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
, spa
,
374 offsetof(struct metaslab
, ms_txg_node
));
375 txg_list_create(&vd
->vdev_dtl_list
, spa
,
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
;
400 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
402 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
403 return (SET_ERROR(EINVAL
));
405 if ((ops
= vdev_getops(type
)) == NULL
)
406 return (SET_ERROR(EINVAL
));
409 * If this is a load, get the vdev guid from the nvlist.
410 * Otherwise, vdev_alloc_common() will generate one for us.
412 if (alloctype
== VDEV_ALLOC_LOAD
) {
415 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
417 return (SET_ERROR(EINVAL
));
419 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
420 return (SET_ERROR(EINVAL
));
421 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
422 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
423 return (SET_ERROR(EINVAL
));
424 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
425 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
426 return (SET_ERROR(EINVAL
));
427 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
428 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
429 return (SET_ERROR(EINVAL
));
433 * The first allocated vdev must be of type 'root'.
435 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
436 return (SET_ERROR(EINVAL
));
439 * Determine whether we're a log vdev.
442 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
443 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
444 return (SET_ERROR(ENOTSUP
));
446 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
447 return (SET_ERROR(ENOTSUP
));
450 * Set the nparity property for RAID-Z vdevs.
453 if (ops
== &vdev_raidz_ops
) {
454 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
456 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
457 return (SET_ERROR(EINVAL
));
459 * Previous versions could only support 1 or 2 parity
463 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
464 return (SET_ERROR(ENOTSUP
));
466 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
467 return (SET_ERROR(ENOTSUP
));
470 * We require the parity to be specified for SPAs that
471 * support multiple parity levels.
473 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
474 return (SET_ERROR(EINVAL
));
476 * Otherwise, we default to 1 parity device for RAID-Z.
483 ASSERT(nparity
!= -1ULL);
485 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
487 vd
->vdev_islog
= islog
;
488 vd
->vdev_nparity
= nparity
;
490 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
491 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
494 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
495 * fault on a vdev and want it to persist across imports (like with
498 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
499 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
500 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
501 vd
->vdev_faulted
= 1;
502 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
505 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
506 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
507 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
508 &vd
->vdev_physpath
) == 0)
509 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
511 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
512 &vd
->vdev_enc_sysfs_path
) == 0)
513 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
515 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
516 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
519 * Set the whole_disk property. If it's not specified, leave the value
522 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
523 &vd
->vdev_wholedisk
) != 0)
524 vd
->vdev_wholedisk
= -1ULL;
527 * Look for the 'not present' flag. This will only be set if the device
528 * was not present at the time of import.
530 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
531 &vd
->vdev_not_present
);
534 * Get the alignment requirement.
536 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
539 * Retrieve the vdev creation time.
541 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
545 * If we're a top-level vdev, try to load the allocation parameters.
547 if (parent
&& !parent
->vdev_parent
&&
548 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
549 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
551 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
553 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
555 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
557 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
560 ASSERT0(vd
->vdev_top_zap
);
563 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
564 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
565 alloctype
== VDEV_ALLOC_ADD
||
566 alloctype
== VDEV_ALLOC_SPLIT
||
567 alloctype
== VDEV_ALLOC_ROOTPOOL
);
568 vd
->vdev_mg
= metaslab_group_create(islog
?
569 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
572 if (vd
->vdev_ops
->vdev_op_leaf
&&
573 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
574 (void) nvlist_lookup_uint64(nv
,
575 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
577 ASSERT0(vd
->vdev_leaf_zap
);
581 * If we're a leaf vdev, try to load the DTL object and other state.
584 if (vd
->vdev_ops
->vdev_op_leaf
&&
585 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
586 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
587 if (alloctype
== VDEV_ALLOC_LOAD
) {
588 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
589 &vd
->vdev_dtl_object
);
590 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
594 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
597 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
598 &spare
) == 0 && spare
)
602 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
605 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
606 &vd
->vdev_resilver_txg
);
609 * In general, when importing a pool we want to ignore the
610 * persistent fault state, as the diagnosis made on another
611 * system may not be valid in the current context. The only
612 * exception is if we forced a vdev to a persistently faulted
613 * state with 'zpool offline -f'. The persistent fault will
614 * remain across imports until cleared.
616 * Local vdevs will remain in the faulted state.
618 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
619 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
620 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
622 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
624 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
627 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
631 VDEV_AUX_ERR_EXCEEDED
;
632 if (nvlist_lookup_string(nv
,
633 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
634 strcmp(aux
, "external") == 0)
635 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
641 * Add ourselves to the parent's list of children.
643 vdev_add_child(parent
, vd
);
651 vdev_free(vdev_t
*vd
)
654 spa_t
*spa
= vd
->vdev_spa
;
657 * vdev_free() implies closing the vdev first. This is simpler than
658 * trying to ensure complicated semantics for all callers.
662 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
663 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
668 for (c
= 0; c
< vd
->vdev_children
; c
++)
669 vdev_free(vd
->vdev_child
[c
]);
671 ASSERT(vd
->vdev_child
== NULL
);
672 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
675 * Discard allocation state.
677 if (vd
->vdev_mg
!= NULL
) {
678 vdev_metaslab_fini(vd
);
679 metaslab_group_destroy(vd
->vdev_mg
);
682 ASSERT0(vd
->vdev_stat
.vs_space
);
683 ASSERT0(vd
->vdev_stat
.vs_dspace
);
684 ASSERT0(vd
->vdev_stat
.vs_alloc
);
687 * Remove this vdev from its parent's child list.
689 vdev_remove_child(vd
->vdev_parent
, vd
);
691 ASSERT(vd
->vdev_parent
== NULL
);
694 * Clean up vdev structure.
700 spa_strfree(vd
->vdev_path
);
702 spa_strfree(vd
->vdev_devid
);
703 if (vd
->vdev_physpath
)
704 spa_strfree(vd
->vdev_physpath
);
706 if (vd
->vdev_enc_sysfs_path
)
707 spa_strfree(vd
->vdev_enc_sysfs_path
);
710 spa_strfree(vd
->vdev_fru
);
712 if (vd
->vdev_isspare
)
713 spa_spare_remove(vd
);
714 if (vd
->vdev_isl2cache
)
715 spa_l2cache_remove(vd
);
717 txg_list_destroy(&vd
->vdev_ms_list
);
718 txg_list_destroy(&vd
->vdev_dtl_list
);
720 mutex_enter(&vd
->vdev_dtl_lock
);
721 space_map_close(vd
->vdev_dtl_sm
);
722 for (t
= 0; t
< DTL_TYPES
; t
++) {
723 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
724 range_tree_destroy(vd
->vdev_dtl
[t
]);
726 mutex_exit(&vd
->vdev_dtl_lock
);
728 mutex_destroy(&vd
->vdev_queue_lock
);
729 mutex_destroy(&vd
->vdev_dtl_lock
);
730 mutex_destroy(&vd
->vdev_stat_lock
);
731 mutex_destroy(&vd
->vdev_probe_lock
);
733 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
734 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
736 if (vd
== spa
->spa_root_vdev
)
737 spa
->spa_root_vdev
= NULL
;
739 kmem_free(vd
, sizeof (vdev_t
));
743 * Transfer top-level vdev state from svd to tvd.
746 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
748 spa_t
*spa
= svd
->vdev_spa
;
753 ASSERT(tvd
== tvd
->vdev_top
);
755 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
756 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
757 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
758 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
759 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
761 svd
->vdev_ms_array
= 0;
762 svd
->vdev_ms_shift
= 0;
763 svd
->vdev_ms_count
= 0;
764 svd
->vdev_top_zap
= 0;
767 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
768 tvd
->vdev_mg
= svd
->vdev_mg
;
769 tvd
->vdev_ms
= svd
->vdev_ms
;
774 if (tvd
->vdev_mg
!= NULL
)
775 tvd
->vdev_mg
->mg_vd
= tvd
;
777 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
778 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
779 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
781 svd
->vdev_stat
.vs_alloc
= 0;
782 svd
->vdev_stat
.vs_space
= 0;
783 svd
->vdev_stat
.vs_dspace
= 0;
785 for (t
= 0; t
< TXG_SIZE
; t
++) {
786 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
787 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
788 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
789 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
790 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
791 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
794 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
795 vdev_config_clean(svd
);
796 vdev_config_dirty(tvd
);
799 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
800 vdev_state_clean(svd
);
801 vdev_state_dirty(tvd
);
804 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
805 svd
->vdev_deflate_ratio
= 0;
807 tvd
->vdev_islog
= svd
->vdev_islog
;
812 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
821 for (c
= 0; c
< vd
->vdev_children
; c
++)
822 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
826 * Add a mirror/replacing vdev above an existing vdev.
829 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
831 spa_t
*spa
= cvd
->vdev_spa
;
832 vdev_t
*pvd
= cvd
->vdev_parent
;
835 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
837 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
839 mvd
->vdev_asize
= cvd
->vdev_asize
;
840 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
841 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
842 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
843 mvd
->vdev_state
= cvd
->vdev_state
;
844 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
846 vdev_remove_child(pvd
, cvd
);
847 vdev_add_child(pvd
, mvd
);
848 cvd
->vdev_id
= mvd
->vdev_children
;
849 vdev_add_child(mvd
, cvd
);
850 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
852 if (mvd
== mvd
->vdev_top
)
853 vdev_top_transfer(cvd
, mvd
);
859 * Remove a 1-way mirror/replacing vdev from the tree.
862 vdev_remove_parent(vdev_t
*cvd
)
864 vdev_t
*mvd
= cvd
->vdev_parent
;
865 vdev_t
*pvd
= mvd
->vdev_parent
;
867 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
869 ASSERT(mvd
->vdev_children
== 1);
870 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
871 mvd
->vdev_ops
== &vdev_replacing_ops
||
872 mvd
->vdev_ops
== &vdev_spare_ops
);
873 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
875 vdev_remove_child(mvd
, cvd
);
876 vdev_remove_child(pvd
, mvd
);
879 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
880 * Otherwise, we could have detached an offline device, and when we
881 * go to import the pool we'll think we have two top-level vdevs,
882 * instead of a different version of the same top-level vdev.
884 if (mvd
->vdev_top
== mvd
) {
885 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
886 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
887 cvd
->vdev_guid
+= guid_delta
;
888 cvd
->vdev_guid_sum
+= guid_delta
;
891 * If pool not set for autoexpand, we need to also preserve
892 * mvd's asize to prevent automatic expansion of cvd.
893 * Otherwise if we are adjusting the mirror by attaching and
894 * detaching children of non-uniform sizes, the mirror could
895 * autoexpand, unexpectedly requiring larger devices to
896 * re-establish the mirror.
898 if (!cvd
->vdev_spa
->spa_autoexpand
)
899 cvd
->vdev_asize
= mvd
->vdev_asize
;
901 cvd
->vdev_id
= mvd
->vdev_id
;
902 vdev_add_child(pvd
, cvd
);
903 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
905 if (cvd
== cvd
->vdev_top
)
906 vdev_top_transfer(mvd
, cvd
);
908 ASSERT(mvd
->vdev_children
== 0);
913 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
915 spa_t
*spa
= vd
->vdev_spa
;
916 objset_t
*mos
= spa
->spa_meta_objset
;
918 uint64_t oldc
= vd
->vdev_ms_count
;
919 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
923 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
926 * This vdev is not being allocated from yet or is a hole.
928 if (vd
->vdev_ms_shift
== 0)
931 ASSERT(!vd
->vdev_ishole
);
934 * Compute the raidz-deflation ratio. Note, we hard-code
935 * in 128k (1 << 17) because it is the "typical" blocksize.
936 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
937 * otherwise it would inconsistently account for existing bp's.
939 vd
->vdev_deflate_ratio
= (1 << 17) /
940 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
942 ASSERT(oldc
<= newc
);
944 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
947 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
948 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
952 vd
->vdev_ms_count
= newc
;
954 for (m
= oldc
; m
< newc
; m
++) {
958 error
= dmu_read(mos
, vd
->vdev_ms_array
,
959 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
965 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
972 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
975 * If the vdev is being removed we don't activate
976 * the metaslabs since we want to ensure that no new
977 * allocations are performed on this device.
979 if (oldc
== 0 && !vd
->vdev_removing
)
980 metaslab_group_activate(vd
->vdev_mg
);
983 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
989 vdev_metaslab_fini(vdev_t
*vd
)
992 uint64_t count
= vd
->vdev_ms_count
;
994 if (vd
->vdev_ms
!= NULL
) {
995 metaslab_group_passivate(vd
->vdev_mg
);
996 for (m
= 0; m
< count
; m
++) {
997 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1002 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1006 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1009 typedef struct vdev_probe_stats
{
1010 boolean_t vps_readable
;
1011 boolean_t vps_writeable
;
1013 } vdev_probe_stats_t
;
1016 vdev_probe_done(zio_t
*zio
)
1018 spa_t
*spa
= zio
->io_spa
;
1019 vdev_t
*vd
= zio
->io_vd
;
1020 vdev_probe_stats_t
*vps
= zio
->io_private
;
1022 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1024 if (zio
->io_type
== ZIO_TYPE_READ
) {
1025 if (zio
->io_error
== 0)
1026 vps
->vps_readable
= 1;
1027 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1028 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1029 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1030 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1031 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1033 abd_free(zio
->io_abd
);
1035 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1036 if (zio
->io_error
== 0)
1037 vps
->vps_writeable
= 1;
1038 abd_free(zio
->io_abd
);
1039 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1043 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1044 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1046 if (vdev_readable(vd
) &&
1047 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1050 ASSERT(zio
->io_error
!= 0);
1051 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1052 spa
, vd
, NULL
, 0, 0);
1053 zio
->io_error
= SET_ERROR(ENXIO
);
1056 mutex_enter(&vd
->vdev_probe_lock
);
1057 ASSERT(vd
->vdev_probe_zio
== zio
);
1058 vd
->vdev_probe_zio
= NULL
;
1059 mutex_exit(&vd
->vdev_probe_lock
);
1062 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1063 if (!vdev_accessible(vd
, pio
))
1064 pio
->io_error
= SET_ERROR(ENXIO
);
1066 kmem_free(vps
, sizeof (*vps
));
1071 * Determine whether this device is accessible.
1073 * Read and write to several known locations: the pad regions of each
1074 * vdev label but the first, which we leave alone in case it contains
1078 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1080 spa_t
*spa
= vd
->vdev_spa
;
1081 vdev_probe_stats_t
*vps
= NULL
;
1085 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1088 * Don't probe the probe.
1090 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1094 * To prevent 'probe storms' when a device fails, we create
1095 * just one probe i/o at a time. All zios that want to probe
1096 * this vdev will become parents of the probe io.
1098 mutex_enter(&vd
->vdev_probe_lock
);
1100 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1101 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1103 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1104 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1107 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1109 * vdev_cant_read and vdev_cant_write can only
1110 * transition from TRUE to FALSE when we have the
1111 * SCL_ZIO lock as writer; otherwise they can only
1112 * transition from FALSE to TRUE. This ensures that
1113 * any zio looking at these values can assume that
1114 * failures persist for the life of the I/O. That's
1115 * important because when a device has intermittent
1116 * connectivity problems, we want to ensure that
1117 * they're ascribed to the device (ENXIO) and not
1120 * Since we hold SCL_ZIO as writer here, clear both
1121 * values so the probe can reevaluate from first
1124 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1125 vd
->vdev_cant_read
= B_FALSE
;
1126 vd
->vdev_cant_write
= B_FALSE
;
1129 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1130 vdev_probe_done
, vps
,
1131 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1134 * We can't change the vdev state in this context, so we
1135 * kick off an async task to do it on our behalf.
1138 vd
->vdev_probe_wanted
= B_TRUE
;
1139 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1144 zio_add_child(zio
, pio
);
1146 mutex_exit(&vd
->vdev_probe_lock
);
1149 ASSERT(zio
!= NULL
);
1153 for (l
= 1; l
< VDEV_LABELS
; l
++) {
1154 zio_nowait(zio_read_phys(pio
, vd
,
1155 vdev_label_offset(vd
->vdev_psize
, l
,
1156 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1157 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1158 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1159 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1170 vdev_open_child(void *arg
)
1174 vd
->vdev_open_thread
= curthread
;
1175 vd
->vdev_open_error
= vdev_open(vd
);
1176 vd
->vdev_open_thread
= NULL
;
1180 vdev_uses_zvols(vdev_t
*vd
)
1185 if (zvol_is_zvol(vd
->vdev_path
))
1189 for (c
= 0; c
< vd
->vdev_children
; c
++)
1190 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1197 vdev_open_children(vdev_t
*vd
)
1200 int children
= vd
->vdev_children
;
1204 * in order to handle pools on top of zvols, do the opens
1205 * in a single thread so that the same thread holds the
1206 * spa_namespace_lock
1208 if (vdev_uses_zvols(vd
)) {
1210 for (c
= 0; c
< children
; c
++)
1211 vd
->vdev_child
[c
]->vdev_open_error
=
1212 vdev_open(vd
->vdev_child
[c
]);
1214 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1215 children
, children
, TASKQ_PREPOPULATE
);
1219 for (c
= 0; c
< children
; c
++)
1220 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1221 vd
->vdev_child
[c
], TQ_SLEEP
) != TASKQID_INVALID
);
1226 vd
->vdev_nonrot
= B_TRUE
;
1228 for (c
= 0; c
< children
; c
++)
1229 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1233 * Prepare a virtual device for access.
1236 vdev_open(vdev_t
*vd
)
1238 spa_t
*spa
= vd
->vdev_spa
;
1241 uint64_t max_osize
= 0;
1242 uint64_t asize
, max_asize
, psize
;
1243 uint64_t ashift
= 0;
1246 ASSERT(vd
->vdev_open_thread
== curthread
||
1247 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1248 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1249 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1250 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1252 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1253 vd
->vdev_cant_read
= B_FALSE
;
1254 vd
->vdev_cant_write
= B_FALSE
;
1255 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1258 * If this vdev is not removed, check its fault status. If it's
1259 * faulted, bail out of the open.
1261 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1262 ASSERT(vd
->vdev_children
== 0);
1263 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1264 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1265 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1266 vd
->vdev_label_aux
);
1267 return (SET_ERROR(ENXIO
));
1268 } else if (vd
->vdev_offline
) {
1269 ASSERT(vd
->vdev_children
== 0);
1270 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1271 return (SET_ERROR(ENXIO
));
1274 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1277 * Reset the vdev_reopening flag so that we actually close
1278 * the vdev on error.
1280 vd
->vdev_reopening
= B_FALSE
;
1281 if (zio_injection_enabled
&& error
== 0)
1282 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1285 if (vd
->vdev_removed
&&
1286 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1287 vd
->vdev_removed
= B_FALSE
;
1289 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1290 vd
->vdev_stat
.vs_aux
);
1294 vd
->vdev_removed
= B_FALSE
;
1297 * Recheck the faulted flag now that we have confirmed that
1298 * the vdev is accessible. If we're faulted, bail.
1300 if (vd
->vdev_faulted
) {
1301 ASSERT(vd
->vdev_children
== 0);
1302 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1303 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1304 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1305 vd
->vdev_label_aux
);
1306 return (SET_ERROR(ENXIO
));
1309 if (vd
->vdev_degraded
) {
1310 ASSERT(vd
->vdev_children
== 0);
1311 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1312 VDEV_AUX_ERR_EXCEEDED
);
1314 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1318 * For hole or missing vdevs we just return success.
1320 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1323 for (c
= 0; c
< vd
->vdev_children
; c
++) {
1324 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1325 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1331 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1332 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1334 if (vd
->vdev_children
== 0) {
1335 if (osize
< SPA_MINDEVSIZE
) {
1336 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1337 VDEV_AUX_TOO_SMALL
);
1338 return (SET_ERROR(EOVERFLOW
));
1341 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1342 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1343 VDEV_LABEL_END_SIZE
);
1345 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1346 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1347 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1348 VDEV_AUX_TOO_SMALL
);
1349 return (SET_ERROR(EOVERFLOW
));
1353 max_asize
= max_osize
;
1357 * If the vdev was expanded, record this so that we can re-create the
1358 * uberblock rings in labels {2,3}, during the next sync.
1360 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
1361 vd
->vdev_copy_uberblocks
= B_TRUE
;
1363 vd
->vdev_psize
= psize
;
1366 * Make sure the allocatable size hasn't shrunk too much.
1368 if (asize
< vd
->vdev_min_asize
) {
1369 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1370 VDEV_AUX_BAD_LABEL
);
1371 return (SET_ERROR(EINVAL
));
1374 if (vd
->vdev_asize
== 0) {
1376 * This is the first-ever open, so use the computed values.
1377 * For compatibility, a different ashift can be requested.
1379 vd
->vdev_asize
= asize
;
1380 vd
->vdev_max_asize
= max_asize
;
1381 if (vd
->vdev_ashift
== 0) {
1382 vd
->vdev_ashift
= ashift
; /* use detected value */
1384 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1385 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1386 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1387 VDEV_AUX_BAD_ASHIFT
);
1388 return (SET_ERROR(EDOM
));
1392 * Detect if the alignment requirement has increased.
1393 * We don't want to make the pool unavailable, just
1394 * post an event instead.
1396 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1397 vd
->vdev_ops
->vdev_op_leaf
) {
1398 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1399 spa
, vd
, NULL
, 0, 0);
1402 vd
->vdev_max_asize
= max_asize
;
1406 * If all children are healthy we update asize if either:
1407 * The asize has increased, due to a device expansion caused by dynamic
1408 * LUN growth or vdev replacement, and automatic expansion is enabled;
1409 * making the additional space available.
1411 * The asize has decreased, due to a device shrink usually caused by a
1412 * vdev replace with a smaller device. This ensures that calculations
1413 * based of max_asize and asize e.g. esize are always valid. It's safe
1414 * to do this as we've already validated that asize is greater than
1417 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1418 ((asize
> vd
->vdev_asize
&&
1419 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1420 (asize
< vd
->vdev_asize
)))
1421 vd
->vdev_asize
= asize
;
1423 vdev_set_min_asize(vd
);
1426 * Ensure we can issue some IO before declaring the
1427 * vdev open for business.
1429 if (vd
->vdev_ops
->vdev_op_leaf
&&
1430 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1431 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1432 VDEV_AUX_ERR_EXCEEDED
);
1437 * Track the min and max ashift values for normal data devices.
1439 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1440 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1441 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1442 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1443 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1444 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1448 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1449 * resilver. But don't do this if we are doing a reopen for a scrub,
1450 * since this would just restart the scrub we are already doing.
1452 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1453 vdev_resilver_needed(vd
, NULL
, NULL
))
1454 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1460 * Called once the vdevs are all opened, this routine validates the label
1461 * contents. This needs to be done before vdev_load() so that we don't
1462 * inadvertently do repair I/Os to the wrong device.
1464 * If 'strict' is false ignore the spa guid check. This is necessary because
1465 * if the machine crashed during a re-guid the new guid might have been written
1466 * to all of the vdev labels, but not the cached config. The strict check
1467 * will be performed when the pool is opened again using the mos config.
1469 * This function will only return failure if one of the vdevs indicates that it
1470 * has since been destroyed or exported. This is only possible if
1471 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1472 * will be updated but the function will return 0.
1475 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1477 spa_t
*spa
= vd
->vdev_spa
;
1479 uint64_t guid
= 0, top_guid
;
1483 for (c
= 0; c
< vd
->vdev_children
; c
++)
1484 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1485 return (SET_ERROR(EBADF
));
1488 * If the device has already failed, or was marked offline, don't do
1489 * any further validation. Otherwise, label I/O will fail and we will
1490 * overwrite the previous state.
1492 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1493 uint64_t aux_guid
= 0;
1495 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1496 spa_last_synced_txg(spa
) : -1ULL;
1498 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1499 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1500 VDEV_AUX_BAD_LABEL
);
1505 * Determine if this vdev has been split off into another
1506 * pool. If so, then refuse to open it.
1508 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1509 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1510 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1511 VDEV_AUX_SPLIT_POOL
);
1516 if (strict
&& (nvlist_lookup_uint64(label
,
1517 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1518 guid
!= spa_guid(spa
))) {
1519 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1520 VDEV_AUX_CORRUPT_DATA
);
1525 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1526 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1531 * If this vdev just became a top-level vdev because its
1532 * sibling was detached, it will have adopted the parent's
1533 * vdev guid -- but the label may or may not be on disk yet.
1534 * Fortunately, either version of the label will have the
1535 * same top guid, so if we're a top-level vdev, we can
1536 * safely compare to that instead.
1538 * If we split this vdev off instead, then we also check the
1539 * original pool's guid. We don't want to consider the vdev
1540 * corrupt if it is partway through a split operation.
1542 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1544 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1546 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1547 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1548 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1549 VDEV_AUX_CORRUPT_DATA
);
1554 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1556 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1557 VDEV_AUX_CORRUPT_DATA
);
1565 * If this is a verbatim import, no need to check the
1566 * state of the pool.
1568 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1569 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1570 state
!= POOL_STATE_ACTIVE
)
1571 return (SET_ERROR(EBADF
));
1574 * If we were able to open and validate a vdev that was
1575 * previously marked permanently unavailable, clear that state
1578 if (vd
->vdev_not_present
)
1579 vd
->vdev_not_present
= 0;
1586 * Close a virtual device.
1589 vdev_close(vdev_t
*vd
)
1591 vdev_t
*pvd
= vd
->vdev_parent
;
1592 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
1594 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1597 * If our parent is reopening, then we are as well, unless we are
1600 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1601 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1603 vd
->vdev_ops
->vdev_op_close(vd
);
1605 vdev_cache_purge(vd
);
1608 * We record the previous state before we close it, so that if we are
1609 * doing a reopen(), we don't generate FMA ereports if we notice that
1610 * it's still faulted.
1612 vd
->vdev_prevstate
= vd
->vdev_state
;
1614 if (vd
->vdev_offline
)
1615 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1617 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1618 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1622 vdev_hold(vdev_t
*vd
)
1624 spa_t
*spa
= vd
->vdev_spa
;
1627 ASSERT(spa_is_root(spa
));
1628 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1631 for (c
= 0; c
< vd
->vdev_children
; c
++)
1632 vdev_hold(vd
->vdev_child
[c
]);
1634 if (vd
->vdev_ops
->vdev_op_leaf
)
1635 vd
->vdev_ops
->vdev_op_hold(vd
);
1639 vdev_rele(vdev_t
*vd
)
1643 ASSERT(spa_is_root(vd
->vdev_spa
));
1644 for (c
= 0; c
< vd
->vdev_children
; c
++)
1645 vdev_rele(vd
->vdev_child
[c
]);
1647 if (vd
->vdev_ops
->vdev_op_leaf
)
1648 vd
->vdev_ops
->vdev_op_rele(vd
);
1652 * Reopen all interior vdevs and any unopened leaves. We don't actually
1653 * reopen leaf vdevs which had previously been opened as they might deadlock
1654 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1655 * If the leaf has never been opened then open it, as usual.
1658 vdev_reopen(vdev_t
*vd
)
1660 spa_t
*spa
= vd
->vdev_spa
;
1662 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1664 /* set the reopening flag unless we're taking the vdev offline */
1665 vd
->vdev_reopening
= !vd
->vdev_offline
;
1667 (void) vdev_open(vd
);
1670 * Call vdev_validate() here to make sure we have the same device.
1671 * Otherwise, a device with an invalid label could be successfully
1672 * opened in response to vdev_reopen().
1675 (void) vdev_validate_aux(vd
);
1676 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1677 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1678 !l2arc_vdev_present(vd
))
1679 l2arc_add_vdev(spa
, vd
);
1681 (void) vdev_validate(vd
, B_TRUE
);
1685 * Reassess parent vdev's health.
1687 vdev_propagate_state(vd
);
1691 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1696 * Normally, partial opens (e.g. of a mirror) are allowed.
1697 * For a create, however, we want to fail the request if
1698 * there are any components we can't open.
1700 error
= vdev_open(vd
);
1702 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1704 return (error
? error
: ENXIO
);
1708 * Recursively load DTLs and initialize all labels.
1710 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1711 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1712 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1721 vdev_metaslab_set_size(vdev_t
*vd
)
1724 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1726 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1727 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1731 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1733 ASSERT(vd
== vd
->vdev_top
);
1734 ASSERT(!vd
->vdev_ishole
);
1735 ASSERT(ISP2(flags
));
1736 ASSERT(spa_writeable(vd
->vdev_spa
));
1738 if (flags
& VDD_METASLAB
)
1739 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1741 if (flags
& VDD_DTL
)
1742 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1744 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1748 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1752 for (c
= 0; c
< vd
->vdev_children
; c
++)
1753 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1755 if (vd
->vdev_ops
->vdev_op_leaf
)
1756 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1762 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1763 * the vdev has less than perfect replication. There are four kinds of DTL:
1765 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1767 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1769 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1770 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1771 * txgs that was scrubbed.
1773 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1774 * persistent errors or just some device being offline.
1775 * Unlike the other three, the DTL_OUTAGE map is not generally
1776 * maintained; it's only computed when needed, typically to
1777 * determine whether a device can be detached.
1779 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1780 * either has the data or it doesn't.
1782 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1783 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1784 * if any child is less than fully replicated, then so is its parent.
1785 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1786 * comprising only those txgs which appear in 'maxfaults' or more children;
1787 * those are the txgs we don't have enough replication to read. For example,
1788 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1789 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1790 * two child DTL_MISSING maps.
1792 * It should be clear from the above that to compute the DTLs and outage maps
1793 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1794 * Therefore, that is all we keep on disk. When loading the pool, or after
1795 * a configuration change, we generate all other DTLs from first principles.
1798 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1800 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1802 ASSERT(t
< DTL_TYPES
);
1803 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1804 ASSERT(spa_writeable(vd
->vdev_spa
));
1806 mutex_enter(rt
->rt_lock
);
1807 if (!range_tree_contains(rt
, txg
, size
))
1808 range_tree_add(rt
, txg
, size
);
1809 mutex_exit(rt
->rt_lock
);
1813 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1815 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1816 boolean_t dirty
= B_FALSE
;
1818 ASSERT(t
< DTL_TYPES
);
1819 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1821 mutex_enter(rt
->rt_lock
);
1822 if (range_tree_space(rt
) != 0)
1823 dirty
= range_tree_contains(rt
, txg
, size
);
1824 mutex_exit(rt
->rt_lock
);
1830 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1832 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1835 mutex_enter(rt
->rt_lock
);
1836 empty
= (range_tree_space(rt
) == 0);
1837 mutex_exit(rt
->rt_lock
);
1843 * Returns B_TRUE if vdev determines offset needs to be resilvered.
1846 vdev_dtl_need_resilver(vdev_t
*vd
, uint64_t offset
, size_t psize
)
1848 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1850 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
1851 vd
->vdev_ops
->vdev_op_leaf
)
1854 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, offset
, psize
));
1858 * Returns the lowest txg in the DTL range.
1861 vdev_dtl_min(vdev_t
*vd
)
1865 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1866 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1867 ASSERT0(vd
->vdev_children
);
1869 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1870 return (rs
->rs_start
- 1);
1874 * Returns the highest txg in the DTL.
1877 vdev_dtl_max(vdev_t
*vd
)
1881 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1882 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1883 ASSERT0(vd
->vdev_children
);
1885 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1886 return (rs
->rs_end
);
1890 * Determine if a resilvering vdev should remove any DTL entries from
1891 * its range. If the vdev was resilvering for the entire duration of the
1892 * scan then it should excise that range from its DTLs. Otherwise, this
1893 * vdev is considered partially resilvered and should leave its DTL
1894 * entries intact. The comment in vdev_dtl_reassess() describes how we
1898 vdev_dtl_should_excise(vdev_t
*vd
)
1900 spa_t
*spa
= vd
->vdev_spa
;
1901 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1903 ASSERT0(scn
->scn_phys
.scn_errors
);
1904 ASSERT0(vd
->vdev_children
);
1906 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
1909 if (vd
->vdev_resilver_txg
== 0 ||
1910 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1914 * When a resilver is initiated the scan will assign the scn_max_txg
1915 * value to the highest txg value that exists in all DTLs. If this
1916 * device's max DTL is not part of this scan (i.e. it is not in
1917 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1920 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1921 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1922 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1923 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1930 * Reassess DTLs after a config change or scrub completion.
1933 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1935 spa_t
*spa
= vd
->vdev_spa
;
1939 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1941 for (c
= 0; c
< vd
->vdev_children
; c
++)
1942 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1943 scrub_txg
, scrub_done
);
1945 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1948 if (vd
->vdev_ops
->vdev_op_leaf
) {
1949 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1951 mutex_enter(&vd
->vdev_dtl_lock
);
1954 * If we've completed a scan cleanly then determine
1955 * if this vdev should remove any DTLs. We only want to
1956 * excise regions on vdevs that were available during
1957 * the entire duration of this scan.
1959 if (scrub_txg
!= 0 &&
1960 (spa
->spa_scrub_started
||
1961 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1962 vdev_dtl_should_excise(vd
)) {
1964 * We completed a scrub up to scrub_txg. If we
1965 * did it without rebooting, then the scrub dtl
1966 * will be valid, so excise the old region and
1967 * fold in the scrub dtl. Otherwise, leave the
1968 * dtl as-is if there was an error.
1970 * There's little trick here: to excise the beginning
1971 * of the DTL_MISSING map, we put it into a reference
1972 * tree and then add a segment with refcnt -1 that
1973 * covers the range [0, scrub_txg). This means
1974 * that each txg in that range has refcnt -1 or 0.
1975 * We then add DTL_SCRUB with a refcnt of 2, so that
1976 * entries in the range [0, scrub_txg) will have a
1977 * positive refcnt -- either 1 or 2. We then convert
1978 * the reference tree into the new DTL_MISSING map.
1980 space_reftree_create(&reftree
);
1981 space_reftree_add_map(&reftree
,
1982 vd
->vdev_dtl
[DTL_MISSING
], 1);
1983 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1984 space_reftree_add_map(&reftree
,
1985 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1986 space_reftree_generate_map(&reftree
,
1987 vd
->vdev_dtl
[DTL_MISSING
], 1);
1988 space_reftree_destroy(&reftree
);
1990 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1991 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1992 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1994 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1995 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1996 if (!vdev_readable(vd
))
1997 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1999 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2000 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
2003 * If the vdev was resilvering and no longer has any
2004 * DTLs then reset its resilvering flag and dirty
2005 * the top level so that we persist the change.
2007 if (vd
->vdev_resilver_txg
!= 0 &&
2008 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
2009 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0) {
2010 vd
->vdev_resilver_txg
= 0;
2011 vdev_config_dirty(vd
->vdev_top
);
2014 mutex_exit(&vd
->vdev_dtl_lock
);
2017 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
2021 mutex_enter(&vd
->vdev_dtl_lock
);
2022 for (t
= 0; t
< DTL_TYPES
; t
++) {
2025 /* account for child's outage in parent's missing map */
2026 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
2028 continue; /* leaf vdevs only */
2029 if (t
== DTL_PARTIAL
)
2030 minref
= 1; /* i.e. non-zero */
2031 else if (vd
->vdev_nparity
!= 0)
2032 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
2034 minref
= vd
->vdev_children
; /* any kind of mirror */
2035 space_reftree_create(&reftree
);
2036 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2037 vdev_t
*cvd
= vd
->vdev_child
[c
];
2038 mutex_enter(&cvd
->vdev_dtl_lock
);
2039 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
2040 mutex_exit(&cvd
->vdev_dtl_lock
);
2042 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
2043 space_reftree_destroy(&reftree
);
2045 mutex_exit(&vd
->vdev_dtl_lock
);
2049 vdev_dtl_load(vdev_t
*vd
)
2051 spa_t
*spa
= vd
->vdev_spa
;
2052 objset_t
*mos
= spa
->spa_meta_objset
;
2056 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2057 ASSERT(!vd
->vdev_ishole
);
2059 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2060 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
2063 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2065 mutex_enter(&vd
->vdev_dtl_lock
);
2068 * Now that we've opened the space_map we need to update
2071 space_map_update(vd
->vdev_dtl_sm
);
2073 error
= space_map_load(vd
->vdev_dtl_sm
,
2074 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2075 mutex_exit(&vd
->vdev_dtl_lock
);
2080 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2081 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2090 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2092 spa_t
*spa
= vd
->vdev_spa
;
2094 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2095 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2100 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2102 spa_t
*spa
= vd
->vdev_spa
;
2103 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2104 DMU_OT_NONE
, 0, tx
);
2107 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2114 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2118 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2119 vd
->vdev_ops
!= &vdev_missing_ops
&&
2120 vd
->vdev_ops
!= &vdev_root_ops
&&
2121 !vd
->vdev_top
->vdev_removing
) {
2122 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2123 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2125 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2126 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2129 for (i
= 0; i
< vd
->vdev_children
; i
++) {
2130 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2135 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2137 spa_t
*spa
= vd
->vdev_spa
;
2138 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2139 objset_t
*mos
= spa
->spa_meta_objset
;
2140 range_tree_t
*rtsync
;
2143 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2145 ASSERT(!vd
->vdev_ishole
);
2146 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2148 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2150 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2151 mutex_enter(&vd
->vdev_dtl_lock
);
2152 space_map_free(vd
->vdev_dtl_sm
, tx
);
2153 space_map_close(vd
->vdev_dtl_sm
);
2154 vd
->vdev_dtl_sm
= NULL
;
2155 mutex_exit(&vd
->vdev_dtl_lock
);
2158 * We only destroy the leaf ZAP for detached leaves or for
2159 * removed log devices. Removed data devices handle leaf ZAP
2160 * cleanup later, once cancellation is no longer possible.
2162 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2163 vd
->vdev_top
->vdev_islog
)) {
2164 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2165 vd
->vdev_leaf_zap
= 0;
2172 if (vd
->vdev_dtl_sm
== NULL
) {
2173 uint64_t new_object
;
2175 new_object
= space_map_alloc(mos
, tx
);
2176 VERIFY3U(new_object
, !=, 0);
2178 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2179 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2180 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2183 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2185 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2187 mutex_enter(&rtlock
);
2189 mutex_enter(&vd
->vdev_dtl_lock
);
2190 range_tree_walk(rt
, range_tree_add
, rtsync
);
2191 mutex_exit(&vd
->vdev_dtl_lock
);
2193 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2194 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2195 range_tree_vacate(rtsync
, NULL
, NULL
);
2197 range_tree_destroy(rtsync
);
2199 mutex_exit(&rtlock
);
2200 mutex_destroy(&rtlock
);
2203 * If the object for the space map has changed then dirty
2204 * the top level so that we update the config.
2206 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2207 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2208 "new object %llu", txg
, spa_name(spa
), object
,
2209 space_map_object(vd
->vdev_dtl_sm
));
2210 vdev_config_dirty(vd
->vdev_top
);
2215 mutex_enter(&vd
->vdev_dtl_lock
);
2216 space_map_update(vd
->vdev_dtl_sm
);
2217 mutex_exit(&vd
->vdev_dtl_lock
);
2221 * Determine whether the specified vdev can be offlined/detached/removed
2222 * without losing data.
2225 vdev_dtl_required(vdev_t
*vd
)
2227 spa_t
*spa
= vd
->vdev_spa
;
2228 vdev_t
*tvd
= vd
->vdev_top
;
2229 uint8_t cant_read
= vd
->vdev_cant_read
;
2232 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2234 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2238 * Temporarily mark the device as unreadable, and then determine
2239 * whether this results in any DTL outages in the top-level vdev.
2240 * If not, we can safely offline/detach/remove the device.
2242 vd
->vdev_cant_read
= B_TRUE
;
2243 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2244 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2245 vd
->vdev_cant_read
= cant_read
;
2246 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2248 if (!required
&& zio_injection_enabled
)
2249 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2255 * Determine if resilver is needed, and if so the txg range.
2258 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2260 boolean_t needed
= B_FALSE
;
2261 uint64_t thismin
= UINT64_MAX
;
2262 uint64_t thismax
= 0;
2265 if (vd
->vdev_children
== 0) {
2266 mutex_enter(&vd
->vdev_dtl_lock
);
2267 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2268 vdev_writeable(vd
)) {
2270 thismin
= vdev_dtl_min(vd
);
2271 thismax
= vdev_dtl_max(vd
);
2274 mutex_exit(&vd
->vdev_dtl_lock
);
2276 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2277 vdev_t
*cvd
= vd
->vdev_child
[c
];
2278 uint64_t cmin
, cmax
;
2280 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2281 thismin
= MIN(thismin
, cmin
);
2282 thismax
= MAX(thismax
, cmax
);
2288 if (needed
&& minp
) {
2296 vdev_load(vdev_t
*vd
)
2301 * Recursively load all children.
2303 for (c
= 0; c
< vd
->vdev_children
; c
++)
2304 vdev_load(vd
->vdev_child
[c
]);
2307 * If this is a top-level vdev, initialize its metaslabs.
2309 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2310 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2311 vdev_metaslab_init(vd
, 0) != 0))
2312 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2313 VDEV_AUX_CORRUPT_DATA
);
2315 * If this is a leaf vdev, load its DTL.
2317 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2318 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2319 VDEV_AUX_CORRUPT_DATA
);
2323 * The special vdev case is used for hot spares and l2cache devices. Its
2324 * sole purpose it to set the vdev state for the associated vdev. To do this,
2325 * we make sure that we can open the underlying device, then try to read the
2326 * label, and make sure that the label is sane and that it hasn't been
2327 * repurposed to another pool.
2330 vdev_validate_aux(vdev_t
*vd
)
2333 uint64_t guid
, version
;
2336 if (!vdev_readable(vd
))
2339 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2340 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2341 VDEV_AUX_CORRUPT_DATA
);
2345 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2346 !SPA_VERSION_IS_SUPPORTED(version
) ||
2347 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2348 guid
!= vd
->vdev_guid
||
2349 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2350 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2351 VDEV_AUX_CORRUPT_DATA
);
2357 * We don't actually check the pool state here. If it's in fact in
2358 * use by another pool, we update this fact on the fly when requested.
2365 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2367 spa_t
*spa
= vd
->vdev_spa
;
2368 objset_t
*mos
= spa
->spa_meta_objset
;
2372 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2373 ASSERT(vd
== vd
->vdev_top
);
2374 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2376 if (vd
->vdev_ms
!= NULL
) {
2377 metaslab_group_t
*mg
= vd
->vdev_mg
;
2379 metaslab_group_histogram_verify(mg
);
2380 metaslab_class_histogram_verify(mg
->mg_class
);
2382 for (m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2383 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2385 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2388 mutex_enter(&msp
->ms_lock
);
2390 * If the metaslab was not loaded when the vdev
2391 * was removed then the histogram accounting may
2392 * not be accurate. Update the histogram information
2393 * here so that we ensure that the metaslab group
2394 * and metaslab class are up-to-date.
2396 metaslab_group_histogram_remove(mg
, msp
);
2398 VERIFY0(space_map_allocated(msp
->ms_sm
));
2399 space_map_free(msp
->ms_sm
, tx
);
2400 space_map_close(msp
->ms_sm
);
2402 mutex_exit(&msp
->ms_lock
);
2405 metaslab_group_histogram_verify(mg
);
2406 metaslab_class_histogram_verify(mg
->mg_class
);
2407 for (i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2408 ASSERT0(mg
->mg_histogram
[i
]);
2412 if (vd
->vdev_ms_array
) {
2413 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2414 vd
->vdev_ms_array
= 0;
2417 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2418 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2419 vd
->vdev_top_zap
= 0;
2425 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2428 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2430 ASSERT(!vd
->vdev_ishole
);
2432 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
2433 metaslab_sync_done(msp
, txg
);
2436 metaslab_sync_reassess(vd
->vdev_mg
);
2440 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2442 spa_t
*spa
= vd
->vdev_spa
;
2447 ASSERT(!vd
->vdev_ishole
);
2449 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2450 ASSERT(vd
== vd
->vdev_top
);
2451 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2452 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2453 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2454 ASSERT(vd
->vdev_ms_array
!= 0);
2455 vdev_config_dirty(vd
);
2460 * Remove the metadata associated with this vdev once it's empty.
2462 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2463 vdev_remove(vd
, txg
);
2465 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2466 metaslab_sync(msp
, txg
);
2467 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2470 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2471 vdev_dtl_sync(lvd
, txg
);
2473 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2477 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2479 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2483 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2484 * not be opened, and no I/O is attempted.
2487 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2491 spa_vdev_state_enter(spa
, SCL_NONE
);
2493 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2494 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2496 if (!vd
->vdev_ops
->vdev_op_leaf
)
2497 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2502 * If user did a 'zpool offline -f' then make the fault persist across
2505 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
2507 * There are two kinds of forced faults: temporary and
2508 * persistent. Temporary faults go away at pool import, while
2509 * persistent faults stay set. Both types of faults can be
2510 * cleared with a zpool clear.
2512 * We tell if a vdev is persistently faulted by looking at the
2513 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
2514 * import then it's a persistent fault. Otherwise, it's
2515 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
2516 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
2517 * tells vdev_config_generate() (which gets run later) to set
2518 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
2520 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
2521 vd
->vdev_tmpoffline
= B_FALSE
;
2522 aux
= VDEV_AUX_EXTERNAL
;
2524 vd
->vdev_tmpoffline
= B_TRUE
;
2528 * We don't directly use the aux state here, but if we do a
2529 * vdev_reopen(), we need this value to be present to remember why we
2532 vd
->vdev_label_aux
= aux
;
2535 * Faulted state takes precedence over degraded.
2537 vd
->vdev_delayed_close
= B_FALSE
;
2538 vd
->vdev_faulted
= 1ULL;
2539 vd
->vdev_degraded
= 0ULL;
2540 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2543 * If this device has the only valid copy of the data, then
2544 * back off and simply mark the vdev as degraded instead.
2546 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2547 vd
->vdev_degraded
= 1ULL;
2548 vd
->vdev_faulted
= 0ULL;
2551 * If we reopen the device and it's not dead, only then do we
2556 if (vdev_readable(vd
))
2557 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2560 return (spa_vdev_state_exit(spa
, vd
, 0));
2564 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2565 * user that something is wrong. The vdev continues to operate as normal as far
2566 * as I/O is concerned.
2569 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2573 spa_vdev_state_enter(spa
, SCL_NONE
);
2575 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2576 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2578 if (!vd
->vdev_ops
->vdev_op_leaf
)
2579 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2582 * If the vdev is already faulted, then don't do anything.
2584 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2585 return (spa_vdev_state_exit(spa
, NULL
, 0));
2587 vd
->vdev_degraded
= 1ULL;
2588 if (!vdev_is_dead(vd
))
2589 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2592 return (spa_vdev_state_exit(spa
, vd
, 0));
2596 * Online the given vdev.
2598 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2599 * spare device should be detached when the device finishes resilvering.
2600 * Second, the online should be treated like a 'test' online case, so no FMA
2601 * events are generated if the device fails to open.
2604 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2606 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2607 boolean_t wasoffline
;
2608 vdev_state_t oldstate
;
2610 spa_vdev_state_enter(spa
, SCL_NONE
);
2612 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2613 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2615 if (!vd
->vdev_ops
->vdev_op_leaf
)
2616 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2618 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
2619 oldstate
= vd
->vdev_state
;
2622 vd
->vdev_offline
= B_FALSE
;
2623 vd
->vdev_tmpoffline
= B_FALSE
;
2624 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2625 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2627 /* XXX - L2ARC 1.0 does not support expansion */
2628 if (!vd
->vdev_aux
) {
2629 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2630 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2634 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2636 if (!vd
->vdev_aux
) {
2637 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2638 pvd
->vdev_expanding
= B_FALSE
;
2642 *newstate
= vd
->vdev_state
;
2643 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2644 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2645 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2646 vd
->vdev_parent
->vdev_child
[0] == vd
)
2647 vd
->vdev_unspare
= B_TRUE
;
2649 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2651 /* XXX - L2ARC 1.0 does not support expansion */
2653 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2654 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2658 (oldstate
< VDEV_STATE_DEGRADED
&&
2659 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
2660 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2662 return (spa_vdev_state_exit(spa
, vd
, 0));
2666 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2670 uint64_t generation
;
2671 metaslab_group_t
*mg
;
2674 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2676 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2677 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2679 if (!vd
->vdev_ops
->vdev_op_leaf
)
2680 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2684 generation
= spa
->spa_config_generation
+ 1;
2687 * If the device isn't already offline, try to offline it.
2689 if (!vd
->vdev_offline
) {
2691 * If this device has the only valid copy of some data,
2692 * don't allow it to be offlined. Log devices are always
2695 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2696 vdev_dtl_required(vd
))
2697 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2700 * If the top-level is a slog and it has had allocations
2701 * then proceed. We check that the vdev's metaslab group
2702 * is not NULL since it's possible that we may have just
2703 * added this vdev but not yet initialized its metaslabs.
2705 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2707 * Prevent any future allocations.
2709 metaslab_group_passivate(mg
);
2710 (void) spa_vdev_state_exit(spa
, vd
, 0);
2712 error
= spa_offline_log(spa
);
2714 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2717 * Check to see if the config has changed.
2719 if (error
|| generation
!= spa
->spa_config_generation
) {
2720 metaslab_group_activate(mg
);
2722 return (spa_vdev_state_exit(spa
,
2724 (void) spa_vdev_state_exit(spa
, vd
, 0);
2727 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2731 * Offline this device and reopen its top-level vdev.
2732 * If the top-level vdev is a log device then just offline
2733 * it. Otherwise, if this action results in the top-level
2734 * vdev becoming unusable, undo it and fail the request.
2736 vd
->vdev_offline
= B_TRUE
;
2739 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2740 vdev_is_dead(tvd
)) {
2741 vd
->vdev_offline
= B_FALSE
;
2743 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2747 * Add the device back into the metaslab rotor so that
2748 * once we online the device it's open for business.
2750 if (tvd
->vdev_islog
&& mg
!= NULL
)
2751 metaslab_group_activate(mg
);
2754 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2756 return (spa_vdev_state_exit(spa
, vd
, 0));
2760 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2764 mutex_enter(&spa
->spa_vdev_top_lock
);
2765 error
= vdev_offline_locked(spa
, guid
, flags
);
2766 mutex_exit(&spa
->spa_vdev_top_lock
);
2772 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2773 * vdev_offline(), we assume the spa config is locked. We also clear all
2774 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2777 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2779 vdev_t
*rvd
= spa
->spa_root_vdev
;
2782 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2787 vd
->vdev_stat
.vs_read_errors
= 0;
2788 vd
->vdev_stat
.vs_write_errors
= 0;
2789 vd
->vdev_stat
.vs_checksum_errors
= 0;
2791 for (c
= 0; c
< vd
->vdev_children
; c
++)
2792 vdev_clear(spa
, vd
->vdev_child
[c
]);
2795 * If we're in the FAULTED state or have experienced failed I/O, then
2796 * clear the persistent state and attempt to reopen the device. We
2797 * also mark the vdev config dirty, so that the new faulted state is
2798 * written out to disk.
2800 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2801 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2803 * When reopening in response to a clear event, it may be due to
2804 * a fmadm repair request. In this case, if the device is
2805 * still broken, we want to still post the ereport again.
2807 vd
->vdev_forcefault
= B_TRUE
;
2809 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2810 vd
->vdev_cant_read
= B_FALSE
;
2811 vd
->vdev_cant_write
= B_FALSE
;
2812 vd
->vdev_stat
.vs_aux
= 0;
2814 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2816 vd
->vdev_forcefault
= B_FALSE
;
2818 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2819 vdev_state_dirty(vd
->vdev_top
);
2821 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2822 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2824 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2828 * When clearing a FMA-diagnosed fault, we always want to
2829 * unspare the device, as we assume that the original spare was
2830 * done in response to the FMA fault.
2832 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2833 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2834 vd
->vdev_parent
->vdev_child
[0] == vd
)
2835 vd
->vdev_unspare
= B_TRUE
;
2839 vdev_is_dead(vdev_t
*vd
)
2842 * Holes and missing devices are always considered "dead".
2843 * This simplifies the code since we don't have to check for
2844 * these types of devices in the various code paths.
2845 * Instead we rely on the fact that we skip over dead devices
2846 * before issuing I/O to them.
2848 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2849 vd
->vdev_ops
== &vdev_missing_ops
);
2853 vdev_readable(vdev_t
*vd
)
2855 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2859 vdev_writeable(vdev_t
*vd
)
2861 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2865 vdev_allocatable(vdev_t
*vd
)
2867 uint64_t state
= vd
->vdev_state
;
2870 * We currently allow allocations from vdevs which may be in the
2871 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2872 * fails to reopen then we'll catch it later when we're holding
2873 * the proper locks. Note that we have to get the vdev state
2874 * in a local variable because although it changes atomically,
2875 * we're asking two separate questions about it.
2877 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2878 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
&&
2879 vd
->vdev_mg
->mg_initialized
);
2883 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2885 ASSERT(zio
->io_vd
== vd
);
2887 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2890 if (zio
->io_type
== ZIO_TYPE_READ
)
2891 return (!vd
->vdev_cant_read
);
2893 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2894 return (!vd
->vdev_cant_write
);
2900 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
2903 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2904 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2905 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2908 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2912 * Get extended stats
2915 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
2918 for (t
= 0; t
< ZIO_TYPES
; t
++) {
2919 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
2920 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
2922 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
2923 vsx
->vsx_total_histo
[t
][b
] +=
2924 cvsx
->vsx_total_histo
[t
][b
];
2928 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
2929 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
2930 vsx
->vsx_queue_histo
[t
][b
] +=
2931 cvsx
->vsx_queue_histo
[t
][b
];
2933 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
2934 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
2936 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
2937 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
2939 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
2940 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
2946 * Get statistics for the given vdev.
2949 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2953 * If we're getting stats on the root vdev, aggregate the I/O counts
2954 * over all top-level vdevs (i.e. the direct children of the root).
2956 if (!vd
->vdev_ops
->vdev_op_leaf
) {
2958 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
2959 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
2962 memset(vsx
, 0, sizeof (*vsx
));
2964 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2965 vdev_t
*cvd
= vd
->vdev_child
[c
];
2966 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2967 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
2969 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
2971 vdev_get_child_stat(cvd
, vs
, cvs
);
2973 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
2978 * We're a leaf. Just copy our ZIO active queue stats in. The
2979 * other leaf stats are updated in vdev_stat_update().
2984 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
2986 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
2987 vsx
->vsx_active_queue
[t
] =
2988 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
2989 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
2990 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
2996 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
2998 vdev_t
*tvd
= vd
->vdev_top
;
2999 mutex_enter(&vd
->vdev_stat_lock
);
3001 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
3002 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
3003 vs
->vs_state
= vd
->vdev_state
;
3004 vs
->vs_rsize
= vdev_get_min_asize(vd
);
3005 if (vd
->vdev_ops
->vdev_op_leaf
)
3006 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
3007 VDEV_LABEL_END_SIZE
;
3009 * Report expandable space on top-level, non-auxillary devices
3010 * only. The expandable space is reported in terms of metaslab
3011 * sized units since that determines how much space the pool
3014 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
3015 vs
->vs_esize
= P2ALIGN(
3016 vd
->vdev_max_asize
- vd
->vdev_asize
,
3017 1ULL << tvd
->vdev_ms_shift
);
3019 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
3020 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
3022 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
3026 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
3027 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
3028 mutex_exit(&vd
->vdev_stat_lock
);
3032 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
3034 return (vdev_get_stats_ex(vd
, vs
, NULL
));
3038 vdev_clear_stats(vdev_t
*vd
)
3040 mutex_enter(&vd
->vdev_stat_lock
);
3041 vd
->vdev_stat
.vs_space
= 0;
3042 vd
->vdev_stat
.vs_dspace
= 0;
3043 vd
->vdev_stat
.vs_alloc
= 0;
3044 mutex_exit(&vd
->vdev_stat_lock
);
3048 vdev_scan_stat_init(vdev_t
*vd
)
3050 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3053 for (c
= 0; c
< vd
->vdev_children
; c
++)
3054 vdev_scan_stat_init(vd
->vdev_child
[c
]);
3056 mutex_enter(&vd
->vdev_stat_lock
);
3057 vs
->vs_scan_processed
= 0;
3058 mutex_exit(&vd
->vdev_stat_lock
);
3062 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
3064 spa_t
*spa
= zio
->io_spa
;
3065 vdev_t
*rvd
= spa
->spa_root_vdev
;
3066 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
3068 uint64_t txg
= zio
->io_txg
;
3069 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3070 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
3071 zio_type_t type
= zio
->io_type
;
3072 int flags
= zio
->io_flags
;
3075 * If this i/o is a gang leader, it didn't do any actual work.
3077 if (zio
->io_gang_tree
)
3080 if (zio
->io_error
== 0) {
3082 * If this is a root i/o, don't count it -- we've already
3083 * counted the top-level vdevs, and vdev_get_stats() will
3084 * aggregate them when asked. This reduces contention on
3085 * the root vdev_stat_lock and implicitly handles blocks
3086 * that compress away to holes, for which there is no i/o.
3087 * (Holes never create vdev children, so all the counters
3088 * remain zero, which is what we want.)
3090 * Note: this only applies to successful i/o (io_error == 0)
3091 * because unlike i/o counts, errors are not additive.
3092 * When reading a ditto block, for example, failure of
3093 * one top-level vdev does not imply a root-level error.
3098 ASSERT(vd
== zio
->io_vd
);
3100 if (flags
& ZIO_FLAG_IO_BYPASS
)
3103 mutex_enter(&vd
->vdev_stat_lock
);
3105 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3106 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3107 dsl_scan_phys_t
*scn_phys
=
3108 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3109 uint64_t *processed
= &scn_phys
->scn_processed
;
3112 if (vd
->vdev_ops
->vdev_op_leaf
)
3113 atomic_add_64(processed
, psize
);
3114 vs
->vs_scan_processed
+= psize
;
3117 if (flags
& ZIO_FLAG_SELF_HEAL
)
3118 vs
->vs_self_healed
+= psize
;
3122 * The bytes/ops/histograms are recorded at the leaf level and
3123 * aggregated into the higher level vdevs in vdev_get_stats().
3125 if (vd
->vdev_ops
->vdev_op_leaf
&&
3126 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3129 vs
->vs_bytes
[type
] += psize
;
3131 if (flags
& ZIO_FLAG_DELEGATED
) {
3132 vsx
->vsx_agg_histo
[zio
->io_priority
]
3133 [RQ_HISTO(zio
->io_size
)]++;
3135 vsx
->vsx_ind_histo
[zio
->io_priority
]
3136 [RQ_HISTO(zio
->io_size
)]++;
3139 if (zio
->io_delta
&& zio
->io_delay
) {
3140 vsx
->vsx_queue_histo
[zio
->io_priority
]
3141 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3142 vsx
->vsx_disk_histo
[type
]
3143 [L_HISTO(zio
->io_delay
)]++;
3144 vsx
->vsx_total_histo
[type
]
3145 [L_HISTO(zio
->io_delta
)]++;
3149 mutex_exit(&vd
->vdev_stat_lock
);
3153 if (flags
& ZIO_FLAG_SPECULATIVE
)
3157 * If this is an I/O error that is going to be retried, then ignore the
3158 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3159 * hard errors, when in reality they can happen for any number of
3160 * innocuous reasons (bus resets, MPxIO link failure, etc).
3162 if (zio
->io_error
== EIO
&&
3163 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3167 * Intent logs writes won't propagate their error to the root
3168 * I/O so don't mark these types of failures as pool-level
3171 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3174 mutex_enter(&vd
->vdev_stat_lock
);
3175 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3176 if (zio
->io_error
== ECKSUM
)
3177 vs
->vs_checksum_errors
++;
3179 vs
->vs_read_errors
++;
3181 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3182 vs
->vs_write_errors
++;
3183 mutex_exit(&vd
->vdev_stat_lock
);
3185 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3186 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3187 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3188 spa
->spa_claiming
)) {
3190 * This is either a normal write (not a repair), or it's
3191 * a repair induced by the scrub thread, or it's a repair
3192 * made by zil_claim() during spa_load() in the first txg.
3193 * In the normal case, we commit the DTL change in the same
3194 * txg as the block was born. In the scrub-induced repair
3195 * case, we know that scrubs run in first-pass syncing context,
3196 * so we commit the DTL change in spa_syncing_txg(spa).
3197 * In the zil_claim() case, we commit in spa_first_txg(spa).
3199 * We currently do not make DTL entries for failed spontaneous
3200 * self-healing writes triggered by normal (non-scrubbing)
3201 * reads, because we have no transactional context in which to
3202 * do so -- and it's not clear that it'd be desirable anyway.
3204 if (vd
->vdev_ops
->vdev_op_leaf
) {
3205 uint64_t commit_txg
= txg
;
3206 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3207 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3208 ASSERT(spa_sync_pass(spa
) == 1);
3209 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3210 commit_txg
= spa_syncing_txg(spa
);
3211 } else if (spa
->spa_claiming
) {
3212 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3213 commit_txg
= spa_first_txg(spa
);
3215 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3216 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3218 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3219 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3220 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3223 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3228 * Update the in-core space usage stats for this vdev, its metaslab class,
3229 * and the root vdev.
3232 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3233 int64_t space_delta
)
3235 int64_t dspace_delta
= space_delta
;
3236 spa_t
*spa
= vd
->vdev_spa
;
3237 vdev_t
*rvd
= spa
->spa_root_vdev
;
3238 metaslab_group_t
*mg
= vd
->vdev_mg
;
3239 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3241 ASSERT(vd
== vd
->vdev_top
);
3244 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3245 * factor. We must calculate this here and not at the root vdev
3246 * because the root vdev's psize-to-asize is simply the max of its
3247 * childrens', thus not accurate enough for us.
3249 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3250 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3251 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3252 vd
->vdev_deflate_ratio
;
3254 mutex_enter(&vd
->vdev_stat_lock
);
3255 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3256 vd
->vdev_stat
.vs_space
+= space_delta
;
3257 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3258 mutex_exit(&vd
->vdev_stat_lock
);
3260 if (mc
== spa_normal_class(spa
)) {
3261 mutex_enter(&rvd
->vdev_stat_lock
);
3262 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3263 rvd
->vdev_stat
.vs_space
+= space_delta
;
3264 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3265 mutex_exit(&rvd
->vdev_stat_lock
);
3269 ASSERT(rvd
== vd
->vdev_parent
);
3270 ASSERT(vd
->vdev_ms_count
!= 0);
3272 metaslab_class_space_update(mc
,
3273 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3278 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3279 * so that it will be written out next time the vdev configuration is synced.
3280 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3283 vdev_config_dirty(vdev_t
*vd
)
3285 spa_t
*spa
= vd
->vdev_spa
;
3286 vdev_t
*rvd
= spa
->spa_root_vdev
;
3289 ASSERT(spa_writeable(spa
));
3292 * If this is an aux vdev (as with l2cache and spare devices), then we
3293 * update the vdev config manually and set the sync flag.
3295 if (vd
->vdev_aux
!= NULL
) {
3296 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3300 for (c
= 0; c
< sav
->sav_count
; c
++) {
3301 if (sav
->sav_vdevs
[c
] == vd
)
3305 if (c
== sav
->sav_count
) {
3307 * We're being removed. There's nothing more to do.
3309 ASSERT(sav
->sav_sync
== B_TRUE
);
3313 sav
->sav_sync
= B_TRUE
;
3315 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3316 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3317 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3318 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3324 * Setting the nvlist in the middle if the array is a little
3325 * sketchy, but it will work.
3327 nvlist_free(aux
[c
]);
3328 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3334 * The dirty list is protected by the SCL_CONFIG lock. The caller
3335 * must either hold SCL_CONFIG as writer, or must be the sync thread
3336 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3337 * so this is sufficient to ensure mutual exclusion.
3339 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3340 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3341 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3344 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3345 vdev_config_dirty(rvd
->vdev_child
[c
]);
3347 ASSERT(vd
== vd
->vdev_top
);
3349 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3351 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3356 vdev_config_clean(vdev_t
*vd
)
3358 spa_t
*spa
= vd
->vdev_spa
;
3360 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3361 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3362 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3364 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3365 list_remove(&spa
->spa_config_dirty_list
, vd
);
3369 * Mark a top-level vdev's state as dirty, so that the next pass of
3370 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3371 * the state changes from larger config changes because they require
3372 * much less locking, and are often needed for administrative actions.
3375 vdev_state_dirty(vdev_t
*vd
)
3377 spa_t
*spa
= vd
->vdev_spa
;
3379 ASSERT(spa_writeable(spa
));
3380 ASSERT(vd
== vd
->vdev_top
);
3383 * The state list is protected by the SCL_STATE lock. The caller
3384 * must either hold SCL_STATE as writer, or must be the sync thread
3385 * (which holds SCL_STATE as reader). There's only one sync thread,
3386 * so this is sufficient to ensure mutual exclusion.
3388 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3389 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3390 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3392 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3393 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3397 vdev_state_clean(vdev_t
*vd
)
3399 spa_t
*spa
= vd
->vdev_spa
;
3401 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3402 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3403 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3405 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3406 list_remove(&spa
->spa_state_dirty_list
, vd
);
3410 * Propagate vdev state up from children to parent.
3413 vdev_propagate_state(vdev_t
*vd
)
3415 spa_t
*spa
= vd
->vdev_spa
;
3416 vdev_t
*rvd
= spa
->spa_root_vdev
;
3417 int degraded
= 0, faulted
= 0;
3422 if (vd
->vdev_children
> 0) {
3423 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3424 child
= vd
->vdev_child
[c
];
3427 * Don't factor holes into the decision.
3429 if (child
->vdev_ishole
)
3432 if (!vdev_readable(child
) ||
3433 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3435 * Root special: if there is a top-level log
3436 * device, treat the root vdev as if it were
3439 if (child
->vdev_islog
&& vd
== rvd
)
3443 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3447 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3451 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3454 * Root special: if there is a top-level vdev that cannot be
3455 * opened due to corrupted metadata, then propagate the root
3456 * vdev's aux state as 'corrupt' rather than 'insufficient
3459 if (corrupted
&& vd
== rvd
&&
3460 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3461 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3462 VDEV_AUX_CORRUPT_DATA
);
3465 if (vd
->vdev_parent
)
3466 vdev_propagate_state(vd
->vdev_parent
);
3470 * Set a vdev's state. If this is during an open, we don't update the parent
3471 * state, because we're in the process of opening children depth-first.
3472 * Otherwise, we propagate the change to the parent.
3474 * If this routine places a device in a faulted state, an appropriate ereport is
3478 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3480 uint64_t save_state
;
3481 spa_t
*spa
= vd
->vdev_spa
;
3483 if (state
== vd
->vdev_state
) {
3485 * Since vdev_offline() code path is already in an offline
3486 * state we can miss a statechange event to OFFLINE. Check
3487 * the previous state to catch this condition.
3489 if (vd
->vdev_ops
->vdev_op_leaf
&&
3490 (state
== VDEV_STATE_OFFLINE
) &&
3491 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
3492 /* post an offline state change */
3493 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
3495 vd
->vdev_stat
.vs_aux
= aux
;
3499 save_state
= vd
->vdev_state
;
3501 vd
->vdev_state
= state
;
3502 vd
->vdev_stat
.vs_aux
= aux
;
3505 * If we are setting the vdev state to anything but an open state, then
3506 * always close the underlying device unless the device has requested
3507 * a delayed close (i.e. we're about to remove or fault the device).
3508 * Otherwise, we keep accessible but invalid devices open forever.
3509 * We don't call vdev_close() itself, because that implies some extra
3510 * checks (offline, etc) that we don't want here. This is limited to
3511 * leaf devices, because otherwise closing the device will affect other
3514 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3515 vd
->vdev_ops
->vdev_op_leaf
)
3516 vd
->vdev_ops
->vdev_op_close(vd
);
3518 if (vd
->vdev_removed
&&
3519 state
== VDEV_STATE_CANT_OPEN
&&
3520 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3522 * If the previous state is set to VDEV_STATE_REMOVED, then this
3523 * device was previously marked removed and someone attempted to
3524 * reopen it. If this failed due to a nonexistent device, then
3525 * keep the device in the REMOVED state. We also let this be if
3526 * it is one of our special test online cases, which is only
3527 * attempting to online the device and shouldn't generate an FMA
3530 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3531 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3532 } else if (state
== VDEV_STATE_REMOVED
) {
3533 vd
->vdev_removed
= B_TRUE
;
3534 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3536 * If we fail to open a vdev during an import or recovery, we
3537 * mark it as "not available", which signifies that it was
3538 * never there to begin with. Failure to open such a device
3539 * is not considered an error.
3541 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3542 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3543 vd
->vdev_ops
->vdev_op_leaf
)
3544 vd
->vdev_not_present
= 1;
3547 * Post the appropriate ereport. If the 'prevstate' field is
3548 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3549 * that this is part of a vdev_reopen(). In this case, we don't
3550 * want to post the ereport if the device was already in the
3551 * CANT_OPEN state beforehand.
3553 * If the 'checkremove' flag is set, then this is an attempt to
3554 * online the device in response to an insertion event. If we
3555 * hit this case, then we have detected an insertion event for a
3556 * faulted or offline device that wasn't in the removed state.
3557 * In this scenario, we don't post an ereport because we are
3558 * about to replace the device, or attempt an online with
3559 * vdev_forcefault, which will generate the fault for us.
3561 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3562 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3563 vd
!= spa
->spa_root_vdev
) {
3567 case VDEV_AUX_OPEN_FAILED
:
3568 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3570 case VDEV_AUX_CORRUPT_DATA
:
3571 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3573 case VDEV_AUX_NO_REPLICAS
:
3574 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3576 case VDEV_AUX_BAD_GUID_SUM
:
3577 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3579 case VDEV_AUX_TOO_SMALL
:
3580 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3582 case VDEV_AUX_BAD_LABEL
:
3583 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3585 case VDEV_AUX_BAD_ASHIFT
:
3586 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
3589 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3592 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3595 /* Erase any notion of persistent removed state */
3596 vd
->vdev_removed
= B_FALSE
;
3598 vd
->vdev_removed
= B_FALSE
;
3602 * Notify ZED of any significant state-change on a leaf vdev.
3605 if (vd
->vdev_ops
->vdev_op_leaf
) {
3606 /* preserve original state from a vdev_reopen() */
3607 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
3608 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
3609 (save_state
<= VDEV_STATE_CLOSED
))
3610 save_state
= vd
->vdev_prevstate
;
3612 /* filter out state change due to initial vdev_open */
3613 if (save_state
> VDEV_STATE_CLOSED
)
3614 zfs_post_state_change(spa
, vd
, save_state
);
3617 if (!isopen
&& vd
->vdev_parent
)
3618 vdev_propagate_state(vd
->vdev_parent
);
3622 * Check the vdev configuration to ensure that it's capable of supporting
3623 * a root pool. We do not support partial configuration.
3626 vdev_is_bootable(vdev_t
*vd
)
3628 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3629 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3631 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0)
3635 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3636 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3643 * Load the state from the original vdev tree (ovd) which
3644 * we've retrieved from the MOS config object. If the original
3645 * vdev was offline or faulted then we transfer that state to the
3646 * device in the current vdev tree (nvd).
3649 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3653 ASSERT(nvd
->vdev_top
->vdev_islog
);
3654 ASSERT(spa_config_held(nvd
->vdev_spa
,
3655 SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3656 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3658 for (c
= 0; c
< nvd
->vdev_children
; c
++)
3659 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3661 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3663 * Restore the persistent vdev state
3665 nvd
->vdev_offline
= ovd
->vdev_offline
;
3666 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3667 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3668 nvd
->vdev_removed
= ovd
->vdev_removed
;
3673 * Determine if a log device has valid content. If the vdev was
3674 * removed or faulted in the MOS config then we know that
3675 * the content on the log device has already been written to the pool.
3678 vdev_log_state_valid(vdev_t
*vd
)
3682 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3686 for (c
= 0; c
< vd
->vdev_children
; c
++)
3687 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3694 * Expand a vdev if possible.
3697 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3699 ASSERT(vd
->vdev_top
== vd
);
3700 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3702 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3703 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3704 vdev_config_dirty(vd
);
3712 vdev_split(vdev_t
*vd
)
3714 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3716 vdev_remove_child(pvd
, vd
);
3717 vdev_compact_children(pvd
);
3719 cvd
= pvd
->vdev_child
[0];
3720 if (pvd
->vdev_children
== 1) {
3721 vdev_remove_parent(cvd
);
3722 cvd
->vdev_splitting
= B_TRUE
;
3724 vdev_propagate_state(cvd
);
3728 vdev_deadman(vdev_t
*vd
)
3732 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3733 vdev_t
*cvd
= vd
->vdev_child
[c
];
3738 if (vd
->vdev_ops
->vdev_op_leaf
) {
3739 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3741 mutex_enter(&vq
->vq_lock
);
3742 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3743 spa_t
*spa
= vd
->vdev_spa
;
3748 * Look at the head of all the pending queues,
3749 * if any I/O has been outstanding for longer than
3750 * the spa_deadman_synctime we log a zevent.
3752 fio
= avl_first(&vq
->vq_active_tree
);
3753 delta
= gethrtime() - fio
->io_timestamp
;
3754 if (delta
> spa_deadman_synctime(spa
)) {
3755 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3756 "delta %lluns, last io %lluns",
3757 fio
->io_timestamp
, delta
,
3758 vq
->vq_io_complete_ts
);
3759 zfs_ereport_post(FM_EREPORT_ZFS_DELAY
,
3760 spa
, vd
, fio
, 0, 0);
3763 mutex_exit(&vq
->vq_lock
);
3767 #if defined(_KERNEL) && defined(HAVE_SPL)
3768 EXPORT_SYMBOL(vdev_fault
);
3769 EXPORT_SYMBOL(vdev_degrade
);
3770 EXPORT_SYMBOL(vdev_online
);
3771 EXPORT_SYMBOL(vdev_offline
);
3772 EXPORT_SYMBOL(vdev_clear
);
3774 module_param(metaslabs_per_vdev
, int, 0644);
3775 MODULE_PARM_DESC(metaslabs_per_vdev
,
3776 "Divide added vdev into approximately (but no more than) this number "