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, 2018 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Toomas Soome <tsoome@me.com>
28 * Copyright 2017 Joyent, Inc.
31 #include <sys/zfs_context.h>
32 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
35 #include <sys/bpobj.h>
37 #include <sys/dmu_tx.h>
38 #include <sys/dsl_dir.h>
39 #include <sys/vdev_impl.h>
40 #include <sys/uberblock_impl.h>
41 #include <sys/metaslab.h>
42 #include <sys/metaslab_impl.h>
43 #include <sys/space_map.h>
44 #include <sys/space_reftree.h>
47 #include <sys/fs/zfs.h>
50 #include <sys/dsl_scan.h>
54 * Virtual device management.
57 static vdev_ops_t
*vdev_ops_table
[] = {
71 /* maximum scrub/resilver I/O queue per leaf vdev */
72 int zfs_scrub_limit
= 10;
75 * When a vdev is added, it will be divided into approximately (but no
76 * more than) this number of metaslabs.
78 int metaslabs_per_vdev
= 200;
82 vdev_dbgmsg(vdev_t
*vd
, const char *fmt
, ...)
88 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
91 if (vd
->vdev_path
!= NULL
) {
92 zfs_dbgmsg("%s vdev '%s': %s", vd
->vdev_ops
->vdev_op_type
,
95 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
96 vd
->vdev_ops
->vdev_op_type
,
97 (u_longlong_t
)vd
->vdev_id
,
98 (u_longlong_t
)vd
->vdev_guid
, buf
);
103 * Given a vdev type, return the appropriate ops vector.
106 vdev_getops(const char *type
)
108 vdev_ops_t
*ops
, **opspp
;
110 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
111 if (strcmp(ops
->vdev_op_type
, type
) == 0)
118 * Default asize function: return the MAX of psize with the asize of
119 * all children. This is what's used by anything other than RAID-Z.
122 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
124 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
127 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
128 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
129 asize
= MAX(asize
, csize
);
136 * Get the minimum allocatable size. We define the allocatable size as
137 * the vdev's asize rounded to the nearest metaslab. This allows us to
138 * replace or attach devices which don't have the same physical size but
139 * can still satisfy the same number of allocations.
142 vdev_get_min_asize(vdev_t
*vd
)
144 vdev_t
*pvd
= vd
->vdev_parent
;
147 * If our parent is NULL (inactive spare or cache) or is the root,
148 * just return our own asize.
151 return (vd
->vdev_asize
);
154 * The top-level vdev just returns the allocatable size rounded
155 * to the nearest metaslab.
157 if (vd
== vd
->vdev_top
)
158 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
161 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
162 * so each child must provide at least 1/Nth of its asize.
164 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
165 return ((pvd
->vdev_min_asize
+ pvd
->vdev_children
- 1) /
168 return (pvd
->vdev_min_asize
);
172 vdev_set_min_asize(vdev_t
*vd
)
174 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
176 for (int c
= 0; c
< vd
->vdev_children
; c
++)
177 vdev_set_min_asize(vd
->vdev_child
[c
]);
181 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
183 vdev_t
*rvd
= spa
->spa_root_vdev
;
185 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
187 if (vdev
< rvd
->vdev_children
) {
188 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
189 return (rvd
->vdev_child
[vdev
]);
196 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
200 if (vd
->vdev_guid
== guid
)
203 for (int c
= 0; c
< vd
->vdev_children
; c
++)
204 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
212 vdev_count_leaves_impl(vdev_t
*vd
)
216 if (vd
->vdev_ops
->vdev_op_leaf
)
219 for (int c
= 0; c
< vd
->vdev_children
; c
++)
220 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
226 vdev_count_leaves(spa_t
*spa
)
228 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
232 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
234 size_t oldsize
, newsize
;
235 uint64_t id
= cvd
->vdev_id
;
237 spa_t
*spa
= cvd
->vdev_spa
;
239 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
240 ASSERT(cvd
->vdev_parent
== NULL
);
242 cvd
->vdev_parent
= pvd
;
247 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
249 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
250 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
251 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
253 newchild
= kmem_zalloc(newsize
, KM_SLEEP
);
254 if (pvd
->vdev_child
!= NULL
) {
255 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
256 kmem_free(pvd
->vdev_child
, oldsize
);
259 pvd
->vdev_child
= newchild
;
260 pvd
->vdev_child
[id
] = cvd
;
262 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
263 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
266 * Walk up all ancestors to update guid sum.
268 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
269 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
273 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
276 uint_t id
= cvd
->vdev_id
;
278 ASSERT(cvd
->vdev_parent
== pvd
);
283 ASSERT(id
< pvd
->vdev_children
);
284 ASSERT(pvd
->vdev_child
[id
] == cvd
);
286 pvd
->vdev_child
[id
] = NULL
;
287 cvd
->vdev_parent
= NULL
;
289 for (c
= 0; c
< pvd
->vdev_children
; c
++)
290 if (pvd
->vdev_child
[c
])
293 if (c
== pvd
->vdev_children
) {
294 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
295 pvd
->vdev_child
= NULL
;
296 pvd
->vdev_children
= 0;
300 * Walk up all ancestors to update guid sum.
302 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
303 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
307 * Remove any holes in the child array.
310 vdev_compact_children(vdev_t
*pvd
)
312 vdev_t
**newchild
, *cvd
;
313 int oldc
= pvd
->vdev_children
;
316 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
318 for (int c
= newc
= 0; c
< oldc
; c
++)
319 if (pvd
->vdev_child
[c
])
322 newchild
= kmem_alloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
324 for (int c
= newc
= 0; c
< oldc
; c
++) {
325 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
326 newchild
[newc
] = cvd
;
327 cvd
->vdev_id
= newc
++;
331 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
332 pvd
->vdev_child
= newchild
;
333 pvd
->vdev_children
= newc
;
337 * Allocate and minimally initialize a vdev_t.
340 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
343 vdev_indirect_config_t
*vic
;
345 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
346 vic
= &vd
->vdev_indirect_config
;
348 if (spa
->spa_root_vdev
== NULL
) {
349 ASSERT(ops
== &vdev_root_ops
);
350 spa
->spa_root_vdev
= vd
;
351 spa
->spa_load_guid
= spa_generate_guid(NULL
);
354 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
355 if (spa
->spa_root_vdev
== vd
) {
357 * The root vdev's guid will also be the pool guid,
358 * which must be unique among all pools.
360 guid
= spa_generate_guid(NULL
);
363 * Any other vdev's guid must be unique within the pool.
365 guid
= spa_generate_guid(spa
);
367 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
372 vd
->vdev_guid
= guid
;
373 vd
->vdev_guid_sum
= guid
;
375 vd
->vdev_state
= VDEV_STATE_CLOSED
;
376 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
377 vic
->vic_prev_indirect_vdev
= UINT64_MAX
;
379 rw_init(&vd
->vdev_indirect_rwlock
, NULL
, RW_DEFAULT
, NULL
);
380 mutex_init(&vd
->vdev_obsolete_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
381 vd
->vdev_obsolete_segments
= range_tree_create(NULL
, NULL
);
383 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
384 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
385 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
386 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
387 for (int t
= 0; t
< DTL_TYPES
; t
++) {
388 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
);
390 txg_list_create(&vd
->vdev_ms_list
, spa
,
391 offsetof(struct metaslab
, ms_txg_node
));
392 txg_list_create(&vd
->vdev_dtl_list
, spa
,
393 offsetof(struct vdev
, vdev_dtl_node
));
394 vd
->vdev_stat
.vs_timestamp
= gethrtime();
402 * Allocate a new vdev. The 'alloctype' is used to control whether we are
403 * creating a new vdev or loading an existing one - the behavior is slightly
404 * different for each case.
407 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
412 uint64_t guid
= 0, islog
, nparity
;
414 vdev_indirect_config_t
*vic
;
416 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
418 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
419 return (SET_ERROR(EINVAL
));
421 if ((ops
= vdev_getops(type
)) == NULL
)
422 return (SET_ERROR(EINVAL
));
425 * If this is a load, get the vdev guid from the nvlist.
426 * Otherwise, vdev_alloc_common() will generate one for us.
428 if (alloctype
== VDEV_ALLOC_LOAD
) {
431 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
433 return (SET_ERROR(EINVAL
));
435 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
436 return (SET_ERROR(EINVAL
));
437 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
438 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
439 return (SET_ERROR(EINVAL
));
440 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
441 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
442 return (SET_ERROR(EINVAL
));
443 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
444 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
445 return (SET_ERROR(EINVAL
));
449 * The first allocated vdev must be of type 'root'.
451 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
452 return (SET_ERROR(EINVAL
));
455 * Determine whether we're a log vdev.
458 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
459 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
460 return (SET_ERROR(ENOTSUP
));
462 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
463 return (SET_ERROR(ENOTSUP
));
466 * Set the nparity property for RAID-Z vdevs.
469 if (ops
== &vdev_raidz_ops
) {
470 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
472 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
473 return (SET_ERROR(EINVAL
));
475 * Previous versions could only support 1 or 2 parity
479 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
480 return (SET_ERROR(ENOTSUP
));
482 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
483 return (SET_ERROR(ENOTSUP
));
486 * We require the parity to be specified for SPAs that
487 * support multiple parity levels.
489 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
490 return (SET_ERROR(EINVAL
));
492 * Otherwise, we default to 1 parity device for RAID-Z.
499 ASSERT(nparity
!= -1ULL);
501 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
502 vic
= &vd
->vdev_indirect_config
;
504 vd
->vdev_islog
= islog
;
505 vd
->vdev_nparity
= nparity
;
507 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
508 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
509 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
510 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
511 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
512 &vd
->vdev_physpath
) == 0)
513 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
514 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
515 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
518 * Set the whole_disk property. If it's not specified, leave the value
521 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
522 &vd
->vdev_wholedisk
) != 0)
523 vd
->vdev_wholedisk
= -1ULL;
525 ASSERT0(vic
->vic_mapping_object
);
526 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
527 &vic
->vic_mapping_object
);
528 ASSERT0(vic
->vic_births_object
);
529 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
530 &vic
->vic_births_object
);
531 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
532 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
533 &vic
->vic_prev_indirect_vdev
);
536 * Look for the 'not present' flag. This will only be set if the device
537 * was not present at the time of import.
539 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
540 &vd
->vdev_not_present
);
543 * Get the alignment requirement.
545 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
548 * Retrieve the vdev creation time.
550 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
554 * If we're a top-level vdev, try to load the allocation parameters.
556 if (parent
&& !parent
->vdev_parent
&&
557 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
558 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
560 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
562 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
564 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
566 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
569 ASSERT0(vd
->vdev_top_zap
);
572 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
573 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
574 alloctype
== VDEV_ALLOC_ADD
||
575 alloctype
== VDEV_ALLOC_SPLIT
||
576 alloctype
== VDEV_ALLOC_ROOTPOOL
);
577 vd
->vdev_mg
= metaslab_group_create(islog
?
578 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
581 if (vd
->vdev_ops
->vdev_op_leaf
&&
582 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
583 (void) nvlist_lookup_uint64(nv
,
584 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
586 ASSERT0(vd
->vdev_leaf_zap
);
590 * If we're a leaf vdev, try to load the DTL object and other state.
593 if (vd
->vdev_ops
->vdev_op_leaf
&&
594 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
595 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
596 if (alloctype
== VDEV_ALLOC_LOAD
) {
597 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
598 &vd
->vdev_dtl_object
);
599 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
603 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
606 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
607 &spare
) == 0 && spare
)
611 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
614 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
615 &vd
->vdev_resilver_txg
);
618 * When importing a pool, we want to ignore the persistent fault
619 * state, as the diagnosis made on another system may not be
620 * valid in the current context. Local vdevs will
621 * remain in the faulted state.
623 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
624 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
626 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
628 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
631 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
635 VDEV_AUX_ERR_EXCEEDED
;
636 if (nvlist_lookup_string(nv
,
637 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
638 strcmp(aux
, "external") == 0)
639 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
645 * Add ourselves to the parent's list of children.
647 vdev_add_child(parent
, vd
);
655 vdev_free(vdev_t
*vd
)
657 spa_t
*spa
= vd
->vdev_spa
;
660 * vdev_free() implies closing the vdev first. This is simpler than
661 * trying to ensure complicated semantics for all callers.
665 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
666 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
671 for (int c
= 0; c
< vd
->vdev_children
; c
++)
672 vdev_free(vd
->vdev_child
[c
]);
674 ASSERT(vd
->vdev_child
== NULL
);
675 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
678 * Discard allocation state.
680 if (vd
->vdev_mg
!= NULL
) {
681 vdev_metaslab_fini(vd
);
682 metaslab_group_destroy(vd
->vdev_mg
);
685 ASSERT0(vd
->vdev_stat
.vs_space
);
686 ASSERT0(vd
->vdev_stat
.vs_dspace
);
687 ASSERT0(vd
->vdev_stat
.vs_alloc
);
690 * Remove this vdev from its parent's child list.
692 vdev_remove_child(vd
->vdev_parent
, vd
);
694 ASSERT(vd
->vdev_parent
== NULL
);
697 * Clean up vdev structure.
703 spa_strfree(vd
->vdev_path
);
705 spa_strfree(vd
->vdev_devid
);
706 if (vd
->vdev_physpath
)
707 spa_strfree(vd
->vdev_physpath
);
709 spa_strfree(vd
->vdev_fru
);
711 if (vd
->vdev_isspare
)
712 spa_spare_remove(vd
);
713 if (vd
->vdev_isl2cache
)
714 spa_l2cache_remove(vd
);
716 txg_list_destroy(&vd
->vdev_ms_list
);
717 txg_list_destroy(&vd
->vdev_dtl_list
);
719 mutex_enter(&vd
->vdev_dtl_lock
);
720 space_map_close(vd
->vdev_dtl_sm
);
721 for (int t
= 0; t
< DTL_TYPES
; t
++) {
722 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
723 range_tree_destroy(vd
->vdev_dtl
[t
]);
725 mutex_exit(&vd
->vdev_dtl_lock
);
727 EQUIV(vd
->vdev_indirect_births
!= NULL
,
728 vd
->vdev_indirect_mapping
!= NULL
);
729 if (vd
->vdev_indirect_births
!= NULL
) {
730 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
731 vdev_indirect_births_close(vd
->vdev_indirect_births
);
734 if (vd
->vdev_obsolete_sm
!= NULL
) {
735 ASSERT(vd
->vdev_removing
||
736 vd
->vdev_ops
== &vdev_indirect_ops
);
737 space_map_close(vd
->vdev_obsolete_sm
);
738 vd
->vdev_obsolete_sm
= NULL
;
740 range_tree_destroy(vd
->vdev_obsolete_segments
);
741 rw_destroy(&vd
->vdev_indirect_rwlock
);
742 mutex_destroy(&vd
->vdev_obsolete_lock
);
744 mutex_destroy(&vd
->vdev_queue_lock
);
745 mutex_destroy(&vd
->vdev_dtl_lock
);
746 mutex_destroy(&vd
->vdev_stat_lock
);
747 mutex_destroy(&vd
->vdev_probe_lock
);
749 if (vd
== spa
->spa_root_vdev
)
750 spa
->spa_root_vdev
= NULL
;
752 kmem_free(vd
, sizeof (vdev_t
));
756 * Transfer top-level vdev state from svd to tvd.
759 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
761 spa_t
*spa
= svd
->vdev_spa
;
766 ASSERT(tvd
== tvd
->vdev_top
);
768 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
769 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
770 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
771 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
773 svd
->vdev_ms_array
= 0;
774 svd
->vdev_ms_shift
= 0;
775 svd
->vdev_ms_count
= 0;
776 svd
->vdev_top_zap
= 0;
779 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
780 tvd
->vdev_mg
= svd
->vdev_mg
;
781 tvd
->vdev_ms
= svd
->vdev_ms
;
786 if (tvd
->vdev_mg
!= NULL
)
787 tvd
->vdev_mg
->mg_vd
= tvd
;
789 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
790 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
791 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
793 svd
->vdev_stat
.vs_alloc
= 0;
794 svd
->vdev_stat
.vs_space
= 0;
795 svd
->vdev_stat
.vs_dspace
= 0;
797 for (t
= 0; t
< TXG_SIZE
; t
++) {
798 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
799 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
800 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
801 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
802 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
803 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
806 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
807 vdev_config_clean(svd
);
808 vdev_config_dirty(tvd
);
811 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
812 vdev_state_clean(svd
);
813 vdev_state_dirty(tvd
);
816 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
817 svd
->vdev_deflate_ratio
= 0;
819 tvd
->vdev_islog
= svd
->vdev_islog
;
824 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
831 for (int c
= 0; c
< vd
->vdev_children
; c
++)
832 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
836 * Add a mirror/replacing vdev above an existing vdev.
839 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
841 spa_t
*spa
= cvd
->vdev_spa
;
842 vdev_t
*pvd
= cvd
->vdev_parent
;
845 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
847 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
849 mvd
->vdev_asize
= cvd
->vdev_asize
;
850 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
851 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
852 mvd
->vdev_psize
= cvd
->vdev_psize
;
853 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
854 mvd
->vdev_state
= cvd
->vdev_state
;
855 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
857 vdev_remove_child(pvd
, cvd
);
858 vdev_add_child(pvd
, mvd
);
859 cvd
->vdev_id
= mvd
->vdev_children
;
860 vdev_add_child(mvd
, cvd
);
861 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
863 if (mvd
== mvd
->vdev_top
)
864 vdev_top_transfer(cvd
, mvd
);
870 * Remove a 1-way mirror/replacing vdev from the tree.
873 vdev_remove_parent(vdev_t
*cvd
)
875 vdev_t
*mvd
= cvd
->vdev_parent
;
876 vdev_t
*pvd
= mvd
->vdev_parent
;
878 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
880 ASSERT(mvd
->vdev_children
== 1);
881 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
882 mvd
->vdev_ops
== &vdev_replacing_ops
||
883 mvd
->vdev_ops
== &vdev_spare_ops
);
884 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
886 vdev_remove_child(mvd
, cvd
);
887 vdev_remove_child(pvd
, mvd
);
890 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
891 * Otherwise, we could have detached an offline device, and when we
892 * go to import the pool we'll think we have two top-level vdevs,
893 * instead of a different version of the same top-level vdev.
895 if (mvd
->vdev_top
== mvd
) {
896 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
897 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
898 cvd
->vdev_guid
+= guid_delta
;
899 cvd
->vdev_guid_sum
+= guid_delta
;
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
);
933 ASSERT(oldc
<= newc
);
935 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
938 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
939 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
943 vd
->vdev_ms_count
= newc
;
945 for (m
= oldc
; m
< newc
; m
++) {
949 * vdev_ms_array may be 0 if we are creating the "fake"
950 * metaslabs for an indirect vdev for zdb's leak detection.
951 * See zdb_leak_init().
953 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
954 error
= dmu_read(mos
, vd
->vdev_ms_array
,
955 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
958 vdev_dbgmsg(vd
, "unable to read the metaslab "
959 "array [error=%d]", error
);
964 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
967 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
974 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
977 * If the vdev is being removed we don't activate
978 * the metaslabs since we want to ensure that no new
979 * allocations are performed on this device.
981 if (oldc
== 0 && !vd
->vdev_removing
)
982 metaslab_group_activate(vd
->vdev_mg
);
985 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
991 vdev_metaslab_fini(vdev_t
*vd
)
993 if (vd
->vdev_ms
!= NULL
) {
994 uint64_t count
= vd
->vdev_ms_count
;
996 metaslab_group_passivate(vd
->vdev_mg
);
997 for (uint64_t m
= 0; m
< count
; m
++) {
998 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1003 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1006 vd
->vdev_ms_count
= 0;
1008 ASSERT0(vd
->vdev_ms_count
);
1011 typedef struct vdev_probe_stats
{
1012 boolean_t vps_readable
;
1013 boolean_t vps_writeable
;
1015 } vdev_probe_stats_t
;
1018 vdev_probe_done(zio_t
*zio
)
1020 spa_t
*spa
= zio
->io_spa
;
1021 vdev_t
*vd
= zio
->io_vd
;
1022 vdev_probe_stats_t
*vps
= zio
->io_private
;
1024 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1026 if (zio
->io_type
== ZIO_TYPE_READ
) {
1027 if (zio
->io_error
== 0)
1028 vps
->vps_readable
= 1;
1029 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1030 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1031 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1032 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1033 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1035 abd_free(zio
->io_abd
);
1037 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1038 if (zio
->io_error
== 0)
1039 vps
->vps_writeable
= 1;
1040 abd_free(zio
->io_abd
);
1041 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1044 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1045 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1047 if (vdev_readable(vd
) &&
1048 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1051 ASSERT(zio
->io_error
!= 0);
1052 vdev_dbgmsg(vd
, "failed probe");
1053 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1054 spa
, vd
, NULL
, 0, 0);
1055 zio
->io_error
= SET_ERROR(ENXIO
);
1058 mutex_enter(&vd
->vdev_probe_lock
);
1059 ASSERT(vd
->vdev_probe_zio
== zio
);
1060 vd
->vdev_probe_zio
= NULL
;
1061 mutex_exit(&vd
->vdev_probe_lock
);
1063 zio_link_t
*zl
= NULL
;
1064 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1065 if (!vdev_accessible(vd
, pio
))
1066 pio
->io_error
= SET_ERROR(ENXIO
);
1068 kmem_free(vps
, sizeof (*vps
));
1073 * Determine whether this device is accessible.
1075 * Read and write to several known locations: the pad regions of each
1076 * vdev label but the first, which we leave alone in case it contains
1080 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1082 spa_t
*spa
= vd
->vdev_spa
;
1083 vdev_probe_stats_t
*vps
= NULL
;
1086 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1089 * Don't probe the probe.
1091 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1095 * To prevent 'probe storms' when a device fails, we create
1096 * just one probe i/o at a time. All zios that want to probe
1097 * this vdev will become parents of the probe io.
1099 mutex_enter(&vd
->vdev_probe_lock
);
1101 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1102 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1104 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1105 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1108 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1110 * vdev_cant_read and vdev_cant_write can only
1111 * transition from TRUE to FALSE when we have the
1112 * SCL_ZIO lock as writer; otherwise they can only
1113 * transition from FALSE to TRUE. This ensures that
1114 * any zio looking at these values can assume that
1115 * failures persist for the life of the I/O. That's
1116 * important because when a device has intermittent
1117 * connectivity problems, we want to ensure that
1118 * they're ascribed to the device (ENXIO) and not
1121 * Since we hold SCL_ZIO as writer here, clear both
1122 * values so the probe can reevaluate from first
1125 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1126 vd
->vdev_cant_read
= B_FALSE
;
1127 vd
->vdev_cant_write
= B_FALSE
;
1130 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1131 vdev_probe_done
, vps
,
1132 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1135 * We can't change the vdev state in this context, so we
1136 * kick off an async task to do it on our behalf.
1139 vd
->vdev_probe_wanted
= B_TRUE
;
1140 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1145 zio_add_child(zio
, pio
);
1147 mutex_exit(&vd
->vdev_probe_lock
);
1150 ASSERT(zio
!= NULL
);
1154 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1155 zio_nowait(zio_read_phys(pio
, vd
,
1156 vdev_label_offset(vd
->vdev_psize
, l
,
1157 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1158 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1159 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1160 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1171 vdev_open_child(void *arg
)
1175 vd
->vdev_open_thread
= curthread
;
1176 vd
->vdev_open_error
= vdev_open(vd
);
1177 vd
->vdev_open_thread
= NULL
;
1181 vdev_uses_zvols(vdev_t
*vd
)
1183 if (vd
->vdev_path
&& strncmp(vd
->vdev_path
, ZVOL_DIR
,
1184 strlen(ZVOL_DIR
)) == 0)
1186 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1187 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1193 vdev_open_children(vdev_t
*vd
)
1196 int children
= vd
->vdev_children
;
1199 * in order to handle pools on top of zvols, do the opens
1200 * in a single thread so that the same thread holds the
1201 * spa_namespace_lock
1203 if (vdev_uses_zvols(vd
)) {
1204 for (int c
= 0; c
< children
; c
++)
1205 vd
->vdev_child
[c
]->vdev_open_error
=
1206 vdev_open(vd
->vdev_child
[c
]);
1209 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1210 children
, children
, TASKQ_PREPOPULATE
);
1212 for (int c
= 0; c
< children
; c
++)
1213 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1220 * Compute the raidz-deflation ratio. Note, we hard-code
1221 * in 128k (1 << 17) because it is the "typical" blocksize.
1222 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1223 * otherwise it would inconsistently account for existing bp's.
1226 vdev_set_deflate_ratio(vdev_t
*vd
)
1228 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1229 vd
->vdev_deflate_ratio
= (1 << 17) /
1230 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1235 * Prepare a virtual device for access.
1238 vdev_open(vdev_t
*vd
)
1240 spa_t
*spa
= vd
->vdev_spa
;
1243 uint64_t max_osize
= 0;
1244 uint64_t asize
, max_asize
, psize
;
1245 uint64_t ashift
= 0;
1247 ASSERT(vd
->vdev_open_thread
== curthread
||
1248 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1249 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1250 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1251 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1253 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1254 vd
->vdev_cant_read
= B_FALSE
;
1255 vd
->vdev_cant_write
= B_FALSE
;
1256 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1259 * If this vdev is not removed, check its fault status. If it's
1260 * faulted, bail out of the open.
1262 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1263 ASSERT(vd
->vdev_children
== 0);
1264 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1265 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1266 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1267 vd
->vdev_label_aux
);
1268 return (SET_ERROR(ENXIO
));
1269 } else if (vd
->vdev_offline
) {
1270 ASSERT(vd
->vdev_children
== 0);
1271 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1272 return (SET_ERROR(ENXIO
));
1275 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1278 * Reset the vdev_reopening flag so that we actually close
1279 * the vdev on error.
1281 vd
->vdev_reopening
= B_FALSE
;
1282 if (zio_injection_enabled
&& error
== 0)
1283 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1286 if (vd
->vdev_removed
&&
1287 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1288 vd
->vdev_removed
= B_FALSE
;
1290 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1291 vd
->vdev_stat
.vs_aux
);
1295 vd
->vdev_removed
= B_FALSE
;
1298 * Recheck the faulted flag now that we have confirmed that
1299 * the vdev is accessible. If we're faulted, bail.
1301 if (vd
->vdev_faulted
) {
1302 ASSERT(vd
->vdev_children
== 0);
1303 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1304 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1305 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1306 vd
->vdev_label_aux
);
1307 return (SET_ERROR(ENXIO
));
1310 if (vd
->vdev_degraded
) {
1311 ASSERT(vd
->vdev_children
== 0);
1312 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1313 VDEV_AUX_ERR_EXCEEDED
);
1315 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1319 * For hole or missing vdevs we just return success.
1321 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1324 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1325 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1326 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1332 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1333 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1335 if (vd
->vdev_children
== 0) {
1336 if (osize
< SPA_MINDEVSIZE
) {
1337 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1338 VDEV_AUX_TOO_SMALL
);
1339 return (SET_ERROR(EOVERFLOW
));
1342 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1343 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1344 VDEV_LABEL_END_SIZE
);
1346 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1347 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1348 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1349 VDEV_AUX_TOO_SMALL
);
1350 return (SET_ERROR(EOVERFLOW
));
1354 max_asize
= max_osize
;
1357 vd
->vdev_psize
= psize
;
1360 * Make sure the allocatable size hasn't shrunk too much.
1362 if (asize
< vd
->vdev_min_asize
) {
1363 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1364 VDEV_AUX_BAD_LABEL
);
1365 return (SET_ERROR(EINVAL
));
1368 if (vd
->vdev_asize
== 0) {
1370 * This is the first-ever open, so use the computed values.
1371 * For testing purposes, a higher ashift can be requested.
1373 vd
->vdev_asize
= asize
;
1374 vd
->vdev_max_asize
= max_asize
;
1375 vd
->vdev_ashift
= MAX(ashift
, vd
->vdev_ashift
);
1378 * Detect if the alignment requirement has increased.
1379 * We don't want to make the pool unavailable, just
1380 * issue a warning instead.
1382 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1383 vd
->vdev_ops
->vdev_op_leaf
) {
1385 "Disk, '%s', has a block alignment that is "
1386 "larger than the pool's alignment\n",
1389 vd
->vdev_max_asize
= max_asize
;
1393 * If all children are healthy we update asize if either:
1394 * The asize has increased, due to a device expansion caused by dynamic
1395 * LUN growth or vdev replacement, and automatic expansion is enabled;
1396 * making the additional space available.
1398 * The asize has decreased, due to a device shrink usually caused by a
1399 * vdev replace with a smaller device. This ensures that calculations
1400 * based of max_asize and asize e.g. esize are always valid. It's safe
1401 * to do this as we've already validated that asize is greater than
1404 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1405 ((asize
> vd
->vdev_asize
&&
1406 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1407 (asize
< vd
->vdev_asize
)))
1408 vd
->vdev_asize
= asize
;
1410 vdev_set_min_asize(vd
);
1413 * Ensure we can issue some IO before declaring the
1414 * vdev open for business.
1416 if (vd
->vdev_ops
->vdev_op_leaf
&&
1417 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1418 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1419 VDEV_AUX_ERR_EXCEEDED
);
1423 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1424 !vd
->vdev_isl2cache
&& !vd
->vdev_islog
) {
1425 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1426 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1427 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1428 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1432 * Track the min and max ashift values for normal data devices.
1434 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1435 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1436 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1437 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1438 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1439 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1443 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1444 * resilver. But don't do this if we are doing a reopen for a scrub,
1445 * since this would just restart the scrub we are already doing.
1447 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1448 vdev_resilver_needed(vd
, NULL
, NULL
))
1449 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1455 * Called once the vdevs are all opened, this routine validates the label
1456 * contents. This needs to be done before vdev_load() so that we don't
1457 * inadvertently do repair I/Os to the wrong device.
1459 * If 'strict' is false ignore the spa guid check. This is necessary because
1460 * if the machine crashed during a re-guid the new guid might have been written
1461 * to all of the vdev labels, but not the cached config. The strict check
1462 * will be performed when the pool is opened again using the mos config.
1464 * This function will only return failure if one of the vdevs indicates that it
1465 * has since been destroyed or exported. This is only possible if
1466 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1467 * will be updated but the function will return 0.
1470 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1472 spa_t
*spa
= vd
->vdev_spa
;
1474 uint64_t guid
= 0, top_guid
;
1477 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1478 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1479 return (SET_ERROR(EBADF
));
1482 * If the device has already failed, or was marked offline, don't do
1483 * any further validation. Otherwise, label I/O will fail and we will
1484 * overwrite the previous state.
1486 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1487 uint64_t aux_guid
= 0;
1489 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1490 spa_last_synced_txg(spa
) : -1ULL;
1492 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1493 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1494 VDEV_AUX_BAD_LABEL
);
1495 vdev_dbgmsg(vd
, "vdev_validate: failed reading config");
1500 * Determine if this vdev has been split off into another
1501 * pool. If so, then refuse to open it.
1503 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1504 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1505 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1506 VDEV_AUX_SPLIT_POOL
);
1508 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other "
1513 if (strict
&& (nvlist_lookup_uint64(label
,
1514 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1515 guid
!= spa_guid(spa
))) {
1516 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1517 VDEV_AUX_CORRUPT_DATA
);
1519 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid "
1520 "doesn't match config (%llu != %llu)",
1522 (u_longlong_t
)spa_guid(spa
));
1526 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1527 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1532 * If this vdev just became a top-level vdev because its
1533 * sibling was detached, it will have adopted the parent's
1534 * vdev guid -- but the label may or may not be on disk yet.
1535 * Fortunately, either version of the label will have the
1536 * same top guid, so if we're a top-level vdev, we can
1537 * safely compare to that instead.
1539 * If we split this vdev off instead, then we also check the
1540 * original pool's guid. We don't want to consider the vdev
1541 * corrupt if it is partway through a split operation.
1543 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1545 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1547 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1548 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1549 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1550 VDEV_AUX_CORRUPT_DATA
);
1552 vdev_dbgmsg(vd
, "vdev_validate: config guid doesn't "
1553 "match label guid (%llu != %llu)",
1554 (u_longlong_t
)vd
->vdev_guid
, (u_longlong_t
)guid
);
1558 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1560 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1561 VDEV_AUX_CORRUPT_DATA
);
1563 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing",
1564 ZPOOL_CONFIG_POOL_STATE
);
1571 * If this is a verbatim import, no need to check the
1572 * state of the pool.
1574 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1575 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1576 state
!= POOL_STATE_ACTIVE
) {
1577 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state "
1578 "(%llu) for spa %s", (u_longlong_t
)state
,
1580 return (SET_ERROR(EBADF
));
1584 * If we were able to open and validate a vdev that was
1585 * previously marked permanently unavailable, clear that state
1588 if (vd
->vdev_not_present
)
1589 vd
->vdev_not_present
= 0;
1596 * Close a virtual device.
1599 vdev_close(vdev_t
*vd
)
1601 spa_t
*spa
= vd
->vdev_spa
;
1602 vdev_t
*pvd
= vd
->vdev_parent
;
1604 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1607 * If our parent is reopening, then we are as well, unless we are
1610 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1611 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1613 vd
->vdev_ops
->vdev_op_close(vd
);
1615 vdev_cache_purge(vd
);
1618 * We record the previous state before we close it, so that if we are
1619 * doing a reopen(), we don't generate FMA ereports if we notice that
1620 * it's still faulted.
1622 vd
->vdev_prevstate
= vd
->vdev_state
;
1624 if (vd
->vdev_offline
)
1625 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1627 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1628 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1632 vdev_hold(vdev_t
*vd
)
1634 spa_t
*spa
= vd
->vdev_spa
;
1636 ASSERT(spa_is_root(spa
));
1637 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1640 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1641 vdev_hold(vd
->vdev_child
[c
]);
1643 if (vd
->vdev_ops
->vdev_op_leaf
)
1644 vd
->vdev_ops
->vdev_op_hold(vd
);
1648 vdev_rele(vdev_t
*vd
)
1650 spa_t
*spa
= vd
->vdev_spa
;
1652 ASSERT(spa_is_root(spa
));
1653 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1654 vdev_rele(vd
->vdev_child
[c
]);
1656 if (vd
->vdev_ops
->vdev_op_leaf
)
1657 vd
->vdev_ops
->vdev_op_rele(vd
);
1661 * Reopen all interior vdevs and any unopened leaves. We don't actually
1662 * reopen leaf vdevs which had previously been opened as they might deadlock
1663 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1664 * If the leaf has never been opened then open it, as usual.
1667 vdev_reopen(vdev_t
*vd
)
1669 spa_t
*spa
= vd
->vdev_spa
;
1671 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1673 /* set the reopening flag unless we're taking the vdev offline */
1674 vd
->vdev_reopening
= !vd
->vdev_offline
;
1676 (void) vdev_open(vd
);
1679 * Call vdev_validate() here to make sure we have the same device.
1680 * Otherwise, a device with an invalid label could be successfully
1681 * opened in response to vdev_reopen().
1684 (void) vdev_validate_aux(vd
);
1685 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1686 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1687 !l2arc_vdev_present(vd
))
1688 l2arc_add_vdev(spa
, vd
);
1690 (void) vdev_validate(vd
, B_TRUE
);
1694 * Reassess parent vdev's health.
1696 vdev_propagate_state(vd
);
1700 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1705 * Normally, partial opens (e.g. of a mirror) are allowed.
1706 * For a create, however, we want to fail the request if
1707 * there are any components we can't open.
1709 error
= vdev_open(vd
);
1711 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1713 return (error
? error
: ENXIO
);
1717 * Recursively load DTLs and initialize all labels.
1719 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1720 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1721 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1730 vdev_metaslab_set_size(vdev_t
*vd
)
1733 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1735 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1736 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1740 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1742 ASSERT(vd
== vd
->vdev_top
);
1743 /* indirect vdevs don't have metaslabs or dtls */
1744 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
1745 ASSERT(ISP2(flags
));
1746 ASSERT(spa_writeable(vd
->vdev_spa
));
1748 if (flags
& VDD_METASLAB
)
1749 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1751 if (flags
& VDD_DTL
)
1752 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1754 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1758 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1760 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1761 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1763 if (vd
->vdev_ops
->vdev_op_leaf
)
1764 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1770 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1771 * the vdev has less than perfect replication. There are four kinds of DTL:
1773 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1775 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1777 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1778 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1779 * txgs that was scrubbed.
1781 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1782 * persistent errors or just some device being offline.
1783 * Unlike the other three, the DTL_OUTAGE map is not generally
1784 * maintained; it's only computed when needed, typically to
1785 * determine whether a device can be detached.
1787 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1788 * either has the data or it doesn't.
1790 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1791 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1792 * if any child is less than fully replicated, then so is its parent.
1793 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1794 * comprising only those txgs which appear in 'maxfaults' or more children;
1795 * those are the txgs we don't have enough replication to read. For example,
1796 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1797 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1798 * two child DTL_MISSING maps.
1800 * It should be clear from the above that to compute the DTLs and outage maps
1801 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1802 * Therefore, that is all we keep on disk. When loading the pool, or after
1803 * a configuration change, we generate all other DTLs from first principles.
1806 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1808 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1810 ASSERT(t
< DTL_TYPES
);
1811 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1812 ASSERT(spa_writeable(vd
->vdev_spa
));
1814 mutex_enter(&vd
->vdev_dtl_lock
);
1815 if (!range_tree_contains(rt
, txg
, size
))
1816 range_tree_add(rt
, txg
, size
);
1817 mutex_exit(&vd
->vdev_dtl_lock
);
1821 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1823 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1824 boolean_t dirty
= B_FALSE
;
1826 ASSERT(t
< DTL_TYPES
);
1827 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1830 * While we are loading the pool, the DTLs have not been loaded yet.
1831 * Ignore the DTLs and try all devices. This avoids a recursive
1832 * mutex enter on the vdev_dtl_lock, and also makes us try hard
1833 * when loading the pool (relying on the checksum to ensure that
1834 * we get the right data -- note that we while loading, we are
1835 * only reading the MOS, which is always checksummed).
1837 if (vd
->vdev_spa
->spa_load_state
!= SPA_LOAD_NONE
)
1840 mutex_enter(&vd
->vdev_dtl_lock
);
1841 if (range_tree_space(rt
) != 0)
1842 dirty
= range_tree_contains(rt
, txg
, size
);
1843 mutex_exit(&vd
->vdev_dtl_lock
);
1849 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1851 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1854 mutex_enter(&vd
->vdev_dtl_lock
);
1855 empty
= (range_tree_space(rt
) == 0);
1856 mutex_exit(&vd
->vdev_dtl_lock
);
1862 * Returns the lowest txg in the DTL range.
1865 vdev_dtl_min(vdev_t
*vd
)
1869 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1870 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1871 ASSERT0(vd
->vdev_children
);
1873 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1874 return (rs
->rs_start
- 1);
1878 * Returns the highest txg in the DTL.
1881 vdev_dtl_max(vdev_t
*vd
)
1885 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1886 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1887 ASSERT0(vd
->vdev_children
);
1889 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1890 return (rs
->rs_end
);
1894 * Determine if a resilvering vdev should remove any DTL entries from
1895 * its range. If the vdev was resilvering for the entire duration of the
1896 * scan then it should excise that range from its DTLs. Otherwise, this
1897 * vdev is considered partially resilvered and should leave its DTL
1898 * entries intact. The comment in vdev_dtl_reassess() describes how we
1902 vdev_dtl_should_excise(vdev_t
*vd
)
1904 spa_t
*spa
= vd
->vdev_spa
;
1905 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1907 ASSERT0(scn
->scn_phys
.scn_errors
);
1908 ASSERT0(vd
->vdev_children
);
1910 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
1913 if (vd
->vdev_resilver_txg
== 0 ||
1914 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1918 * When a resilver is initiated the scan will assign the scn_max_txg
1919 * value to the highest txg value that exists in all DTLs. If this
1920 * device's max DTL is not part of this scan (i.e. it is not in
1921 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1924 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1925 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1926 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1927 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1934 * Reassess DTLs after a config change or scrub completion.
1937 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1939 spa_t
*spa
= vd
->vdev_spa
;
1943 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1945 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1946 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1947 scrub_txg
, scrub_done
);
1949 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
1952 if (vd
->vdev_ops
->vdev_op_leaf
) {
1953 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1955 mutex_enter(&vd
->vdev_dtl_lock
);
1958 * If we've completed a scan cleanly then determine
1959 * if this vdev should remove any DTLs. We only want to
1960 * excise regions on vdevs that were available during
1961 * the entire duration of this scan.
1963 if (scrub_txg
!= 0 &&
1964 (spa
->spa_scrub_started
||
1965 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1966 vdev_dtl_should_excise(vd
)) {
1968 * We completed a scrub up to scrub_txg. If we
1969 * did it without rebooting, then the scrub dtl
1970 * will be valid, so excise the old region and
1971 * fold in the scrub dtl. Otherwise, leave the
1972 * dtl as-is if there was an error.
1974 * There's little trick here: to excise the beginning
1975 * of the DTL_MISSING map, we put it into a reference
1976 * tree and then add a segment with refcnt -1 that
1977 * covers the range [0, scrub_txg). This means
1978 * that each txg in that range has refcnt -1 or 0.
1979 * We then add DTL_SCRUB with a refcnt of 2, so that
1980 * entries in the range [0, scrub_txg) will have a
1981 * positive refcnt -- either 1 or 2. We then convert
1982 * the reference tree into the new DTL_MISSING map.
1984 space_reftree_create(&reftree
);
1985 space_reftree_add_map(&reftree
,
1986 vd
->vdev_dtl
[DTL_MISSING
], 1);
1987 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1988 space_reftree_add_map(&reftree
,
1989 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1990 space_reftree_generate_map(&reftree
,
1991 vd
->vdev_dtl
[DTL_MISSING
], 1);
1992 space_reftree_destroy(&reftree
);
1994 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1995 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1996 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1998 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1999 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
2000 if (!vdev_readable(vd
))
2001 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
2003 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2004 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
2007 * If the vdev was resilvering and no longer has any
2008 * DTLs then reset its resilvering flag.
2010 if (vd
->vdev_resilver_txg
!= 0 &&
2011 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
2012 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0)
2013 vd
->vdev_resilver_txg
= 0;
2015 mutex_exit(&vd
->vdev_dtl_lock
);
2018 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
2022 mutex_enter(&vd
->vdev_dtl_lock
);
2023 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2024 /* account for child's outage in parent's missing map */
2025 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
2027 continue; /* leaf vdevs only */
2028 if (t
== DTL_PARTIAL
)
2029 minref
= 1; /* i.e. non-zero */
2030 else if (vd
->vdev_nparity
!= 0)
2031 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
2033 minref
= vd
->vdev_children
; /* any kind of mirror */
2034 space_reftree_create(&reftree
);
2035 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2036 vdev_t
*cvd
= vd
->vdev_child
[c
];
2037 mutex_enter(&cvd
->vdev_dtl_lock
);
2038 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
2039 mutex_exit(&cvd
->vdev_dtl_lock
);
2041 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
2042 space_reftree_destroy(&reftree
);
2044 mutex_exit(&vd
->vdev_dtl_lock
);
2048 vdev_dtl_load(vdev_t
*vd
)
2050 spa_t
*spa
= vd
->vdev_spa
;
2051 objset_t
*mos
= spa
->spa_meta_objset
;
2054 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2055 ASSERT(vdev_is_concrete(vd
));
2057 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2058 vd
->vdev_dtl_object
, 0, -1ULL, 0);
2061 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2063 mutex_enter(&vd
->vdev_dtl_lock
);
2066 * Now that we've opened the space_map we need to update
2069 space_map_update(vd
->vdev_dtl_sm
);
2071 error
= space_map_load(vd
->vdev_dtl_sm
,
2072 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2073 mutex_exit(&vd
->vdev_dtl_lock
);
2078 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2079 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2088 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2090 spa_t
*spa
= vd
->vdev_spa
;
2092 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2093 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2098 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2100 spa_t
*spa
= vd
->vdev_spa
;
2101 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2102 DMU_OT_NONE
, 0, tx
);
2105 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2112 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2114 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2115 vd
->vdev_ops
!= &vdev_missing_ops
&&
2116 vd
->vdev_ops
!= &vdev_root_ops
&&
2117 !vd
->vdev_top
->vdev_removing
) {
2118 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2119 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2121 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2122 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2125 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2126 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2131 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2133 spa_t
*spa
= vd
->vdev_spa
;
2134 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2135 objset_t
*mos
= spa
->spa_meta_objset
;
2136 range_tree_t
*rtsync
;
2138 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2140 ASSERT(vdev_is_concrete(vd
));
2141 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2143 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2145 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2146 mutex_enter(&vd
->vdev_dtl_lock
);
2147 space_map_free(vd
->vdev_dtl_sm
, tx
);
2148 space_map_close(vd
->vdev_dtl_sm
);
2149 vd
->vdev_dtl_sm
= NULL
;
2150 mutex_exit(&vd
->vdev_dtl_lock
);
2153 * We only destroy the leaf ZAP for detached leaves or for
2154 * removed log devices. Removed data devices handle leaf ZAP
2155 * cleanup later, once cancellation is no longer possible.
2157 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2158 vd
->vdev_top
->vdev_islog
)) {
2159 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2160 vd
->vdev_leaf_zap
= 0;
2167 if (vd
->vdev_dtl_sm
== NULL
) {
2168 uint64_t new_object
;
2170 new_object
= space_map_alloc(mos
, tx
);
2171 VERIFY3U(new_object
, !=, 0);
2173 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2175 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2178 rtsync
= range_tree_create(NULL
, NULL
);
2180 mutex_enter(&vd
->vdev_dtl_lock
);
2181 range_tree_walk(rt
, range_tree_add
, rtsync
);
2182 mutex_exit(&vd
->vdev_dtl_lock
);
2184 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2185 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2186 range_tree_vacate(rtsync
, NULL
, NULL
);
2188 range_tree_destroy(rtsync
);
2191 * If the object for the space map has changed then dirty
2192 * the top level so that we update the config.
2194 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2195 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
2196 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
2197 (u_longlong_t
)object
,
2198 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
2199 vdev_config_dirty(vd
->vdev_top
);
2204 mutex_enter(&vd
->vdev_dtl_lock
);
2205 space_map_update(vd
->vdev_dtl_sm
);
2206 mutex_exit(&vd
->vdev_dtl_lock
);
2210 * Determine whether the specified vdev can be offlined/detached/removed
2211 * without losing data.
2214 vdev_dtl_required(vdev_t
*vd
)
2216 spa_t
*spa
= vd
->vdev_spa
;
2217 vdev_t
*tvd
= vd
->vdev_top
;
2218 uint8_t cant_read
= vd
->vdev_cant_read
;
2221 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2223 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2227 * Temporarily mark the device as unreadable, and then determine
2228 * whether this results in any DTL outages in the top-level vdev.
2229 * If not, we can safely offline/detach/remove the device.
2231 vd
->vdev_cant_read
= B_TRUE
;
2232 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2233 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2234 vd
->vdev_cant_read
= cant_read
;
2235 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2237 if (!required
&& zio_injection_enabled
)
2238 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2244 * Determine if resilver is needed, and if so the txg range.
2247 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2249 boolean_t needed
= B_FALSE
;
2250 uint64_t thismin
= UINT64_MAX
;
2251 uint64_t thismax
= 0;
2253 if (vd
->vdev_children
== 0) {
2254 mutex_enter(&vd
->vdev_dtl_lock
);
2255 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2256 vdev_writeable(vd
)) {
2258 thismin
= vdev_dtl_min(vd
);
2259 thismax
= vdev_dtl_max(vd
);
2262 mutex_exit(&vd
->vdev_dtl_lock
);
2264 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2265 vdev_t
*cvd
= vd
->vdev_child
[c
];
2266 uint64_t cmin
, cmax
;
2268 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2269 thismin
= MIN(thismin
, cmin
);
2270 thismax
= MAX(thismax
, cmax
);
2276 if (needed
&& minp
) {
2284 vdev_load(vdev_t
*vd
)
2288 * Recursively load all children.
2290 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2291 error
= vdev_load(vd
->vdev_child
[c
]);
2297 vdev_set_deflate_ratio(vd
);
2300 * If this is a top-level vdev, initialize its metaslabs.
2302 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
2303 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
2304 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2305 VDEV_AUX_CORRUPT_DATA
);
2306 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
2307 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
2308 (u_longlong_t
)vd
->vdev_asize
);
2309 return (SET_ERROR(ENXIO
));
2310 } else if ((error
= vdev_metaslab_init(vd
, 0)) != 0) {
2311 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
2312 "[error=%d]", error
);
2313 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2314 VDEV_AUX_CORRUPT_DATA
);
2320 * If this is a leaf vdev, load its DTL.
2322 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
2323 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2324 VDEV_AUX_CORRUPT_DATA
);
2325 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
2326 "[error=%d]", error
);
2330 uint64_t obsolete_sm_object
= vdev_obsolete_sm_object(vd
);
2331 if (obsolete_sm_object
!= 0) {
2332 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
2333 ASSERT(vd
->vdev_asize
!= 0);
2334 ASSERT(vd
->vdev_obsolete_sm
== NULL
);
2336 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
2337 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
2338 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2339 VDEV_AUX_CORRUPT_DATA
);
2340 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
2341 "obsolete spacemap (obj %llu) [error=%d]",
2342 (u_longlong_t
)obsolete_sm_object
, error
);
2345 space_map_update(vd
->vdev_obsolete_sm
);
2352 * The special vdev case is used for hot spares and l2cache devices. Its
2353 * sole purpose it to set the vdev state for the associated vdev. To do this,
2354 * we make sure that we can open the underlying device, then try to read the
2355 * label, and make sure that the label is sane and that it hasn't been
2356 * repurposed to another pool.
2359 vdev_validate_aux(vdev_t
*vd
)
2362 uint64_t guid
, version
;
2365 if (!vdev_readable(vd
))
2368 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2369 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2370 VDEV_AUX_CORRUPT_DATA
);
2374 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2375 !SPA_VERSION_IS_SUPPORTED(version
) ||
2376 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2377 guid
!= vd
->vdev_guid
||
2378 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2379 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2380 VDEV_AUX_CORRUPT_DATA
);
2386 * We don't actually check the pool state here. If it's in fact in
2387 * use by another pool, we update this fact on the fly when requested.
2394 * Free the objects used to store this vdev's spacemaps, and the array
2395 * that points to them.
2398 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2400 if (vd
->vdev_ms_array
== 0)
2403 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
2404 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
2405 size_t array_bytes
= array_count
* sizeof (uint64_t);
2406 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
2407 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
2408 array_bytes
, smobj_array
, 0));
2410 for (uint64_t i
= 0; i
< array_count
; i
++) {
2411 uint64_t smobj
= smobj_array
[i
];
2415 space_map_free_obj(mos
, smobj
, tx
);
2418 kmem_free(smobj_array
, array_bytes
);
2419 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
2420 vd
->vdev_ms_array
= 0;
2424 vdev_remove_empty(vdev_t
*vd
, uint64_t txg
)
2426 spa_t
*spa
= vd
->vdev_spa
;
2429 ASSERT(vd
== vd
->vdev_top
);
2430 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2432 if (vd
->vdev_ms
!= NULL
) {
2433 metaslab_group_t
*mg
= vd
->vdev_mg
;
2435 metaslab_group_histogram_verify(mg
);
2436 metaslab_class_histogram_verify(mg
->mg_class
);
2438 for (int m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2439 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2441 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2444 mutex_enter(&msp
->ms_lock
);
2446 * If the metaslab was not loaded when the vdev
2447 * was removed then the histogram accounting may
2448 * not be accurate. Update the histogram information
2449 * here so that we ensure that the metaslab group
2450 * and metaslab class are up-to-date.
2452 metaslab_group_histogram_remove(mg
, msp
);
2454 VERIFY0(space_map_allocated(msp
->ms_sm
));
2455 space_map_close(msp
->ms_sm
);
2457 mutex_exit(&msp
->ms_lock
);
2460 metaslab_group_histogram_verify(mg
);
2461 metaslab_class_histogram_verify(mg
->mg_class
);
2462 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2463 ASSERT0(mg
->mg_histogram
[i
]);
2466 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2467 vdev_destroy_spacemaps(vd
, tx
);
2469 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2470 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2471 vd
->vdev_top_zap
= 0;
2477 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2480 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2482 ASSERT(vdev_is_concrete(vd
));
2484 while (msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
2485 metaslab_sync_done(msp
, txg
);
2488 metaslab_sync_reassess(vd
->vdev_mg
);
2492 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2494 spa_t
*spa
= vd
->vdev_spa
;
2499 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
2502 ASSERT(vd
->vdev_removing
||
2503 vd
->vdev_ops
== &vdev_indirect_ops
);
2505 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2506 vdev_indirect_sync_obsolete(vd
, tx
);
2510 * If the vdev is indirect, it can't have dirty
2511 * metaslabs or DTLs.
2513 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
2514 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
2515 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
2520 ASSERT(vdev_is_concrete(vd
));
2522 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
2523 !vd
->vdev_removing
) {
2524 ASSERT(vd
== vd
->vdev_top
);
2525 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
2526 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2527 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2528 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2529 ASSERT(vd
->vdev_ms_array
!= 0);
2530 vdev_config_dirty(vd
);
2534 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2535 metaslab_sync(msp
, txg
);
2536 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2539 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2540 vdev_dtl_sync(lvd
, txg
);
2543 * Remove the metadata associated with this vdev once it's empty.
2544 * Note that this is typically used for log/cache device removal;
2545 * we don't empty toplevel vdevs when removing them. But if
2546 * a toplevel happens to be emptied, this is not harmful.
2548 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
) {
2549 vdev_remove_empty(vd
, txg
);
2552 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2556 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2558 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2562 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2563 * not be opened, and no I/O is attempted.
2566 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2570 spa_vdev_state_enter(spa
, SCL_NONE
);
2572 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2573 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2575 if (!vd
->vdev_ops
->vdev_op_leaf
)
2576 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2581 * We don't directly use the aux state here, but if we do a
2582 * vdev_reopen(), we need this value to be present to remember why we
2585 vd
->vdev_label_aux
= aux
;
2588 * Faulted state takes precedence over degraded.
2590 vd
->vdev_delayed_close
= B_FALSE
;
2591 vd
->vdev_faulted
= 1ULL;
2592 vd
->vdev_degraded
= 0ULL;
2593 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2596 * If this device has the only valid copy of the data, then
2597 * back off and simply mark the vdev as degraded instead.
2599 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2600 vd
->vdev_degraded
= 1ULL;
2601 vd
->vdev_faulted
= 0ULL;
2604 * If we reopen the device and it's not dead, only then do we
2609 if (vdev_readable(vd
))
2610 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2613 return (spa_vdev_state_exit(spa
, vd
, 0));
2617 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2618 * user that something is wrong. The vdev continues to operate as normal as far
2619 * as I/O is concerned.
2622 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2626 spa_vdev_state_enter(spa
, SCL_NONE
);
2628 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2629 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2631 if (!vd
->vdev_ops
->vdev_op_leaf
)
2632 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2635 * If the vdev is already faulted, then don't do anything.
2637 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2638 return (spa_vdev_state_exit(spa
, NULL
, 0));
2640 vd
->vdev_degraded
= 1ULL;
2641 if (!vdev_is_dead(vd
))
2642 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2645 return (spa_vdev_state_exit(spa
, vd
, 0));
2649 * Online the given vdev.
2651 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2652 * spare device should be detached when the device finishes resilvering.
2653 * Second, the online should be treated like a 'test' online case, so no FMA
2654 * events are generated if the device fails to open.
2657 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2659 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2660 boolean_t wasoffline
;
2661 vdev_state_t oldstate
;
2663 spa_vdev_state_enter(spa
, SCL_NONE
);
2665 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2666 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2668 if (!vd
->vdev_ops
->vdev_op_leaf
)
2669 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2671 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
2672 oldstate
= vd
->vdev_state
;
2675 vd
->vdev_offline
= B_FALSE
;
2676 vd
->vdev_tmpoffline
= B_FALSE
;
2677 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2678 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2680 /* XXX - L2ARC 1.0 does not support expansion */
2681 if (!vd
->vdev_aux
) {
2682 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2683 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2687 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2689 if (!vd
->vdev_aux
) {
2690 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2691 pvd
->vdev_expanding
= B_FALSE
;
2695 *newstate
= vd
->vdev_state
;
2696 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2697 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2698 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2699 vd
->vdev_parent
->vdev_child
[0] == vd
)
2700 vd
->vdev_unspare
= B_TRUE
;
2702 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2704 /* XXX - L2ARC 1.0 does not support expansion */
2706 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2707 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2711 (oldstate
< VDEV_STATE_DEGRADED
&&
2712 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
2713 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
2715 return (spa_vdev_state_exit(spa
, vd
, 0));
2719 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2723 uint64_t generation
;
2724 metaslab_group_t
*mg
;
2727 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2729 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2730 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2732 if (!vd
->vdev_ops
->vdev_op_leaf
)
2733 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2737 generation
= spa
->spa_config_generation
+ 1;
2740 * If the device isn't already offline, try to offline it.
2742 if (!vd
->vdev_offline
) {
2744 * If this device has the only valid copy of some data,
2745 * don't allow it to be offlined. Log devices are always
2748 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2749 vdev_dtl_required(vd
))
2750 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2753 * If the top-level is a slog and it has had allocations
2754 * then proceed. We check that the vdev's metaslab group
2755 * is not NULL since it's possible that we may have just
2756 * added this vdev but not yet initialized its metaslabs.
2758 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2760 * Prevent any future allocations.
2762 metaslab_group_passivate(mg
);
2763 (void) spa_vdev_state_exit(spa
, vd
, 0);
2765 error
= spa_reset_logs(spa
);
2767 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2770 * Check to see if the config has changed.
2772 if (error
|| generation
!= spa
->spa_config_generation
) {
2773 metaslab_group_activate(mg
);
2775 return (spa_vdev_state_exit(spa
,
2777 (void) spa_vdev_state_exit(spa
, vd
, 0);
2780 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2784 * Offline this device and reopen its top-level vdev.
2785 * If the top-level vdev is a log device then just offline
2786 * it. Otherwise, if this action results in the top-level
2787 * vdev becoming unusable, undo it and fail the request.
2789 vd
->vdev_offline
= B_TRUE
;
2792 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2793 vdev_is_dead(tvd
)) {
2794 vd
->vdev_offline
= B_FALSE
;
2796 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2800 * Add the device back into the metaslab rotor so that
2801 * once we online the device it's open for business.
2803 if (tvd
->vdev_islog
&& mg
!= NULL
)
2804 metaslab_group_activate(mg
);
2807 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2809 return (spa_vdev_state_exit(spa
, vd
, 0));
2813 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2817 mutex_enter(&spa
->spa_vdev_top_lock
);
2818 error
= vdev_offline_locked(spa
, guid
, flags
);
2819 mutex_exit(&spa
->spa_vdev_top_lock
);
2825 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2826 * vdev_offline(), we assume the spa config is locked. We also clear all
2827 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2830 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2832 vdev_t
*rvd
= spa
->spa_root_vdev
;
2834 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2839 vd
->vdev_stat
.vs_read_errors
= 0;
2840 vd
->vdev_stat
.vs_write_errors
= 0;
2841 vd
->vdev_stat
.vs_checksum_errors
= 0;
2843 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2844 vdev_clear(spa
, vd
->vdev_child
[c
]);
2847 * It makes no sense to "clear" an indirect vdev.
2849 if (!vdev_is_concrete(vd
))
2853 * If we're in the FAULTED state or have experienced failed I/O, then
2854 * clear the persistent state and attempt to reopen the device. We
2855 * also mark the vdev config dirty, so that the new faulted state is
2856 * written out to disk.
2858 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2859 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2862 * When reopening in reponse to a clear event, it may be due to
2863 * a fmadm repair request. In this case, if the device is
2864 * still broken, we want to still post the ereport again.
2866 vd
->vdev_forcefault
= B_TRUE
;
2868 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2869 vd
->vdev_cant_read
= B_FALSE
;
2870 vd
->vdev_cant_write
= B_FALSE
;
2872 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2874 vd
->vdev_forcefault
= B_FALSE
;
2876 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2877 vdev_state_dirty(vd
->vdev_top
);
2879 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2880 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2882 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
2886 * When clearing a FMA-diagnosed fault, we always want to
2887 * unspare the device, as we assume that the original spare was
2888 * done in response to the FMA fault.
2890 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2891 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2892 vd
->vdev_parent
->vdev_child
[0] == vd
)
2893 vd
->vdev_unspare
= B_TRUE
;
2897 vdev_is_dead(vdev_t
*vd
)
2900 * Holes and missing devices are always considered "dead".
2901 * This simplifies the code since we don't have to check for
2902 * these types of devices in the various code paths.
2903 * Instead we rely on the fact that we skip over dead devices
2904 * before issuing I/O to them.
2906 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
2907 vd
->vdev_ops
== &vdev_hole_ops
||
2908 vd
->vdev_ops
== &vdev_missing_ops
);
2912 vdev_readable(vdev_t
*vd
)
2914 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2918 vdev_writeable(vdev_t
*vd
)
2920 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
2921 vdev_is_concrete(vd
));
2925 vdev_allocatable(vdev_t
*vd
)
2927 uint64_t state
= vd
->vdev_state
;
2930 * We currently allow allocations from vdevs which may be in the
2931 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2932 * fails to reopen then we'll catch it later when we're holding
2933 * the proper locks. Note that we have to get the vdev state
2934 * in a local variable because although it changes atomically,
2935 * we're asking two separate questions about it.
2937 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2938 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
2939 vd
->vdev_mg
->mg_initialized
);
2943 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2945 ASSERT(zio
->io_vd
== vd
);
2947 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2950 if (zio
->io_type
== ZIO_TYPE_READ
)
2951 return (!vd
->vdev_cant_read
);
2953 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2954 return (!vd
->vdev_cant_write
);
2960 * Get statistics for the given vdev.
2963 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2965 spa_t
*spa
= vd
->vdev_spa
;
2966 vdev_t
*rvd
= spa
->spa_root_vdev
;
2967 vdev_t
*tvd
= vd
->vdev_top
;
2969 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2971 mutex_enter(&vd
->vdev_stat_lock
);
2972 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2973 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2974 vs
->vs_state
= vd
->vdev_state
;
2975 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2976 if (vd
->vdev_ops
->vdev_op_leaf
)
2977 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2979 * Report expandable space on top-level, non-auxillary devices only.
2980 * The expandable space is reported in terms of metaslab sized units
2981 * since that determines how much space the pool can expand.
2983 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
2984 vs
->vs_esize
= P2ALIGN(vd
->vdev_max_asize
- vd
->vdev_asize
-
2985 spa
->spa_bootsize
, 1ULL << tvd
->vdev_ms_shift
);
2987 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
2988 vdev_is_concrete(vd
)) {
2989 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2993 * If we're getting stats on the root vdev, aggregate the I/O counts
2994 * over all top-level vdevs (i.e. the direct children of the root).
2997 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2998 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2999 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
3001 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
3002 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
3003 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
3005 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
3008 mutex_exit(&vd
->vdev_stat_lock
);
3012 vdev_clear_stats(vdev_t
*vd
)
3014 mutex_enter(&vd
->vdev_stat_lock
);
3015 vd
->vdev_stat
.vs_space
= 0;
3016 vd
->vdev_stat
.vs_dspace
= 0;
3017 vd
->vdev_stat
.vs_alloc
= 0;
3018 mutex_exit(&vd
->vdev_stat_lock
);
3022 vdev_scan_stat_init(vdev_t
*vd
)
3024 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3026 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3027 vdev_scan_stat_init(vd
->vdev_child
[c
]);
3029 mutex_enter(&vd
->vdev_stat_lock
);
3030 vs
->vs_scan_processed
= 0;
3031 mutex_exit(&vd
->vdev_stat_lock
);
3035 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
3037 spa_t
*spa
= zio
->io_spa
;
3038 vdev_t
*rvd
= spa
->spa_root_vdev
;
3039 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
3041 uint64_t txg
= zio
->io_txg
;
3042 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3043 zio_type_t type
= zio
->io_type
;
3044 int flags
= zio
->io_flags
;
3047 * If this i/o is a gang leader, it didn't do any actual work.
3049 if (zio
->io_gang_tree
)
3052 if (zio
->io_error
== 0) {
3054 * If this is a root i/o, don't count it -- we've already
3055 * counted the top-level vdevs, and vdev_get_stats() will
3056 * aggregate them when asked. This reduces contention on
3057 * the root vdev_stat_lock and implicitly handles blocks
3058 * that compress away to holes, for which there is no i/o.
3059 * (Holes never create vdev children, so all the counters
3060 * remain zero, which is what we want.)
3062 * Note: this only applies to successful i/o (io_error == 0)
3063 * because unlike i/o counts, errors are not additive.
3064 * When reading a ditto block, for example, failure of
3065 * one top-level vdev does not imply a root-level error.
3070 ASSERT(vd
== zio
->io_vd
);
3072 if (flags
& ZIO_FLAG_IO_BYPASS
)
3075 mutex_enter(&vd
->vdev_stat_lock
);
3077 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3078 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3079 dsl_scan_phys_t
*scn_phys
=
3080 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3081 uint64_t *processed
= &scn_phys
->scn_processed
;
3084 if (vd
->vdev_ops
->vdev_op_leaf
)
3085 atomic_add_64(processed
, psize
);
3086 vs
->vs_scan_processed
+= psize
;
3089 if (flags
& ZIO_FLAG_SELF_HEAL
)
3090 vs
->vs_self_healed
+= psize
;
3094 vs
->vs_bytes
[type
] += psize
;
3096 mutex_exit(&vd
->vdev_stat_lock
);
3100 if (flags
& ZIO_FLAG_SPECULATIVE
)
3104 * If this is an I/O error that is going to be retried, then ignore the
3105 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3106 * hard errors, when in reality they can happen for any number of
3107 * innocuous reasons (bus resets, MPxIO link failure, etc).
3109 if (zio
->io_error
== EIO
&&
3110 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3114 * Intent logs writes won't propagate their error to the root
3115 * I/O so don't mark these types of failures as pool-level
3118 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3121 mutex_enter(&vd
->vdev_stat_lock
);
3122 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3123 if (zio
->io_error
== ECKSUM
)
3124 vs
->vs_checksum_errors
++;
3126 vs
->vs_read_errors
++;
3128 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3129 vs
->vs_write_errors
++;
3130 mutex_exit(&vd
->vdev_stat_lock
);
3132 if (spa
->spa_load_state
== SPA_LOAD_NONE
&&
3133 type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3134 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3135 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3136 spa
->spa_claiming
)) {
3138 * This is either a normal write (not a repair), or it's
3139 * a repair induced by the scrub thread, or it's a repair
3140 * made by zil_claim() during spa_load() in the first txg.
3141 * In the normal case, we commit the DTL change in the same
3142 * txg as the block was born. In the scrub-induced repair
3143 * case, we know that scrubs run in first-pass syncing context,
3144 * so we commit the DTL change in spa_syncing_txg(spa).
3145 * In the zil_claim() case, we commit in spa_first_txg(spa).
3147 * We currently do not make DTL entries for failed spontaneous
3148 * self-healing writes triggered by normal (non-scrubbing)
3149 * reads, because we have no transactional context in which to
3150 * do so -- and it's not clear that it'd be desirable anyway.
3152 if (vd
->vdev_ops
->vdev_op_leaf
) {
3153 uint64_t commit_txg
= txg
;
3154 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3155 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3156 ASSERT(spa_sync_pass(spa
) == 1);
3157 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3158 commit_txg
= spa_syncing_txg(spa
);
3159 } else if (spa
->spa_claiming
) {
3160 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3161 commit_txg
= spa_first_txg(spa
);
3163 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3164 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3166 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3167 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3168 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3171 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3176 * Update the in-core space usage stats for this vdev, its metaslab class,
3177 * and the root vdev.
3180 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3181 int64_t space_delta
)
3183 int64_t dspace_delta
= space_delta
;
3184 spa_t
*spa
= vd
->vdev_spa
;
3185 vdev_t
*rvd
= spa
->spa_root_vdev
;
3186 metaslab_group_t
*mg
= vd
->vdev_mg
;
3187 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3189 ASSERT(vd
== vd
->vdev_top
);
3192 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3193 * factor. We must calculate this here and not at the root vdev
3194 * because the root vdev's psize-to-asize is simply the max of its
3195 * childrens', thus not accurate enough for us.
3197 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3198 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3199 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3200 vd
->vdev_deflate_ratio
;
3202 mutex_enter(&vd
->vdev_stat_lock
);
3203 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3204 vd
->vdev_stat
.vs_space
+= space_delta
;
3205 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3206 mutex_exit(&vd
->vdev_stat_lock
);
3208 if (mc
== spa_normal_class(spa
)) {
3209 mutex_enter(&rvd
->vdev_stat_lock
);
3210 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3211 rvd
->vdev_stat
.vs_space
+= space_delta
;
3212 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3213 mutex_exit(&rvd
->vdev_stat_lock
);
3217 ASSERT(rvd
== vd
->vdev_parent
);
3218 ASSERT(vd
->vdev_ms_count
!= 0);
3220 metaslab_class_space_update(mc
,
3221 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3226 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3227 * so that it will be written out next time the vdev configuration is synced.
3228 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3231 vdev_config_dirty(vdev_t
*vd
)
3233 spa_t
*spa
= vd
->vdev_spa
;
3234 vdev_t
*rvd
= spa
->spa_root_vdev
;
3237 ASSERT(spa_writeable(spa
));
3240 * If this is an aux vdev (as with l2cache and spare devices), then we
3241 * update the vdev config manually and set the sync flag.
3243 if (vd
->vdev_aux
!= NULL
) {
3244 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3248 for (c
= 0; c
< sav
->sav_count
; c
++) {
3249 if (sav
->sav_vdevs
[c
] == vd
)
3253 if (c
== sav
->sav_count
) {
3255 * We're being removed. There's nothing more to do.
3257 ASSERT(sav
->sav_sync
== B_TRUE
);
3261 sav
->sav_sync
= B_TRUE
;
3263 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3264 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3265 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3266 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3272 * Setting the nvlist in the middle if the array is a little
3273 * sketchy, but it will work.
3275 nvlist_free(aux
[c
]);
3276 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3282 * The dirty list is protected by the SCL_CONFIG lock. The caller
3283 * must either hold SCL_CONFIG as writer, or must be the sync thread
3284 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3285 * so this is sufficient to ensure mutual exclusion.
3287 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3288 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3289 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3292 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3293 vdev_config_dirty(rvd
->vdev_child
[c
]);
3295 ASSERT(vd
== vd
->vdev_top
);
3297 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3298 vdev_is_concrete(vd
)) {
3299 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3305 vdev_config_clean(vdev_t
*vd
)
3307 spa_t
*spa
= vd
->vdev_spa
;
3309 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3310 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3311 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3313 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3314 list_remove(&spa
->spa_config_dirty_list
, vd
);
3318 * Mark a top-level vdev's state as dirty, so that the next pass of
3319 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3320 * the state changes from larger config changes because they require
3321 * much less locking, and are often needed for administrative actions.
3324 vdev_state_dirty(vdev_t
*vd
)
3326 spa_t
*spa
= vd
->vdev_spa
;
3328 ASSERT(spa_writeable(spa
));
3329 ASSERT(vd
== vd
->vdev_top
);
3332 * The state list is protected by the SCL_STATE lock. The caller
3333 * must either hold SCL_STATE as writer, or must be the sync thread
3334 * (which holds SCL_STATE as reader). There's only one sync thread,
3335 * so this is sufficient to ensure mutual exclusion.
3337 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3338 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3339 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3341 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
3342 vdev_is_concrete(vd
))
3343 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3347 vdev_state_clean(vdev_t
*vd
)
3349 spa_t
*spa
= vd
->vdev_spa
;
3351 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3352 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3353 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3355 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3356 list_remove(&spa
->spa_state_dirty_list
, vd
);
3360 * Propagate vdev state up from children to parent.
3363 vdev_propagate_state(vdev_t
*vd
)
3365 spa_t
*spa
= vd
->vdev_spa
;
3366 vdev_t
*rvd
= spa
->spa_root_vdev
;
3367 int degraded
= 0, faulted
= 0;
3371 if (vd
->vdev_children
> 0) {
3372 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3373 child
= vd
->vdev_child
[c
];
3376 * Don't factor holes or indirect vdevs into the
3379 if (!vdev_is_concrete(child
))
3382 if (!vdev_readable(child
) ||
3383 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3385 * Root special: if there is a top-level log
3386 * device, treat the root vdev as if it were
3389 if (child
->vdev_islog
&& vd
== rvd
)
3393 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3397 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3401 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3404 * Root special: if there is a top-level vdev that cannot be
3405 * opened due to corrupted metadata, then propagate the root
3406 * vdev's aux state as 'corrupt' rather than 'insufficient
3409 if (corrupted
&& vd
== rvd
&&
3410 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3411 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3412 VDEV_AUX_CORRUPT_DATA
);
3415 if (vd
->vdev_parent
)
3416 vdev_propagate_state(vd
->vdev_parent
);
3420 * Set a vdev's state. If this is during an open, we don't update the parent
3421 * state, because we're in the process of opening children depth-first.
3422 * Otherwise, we propagate the change to the parent.
3424 * If this routine places a device in a faulted state, an appropriate ereport is
3428 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3430 uint64_t save_state
;
3431 spa_t
*spa
= vd
->vdev_spa
;
3433 if (state
== vd
->vdev_state
) {
3434 vd
->vdev_stat
.vs_aux
= aux
;
3438 save_state
= vd
->vdev_state
;
3440 vd
->vdev_state
= state
;
3441 vd
->vdev_stat
.vs_aux
= aux
;
3444 * If we are setting the vdev state to anything but an open state, then
3445 * always close the underlying device unless the device has requested
3446 * a delayed close (i.e. we're about to remove or fault the device).
3447 * Otherwise, we keep accessible but invalid devices open forever.
3448 * We don't call vdev_close() itself, because that implies some extra
3449 * checks (offline, etc) that we don't want here. This is limited to
3450 * leaf devices, because otherwise closing the device will affect other
3453 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3454 vd
->vdev_ops
->vdev_op_leaf
)
3455 vd
->vdev_ops
->vdev_op_close(vd
);
3458 * If we have brought this vdev back into service, we need
3459 * to notify fmd so that it can gracefully repair any outstanding
3460 * cases due to a missing device. We do this in all cases, even those
3461 * that probably don't correlate to a repaired fault. This is sure to
3462 * catch all cases, and we let the zfs-retire agent sort it out. If
3463 * this is a transient state it's OK, as the retire agent will
3464 * double-check the state of the vdev before repairing it.
3466 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
3467 vd
->vdev_prevstate
!= state
)
3468 zfs_post_state_change(spa
, vd
);
3470 if (vd
->vdev_removed
&&
3471 state
== VDEV_STATE_CANT_OPEN
&&
3472 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3474 * If the previous state is set to VDEV_STATE_REMOVED, then this
3475 * device was previously marked removed and someone attempted to
3476 * reopen it. If this failed due to a nonexistent device, then
3477 * keep the device in the REMOVED state. We also let this be if
3478 * it is one of our special test online cases, which is only
3479 * attempting to online the device and shouldn't generate an FMA
3482 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3483 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3484 } else if (state
== VDEV_STATE_REMOVED
) {
3485 vd
->vdev_removed
= B_TRUE
;
3486 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3488 * If we fail to open a vdev during an import or recovery, we
3489 * mark it as "not available", which signifies that it was
3490 * never there to begin with. Failure to open such a device
3491 * is not considered an error.
3493 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3494 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3495 vd
->vdev_ops
->vdev_op_leaf
)
3496 vd
->vdev_not_present
= 1;
3499 * Post the appropriate ereport. If the 'prevstate' field is
3500 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3501 * that this is part of a vdev_reopen(). In this case, we don't
3502 * want to post the ereport if the device was already in the
3503 * CANT_OPEN state beforehand.
3505 * If the 'checkremove' flag is set, then this is an attempt to
3506 * online the device in response to an insertion event. If we
3507 * hit this case, then we have detected an insertion event for a
3508 * faulted or offline device that wasn't in the removed state.
3509 * In this scenario, we don't post an ereport because we are
3510 * about to replace the device, or attempt an online with
3511 * vdev_forcefault, which will generate the fault for us.
3513 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3514 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3515 vd
!= spa
->spa_root_vdev
) {
3519 case VDEV_AUX_OPEN_FAILED
:
3520 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3522 case VDEV_AUX_CORRUPT_DATA
:
3523 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3525 case VDEV_AUX_NO_REPLICAS
:
3526 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3528 case VDEV_AUX_BAD_GUID_SUM
:
3529 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3531 case VDEV_AUX_TOO_SMALL
:
3532 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3534 case VDEV_AUX_BAD_LABEL
:
3535 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3538 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3541 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3544 /* Erase any notion of persistent removed state */
3545 vd
->vdev_removed
= B_FALSE
;
3547 vd
->vdev_removed
= B_FALSE
;
3550 if (!isopen
&& vd
->vdev_parent
)
3551 vdev_propagate_state(vd
->vdev_parent
);
3555 * Check the vdev configuration to ensure that it's capable of supporting
3556 * a root pool. We do not support partial configuration.
3557 * In addition, only a single top-level vdev is allowed.
3560 vdev_is_bootable(vdev_t
*vd
)
3562 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3563 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3565 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3566 vd
->vdev_children
> 1) {
3568 } else if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0 ||
3569 strcmp(vdev_type
, VDEV_TYPE_INDIRECT
) == 0) {
3574 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3575 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3582 vdev_is_concrete(vdev_t
*vd
)
3584 vdev_ops_t
*ops
= vd
->vdev_ops
;
3585 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
3586 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
3594 * Load the state from the original vdev tree (ovd) which
3595 * we've retrieved from the MOS config object. If the original
3596 * vdev was offline or faulted then we transfer that state to the
3597 * device in the current vdev tree (nvd).
3600 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3602 spa_t
*spa
= nvd
->vdev_spa
;
3604 ASSERT(nvd
->vdev_top
->vdev_islog
);
3605 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3606 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3608 for (int c
= 0; c
< nvd
->vdev_children
; c
++)
3609 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3611 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3613 * Restore the persistent vdev state
3615 nvd
->vdev_offline
= ovd
->vdev_offline
;
3616 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3617 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3618 nvd
->vdev_removed
= ovd
->vdev_removed
;
3623 * Determine if a log device has valid content. If the vdev was
3624 * removed or faulted in the MOS config then we know that
3625 * the content on the log device has already been written to the pool.
3628 vdev_log_state_valid(vdev_t
*vd
)
3630 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3634 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3635 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3642 * Expand a vdev if possible.
3645 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3647 ASSERT(vd
->vdev_top
== vd
);
3648 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3650 vdev_set_deflate_ratio(vd
);
3652 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
3653 vdev_is_concrete(vd
)) {
3654 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3655 vdev_config_dirty(vd
);
3663 vdev_split(vdev_t
*vd
)
3665 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3667 vdev_remove_child(pvd
, vd
);
3668 vdev_compact_children(pvd
);
3670 cvd
= pvd
->vdev_child
[0];
3671 if (pvd
->vdev_children
== 1) {
3672 vdev_remove_parent(cvd
);
3673 cvd
->vdev_splitting
= B_TRUE
;
3675 vdev_propagate_state(cvd
);
3679 vdev_deadman(vdev_t
*vd
)
3681 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3682 vdev_t
*cvd
= vd
->vdev_child
[c
];
3687 if (vd
->vdev_ops
->vdev_op_leaf
) {
3688 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3690 mutex_enter(&vq
->vq_lock
);
3691 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3692 spa_t
*spa
= vd
->vdev_spa
;
3697 * Look at the head of all the pending queues,
3698 * if any I/O has been outstanding for longer than
3699 * the spa_deadman_synctime we panic the system.
3701 fio
= avl_first(&vq
->vq_active_tree
);
3702 delta
= gethrtime() - fio
->io_timestamp
;
3703 if (delta
> spa_deadman_synctime(spa
)) {
3704 vdev_dbgmsg(vd
, "SLOW IO: zio timestamp "
3705 "%lluns, delta %lluns, last io %lluns",
3706 fio
->io_timestamp
, (u_longlong_t
)delta
,
3707 vq
->vq_io_complete_ts
);
3708 fm_panic("I/O to pool '%s' appears to be "
3709 "hung.", spa_name(spa
));
3712 mutex_exit(&vq
->vq_lock
);