
3416 (state
== VDEV_STATE_OFFLINE
) &&
3417 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
3418 /* post an offline state change */
3419 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
3421 vd
->vdev_stat
.vs_aux
= aux
;
3425 save_state
= vd
->vdev_state
;
3427 vd
->vdev_state
= state
;
3428 vd
->vdev_stat
.vs_aux
= aux
;
3431 * If we are setting the vdev state to anything but an open state, then
3432 * always close the underlying device unless the device has requested
3433 * a delayed close (i.e. we're about to remove or fault the device).
3434 * Otherwise, we keep accessible but invalid devices open forever.
3435 * We don't call vdev_close() itself, because that implies some extra
3436 * checks (offline, etc) that we don't want here. This is limited to
3437 * leaf devices, because otherwise closing the device will affect other
3440 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3441 vd
->vdev_ops
->vdev_op_leaf
)
3442 vd
->vdev_ops
->vdev_op_close(vd
);
3444 if (vd
->vdev_removed
&&
3445 state
== VDEV_STATE_CANT_OPEN
&&
3446 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3448 * If the previous state is set to VDEV_STATE_REMOVED, then this
3449 * device was previously marked removed and someone attempted to
3450 * reopen it. If this failed due to a nonexistent device, then
3451 * keep the device in the REMOVED state. We also let this be if
3452 * it is one of our special test online cases, which is only
3453 * attempting to online the device and shouldn't generate an FMA
3456 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3457 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3458 } else if (state
== VDEV_STATE_REMOVED
) {
3459 vd
->vdev_removed
= B_TRUE
;
3460 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3462 * If we fail to open a vdev during an import or recovery, we
3463 * mark it as "not available", which signifies that it was
3464 * never there to begin with. Failure to open such a device
3465 * is not considered an error.
3467 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3468 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3469 vd
->vdev_ops
->vdev_op_leaf
)
3470 vd
->vdev_not_present
= 1;
3473 * Post the appropriate ereport. If the 'prevstate' field is
3474 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3475 * that this is part of a vdev_reopen(). In this case, we don't
3476 * want to post the ereport if the device was already in the
3477 * CANT_OPEN state beforehand.
3479 * If the 'checkremove' flag is set, then this is an attempt to
3480 * online the device in response to an insertion event. If we
3481 * hit this case, then we have detected an insertion event for a
3482 * faulted or offline device that wasn't in the removed state.
3483 * In this scenario, we don't post an ereport because we are
3484 * about to replace the device, or attempt an online with
3485 * vdev_forcefault, which will generate the fault for us.
3487 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3488 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3489 vd
!= spa
->spa_root_vdev
) {
3493 case VDEV_AUX_OPEN_FAILED
:
3494 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3496 case VDEV_AUX_CORRUPT_DATA
:
3497 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3499 case VDEV_AUX_NO_REPLICAS
:
3500 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3502 case VDEV_AUX_BAD_GUID_SUM
:
3503 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3505 case VDEV_AUX_TOO_SMALL
:
3506 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3508 case VDEV_AUX_BAD_LABEL
:
3509 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3511 case VDEV_AUX_BAD_ASHIFT
:
3512 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
3515 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3518 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3521 /* Erase any notion of persistent removed state */
3522 vd
->vdev_removed
= B_FALSE
;
3524 vd
->vdev_removed
= B_FALSE
;
3528 * Notify ZED of any significant state-change on a leaf vdev.
3531 if (vd
->vdev_ops
->vdev_op_leaf
) {
3532 /* preserve original state from a vdev_reopen() */
3533 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
3534 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
3535 (save_state
<= VDEV_STATE_CLOSED
))
3536 save_state
= vd
->vdev_prevstate
;
3538 /* filter out state change due to initial vdev_open */
3539 if (save_state
> VDEV_STATE_CLOSED
)
3540 zfs_post_state_change(spa
, vd
, save_state
);
3543 if (!isopen
&& vd
->vdev_parent
)
3544 vdev_propagate_state(vd
->vdev_parent
);
3548 * Check the vdev configuration to ensure that it's capable of supporting
3549 * a root pool. We do not support partial configuration.
3552 vdev_is_bootable(vdev_t
*vd
)
3554 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3555 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3557 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0)
3561 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3562 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3569 * Load the state from the original vdev tree (ovd) which
3570 * we've retrieved from the MOS config object. If the original
3571 * vdev was offline or faulted then we transfer that state to the
3572 * device in the current vdev tree (nvd).
3575 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3579 ASSERT(nvd
->vdev_top
->vdev_islog
);
3580 ASSERT(spa_config_held(nvd
->vdev_spa
,
3581 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3582 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3584 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3585 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3587 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3589 * Restore the persistent vdev state
3591 nvd
->vdev_offline
= ovd
->vdev_offline
;
3592 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3593 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3594 nvd
->vdev_removed
= ovd
->vdev_removed
;
3599 * Determine if a log device has valid content. If the vdev was
3600 * removed or faulted in the MOS config then we know that
3601 * the content on the log device has already been written to the pool.
3604 vdev_log_state_valid(vdev_t
*vd
)
3608 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3612 for (c
= 0; c
< vd
->vdev_children
; c
++)
3613 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3620 * Expand a vdev if possible.
3623 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3625 ASSERT(vd
->vdev_top
== vd
);
3626 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3628 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3629 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3630 vdev_config_dirty(vd
);
3638 vdev_split(vdev_t
*vd
)
3640 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3642 vdev_remove_child(pvd
, vd
);
3643 vdev_compact_children(pvd
);
3645 cvd
= pvd
->vdev_child
[0];
3646 if (pvd
->vdev_children
== 1) {
3647 vdev_remove_parent(cvd
);
3648 cvd
->vdev_splitting
= B_TRUE
;
3650 vdev_propagate_state(cvd
);
3654 vdev_deadman(vdev_t
*vd
)
3658 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3659 vdev_t
*cvd
= vd
->vdev_child
[c
];
3664 if (vd
->vdev_ops
->vdev_op_leaf
) {
3665 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3667 mutex_enter(&vq
->vq_lock
);
3668 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3669 spa_t
*spa
= vd
->vdev_spa
;
3674 * Look at the head of all the pending queues,
3675 * if any I/O has been outstanding for longer than
3676 * the spa_deadman_synctime we log a zevent.
3678 fio
= avl_first(&vq
->vq_active_tree
);
3679 delta
= gethrtime() - fio
->io_timestamp
;
3680 if (delta
> spa_deadman_synctime(spa
)) {
3681 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3682 "delta %lluns, last io %lluns",
3683 fio
->io_timestamp
, delta
,
3684 vq
->vq_io_complete_ts
);
3685 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3686 spa
, vd
, fio
, 0, 0);
3689 mutex_exit(&vq
->vq_lock
);
3693 #if defined(_KERNEL) && defined(HAVE_SPL)
3694 EXPORT_SYMBOL(vdev_fault
);
3695 EXPORT_SYMBOL(vdev_degrade
);
3696 EXPORT_SYMBOL(vdev_online
);
3697 EXPORT_SYMBOL(vdev_offline
);
3698 EXPORT_SYMBOL(vdev_clear
);
3700 module_param(metaslabs_per_vdev
, int, 0644);
3701 MODULE_PARM_DESC(metaslabs_per_vdev
,
3702 "Divide added vdev into approximately (but no more than) this number "