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
;
1334 * If the vdev was expanded, record this so that we can re-create the
1335 * uberblock rings in labels {2,3}, during the next sync.
1337 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
1338 vd
->vdev_copy_uberblocks
= B_TRUE
;
1340 vd
->vdev_psize
= psize
;
1343 * Make sure the allocatable size hasn't shrunk too much.
1345 if (asize
< vd
->vdev_min_asize
) {
1346 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1347 VDEV_AUX_BAD_LABEL
);
1348 return (SET_ERROR(EINVAL
));
1351 if (vd
->vdev_asize
== 0) {
1353 * This is the first-ever open, so use the computed values.
1354 * For compatibility, a different ashift can be requested.
1356 vd
->vdev_asize
= asize
;
1357 vd
->vdev_max_asize
= max_asize
;
1358 if (vd
->vdev_ashift
== 0) {
1359 vd
->vdev_ashift
= ashift
; /* use detected value */
1361 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1362 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1363 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1364 VDEV_AUX_BAD_ASHIFT
);
1365 return (SET_ERROR(EDOM
));
1369 * Detect if the alignment requirement has increased.
1370 * We don't want to make the pool unavailable, just
1371 * post an event instead.
1373 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1374 vd
->vdev_ops
->vdev_op_leaf
) {
1375 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1376 spa
, vd
, NULL
, 0, 0);
1379 vd
->vdev_max_asize
= max_asize
;
1383 * If all children are healthy we update asize if either:
1384 * The asize has increased, due to a device expansion caused by dynamic
1385 * LUN growth or vdev replacement, and automatic expansion is enabled;
1386 * making the additional space available.
1388 * The asize has decreased, due to a device shrink usually caused by a
1389 * vdev replace with a smaller device. This ensures that calculations
1390 * based of max_asize and asize e.g. esize are always valid. It's safe
1391 * to do this as we've already validated that asize is greater than
1394 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1395 ((asize
> vd
->vdev_asize
&&
1396 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1397 (asize
< vd
->vdev_asize
)))
1398 vd
->vdev_asize
= asize
;
1400 vdev_set_min_asize(vd
);
1403 * Ensure we can issue some IO before declaring the
1404 * vdev open for business.
1406 if (vd
->vdev_ops
->vdev_op_leaf
&&
1407 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1408 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1409 VDEV_AUX_ERR_EXCEEDED
);
1414 * Track the min and max ashift values for normal data devices.
1416 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1417 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1418 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1419 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1420 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1421 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1425 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1426 * resilver. But don't do this if we are doing a reopen for a scrub,
1427 * since this would just restart the scrub we are already doing.
1429 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1430 vdev_resilver_needed(vd
, NULL
, NULL
))
1431 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1437 * Called once the vdevs are all opened, this routine validates the label
1438 * contents. This needs to be done before vdev_load() so that we don't
1439 * inadvertently do repair I/Os to the wrong device.
1441 * If 'strict' is false ignore the spa guid check. This is necessary because
1442 * if the machine crashed during a re-guid the new guid might have been written
1443 * to all of the vdev labels, but not the cached config. The strict check
1444 * will be performed when the pool is opened again using the mos config.
1446 * This function will only return failure if one of the vdevs indicates that it
1447 * has since been destroyed or exported. This is only possible if
1448 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1449 * will be updated but the function will return 0.
1452 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1454 spa_t
*spa
= vd
->vdev_spa
;
1456 uint64_t guid
= 0, top_guid
;
1460 for (c
= 0; c
< vd
->vdev_children
; c
++)
1461 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1462 return (SET_ERROR(EBADF
));
1465 * If the device has already failed, or was marked offline, don't do
1466 * any further validation. Otherwise, label I/O will fail and we will
1467 * overwrite the previous state.
1469 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1470 uint64_t aux_guid
= 0;
1472 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1473 spa_last_synced_txg(spa
) : -1ULL;
1475 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1476 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1477 VDEV_AUX_BAD_LABEL
);
1482 * Determine if this vdev has been split off into another
1483 * pool. If so, then refuse to open it.
1485 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1486 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1487 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1488 VDEV_AUX_SPLIT_POOL
);
1493 if (strict
&& (nvlist_lookup_uint64(label
,
1494 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1495 guid
!= spa_guid(spa
))) {
1496 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1497 VDEV_AUX_CORRUPT_DATA
);
1502 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1503 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1508 * If this vdev just became a top-level vdev because its
1509 * sibling was detached, it will have adopted the parent's
1510 * vdev guid -- but the label may or may not be on disk yet.
1511 * Fortunately, either version of the label will have the
1512 * same top guid, so if we're a top-level vdev, we can
1513 * safely compare to that instead.
1515 * If we split this vdev off instead, then we also check the
1516 * original pool's guid. We don't want to consider the vdev
1517 * corrupt if it is partway through a split operation.
1519 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1521 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1523 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1524 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1525 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1526 VDEV_AUX_CORRUPT_DATA
);
1531 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1533 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1534 VDEV_AUX_CORRUPT_DATA
);
1542 * If this is a verbatim import, no need to check the
1543 * state of the pool.
1545 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1546 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1547 state
!= POOL_STATE_ACTIVE
)
1548 return (SET_ERROR(EBADF
));
1551 * If we were able to open and validate a vdev that was
1552 * previously marked permanently unavailable, clear that state
1555 if (vd
->vdev_not_present
)
1556 vd
->vdev_not_present
= 0;
1563 * Close a virtual device.
1566 vdev_close(vdev_t
*vd
)
1568 vdev_t
*pvd
= vd
->vdev_parent
;
1569 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1571 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1574 * If our parent is reopening, then we are as well, unless we are
1577 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1578 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1580 vd
->vdev_ops
->vdev_op_close(vd
);
1582 vdev_cache_purge(vd
);
1585 * We record the previous state before we close it, so that if we are
1586 * doing a reopen(), we don't generate FMA ereports if we notice that
1587 * it's still faulted.
1589 vd
->vdev_prevstate
= vd
->vdev_state
;
1591 if (vd
->vdev_offline
)
1592 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1594 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1595 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1599 vdev_hold(vdev_t
*vd
)
1601 spa_t
*spa
= vd
->vdev_spa
;
1604 ASSERT(spa_is_root(spa
));
1605 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1608 for (c
= 0; c
< vd
->vdev_children
; c
++)
1609 vdev_hold(vd
->vdev_child
[c
]);
1611 if (vd
->vdev_ops
->vdev_op_leaf
)
1612 vd
->vdev_ops
->vdev_op_hold(vd
);
1616 vdev_rele(vdev_t
*vd
)
1620 ASSERT(spa_is_root(vd
->vdev_spa
));
1621 for (c
= 0; c
< vd
->vdev_children
; c
++)
1622 vdev_rele(vd
->vdev_child
[c
]);
1624 if (vd
->vdev_ops
->vdev_op_leaf
)
1625 vd
->vdev_ops
->vdev_op_rele(vd
);
1629 * Reopen all interior vdevs and any unopened leaves. We don't actually
1630 * reopen leaf vdevs which had previously been opened as they might deadlock
1631 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1632 * If the leaf has never been opened then open it, as usual.
1635 vdev_reopen(vdev_t
*vd
)
1637 spa_t
*spa
= vd
->vdev_spa
;
1639 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1641 /* set the reopening flag unless we're taking the vdev offline */
1642 vd
->vdev_reopening
= !vd
->vdev_offline
;
1644 (void) vdev_open(vd
);
1647 * Call vdev_validate() here to make sure we have the same device.
1648 * Otherwise, a device with an invalid label could be successfully
1649 * opened in response to vdev_reopen().
1652 (void) vdev_validate_aux(vd
);
1653 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1654 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1655 !l2arc_vdev_present(vd
))
1656 l2arc_add_vdev(spa
, vd
);
1658 (void) vdev_validate(vd
, B_TRUE
);
1662 * Reassess parent vdev's health.
1664 vdev_propagate_state(vd
);
1668 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1673 * Normally, partial opens (e.g. of a mirror) are allowed.
1674 * For a create, however, we want to fail the request if
1675 * there are any components we can't open.
1677 error
= vdev_open(vd
);
1679 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1681 return (error
? error
: ENXIO
);
1685 * Recursively load DTLs and initialize all labels.
1687 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1688 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1689 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1698 vdev_metaslab_set_size(vdev_t
*vd
)
1701 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1703 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1704 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1708 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1710 ASSERT(vd
== vd
->vdev_top
);
1711 ASSERT(!vd
->vdev_ishole
);
1712 ASSERT(ISP2(flags
));
1713 ASSERT(spa_writeable(vd
->vdev_spa
));
1715 if (flags
& VDD_METASLAB
)
1716 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1718 if (flags
& VDD_DTL
)
1719 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1721 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1725 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1729 for (c
= 0; c
< vd
->vdev_children
; c
++)
1730 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1732 if (vd
->vdev_ops
->vdev_op_leaf
)
1733 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1739 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1740 * the vdev has less than perfect replication. There are four kinds of DTL:
1742 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1744 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1746 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1747 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1748 * txgs that was scrubbed.
1750 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1751 * persistent errors or just some device being offline.
1752 * Unlike the other three, the DTL_OUTAGE map is not generally
1753 * maintained; it's only computed when needed, typically to
1754 * determine whether a device can be detached.
1756 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1757 * either has the data or it doesn't.
1759 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1760 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1761 * if any child is less than fully replicated, then so is its parent.
1762 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1763 * comprising only those txgs which appear in 'maxfaults' or more children;
1764 * those are the txgs we don't have enough replication to read. For example,
1765 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1766 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1767 * two child DTL_MISSING maps.
1769 * It should be clear from the above that to compute the DTLs and outage maps
1770 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1771 * Therefore, that is all we keep on disk. When loading the pool, or after
1772 * a configuration change, we generate all other DTLs from first principles.
1775 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1777 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1779 ASSERT(t
< DTL_TYPES
);
1780 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1781 ASSERT(spa_writeable(vd
->vdev_spa
));
1783 mutex_enter(rt
->rt_lock
);
1784 if (!range_tree_contains(rt
, txg
, size
))
1785 range_tree_add(rt
, txg
, size
);
1786 mutex_exit(rt
->rt_lock
);
1790 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1792 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1793 boolean_t dirty
= B_FALSE
;
1795 ASSERT(t
< DTL_TYPES
);
1796 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1798 mutex_enter(rt
->rt_lock
);
1799 if (range_tree_space(rt
) != 0)
1800 dirty
= range_tree_contains(rt
, txg
, size
);
1801 mutex_exit(rt
->rt_lock
);
1807 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1809 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1812 mutex_enter(rt
->rt_lock
);
1813 empty
= (range_tree_space(rt
) == 0);
1814 mutex_exit(rt
->rt_lock
);
1820 * Returns the lowest txg in the DTL range.
1823 vdev_dtl_min(vdev_t
*vd
)
1827 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1828 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1829 ASSERT0(vd
->vdev_children
);
1831 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1832 return (rs
->rs_start
- 1);
1836 * Returns the highest txg in the DTL.
1839 vdev_dtl_max(vdev_t
*vd
)
1843 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1844 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1845 ASSERT0(vd
->vdev_children
);
1847 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1848 return (rs
->rs_end
);
1852 * Determine if a resilvering vdev should remove any DTL entries from
1853 * its range. If the vdev was resilvering for the entire duration of the
1854 * scan then it should excise that range from its DTLs. Otherwise, this
1855 * vdev is considered partially resilvered and should leave its DTL
1856 * entries intact. The comment in vdev_dtl_reassess() describes how we
1860 vdev_dtl_should_excise(vdev_t
*vd
)
1862 spa_t
*spa
= vd
->vdev_spa
;
1863 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1865 ASSERT0(scn
->scn_phys
.scn_errors
);
1866 ASSERT0(vd
->vdev_children
);
1868 if (vd
->vdev_resilver_txg
== 0 ||
1869 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1873 * When a resilver is initiated the scan will assign the scn_max_txg
1874 * value to the highest txg value that exists in all DTLs. If this
1875 * device's max DTL is not part of this scan (i.e. it is not in
1876 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1879 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1880 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1881 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1882 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1889 * Reassess DTLs after a config change or scrub completion.
1892 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1894 spa_t
*spa
= vd
->vdev_spa
;
1898 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1900 for (c
= 0; c
< vd
->vdev_children
; c
++)
1901 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1902 scrub_txg
, scrub_done
);
1904 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1907 if (vd
->vdev_ops
->vdev_op_leaf
) {
1908 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1910 mutex_enter(&vd
->vdev_dtl_lock
);
1913 * If we've completed a scan cleanly then determine
1914 * if this vdev should remove any DTLs. We only want to
1915 * excise regions on vdevs that were available during
1916 * the entire duration of this scan.
1918 if (scrub_txg
!= 0 &&
1919 (spa
->spa_scrub_started
||
1920 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1921 vdev_dtl_should_excise(vd
)) {
1923 * We completed a scrub up to scrub_txg. If we
1924 * did it without rebooting, then the scrub dtl
1925 * will be valid, so excise the old region and
1926 * fold in the scrub dtl. Otherwise, leave the
1927 * dtl as-is if there was an error.
1929 * There's little trick here: to excise the beginning
1930 * of the DTL_MISSING map, we put it into a reference
1931 * tree and then add a segment with refcnt -1 that
1932 * covers the range [0, scrub_txg). This means
1933 * that each txg in that range has refcnt -1 or 0.
1934 * We then add DTL_SCRUB with a refcnt of 2, so that
1935 * entries in the range [0, scrub_txg) will have a
1936 * positive refcnt -- either 1 or 2. We then convert
1937 * the reference tree into the new DTL_MISSING map.
1939 space_reftree_create(&reftree
);
1940 space_reftree_add_map(&reftree
,
1941 vd
->vdev_dtl
[DTL_MISSING
], 1);
1942 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1943 space_reftree_add_map(&reftree
,
1944 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1945 space_reftree_generate_map(&reftree
,
1946 vd
->vdev_dtl
[DTL_MISSING
], 1);
1947 space_reftree_destroy(&reftree
);
1949 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1950 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1951 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1953 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1954 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1955 if (!vdev_readable(vd
))
1956 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1958 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1959 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1962 * If the vdev was resilvering and no longer has any
1963 * DTLs then reset its resilvering flag and dirty
1964 * the top level so that we persist the change.
1966 if (vd
->vdev_resilver_txg
!= 0 &&
1967 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1968 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
1969 vd
->vdev_resilver_txg
= 0;
1970 vdev_config_dirty(vd
->vdev_top
);
1973 mutex_exit(&vd
->vdev_dtl_lock
);
1976 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1980 mutex_enter(&vd
->vdev_dtl_lock
);
1981 for (t
= 0; t
< DTL_TYPES
; t
++) {
1984 /* account for child's outage in parent's missing map */
1985 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1987 continue; /* leaf vdevs only */
1988 if (t
== DTL_PARTIAL
)
1989 minref
= 1; /* i.e. non-zero */
1990 else if (vd
->vdev_nparity
!= 0)
1991 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1993 minref
= vd
->vdev_children
; /* any kind of mirror */
1994 space_reftree_create(&reftree
);
1995 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1996 vdev_t
*cvd
= vd
->vdev_child
[c
];
1997 mutex_enter(&cvd
->vdev_dtl_lock
);
1998 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1999 mutex_exit(&cvd
->vdev_dtl_lock
);
2001 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
2002 space_reftree_destroy(&reftree
);
2004 mutex_exit(&vd
->vdev_dtl_lock
);
2008 vdev_dtl_load(vdev_t
*vd
)
2010 spa_t
*spa
= vd
->vdev_spa
;
2011 objset_t
*mos
= spa
->spa_meta_objset
;
2015 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2016 ASSERT(!vd
->vdev_ishole
);
2018 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2019 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
2022 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2024 mutex_enter(&vd
->vdev_dtl_lock
);
2027 * Now that we've opened the space_map we need to update
2030 space_map_update(vd
->vdev_dtl_sm
);
2032 error
= space_map_load(vd
->vdev_dtl_sm
,
2033 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2034 mutex_exit(&vd
->vdev_dtl_lock
);
2039 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2040 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2049 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2051 spa_t
*spa
= vd
->vdev_spa
;
2053 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2054 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2059 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2061 spa_t
*spa
= vd
->vdev_spa
;
2062 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2063 DMU_OT_NONE
, 0, tx
);
2066 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2073 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2077 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2078 vd
->vdev_ops
!= &vdev_missing_ops
&&
2079 vd
->vdev_ops
!= &vdev_root_ops
&&
2080 !vd
->vdev_top
->vdev_removing
) {
2081 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2082 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2084 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2085 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2088 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2089 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2094 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2096 spa_t
*spa
= vd
->vdev_spa
;
2097 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2098 objset_t
*mos
= spa
->spa_meta_objset
;
2099 range_tree_t
*rtsync
;
2102 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2104 ASSERT(!vd
->vdev_ishole
);
2105 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2107 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2109 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2110 mutex_enter(&vd
->vdev_dtl_lock
);
2111 space_map_free(vd
->vdev_dtl_sm
, tx
);
2112 space_map_close(vd
->vdev_dtl_sm
);
2113 vd
->vdev_dtl_sm
= NULL
;
2114 mutex_exit(&vd
->vdev_dtl_lock
);
2117 * We only destroy the leaf ZAP for detached leaves or for
2118 * removed log devices. Removed data devices handle leaf ZAP
2119 * cleanup later, once cancellation is no longer possible.
2121 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2122 vd
->vdev_top
->vdev_islog
)) {
2123 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2124 vd
->vdev_leaf_zap
= 0;
2131 if (vd
->vdev_dtl_sm
== NULL
) {
2132 uint64_t new_object
;
2134 new_object
= space_map_alloc(mos
, tx
);
2135 VERIFY3U(new_object
, !=, 0);
2137 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2138 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2139 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2142 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2144 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2146 mutex_enter(&rtlock
);
2148 mutex_enter(&vd
->vdev_dtl_lock
);
2149 range_tree_walk(rt
, range_tree_add
, rtsync
);
2150 mutex_exit(&vd
->vdev_dtl_lock
);
2152 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2153 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2154 range_tree_vacate(rtsync
, NULL
, NULL
);
2156 range_tree_destroy(rtsync
);
2158 mutex_exit(&rtlock
);
2159 mutex_destroy(&rtlock
);
2162 * If the object for the space map has changed then dirty
2163 * the top level so that we update the config.
2165 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2166 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2167 "new object %llu", txg
, spa_name(spa
), object
,
2168 space_map_object(vd
->vdev_dtl_sm
));
2169 vdev_config_dirty(vd
->vdev_top
);
2174 mutex_enter(&vd
->vdev_dtl_lock
);
2175 space_map_update(vd
->vdev_dtl_sm
);
2176 mutex_exit(&vd
->vdev_dtl_lock
);
2180 * Determine whether the specified vdev can be offlined/detached/removed
2181 * without losing data.
2184 vdev_dtl_required(vdev_t
*vd
)
2186 spa_t
*spa
= vd
->vdev_spa
;
2187 vdev_t
*tvd
= vd
->vdev_top
;
2188 uint8_t cant_read
= vd
->vdev_cant_read
;
2191 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2193 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2197 * Temporarily mark the device as unreadable, and then determine
2198 * whether this results in any DTL outages in the top-level vdev.
2199 * If not, we can safely offline/detach/remove the device.
2201 vd
->vdev_cant_read
= B_TRUE
;
2202 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2203 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2204 vd
->vdev_cant_read
= cant_read
;
2205 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2207 if (!required
&& zio_injection_enabled
)
2208 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2214 * Determine if resilver is needed, and if so the txg range.
2217 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2219 boolean_t needed
= B_FALSE
;
2220 uint64_t thismin
= UINT64_MAX
;
2221 uint64_t thismax
= 0;
2224 if (vd
->vdev_children
== 0) {
2225 mutex_enter(&vd
->vdev_dtl_lock
);
2226 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2227 vdev_writeable(vd
)) {
2229 thismin
= vdev_dtl_min(vd
);
2230 thismax
= vdev_dtl_max(vd
);
2233 mutex_exit(&vd
->vdev_dtl_lock
);
2235 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2236 vdev_t
*cvd
= vd
->vdev_child
[c
];
2237 uint64_t cmin
, cmax
;
2239 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2240 thismin
= MIN(thismin
, cmin
);
2241 thismax
= MAX(thismax
, cmax
);
2247 if (needed
&& minp
) {
2255 vdev_load(vdev_t
*vd
)
2260 * Recursively load all children.
2262 for (c
= 0; c
< vd
->vdev_children
; c
++)
2263 vdev_load(vd
->vdev_child
[c
]);
2266 * If this is a top-level vdev, initialize its metaslabs.
2268 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2269 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2270 vdev_metaslab_init(vd
, 0) != 0))
2271 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2272 VDEV_AUX_CORRUPT_DATA
);
2274 * If this is a leaf vdev, load its DTL.
2276 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2277 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2278 VDEV_AUX_CORRUPT_DATA
);
2282 * The special vdev case is used for hot spares and l2cache devices. Its
2283 * sole purpose it to set the vdev state for the associated vdev. To do this,
2284 * we make sure that we can open the underlying device, then try to read the
2285 * label, and make sure that the label is sane and that it hasn't been
2286 * repurposed to another pool.
2289 vdev_validate_aux(vdev_t
*vd
)
2292 uint64_t guid
, version
;
2295 if (!vdev_readable(vd
))
2298 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2299 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2300 VDEV_AUX_CORRUPT_DATA
);
2304 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2305 !SPA_VERSION_IS_SUPPORTED(version
) ||
2306 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2307 guid
!= vd
->vdev_guid
||
2308 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2309 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2310 VDEV_AUX_CORRUPT_DATA
);
2316 * We don't actually check the pool state here. If it's in fact in
2317 * use by another pool, we update this fact on the fly when requested.
2324 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2326 spa_t
*spa
= vd
->vdev_spa
;
2327 objset_t
*mos
= spa
->spa_meta_objset
;
2331 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2332 ASSERT(vd
== vd
->vdev_top
);
2333 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2335 if (vd
->vdev_ms
!= NULL
) {
2336 metaslab_group_t
*mg
= vd
->vdev_mg
;
2338 metaslab_group_histogram_verify(mg
);
2339 metaslab_class_histogram_verify(mg
->mg_class
);
2341 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2342 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2344 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2347 mutex_enter(&msp
->ms_lock
);
2349 * If the metaslab was not loaded when the vdev
2350 * was removed then the histogram accounting may
2351 * not be accurate. Update the histogram information
2352 * here so that we ensure that the metaslab group
2353 * and metaslab class are up-to-date.
2355 metaslab_group_histogram_remove(mg
, msp
);
2357 VERIFY0(space_map_allocated(msp
->ms_sm
));
2358 space_map_free(msp
->ms_sm
, tx
);
2359 space_map_close(msp
->ms_sm
);
2361 mutex_exit(&msp
->ms_lock
);
2364 metaslab_group_histogram_verify(mg
);
2365 metaslab_class_histogram_verify(mg
->mg_class
);
2366 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2367 ASSERT0(mg
->mg_histogram
[i
]);
2371 if (vd
->vdev_ms_array
) {
2372 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2373 vd
->vdev_ms_array
= 0;
2376 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2377 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2378 vd
->vdev_top_zap
= 0;
2384 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2387 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2389 ASSERT(!vd
->vdev_ishole
);
2391 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2392 metaslab_sync_done(msp
, txg
);
2395 metaslab_sync_reassess(vd
->vdev_mg
);
2399 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2401 spa_t
*spa
= vd
->vdev_spa
;
2406 ASSERT(!vd
->vdev_ishole
);
2408 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2409 ASSERT(vd
== vd
->vdev_top
);
2410 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2411 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2412 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2413 ASSERT(vd
->vdev_ms_array
!= 0);
2414 vdev_config_dirty(vd
);
2419 * Remove the metadata associated with this vdev once it's empty.
2421 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2422 vdev_remove(vd
, txg
);
2424 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2425 metaslab_sync(msp
, txg
);
2426 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2429 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2430 vdev_dtl_sync(lvd
, txg
);
2432 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2436 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2438 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2442 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2443 * not be opened, and no I/O is attempted.
2446 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2450 spa_vdev_state_enter(spa
, SCL_NONE
);
2452 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2453 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2455 if (!vd
->vdev_ops
->vdev_op_leaf
)
2456 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2461 * We don't directly use the aux state here, but if we do a
2462 * vdev_reopen(), we need this value to be present to remember why we
2465 vd
->vdev_label_aux
= aux
;
2468 * Faulted state takes precedence over degraded.
2470 vd
->vdev_delayed_close
= B_FALSE
;
2471 vd
->vdev_faulted
= 1ULL;
2472 vd
->vdev_degraded
= 0ULL;
2473 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2476 * If this device has the only valid copy of the data, then
2477 * back off and simply mark the vdev as degraded instead.
2479 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2480 vd
->vdev_degraded
= 1ULL;
2481 vd
->vdev_faulted
= 0ULL;
2484 * If we reopen the device and it's not dead, only then do we
2489 if (vdev_readable(vd
))
2490 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2493 return (spa_vdev_state_exit(spa
, vd
, 0));
2497 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2498 * user that something is wrong. The vdev continues to operate as normal as far
2499 * as I/O is concerned.
2502 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2506 spa_vdev_state_enter(spa
, SCL_NONE
);
2508 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2509 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2511 if (!vd
->vdev_ops
->vdev_op_leaf
)
2512 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2515 * If the vdev is already faulted, then don't do anything.
2517 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2518 return (spa_vdev_state_exit(spa
, NULL
, 0));
2520 vd
->vdev_degraded
= 1ULL;
2521 if (!vdev_is_dead(vd
))
2522 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2525 return (spa_vdev_state_exit(spa
, vd
, 0));
2529 * Online the given vdev.
2531 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2532 * spare device should be detached when the device finishes resilvering.
2533 * Second, the online should be treated like a 'test' online case, so no FMA
2534 * events are generated if the device fails to open.
2537 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2539 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2540 boolean_t wasoffline
;
2541 vdev_state_t oldstate
;
2543 spa_vdev_state_enter(spa
, SCL_NONE
);
2545 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2546 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2548 if (!vd
->vdev_ops
->vdev_op_leaf
)
2549 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2551 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
2552 oldstate
= vd
->vdev_state
;
2555 vd
->vdev_offline
= B_FALSE
;
2556 vd
->vdev_tmpoffline
= B_FALSE
;
2557 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2558 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2560 /* XXX - L2ARC 1.0 does not support expansion */
2561 if (!vd
->vdev_aux
) {
2562 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2563 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2567 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2569 if (!vd
->vdev_aux
) {
2570 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2571 pvd
->vdev_expanding
= B_FALSE
;
2575 *newstate
= vd
->vdev_state
;
2576 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2577 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2578 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2579 vd
->vdev_parent
->vdev_child
[0] == vd
)
2580 vd
->vdev_unspare
= B_TRUE
;
2582 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2584 /* XXX - L2ARC 1.0 does not support expansion */
2586 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2587 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2591 (oldstate
< VDEV_STATE_DEGRADED
&&
2592 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
2593 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2595 return (spa_vdev_state_exit(spa
, vd
, 0));
2599 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2603 uint64_t generation
;
2604 metaslab_group_t
*mg
;
2607 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2609 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2610 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2612 if (!vd
->vdev_ops
->vdev_op_leaf
)
2613 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2617 generation
= spa
->spa_config_generation
+ 1;
2620 * If the device isn't already offline, try to offline it.
2622 if (!vd
->vdev_offline
) {
2624 * If this device has the only valid copy of some data,
2625 * don't allow it to be offlined. Log devices are always
2628 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2629 vdev_dtl_required(vd
))
2630 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2633 * If the top-level is a slog and it has had allocations
2634 * then proceed. We check that the vdev's metaslab group
2635 * is not NULL since it's possible that we may have just
2636 * added this vdev but not yet initialized its metaslabs.
2638 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2640 * Prevent any future allocations.
2642 metaslab_group_passivate(mg
);
2643 (void) spa_vdev_state_exit(spa
, vd
, 0);
2645 error
= spa_offline_log(spa
);
2647 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2650 * Check to see if the config has changed.
2652 if (error
|| generation
!= spa
->spa_config_generation
) {
2653 metaslab_group_activate(mg
);
2655 return (spa_vdev_state_exit(spa
,
2657 (void) spa_vdev_state_exit(spa
, vd
, 0);
2660 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2664 * Offline this device and reopen its top-level vdev.
2665 * If the top-level vdev is a log device then just offline
2666 * it. Otherwise, if this action results in the top-level
2667 * vdev becoming unusable, undo it and fail the request.
2669 vd
->vdev_offline
= B_TRUE
;
2672 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2673 vdev_is_dead(tvd
)) {
2674 vd
->vdev_offline
= B_FALSE
;
2676 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2680 * Add the device back into the metaslab rotor so that
2681 * once we online the device it's open for business.
2683 if (tvd
->vdev_islog
&& mg
!= NULL
)
2684 metaslab_group_activate(mg
);
2687 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2689 return (spa_vdev_state_exit(spa
, vd
, 0));
2693 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2697 mutex_enter(&spa
->spa_vdev_top_lock
);
2698 error
= vdev_offline_locked(spa
, guid
, flags
);
2699 mutex_exit(&spa
->spa_vdev_top_lock
);
2705 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2706 * vdev_offline(), we assume the spa config is locked. We also clear all
2707 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2710 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2712 vdev_t
*rvd
= spa
->spa_root_vdev
;
2715 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2720 vd
->vdev_stat
.vs_read_errors
= 0;
2721 vd
->vdev_stat
.vs_write_errors
= 0;
2722 vd
->vdev_stat
.vs_checksum_errors
= 0;
2724 for (c
= 0; c
< vd
->vdev_children
; c
++)
2725 vdev_clear(spa
, vd
->vdev_child
[c
]);
2728 * If we're in the FAULTED state or have experienced failed I/O, then
2729 * clear the persistent state and attempt to reopen the device. We
2730 * also mark the vdev config dirty, so that the new faulted state is
2731 * written out to disk.
2733 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2734 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2737 * When reopening in response to a clear event, it may be due to
2738 * a fmadm repair request. In this case, if the device is
2739 * still broken, we want to still post the ereport again.
2741 vd
->vdev_forcefault
= B_TRUE
;
2743 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2744 vd
->vdev_cant_read
= B_FALSE
;
2745 vd
->vdev_cant_write
= B_FALSE
;
2747 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2749 vd
->vdev_forcefault
= B_FALSE
;
2751 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2752 vdev_state_dirty(vd
->vdev_top
);
2754 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2755 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2757 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2761 * When clearing a FMA-diagnosed fault, we always want to
2762 * unspare the device, as we assume that the original spare was
2763 * done in response to the FMA fault.
2765 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2766 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2767 vd
->vdev_parent
->vdev_child
[0] == vd
)
2768 vd
->vdev_unspare
= B_TRUE
;
2772 vdev_is_dead(vdev_t
*vd
)
2775 * Holes and missing devices are always considered "dead".
2776 * This simplifies the code since we don't have to check for
2777 * these types of devices in the various code paths.
2778 * Instead we rely on the fact that we skip over dead devices
2779 * before issuing I/O to them.
2781 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2782 vd
->vdev_ops
== &vdev_missing_ops
);
2786 vdev_readable(vdev_t
*vd
)
2788 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2792 vdev_writeable(vdev_t
*vd
)
2794 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2798 vdev_allocatable(vdev_t
*vd
)
2800 uint64_t state
= vd
->vdev_state
;
2803 * We currently allow allocations from vdevs which may be in the
2804 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2805 * fails to reopen then we'll catch it later when we're holding
2806 * the proper locks. Note that we have to get the vdev state
2807 * in a local variable because although it changes atomically,
2808 * we're asking two separate questions about it.
2810 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2811 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
&&
2812 vd
->vdev_mg
->mg_initialized
);
2816 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2818 ASSERT(zio
->io_vd
== vd
);
2820 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2823 if (zio
->io_type
== ZIO_TYPE_READ
)
2824 return (!vd
->vdev_cant_read
);
2826 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2827 return (!vd
->vdev_cant_write
);
2833 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2836 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2837 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2838 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2841 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2845 * Get extended stats
2848 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2851 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2852 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2853 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2855 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2856 vsx
->vsx_total_histo
[t
][b
] +=
2857 cvsx
->vsx_total_histo
[t
][b
];
2861 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2862 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2863 vsx
->vsx_queue_histo
[t
][b
] +=
2864 cvsx
->vsx_queue_histo
[t
][b
];
2866 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2867 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2869 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2870 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2872 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2873 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2879 * Get statistics for the given vdev.
2882 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2886 * If we're getting stats on the root vdev, aggregate the I/O counts
2887 * over all top-level vdevs (i.e. the direct children of the root).
2889 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2891 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2892 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2895 memset(vsx
, 0, sizeof (*vsx
));
2897 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2898 vdev_t
*cvd
= vd
->vdev_child
[c
];
2899 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2900 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2902 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2904 vdev_get_child_stat(cvd
, vs
, cvs
);
2906 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2911 * We're a leaf. Just copy our ZIO active queue stats in. The
2912 * other leaf stats are updated in vdev_stat_update().
2917 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2919 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2920 vsx
->vsx_active_queue
[t
] =
2921 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2922 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2923 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2929 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2931 vdev_t
*tvd
= vd
->vdev_top
;
2932 mutex_enter(&vd
->vdev_stat_lock
);
2934 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2935 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2936 vs
->vs_state
= vd
->vdev_state
;
2937 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2938 if (vd
->vdev_ops
->vdev_op_leaf
)
2939 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
2940 VDEV_LABEL_END_SIZE
;
2942 * Report expandable space on top-level, non-auxillary devices
2943 * only. The expandable space is reported in terms of metaslab
2944 * sized units since that determines how much space the pool
2947 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
2948 vs
->vs_esize
= P2ALIGN(
2949 vd
->vdev_max_asize
- vd
->vdev_asize
,
2950 1ULL << tvd
->vdev_ms_shift
);
2952 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2953 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
2955 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2959 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
2960 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
2961 mutex_exit(&vd
->vdev_stat_lock
);
2965 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2967 return (vdev_get_stats_ex(vd
, vs
, NULL
));
2971 vdev_clear_stats(vdev_t
*vd
)
2973 mutex_enter(&vd
->vdev_stat_lock
);
2974 vd
->vdev_stat
.vs_space
= 0;
2975 vd
->vdev_stat
.vs_dspace
= 0;
2976 vd
->vdev_stat
.vs_alloc
= 0;
2977 mutex_exit(&vd
->vdev_stat_lock
);
2981 vdev_scan_stat_init(vdev_t
*vd
)
2983 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2986 for (c
= 0; c
< vd
->vdev_children
; c
++)
2987 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2989 mutex_enter(&vd
->vdev_stat_lock
);
2990 vs
->vs_scan_processed
= 0;
2991 mutex_exit(&vd
->vdev_stat_lock
);
2995 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2997 spa_t
*spa
= zio
->io_spa
;
2998 vdev_t
*rvd
= spa
->spa_root_vdev
;
2999 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
3001 uint64_t txg
= zio
->io_txg
;
3002 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3003 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
3004 zio_type_t type
= zio
->io_type
;
3005 int flags
= zio
->io_flags
;
3008 * If this i/o is a gang leader, it didn't do any actual work.
3010 if (zio
->io_gang_tree
)
3013 if (zio
->io_error
== 0) {
3015 * If this is a root i/o, don't count it -- we've already
3016 * counted the top-level vdevs, and vdev_get_stats() will
3017 * aggregate them when asked. This reduces contention on
3018 * the root vdev_stat_lock and implicitly handles blocks
3019 * that compress away to holes, for which there is no i/o.
3020 * (Holes never create vdev children, so all the counters
3021 * remain zero, which is what we want.)
3023 * Note: this only applies to successful i/o (io_error == 0)
3024 * because unlike i/o counts, errors are not additive.
3025 * When reading a ditto block, for example, failure of
3026 * one top-level vdev does not imply a root-level error.
3031 ASSERT(vd
== zio
->io_vd
);
3033 if (flags
& ZIO_FLAG_IO_BYPASS
)
3036 mutex_enter(&vd
->vdev_stat_lock
);
3038 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3039 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3040 dsl_scan_phys_t
*scn_phys
=
3041 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3042 uint64_t *processed
= &scn_phys
->scn_processed
;
3045 if (vd
->vdev_ops
->vdev_op_leaf
)
3046 atomic_add_64(processed
, psize
);
3047 vs
->vs_scan_processed
+= psize
;
3050 if (flags
& ZIO_FLAG_SELF_HEAL
)
3051 vs
->vs_self_healed
+= psize
;
3055 * The bytes/ops/histograms are recorded at the leaf level and
3056 * aggregated into the higher level vdevs in vdev_get_stats().
3058 if (vd
->vdev_ops
->vdev_op_leaf
&&
3059 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3062 vs
->vs_bytes
[type
] += psize
;
3064 if (flags
& ZIO_FLAG_DELEGATED
) {
3065 vsx
->vsx_agg_histo
[zio
->io_priority
]
3066 [RQ_HISTO(zio
->io_size
)]++;
3068 vsx
->vsx_ind_histo
[zio
->io_priority
]
3069 [RQ_HISTO(zio
->io_size
)]++;
3072 if (zio
->io_delta
&& zio
->io_delay
) {
3073 vsx
->vsx_queue_histo
[zio
->io_priority
]
3074 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3075 vsx
->vsx_disk_histo
[type
]
3076 [L_HISTO(zio
->io_delay
)]++;
3077 vsx
->vsx_total_histo
[type
]
3078 [L_HISTO(zio
->io_delta
)]++;
3082 mutex_exit(&vd
->vdev_stat_lock
);
3086 if (flags
& ZIO_FLAG_SPECULATIVE
)
3090 * If this is an I/O error that is going to be retried, then ignore the
3091 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3092 * hard errors, when in reality they can happen for any number of
3093 * innocuous reasons (bus resets, MPxIO link failure, etc).
3095 if (zio
->io_error
== EIO
&&
3096 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3100 * Intent logs writes won't propagate their error to the root
3101 * I/O so don't mark these types of failures as pool-level
3104 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3107 mutex_enter(&vd
->vdev_stat_lock
);
3108 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3109 if (zio
->io_error
== ECKSUM
)
3110 vs
->vs_checksum_errors
++;
3112 vs
->vs_read_errors
++;
3114 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3115 vs
->vs_write_errors
++;
3116 mutex_exit(&vd
->vdev_stat_lock
);
3118 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3119 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3120 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3121 spa
->spa_claiming
)) {
3123 * This is either a normal write (not a repair), or it's
3124 * a repair induced by the scrub thread, or it's a repair
3125 * made by zil_claim() during spa_load() in the first txg.
3126 * In the normal case, we commit the DTL change in the same
3127 * txg as the block was born. In the scrub-induced repair
3128 * case, we know that scrubs run in first-pass syncing context,
3129 * so we commit the DTL change in spa_syncing_txg(spa).
3130 * In the zil_claim() case, we commit in spa_first_txg(spa).
3132 * We currently do not make DTL entries for failed spontaneous
3133 * self-healing writes triggered by normal (non-scrubbing)
3134 * reads, because we have no transactional context in which to
3135 * do so -- and it's not clear that it'd be desirable anyway.
3137 if (vd
->vdev_ops
->vdev_op_leaf
) {
3138 uint64_t commit_txg
= txg
;
3139 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3140 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3141 ASSERT(spa_sync_pass(spa
) == 1);
3142 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3143 commit_txg
= spa_syncing_txg(spa
);
3144 } else if (spa
->spa_claiming
) {
3145 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3146 commit_txg
= spa_first_txg(spa
);
3148 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3149 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3151 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3152 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3153 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3156 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3161 * Update the in-core space usage stats for this vdev, its metaslab class,
3162 * and the root vdev.
3165 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3166 int64_t space_delta
)
3168 int64_t dspace_delta
= space_delta
;
3169 spa_t
*spa
= vd
->vdev_spa
;
3170 vdev_t
*rvd
= spa
->spa_root_vdev
;
3171 metaslab_group_t
*mg
= vd
->vdev_mg
;
3172 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3174 ASSERT(vd
== vd
->vdev_top
);
3177 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3178 * factor. We must calculate this here and not at the root vdev
3179 * because the root vdev's psize-to-asize is simply the max of its
3180 * childrens', thus not accurate enough for us.
3182 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3183 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3184 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3185 vd
->vdev_deflate_ratio
;
3187 mutex_enter(&vd
->vdev_stat_lock
);
3188 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3189 vd
->vdev_stat
.vs_space
+= space_delta
;
3190 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3191 mutex_exit(&vd
->vdev_stat_lock
);
3193 if (mc
== spa_normal_class(spa
)) {
3194 mutex_enter(&rvd
->vdev_stat_lock
);
3195 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3196 rvd
->vdev_stat
.vs_space
+= space_delta
;
3197 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3198 mutex_exit(&rvd
->vdev_stat_lock
);
3202 ASSERT(rvd
== vd
->vdev_parent
);
3203 ASSERT(vd
->vdev_ms_count
!= 0);
3205 metaslab_class_space_update(mc
,
3206 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3211 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3212 * so that it will be written out next time the vdev configuration is synced.
3213 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3216 vdev_config_dirty(vdev_t
*vd
)
3218 spa_t
*spa
= vd
->vdev_spa
;
3219 vdev_t
*rvd
= spa
->spa_root_vdev
;
3222 ASSERT(spa_writeable(spa
));
3225 * If this is an aux vdev (as with l2cache and spare devices), then we
3226 * update the vdev config manually and set the sync flag.
3228 if (vd
->vdev_aux
!= NULL
) {
3229 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3233 for (c
= 0; c
< sav
->sav_count
; c
++) {
3234 if (sav
->sav_vdevs
[c
] == vd
)
3238 if (c
== sav
->sav_count
) {
3240 * We're being removed. There's nothing more to do.
3242 ASSERT(sav
->sav_sync
== B_TRUE
);
3246 sav
->sav_sync
= B_TRUE
;
3248 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3249 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3250 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3251 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3257 * Setting the nvlist in the middle if the array is a little
3258 * sketchy, but it will work.
3260 nvlist_free(aux
[c
]);
3261 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3267 * The dirty list is protected by the SCL_CONFIG lock. The caller
3268 * must either hold SCL_CONFIG as writer, or must be the sync thread
3269 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3270 * so this is sufficient to ensure mutual exclusion.
3272 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3273 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3274 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3277 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3278 vdev_config_dirty(rvd
->vdev_child
[c
]);
3280 ASSERT(vd
== vd
->vdev_top
);
3282 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3284 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3289 vdev_config_clean(vdev_t
*vd
)
3291 spa_t
*spa
= vd
->vdev_spa
;
3293 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3294 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3295 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3297 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3298 list_remove(&spa
->spa_config_dirty_list
, vd
);
3302 * Mark a top-level vdev's state as dirty, so that the next pass of
3303 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3304 * the state changes from larger config changes because they require
3305 * much less locking, and are often needed for administrative actions.
3308 vdev_state_dirty(vdev_t
*vd
)
3310 spa_t
*spa
= vd
->vdev_spa
;
3312 ASSERT(spa_writeable(spa
));
3313 ASSERT(vd
== vd
->vdev_top
);
3316 * The state list is protected by the SCL_STATE lock. The caller
3317 * must either hold SCL_STATE as writer, or must be the sync thread
3318 * (which holds SCL_STATE as reader). There's only one sync thread,
3319 * so this is sufficient to ensure mutual exclusion.
3321 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3322 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3323 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3325 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3326 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3330 vdev_state_clean(vdev_t
*vd
)
3332 spa_t
*spa
= vd
->vdev_spa
;
3334 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3335 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3336 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3338 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3339 list_remove(&spa
->spa_state_dirty_list
, vd
);
3343 * Propagate vdev state up from children to parent.
3346 vdev_propagate_state(vdev_t
*vd
)
3348 spa_t
*spa
= vd
->vdev_spa
;
3349 vdev_t
*rvd
= spa
->spa_root_vdev
;
3350 int degraded
= 0, faulted
= 0;
3355 if (vd
->vdev_children
> 0) {
3356 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3357 child
= vd
->vdev_child
[c
];
3360 * Don't factor holes into the decision.
3362 if (child
->vdev_ishole
)
3365 if (!vdev_readable(child
) ||
3366 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3368 * Root special: if there is a top-level log
3369 * device, treat the root vdev as if it were
3372 if (child
->vdev_islog
&& vd
== rvd
)
3376 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3380 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3384 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3387 * Root special: if there is a top-level vdev that cannot be
3388 * opened due to corrupted metadata, then propagate the root
3389 * vdev's aux state as 'corrupt' rather than 'insufficient
3392 if (corrupted
&& vd
== rvd
&&
3393 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3394 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3395 VDEV_AUX_CORRUPT_DATA
);
3398 if (vd
->vdev_parent
)
3399 vdev_propagate_state(vd
->vdev_parent
);
3403 * Set a vdev's state. If this is during an open, we don't update the parent
3404 * state, because we're in the process of opening children depth-first.
3405 * Otherwise, we propagate the change to the parent.
3407 * If this routine places a device in a faulted state, an appropriate ereport is
3411 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3413 uint64_t save_state
;
3414 spa_t
*spa
= vd
->vdev_spa
;
3416 if (state
== vd
->vdev_state
) {
3418 * Since vdev_offline() code path is already in an offline
3419 * state we can miss a statechange event to OFFLINE. Check
3420 * the previous state to catch this condition.
3422 if (vd
->vdev_ops
->vdev_op_leaf
&&
3423 (state
== VDEV_STATE_OFFLINE
) &&
3424 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
3425 /* post an offline state change */
3426 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
3428 vd
->vdev_stat
.vs_aux
= aux
;
3432 save_state
= vd
->vdev_state
;
3434 vd
->vdev_state
= state
;
3435 vd
->vdev_stat
.vs_aux
= aux
;
3438 * If we are setting the vdev state to anything but an open state, then
3439 * always close the underlying device unless the device has requested
3440 * a delayed close (i.e. we're about to remove or fault the device).
3441 * Otherwise, we keep accessible but invalid devices open forever.
3442 * We don't call vdev_close() itself, because that implies some extra
3443 * checks (offline, etc) that we don't want here. This is limited to
3444 * leaf devices, because otherwise closing the device will affect other
3447 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3448 vd
->vdev_ops
->vdev_op_leaf
)
3449 vd
->vdev_ops
->vdev_op_close(vd
);
3451 if (vd
->vdev_removed
&&
3452 state
== VDEV_STATE_CANT_OPEN
&&
3453 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3455 * If the previous state is set to VDEV_STATE_REMOVED, then this
3456 * device was previously marked removed and someone attempted to
3457 * reopen it. If this failed due to a nonexistent device, then
3458 * keep the device in the REMOVED state. We also let this be if
3459 * it is one of our special test online cases, which is only
3460 * attempting to online the device and shouldn't generate an FMA
3463 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3464 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3465 } else if (state
== VDEV_STATE_REMOVED
) {
3466 vd
->vdev_removed
= B_TRUE
;
3467 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3469 * If we fail to open a vdev during an import or recovery, we
3470 * mark it as "not available", which signifies that it was
3471 * never there to begin with. Failure to open such a device
3472 * is not considered an error.
3474 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3475 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3476 vd
->vdev_ops
->vdev_op_leaf
)
3477 vd
->vdev_not_present
= 1;
3480 * Post the appropriate ereport. If the 'prevstate' field is
3481 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3482 * that this is part of a vdev_reopen(). In this case, we don't
3483 * want to post the ereport if the device was already in the
3484 * CANT_OPEN state beforehand.
3486 * If the 'checkremove' flag is set, then this is an attempt to
3487 * online the device in response to an insertion event. If we
3488 * hit this case, then we have detected an insertion event for a
3489 * faulted or offline device that wasn't in the removed state.
3490 * In this scenario, we don't post an ereport because we are
3491 * about to replace the device, or attempt an online with
3492 * vdev_forcefault, which will generate the fault for us.
3494 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3495 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3496 vd
!= spa
->spa_root_vdev
) {
3500 case VDEV_AUX_OPEN_FAILED
:
3501 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3503 case VDEV_AUX_CORRUPT_DATA
:
3504 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3506 case VDEV_AUX_NO_REPLICAS
:
3507 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3509 case VDEV_AUX_BAD_GUID_SUM
:
3510 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3512 case VDEV_AUX_TOO_SMALL
:
3513 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3515 case VDEV_AUX_BAD_LABEL
:
3516 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3518 case VDEV_AUX_BAD_ASHIFT
:
3519 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
3522 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3525 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3528 /* Erase any notion of persistent removed state */
3529 vd
->vdev_removed
= B_FALSE
;
3531 vd
->vdev_removed
= B_FALSE
;
3535 * Notify ZED of any significant state-change on a leaf vdev.
3538 if (vd
->vdev_ops
->vdev_op_leaf
) {
3539 /* preserve original state from a vdev_reopen() */
3540 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
3541 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
3542 (save_state
<= VDEV_STATE_CLOSED
))
3543 save_state
= vd
->vdev_prevstate
;
3545 /* filter out state change due to initial vdev_open */
3546 if (save_state
> VDEV_STATE_CLOSED
)
3547 zfs_post_state_change(spa
, vd
, save_state
);
3550 if (!isopen
&& vd
->vdev_parent
)
3551 vdev_propagate_state(vd
->vdev_parent
);
3555 * Check the vdev configuration to ensure that it's capable of supporting
3556 * a root pool. We do not support partial configuration.
3559 vdev_is_bootable(vdev_t
*vd
)
3561 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3562 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3564 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0)
3568 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3569 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3576 * Load the state from the original vdev tree (ovd) which
3577 * we've retrieved from the MOS config object. If the original
3578 * vdev was offline or faulted then we transfer that state to the
3579 * device in the current vdev tree (nvd).
3582 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3586 ASSERT(nvd
->vdev_top
->vdev_islog
);
3587 ASSERT(spa_config_held(nvd
->vdev_spa
,
3588 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3589 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3591 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3592 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3594 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3596 * Restore the persistent vdev state
3598 nvd
->vdev_offline
= ovd
->vdev_offline
;
3599 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3600 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3601 nvd
->vdev_removed
= ovd
->vdev_removed
;
3606 * Determine if a log device has valid content. If the vdev was
3607 * removed or faulted in the MOS config then we know that
3608 * the content on the log device has already been written to the pool.
3611 vdev_log_state_valid(vdev_t
*vd
)
3615 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3619 for (c
= 0; c
< vd
->vdev_children
; c
++)
3620 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3627 * Expand a vdev if possible.
3630 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3632 ASSERT(vd
->vdev_top
== vd
);
3633 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3635 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3636 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3637 vdev_config_dirty(vd
);
3645 vdev_split(vdev_t
*vd
)
3647 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3649 vdev_remove_child(pvd
, vd
);
3650 vdev_compact_children(pvd
);
3652 cvd
= pvd
->vdev_child
[0];
3653 if (pvd
->vdev_children
== 1) {
3654 vdev_remove_parent(cvd
);
3655 cvd
->vdev_splitting
= B_TRUE
;
3657 vdev_propagate_state(cvd
);
3661 vdev_deadman(vdev_t
*vd
)
3665 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3666 vdev_t
*cvd
= vd
->vdev_child
[c
];
3671 if (vd
->vdev_ops
->vdev_op_leaf
) {
3672 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3674 mutex_enter(&vq
->vq_lock
);
3675 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3676 spa_t
*spa
= vd
->vdev_spa
;
3681 * Look at the head of all the pending queues,
3682 * if any I/O has been outstanding for longer than
3683 * the spa_deadman_synctime we log a zevent.
3685 fio
= avl_first(&vq
->vq_active_tree
);
3686 delta
= gethrtime() - fio
->io_timestamp
;
3687 if (delta
> spa_deadman_synctime(spa
)) {
3688 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3689 "delta %lluns, last io %lluns",
3690 fio
->io_timestamp
, delta
,
3691 vq
->vq_io_complete_ts
);
3692 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3693 spa
, vd
, fio
, 0, 0);
3696 mutex_exit(&vq
->vq_lock
);
3700 #if defined(_KERNEL) && defined(HAVE_SPL)
3701 EXPORT_SYMBOL(vdev_fault
);
3702 EXPORT_SYMBOL(vdev_degrade
);
3703 EXPORT_SYMBOL(vdev_online
);
3704 EXPORT_SYMBOL(vdev_offline
);
3705 EXPORT_SYMBOL(vdev_clear
);
3707 module_param(metaslabs_per_vdev
, int, 0644);
3708 MODULE_PARM_DESC(metaslabs_per_vdev
,
3709 "Divide added vdev into approximately (but no more than) this number "