4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
25 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Toomas Soome <tsoome@me.com>
30 #include <sys/zfs_context.h>
31 #include <sys/fm/fs/zfs.h>
33 #include <sys/spa_impl.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/vdev_impl.h>
37 #include <sys/uberblock_impl.h>
38 #include <sys/metaslab.h>
39 #include <sys/metaslab_impl.h>
40 #include <sys/space_map.h>
41 #include <sys/space_reftree.h>
44 #include <sys/fs/zfs.h>
47 #include <sys/dsl_scan.h>
50 * Virtual device management.
53 static vdev_ops_t
*vdev_ops_table
[] = {
66 /* maximum scrub/resilver I/O queue per leaf vdev */
67 int zfs_scrub_limit
= 10;
70 * When a vdev is added, it will be divided into approximately (but no
71 * more than) this number of metaslabs.
73 int metaslabs_per_vdev
= 200;
76 * Given a vdev type, return the appropriate ops vector.
79 vdev_getops(const char *type
)
81 vdev_ops_t
*ops
, **opspp
;
83 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
84 if (strcmp(ops
->vdev_op_type
, type
) == 0)
91 * Default asize function: return the MAX of psize with the asize of
92 * all children. This is what's used by anything other than RAID-Z.
95 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
97 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
100 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
101 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
102 asize
= MAX(asize
, csize
);
109 * Get the minimum allocatable size. We define the allocatable size as
110 * the vdev's asize rounded to the nearest metaslab. This allows us to
111 * replace or attach devices which don't have the same physical size but
112 * can still satisfy the same number of allocations.
115 vdev_get_min_asize(vdev_t
*vd
)
117 vdev_t
*pvd
= vd
->vdev_parent
;
120 * If our parent is NULL (inactive spare or cache) or is the root,
121 * just return our own asize.
124 return (vd
->vdev_asize
);
127 * The top-level vdev just returns the allocatable size rounded
128 * to the nearest metaslab.
130 if (vd
== vd
->vdev_top
)
131 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
134 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
135 * so each child must provide at least 1/Nth of its asize.
137 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
138 return (pvd
->vdev_min_asize
/ pvd
->vdev_children
);
140 return (pvd
->vdev_min_asize
);
144 vdev_set_min_asize(vdev_t
*vd
)
146 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
148 for (int c
= 0; c
< vd
->vdev_children
; c
++)
149 vdev_set_min_asize(vd
->vdev_child
[c
]);
153 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
155 vdev_t
*rvd
= spa
->spa_root_vdev
;
157 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
159 if (vdev
< rvd
->vdev_children
) {
160 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
161 return (rvd
->vdev_child
[vdev
]);
168 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
172 if (vd
->vdev_guid
== guid
)
175 for (int c
= 0; c
< vd
->vdev_children
; c
++)
176 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
184 vdev_count_leaves_impl(vdev_t
*vd
)
188 if (vd
->vdev_ops
->vdev_op_leaf
)
191 for (int c
= 0; c
< vd
->vdev_children
; c
++)
192 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
198 vdev_count_leaves(spa_t
*spa
)
200 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
204 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
206 size_t oldsize
, newsize
;
207 uint64_t id
= cvd
->vdev_id
;
209 spa_t
*spa
= cvd
->vdev_spa
;
211 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
212 ASSERT(cvd
->vdev_parent
== NULL
);
214 cvd
->vdev_parent
= pvd
;
219 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
221 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
222 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
223 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
225 newchild
= kmem_zalloc(newsize
, KM_SLEEP
);
226 if (pvd
->vdev_child
!= NULL
) {
227 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
228 kmem_free(pvd
->vdev_child
, oldsize
);
231 pvd
->vdev_child
= newchild
;
232 pvd
->vdev_child
[id
] = cvd
;
234 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
235 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
238 * Walk up all ancestors to update guid sum.
240 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
241 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
245 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
248 uint_t id
= cvd
->vdev_id
;
250 ASSERT(cvd
->vdev_parent
== pvd
);
255 ASSERT(id
< pvd
->vdev_children
);
256 ASSERT(pvd
->vdev_child
[id
] == cvd
);
258 pvd
->vdev_child
[id
] = NULL
;
259 cvd
->vdev_parent
= NULL
;
261 for (c
= 0; c
< pvd
->vdev_children
; c
++)
262 if (pvd
->vdev_child
[c
])
265 if (c
== pvd
->vdev_children
) {
266 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
267 pvd
->vdev_child
= NULL
;
268 pvd
->vdev_children
= 0;
272 * Walk up all ancestors to update guid sum.
274 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
275 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
279 * Remove any holes in the child array.
282 vdev_compact_children(vdev_t
*pvd
)
284 vdev_t
**newchild
, *cvd
;
285 int oldc
= pvd
->vdev_children
;
288 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
290 for (int c
= newc
= 0; c
< oldc
; c
++)
291 if (pvd
->vdev_child
[c
])
294 newchild
= kmem_alloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
296 for (int c
= newc
= 0; c
< oldc
; c
++) {
297 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
298 newchild
[newc
] = cvd
;
299 cvd
->vdev_id
= newc
++;
303 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
304 pvd
->vdev_child
= newchild
;
305 pvd
->vdev_children
= newc
;
309 * Allocate and minimally initialize a vdev_t.
312 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
316 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
318 if (spa
->spa_root_vdev
== NULL
) {
319 ASSERT(ops
== &vdev_root_ops
);
320 spa
->spa_root_vdev
= vd
;
321 spa
->spa_load_guid
= spa_generate_guid(NULL
);
324 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
325 if (spa
->spa_root_vdev
== vd
) {
327 * The root vdev's guid will also be the pool guid,
328 * which must be unique among all pools.
330 guid
= spa_generate_guid(NULL
);
333 * Any other vdev's guid must be unique within the pool.
335 guid
= spa_generate_guid(spa
);
337 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
342 vd
->vdev_guid
= guid
;
343 vd
->vdev_guid_sum
= guid
;
345 vd
->vdev_state
= VDEV_STATE_CLOSED
;
346 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
348 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
349 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
350 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
351 for (int t
= 0; t
< DTL_TYPES
; t
++) {
352 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
,
355 txg_list_create(&vd
->vdev_ms_list
,
356 offsetof(struct metaslab
, ms_txg_node
));
357 txg_list_create(&vd
->vdev_dtl_list
,
358 offsetof(struct vdev
, vdev_dtl_node
));
359 vd
->vdev_stat
.vs_timestamp
= gethrtime();
367 * Allocate a new vdev. The 'alloctype' is used to control whether we are
368 * creating a new vdev or loading an existing one - the behavior is slightly
369 * different for each case.
372 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
377 uint64_t guid
= 0, islog
, nparity
;
380 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
382 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
383 return (SET_ERROR(EINVAL
));
385 if ((ops
= vdev_getops(type
)) == NULL
)
386 return (SET_ERROR(EINVAL
));
389 * If this is a load, get the vdev guid from the nvlist.
390 * Otherwise, vdev_alloc_common() will generate one for us.
392 if (alloctype
== VDEV_ALLOC_LOAD
) {
395 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
397 return (SET_ERROR(EINVAL
));
399 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
400 return (SET_ERROR(EINVAL
));
401 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
402 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
403 return (SET_ERROR(EINVAL
));
404 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
405 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
406 return (SET_ERROR(EINVAL
));
407 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
408 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
409 return (SET_ERROR(EINVAL
));
413 * The first allocated vdev must be of type 'root'.
415 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
416 return (SET_ERROR(EINVAL
));
419 * Determine whether we're a log vdev.
422 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
423 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
424 return (SET_ERROR(ENOTSUP
));
426 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
427 return (SET_ERROR(ENOTSUP
));
430 * Set the nparity property for RAID-Z vdevs.
433 if (ops
== &vdev_raidz_ops
) {
434 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
436 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
437 return (SET_ERROR(EINVAL
));
439 * Previous versions could only support 1 or 2 parity
443 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
444 return (SET_ERROR(ENOTSUP
));
446 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
447 return (SET_ERROR(ENOTSUP
));
450 * We require the parity to be specified for SPAs that
451 * support multiple parity levels.
453 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
454 return (SET_ERROR(EINVAL
));
456 * Otherwise, we default to 1 parity device for RAID-Z.
463 ASSERT(nparity
!= -1ULL);
465 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
467 vd
->vdev_islog
= islog
;
468 vd
->vdev_nparity
= nparity
;
470 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
471 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
472 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
473 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
474 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
475 &vd
->vdev_physpath
) == 0)
476 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
477 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
478 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
481 * Set the whole_disk property. If it's not specified, leave the value
484 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
485 &vd
->vdev_wholedisk
) != 0)
486 vd
->vdev_wholedisk
= -1ULL;
489 * Look for the 'not present' flag. This will only be set if the device
490 * was not present at the time of import.
492 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
493 &vd
->vdev_not_present
);
496 * Get the alignment requirement.
498 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
501 * Retrieve the vdev creation time.
503 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
507 * If we're a top-level vdev, try to load the allocation parameters.
509 if (parent
&& !parent
->vdev_parent
&&
510 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
511 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
513 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
515 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
517 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
519 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
522 ASSERT0(vd
->vdev_top_zap
);
525 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
526 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
527 alloctype
== VDEV_ALLOC_ADD
||
528 alloctype
== VDEV_ALLOC_SPLIT
||
529 alloctype
== VDEV_ALLOC_ROOTPOOL
);
530 vd
->vdev_mg
= metaslab_group_create(islog
?
531 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
534 if (vd
->vdev_ops
->vdev_op_leaf
&&
535 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
536 (void) nvlist_lookup_uint64(nv
,
537 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
539 ASSERT0(vd
->vdev_leaf_zap
);
543 * If we're a leaf vdev, try to load the DTL object and other state.
546 if (vd
->vdev_ops
->vdev_op_leaf
&&
547 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
548 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
549 if (alloctype
== VDEV_ALLOC_LOAD
) {
550 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
551 &vd
->vdev_dtl_object
);
552 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
556 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
559 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
560 &spare
) == 0 && spare
)
564 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
567 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
568 &vd
->vdev_resilver_txg
);
571 * When importing a pool, we want to ignore the persistent fault
572 * state, as the diagnosis made on another system may not be
573 * valid in the current context. Local vdevs will
574 * remain in the faulted state.
576 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
577 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
579 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
581 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
584 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
588 VDEV_AUX_ERR_EXCEEDED
;
589 if (nvlist_lookup_string(nv
,
590 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
591 strcmp(aux
, "external") == 0)
592 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
598 * Add ourselves to the parent's list of children.
600 vdev_add_child(parent
, vd
);
608 vdev_free(vdev_t
*vd
)
610 spa_t
*spa
= vd
->vdev_spa
;
613 * vdev_free() implies closing the vdev first. This is simpler than
614 * trying to ensure complicated semantics for all callers.
618 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
619 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
624 for (int c
= 0; c
< vd
->vdev_children
; c
++)
625 vdev_free(vd
->vdev_child
[c
]);
627 ASSERT(vd
->vdev_child
== NULL
);
628 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
631 * Discard allocation state.
633 if (vd
->vdev_mg
!= NULL
) {
634 vdev_metaslab_fini(vd
);
635 metaslab_group_destroy(vd
->vdev_mg
);
638 ASSERT0(vd
->vdev_stat
.vs_space
);
639 ASSERT0(vd
->vdev_stat
.vs_dspace
);
640 ASSERT0(vd
->vdev_stat
.vs_alloc
);
643 * Remove this vdev from its parent's child list.
645 vdev_remove_child(vd
->vdev_parent
, vd
);
647 ASSERT(vd
->vdev_parent
== NULL
);
650 * Clean up vdev structure.
656 spa_strfree(vd
->vdev_path
);
658 spa_strfree(vd
->vdev_devid
);
659 if (vd
->vdev_physpath
)
660 spa_strfree(vd
->vdev_physpath
);
662 spa_strfree(vd
->vdev_fru
);
664 if (vd
->vdev_isspare
)
665 spa_spare_remove(vd
);
666 if (vd
->vdev_isl2cache
)
667 spa_l2cache_remove(vd
);
669 txg_list_destroy(&vd
->vdev_ms_list
);
670 txg_list_destroy(&vd
->vdev_dtl_list
);
672 mutex_enter(&vd
->vdev_dtl_lock
);
673 space_map_close(vd
->vdev_dtl_sm
);
674 for (int t
= 0; t
< DTL_TYPES
; t
++) {
675 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
676 range_tree_destroy(vd
->vdev_dtl
[t
]);
678 mutex_exit(&vd
->vdev_dtl_lock
);
680 mutex_destroy(&vd
->vdev_dtl_lock
);
681 mutex_destroy(&vd
->vdev_stat_lock
);
682 mutex_destroy(&vd
->vdev_probe_lock
);
684 if (vd
== spa
->spa_root_vdev
)
685 spa
->spa_root_vdev
= NULL
;
687 kmem_free(vd
, sizeof (vdev_t
));
691 * Transfer top-level vdev state from svd to tvd.
694 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
696 spa_t
*spa
= svd
->vdev_spa
;
701 ASSERT(tvd
== tvd
->vdev_top
);
703 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
704 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
705 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
706 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
708 svd
->vdev_ms_array
= 0;
709 svd
->vdev_ms_shift
= 0;
710 svd
->vdev_ms_count
= 0;
711 svd
->vdev_top_zap
= 0;
714 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
715 tvd
->vdev_mg
= svd
->vdev_mg
;
716 tvd
->vdev_ms
= svd
->vdev_ms
;
721 if (tvd
->vdev_mg
!= NULL
)
722 tvd
->vdev_mg
->mg_vd
= tvd
;
724 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
725 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
726 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
728 svd
->vdev_stat
.vs_alloc
= 0;
729 svd
->vdev_stat
.vs_space
= 0;
730 svd
->vdev_stat
.vs_dspace
= 0;
732 for (t
= 0; t
< TXG_SIZE
; t
++) {
733 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
734 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
735 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
736 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
737 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
738 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
741 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
742 vdev_config_clean(svd
);
743 vdev_config_dirty(tvd
);
746 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
747 vdev_state_clean(svd
);
748 vdev_state_dirty(tvd
);
751 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
752 svd
->vdev_deflate_ratio
= 0;
754 tvd
->vdev_islog
= svd
->vdev_islog
;
759 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
766 for (int c
= 0; c
< vd
->vdev_children
; c
++)
767 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
771 * Add a mirror/replacing vdev above an existing vdev.
774 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
776 spa_t
*spa
= cvd
->vdev_spa
;
777 vdev_t
*pvd
= cvd
->vdev_parent
;
780 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
782 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
784 mvd
->vdev_asize
= cvd
->vdev_asize
;
785 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
786 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
787 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
788 mvd
->vdev_state
= cvd
->vdev_state
;
789 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
791 vdev_remove_child(pvd
, cvd
);
792 vdev_add_child(pvd
, mvd
);
793 cvd
->vdev_id
= mvd
->vdev_children
;
794 vdev_add_child(mvd
, cvd
);
795 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
797 if (mvd
== mvd
->vdev_top
)
798 vdev_top_transfer(cvd
, mvd
);
804 * Remove a 1-way mirror/replacing vdev from the tree.
807 vdev_remove_parent(vdev_t
*cvd
)
809 vdev_t
*mvd
= cvd
->vdev_parent
;
810 vdev_t
*pvd
= mvd
->vdev_parent
;
812 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
814 ASSERT(mvd
->vdev_children
== 1);
815 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
816 mvd
->vdev_ops
== &vdev_replacing_ops
||
817 mvd
->vdev_ops
== &vdev_spare_ops
);
818 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
820 vdev_remove_child(mvd
, cvd
);
821 vdev_remove_child(pvd
, mvd
);
824 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
825 * Otherwise, we could have detached an offline device, and when we
826 * go to import the pool we'll think we have two top-level vdevs,
827 * instead of a different version of the same top-level vdev.
829 if (mvd
->vdev_top
== mvd
) {
830 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
831 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
832 cvd
->vdev_guid
+= guid_delta
;
833 cvd
->vdev_guid_sum
+= guid_delta
;
835 cvd
->vdev_id
= mvd
->vdev_id
;
836 vdev_add_child(pvd
, cvd
);
837 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
839 if (cvd
== cvd
->vdev_top
)
840 vdev_top_transfer(mvd
, cvd
);
842 ASSERT(mvd
->vdev_children
== 0);
847 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
849 spa_t
*spa
= vd
->vdev_spa
;
850 objset_t
*mos
= spa
->spa_meta_objset
;
852 uint64_t oldc
= vd
->vdev_ms_count
;
853 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
857 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
860 * This vdev is not being allocated from yet or is a hole.
862 if (vd
->vdev_ms_shift
== 0)
865 ASSERT(!vd
->vdev_ishole
);
868 * Compute the raidz-deflation ratio. Note, we hard-code
869 * in 128k (1 << 17) because it is the "typical" blocksize.
870 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
871 * otherwise it would inconsistently account for existing bp's.
873 vd
->vdev_deflate_ratio
= (1 << 17) /
874 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
876 ASSERT(oldc
<= newc
);
878 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
881 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
882 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
886 vd
->vdev_ms_count
= newc
;
888 for (m
= oldc
; m
< newc
; m
++) {
892 error
= dmu_read(mos
, vd
->vdev_ms_array
,
893 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
899 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
906 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
909 * If the vdev is being removed we don't activate
910 * the metaslabs since we want to ensure that no new
911 * allocations are performed on this device.
913 if (oldc
== 0 && !vd
->vdev_removing
)
914 metaslab_group_activate(vd
->vdev_mg
);
917 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
923 vdev_metaslab_fini(vdev_t
*vd
)
926 uint64_t count
= vd
->vdev_ms_count
;
928 if (vd
->vdev_ms
!= NULL
) {
929 metaslab_group_passivate(vd
->vdev_mg
);
930 for (m
= 0; m
< count
; m
++) {
931 metaslab_t
*msp
= vd
->vdev_ms
[m
];
936 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
941 typedef struct vdev_probe_stats
{
942 boolean_t vps_readable
;
943 boolean_t vps_writeable
;
945 } vdev_probe_stats_t
;
948 vdev_probe_done(zio_t
*zio
)
950 spa_t
*spa
= zio
->io_spa
;
951 vdev_t
*vd
= zio
->io_vd
;
952 vdev_probe_stats_t
*vps
= zio
->io_private
;
954 ASSERT(vd
->vdev_probe_zio
!= NULL
);
956 if (zio
->io_type
== ZIO_TYPE_READ
) {
957 if (zio
->io_error
== 0)
958 vps
->vps_readable
= 1;
959 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
960 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
961 zio
->io_offset
, zio
->io_size
, zio
->io_data
,
962 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
963 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
965 zio_buf_free(zio
->io_data
, zio
->io_size
);
967 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
968 if (zio
->io_error
== 0)
969 vps
->vps_writeable
= 1;
970 zio_buf_free(zio
->io_data
, zio
->io_size
);
971 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
974 vd
->vdev_cant_read
|= !vps
->vps_readable
;
975 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
977 if (vdev_readable(vd
) &&
978 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
981 ASSERT(zio
->io_error
!= 0);
982 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
983 spa
, vd
, NULL
, 0, 0);
984 zio
->io_error
= SET_ERROR(ENXIO
);
987 mutex_enter(&vd
->vdev_probe_lock
);
988 ASSERT(vd
->vdev_probe_zio
== zio
);
989 vd
->vdev_probe_zio
= NULL
;
990 mutex_exit(&vd
->vdev_probe_lock
);
992 while ((pio
= zio_walk_parents(zio
)) != NULL
)
993 if (!vdev_accessible(vd
, pio
))
994 pio
->io_error
= SET_ERROR(ENXIO
);
996 kmem_free(vps
, sizeof (*vps
));
1001 * Determine whether this device is accessible.
1003 * Read and write to several known locations: the pad regions of each
1004 * vdev label but the first, which we leave alone in case it contains
1008 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1010 spa_t
*spa
= vd
->vdev_spa
;
1011 vdev_probe_stats_t
*vps
= NULL
;
1014 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1017 * Don't probe the probe.
1019 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1023 * To prevent 'probe storms' when a device fails, we create
1024 * just one probe i/o at a time. All zios that want to probe
1025 * this vdev will become parents of the probe io.
1027 mutex_enter(&vd
->vdev_probe_lock
);
1029 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1030 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1032 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1033 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1036 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1038 * vdev_cant_read and vdev_cant_write can only
1039 * transition from TRUE to FALSE when we have the
1040 * SCL_ZIO lock as writer; otherwise they can only
1041 * transition from FALSE to TRUE. This ensures that
1042 * any zio looking at these values can assume that
1043 * failures persist for the life of the I/O. That's
1044 * important because when a device has intermittent
1045 * connectivity problems, we want to ensure that
1046 * they're ascribed to the device (ENXIO) and not
1049 * Since we hold SCL_ZIO as writer here, clear both
1050 * values so the probe can reevaluate from first
1053 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1054 vd
->vdev_cant_read
= B_FALSE
;
1055 vd
->vdev_cant_write
= B_FALSE
;
1058 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1059 vdev_probe_done
, vps
,
1060 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1063 * We can't change the vdev state in this context, so we
1064 * kick off an async task to do it on our behalf.
1067 vd
->vdev_probe_wanted
= B_TRUE
;
1068 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1073 zio_add_child(zio
, pio
);
1075 mutex_exit(&vd
->vdev_probe_lock
);
1078 ASSERT(zio
!= NULL
);
1082 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1083 zio_nowait(zio_read_phys(pio
, vd
,
1084 vdev_label_offset(vd
->vdev_psize
, l
,
1085 offsetof(vdev_label_t
, vl_pad2
)),
1086 VDEV_PAD_SIZE
, zio_buf_alloc(VDEV_PAD_SIZE
),
1087 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1088 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1099 vdev_open_child(void *arg
)
1103 vd
->vdev_open_thread
= curthread
;
1104 vd
->vdev_open_error
= vdev_open(vd
);
1105 vd
->vdev_open_thread
= NULL
;
1109 vdev_uses_zvols(vdev_t
*vd
)
1111 if (vd
->vdev_path
&& strncmp(vd
->vdev_path
, ZVOL_DIR
,
1112 strlen(ZVOL_DIR
)) == 0)
1114 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1115 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1121 vdev_open_children(vdev_t
*vd
)
1124 int children
= vd
->vdev_children
;
1127 * in order to handle pools on top of zvols, do the opens
1128 * in a single thread so that the same thread holds the
1129 * spa_namespace_lock
1131 if (vdev_uses_zvols(vd
)) {
1132 for (int c
= 0; c
< children
; c
++)
1133 vd
->vdev_child
[c
]->vdev_open_error
=
1134 vdev_open(vd
->vdev_child
[c
]);
1137 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1138 children
, children
, TASKQ_PREPOPULATE
);
1140 for (int c
= 0; c
< children
; c
++)
1141 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1148 * Prepare a virtual device for access.
1151 vdev_open(vdev_t
*vd
)
1153 spa_t
*spa
= vd
->vdev_spa
;
1156 uint64_t max_osize
= 0;
1157 uint64_t asize
, max_asize
, psize
;
1158 uint64_t ashift
= 0;
1160 ASSERT(vd
->vdev_open_thread
== curthread
||
1161 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1162 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1163 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1164 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1166 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1167 vd
->vdev_cant_read
= B_FALSE
;
1168 vd
->vdev_cant_write
= B_FALSE
;
1169 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1172 * If this vdev is not removed, check its fault status. If it's
1173 * faulted, bail out of the open.
1175 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1176 ASSERT(vd
->vdev_children
== 0);
1177 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1178 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1179 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1180 vd
->vdev_label_aux
);
1181 return (SET_ERROR(ENXIO
));
1182 } else if (vd
->vdev_offline
) {
1183 ASSERT(vd
->vdev_children
== 0);
1184 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1185 return (SET_ERROR(ENXIO
));
1188 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1191 * Reset the vdev_reopening flag so that we actually close
1192 * the vdev on error.
1194 vd
->vdev_reopening
= B_FALSE
;
1195 if (zio_injection_enabled
&& error
== 0)
1196 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1199 if (vd
->vdev_removed
&&
1200 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1201 vd
->vdev_removed
= B_FALSE
;
1203 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1204 vd
->vdev_stat
.vs_aux
);
1208 vd
->vdev_removed
= B_FALSE
;
1211 * Recheck the faulted flag now that we have confirmed that
1212 * the vdev is accessible. If we're faulted, bail.
1214 if (vd
->vdev_faulted
) {
1215 ASSERT(vd
->vdev_children
== 0);
1216 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1217 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1218 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1219 vd
->vdev_label_aux
);
1220 return (SET_ERROR(ENXIO
));
1223 if (vd
->vdev_degraded
) {
1224 ASSERT(vd
->vdev_children
== 0);
1225 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1226 VDEV_AUX_ERR_EXCEEDED
);
1228 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1232 * For hole or missing vdevs we just return success.
1234 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1237 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1238 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1239 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1245 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1246 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1248 if (vd
->vdev_children
== 0) {
1249 if (osize
< SPA_MINDEVSIZE
) {
1250 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1251 VDEV_AUX_TOO_SMALL
);
1252 return (SET_ERROR(EOVERFLOW
));
1255 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1256 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1257 VDEV_LABEL_END_SIZE
);
1259 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1260 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1261 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1262 VDEV_AUX_TOO_SMALL
);
1263 return (SET_ERROR(EOVERFLOW
));
1267 max_asize
= max_osize
;
1270 vd
->vdev_psize
= psize
;
1273 * Make sure the allocatable size hasn't shrunk.
1275 if (asize
< vd
->vdev_min_asize
) {
1276 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1277 VDEV_AUX_BAD_LABEL
);
1278 return (SET_ERROR(EINVAL
));
1281 if (vd
->vdev_asize
== 0) {
1283 * This is the first-ever open, so use the computed values.
1284 * For testing purposes, a higher ashift can be requested.
1286 vd
->vdev_asize
= asize
;
1287 vd
->vdev_max_asize
= max_asize
;
1288 vd
->vdev_ashift
= MAX(ashift
, vd
->vdev_ashift
);
1291 * Detect if the alignment requirement has increased.
1292 * We don't want to make the pool unavailable, just
1293 * issue a warning instead.
1295 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1296 vd
->vdev_ops
->vdev_op_leaf
) {
1298 "Disk, '%s', has a block alignment that is "
1299 "larger than the pool's alignment\n",
1302 vd
->vdev_max_asize
= max_asize
;
1306 * If all children are healthy and the asize has increased,
1307 * then we've experienced dynamic LUN growth. If automatic
1308 * expansion is enabled then use the additional space.
1310 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&& asize
> vd
->vdev_asize
&&
1311 (vd
->vdev_expanding
|| spa
->spa_autoexpand
))
1312 vd
->vdev_asize
= asize
;
1314 vdev_set_min_asize(vd
);
1317 * Ensure we can issue some IO before declaring the
1318 * vdev open for business.
1320 if (vd
->vdev_ops
->vdev_op_leaf
&&
1321 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1322 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1323 VDEV_AUX_ERR_EXCEEDED
);
1328 * Track the min and max ashift values for normal data devices.
1330 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1331 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1332 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1333 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1334 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1335 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1339 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1340 * resilver. But don't do this if we are doing a reopen for a scrub,
1341 * since this would just restart the scrub we are already doing.
1343 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1344 vdev_resilver_needed(vd
, NULL
, NULL
))
1345 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1351 * Called once the vdevs are all opened, this routine validates the label
1352 * contents. This needs to be done before vdev_load() so that we don't
1353 * inadvertently do repair I/Os to the wrong device.
1355 * If 'strict' is false ignore the spa guid check. This is necessary because
1356 * if the machine crashed during a re-guid the new guid might have been written
1357 * to all of the vdev labels, but not the cached config. The strict check
1358 * will be performed when the pool is opened again using the mos config.
1360 * This function will only return failure if one of the vdevs indicates that it
1361 * has since been destroyed or exported. This is only possible if
1362 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1363 * will be updated but the function will return 0.
1366 vdev_validate(vdev_t
*vd
, boolean_t strict
)
1368 spa_t
*spa
= vd
->vdev_spa
;
1370 uint64_t guid
= 0, top_guid
;
1373 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1374 if (vdev_validate(vd
->vdev_child
[c
], strict
) != 0)
1375 return (SET_ERROR(EBADF
));
1378 * If the device has already failed, or was marked offline, don't do
1379 * any further validation. Otherwise, label I/O will fail and we will
1380 * overwrite the previous state.
1382 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1383 uint64_t aux_guid
= 0;
1385 uint64_t txg
= spa_last_synced_txg(spa
) != 0 ?
1386 spa_last_synced_txg(spa
) : -1ULL;
1388 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1389 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1390 VDEV_AUX_BAD_LABEL
);
1395 * Determine if this vdev has been split off into another
1396 * pool. If so, then refuse to open it.
1398 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1399 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1400 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1401 VDEV_AUX_SPLIT_POOL
);
1406 if (strict
&& (nvlist_lookup_uint64(label
,
1407 ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0 ||
1408 guid
!= spa_guid(spa
))) {
1409 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1410 VDEV_AUX_CORRUPT_DATA
);
1415 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1416 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1421 * If this vdev just became a top-level vdev because its
1422 * sibling was detached, it will have adopted the parent's
1423 * vdev guid -- but the label may or may not be on disk yet.
1424 * Fortunately, either version of the label will have the
1425 * same top guid, so if we're a top-level vdev, we can
1426 * safely compare to that instead.
1428 * If we split this vdev off instead, then we also check the
1429 * original pool's guid. We don't want to consider the vdev
1430 * corrupt if it is partway through a split operation.
1432 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
1434 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
,
1436 ((vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) &&
1437 (vd
->vdev_guid
!= top_guid
|| vd
!= vd
->vdev_top
))) {
1438 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1439 VDEV_AUX_CORRUPT_DATA
);
1444 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1446 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1447 VDEV_AUX_CORRUPT_DATA
);
1455 * If this is a verbatim import, no need to check the
1456 * state of the pool.
1458 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1459 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1460 state
!= POOL_STATE_ACTIVE
)
1461 return (SET_ERROR(EBADF
));
1464 * If we were able to open and validate a vdev that was
1465 * previously marked permanently unavailable, clear that state
1468 if (vd
->vdev_not_present
)
1469 vd
->vdev_not_present
= 0;
1476 * Close a virtual device.
1479 vdev_close(vdev_t
*vd
)
1481 spa_t
*spa
= vd
->vdev_spa
;
1482 vdev_t
*pvd
= vd
->vdev_parent
;
1484 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1487 * If our parent is reopening, then we are as well, unless we are
1490 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1491 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1493 vd
->vdev_ops
->vdev_op_close(vd
);
1495 vdev_cache_purge(vd
);
1498 * We record the previous state before we close it, so that if we are
1499 * doing a reopen(), we don't generate FMA ereports if we notice that
1500 * it's still faulted.
1502 vd
->vdev_prevstate
= vd
->vdev_state
;
1504 if (vd
->vdev_offline
)
1505 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1507 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1508 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1512 vdev_hold(vdev_t
*vd
)
1514 spa_t
*spa
= vd
->vdev_spa
;
1516 ASSERT(spa_is_root(spa
));
1517 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1520 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1521 vdev_hold(vd
->vdev_child
[c
]);
1523 if (vd
->vdev_ops
->vdev_op_leaf
)
1524 vd
->vdev_ops
->vdev_op_hold(vd
);
1528 vdev_rele(vdev_t
*vd
)
1530 spa_t
*spa
= vd
->vdev_spa
;
1532 ASSERT(spa_is_root(spa
));
1533 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1534 vdev_rele(vd
->vdev_child
[c
]);
1536 if (vd
->vdev_ops
->vdev_op_leaf
)
1537 vd
->vdev_ops
->vdev_op_rele(vd
);
1541 * Reopen all interior vdevs and any unopened leaves. We don't actually
1542 * reopen leaf vdevs which had previously been opened as they might deadlock
1543 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1544 * If the leaf has never been opened then open it, as usual.
1547 vdev_reopen(vdev_t
*vd
)
1549 spa_t
*spa
= vd
->vdev_spa
;
1551 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1553 /* set the reopening flag unless we're taking the vdev offline */
1554 vd
->vdev_reopening
= !vd
->vdev_offline
;
1556 (void) vdev_open(vd
);
1559 * Call vdev_validate() here to make sure we have the same device.
1560 * Otherwise, a device with an invalid label could be successfully
1561 * opened in response to vdev_reopen().
1564 (void) vdev_validate_aux(vd
);
1565 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1566 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1567 !l2arc_vdev_present(vd
))
1568 l2arc_add_vdev(spa
, vd
);
1570 (void) vdev_validate(vd
, B_TRUE
);
1574 * Reassess parent vdev's health.
1576 vdev_propagate_state(vd
);
1580 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1585 * Normally, partial opens (e.g. of a mirror) are allowed.
1586 * For a create, however, we want to fail the request if
1587 * there are any components we can't open.
1589 error
= vdev_open(vd
);
1591 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1593 return (error
? error
: ENXIO
);
1597 * Recursively load DTLs and initialize all labels.
1599 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1600 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1601 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1610 vdev_metaslab_set_size(vdev_t
*vd
)
1613 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1615 vd
->vdev_ms_shift
= highbit64(vd
->vdev_asize
/ metaslabs_per_vdev
);
1616 vd
->vdev_ms_shift
= MAX(vd
->vdev_ms_shift
, SPA_MAXBLOCKSHIFT
);
1620 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
1622 ASSERT(vd
== vd
->vdev_top
);
1623 ASSERT(!vd
->vdev_ishole
);
1624 ASSERT(ISP2(flags
));
1625 ASSERT(spa_writeable(vd
->vdev_spa
));
1627 if (flags
& VDD_METASLAB
)
1628 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
1630 if (flags
& VDD_DTL
)
1631 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
1633 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
1637 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
1639 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1640 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
1642 if (vd
->vdev_ops
->vdev_op_leaf
)
1643 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
1649 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1650 * the vdev has less than perfect replication. There are four kinds of DTL:
1652 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1654 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1656 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1657 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1658 * txgs that was scrubbed.
1660 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1661 * persistent errors or just some device being offline.
1662 * Unlike the other three, the DTL_OUTAGE map is not generally
1663 * maintained; it's only computed when needed, typically to
1664 * determine whether a device can be detached.
1666 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1667 * either has the data or it doesn't.
1669 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1670 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1671 * if any child is less than fully replicated, then so is its parent.
1672 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1673 * comprising only those txgs which appear in 'maxfaults' or more children;
1674 * those are the txgs we don't have enough replication to read. For example,
1675 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1676 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1677 * two child DTL_MISSING maps.
1679 * It should be clear from the above that to compute the DTLs and outage maps
1680 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1681 * Therefore, that is all we keep on disk. When loading the pool, or after
1682 * a configuration change, we generate all other DTLs from first principles.
1685 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1687 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1689 ASSERT(t
< DTL_TYPES
);
1690 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1691 ASSERT(spa_writeable(vd
->vdev_spa
));
1693 mutex_enter(rt
->rt_lock
);
1694 if (!range_tree_contains(rt
, txg
, size
))
1695 range_tree_add(rt
, txg
, size
);
1696 mutex_exit(rt
->rt_lock
);
1700 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
1702 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1703 boolean_t dirty
= B_FALSE
;
1705 ASSERT(t
< DTL_TYPES
);
1706 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
1708 mutex_enter(rt
->rt_lock
);
1709 if (range_tree_space(rt
) != 0)
1710 dirty
= range_tree_contains(rt
, txg
, size
);
1711 mutex_exit(rt
->rt_lock
);
1717 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
1719 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
1722 mutex_enter(rt
->rt_lock
);
1723 empty
= (range_tree_space(rt
) == 0);
1724 mutex_exit(rt
->rt_lock
);
1730 * Returns the lowest txg in the DTL range.
1733 vdev_dtl_min(vdev_t
*vd
)
1737 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1738 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1739 ASSERT0(vd
->vdev_children
);
1741 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1742 return (rs
->rs_start
- 1);
1746 * Returns the highest txg in the DTL.
1749 vdev_dtl_max(vdev_t
*vd
)
1753 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
1754 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
1755 ASSERT0(vd
->vdev_children
);
1757 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
1758 return (rs
->rs_end
);
1762 * Determine if a resilvering vdev should remove any DTL entries from
1763 * its range. If the vdev was resilvering for the entire duration of the
1764 * scan then it should excise that range from its DTLs. Otherwise, this
1765 * vdev is considered partially resilvered and should leave its DTL
1766 * entries intact. The comment in vdev_dtl_reassess() describes how we
1770 vdev_dtl_should_excise(vdev_t
*vd
)
1772 spa_t
*spa
= vd
->vdev_spa
;
1773 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1775 ASSERT0(scn
->scn_phys
.scn_errors
);
1776 ASSERT0(vd
->vdev_children
);
1778 if (vd
->vdev_resilver_txg
== 0 ||
1779 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0)
1783 * When a resilver is initiated the scan will assign the scn_max_txg
1784 * value to the highest txg value that exists in all DTLs. If this
1785 * device's max DTL is not part of this scan (i.e. it is not in
1786 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1789 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
1790 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
1791 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
1792 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
1799 * Reassess DTLs after a config change or scrub completion.
1802 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
1804 spa_t
*spa
= vd
->vdev_spa
;
1808 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1810 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1811 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
1812 scrub_txg
, scrub_done
);
1814 if (vd
== spa
->spa_root_vdev
|| vd
->vdev_ishole
|| vd
->vdev_aux
)
1817 if (vd
->vdev_ops
->vdev_op_leaf
) {
1818 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
1820 mutex_enter(&vd
->vdev_dtl_lock
);
1823 * If we've completed a scan cleanly then determine
1824 * if this vdev should remove any DTLs. We only want to
1825 * excise regions on vdevs that were available during
1826 * the entire duration of this scan.
1828 if (scrub_txg
!= 0 &&
1829 (spa
->spa_scrub_started
||
1830 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
1831 vdev_dtl_should_excise(vd
)) {
1833 * We completed a scrub up to scrub_txg. If we
1834 * did it without rebooting, then the scrub dtl
1835 * will be valid, so excise the old region and
1836 * fold in the scrub dtl. Otherwise, leave the
1837 * dtl as-is if there was an error.
1839 * There's little trick here: to excise the beginning
1840 * of the DTL_MISSING map, we put it into a reference
1841 * tree and then add a segment with refcnt -1 that
1842 * covers the range [0, scrub_txg). This means
1843 * that each txg in that range has refcnt -1 or 0.
1844 * We then add DTL_SCRUB with a refcnt of 2, so that
1845 * entries in the range [0, scrub_txg) will have a
1846 * positive refcnt -- either 1 or 2. We then convert
1847 * the reference tree into the new DTL_MISSING map.
1849 space_reftree_create(&reftree
);
1850 space_reftree_add_map(&reftree
,
1851 vd
->vdev_dtl
[DTL_MISSING
], 1);
1852 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
1853 space_reftree_add_map(&reftree
,
1854 vd
->vdev_dtl
[DTL_SCRUB
], 2);
1855 space_reftree_generate_map(&reftree
,
1856 vd
->vdev_dtl
[DTL_MISSING
], 1);
1857 space_reftree_destroy(&reftree
);
1859 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
1860 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1861 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
1863 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
1864 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
1865 if (!vdev_readable(vd
))
1866 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
1868 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
1869 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
1872 * If the vdev was resilvering and no longer has any
1873 * DTLs then reset its resilvering flag.
1875 if (vd
->vdev_resilver_txg
!= 0 &&
1876 range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) == 0 &&
1877 range_tree_space(vd
->vdev_dtl
[DTL_OUTAGE
]) == 0)
1878 vd
->vdev_resilver_txg
= 0;
1880 mutex_exit(&vd
->vdev_dtl_lock
);
1883 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
1887 mutex_enter(&vd
->vdev_dtl_lock
);
1888 for (int t
= 0; t
< DTL_TYPES
; t
++) {
1889 /* account for child's outage in parent's missing map */
1890 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
1892 continue; /* leaf vdevs only */
1893 if (t
== DTL_PARTIAL
)
1894 minref
= 1; /* i.e. non-zero */
1895 else if (vd
->vdev_nparity
!= 0)
1896 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
1898 minref
= vd
->vdev_children
; /* any kind of mirror */
1899 space_reftree_create(&reftree
);
1900 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1901 vdev_t
*cvd
= vd
->vdev_child
[c
];
1902 mutex_enter(&cvd
->vdev_dtl_lock
);
1903 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
1904 mutex_exit(&cvd
->vdev_dtl_lock
);
1906 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
1907 space_reftree_destroy(&reftree
);
1909 mutex_exit(&vd
->vdev_dtl_lock
);
1913 vdev_dtl_load(vdev_t
*vd
)
1915 spa_t
*spa
= vd
->vdev_spa
;
1916 objset_t
*mos
= spa
->spa_meta_objset
;
1919 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
1920 ASSERT(!vd
->vdev_ishole
);
1922 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
1923 vd
->vdev_dtl_object
, 0, -1ULL, 0, &vd
->vdev_dtl_lock
);
1926 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
1928 mutex_enter(&vd
->vdev_dtl_lock
);
1931 * Now that we've opened the space_map we need to update
1934 space_map_update(vd
->vdev_dtl_sm
);
1936 error
= space_map_load(vd
->vdev_dtl_sm
,
1937 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
1938 mutex_exit(&vd
->vdev_dtl_lock
);
1943 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1944 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
1953 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
1955 spa_t
*spa
= vd
->vdev_spa
;
1957 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
1958 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
1963 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
1965 spa_t
*spa
= vd
->vdev_spa
;
1966 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
1967 DMU_OT_NONE
, 0, tx
);
1970 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
1977 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
1979 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
1980 vd
->vdev_ops
!= &vdev_missing_ops
&&
1981 vd
->vdev_ops
!= &vdev_root_ops
&&
1982 !vd
->vdev_top
->vdev_removing
) {
1983 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
1984 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
1986 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
1987 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
1990 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
1991 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
1996 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
1998 spa_t
*spa
= vd
->vdev_spa
;
1999 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2000 objset_t
*mos
= spa
->spa_meta_objset
;
2001 range_tree_t
*rtsync
;
2004 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2006 ASSERT(!vd
->vdev_ishole
);
2007 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2009 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2011 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2012 mutex_enter(&vd
->vdev_dtl_lock
);
2013 space_map_free(vd
->vdev_dtl_sm
, tx
);
2014 space_map_close(vd
->vdev_dtl_sm
);
2015 vd
->vdev_dtl_sm
= NULL
;
2016 mutex_exit(&vd
->vdev_dtl_lock
);
2019 * We only destroy the leaf ZAP for detached leaves or for
2020 * removed log devices. Removed data devices handle leaf ZAP
2021 * cleanup later, once cancellation is no longer possible.
2023 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2024 vd
->vdev_top
->vdev_islog
)) {
2025 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2026 vd
->vdev_leaf_zap
= 0;
2033 if (vd
->vdev_dtl_sm
== NULL
) {
2034 uint64_t new_object
;
2036 new_object
= space_map_alloc(mos
, tx
);
2037 VERIFY3U(new_object
, !=, 0);
2039 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2040 0, -1ULL, 0, &vd
->vdev_dtl_lock
));
2041 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2044 mutex_init(&rtlock
, NULL
, MUTEX_DEFAULT
, NULL
);
2046 rtsync
= range_tree_create(NULL
, NULL
, &rtlock
);
2048 mutex_enter(&rtlock
);
2050 mutex_enter(&vd
->vdev_dtl_lock
);
2051 range_tree_walk(rt
, range_tree_add
, rtsync
);
2052 mutex_exit(&vd
->vdev_dtl_lock
);
2054 space_map_truncate(vd
->vdev_dtl_sm
, tx
);
2055 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, tx
);
2056 range_tree_vacate(rtsync
, NULL
, NULL
);
2058 range_tree_destroy(rtsync
);
2060 mutex_exit(&rtlock
);
2061 mutex_destroy(&rtlock
);
2064 * If the object for the space map has changed then dirty
2065 * the top level so that we update the config.
2067 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2068 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2069 "new object %llu", txg
, spa_name(spa
), object
,
2070 space_map_object(vd
->vdev_dtl_sm
));
2071 vdev_config_dirty(vd
->vdev_top
);
2076 mutex_enter(&vd
->vdev_dtl_lock
);
2077 space_map_update(vd
->vdev_dtl_sm
);
2078 mutex_exit(&vd
->vdev_dtl_lock
);
2082 * Determine whether the specified vdev can be offlined/detached/removed
2083 * without losing data.
2086 vdev_dtl_required(vdev_t
*vd
)
2088 spa_t
*spa
= vd
->vdev_spa
;
2089 vdev_t
*tvd
= vd
->vdev_top
;
2090 uint8_t cant_read
= vd
->vdev_cant_read
;
2093 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2095 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2099 * Temporarily mark the device as unreadable, and then determine
2100 * whether this results in any DTL outages in the top-level vdev.
2101 * If not, we can safely offline/detach/remove the device.
2103 vd
->vdev_cant_read
= B_TRUE
;
2104 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2105 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2106 vd
->vdev_cant_read
= cant_read
;
2107 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2109 if (!required
&& zio_injection_enabled
)
2110 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2116 * Determine if resilver is needed, and if so the txg range.
2119 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2121 boolean_t needed
= B_FALSE
;
2122 uint64_t thismin
= UINT64_MAX
;
2123 uint64_t thismax
= 0;
2125 if (vd
->vdev_children
== 0) {
2126 mutex_enter(&vd
->vdev_dtl_lock
);
2127 if (range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]) != 0 &&
2128 vdev_writeable(vd
)) {
2130 thismin
= vdev_dtl_min(vd
);
2131 thismax
= vdev_dtl_max(vd
);
2134 mutex_exit(&vd
->vdev_dtl_lock
);
2136 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2137 vdev_t
*cvd
= vd
->vdev_child
[c
];
2138 uint64_t cmin
, cmax
;
2140 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2141 thismin
= MIN(thismin
, cmin
);
2142 thismax
= MAX(thismax
, cmax
);
2148 if (needed
&& minp
) {
2156 vdev_load(vdev_t
*vd
)
2159 * Recursively load all children.
2161 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2162 vdev_load(vd
->vdev_child
[c
]);
2165 * If this is a top-level vdev, initialize its metaslabs.
2167 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&&
2168 (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0 ||
2169 vdev_metaslab_init(vd
, 0) != 0))
2170 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2171 VDEV_AUX_CORRUPT_DATA
);
2174 * If this is a leaf vdev, load its DTL.
2176 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_dtl_load(vd
) != 0)
2177 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2178 VDEV_AUX_CORRUPT_DATA
);
2182 * The special vdev case is used for hot spares and l2cache devices. Its
2183 * sole purpose it to set the vdev state for the associated vdev. To do this,
2184 * we make sure that we can open the underlying device, then try to read the
2185 * label, and make sure that the label is sane and that it hasn't been
2186 * repurposed to another pool.
2189 vdev_validate_aux(vdev_t
*vd
)
2192 uint64_t guid
, version
;
2195 if (!vdev_readable(vd
))
2198 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2199 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2200 VDEV_AUX_CORRUPT_DATA
);
2204 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2205 !SPA_VERSION_IS_SUPPORTED(version
) ||
2206 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2207 guid
!= vd
->vdev_guid
||
2208 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2209 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2210 VDEV_AUX_CORRUPT_DATA
);
2216 * We don't actually check the pool state here. If it's in fact in
2217 * use by another pool, we update this fact on the fly when requested.
2224 vdev_remove(vdev_t
*vd
, uint64_t txg
)
2226 spa_t
*spa
= vd
->vdev_spa
;
2227 objset_t
*mos
= spa
->spa_meta_objset
;
2230 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2231 ASSERT(vd
== vd
->vdev_top
);
2232 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2234 if (vd
->vdev_ms
!= NULL
) {
2235 metaslab_group_t
*mg
= vd
->vdev_mg
;
2237 metaslab_group_histogram_verify(mg
);
2238 metaslab_class_histogram_verify(mg
->mg_class
);
2240 for (int m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2241 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2243 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2246 mutex_enter(&msp
->ms_lock
);
2248 * If the metaslab was not loaded when the vdev
2249 * was removed then the histogram accounting may
2250 * not be accurate. Update the histogram information
2251 * here so that we ensure that the metaslab group
2252 * and metaslab class are up-to-date.
2254 metaslab_group_histogram_remove(mg
, msp
);
2256 VERIFY0(space_map_allocated(msp
->ms_sm
));
2257 space_map_free(msp
->ms_sm
, tx
);
2258 space_map_close(msp
->ms_sm
);
2260 mutex_exit(&msp
->ms_lock
);
2263 metaslab_group_histogram_verify(mg
);
2264 metaslab_class_histogram_verify(mg
->mg_class
);
2265 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2266 ASSERT0(mg
->mg_histogram
[i
]);
2270 if (vd
->vdev_ms_array
) {
2271 (void) dmu_object_free(mos
, vd
->vdev_ms_array
, tx
);
2272 vd
->vdev_ms_array
= 0;
2275 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2276 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2277 vd
->vdev_top_zap
= 0;
2283 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2286 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2288 ASSERT(!vd
->vdev_ishole
);
2290 while (msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
2291 metaslab_sync_done(msp
, txg
);
2294 metaslab_sync_reassess(vd
->vdev_mg
);
2298 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2300 spa_t
*spa
= vd
->vdev_spa
;
2305 ASSERT(!vd
->vdev_ishole
);
2307 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0) {
2308 ASSERT(vd
== vd
->vdev_top
);
2309 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2310 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2311 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2312 ASSERT(vd
->vdev_ms_array
!= 0);
2313 vdev_config_dirty(vd
);
2318 * Remove the metadata associated with this vdev once it's empty.
2320 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
2321 vdev_remove(vd
, txg
);
2323 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2324 metaslab_sync(msp
, txg
);
2325 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2328 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2329 vdev_dtl_sync(lvd
, txg
);
2331 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2335 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2337 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2341 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2342 * not be opened, and no I/O is attempted.
2345 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2349 spa_vdev_state_enter(spa
, SCL_NONE
);
2351 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2352 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2354 if (!vd
->vdev_ops
->vdev_op_leaf
)
2355 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2360 * We don't directly use the aux state here, but if we do a
2361 * vdev_reopen(), we need this value to be present to remember why we
2364 vd
->vdev_label_aux
= aux
;
2367 * Faulted state takes precedence over degraded.
2369 vd
->vdev_delayed_close
= B_FALSE
;
2370 vd
->vdev_faulted
= 1ULL;
2371 vd
->vdev_degraded
= 0ULL;
2372 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2375 * If this device has the only valid copy of the data, then
2376 * back off and simply mark the vdev as degraded instead.
2378 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2379 vd
->vdev_degraded
= 1ULL;
2380 vd
->vdev_faulted
= 0ULL;
2383 * If we reopen the device and it's not dead, only then do we
2388 if (vdev_readable(vd
))
2389 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2392 return (spa_vdev_state_exit(spa
, vd
, 0));
2396 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2397 * user that something is wrong. The vdev continues to operate as normal as far
2398 * as I/O is concerned.
2401 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2405 spa_vdev_state_enter(spa
, SCL_NONE
);
2407 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2408 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2410 if (!vd
->vdev_ops
->vdev_op_leaf
)
2411 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2414 * If the vdev is already faulted, then don't do anything.
2416 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2417 return (spa_vdev_state_exit(spa
, NULL
, 0));
2419 vd
->vdev_degraded
= 1ULL;
2420 if (!vdev_is_dead(vd
))
2421 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2424 return (spa_vdev_state_exit(spa
, vd
, 0));
2428 * Online the given vdev.
2430 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2431 * spare device should be detached when the device finishes resilvering.
2432 * Second, the online should be treated like a 'test' online case, so no FMA
2433 * events are generated if the device fails to open.
2436 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2438 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2439 boolean_t postevent
= B_FALSE
;
2441 spa_vdev_state_enter(spa
, SCL_NONE
);
2443 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2444 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2446 if (!vd
->vdev_ops
->vdev_op_leaf
)
2447 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2450 (vd
->vdev_offline
== B_TRUE
|| vd
->vdev_tmpoffline
== B_TRUE
) ?
2454 vd
->vdev_offline
= B_FALSE
;
2455 vd
->vdev_tmpoffline
= B_FALSE
;
2456 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
2457 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
2459 /* XXX - L2ARC 1.0 does not support expansion */
2460 if (!vd
->vdev_aux
) {
2461 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2462 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
2466 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
2468 if (!vd
->vdev_aux
) {
2469 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2470 pvd
->vdev_expanding
= B_FALSE
;
2474 *newstate
= vd
->vdev_state
;
2475 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
2476 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
2477 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2478 vd
->vdev_parent
->vdev_child
[0] == vd
)
2479 vd
->vdev_unspare
= B_TRUE
;
2481 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
2483 /* XXX - L2ARC 1.0 does not support expansion */
2485 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
2486 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
2490 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_ONLINE
);
2492 return (spa_vdev_state_exit(spa
, vd
, 0));
2496 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2500 uint64_t generation
;
2501 metaslab_group_t
*mg
;
2504 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2506 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2507 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2509 if (!vd
->vdev_ops
->vdev_op_leaf
)
2510 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2514 generation
= spa
->spa_config_generation
+ 1;
2517 * If the device isn't already offline, try to offline it.
2519 if (!vd
->vdev_offline
) {
2521 * If this device has the only valid copy of some data,
2522 * don't allow it to be offlined. Log devices are always
2525 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2526 vdev_dtl_required(vd
))
2527 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2530 * If the top-level is a slog and it has had allocations
2531 * then proceed. We check that the vdev's metaslab group
2532 * is not NULL since it's possible that we may have just
2533 * added this vdev but not yet initialized its metaslabs.
2535 if (tvd
->vdev_islog
&& mg
!= NULL
) {
2537 * Prevent any future allocations.
2539 metaslab_group_passivate(mg
);
2540 (void) spa_vdev_state_exit(spa
, vd
, 0);
2542 error
= spa_offline_log(spa
);
2544 spa_vdev_state_enter(spa
, SCL_ALLOC
);
2547 * Check to see if the config has changed.
2549 if (error
|| generation
!= spa
->spa_config_generation
) {
2550 metaslab_group_activate(mg
);
2552 return (spa_vdev_state_exit(spa
,
2554 (void) spa_vdev_state_exit(spa
, vd
, 0);
2557 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
2561 * Offline this device and reopen its top-level vdev.
2562 * If the top-level vdev is a log device then just offline
2563 * it. Otherwise, if this action results in the top-level
2564 * vdev becoming unusable, undo it and fail the request.
2566 vd
->vdev_offline
= B_TRUE
;
2569 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
2570 vdev_is_dead(tvd
)) {
2571 vd
->vdev_offline
= B_FALSE
;
2573 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
2577 * Add the device back into the metaslab rotor so that
2578 * once we online the device it's open for business.
2580 if (tvd
->vdev_islog
&& mg
!= NULL
)
2581 metaslab_group_activate(mg
);
2584 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
2586 return (spa_vdev_state_exit(spa
, vd
, 0));
2590 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
2594 mutex_enter(&spa
->spa_vdev_top_lock
);
2595 error
= vdev_offline_locked(spa
, guid
, flags
);
2596 mutex_exit(&spa
->spa_vdev_top_lock
);
2602 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2603 * vdev_offline(), we assume the spa config is locked. We also clear all
2604 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2607 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
2609 vdev_t
*rvd
= spa
->spa_root_vdev
;
2611 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2616 vd
->vdev_stat
.vs_read_errors
= 0;
2617 vd
->vdev_stat
.vs_write_errors
= 0;
2618 vd
->vdev_stat
.vs_checksum_errors
= 0;
2620 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2621 vdev_clear(spa
, vd
->vdev_child
[c
]);
2624 * If we're in the FAULTED state or have experienced failed I/O, then
2625 * clear the persistent state and attempt to reopen the device. We
2626 * also mark the vdev config dirty, so that the new faulted state is
2627 * written out to disk.
2629 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
2630 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
2633 * When reopening in reponse to a clear event, it may be due to
2634 * a fmadm repair request. In this case, if the device is
2635 * still broken, we want to still post the ereport again.
2637 vd
->vdev_forcefault
= B_TRUE
;
2639 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
2640 vd
->vdev_cant_read
= B_FALSE
;
2641 vd
->vdev_cant_write
= B_FALSE
;
2643 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
2645 vd
->vdev_forcefault
= B_FALSE
;
2647 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
2648 vdev_state_dirty(vd
->vdev_top
);
2650 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
2651 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
2653 spa_event_notify(spa
, vd
, ESC_ZFS_VDEV_CLEAR
);
2657 * When clearing a FMA-diagnosed fault, we always want to
2658 * unspare the device, as we assume that the original spare was
2659 * done in response to the FMA fault.
2661 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
2662 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
2663 vd
->vdev_parent
->vdev_child
[0] == vd
)
2664 vd
->vdev_unspare
= B_TRUE
;
2668 vdev_is_dead(vdev_t
*vd
)
2671 * Holes and missing devices are always considered "dead".
2672 * This simplifies the code since we don't have to check for
2673 * these types of devices in the various code paths.
2674 * Instead we rely on the fact that we skip over dead devices
2675 * before issuing I/O to them.
2677 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
|| vd
->vdev_ishole
||
2678 vd
->vdev_ops
== &vdev_missing_ops
);
2682 vdev_readable(vdev_t
*vd
)
2684 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
2688 vdev_writeable(vdev_t
*vd
)
2690 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
);
2694 vdev_allocatable(vdev_t
*vd
)
2696 uint64_t state
= vd
->vdev_state
;
2699 * We currently allow allocations from vdevs which may be in the
2700 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2701 * fails to reopen then we'll catch it later when we're holding
2702 * the proper locks. Note that we have to get the vdev state
2703 * in a local variable because although it changes atomically,
2704 * we're asking two separate questions about it.
2706 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
2707 !vd
->vdev_cant_write
&& !vd
->vdev_ishole
);
2711 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
2713 ASSERT(zio
->io_vd
== vd
);
2715 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
2718 if (zio
->io_type
== ZIO_TYPE_READ
)
2719 return (!vd
->vdev_cant_read
);
2721 if (zio
->io_type
== ZIO_TYPE_WRITE
)
2722 return (!vd
->vdev_cant_write
);
2728 * Get statistics for the given vdev.
2731 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
2733 spa_t
*spa
= vd
->vdev_spa
;
2734 vdev_t
*rvd
= spa
->spa_root_vdev
;
2736 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2738 mutex_enter(&vd
->vdev_stat_lock
);
2739 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
2740 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
2741 vs
->vs_state
= vd
->vdev_state
;
2742 vs
->vs_rsize
= vdev_get_min_asize(vd
);
2743 if (vd
->vdev_ops
->vdev_op_leaf
)
2744 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
2745 vs
->vs_esize
= vd
->vdev_max_asize
- vd
->vdev_asize
;
2746 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&& !vd
->vdev_ishole
) {
2747 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
2751 * If we're getting stats on the root vdev, aggregate the I/O counts
2752 * over all top-level vdevs (i.e. the direct children of the root).
2755 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
2756 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2757 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
2759 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
2760 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
2761 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
2763 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
2766 mutex_exit(&vd
->vdev_stat_lock
);
2770 vdev_clear_stats(vdev_t
*vd
)
2772 mutex_enter(&vd
->vdev_stat_lock
);
2773 vd
->vdev_stat
.vs_space
= 0;
2774 vd
->vdev_stat
.vs_dspace
= 0;
2775 vd
->vdev_stat
.vs_alloc
= 0;
2776 mutex_exit(&vd
->vdev_stat_lock
);
2780 vdev_scan_stat_init(vdev_t
*vd
)
2782 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2784 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2785 vdev_scan_stat_init(vd
->vdev_child
[c
]);
2787 mutex_enter(&vd
->vdev_stat_lock
);
2788 vs
->vs_scan_processed
= 0;
2789 mutex_exit(&vd
->vdev_stat_lock
);
2793 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
2795 spa_t
*spa
= zio
->io_spa
;
2796 vdev_t
*rvd
= spa
->spa_root_vdev
;
2797 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
2799 uint64_t txg
= zio
->io_txg
;
2800 vdev_stat_t
*vs
= &vd
->vdev_stat
;
2801 zio_type_t type
= zio
->io_type
;
2802 int flags
= zio
->io_flags
;
2805 * If this i/o is a gang leader, it didn't do any actual work.
2807 if (zio
->io_gang_tree
)
2810 if (zio
->io_error
== 0) {
2812 * If this is a root i/o, don't count it -- we've already
2813 * counted the top-level vdevs, and vdev_get_stats() will
2814 * aggregate them when asked. This reduces contention on
2815 * the root vdev_stat_lock and implicitly handles blocks
2816 * that compress away to holes, for which there is no i/o.
2817 * (Holes never create vdev children, so all the counters
2818 * remain zero, which is what we want.)
2820 * Note: this only applies to successful i/o (io_error == 0)
2821 * because unlike i/o counts, errors are not additive.
2822 * When reading a ditto block, for example, failure of
2823 * one top-level vdev does not imply a root-level error.
2828 ASSERT(vd
== zio
->io_vd
);
2830 if (flags
& ZIO_FLAG_IO_BYPASS
)
2833 mutex_enter(&vd
->vdev_stat_lock
);
2835 if (flags
& ZIO_FLAG_IO_REPAIR
) {
2836 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2837 dsl_scan_phys_t
*scn_phys
=
2838 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
2839 uint64_t *processed
= &scn_phys
->scn_processed
;
2842 if (vd
->vdev_ops
->vdev_op_leaf
)
2843 atomic_add_64(processed
, psize
);
2844 vs
->vs_scan_processed
+= psize
;
2847 if (flags
& ZIO_FLAG_SELF_HEAL
)
2848 vs
->vs_self_healed
+= psize
;
2852 vs
->vs_bytes
[type
] += psize
;
2854 mutex_exit(&vd
->vdev_stat_lock
);
2858 if (flags
& ZIO_FLAG_SPECULATIVE
)
2862 * If this is an I/O error that is going to be retried, then ignore the
2863 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
2864 * hard errors, when in reality they can happen for any number of
2865 * innocuous reasons (bus resets, MPxIO link failure, etc).
2867 if (zio
->io_error
== EIO
&&
2868 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
2872 * Intent logs writes won't propagate their error to the root
2873 * I/O so don't mark these types of failures as pool-level
2876 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
2879 mutex_enter(&vd
->vdev_stat_lock
);
2880 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
2881 if (zio
->io_error
== ECKSUM
)
2882 vs
->vs_checksum_errors
++;
2884 vs
->vs_read_errors
++;
2886 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
2887 vs
->vs_write_errors
++;
2888 mutex_exit(&vd
->vdev_stat_lock
);
2890 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
2891 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
2892 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
2893 spa
->spa_claiming
)) {
2895 * This is either a normal write (not a repair), or it's
2896 * a repair induced by the scrub thread, or it's a repair
2897 * made by zil_claim() during spa_load() in the first txg.
2898 * In the normal case, we commit the DTL change in the same
2899 * txg as the block was born. In the scrub-induced repair
2900 * case, we know that scrubs run in first-pass syncing context,
2901 * so we commit the DTL change in spa_syncing_txg(spa).
2902 * In the zil_claim() case, we commit in spa_first_txg(spa).
2904 * We currently do not make DTL entries for failed spontaneous
2905 * self-healing writes triggered by normal (non-scrubbing)
2906 * reads, because we have no transactional context in which to
2907 * do so -- and it's not clear that it'd be desirable anyway.
2909 if (vd
->vdev_ops
->vdev_op_leaf
) {
2910 uint64_t commit_txg
= txg
;
2911 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
2912 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2913 ASSERT(spa_sync_pass(spa
) == 1);
2914 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
2915 commit_txg
= spa_syncing_txg(spa
);
2916 } else if (spa
->spa_claiming
) {
2917 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
2918 commit_txg
= spa_first_txg(spa
);
2920 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
2921 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
2923 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
2924 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
2925 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
2928 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
2933 * Update the in-core space usage stats for this vdev, its metaslab class,
2934 * and the root vdev.
2937 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
2938 int64_t space_delta
)
2940 int64_t dspace_delta
= space_delta
;
2941 spa_t
*spa
= vd
->vdev_spa
;
2942 vdev_t
*rvd
= spa
->spa_root_vdev
;
2943 metaslab_group_t
*mg
= vd
->vdev_mg
;
2944 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
2946 ASSERT(vd
== vd
->vdev_top
);
2949 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2950 * factor. We must calculate this here and not at the root vdev
2951 * because the root vdev's psize-to-asize is simply the max of its
2952 * childrens', thus not accurate enough for us.
2954 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
2955 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
2956 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
2957 vd
->vdev_deflate_ratio
;
2959 mutex_enter(&vd
->vdev_stat_lock
);
2960 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2961 vd
->vdev_stat
.vs_space
+= space_delta
;
2962 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2963 mutex_exit(&vd
->vdev_stat_lock
);
2965 if (mc
== spa_normal_class(spa
)) {
2966 mutex_enter(&rvd
->vdev_stat_lock
);
2967 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
2968 rvd
->vdev_stat
.vs_space
+= space_delta
;
2969 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
2970 mutex_exit(&rvd
->vdev_stat_lock
);
2974 ASSERT(rvd
== vd
->vdev_parent
);
2975 ASSERT(vd
->vdev_ms_count
!= 0);
2977 metaslab_class_space_update(mc
,
2978 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
2983 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2984 * so that it will be written out next time the vdev configuration is synced.
2985 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2988 vdev_config_dirty(vdev_t
*vd
)
2990 spa_t
*spa
= vd
->vdev_spa
;
2991 vdev_t
*rvd
= spa
->spa_root_vdev
;
2994 ASSERT(spa_writeable(spa
));
2997 * If this is an aux vdev (as with l2cache and spare devices), then we
2998 * update the vdev config manually and set the sync flag.
3000 if (vd
->vdev_aux
!= NULL
) {
3001 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3005 for (c
= 0; c
< sav
->sav_count
; c
++) {
3006 if (sav
->sav_vdevs
[c
] == vd
)
3010 if (c
== sav
->sav_count
) {
3012 * We're being removed. There's nothing more to do.
3014 ASSERT(sav
->sav_sync
== B_TRUE
);
3018 sav
->sav_sync
= B_TRUE
;
3020 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3021 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3022 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3023 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3029 * Setting the nvlist in the middle if the array is a little
3030 * sketchy, but it will work.
3032 nvlist_free(aux
[c
]);
3033 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3039 * The dirty list is protected by the SCL_CONFIG lock. The caller
3040 * must either hold SCL_CONFIG as writer, or must be the sync thread
3041 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3042 * so this is sufficient to ensure mutual exclusion.
3044 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3045 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3046 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3049 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3050 vdev_config_dirty(rvd
->vdev_child
[c
]);
3052 ASSERT(vd
== vd
->vdev_top
);
3054 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3056 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3061 vdev_config_clean(vdev_t
*vd
)
3063 spa_t
*spa
= vd
->vdev_spa
;
3065 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3066 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3067 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3069 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3070 list_remove(&spa
->spa_config_dirty_list
, vd
);
3074 * Mark a top-level vdev's state as dirty, so that the next pass of
3075 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3076 * the state changes from larger config changes because they require
3077 * much less locking, and are often needed for administrative actions.
3080 vdev_state_dirty(vdev_t
*vd
)
3082 spa_t
*spa
= vd
->vdev_spa
;
3084 ASSERT(spa_writeable(spa
));
3085 ASSERT(vd
== vd
->vdev_top
);
3088 * The state list is protected by the SCL_STATE lock. The caller
3089 * must either hold SCL_STATE as writer, or must be the sync thread
3090 * (which holds SCL_STATE as reader). There's only one sync thread,
3091 * so this is sufficient to ensure mutual exclusion.
3093 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3094 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3095 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3097 if (!list_link_active(&vd
->vdev_state_dirty_node
) && !vd
->vdev_ishole
)
3098 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3102 vdev_state_clean(vdev_t
*vd
)
3104 spa_t
*spa
= vd
->vdev_spa
;
3106 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3107 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3108 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3110 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3111 list_remove(&spa
->spa_state_dirty_list
, vd
);
3115 * Propagate vdev state up from children to parent.
3118 vdev_propagate_state(vdev_t
*vd
)
3120 spa_t
*spa
= vd
->vdev_spa
;
3121 vdev_t
*rvd
= spa
->spa_root_vdev
;
3122 int degraded
= 0, faulted
= 0;
3126 if (vd
->vdev_children
> 0) {
3127 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3128 child
= vd
->vdev_child
[c
];
3131 * Don't factor holes into the decision.
3133 if (child
->vdev_ishole
)
3136 if (!vdev_readable(child
) ||
3137 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3139 * Root special: if there is a top-level log
3140 * device, treat the root vdev as if it were
3143 if (child
->vdev_islog
&& vd
== rvd
)
3147 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3151 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3155 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3158 * Root special: if there is a top-level vdev that cannot be
3159 * opened due to corrupted metadata, then propagate the root
3160 * vdev's aux state as 'corrupt' rather than 'insufficient
3163 if (corrupted
&& vd
== rvd
&&
3164 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3165 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3166 VDEV_AUX_CORRUPT_DATA
);
3169 if (vd
->vdev_parent
)
3170 vdev_propagate_state(vd
->vdev_parent
);
3174 * Set a vdev's state. If this is during an open, we don't update the parent
3175 * state, because we're in the process of opening children depth-first.
3176 * Otherwise, we propagate the change to the parent.
3178 * If this routine places a device in a faulted state, an appropriate ereport is
3182 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3184 uint64_t save_state
;
3185 spa_t
*spa
= vd
->vdev_spa
;
3187 if (state
== vd
->vdev_state
) {
3188 vd
->vdev_stat
.vs_aux
= aux
;
3192 save_state
= vd
->vdev_state
;
3194 vd
->vdev_state
= state
;
3195 vd
->vdev_stat
.vs_aux
= aux
;
3198 * If we are setting the vdev state to anything but an open state, then
3199 * always close the underlying device unless the device has requested
3200 * a delayed close (i.e. we're about to remove or fault the device).
3201 * Otherwise, we keep accessible but invalid devices open forever.
3202 * We don't call vdev_close() itself, because that implies some extra
3203 * checks (offline, etc) that we don't want here. This is limited to
3204 * leaf devices, because otherwise closing the device will affect other
3207 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3208 vd
->vdev_ops
->vdev_op_leaf
)
3209 vd
->vdev_ops
->vdev_op_close(vd
);
3212 * If we have brought this vdev back into service, we need
3213 * to notify fmd so that it can gracefully repair any outstanding
3214 * cases due to a missing device. We do this in all cases, even those
3215 * that probably don't correlate to a repaired fault. This is sure to
3216 * catch all cases, and we let the zfs-retire agent sort it out. If
3217 * this is a transient state it's OK, as the retire agent will
3218 * double-check the state of the vdev before repairing it.
3220 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
3221 vd
->vdev_prevstate
!= state
)
3222 zfs_post_state_change(spa
, vd
);
3224 if (vd
->vdev_removed
&&
3225 state
== VDEV_STATE_CANT_OPEN
&&
3226 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3228 * If the previous state is set to VDEV_STATE_REMOVED, then this
3229 * device was previously marked removed and someone attempted to
3230 * reopen it. If this failed due to a nonexistent device, then
3231 * keep the device in the REMOVED state. We also let this be if
3232 * it is one of our special test online cases, which is only
3233 * attempting to online the device and shouldn't generate an FMA
3236 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3237 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3238 } else if (state
== VDEV_STATE_REMOVED
) {
3239 vd
->vdev_removed
= B_TRUE
;
3240 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3242 * If we fail to open a vdev during an import or recovery, we
3243 * mark it as "not available", which signifies that it was
3244 * never there to begin with. Failure to open such a device
3245 * is not considered an error.
3247 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3248 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3249 vd
->vdev_ops
->vdev_op_leaf
)
3250 vd
->vdev_not_present
= 1;
3253 * Post the appropriate ereport. If the 'prevstate' field is
3254 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3255 * that this is part of a vdev_reopen(). In this case, we don't
3256 * want to post the ereport if the device was already in the
3257 * CANT_OPEN state beforehand.
3259 * If the 'checkremove' flag is set, then this is an attempt to
3260 * online the device in response to an insertion event. If we
3261 * hit this case, then we have detected an insertion event for a
3262 * faulted or offline device that wasn't in the removed state.
3263 * In this scenario, we don't post an ereport because we are
3264 * about to replace the device, or attempt an online with
3265 * vdev_forcefault, which will generate the fault for us.
3267 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3268 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3269 vd
!= spa
->spa_root_vdev
) {
3273 case VDEV_AUX_OPEN_FAILED
:
3274 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3276 case VDEV_AUX_CORRUPT_DATA
:
3277 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3279 case VDEV_AUX_NO_REPLICAS
:
3280 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3282 case VDEV_AUX_BAD_GUID_SUM
:
3283 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3285 case VDEV_AUX_TOO_SMALL
:
3286 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3288 case VDEV_AUX_BAD_LABEL
:
3289 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3292 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3295 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3298 /* Erase any notion of persistent removed state */
3299 vd
->vdev_removed
= B_FALSE
;
3301 vd
->vdev_removed
= B_FALSE
;
3304 if (!isopen
&& vd
->vdev_parent
)
3305 vdev_propagate_state(vd
->vdev_parent
);
3309 * Check the vdev configuration to ensure that it's capable of supporting
3310 * a root pool. We do not support partial configuration.
3311 * In addition, only a single top-level vdev is allowed.
3314 vdev_is_bootable(vdev_t
*vd
)
3316 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3317 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3319 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3320 vd
->vdev_children
> 1) {
3322 } else if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0) {
3327 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3328 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3335 * Load the state from the original vdev tree (ovd) which
3336 * we've retrieved from the MOS config object. If the original
3337 * vdev was offline or faulted then we transfer that state to the
3338 * device in the current vdev tree (nvd).
3341 vdev_load_log_state(vdev_t
*nvd
, vdev_t
*ovd
)
3343 spa_t
*spa
= nvd
->vdev_spa
;
3345 ASSERT(nvd
->vdev_top
->vdev_islog
);
3346 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3347 ASSERT3U(nvd
->vdev_guid
, ==, ovd
->vdev_guid
);
3349 for (int c
= 0; c
< nvd
->vdev_children
; c
++)
3350 vdev_load_log_state(nvd
->vdev_child
[c
], ovd
->vdev_child
[c
]);
3352 if (nvd
->vdev_ops
->vdev_op_leaf
) {
3354 * Restore the persistent vdev state
3356 nvd
->vdev_offline
= ovd
->vdev_offline
;
3357 nvd
->vdev_faulted
= ovd
->vdev_faulted
;
3358 nvd
->vdev_degraded
= ovd
->vdev_degraded
;
3359 nvd
->vdev_removed
= ovd
->vdev_removed
;
3364 * Determine if a log device has valid content. If the vdev was
3365 * removed or faulted in the MOS config then we know that
3366 * the content on the log device has already been written to the pool.
3369 vdev_log_state_valid(vdev_t
*vd
)
3371 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3375 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3376 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3383 * Expand a vdev if possible.
3386 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3388 ASSERT(vd
->vdev_top
== vd
);
3389 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3391 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
3392 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
3393 vdev_config_dirty(vd
);
3401 vdev_split(vdev_t
*vd
)
3403 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
3405 vdev_remove_child(pvd
, vd
);
3406 vdev_compact_children(pvd
);
3408 cvd
= pvd
->vdev_child
[0];
3409 if (pvd
->vdev_children
== 1) {
3410 vdev_remove_parent(cvd
);
3411 cvd
->vdev_splitting
= B_TRUE
;
3413 vdev_propagate_state(cvd
);
3417 vdev_deadman(vdev_t
*vd
)
3419 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3420 vdev_t
*cvd
= vd
->vdev_child
[c
];
3425 if (vd
->vdev_ops
->vdev_op_leaf
) {
3426 vdev_queue_t
*vq
= &vd
->vdev_queue
;
3428 mutex_enter(&vq
->vq_lock
);
3429 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
3430 spa_t
*spa
= vd
->vdev_spa
;
3435 * Look at the head of all the pending queues,
3436 * if any I/O has been outstanding for longer than
3437 * the spa_deadman_synctime we panic the system.
3439 fio
= avl_first(&vq
->vq_active_tree
);
3440 delta
= gethrtime() - fio
->io_timestamp
;
3441 if (delta
> spa_deadman_synctime(spa
)) {
3442 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3443 "delta %lluns, last io %lluns",
3444 fio
->io_timestamp
, delta
,
3445 vq
->vq_io_complete_ts
);
3446 fm_panic("I/O to pool '%s' appears to be "
3447 "hung.", spa_name(spa
));
3450 mutex_exit(&vq
->vq_lock
);