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;
74 /* maximum number of metaslabs per top-level vdev */
75 int vdev_max_ms_count
= 200;
77 /* minimum amount of metaslabs per top-level vdev */
78 int vdev_min_ms_count
= 16;
80 /* see comment in vdev_metaslab_set_size() */
81 int vdev_default_ms_shift
= 29;
83 boolean_t vdev_validate_skip
= B_FALSE
;
86 * Since the DTL space map of a vdev is not expected to have a lot of
87 * entries, we default its block size to 4K.
89 int vdev_dtl_sm_blksz
= (1 << 12);
92 * vdev-wide space maps that have lots of entries written to them at
93 * the end of each transaction can benefit from a higher I/O bandwidth
94 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
96 int vdev_standard_sm_blksz
= (1 << 17);
100 vdev_dbgmsg(vdev_t
*vd
, const char *fmt
, ...)
106 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
109 if (vd
->vdev_path
!= NULL
) {
110 zfs_dbgmsg("%s vdev '%s': %s", vd
->vdev_ops
->vdev_op_type
,
113 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
114 vd
->vdev_ops
->vdev_op_type
,
115 (u_longlong_t
)vd
->vdev_id
,
116 (u_longlong_t
)vd
->vdev_guid
, buf
);
121 vdev_dbgmsg_print_tree(vdev_t
*vd
, int indent
)
125 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
) {
126 zfs_dbgmsg("%*svdev %u: %s", indent
, "", vd
->vdev_id
,
127 vd
->vdev_ops
->vdev_op_type
);
131 switch (vd
->vdev_state
) {
132 case VDEV_STATE_UNKNOWN
:
133 (void) snprintf(state
, sizeof (state
), "unknown");
135 case VDEV_STATE_CLOSED
:
136 (void) snprintf(state
, sizeof (state
), "closed");
138 case VDEV_STATE_OFFLINE
:
139 (void) snprintf(state
, sizeof (state
), "offline");
141 case VDEV_STATE_REMOVED
:
142 (void) snprintf(state
, sizeof (state
), "removed");
144 case VDEV_STATE_CANT_OPEN
:
145 (void) snprintf(state
, sizeof (state
), "can't open");
147 case VDEV_STATE_FAULTED
:
148 (void) snprintf(state
, sizeof (state
), "faulted");
150 case VDEV_STATE_DEGRADED
:
151 (void) snprintf(state
, sizeof (state
), "degraded");
153 case VDEV_STATE_HEALTHY
:
154 (void) snprintf(state
, sizeof (state
), "healthy");
157 (void) snprintf(state
, sizeof (state
), "<state %u>",
158 (uint_t
)vd
->vdev_state
);
161 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent
,
162 "", vd
->vdev_id
, vd
->vdev_ops
->vdev_op_type
,
163 vd
->vdev_islog
? " (log)" : "",
164 (u_longlong_t
)vd
->vdev_guid
,
165 vd
->vdev_path
? vd
->vdev_path
: "N/A", state
);
167 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++)
168 vdev_dbgmsg_print_tree(vd
->vdev_child
[i
], indent
+ 2);
172 * Given a vdev type, return the appropriate ops vector.
175 vdev_getops(const char *type
)
177 vdev_ops_t
*ops
, **opspp
;
179 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
180 if (strcmp(ops
->vdev_op_type
, type
) == 0)
187 * Default asize function: return the MAX of psize with the asize of
188 * all children. This is what's used by anything other than RAID-Z.
191 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
193 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
196 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
197 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
198 asize
= MAX(asize
, csize
);
205 * Get the minimum allocatable size. We define the allocatable size as
206 * the vdev's asize rounded to the nearest metaslab. This allows us to
207 * replace or attach devices which don't have the same physical size but
208 * can still satisfy the same number of allocations.
211 vdev_get_min_asize(vdev_t
*vd
)
213 vdev_t
*pvd
= vd
->vdev_parent
;
216 * If our parent is NULL (inactive spare or cache) or is the root,
217 * just return our own asize.
220 return (vd
->vdev_asize
);
223 * The top-level vdev just returns the allocatable size rounded
224 * to the nearest metaslab.
226 if (vd
== vd
->vdev_top
)
227 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
230 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
231 * so each child must provide at least 1/Nth of its asize.
233 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
234 return ((pvd
->vdev_min_asize
+ pvd
->vdev_children
- 1) /
237 return (pvd
->vdev_min_asize
);
241 vdev_set_min_asize(vdev_t
*vd
)
243 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
245 for (int c
= 0; c
< vd
->vdev_children
; c
++)
246 vdev_set_min_asize(vd
->vdev_child
[c
]);
250 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
252 vdev_t
*rvd
= spa
->spa_root_vdev
;
254 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
256 if (vdev
< rvd
->vdev_children
) {
257 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
258 return (rvd
->vdev_child
[vdev
]);
265 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
269 if (vd
->vdev_guid
== guid
)
272 for (int c
= 0; c
< vd
->vdev_children
; c
++)
273 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
281 vdev_count_leaves_impl(vdev_t
*vd
)
285 if (vd
->vdev_ops
->vdev_op_leaf
)
288 for (int c
= 0; c
< vd
->vdev_children
; c
++)
289 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
295 vdev_count_leaves(spa_t
*spa
)
297 return (vdev_count_leaves_impl(spa
->spa_root_vdev
));
301 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
303 size_t oldsize
, newsize
;
304 uint64_t id
= cvd
->vdev_id
;
306 spa_t
*spa
= cvd
->vdev_spa
;
308 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
309 ASSERT(cvd
->vdev_parent
== NULL
);
311 cvd
->vdev_parent
= pvd
;
316 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
318 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
319 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
320 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
322 newchild
= kmem_zalloc(newsize
, KM_SLEEP
);
323 if (pvd
->vdev_child
!= NULL
) {
324 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
325 kmem_free(pvd
->vdev_child
, oldsize
);
328 pvd
->vdev_child
= newchild
;
329 pvd
->vdev_child
[id
] = cvd
;
331 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
332 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
335 * Walk up all ancestors to update guid sum.
337 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
338 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
342 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
345 uint_t id
= cvd
->vdev_id
;
347 ASSERT(cvd
->vdev_parent
== pvd
);
352 ASSERT(id
< pvd
->vdev_children
);
353 ASSERT(pvd
->vdev_child
[id
] == cvd
);
355 pvd
->vdev_child
[id
] = NULL
;
356 cvd
->vdev_parent
= NULL
;
358 for (c
= 0; c
< pvd
->vdev_children
; c
++)
359 if (pvd
->vdev_child
[c
])
362 if (c
== pvd
->vdev_children
) {
363 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
364 pvd
->vdev_child
= NULL
;
365 pvd
->vdev_children
= 0;
369 * Walk up all ancestors to update guid sum.
371 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
372 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
376 * Remove any holes in the child array.
379 vdev_compact_children(vdev_t
*pvd
)
381 vdev_t
**newchild
, *cvd
;
382 int oldc
= pvd
->vdev_children
;
385 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
387 for (int c
= newc
= 0; c
< oldc
; c
++)
388 if (pvd
->vdev_child
[c
])
391 newchild
= kmem_alloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
393 for (int c
= newc
= 0; c
< oldc
; c
++) {
394 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
395 newchild
[newc
] = cvd
;
396 cvd
->vdev_id
= newc
++;
400 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
401 pvd
->vdev_child
= newchild
;
402 pvd
->vdev_children
= newc
;
406 * Allocate and minimally initialize a vdev_t.
409 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
412 vdev_indirect_config_t
*vic
;
414 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
415 vic
= &vd
->vdev_indirect_config
;
417 if (spa
->spa_root_vdev
== NULL
) {
418 ASSERT(ops
== &vdev_root_ops
);
419 spa
->spa_root_vdev
= vd
;
420 spa
->spa_load_guid
= spa_generate_guid(NULL
);
423 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
424 if (spa
->spa_root_vdev
== vd
) {
426 * The root vdev's guid will also be the pool guid,
427 * which must be unique among all pools.
429 guid
= spa_generate_guid(NULL
);
432 * Any other vdev's guid must be unique within the pool.
434 guid
= spa_generate_guid(spa
);
436 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
441 vd
->vdev_guid
= guid
;
442 vd
->vdev_guid_sum
= guid
;
444 vd
->vdev_state
= VDEV_STATE_CLOSED
;
445 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
446 vic
->vic_prev_indirect_vdev
= UINT64_MAX
;
448 rw_init(&vd
->vdev_indirect_rwlock
, NULL
, RW_DEFAULT
, NULL
);
449 mutex_init(&vd
->vdev_obsolete_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
450 vd
->vdev_obsolete_segments
= range_tree_create(NULL
, NULL
);
452 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
453 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
454 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
455 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
456 for (int t
= 0; t
< DTL_TYPES
; t
++) {
457 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
);
459 txg_list_create(&vd
->vdev_ms_list
, spa
,
460 offsetof(struct metaslab
, ms_txg_node
));
461 txg_list_create(&vd
->vdev_dtl_list
, spa
,
462 offsetof(struct vdev
, vdev_dtl_node
));
463 vd
->vdev_stat
.vs_timestamp
= gethrtime();
471 * Allocate a new vdev. The 'alloctype' is used to control whether we are
472 * creating a new vdev or loading an existing one - the behavior is slightly
473 * different for each case.
476 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
481 uint64_t guid
= 0, islog
, nparity
;
483 vdev_indirect_config_t
*vic
;
485 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
487 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
488 return (SET_ERROR(EINVAL
));
490 if ((ops
= vdev_getops(type
)) == NULL
)
491 return (SET_ERROR(EINVAL
));
494 * If this is a load, get the vdev guid from the nvlist.
495 * Otherwise, vdev_alloc_common() will generate one for us.
497 if (alloctype
== VDEV_ALLOC_LOAD
) {
500 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
502 return (SET_ERROR(EINVAL
));
504 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
505 return (SET_ERROR(EINVAL
));
506 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
507 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
508 return (SET_ERROR(EINVAL
));
509 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
510 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
511 return (SET_ERROR(EINVAL
));
512 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
513 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
514 return (SET_ERROR(EINVAL
));
518 * The first allocated vdev must be of type 'root'.
520 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
521 return (SET_ERROR(EINVAL
));
524 * Determine whether we're a log vdev.
527 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
528 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
529 return (SET_ERROR(ENOTSUP
));
531 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
532 return (SET_ERROR(ENOTSUP
));
535 * Set the nparity property for RAID-Z vdevs.
538 if (ops
== &vdev_raidz_ops
) {
539 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
541 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
542 return (SET_ERROR(EINVAL
));
544 * Previous versions could only support 1 or 2 parity
548 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
549 return (SET_ERROR(ENOTSUP
));
551 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
552 return (SET_ERROR(ENOTSUP
));
555 * We require the parity to be specified for SPAs that
556 * support multiple parity levels.
558 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
559 return (SET_ERROR(EINVAL
));
561 * Otherwise, we default to 1 parity device for RAID-Z.
568 ASSERT(nparity
!= -1ULL);
570 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
571 vic
= &vd
->vdev_indirect_config
;
573 vd
->vdev_islog
= islog
;
574 vd
->vdev_nparity
= nparity
;
576 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
577 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
578 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
579 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
580 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
581 &vd
->vdev_physpath
) == 0)
582 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
583 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
584 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
587 * Set the whole_disk property. If it's not specified, leave the value
590 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
591 &vd
->vdev_wholedisk
) != 0)
592 vd
->vdev_wholedisk
= -1ULL;
594 ASSERT0(vic
->vic_mapping_object
);
595 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
596 &vic
->vic_mapping_object
);
597 ASSERT0(vic
->vic_births_object
);
598 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
599 &vic
->vic_births_object
);
600 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
601 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
602 &vic
->vic_prev_indirect_vdev
);
605 * Look for the 'not present' flag. This will only be set if the device
606 * was not present at the time of import.
608 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
609 &vd
->vdev_not_present
);
612 * Get the alignment requirement.
614 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
617 * Retrieve the vdev creation time.
619 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
623 * If we're a top-level vdev, try to load the allocation parameters.
625 if (parent
&& !parent
->vdev_parent
&&
626 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
627 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
629 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
631 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
633 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
635 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
638 ASSERT0(vd
->vdev_top_zap
);
641 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
642 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
643 alloctype
== VDEV_ALLOC_ADD
||
644 alloctype
== VDEV_ALLOC_SPLIT
||
645 alloctype
== VDEV_ALLOC_ROOTPOOL
);
646 vd
->vdev_mg
= metaslab_group_create(islog
?
647 spa_log_class(spa
) : spa_normal_class(spa
), vd
);
650 if (vd
->vdev_ops
->vdev_op_leaf
&&
651 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
652 (void) nvlist_lookup_uint64(nv
,
653 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
655 ASSERT0(vd
->vdev_leaf_zap
);
659 * If we're a leaf vdev, try to load the DTL object and other state.
662 if (vd
->vdev_ops
->vdev_op_leaf
&&
663 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
664 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
665 if (alloctype
== VDEV_ALLOC_LOAD
) {
666 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
667 &vd
->vdev_dtl_object
);
668 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
672 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
675 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
676 &spare
) == 0 && spare
)
680 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
683 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
684 &vd
->vdev_resilver_txg
);
687 * When importing a pool, we want to ignore the persistent fault
688 * state, as the diagnosis made on another system may not be
689 * valid in the current context. Local vdevs will
690 * remain in the faulted state.
692 if (spa_load_state(spa
) == SPA_LOAD_OPEN
) {
693 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
695 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
697 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
700 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
704 VDEV_AUX_ERR_EXCEEDED
;
705 if (nvlist_lookup_string(nv
,
706 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
707 strcmp(aux
, "external") == 0)
708 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
714 * Add ourselves to the parent's list of children.
716 vdev_add_child(parent
, vd
);
724 vdev_free(vdev_t
*vd
)
726 spa_t
*spa
= vd
->vdev_spa
;
729 * vdev_free() implies closing the vdev first. This is simpler than
730 * trying to ensure complicated semantics for all callers.
734 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
735 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
740 for (int c
= 0; c
< vd
->vdev_children
; c
++)
741 vdev_free(vd
->vdev_child
[c
]);
743 ASSERT(vd
->vdev_child
== NULL
);
744 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
747 * Discard allocation state.
749 if (vd
->vdev_mg
!= NULL
) {
750 vdev_metaslab_fini(vd
);
751 metaslab_group_destroy(vd
->vdev_mg
);
754 ASSERT0(vd
->vdev_stat
.vs_space
);
755 ASSERT0(vd
->vdev_stat
.vs_dspace
);
756 ASSERT0(vd
->vdev_stat
.vs_alloc
);
759 * Remove this vdev from its parent's child list.
761 vdev_remove_child(vd
->vdev_parent
, vd
);
763 ASSERT(vd
->vdev_parent
== NULL
);
766 * Clean up vdev structure.
772 spa_strfree(vd
->vdev_path
);
774 spa_strfree(vd
->vdev_devid
);
775 if (vd
->vdev_physpath
)
776 spa_strfree(vd
->vdev_physpath
);
778 spa_strfree(vd
->vdev_fru
);
780 if (vd
->vdev_isspare
)
781 spa_spare_remove(vd
);
782 if (vd
->vdev_isl2cache
)
783 spa_l2cache_remove(vd
);
785 txg_list_destroy(&vd
->vdev_ms_list
);
786 txg_list_destroy(&vd
->vdev_dtl_list
);
788 mutex_enter(&vd
->vdev_dtl_lock
);
789 space_map_close(vd
->vdev_dtl_sm
);
790 for (int t
= 0; t
< DTL_TYPES
; t
++) {
791 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
792 range_tree_destroy(vd
->vdev_dtl
[t
]);
794 mutex_exit(&vd
->vdev_dtl_lock
);
796 EQUIV(vd
->vdev_indirect_births
!= NULL
,
797 vd
->vdev_indirect_mapping
!= NULL
);
798 if (vd
->vdev_indirect_births
!= NULL
) {
799 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
800 vdev_indirect_births_close(vd
->vdev_indirect_births
);
803 if (vd
->vdev_obsolete_sm
!= NULL
) {
804 ASSERT(vd
->vdev_removing
||
805 vd
->vdev_ops
== &vdev_indirect_ops
);
806 space_map_close(vd
->vdev_obsolete_sm
);
807 vd
->vdev_obsolete_sm
= NULL
;
809 range_tree_destroy(vd
->vdev_obsolete_segments
);
810 rw_destroy(&vd
->vdev_indirect_rwlock
);
811 mutex_destroy(&vd
->vdev_obsolete_lock
);
813 mutex_destroy(&vd
->vdev_queue_lock
);
814 mutex_destroy(&vd
->vdev_dtl_lock
);
815 mutex_destroy(&vd
->vdev_stat_lock
);
816 mutex_destroy(&vd
->vdev_probe_lock
);
818 if (vd
== spa
->spa_root_vdev
)
819 spa
->spa_root_vdev
= NULL
;
821 kmem_free(vd
, sizeof (vdev_t
));
825 * Transfer top-level vdev state from svd to tvd.
828 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
830 spa_t
*spa
= svd
->vdev_spa
;
835 ASSERT(tvd
== tvd
->vdev_top
);
837 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
838 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
839 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
840 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
842 svd
->vdev_ms_array
= 0;
843 svd
->vdev_ms_shift
= 0;
844 svd
->vdev_ms_count
= 0;
845 svd
->vdev_top_zap
= 0;
848 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
849 tvd
->vdev_mg
= svd
->vdev_mg
;
850 tvd
->vdev_ms
= svd
->vdev_ms
;
855 if (tvd
->vdev_mg
!= NULL
)
856 tvd
->vdev_mg
->mg_vd
= tvd
;
858 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
859 svd
->vdev_checkpoint_sm
= NULL
;
861 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
862 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
863 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
865 svd
->vdev_stat
.vs_alloc
= 0;
866 svd
->vdev_stat
.vs_space
= 0;
867 svd
->vdev_stat
.vs_dspace
= 0;
869 for (t
= 0; t
< TXG_SIZE
; t
++) {
870 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
871 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
872 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
873 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
874 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
875 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
878 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
879 vdev_config_clean(svd
);
880 vdev_config_dirty(tvd
);
883 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
884 vdev_state_clean(svd
);
885 vdev_state_dirty(tvd
);
888 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
889 svd
->vdev_deflate_ratio
= 0;
891 tvd
->vdev_islog
= svd
->vdev_islog
;
896 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
903 for (int c
= 0; c
< vd
->vdev_children
; c
++)
904 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
908 * Add a mirror/replacing vdev above an existing vdev.
911 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
913 spa_t
*spa
= cvd
->vdev_spa
;
914 vdev_t
*pvd
= cvd
->vdev_parent
;
917 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
919 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
921 mvd
->vdev_asize
= cvd
->vdev_asize
;
922 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
923 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
924 mvd
->vdev_psize
= cvd
->vdev_psize
;
925 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
926 mvd
->vdev_state
= cvd
->vdev_state
;
927 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
929 vdev_remove_child(pvd
, cvd
);
930 vdev_add_child(pvd
, mvd
);
931 cvd
->vdev_id
= mvd
->vdev_children
;
932 vdev_add_child(mvd
, cvd
);
933 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
935 if (mvd
== mvd
->vdev_top
)
936 vdev_top_transfer(cvd
, mvd
);
942 * Remove a 1-way mirror/replacing vdev from the tree.
945 vdev_remove_parent(vdev_t
*cvd
)
947 vdev_t
*mvd
= cvd
->vdev_parent
;
948 vdev_t
*pvd
= mvd
->vdev_parent
;
950 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
952 ASSERT(mvd
->vdev_children
== 1);
953 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
954 mvd
->vdev_ops
== &vdev_replacing_ops
||
955 mvd
->vdev_ops
== &vdev_spare_ops
);
956 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
958 vdev_remove_child(mvd
, cvd
);
959 vdev_remove_child(pvd
, mvd
);
962 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
963 * Otherwise, we could have detached an offline device, and when we
964 * go to import the pool we'll think we have two top-level vdevs,
965 * instead of a different version of the same top-level vdev.
967 if (mvd
->vdev_top
== mvd
) {
968 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
969 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
970 cvd
->vdev_guid
+= guid_delta
;
971 cvd
->vdev_guid_sum
+= guid_delta
;
973 cvd
->vdev_id
= mvd
->vdev_id
;
974 vdev_add_child(pvd
, cvd
);
975 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
977 if (cvd
== cvd
->vdev_top
)
978 vdev_top_transfer(mvd
, cvd
);
980 ASSERT(mvd
->vdev_children
== 0);
985 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
987 spa_t
*spa
= vd
->vdev_spa
;
988 objset_t
*mos
= spa
->spa_meta_objset
;
990 uint64_t oldc
= vd
->vdev_ms_count
;
991 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
995 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
998 * This vdev is not being allocated from yet or is a hole.
1000 if (vd
->vdev_ms_shift
== 0)
1003 ASSERT(!vd
->vdev_ishole
);
1005 ASSERT(oldc
<= newc
);
1007 mspp
= kmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1010 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
1011 kmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1015 vd
->vdev_ms_count
= newc
;
1017 for (m
= oldc
; m
< newc
; m
++) {
1018 uint64_t object
= 0;
1021 * vdev_ms_array may be 0 if we are creating the "fake"
1022 * metaslabs for an indirect vdev for zdb's leak detection.
1023 * See zdb_leak_init().
1025 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1026 error
= dmu_read(mos
, vd
->vdev_ms_array
,
1027 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1030 vdev_dbgmsg(vd
, "unable to read the metaslab "
1031 "array [error=%d]", error
);
1036 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1039 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1046 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1049 * If the vdev is being removed we don't activate
1050 * the metaslabs since we want to ensure that no new
1051 * allocations are performed on this device.
1053 if (oldc
== 0 && !vd
->vdev_removing
)
1054 metaslab_group_activate(vd
->vdev_mg
);
1057 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1063 vdev_metaslab_fini(vdev_t
*vd
)
1065 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1066 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1067 SPA_FEATURE_POOL_CHECKPOINT
));
1068 space_map_close(vd
->vdev_checkpoint_sm
);
1070 * Even though we close the space map, we need to set its
1071 * pointer to NULL. The reason is that vdev_metaslab_fini()
1072 * may be called multiple times for certain operations
1073 * (i.e. when destroying a pool) so we need to ensure that
1074 * this clause never executes twice. This logic is similar
1075 * to the one used for the vdev_ms clause below.
1077 vd
->vdev_checkpoint_sm
= NULL
;
1080 if (vd
->vdev_ms
!= NULL
) {
1081 uint64_t count
= vd
->vdev_ms_count
;
1083 metaslab_group_passivate(vd
->vdev_mg
);
1084 for (uint64_t m
= 0; m
< count
; m
++) {
1085 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1090 kmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1093 vd
->vdev_ms_count
= 0;
1095 ASSERT0(vd
->vdev_ms_count
);
1098 typedef struct vdev_probe_stats
{
1099 boolean_t vps_readable
;
1100 boolean_t vps_writeable
;
1102 } vdev_probe_stats_t
;
1105 vdev_probe_done(zio_t
*zio
)
1107 spa_t
*spa
= zio
->io_spa
;
1108 vdev_t
*vd
= zio
->io_vd
;
1109 vdev_probe_stats_t
*vps
= zio
->io_private
;
1111 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1113 if (zio
->io_type
== ZIO_TYPE_READ
) {
1114 if (zio
->io_error
== 0)
1115 vps
->vps_readable
= 1;
1116 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1117 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1118 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1119 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1120 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1122 abd_free(zio
->io_abd
);
1124 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1125 if (zio
->io_error
== 0)
1126 vps
->vps_writeable
= 1;
1127 abd_free(zio
->io_abd
);
1128 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1131 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1132 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1134 if (vdev_readable(vd
) &&
1135 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1138 ASSERT(zio
->io_error
!= 0);
1139 vdev_dbgmsg(vd
, "failed probe");
1140 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1141 spa
, vd
, NULL
, 0, 0);
1142 zio
->io_error
= SET_ERROR(ENXIO
);
1145 mutex_enter(&vd
->vdev_probe_lock
);
1146 ASSERT(vd
->vdev_probe_zio
== zio
);
1147 vd
->vdev_probe_zio
= NULL
;
1148 mutex_exit(&vd
->vdev_probe_lock
);
1150 zio_link_t
*zl
= NULL
;
1151 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1152 if (!vdev_accessible(vd
, pio
))
1153 pio
->io_error
= SET_ERROR(ENXIO
);
1155 kmem_free(vps
, sizeof (*vps
));
1160 * Determine whether this device is accessible.
1162 * Read and write to several known locations: the pad regions of each
1163 * vdev label but the first, which we leave alone in case it contains
1167 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1169 spa_t
*spa
= vd
->vdev_spa
;
1170 vdev_probe_stats_t
*vps
= NULL
;
1173 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1176 * Don't probe the probe.
1178 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1182 * To prevent 'probe storms' when a device fails, we create
1183 * just one probe i/o at a time. All zios that want to probe
1184 * this vdev will become parents of the probe io.
1186 mutex_enter(&vd
->vdev_probe_lock
);
1188 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1189 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1191 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1192 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1195 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1197 * vdev_cant_read and vdev_cant_write can only
1198 * transition from TRUE to FALSE when we have the
1199 * SCL_ZIO lock as writer; otherwise they can only
1200 * transition from FALSE to TRUE. This ensures that
1201 * any zio looking at these values can assume that
1202 * failures persist for the life of the I/O. That's
1203 * important because when a device has intermittent
1204 * connectivity problems, we want to ensure that
1205 * they're ascribed to the device (ENXIO) and not
1208 * Since we hold SCL_ZIO as writer here, clear both
1209 * values so the probe can reevaluate from first
1212 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1213 vd
->vdev_cant_read
= B_FALSE
;
1214 vd
->vdev_cant_write
= B_FALSE
;
1217 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1218 vdev_probe_done
, vps
,
1219 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1222 * We can't change the vdev state in this context, so we
1223 * kick off an async task to do it on our behalf.
1226 vd
->vdev_probe_wanted
= B_TRUE
;
1227 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1232 zio_add_child(zio
, pio
);
1234 mutex_exit(&vd
->vdev_probe_lock
);
1237 ASSERT(zio
!= NULL
);
1241 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1242 zio_nowait(zio_read_phys(pio
, vd
,
1243 vdev_label_offset(vd
->vdev_psize
, l
,
1244 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1245 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1246 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1247 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1258 vdev_open_child(void *arg
)
1262 vd
->vdev_open_thread
= curthread
;
1263 vd
->vdev_open_error
= vdev_open(vd
);
1264 vd
->vdev_open_thread
= NULL
;
1268 vdev_uses_zvols(vdev_t
*vd
)
1270 if (vd
->vdev_path
&& strncmp(vd
->vdev_path
, ZVOL_DIR
,
1271 strlen(ZVOL_DIR
)) == 0)
1273 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1274 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1280 vdev_open_children(vdev_t
*vd
)
1283 int children
= vd
->vdev_children
;
1286 * in order to handle pools on top of zvols, do the opens
1287 * in a single thread so that the same thread holds the
1288 * spa_namespace_lock
1290 if (vdev_uses_zvols(vd
)) {
1291 for (int c
= 0; c
< children
; c
++)
1292 vd
->vdev_child
[c
]->vdev_open_error
=
1293 vdev_open(vd
->vdev_child
[c
]);
1296 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1297 children
, children
, TASKQ_PREPOPULATE
);
1299 for (int c
= 0; c
< children
; c
++)
1300 VERIFY(taskq_dispatch(tq
, vdev_open_child
, vd
->vdev_child
[c
],
1307 * Compute the raidz-deflation ratio. Note, we hard-code
1308 * in 128k (1 << 17) because it is the "typical" blocksize.
1309 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1310 * otherwise it would inconsistently account for existing bp's.
1313 vdev_set_deflate_ratio(vdev_t
*vd
)
1315 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1316 vd
->vdev_deflate_ratio
= (1 << 17) /
1317 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1322 * Prepare a virtual device for access.
1325 vdev_open(vdev_t
*vd
)
1327 spa_t
*spa
= vd
->vdev_spa
;
1330 uint64_t max_osize
= 0;
1331 uint64_t asize
, max_asize
, psize
;
1332 uint64_t ashift
= 0;
1334 ASSERT(vd
->vdev_open_thread
== curthread
||
1335 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1336 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1337 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1338 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1340 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1341 vd
->vdev_cant_read
= B_FALSE
;
1342 vd
->vdev_cant_write
= B_FALSE
;
1343 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1346 * If this vdev is not removed, check its fault status. If it's
1347 * faulted, bail out of the open.
1349 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1350 ASSERT(vd
->vdev_children
== 0);
1351 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1352 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1353 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1354 vd
->vdev_label_aux
);
1355 return (SET_ERROR(ENXIO
));
1356 } else if (vd
->vdev_offline
) {
1357 ASSERT(vd
->vdev_children
== 0);
1358 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1359 return (SET_ERROR(ENXIO
));
1362 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1365 * Reset the vdev_reopening flag so that we actually close
1366 * the vdev on error.
1368 vd
->vdev_reopening
= B_FALSE
;
1369 if (zio_injection_enabled
&& error
== 0)
1370 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1373 if (vd
->vdev_removed
&&
1374 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1375 vd
->vdev_removed
= B_FALSE
;
1377 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
1378 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
1379 vd
->vdev_stat
.vs_aux
);
1381 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1382 vd
->vdev_stat
.vs_aux
);
1387 vd
->vdev_removed
= B_FALSE
;
1390 * Recheck the faulted flag now that we have confirmed that
1391 * the vdev is accessible. If we're faulted, bail.
1393 if (vd
->vdev_faulted
) {
1394 ASSERT(vd
->vdev_children
== 0);
1395 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1396 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1397 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1398 vd
->vdev_label_aux
);
1399 return (SET_ERROR(ENXIO
));
1402 if (vd
->vdev_degraded
) {
1403 ASSERT(vd
->vdev_children
== 0);
1404 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1405 VDEV_AUX_ERR_EXCEEDED
);
1407 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1411 * For hole or missing vdevs we just return success.
1413 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1416 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1417 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1418 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1424 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1425 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1427 if (vd
->vdev_children
== 0) {
1428 if (osize
< SPA_MINDEVSIZE
) {
1429 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1430 VDEV_AUX_TOO_SMALL
);
1431 return (SET_ERROR(EOVERFLOW
));
1434 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1435 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1436 VDEV_LABEL_END_SIZE
);
1438 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1439 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1440 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1441 VDEV_AUX_TOO_SMALL
);
1442 return (SET_ERROR(EOVERFLOW
));
1446 max_asize
= max_osize
;
1449 vd
->vdev_psize
= psize
;
1452 * Make sure the allocatable size hasn't shrunk too much.
1454 if (asize
< vd
->vdev_min_asize
) {
1455 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1456 VDEV_AUX_BAD_LABEL
);
1457 return (SET_ERROR(EINVAL
));
1460 if (vd
->vdev_asize
== 0) {
1462 * This is the first-ever open, so use the computed values.
1463 * For testing purposes, a higher ashift can be requested.
1465 vd
->vdev_asize
= asize
;
1466 vd
->vdev_max_asize
= max_asize
;
1467 vd
->vdev_ashift
= MAX(ashift
, vd
->vdev_ashift
);
1470 * Detect if the alignment requirement has increased.
1471 * We don't want to make the pool unavailable, just
1472 * issue a warning instead.
1474 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1475 vd
->vdev_ops
->vdev_op_leaf
) {
1477 "Disk, '%s', has a block alignment that is "
1478 "larger than the pool's alignment\n",
1481 vd
->vdev_max_asize
= max_asize
;
1485 * If all children are healthy we update asize if either:
1486 * The asize has increased, due to a device expansion caused by dynamic
1487 * LUN growth or vdev replacement, and automatic expansion is enabled;
1488 * making the additional space available.
1490 * The asize has decreased, due to a device shrink usually caused by a
1491 * vdev replace with a smaller device. This ensures that calculations
1492 * based of max_asize and asize e.g. esize are always valid. It's safe
1493 * to do this as we've already validated that asize is greater than
1496 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1497 ((asize
> vd
->vdev_asize
&&
1498 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1499 (asize
< vd
->vdev_asize
)))
1500 vd
->vdev_asize
= asize
;
1502 vdev_set_min_asize(vd
);
1505 * Ensure we can issue some IO before declaring the
1506 * vdev open for business.
1508 if (vd
->vdev_ops
->vdev_op_leaf
&&
1509 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1510 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1511 VDEV_AUX_ERR_EXCEEDED
);
1516 * Track the min and max ashift values for normal data devices.
1518 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1519 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1520 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1521 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1522 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1523 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1527 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1528 * resilver. But don't do this if we are doing a reopen for a scrub,
1529 * since this would just restart the scrub we are already doing.
1531 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1532 vdev_resilver_needed(vd
, NULL
, NULL
))
1533 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1539 * Called once the vdevs are all opened, this routine validates the label
1540 * contents. This needs to be done before vdev_load() so that we don't
1541 * inadvertently do repair I/Os to the wrong device.
1543 * This function will only return failure if one of the vdevs indicates that it
1544 * has since been destroyed or exported. This is only possible if
1545 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1546 * will be updated but the function will return 0.
1549 vdev_validate(vdev_t
*vd
)
1551 spa_t
*spa
= vd
->vdev_spa
;
1553 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
1558 if (vdev_validate_skip
)
1561 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1562 if (vdev_validate(vd
->vdev_child
[c
]) != 0)
1563 return (SET_ERROR(EBADF
));
1566 * If the device has already failed, or was marked offline, don't do
1567 * any further validation. Otherwise, label I/O will fail and we will
1568 * overwrite the previous state.
1570 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
1574 * If we are performing an extreme rewind, we allow for a label that
1575 * was modified at a point after the current txg.
1576 * If config lock is not held do not check for the txg. spa_sync could
1577 * be updating the vdev's label before updating spa_last_synced_txg.
1579 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
1580 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
1583 txg
= spa_last_synced_txg(spa
);
1585 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1586 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1587 VDEV_AUX_BAD_LABEL
);
1588 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
1589 "txg %llu", (u_longlong_t
)txg
);
1594 * Determine if this vdev has been split off into another
1595 * pool. If so, then refuse to open it.
1597 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1598 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1599 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1600 VDEV_AUX_SPLIT_POOL
);
1602 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
1606 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
1607 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1608 VDEV_AUX_CORRUPT_DATA
);
1610 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1611 ZPOOL_CONFIG_POOL_GUID
);
1616 * If config is not trusted then ignore the spa guid check. This is
1617 * necessary because if the machine crashed during a re-guid the new
1618 * guid might have been written to all of the vdev labels, but not the
1619 * cached config. The check will be performed again once we have the
1620 * trusted config from the MOS.
1622 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
1623 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1624 VDEV_AUX_CORRUPT_DATA
);
1626 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
1627 "match config (%llu != %llu)", (u_longlong_t
)guid
,
1628 (u_longlong_t
)spa_guid(spa
));
1632 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1633 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1637 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
1638 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1639 VDEV_AUX_CORRUPT_DATA
);
1641 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1646 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
1648 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1649 VDEV_AUX_CORRUPT_DATA
);
1651 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1652 ZPOOL_CONFIG_TOP_GUID
);
1657 * If this vdev just became a top-level vdev because its sibling was
1658 * detached, it will have adopted the parent's vdev guid -- but the
1659 * label may or may not be on disk yet. Fortunately, either version
1660 * of the label will have the same top guid, so if we're a top-level
1661 * vdev, we can safely compare to that instead.
1662 * However, if the config comes from a cachefile that failed to update
1663 * after the detach, a top-level vdev will appear as a non top-level
1664 * vdev in the config. Also relax the constraints if we perform an
1667 * If we split this vdev off instead, then we also check the
1668 * original pool's guid. We don't want to consider the vdev
1669 * corrupt if it is partway through a split operation.
1671 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
1672 boolean_t mismatch
= B_FALSE
;
1673 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
1674 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
1677 if (vd
->vdev_guid
!= top_guid
&&
1678 vd
->vdev_top
->vdev_guid
!= guid
)
1683 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1684 VDEV_AUX_CORRUPT_DATA
);
1686 vdev_dbgmsg(vd
, "vdev_validate: config guid "
1687 "doesn't match label guid");
1688 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
1689 (u_longlong_t
)vd
->vdev_guid
,
1690 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
1691 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
1692 "aux_guid %llu", (u_longlong_t
)guid
,
1693 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
1698 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1700 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1701 VDEV_AUX_CORRUPT_DATA
);
1703 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1704 ZPOOL_CONFIG_POOL_STATE
);
1711 * If this is a verbatim import, no need to check the
1712 * state of the pool.
1714 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1715 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1716 state
!= POOL_STATE_ACTIVE
) {
1717 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
1718 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
1719 return (SET_ERROR(EBADF
));
1723 * If we were able to open and validate a vdev that was
1724 * previously marked permanently unavailable, clear that state
1727 if (vd
->vdev_not_present
)
1728 vd
->vdev_not_present
= 0;
1734 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
1736 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
1737 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
1738 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
1739 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
1740 dvd
->vdev_path
, svd
->vdev_path
);
1741 spa_strfree(dvd
->vdev_path
);
1742 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
1744 } else if (svd
->vdev_path
!= NULL
) {
1745 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
1746 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
1747 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
1752 * Recursively copy vdev paths from one vdev to another. Source and destination
1753 * vdev trees must have same geometry otherwise return error. Intended to copy
1754 * paths from userland config into MOS config.
1757 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
1759 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
1760 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
1761 (dvd
->vdev_ops
== &vdev_indirect_ops
))
1764 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
1765 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
1766 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
1767 return (SET_ERROR(EINVAL
));
1770 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
1771 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
1772 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
1773 (u_longlong_t
)dvd
->vdev_guid
);
1774 return (SET_ERROR(EINVAL
));
1777 if (svd
->vdev_children
!= dvd
->vdev_children
) {
1778 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
1779 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
1780 (u_longlong_t
)dvd
->vdev_children
);
1781 return (SET_ERROR(EINVAL
));
1784 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
1785 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
1786 dvd
->vdev_child
[i
]);
1791 if (svd
->vdev_ops
->vdev_op_leaf
)
1792 vdev_copy_path_impl(svd
, dvd
);
1798 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
1800 ASSERT(stvd
->vdev_top
== stvd
);
1801 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
1803 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
1804 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
1807 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
1811 * The idea here is that while a vdev can shift positions within
1812 * a top vdev (when replacing, attaching mirror, etc.) it cannot
1813 * step outside of it.
1815 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
1817 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
1820 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1822 vdev_copy_path_impl(vd
, dvd
);
1826 * Recursively copy vdev paths from one root vdev to another. Source and
1827 * destination vdev trees may differ in geometry. For each destination leaf
1828 * vdev, search a vdev with the same guid and top vdev id in the source.
1829 * Intended to copy paths from userland config into MOS config.
1832 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
1834 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
1835 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
1836 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
1838 for (uint64_t i
= 0; i
< children
; i
++) {
1839 vdev_copy_path_search(srvd
->vdev_child
[i
],
1840 drvd
->vdev_child
[i
]);
1845 * Close a virtual device.
1848 vdev_close(vdev_t
*vd
)
1850 spa_t
*spa
= vd
->vdev_spa
;
1851 vdev_t
*pvd
= vd
->vdev_parent
;
1853 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1856 * If our parent is reopening, then we are as well, unless we are
1859 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
1860 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
1862 vd
->vdev_ops
->vdev_op_close(vd
);
1864 vdev_cache_purge(vd
);
1867 * We record the previous state before we close it, so that if we are
1868 * doing a reopen(), we don't generate FMA ereports if we notice that
1869 * it's still faulted.
1871 vd
->vdev_prevstate
= vd
->vdev_state
;
1873 if (vd
->vdev_offline
)
1874 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
1876 vd
->vdev_state
= VDEV_STATE_CLOSED
;
1877 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1881 vdev_hold(vdev_t
*vd
)
1883 spa_t
*spa
= vd
->vdev_spa
;
1885 ASSERT(spa_is_root(spa
));
1886 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1889 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1890 vdev_hold(vd
->vdev_child
[c
]);
1892 if (vd
->vdev_ops
->vdev_op_leaf
)
1893 vd
->vdev_ops
->vdev_op_hold(vd
);
1897 vdev_rele(vdev_t
*vd
)
1899 spa_t
*spa
= vd
->vdev_spa
;
1901 ASSERT(spa_is_root(spa
));
1902 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1903 vdev_rele(vd
->vdev_child
[c
]);
1905 if (vd
->vdev_ops
->vdev_op_leaf
)
1906 vd
->vdev_ops
->vdev_op_rele(vd
);
1910 * Reopen all interior vdevs and any unopened leaves. We don't actually
1911 * reopen leaf vdevs which had previously been opened as they might deadlock
1912 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1913 * If the leaf has never been opened then open it, as usual.
1916 vdev_reopen(vdev_t
*vd
)
1918 spa_t
*spa
= vd
->vdev_spa
;
1920 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1922 /* set the reopening flag unless we're taking the vdev offline */
1923 vd
->vdev_reopening
= !vd
->vdev_offline
;
1925 (void) vdev_open(vd
);
1928 * Call vdev_validate() here to make sure we have the same device.
1929 * Otherwise, a device with an invalid label could be successfully
1930 * opened in response to vdev_reopen().
1933 (void) vdev_validate_aux(vd
);
1934 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
1935 vd
->vdev_aux
== &spa
->spa_l2cache
&&
1936 !l2arc_vdev_present(vd
))
1937 l2arc_add_vdev(spa
, vd
);
1939 (void) vdev_validate(vd
);
1943 * Reassess parent vdev's health.
1945 vdev_propagate_state(vd
);
1949 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
1954 * Normally, partial opens (e.g. of a mirror) are allowed.
1955 * For a create, however, we want to fail the request if
1956 * there are any components we can't open.
1958 error
= vdev_open(vd
);
1960 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
1962 return (error
? error
: ENXIO
);
1966 * Recursively load DTLs and initialize all labels.
1968 if ((error
= vdev_dtl_load(vd
)) != 0 ||
1969 (error
= vdev_label_init(vd
, txg
, isreplacing
?
1970 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
1979 vdev_metaslab_set_size(vdev_t
*vd
)
1981 uint64_t asize
= vd
->vdev_asize
;
1982 uint64_t ms_shift
= 0;
1985 * For vdevs that are bigger than 8G the metaslab size varies in
1986 * a way that the number of metaslabs increases in powers of two,
1987 * linearly in terms of vdev_asize, starting from 16 metaslabs.
1988 * So for vdev_asize of 8G we get 16 metaslabs, for 16G, we get 32,
1989 * and so on, until we hit the maximum metaslab count limit
1990 * [vdev_max_ms_count] from which point the metaslab count stays
1993 ms_shift
= vdev_default_ms_shift
;
1995 if ((asize
>> ms_shift
) < vdev_min_ms_count
) {
1997 * For devices that are less than 8G we want to have
1998 * exactly 16 metaslabs. We don't want less as integer
1999 * division rounds down, so less metaslabs mean more
2000 * wasted space. We don't want more as these vdevs are
2001 * small and in the likely event that we are running
2002 * out of space, the SPA will have a hard time finding
2003 * space due to fragmentation.
2005 ms_shift
= highbit64(asize
/ vdev_min_ms_count
);
2006 ms_shift
= MAX(ms_shift
, SPA_MAXBLOCKSHIFT
);
2008 } else if ((asize
>> ms_shift
) > vdev_max_ms_count
) {
2009 ms_shift
= highbit64(asize
/ vdev_max_ms_count
);
2012 vd
->vdev_ms_shift
= ms_shift
;
2013 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2017 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2019 ASSERT(vd
== vd
->vdev_top
);
2020 /* indirect vdevs don't have metaslabs or dtls */
2021 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2022 ASSERT(ISP2(flags
));
2023 ASSERT(spa_writeable(vd
->vdev_spa
));
2025 if (flags
& VDD_METASLAB
)
2026 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2028 if (flags
& VDD_DTL
)
2029 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2031 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2035 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2037 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2038 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2040 if (vd
->vdev_ops
->vdev_op_leaf
)
2041 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2047 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2048 * the vdev has less than perfect replication. There are four kinds of DTL:
2050 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2052 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2054 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2055 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2056 * txgs that was scrubbed.
2058 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2059 * persistent errors or just some device being offline.
2060 * Unlike the other three, the DTL_OUTAGE map is not generally
2061 * maintained; it's only computed when needed, typically to
2062 * determine whether a device can be detached.
2064 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2065 * either has the data or it doesn't.
2067 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2068 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2069 * if any child is less than fully replicated, then so is its parent.
2070 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2071 * comprising only those txgs which appear in 'maxfaults' or more children;
2072 * those are the txgs we don't have enough replication to read. For example,
2073 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2074 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2075 * two child DTL_MISSING maps.
2077 * It should be clear from the above that to compute the DTLs and outage maps
2078 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2079 * Therefore, that is all we keep on disk. When loading the pool, or after
2080 * a configuration change, we generate all other DTLs from first principles.
2083 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2085 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2087 ASSERT(t
< DTL_TYPES
);
2088 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2089 ASSERT(spa_writeable(vd
->vdev_spa
));
2091 mutex_enter(&vd
->vdev_dtl_lock
);
2092 if (!range_tree_contains(rt
, txg
, size
))
2093 range_tree_add(rt
, txg
, size
);
2094 mutex_exit(&vd
->vdev_dtl_lock
);
2098 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2100 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2101 boolean_t dirty
= B_FALSE
;
2103 ASSERT(t
< DTL_TYPES
);
2104 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2107 * While we are loading the pool, the DTLs have not been loaded yet.
2108 * Ignore the DTLs and try all devices. This avoids a recursive
2109 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2110 * when loading the pool (relying on the checksum to ensure that
2111 * we get the right data -- note that we while loading, we are
2112 * only reading the MOS, which is always checksummed).
2114 if (vd
->vdev_spa
->spa_load_state
!= SPA_LOAD_NONE
)
2117 mutex_enter(&vd
->vdev_dtl_lock
);
2118 if (!range_tree_is_empty(rt
))
2119 dirty
= range_tree_contains(rt
, txg
, size
);
2120 mutex_exit(&vd
->vdev_dtl_lock
);
2126 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2128 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2131 mutex_enter(&vd
->vdev_dtl_lock
);
2132 empty
= range_tree_is_empty(rt
);
2133 mutex_exit(&vd
->vdev_dtl_lock
);
2139 * Returns the lowest txg in the DTL range.
2142 vdev_dtl_min(vdev_t
*vd
)
2146 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2147 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2148 ASSERT0(vd
->vdev_children
);
2150 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2151 return (rs
->rs_start
- 1);
2155 * Returns the highest txg in the DTL.
2158 vdev_dtl_max(vdev_t
*vd
)
2162 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2163 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2164 ASSERT0(vd
->vdev_children
);
2166 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2167 return (rs
->rs_end
);
2171 * Determine if a resilvering vdev should remove any DTL entries from
2172 * its range. If the vdev was resilvering for the entire duration of the
2173 * scan then it should excise that range from its DTLs. Otherwise, this
2174 * vdev is considered partially resilvered and should leave its DTL
2175 * entries intact. The comment in vdev_dtl_reassess() describes how we
2179 vdev_dtl_should_excise(vdev_t
*vd
)
2181 spa_t
*spa
= vd
->vdev_spa
;
2182 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2184 ASSERT0(scn
->scn_phys
.scn_errors
);
2185 ASSERT0(vd
->vdev_children
);
2187 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
2190 if (vd
->vdev_resilver_txg
== 0 ||
2191 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
2195 * When a resilver is initiated the scan will assign the scn_max_txg
2196 * value to the highest txg value that exists in all DTLs. If this
2197 * device's max DTL is not part of this scan (i.e. it is not in
2198 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2201 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
2202 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
2203 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
2204 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
2211 * Reassess DTLs after a config change or scrub completion.
2214 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
2216 spa_t
*spa
= vd
->vdev_spa
;
2220 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2222 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2223 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
2224 scrub_txg
, scrub_done
);
2226 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
2229 if (vd
->vdev_ops
->vdev_op_leaf
) {
2230 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2232 mutex_enter(&vd
->vdev_dtl_lock
);
2235 * If we've completed a scan cleanly then determine
2236 * if this vdev should remove any DTLs. We only want to
2237 * excise regions on vdevs that were available during
2238 * the entire duration of this scan.
2240 if (scrub_txg
!= 0 &&
2241 (spa
->spa_scrub_started
||
2242 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
2243 vdev_dtl_should_excise(vd
)) {
2245 * We completed a scrub up to scrub_txg. If we
2246 * did it without rebooting, then the scrub dtl
2247 * will be valid, so excise the old region and
2248 * fold in the scrub dtl. Otherwise, leave the
2249 * dtl as-is if there was an error.
2251 * There's little trick here: to excise the beginning
2252 * of the DTL_MISSING map, we put it into a reference
2253 * tree and then add a segment with refcnt -1 that
2254 * covers the range [0, scrub_txg). This means
2255 * that each txg in that range has refcnt -1 or 0.
2256 * We then add DTL_SCRUB with a refcnt of 2, so that
2257 * entries in the range [0, scrub_txg) will have a
2258 * positive refcnt -- either 1 or 2. We then convert
2259 * the reference tree into the new DTL_MISSING map.
2261 space_reftree_create(&reftree
);
2262 space_reftree_add_map(&reftree
,
2263 vd
->vdev_dtl
[DTL_MISSING
], 1);
2264 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
2265 space_reftree_add_map(&reftree
,
2266 vd
->vdev_dtl
[DTL_SCRUB
], 2);
2267 space_reftree_generate_map(&reftree
,
2268 vd
->vdev_dtl
[DTL_MISSING
], 1);
2269 space_reftree_destroy(&reftree
);
2271 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
2272 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2273 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
2275 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
2276 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
2277 if (!vdev_readable(vd
))
2278 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
2280 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2281 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
2284 * If the vdev was resilvering and no longer has any
2285 * DTLs then reset its resilvering flag.
2287 if (vd
->vdev_resilver_txg
!= 0 &&
2288 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2289 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
]))
2290 vd
->vdev_resilver_txg
= 0;
2292 mutex_exit(&vd
->vdev_dtl_lock
);
2295 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
2299 mutex_enter(&vd
->vdev_dtl_lock
);
2300 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2301 /* account for child's outage in parent's missing map */
2302 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
2304 continue; /* leaf vdevs only */
2305 if (t
== DTL_PARTIAL
)
2306 minref
= 1; /* i.e. non-zero */
2307 else if (vd
->vdev_nparity
!= 0)
2308 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
2310 minref
= vd
->vdev_children
; /* any kind of mirror */
2311 space_reftree_create(&reftree
);
2312 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2313 vdev_t
*cvd
= vd
->vdev_child
[c
];
2314 mutex_enter(&cvd
->vdev_dtl_lock
);
2315 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
2316 mutex_exit(&cvd
->vdev_dtl_lock
);
2318 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
2319 space_reftree_destroy(&reftree
);
2321 mutex_exit(&vd
->vdev_dtl_lock
);
2325 vdev_dtl_load(vdev_t
*vd
)
2327 spa_t
*spa
= vd
->vdev_spa
;
2328 objset_t
*mos
= spa
->spa_meta_objset
;
2331 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2332 ASSERT(vdev_is_concrete(vd
));
2334 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2335 vd
->vdev_dtl_object
, 0, -1ULL, 0);
2338 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2340 mutex_enter(&vd
->vdev_dtl_lock
);
2343 * Now that we've opened the space_map we need to update
2346 space_map_update(vd
->vdev_dtl_sm
);
2348 error
= space_map_load(vd
->vdev_dtl_sm
,
2349 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2350 mutex_exit(&vd
->vdev_dtl_lock
);
2355 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2356 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2365 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2367 spa_t
*spa
= vd
->vdev_spa
;
2369 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2370 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2375 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2377 spa_t
*spa
= vd
->vdev_spa
;
2378 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2379 DMU_OT_NONE
, 0, tx
);
2382 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2389 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2391 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2392 vd
->vdev_ops
!= &vdev_missing_ops
&&
2393 vd
->vdev_ops
!= &vdev_root_ops
&&
2394 !vd
->vdev_top
->vdev_removing
) {
2395 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2396 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2398 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2399 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2402 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2403 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2408 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2410 spa_t
*spa
= vd
->vdev_spa
;
2411 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2412 objset_t
*mos
= spa
->spa_meta_objset
;
2413 range_tree_t
*rtsync
;
2415 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2417 ASSERT(vdev_is_concrete(vd
));
2418 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2420 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2422 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2423 mutex_enter(&vd
->vdev_dtl_lock
);
2424 space_map_free(vd
->vdev_dtl_sm
, tx
);
2425 space_map_close(vd
->vdev_dtl_sm
);
2426 vd
->vdev_dtl_sm
= NULL
;
2427 mutex_exit(&vd
->vdev_dtl_lock
);
2430 * We only destroy the leaf ZAP for detached leaves or for
2431 * removed log devices. Removed data devices handle leaf ZAP
2432 * cleanup later, once cancellation is no longer possible.
2434 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2435 vd
->vdev_top
->vdev_islog
)) {
2436 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2437 vd
->vdev_leaf_zap
= 0;
2444 if (vd
->vdev_dtl_sm
== NULL
) {
2445 uint64_t new_object
;
2447 new_object
= space_map_alloc(mos
, vdev_dtl_sm_blksz
, tx
);
2448 VERIFY3U(new_object
, !=, 0);
2450 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2452 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2455 rtsync
= range_tree_create(NULL
, NULL
);
2457 mutex_enter(&vd
->vdev_dtl_lock
);
2458 range_tree_walk(rt
, range_tree_add
, rtsync
);
2459 mutex_exit(&vd
->vdev_dtl_lock
);
2461 space_map_truncate(vd
->vdev_dtl_sm
, vdev_dtl_sm_blksz
, tx
);
2462 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
2463 range_tree_vacate(rtsync
, NULL
, NULL
);
2465 range_tree_destroy(rtsync
);
2468 * If the object for the space map has changed then dirty
2469 * the top level so that we update the config.
2471 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2472 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
2473 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
2474 (u_longlong_t
)object
,
2475 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
2476 vdev_config_dirty(vd
->vdev_top
);
2481 mutex_enter(&vd
->vdev_dtl_lock
);
2482 space_map_update(vd
->vdev_dtl_sm
);
2483 mutex_exit(&vd
->vdev_dtl_lock
);
2487 * Determine whether the specified vdev can be offlined/detached/removed
2488 * without losing data.
2491 vdev_dtl_required(vdev_t
*vd
)
2493 spa_t
*spa
= vd
->vdev_spa
;
2494 vdev_t
*tvd
= vd
->vdev_top
;
2495 uint8_t cant_read
= vd
->vdev_cant_read
;
2498 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2500 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2504 * Temporarily mark the device as unreadable, and then determine
2505 * whether this results in any DTL outages in the top-level vdev.
2506 * If not, we can safely offline/detach/remove the device.
2508 vd
->vdev_cant_read
= B_TRUE
;
2509 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2510 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2511 vd
->vdev_cant_read
= cant_read
;
2512 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2514 if (!required
&& zio_injection_enabled
)
2515 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2521 * Determine if resilver is needed, and if so the txg range.
2524 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2526 boolean_t needed
= B_FALSE
;
2527 uint64_t thismin
= UINT64_MAX
;
2528 uint64_t thismax
= 0;
2530 if (vd
->vdev_children
== 0) {
2531 mutex_enter(&vd
->vdev_dtl_lock
);
2532 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2533 vdev_writeable(vd
)) {
2535 thismin
= vdev_dtl_min(vd
);
2536 thismax
= vdev_dtl_max(vd
);
2539 mutex_exit(&vd
->vdev_dtl_lock
);
2541 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2542 vdev_t
*cvd
= vd
->vdev_child
[c
];
2543 uint64_t cmin
, cmax
;
2545 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2546 thismin
= MIN(thismin
, cmin
);
2547 thismax
= MAX(thismax
, cmax
);
2553 if (needed
&& minp
) {
2561 * Gets the checkpoint space map object from the vdev's ZAP.
2562 * Returns the spacemap object, or 0 if it wasn't in the ZAP
2563 * or the ZAP doesn't exist yet.
2566 vdev_checkpoint_sm_object(vdev_t
*vd
)
2568 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
2569 if (vd
->vdev_top_zap
== 0) {
2573 uint64_t sm_obj
= 0;
2574 int err
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
2575 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, &sm_obj
);
2577 ASSERT(err
== 0 || err
== ENOENT
);
2583 vdev_load(vdev_t
*vd
)
2587 * Recursively load all children.
2589 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2590 error
= vdev_load(vd
->vdev_child
[c
]);
2596 vdev_set_deflate_ratio(vd
);
2599 * If this is a top-level vdev, initialize its metaslabs.
2601 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
2602 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
2603 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2604 VDEV_AUX_CORRUPT_DATA
);
2605 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
2606 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
2607 (u_longlong_t
)vd
->vdev_asize
);
2608 return (SET_ERROR(ENXIO
));
2609 } else if ((error
= vdev_metaslab_init(vd
, 0)) != 0) {
2610 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
2611 "[error=%d]", error
);
2612 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2613 VDEV_AUX_CORRUPT_DATA
);
2617 uint64_t checkpoint_sm_obj
= vdev_checkpoint_sm_object(vd
);
2618 if (checkpoint_sm_obj
!= 0) {
2619 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
2620 ASSERT(vd
->vdev_asize
!= 0);
2621 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
2623 if ((error
= space_map_open(&vd
->vdev_checkpoint_sm
,
2624 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
2625 vd
->vdev_ashift
))) {
2626 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
2627 "failed for checkpoint spacemap (obj %llu) "
2629 (u_longlong_t
)checkpoint_sm_obj
, error
);
2632 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
2633 space_map_update(vd
->vdev_checkpoint_sm
);
2636 * Since the checkpoint_sm contains free entries
2637 * exclusively we can use sm_alloc to indicate the
2638 * culmulative checkpointed space that has been freed.
2640 vd
->vdev_stat
.vs_checkpoint_space
=
2641 -vd
->vdev_checkpoint_sm
->sm_alloc
;
2642 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
2643 vd
->vdev_stat
.vs_checkpoint_space
;
2648 * If this is a leaf vdev, load its DTL.
2650 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
2651 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2652 VDEV_AUX_CORRUPT_DATA
);
2653 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
2654 "[error=%d]", error
);
2658 uint64_t obsolete_sm_object
= vdev_obsolete_sm_object(vd
);
2659 if (obsolete_sm_object
!= 0) {
2660 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
2661 ASSERT(vd
->vdev_asize
!= 0);
2662 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
2664 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
2665 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
2666 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2667 VDEV_AUX_CORRUPT_DATA
);
2668 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
2669 "obsolete spacemap (obj %llu) [error=%d]",
2670 (u_longlong_t
)obsolete_sm_object
, error
);
2673 space_map_update(vd
->vdev_obsolete_sm
);
2680 * The special vdev case is used for hot spares and l2cache devices. Its
2681 * sole purpose it to set the vdev state for the associated vdev. To do this,
2682 * we make sure that we can open the underlying device, then try to read the
2683 * label, and make sure that the label is sane and that it hasn't been
2684 * repurposed to another pool.
2687 vdev_validate_aux(vdev_t
*vd
)
2690 uint64_t guid
, version
;
2693 if (!vdev_readable(vd
))
2696 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2697 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2698 VDEV_AUX_CORRUPT_DATA
);
2702 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2703 !SPA_VERSION_IS_SUPPORTED(version
) ||
2704 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2705 guid
!= vd
->vdev_guid
||
2706 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2707 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2708 VDEV_AUX_CORRUPT_DATA
);
2714 * We don't actually check the pool state here. If it's in fact in
2715 * use by another pool, we update this fact on the fly when requested.
2722 * Free the objects used to store this vdev's spacemaps, and the array
2723 * that points to them.
2726 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2728 if (vd
->vdev_ms_array
== 0)
2731 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
2732 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
2733 size_t array_bytes
= array_count
* sizeof (uint64_t);
2734 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
2735 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
2736 array_bytes
, smobj_array
, 0));
2738 for (uint64_t i
= 0; i
< array_count
; i
++) {
2739 uint64_t smobj
= smobj_array
[i
];
2743 space_map_free_obj(mos
, smobj
, tx
);
2746 kmem_free(smobj_array
, array_bytes
);
2747 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
2748 vd
->vdev_ms_array
= 0;
2752 vdev_remove_empty(vdev_t
*vd
, uint64_t txg
)
2754 spa_t
*spa
= vd
->vdev_spa
;
2757 ASSERT(vd
== vd
->vdev_top
);
2758 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2760 if (vd
->vdev_ms
!= NULL
) {
2761 metaslab_group_t
*mg
= vd
->vdev_mg
;
2763 metaslab_group_histogram_verify(mg
);
2764 metaslab_class_histogram_verify(mg
->mg_class
);
2766 for (int m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2767 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2769 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2772 mutex_enter(&msp
->ms_lock
);
2774 * If the metaslab was not loaded when the vdev
2775 * was removed then the histogram accounting may
2776 * not be accurate. Update the histogram information
2777 * here so that we ensure that the metaslab group
2778 * and metaslab class are up-to-date.
2780 metaslab_group_histogram_remove(mg
, msp
);
2782 VERIFY0(space_map_allocated(msp
->ms_sm
));
2783 space_map_close(msp
->ms_sm
);
2785 mutex_exit(&msp
->ms_lock
);
2788 if (vd
->vdev_checkpoint_sm
!= NULL
) {
2789 ASSERT(spa_has_checkpoint(spa
));
2790 space_map_close(vd
->vdev_checkpoint_sm
);
2791 vd
->vdev_checkpoint_sm
= NULL
;
2794 metaslab_group_histogram_verify(mg
);
2795 metaslab_class_histogram_verify(mg
->mg_class
);
2796 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2797 ASSERT0(mg
->mg_histogram
[i
]);
2800 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
2801 vdev_destroy_spacemaps(vd
, tx
);
2803 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
2804 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
2805 vd
->vdev_top_zap
= 0;
2811 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
2814 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
2816 ASSERT(vdev_is_concrete(vd
));
2818 while (msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
2819 metaslab_sync_done(msp
, txg
);
2822 metaslab_sync_reassess(vd
->vdev_mg
);
2826 vdev_sync(vdev_t
*vd
, uint64_t txg
)
2828 spa_t
*spa
= vd
->vdev_spa
;
2833 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
2836 ASSERT(vd
->vdev_removing
||
2837 vd
->vdev_ops
== &vdev_indirect_ops
);
2839 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2840 vdev_indirect_sync_obsolete(vd
, tx
);
2844 * If the vdev is indirect, it can't have dirty
2845 * metaslabs or DTLs.
2847 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
2848 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
2849 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
2854 ASSERT(vdev_is_concrete(vd
));
2856 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
2857 !vd
->vdev_removing
) {
2858 ASSERT(vd
== vd
->vdev_top
);
2859 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
2860 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2861 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
2862 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
2863 ASSERT(vd
->vdev_ms_array
!= 0);
2864 vdev_config_dirty(vd
);
2868 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
2869 metaslab_sync(msp
, txg
);
2870 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
2873 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
2874 vdev_dtl_sync(lvd
, txg
);
2877 * Remove the metadata associated with this vdev once it's empty.
2878 * Note that this is typically used for log/cache device removal;
2879 * we don't empty toplevel vdevs when removing them. But if
2880 * a toplevel happens to be emptied, this is not harmful.
2882 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
) {
2883 vdev_remove_empty(vd
, txg
);
2886 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
2890 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
2892 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
2896 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2897 * not be opened, and no I/O is attempted.
2900 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2904 spa_vdev_state_enter(spa
, SCL_NONE
);
2906 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2907 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2909 if (!vd
->vdev_ops
->vdev_op_leaf
)
2910 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2915 * We don't directly use the aux state here, but if we do a
2916 * vdev_reopen(), we need this value to be present to remember why we
2919 vd
->vdev_label_aux
= aux
;
2922 * Faulted state takes precedence over degraded.
2924 vd
->vdev_delayed_close
= B_FALSE
;
2925 vd
->vdev_faulted
= 1ULL;
2926 vd
->vdev_degraded
= 0ULL;
2927 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
2930 * If this device has the only valid copy of the data, then
2931 * back off and simply mark the vdev as degraded instead.
2933 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
2934 vd
->vdev_degraded
= 1ULL;
2935 vd
->vdev_faulted
= 0ULL;
2938 * If we reopen the device and it's not dead, only then do we
2943 if (vdev_readable(vd
))
2944 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
2947 return (spa_vdev_state_exit(spa
, vd
, 0));
2951 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2952 * user that something is wrong. The vdev continues to operate as normal as far
2953 * as I/O is concerned.
2956 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
2960 spa_vdev_state_enter(spa
, SCL_NONE
);
2962 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
2963 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
2965 if (!vd
->vdev_ops
->vdev_op_leaf
)
2966 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
2969 * If the vdev is already faulted, then don't do anything.
2971 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
2972 return (spa_vdev_state_exit(spa
, NULL
, 0));
2974 vd
->vdev_degraded
= 1ULL;
2975 if (!vdev_is_dead(vd
))
2976 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
2979 return (spa_vdev_state_exit(spa
, vd
, 0));
2983 * Online the given vdev.
2985 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2986 * spare device should be detached when the device finishes resilvering.
2987 * Second, the online should be treated like a 'test' online case, so no FMA
2988 * events are generated if the device fails to open.
2991 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
2993 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
2994 boolean_t wasoffline
;
2995 vdev_state_t oldstate
;
2997 spa_vdev_state_enter(spa
, SCL_NONE
);
2999 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3000 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3002 if (!vd
->vdev_ops
->vdev_op_leaf
)
3003 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3005 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
3006 oldstate
= vd
->vdev_state
;
3009 vd
->vdev_offline
= B_FALSE
;
3010 vd
->vdev_tmpoffline
= B_FALSE
;
3011 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
3012 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
3014 /* XXX - L2ARC 1.0 does not support expansion */
3015 if (!vd
->vdev_aux
) {
3016 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3017 pvd
->vdev_expanding
= !!(flags
& ZFS_ONLINE_EXPAND
);
3021 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
3023 if (!vd
->vdev_aux
) {
3024 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3025 pvd
->vdev_expanding
= B_FALSE
;
3029 *newstate
= vd
->vdev_state
;
3030 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
3031 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
3032 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3033 vd
->vdev_parent
->vdev_child
[0] == vd
)
3034 vd
->vdev_unspare
= B_TRUE
;
3036 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
3038 /* XXX - L2ARC 1.0 does not support expansion */
3040 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
3041 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3045 (oldstate
< VDEV_STATE_DEGRADED
&&
3046 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
3047 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
3049 return (spa_vdev_state_exit(spa
, vd
, 0));
3053 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3057 uint64_t generation
;
3058 metaslab_group_t
*mg
;
3061 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3063 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3064 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3066 if (!vd
->vdev_ops
->vdev_op_leaf
)
3067 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3071 generation
= spa
->spa_config_generation
+ 1;
3074 * If the device isn't already offline, try to offline it.
3076 if (!vd
->vdev_offline
) {
3078 * If this device has the only valid copy of some data,
3079 * don't allow it to be offlined. Log devices are always
3082 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3083 vdev_dtl_required(vd
))
3084 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3087 * If the top-level is a slog and it has had allocations
3088 * then proceed. We check that the vdev's metaslab group
3089 * is not NULL since it's possible that we may have just
3090 * added this vdev but not yet initialized its metaslabs.
3092 if (tvd
->vdev_islog
&& mg
!= NULL
) {
3094 * Prevent any future allocations.
3096 metaslab_group_passivate(mg
);
3097 (void) spa_vdev_state_exit(spa
, vd
, 0);
3099 error
= spa_reset_logs(spa
);
3102 * If the log device was successfully reset but has
3103 * checkpointed data, do not offline it.
3106 tvd
->vdev_checkpoint_sm
!= NULL
) {
3107 ASSERT3U(tvd
->vdev_checkpoint_sm
->sm_alloc
,
3109 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
3112 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3115 * Check to see if the config has changed.
3117 if (error
|| generation
!= spa
->spa_config_generation
) {
3118 metaslab_group_activate(mg
);
3120 return (spa_vdev_state_exit(spa
,
3122 (void) spa_vdev_state_exit(spa
, vd
, 0);
3125 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
3129 * Offline this device and reopen its top-level vdev.
3130 * If the top-level vdev is a log device then just offline
3131 * it. Otherwise, if this action results in the top-level
3132 * vdev becoming unusable, undo it and fail the request.
3134 vd
->vdev_offline
= B_TRUE
;
3137 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3138 vdev_is_dead(tvd
)) {
3139 vd
->vdev_offline
= B_FALSE
;
3141 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3145 * Add the device back into the metaslab rotor so that
3146 * once we online the device it's open for business.
3148 if (tvd
->vdev_islog
&& mg
!= NULL
)
3149 metaslab_group_activate(mg
);
3152 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
3154 return (spa_vdev_state_exit(spa
, vd
, 0));
3158 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3162 mutex_enter(&spa
->spa_vdev_top_lock
);
3163 error
= vdev_offline_locked(spa
, guid
, flags
);
3164 mutex_exit(&spa
->spa_vdev_top_lock
);
3170 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3171 * vdev_offline(), we assume the spa config is locked. We also clear all
3172 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3175 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
3177 vdev_t
*rvd
= spa
->spa_root_vdev
;
3179 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3184 vd
->vdev_stat
.vs_read_errors
= 0;
3185 vd
->vdev_stat
.vs_write_errors
= 0;
3186 vd
->vdev_stat
.vs_checksum_errors
= 0;
3188 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3189 vdev_clear(spa
, vd
->vdev_child
[c
]);
3192 * It makes no sense to "clear" an indirect vdev.
3194 if (!vdev_is_concrete(vd
))
3198 * If we're in the FAULTED state or have experienced failed I/O, then
3199 * clear the persistent state and attempt to reopen the device. We
3200 * also mark the vdev config dirty, so that the new faulted state is
3201 * written out to disk.
3203 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
3204 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
3207 * When reopening in reponse to a clear event, it may be due to
3208 * a fmadm repair request. In this case, if the device is
3209 * still broken, we want to still post the ereport again.
3211 vd
->vdev_forcefault
= B_TRUE
;
3213 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
3214 vd
->vdev_cant_read
= B_FALSE
;
3215 vd
->vdev_cant_write
= B_FALSE
;
3217 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
3219 vd
->vdev_forcefault
= B_FALSE
;
3221 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
3222 vdev_state_dirty(vd
->vdev_top
);
3224 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
3225 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3227 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
3231 * When clearing a FMA-diagnosed fault, we always want to
3232 * unspare the device, as we assume that the original spare was
3233 * done in response to the FMA fault.
3235 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
3236 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3237 vd
->vdev_parent
->vdev_child
[0] == vd
)
3238 vd
->vdev_unspare
= B_TRUE
;
3242 vdev_is_dead(vdev_t
*vd
)
3245 * Holes and missing devices are always considered "dead".
3246 * This simplifies the code since we don't have to check for
3247 * these types of devices in the various code paths.
3248 * Instead we rely on the fact that we skip over dead devices
3249 * before issuing I/O to them.
3251 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
3252 vd
->vdev_ops
== &vdev_hole_ops
||
3253 vd
->vdev_ops
== &vdev_missing_ops
);
3257 vdev_readable(vdev_t
*vd
)
3259 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
3263 vdev_writeable(vdev_t
*vd
)
3265 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
3266 vdev_is_concrete(vd
));
3270 vdev_allocatable(vdev_t
*vd
)
3272 uint64_t state
= vd
->vdev_state
;
3275 * We currently allow allocations from vdevs which may be in the
3276 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3277 * fails to reopen then we'll catch it later when we're holding
3278 * the proper locks. Note that we have to get the vdev state
3279 * in a local variable because although it changes atomically,
3280 * we're asking two separate questions about it.
3282 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
3283 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
3284 vd
->vdev_mg
->mg_initialized
);
3288 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
3290 ASSERT(zio
->io_vd
== vd
);
3292 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
3295 if (zio
->io_type
== ZIO_TYPE_READ
)
3296 return (!vd
->vdev_cant_read
);
3298 if (zio
->io_type
== ZIO_TYPE_WRITE
)
3299 return (!vd
->vdev_cant_write
);
3305 vdev_is_spacemap_addressable(vdev_t
*vd
)
3308 * Assuming 47 bits of the space map entry dedicated for the entry's
3309 * offset (see description in space_map.h), we calculate the maximum
3310 * address that can be described by a space map entry for the given
3313 uint64_t shift
= vd
->vdev_ashift
+ 47;
3315 if (shift
>= 63) /* detect potential overflow */
3318 return (vd
->vdev_asize
< (1ULL << shift
));
3322 * Get statistics for the given vdev.
3325 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
3327 spa_t
*spa
= vd
->vdev_spa
;
3328 vdev_t
*rvd
= spa
->spa_root_vdev
;
3329 vdev_t
*tvd
= vd
->vdev_top
;
3331 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
3333 mutex_enter(&vd
->vdev_stat_lock
);
3334 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
3335 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
3336 vs
->vs_state
= vd
->vdev_state
;
3337 vs
->vs_rsize
= vdev_get_min_asize(vd
);
3338 if (vd
->vdev_ops
->vdev_op_leaf
)
3339 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
;
3341 * Report expandable space on top-level, non-auxillary devices only.
3342 * The expandable space is reported in terms of metaslab sized units
3343 * since that determines how much space the pool can expand.
3345 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
3346 vs
->vs_esize
= P2ALIGN(vd
->vdev_max_asize
- vd
->vdev_asize
-
3347 spa
->spa_bootsize
, 1ULL << tvd
->vdev_ms_shift
);
3349 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
3350 vdev_is_concrete(vd
)) {
3351 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
3355 * If we're getting stats on the root vdev, aggregate the I/O counts
3356 * over all top-level vdevs (i.e. the direct children of the root).
3359 for (int c
= 0; c
< rvd
->vdev_children
; c
++) {
3360 vdev_t
*cvd
= rvd
->vdev_child
[c
];
3361 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
3363 for (int t
= 0; t
< ZIO_TYPES
; t
++) {
3364 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
3365 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
3367 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
3370 mutex_exit(&vd
->vdev_stat_lock
);
3374 vdev_clear_stats(vdev_t
*vd
)
3376 mutex_enter(&vd
->vdev_stat_lock
);
3377 vd
->vdev_stat
.vs_space
= 0;
3378 vd
->vdev_stat
.vs_dspace
= 0;
3379 vd
->vdev_stat
.vs_alloc
= 0;
3380 mutex_exit(&vd
->vdev_stat_lock
);
3384 vdev_scan_stat_init(vdev_t
*vd
)
3386 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3388 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3389 vdev_scan_stat_init(vd
->vdev_child
[c
]);
3391 mutex_enter(&vd
->vdev_stat_lock
);
3392 vs
->vs_scan_processed
= 0;
3393 mutex_exit(&vd
->vdev_stat_lock
);
3397 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
3399 spa_t
*spa
= zio
->io_spa
;
3400 vdev_t
*rvd
= spa
->spa_root_vdev
;
3401 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
3403 uint64_t txg
= zio
->io_txg
;
3404 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3405 zio_type_t type
= zio
->io_type
;
3406 int flags
= zio
->io_flags
;
3409 * If this i/o is a gang leader, it didn't do any actual work.
3411 if (zio
->io_gang_tree
)
3414 if (zio
->io_error
== 0) {
3416 * If this is a root i/o, don't count it -- we've already
3417 * counted the top-level vdevs, and vdev_get_stats() will
3418 * aggregate them when asked. This reduces contention on
3419 * the root vdev_stat_lock and implicitly handles blocks
3420 * that compress away to holes, for which there is no i/o.
3421 * (Holes never create vdev children, so all the counters
3422 * remain zero, which is what we want.)
3424 * Note: this only applies to successful i/o (io_error == 0)
3425 * because unlike i/o counts, errors are not additive.
3426 * When reading a ditto block, for example, failure of
3427 * one top-level vdev does not imply a root-level error.
3432 ASSERT(vd
== zio
->io_vd
);
3434 if (flags
& ZIO_FLAG_IO_BYPASS
)
3437 mutex_enter(&vd
->vdev_stat_lock
);
3439 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3440 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3441 dsl_scan_phys_t
*scn_phys
=
3442 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3443 uint64_t *processed
= &scn_phys
->scn_processed
;
3446 if (vd
->vdev_ops
->vdev_op_leaf
)
3447 atomic_add_64(processed
, psize
);
3448 vs
->vs_scan_processed
+= psize
;
3451 if (flags
& ZIO_FLAG_SELF_HEAL
)
3452 vs
->vs_self_healed
+= psize
;
3456 vs
->vs_bytes
[type
] += psize
;
3458 mutex_exit(&vd
->vdev_stat_lock
);
3462 if (flags
& ZIO_FLAG_SPECULATIVE
)
3466 * If this is an I/O error that is going to be retried, then ignore the
3467 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3468 * hard errors, when in reality they can happen for any number of
3469 * innocuous reasons (bus resets, MPxIO link failure, etc).
3471 if (zio
->io_error
== EIO
&&
3472 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3476 * Intent logs writes won't propagate their error to the root
3477 * I/O so don't mark these types of failures as pool-level
3480 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3483 mutex_enter(&vd
->vdev_stat_lock
);
3484 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3485 if (zio
->io_error
== ECKSUM
)
3486 vs
->vs_checksum_errors
++;
3488 vs
->vs_read_errors
++;
3490 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3491 vs
->vs_write_errors
++;
3492 mutex_exit(&vd
->vdev_stat_lock
);
3494 if (spa
->spa_load_state
== SPA_LOAD_NONE
&&
3495 type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3496 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3497 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3498 spa
->spa_claiming
)) {
3500 * This is either a normal write (not a repair), or it's
3501 * a repair induced by the scrub thread, or it's a repair
3502 * made by zil_claim() during spa_load() in the first txg.
3503 * In the normal case, we commit the DTL change in the same
3504 * txg as the block was born. In the scrub-induced repair
3505 * case, we know that scrubs run in first-pass syncing context,
3506 * so we commit the DTL change in spa_syncing_txg(spa).
3507 * In the zil_claim() case, we commit in spa_first_txg(spa).
3509 * We currently do not make DTL entries for failed spontaneous
3510 * self-healing writes triggered by normal (non-scrubbing)
3511 * reads, because we have no transactional context in which to
3512 * do so -- and it's not clear that it'd be desirable anyway.
3514 if (vd
->vdev_ops
->vdev_op_leaf
) {
3515 uint64_t commit_txg
= txg
;
3516 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3517 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3518 ASSERT(spa_sync_pass(spa
) == 1);
3519 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3520 commit_txg
= spa_syncing_txg(spa
);
3521 } else if (spa
->spa_claiming
) {
3522 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3523 commit_txg
= spa_first_txg(spa
);
3525 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3526 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3528 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3529 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3530 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3533 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3538 * Update the in-core space usage stats for this vdev, its metaslab class,
3539 * and the root vdev.
3542 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3543 int64_t space_delta
)
3545 int64_t dspace_delta
= space_delta
;
3546 spa_t
*spa
= vd
->vdev_spa
;
3547 vdev_t
*rvd
= spa
->spa_root_vdev
;
3548 metaslab_group_t
*mg
= vd
->vdev_mg
;
3549 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3551 ASSERT(vd
== vd
->vdev_top
);
3554 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3555 * factor. We must calculate this here and not at the root vdev
3556 * because the root vdev's psize-to-asize is simply the max of its
3557 * childrens', thus not accurate enough for us.
3559 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3560 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3561 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3562 vd
->vdev_deflate_ratio
;
3564 mutex_enter(&vd
->vdev_stat_lock
);
3565 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3566 vd
->vdev_stat
.vs_space
+= space_delta
;
3567 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3568 mutex_exit(&vd
->vdev_stat_lock
);
3570 if (mc
== spa_normal_class(spa
)) {
3571 mutex_enter(&rvd
->vdev_stat_lock
);
3572 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3573 rvd
->vdev_stat
.vs_space
+= space_delta
;
3574 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3575 mutex_exit(&rvd
->vdev_stat_lock
);
3579 ASSERT(rvd
== vd
->vdev_parent
);
3580 ASSERT(vd
->vdev_ms_count
!= 0);
3582 metaslab_class_space_update(mc
,
3583 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3588 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3589 * so that it will be written out next time the vdev configuration is synced.
3590 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3593 vdev_config_dirty(vdev_t
*vd
)
3595 spa_t
*spa
= vd
->vdev_spa
;
3596 vdev_t
*rvd
= spa
->spa_root_vdev
;
3599 ASSERT(spa_writeable(spa
));
3602 * If this is an aux vdev (as with l2cache and spare devices), then we
3603 * update the vdev config manually and set the sync flag.
3605 if (vd
->vdev_aux
!= NULL
) {
3606 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3610 for (c
= 0; c
< sav
->sav_count
; c
++) {
3611 if (sav
->sav_vdevs
[c
] == vd
)
3615 if (c
== sav
->sav_count
) {
3617 * We're being removed. There's nothing more to do.
3619 ASSERT(sav
->sav_sync
== B_TRUE
);
3623 sav
->sav_sync
= B_TRUE
;
3625 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3626 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3627 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3628 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3634 * Setting the nvlist in the middle if the array is a little
3635 * sketchy, but it will work.
3637 nvlist_free(aux
[c
]);
3638 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3644 * The dirty list is protected by the SCL_CONFIG lock. The caller
3645 * must either hold SCL_CONFIG as writer, or must be the sync thread
3646 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3647 * so this is sufficient to ensure mutual exclusion.
3649 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3650 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3651 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3654 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3655 vdev_config_dirty(rvd
->vdev_child
[c
]);
3657 ASSERT(vd
== vd
->vdev_top
);
3659 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3660 vdev_is_concrete(vd
)) {
3661 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
3667 vdev_config_clean(vdev_t
*vd
)
3669 spa_t
*spa
= vd
->vdev_spa
;
3671 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3672 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3673 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3675 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
3676 list_remove(&spa
->spa_config_dirty_list
, vd
);
3680 * Mark a top-level vdev's state as dirty, so that the next pass of
3681 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3682 * the state changes from larger config changes because they require
3683 * much less locking, and are often needed for administrative actions.
3686 vdev_state_dirty(vdev_t
*vd
)
3688 spa_t
*spa
= vd
->vdev_spa
;
3690 ASSERT(spa_writeable(spa
));
3691 ASSERT(vd
== vd
->vdev_top
);
3694 * The state list is protected by the SCL_STATE lock. The caller
3695 * must either hold SCL_STATE as writer, or must be the sync thread
3696 * (which holds SCL_STATE as reader). There's only one sync thread,
3697 * so this is sufficient to ensure mutual exclusion.
3699 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3700 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3701 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3703 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
3704 vdev_is_concrete(vd
))
3705 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
3709 vdev_state_clean(vdev_t
*vd
)
3711 spa_t
*spa
= vd
->vdev_spa
;
3713 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
3714 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3715 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
3717 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
3718 list_remove(&spa
->spa_state_dirty_list
, vd
);
3722 * Propagate vdev state up from children to parent.
3725 vdev_propagate_state(vdev_t
*vd
)
3727 spa_t
*spa
= vd
->vdev_spa
;
3728 vdev_t
*rvd
= spa
->spa_root_vdev
;
3729 int degraded
= 0, faulted
= 0;
3733 if (vd
->vdev_children
> 0) {
3734 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3735 child
= vd
->vdev_child
[c
];
3738 * Don't factor holes or indirect vdevs into the
3741 if (!vdev_is_concrete(child
))
3744 if (!vdev_readable(child
) ||
3745 (!vdev_writeable(child
) && spa_writeable(spa
))) {
3747 * Root special: if there is a top-level log
3748 * device, treat the root vdev as if it were
3751 if (child
->vdev_islog
&& vd
== rvd
)
3755 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
3759 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
3763 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
3766 * Root special: if there is a top-level vdev that cannot be
3767 * opened due to corrupted metadata, then propagate the root
3768 * vdev's aux state as 'corrupt' rather than 'insufficient
3771 if (corrupted
&& vd
== rvd
&&
3772 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
3773 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3774 VDEV_AUX_CORRUPT_DATA
);
3777 if (vd
->vdev_parent
)
3778 vdev_propagate_state(vd
->vdev_parent
);
3782 * Set a vdev's state. If this is during an open, we don't update the parent
3783 * state, because we're in the process of opening children depth-first.
3784 * Otherwise, we propagate the change to the parent.
3786 * If this routine places a device in a faulted state, an appropriate ereport is
3790 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
3792 uint64_t save_state
;
3793 spa_t
*spa
= vd
->vdev_spa
;
3795 if (state
== vd
->vdev_state
) {
3796 vd
->vdev_stat
.vs_aux
= aux
;
3800 save_state
= vd
->vdev_state
;
3802 vd
->vdev_state
= state
;
3803 vd
->vdev_stat
.vs_aux
= aux
;
3806 * If we are setting the vdev state to anything but an open state, then
3807 * always close the underlying device unless the device has requested
3808 * a delayed close (i.e. we're about to remove or fault the device).
3809 * Otherwise, we keep accessible but invalid devices open forever.
3810 * We don't call vdev_close() itself, because that implies some extra
3811 * checks (offline, etc) that we don't want here. This is limited to
3812 * leaf devices, because otherwise closing the device will affect other
3815 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
3816 vd
->vdev_ops
->vdev_op_leaf
)
3817 vd
->vdev_ops
->vdev_op_close(vd
);
3820 * If we have brought this vdev back into service, we need
3821 * to notify fmd so that it can gracefully repair any outstanding
3822 * cases due to a missing device. We do this in all cases, even those
3823 * that probably don't correlate to a repaired fault. This is sure to
3824 * catch all cases, and we let the zfs-retire agent sort it out. If
3825 * this is a transient state it's OK, as the retire agent will
3826 * double-check the state of the vdev before repairing it.
3828 if (state
== VDEV_STATE_HEALTHY
&& vd
->vdev_ops
->vdev_op_leaf
&&
3829 vd
->vdev_prevstate
!= state
)
3830 zfs_post_state_change(spa
, vd
);
3832 if (vd
->vdev_removed
&&
3833 state
== VDEV_STATE_CANT_OPEN
&&
3834 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
3836 * If the previous state is set to VDEV_STATE_REMOVED, then this
3837 * device was previously marked removed and someone attempted to
3838 * reopen it. If this failed due to a nonexistent device, then
3839 * keep the device in the REMOVED state. We also let this be if
3840 * it is one of our special test online cases, which is only
3841 * attempting to online the device and shouldn't generate an FMA
3844 vd
->vdev_state
= VDEV_STATE_REMOVED
;
3845 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
3846 } else if (state
== VDEV_STATE_REMOVED
) {
3847 vd
->vdev_removed
= B_TRUE
;
3848 } else if (state
== VDEV_STATE_CANT_OPEN
) {
3850 * If we fail to open a vdev during an import or recovery, we
3851 * mark it as "not available", which signifies that it was
3852 * never there to begin with. Failure to open such a device
3853 * is not considered an error.
3855 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
3856 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
3857 vd
->vdev_ops
->vdev_op_leaf
)
3858 vd
->vdev_not_present
= 1;
3861 * Post the appropriate ereport. If the 'prevstate' field is
3862 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3863 * that this is part of a vdev_reopen(). In this case, we don't
3864 * want to post the ereport if the device was already in the
3865 * CANT_OPEN state beforehand.
3867 * If the 'checkremove' flag is set, then this is an attempt to
3868 * online the device in response to an insertion event. If we
3869 * hit this case, then we have detected an insertion event for a
3870 * faulted or offline device that wasn't in the removed state.
3871 * In this scenario, we don't post an ereport because we are
3872 * about to replace the device, or attempt an online with
3873 * vdev_forcefault, which will generate the fault for us.
3875 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
3876 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
3877 vd
!= spa
->spa_root_vdev
) {
3881 case VDEV_AUX_OPEN_FAILED
:
3882 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
3884 case VDEV_AUX_CORRUPT_DATA
:
3885 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
3887 case VDEV_AUX_NO_REPLICAS
:
3888 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
3890 case VDEV_AUX_BAD_GUID_SUM
:
3891 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
3893 case VDEV_AUX_TOO_SMALL
:
3894 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
3896 case VDEV_AUX_BAD_LABEL
:
3897 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
3900 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
3903 zfs_ereport_post(class, spa
, vd
, NULL
, save_state
, 0);
3906 /* Erase any notion of persistent removed state */
3907 vd
->vdev_removed
= B_FALSE
;
3909 vd
->vdev_removed
= B_FALSE
;
3912 if (!isopen
&& vd
->vdev_parent
)
3913 vdev_propagate_state(vd
->vdev_parent
);
3917 vdev_children_are_offline(vdev_t
*vd
)
3919 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
3921 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3922 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
3930 * Check the vdev configuration to ensure that it's capable of supporting
3931 * a root pool. We do not support partial configuration.
3932 * In addition, only a single top-level vdev is allowed.
3935 vdev_is_bootable(vdev_t
*vd
)
3937 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3938 char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
3940 if (strcmp(vdev_type
, VDEV_TYPE_ROOT
) == 0 &&
3941 vd
->vdev_children
> 1) {
3943 } else if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0 ||
3944 strcmp(vdev_type
, VDEV_TYPE_INDIRECT
) == 0) {
3949 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3950 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
3957 vdev_is_concrete(vdev_t
*vd
)
3959 vdev_ops_t
*ops
= vd
->vdev_ops
;
3960 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
3961 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
3969 * Determine if a log device has valid content. If the vdev was
3970 * removed or faulted in the MOS config then we know that
3971 * the content on the log device has already been written to the pool.
3974 vdev_log_state_valid(vdev_t
*vd
)
3976 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
3980 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3981 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
3988 * Expand a vdev if possible.
3991 vdev_expand(vdev_t
*vd
, uint64_t txg
)
3993 ASSERT(vd
->vdev_top
== vd
);
3994 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
3996 vdev_set_deflate_ratio(vd
);
3998 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
3999 vdev_is_concrete(vd
)) {
4000 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
4001 vdev_config_dirty(vd
);
4009 vdev_split(vdev_t
*vd
)
4011 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
4013 vdev_remove_child(pvd
, vd
);
4014 vdev_compact_children(pvd
);
4016 cvd
= pvd
->vdev_child
[0];
4017 if (pvd
->vdev_children
== 1) {
4018 vdev_remove_parent(cvd
);
4019 cvd
->vdev_splitting
= B_TRUE
;
4021 vdev_propagate_state(cvd
);
4025 vdev_deadman(vdev_t
*vd
)
4027 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4028 vdev_t
*cvd
= vd
->vdev_child
[c
];
4033 if (vd
->vdev_ops
->vdev_op_leaf
) {
4034 vdev_queue_t
*vq
= &vd
->vdev_queue
;
4036 mutex_enter(&vq
->vq_lock
);
4037 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
4038 spa_t
*spa
= vd
->vdev_spa
;
4043 * Look at the head of all the pending queues,
4044 * if any I/O has been outstanding for longer than
4045 * the spa_deadman_synctime we panic the system.
4047 fio
= avl_first(&vq
->vq_active_tree
);
4048 delta
= gethrtime() - fio
->io_timestamp
;
4049 if (delta
> spa_deadman_synctime(spa
)) {
4050 vdev_dbgmsg(vd
, "SLOW IO: zio timestamp "
4051 "%lluns, delta %lluns, last io %lluns",
4052 fio
->io_timestamp
, (u_longlong_t
)delta
,
4053 vq
->vq_io_complete_ts
);
4054 fm_panic("I/O to pool '%s' appears to be "
4055 "hung.", spa_name(spa
));
4058 mutex_exit(&vq
->vq_lock
);