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>
53 #include <sys/zfs_ratelimit.h>
55 /* target number of metaslabs per top-level vdev */
56 int vdev_max_ms_count
= 200;
58 /* minimum number of metaslabs per top-level vdev */
59 int vdev_min_ms_count
= 16;
61 /* practical upper limit of total metaslabs per top-level vdev */
62 int vdev_ms_count_limit
= 1ULL << 17;
64 /* lower limit for metaslab size (512M) */
65 int vdev_default_ms_shift
= 29;
67 /* upper limit for metaslab size (256G) */
68 int vdev_max_ms_shift
= 38;
70 int vdev_validate_skip
= B_FALSE
;
73 * Since the DTL space map of a vdev is not expected to have a lot of
74 * entries, we default its block size to 4K.
76 int vdev_dtl_sm_blksz
= (1 << 12);
79 * Rate limit delay events to this many IO delays per second.
81 unsigned int zfs_delays_per_second
= 20;
84 * Rate limit checksum events after this many checksum errors per second.
86 unsigned int zfs_checksums_per_second
= 20;
89 * Ignore errors during scrub/resilver. Allows to work around resilver
90 * upon import when there are pool errors.
92 int zfs_scan_ignore_errors
= 0;
95 * vdev-wide space maps that have lots of entries written to them at
96 * the end of each transaction can benefit from a higher I/O bandwidth
97 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
99 int vdev_standard_sm_blksz
= (1 << 17);
103 vdev_dbgmsg(vdev_t
*vd
, const char *fmt
, ...)
109 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
112 if (vd
->vdev_path
!= NULL
) {
113 zfs_dbgmsg("%s vdev '%s': %s", vd
->vdev_ops
->vdev_op_type
,
116 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
117 vd
->vdev_ops
->vdev_op_type
,
118 (u_longlong_t
)vd
->vdev_id
,
119 (u_longlong_t
)vd
->vdev_guid
, buf
);
124 vdev_dbgmsg_print_tree(vdev_t
*vd
, int indent
)
128 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
) {
129 zfs_dbgmsg("%*svdev %u: %s", indent
, "", vd
->vdev_id
,
130 vd
->vdev_ops
->vdev_op_type
);
134 switch (vd
->vdev_state
) {
135 case VDEV_STATE_UNKNOWN
:
136 (void) snprintf(state
, sizeof (state
), "unknown");
138 case VDEV_STATE_CLOSED
:
139 (void) snprintf(state
, sizeof (state
), "closed");
141 case VDEV_STATE_OFFLINE
:
142 (void) snprintf(state
, sizeof (state
), "offline");
144 case VDEV_STATE_REMOVED
:
145 (void) snprintf(state
, sizeof (state
), "removed");
147 case VDEV_STATE_CANT_OPEN
:
148 (void) snprintf(state
, sizeof (state
), "can't open");
150 case VDEV_STATE_FAULTED
:
151 (void) snprintf(state
, sizeof (state
), "faulted");
153 case VDEV_STATE_DEGRADED
:
154 (void) snprintf(state
, sizeof (state
), "degraded");
156 case VDEV_STATE_HEALTHY
:
157 (void) snprintf(state
, sizeof (state
), "healthy");
160 (void) snprintf(state
, sizeof (state
), "<state %u>",
161 (uint_t
)vd
->vdev_state
);
164 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent
,
165 "", (int)vd
->vdev_id
, vd
->vdev_ops
->vdev_op_type
,
166 vd
->vdev_islog
? " (log)" : "",
167 (u_longlong_t
)vd
->vdev_guid
,
168 vd
->vdev_path
? vd
->vdev_path
: "N/A", state
);
170 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++)
171 vdev_dbgmsg_print_tree(vd
->vdev_child
[i
], indent
+ 2);
175 * Virtual device management.
178 static vdev_ops_t
*vdev_ops_table
[] = {
193 * Given a vdev type, return the appropriate ops vector.
196 vdev_getops(const char *type
)
198 vdev_ops_t
*ops
, **opspp
;
200 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
201 if (strcmp(ops
->vdev_op_type
, type
) == 0)
208 * Default asize function: return the MAX of psize with the asize of
209 * all children. This is what's used by anything other than RAID-Z.
212 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
214 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
217 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
218 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
219 asize
= MAX(asize
, csize
);
226 * Get the minimum allocatable size. We define the allocatable size as
227 * the vdev's asize rounded to the nearest metaslab. This allows us to
228 * replace or attach devices which don't have the same physical size but
229 * can still satisfy the same number of allocations.
232 vdev_get_min_asize(vdev_t
*vd
)
234 vdev_t
*pvd
= vd
->vdev_parent
;
237 * If our parent is NULL (inactive spare or cache) or is the root,
238 * just return our own asize.
241 return (vd
->vdev_asize
);
244 * The top-level vdev just returns the allocatable size rounded
245 * to the nearest metaslab.
247 if (vd
== vd
->vdev_top
)
248 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
251 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
252 * so each child must provide at least 1/Nth of its asize.
254 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
255 return ((pvd
->vdev_min_asize
+ pvd
->vdev_children
- 1) /
258 return (pvd
->vdev_min_asize
);
262 vdev_set_min_asize(vdev_t
*vd
)
264 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
266 for (int c
= 0; c
< vd
->vdev_children
; c
++)
267 vdev_set_min_asize(vd
->vdev_child
[c
]);
271 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
273 vdev_t
*rvd
= spa
->spa_root_vdev
;
275 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
277 if (vdev
< rvd
->vdev_children
) {
278 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
279 return (rvd
->vdev_child
[vdev
]);
286 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
290 if (vd
->vdev_guid
== guid
)
293 for (int c
= 0; c
< vd
->vdev_children
; c
++)
294 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
302 vdev_count_leaves_impl(vdev_t
*vd
)
306 if (vd
->vdev_ops
->vdev_op_leaf
)
309 for (int c
= 0; c
< vd
->vdev_children
; c
++)
310 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
316 vdev_count_leaves(spa_t
*spa
)
320 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
321 rc
= vdev_count_leaves_impl(spa
->spa_root_vdev
);
322 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
328 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
330 size_t oldsize
, newsize
;
331 uint64_t id
= cvd
->vdev_id
;
334 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
335 ASSERT(cvd
->vdev_parent
== NULL
);
337 cvd
->vdev_parent
= pvd
;
342 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
344 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
345 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
346 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
348 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
349 if (pvd
->vdev_child
!= NULL
) {
350 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
351 kmem_free(pvd
->vdev_child
, oldsize
);
354 pvd
->vdev_child
= newchild
;
355 pvd
->vdev_child
[id
] = cvd
;
357 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
358 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
361 * Walk up all ancestors to update guid sum.
363 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
364 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
368 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
371 uint_t id
= cvd
->vdev_id
;
373 ASSERT(cvd
->vdev_parent
== pvd
);
378 ASSERT(id
< pvd
->vdev_children
);
379 ASSERT(pvd
->vdev_child
[id
] == cvd
);
381 pvd
->vdev_child
[id
] = NULL
;
382 cvd
->vdev_parent
= NULL
;
384 for (c
= 0; c
< pvd
->vdev_children
; c
++)
385 if (pvd
->vdev_child
[c
])
388 if (c
== pvd
->vdev_children
) {
389 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
390 pvd
->vdev_child
= NULL
;
391 pvd
->vdev_children
= 0;
395 * Walk up all ancestors to update guid sum.
397 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
398 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
402 * Remove any holes in the child array.
405 vdev_compact_children(vdev_t
*pvd
)
407 vdev_t
**newchild
, *cvd
;
408 int oldc
= pvd
->vdev_children
;
411 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
416 for (int c
= newc
= 0; c
< oldc
; c
++)
417 if (pvd
->vdev_child
[c
])
421 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
423 for (int c
= newc
= 0; c
< oldc
; c
++) {
424 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
425 newchild
[newc
] = cvd
;
426 cvd
->vdev_id
= newc
++;
433 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
434 pvd
->vdev_child
= newchild
;
435 pvd
->vdev_children
= newc
;
439 * Allocate and minimally initialize a vdev_t.
442 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
445 vdev_indirect_config_t
*vic
;
447 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
448 vic
= &vd
->vdev_indirect_config
;
450 if (spa
->spa_root_vdev
== NULL
) {
451 ASSERT(ops
== &vdev_root_ops
);
452 spa
->spa_root_vdev
= vd
;
453 spa
->spa_load_guid
= spa_generate_guid(NULL
);
456 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
457 if (spa
->spa_root_vdev
== vd
) {
459 * The root vdev's guid will also be the pool guid,
460 * which must be unique among all pools.
462 guid
= spa_generate_guid(NULL
);
465 * Any other vdev's guid must be unique within the pool.
467 guid
= spa_generate_guid(spa
);
469 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
474 vd
->vdev_guid
= guid
;
475 vd
->vdev_guid_sum
= guid
;
477 vd
->vdev_state
= VDEV_STATE_CLOSED
;
478 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
479 vic
->vic_prev_indirect_vdev
= UINT64_MAX
;
481 rw_init(&vd
->vdev_indirect_rwlock
, NULL
, RW_DEFAULT
, NULL
);
482 mutex_init(&vd
->vdev_obsolete_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
483 vd
->vdev_obsolete_segments
= range_tree_create(NULL
, NULL
);
486 * Initialize rate limit structs for events. We rate limit ZIO delay
487 * and checksum events so that we don't overwhelm ZED with thousands
488 * of events when a disk is acting up.
490 zfs_ratelimit_init(&vd
->vdev_delay_rl
, &zfs_delays_per_second
, 1);
491 zfs_ratelimit_init(&vd
->vdev_checksum_rl
, &zfs_checksums_per_second
, 1);
493 list_link_init(&vd
->vdev_config_dirty_node
);
494 list_link_init(&vd
->vdev_state_dirty_node
);
495 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
496 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
497 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
498 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
499 mutex_init(&vd
->vdev_scan_io_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
501 for (int t
= 0; t
< DTL_TYPES
; t
++) {
502 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
);
504 txg_list_create(&vd
->vdev_ms_list
, spa
,
505 offsetof(struct metaslab
, ms_txg_node
));
506 txg_list_create(&vd
->vdev_dtl_list
, spa
,
507 offsetof(struct vdev
, vdev_dtl_node
));
508 vd
->vdev_stat
.vs_timestamp
= gethrtime();
516 * Allocate a new vdev. The 'alloctype' is used to control whether we are
517 * creating a new vdev or loading an existing one - the behavior is slightly
518 * different for each case.
521 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
526 uint64_t guid
= 0, islog
, nparity
;
528 vdev_indirect_config_t
*vic
;
532 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
534 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
535 return (SET_ERROR(EINVAL
));
537 if ((ops
= vdev_getops(type
)) == NULL
)
538 return (SET_ERROR(EINVAL
));
541 * If this is a load, get the vdev guid from the nvlist.
542 * Otherwise, vdev_alloc_common() will generate one for us.
544 if (alloctype
== VDEV_ALLOC_LOAD
) {
547 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
549 return (SET_ERROR(EINVAL
));
551 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
552 return (SET_ERROR(EINVAL
));
553 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
554 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
555 return (SET_ERROR(EINVAL
));
556 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
557 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
558 return (SET_ERROR(EINVAL
));
559 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
560 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
561 return (SET_ERROR(EINVAL
));
565 * The first allocated vdev must be of type 'root'.
567 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
568 return (SET_ERROR(EINVAL
));
571 * Determine whether we're a log vdev.
574 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
575 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
576 return (SET_ERROR(ENOTSUP
));
578 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
579 return (SET_ERROR(ENOTSUP
));
582 * Set the nparity property for RAID-Z vdevs.
585 if (ops
== &vdev_raidz_ops
) {
586 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
588 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
589 return (SET_ERROR(EINVAL
));
591 * Previous versions could only support 1 or 2 parity
595 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
596 return (SET_ERROR(ENOTSUP
));
598 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
599 return (SET_ERROR(ENOTSUP
));
602 * We require the parity to be specified for SPAs that
603 * support multiple parity levels.
605 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
606 return (SET_ERROR(EINVAL
));
608 * Otherwise, we default to 1 parity device for RAID-Z.
615 ASSERT(nparity
!= -1ULL);
617 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
618 vic
= &vd
->vdev_indirect_config
;
620 vd
->vdev_islog
= islog
;
621 vd
->vdev_nparity
= nparity
;
623 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
624 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
627 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
628 * fault on a vdev and want it to persist across imports (like with
631 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
632 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
633 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
634 vd
->vdev_faulted
= 1;
635 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
638 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
639 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
640 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
641 &vd
->vdev_physpath
) == 0)
642 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
644 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
645 &vd
->vdev_enc_sysfs_path
) == 0)
646 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
648 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
649 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
652 * Set the whole_disk property. If it's not specified, leave the value
655 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
656 &vd
->vdev_wholedisk
) != 0)
657 vd
->vdev_wholedisk
= -1ULL;
659 ASSERT0(vic
->vic_mapping_object
);
660 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
661 &vic
->vic_mapping_object
);
662 ASSERT0(vic
->vic_births_object
);
663 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
664 &vic
->vic_births_object
);
665 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
666 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
667 &vic
->vic_prev_indirect_vdev
);
670 * Look for the 'not present' flag. This will only be set if the device
671 * was not present at the time of import.
673 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
674 &vd
->vdev_not_present
);
677 * Get the alignment requirement.
679 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
682 * Retrieve the vdev creation time.
684 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
688 * If we're a top-level vdev, try to load the allocation parameters.
690 if (parent
&& !parent
->vdev_parent
&&
691 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
692 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
694 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
696 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
698 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
700 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
703 ASSERT0(vd
->vdev_top_zap
);
706 if (parent
&& !parent
->vdev_parent
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
707 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
708 alloctype
== VDEV_ALLOC_ADD
||
709 alloctype
== VDEV_ALLOC_SPLIT
||
710 alloctype
== VDEV_ALLOC_ROOTPOOL
);
711 vd
->vdev_mg
= metaslab_group_create(islog
?
712 spa_log_class(spa
) : spa_normal_class(spa
), vd
,
713 spa
->spa_alloc_count
);
716 if (vd
->vdev_ops
->vdev_op_leaf
&&
717 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
718 (void) nvlist_lookup_uint64(nv
,
719 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
721 ASSERT0(vd
->vdev_leaf_zap
);
725 * If we're a leaf vdev, try to load the DTL object and other state.
728 if (vd
->vdev_ops
->vdev_op_leaf
&&
729 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
730 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
731 if (alloctype
== VDEV_ALLOC_LOAD
) {
732 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
733 &vd
->vdev_dtl_object
);
734 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
738 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
741 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
742 &spare
) == 0 && spare
)
746 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
749 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
750 &vd
->vdev_resilver_txg
);
753 * In general, when importing a pool we want to ignore the
754 * persistent fault state, as the diagnosis made on another
755 * system may not be valid in the current context. The only
756 * exception is if we forced a vdev to a persistently faulted
757 * state with 'zpool offline -f'. The persistent fault will
758 * remain across imports until cleared.
760 * Local vdevs will remain in the faulted state.
762 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
763 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
764 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
766 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
768 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
771 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
775 VDEV_AUX_ERR_EXCEEDED
;
776 if (nvlist_lookup_string(nv
,
777 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
778 strcmp(aux
, "external") == 0)
779 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
781 vd
->vdev_faulted
= 0ULL;
787 * Add ourselves to the parent's list of children.
789 vdev_add_child(parent
, vd
);
797 vdev_free(vdev_t
*vd
)
799 spa_t
*spa
= vd
->vdev_spa
;
802 * Scan queues are normally destroyed at the end of a scan. If the
803 * queue exists here, that implies the vdev is being removed while
804 * the scan is still running.
806 if (vd
->vdev_scan_io_queue
!= NULL
) {
807 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
808 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
809 vd
->vdev_scan_io_queue
= NULL
;
810 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
814 * vdev_free() implies closing the vdev first. This is simpler than
815 * trying to ensure complicated semantics for all callers.
819 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
820 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
825 for (int c
= 0; c
< vd
->vdev_children
; c
++)
826 vdev_free(vd
->vdev_child
[c
]);
828 ASSERT(vd
->vdev_child
== NULL
);
829 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
832 * Discard allocation state.
834 if (vd
->vdev_mg
!= NULL
) {
835 vdev_metaslab_fini(vd
);
836 metaslab_group_destroy(vd
->vdev_mg
);
839 ASSERT0(vd
->vdev_stat
.vs_space
);
840 ASSERT0(vd
->vdev_stat
.vs_dspace
);
841 ASSERT0(vd
->vdev_stat
.vs_alloc
);
844 * Remove this vdev from its parent's child list.
846 vdev_remove_child(vd
->vdev_parent
, vd
);
848 ASSERT(vd
->vdev_parent
== NULL
);
851 * Clean up vdev structure.
857 spa_strfree(vd
->vdev_path
);
859 spa_strfree(vd
->vdev_devid
);
860 if (vd
->vdev_physpath
)
861 spa_strfree(vd
->vdev_physpath
);
863 if (vd
->vdev_enc_sysfs_path
)
864 spa_strfree(vd
->vdev_enc_sysfs_path
);
867 spa_strfree(vd
->vdev_fru
);
869 if (vd
->vdev_isspare
)
870 spa_spare_remove(vd
);
871 if (vd
->vdev_isl2cache
)
872 spa_l2cache_remove(vd
);
874 txg_list_destroy(&vd
->vdev_ms_list
);
875 txg_list_destroy(&vd
->vdev_dtl_list
);
877 mutex_enter(&vd
->vdev_dtl_lock
);
878 space_map_close(vd
->vdev_dtl_sm
);
879 for (int t
= 0; t
< DTL_TYPES
; t
++) {
880 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
881 range_tree_destroy(vd
->vdev_dtl
[t
]);
883 mutex_exit(&vd
->vdev_dtl_lock
);
885 EQUIV(vd
->vdev_indirect_births
!= NULL
,
886 vd
->vdev_indirect_mapping
!= NULL
);
887 if (vd
->vdev_indirect_births
!= NULL
) {
888 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
889 vdev_indirect_births_close(vd
->vdev_indirect_births
);
892 if (vd
->vdev_obsolete_sm
!= NULL
) {
893 ASSERT(vd
->vdev_removing
||
894 vd
->vdev_ops
== &vdev_indirect_ops
);
895 space_map_close(vd
->vdev_obsolete_sm
);
896 vd
->vdev_obsolete_sm
= NULL
;
898 range_tree_destroy(vd
->vdev_obsolete_segments
);
899 rw_destroy(&vd
->vdev_indirect_rwlock
);
900 mutex_destroy(&vd
->vdev_obsolete_lock
);
902 mutex_destroy(&vd
->vdev_queue_lock
);
903 mutex_destroy(&vd
->vdev_dtl_lock
);
904 mutex_destroy(&vd
->vdev_stat_lock
);
905 mutex_destroy(&vd
->vdev_probe_lock
);
906 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
908 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
909 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
911 if (vd
== spa
->spa_root_vdev
)
912 spa
->spa_root_vdev
= NULL
;
914 kmem_free(vd
, sizeof (vdev_t
));
918 * Transfer top-level vdev state from svd to tvd.
921 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
923 spa_t
*spa
= svd
->vdev_spa
;
928 ASSERT(tvd
== tvd
->vdev_top
);
930 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
931 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
932 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
933 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
934 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
936 svd
->vdev_ms_array
= 0;
937 svd
->vdev_ms_shift
= 0;
938 svd
->vdev_ms_count
= 0;
939 svd
->vdev_top_zap
= 0;
942 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
943 tvd
->vdev_mg
= svd
->vdev_mg
;
944 tvd
->vdev_ms
= svd
->vdev_ms
;
949 if (tvd
->vdev_mg
!= NULL
)
950 tvd
->vdev_mg
->mg_vd
= tvd
;
952 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
953 svd
->vdev_checkpoint_sm
= NULL
;
955 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
956 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
957 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
959 svd
->vdev_stat
.vs_alloc
= 0;
960 svd
->vdev_stat
.vs_space
= 0;
961 svd
->vdev_stat
.vs_dspace
= 0;
964 * State which may be set on a top-level vdev that's in the
965 * process of being removed.
967 ASSERT0(tvd
->vdev_indirect_config
.vic_births_object
);
968 ASSERT0(tvd
->vdev_indirect_config
.vic_mapping_object
);
969 ASSERT3U(tvd
->vdev_indirect_config
.vic_prev_indirect_vdev
, ==, -1ULL);
970 ASSERT3P(tvd
->vdev_indirect_mapping
, ==, NULL
);
971 ASSERT3P(tvd
->vdev_indirect_births
, ==, NULL
);
972 ASSERT3P(tvd
->vdev_obsolete_sm
, ==, NULL
);
973 ASSERT0(tvd
->vdev_removing
);
974 tvd
->vdev_removing
= svd
->vdev_removing
;
975 tvd
->vdev_indirect_config
= svd
->vdev_indirect_config
;
976 tvd
->vdev_indirect_mapping
= svd
->vdev_indirect_mapping
;
977 tvd
->vdev_indirect_births
= svd
->vdev_indirect_births
;
978 range_tree_swap(&svd
->vdev_obsolete_segments
,
979 &tvd
->vdev_obsolete_segments
);
980 tvd
->vdev_obsolete_sm
= svd
->vdev_obsolete_sm
;
981 svd
->vdev_indirect_config
.vic_mapping_object
= 0;
982 svd
->vdev_indirect_config
.vic_births_object
= 0;
983 svd
->vdev_indirect_config
.vic_prev_indirect_vdev
= -1ULL;
984 svd
->vdev_indirect_mapping
= NULL
;
985 svd
->vdev_indirect_births
= NULL
;
986 svd
->vdev_obsolete_sm
= NULL
;
987 svd
->vdev_removing
= 0;
989 for (t
= 0; t
< TXG_SIZE
; t
++) {
990 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
991 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
992 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
993 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
994 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
995 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
998 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
999 vdev_config_clean(svd
);
1000 vdev_config_dirty(tvd
);
1003 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
1004 vdev_state_clean(svd
);
1005 vdev_state_dirty(tvd
);
1008 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
1009 svd
->vdev_deflate_ratio
= 0;
1011 tvd
->vdev_islog
= svd
->vdev_islog
;
1012 svd
->vdev_islog
= 0;
1014 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
1018 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
1025 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1026 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
1030 * Add a mirror/replacing vdev above an existing vdev.
1033 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
1035 spa_t
*spa
= cvd
->vdev_spa
;
1036 vdev_t
*pvd
= cvd
->vdev_parent
;
1039 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1041 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
1043 mvd
->vdev_asize
= cvd
->vdev_asize
;
1044 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
1045 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
1046 mvd
->vdev_psize
= cvd
->vdev_psize
;
1047 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
1048 mvd
->vdev_state
= cvd
->vdev_state
;
1049 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
1051 vdev_remove_child(pvd
, cvd
);
1052 vdev_add_child(pvd
, mvd
);
1053 cvd
->vdev_id
= mvd
->vdev_children
;
1054 vdev_add_child(mvd
, cvd
);
1055 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1057 if (mvd
== mvd
->vdev_top
)
1058 vdev_top_transfer(cvd
, mvd
);
1064 * Remove a 1-way mirror/replacing vdev from the tree.
1067 vdev_remove_parent(vdev_t
*cvd
)
1069 vdev_t
*mvd
= cvd
->vdev_parent
;
1070 vdev_t
*pvd
= mvd
->vdev_parent
;
1072 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1074 ASSERT(mvd
->vdev_children
== 1);
1075 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
1076 mvd
->vdev_ops
== &vdev_replacing_ops
||
1077 mvd
->vdev_ops
== &vdev_spare_ops
);
1078 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
1080 vdev_remove_child(mvd
, cvd
);
1081 vdev_remove_child(pvd
, mvd
);
1084 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1085 * Otherwise, we could have detached an offline device, and when we
1086 * go to import the pool we'll think we have two top-level vdevs,
1087 * instead of a different version of the same top-level vdev.
1089 if (mvd
->vdev_top
== mvd
) {
1090 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
1091 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
1092 cvd
->vdev_guid
+= guid_delta
;
1093 cvd
->vdev_guid_sum
+= guid_delta
;
1096 * If pool not set for autoexpand, we need to also preserve
1097 * mvd's asize to prevent automatic expansion of cvd.
1098 * Otherwise if we are adjusting the mirror by attaching and
1099 * detaching children of non-uniform sizes, the mirror could
1100 * autoexpand, unexpectedly requiring larger devices to
1101 * re-establish the mirror.
1103 if (!cvd
->vdev_spa
->spa_autoexpand
)
1104 cvd
->vdev_asize
= mvd
->vdev_asize
;
1106 cvd
->vdev_id
= mvd
->vdev_id
;
1107 vdev_add_child(pvd
, cvd
);
1108 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1110 if (cvd
== cvd
->vdev_top
)
1111 vdev_top_transfer(mvd
, cvd
);
1113 ASSERT(mvd
->vdev_children
== 0);
1118 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
1120 spa_t
*spa
= vd
->vdev_spa
;
1121 objset_t
*mos
= spa
->spa_meta_objset
;
1123 uint64_t oldc
= vd
->vdev_ms_count
;
1124 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
1128 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1131 * This vdev is not being allocated from yet or is a hole.
1133 if (vd
->vdev_ms_shift
== 0)
1136 ASSERT(!vd
->vdev_ishole
);
1138 ASSERT(oldc
<= newc
);
1140 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1143 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
1144 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1148 vd
->vdev_ms_count
= newc
;
1149 for (m
= oldc
; m
< newc
; m
++) {
1150 uint64_t object
= 0;
1153 * vdev_ms_array may be 0 if we are creating the "fake"
1154 * metaslabs for an indirect vdev for zdb's leak detection.
1155 * See zdb_leak_init().
1157 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1158 error
= dmu_read(mos
, vd
->vdev_ms_array
,
1159 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1162 vdev_dbgmsg(vd
, "unable to read the metaslab "
1163 "array [error=%d]", error
);
1168 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1171 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1178 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1181 * If the vdev is being removed we don't activate
1182 * the metaslabs since we want to ensure that no new
1183 * allocations are performed on this device.
1185 if (oldc
== 0 && !vd
->vdev_removing
)
1186 metaslab_group_activate(vd
->vdev_mg
);
1189 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1195 vdev_metaslab_fini(vdev_t
*vd
)
1197 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1198 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1199 SPA_FEATURE_POOL_CHECKPOINT
));
1200 space_map_close(vd
->vdev_checkpoint_sm
);
1202 * Even though we close the space map, we need to set its
1203 * pointer to NULL. The reason is that vdev_metaslab_fini()
1204 * may be called multiple times for certain operations
1205 * (i.e. when destroying a pool) so we need to ensure that
1206 * this clause never executes twice. This logic is similar
1207 * to the one used for the vdev_ms clause below.
1209 vd
->vdev_checkpoint_sm
= NULL
;
1212 if (vd
->vdev_ms
!= NULL
) {
1213 uint64_t count
= vd
->vdev_ms_count
;
1215 metaslab_group_passivate(vd
->vdev_mg
);
1216 for (uint64_t m
= 0; m
< count
; m
++) {
1217 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1222 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1225 vd
->vdev_ms_count
= 0;
1227 ASSERT0(vd
->vdev_ms_count
);
1228 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1231 typedef struct vdev_probe_stats
{
1232 boolean_t vps_readable
;
1233 boolean_t vps_writeable
;
1235 } vdev_probe_stats_t
;
1238 vdev_probe_done(zio_t
*zio
)
1240 spa_t
*spa
= zio
->io_spa
;
1241 vdev_t
*vd
= zio
->io_vd
;
1242 vdev_probe_stats_t
*vps
= zio
->io_private
;
1244 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1246 if (zio
->io_type
== ZIO_TYPE_READ
) {
1247 if (zio
->io_error
== 0)
1248 vps
->vps_readable
= 1;
1249 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1250 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1251 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1252 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1253 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1255 abd_free(zio
->io_abd
);
1257 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1258 if (zio
->io_error
== 0)
1259 vps
->vps_writeable
= 1;
1260 abd_free(zio
->io_abd
);
1261 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1265 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1266 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1268 if (vdev_readable(vd
) &&
1269 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1272 ASSERT(zio
->io_error
!= 0);
1273 vdev_dbgmsg(vd
, "failed probe");
1274 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1275 spa
, vd
, NULL
, NULL
, 0, 0);
1276 zio
->io_error
= SET_ERROR(ENXIO
);
1279 mutex_enter(&vd
->vdev_probe_lock
);
1280 ASSERT(vd
->vdev_probe_zio
== zio
);
1281 vd
->vdev_probe_zio
= NULL
;
1282 mutex_exit(&vd
->vdev_probe_lock
);
1285 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1286 if (!vdev_accessible(vd
, pio
))
1287 pio
->io_error
= SET_ERROR(ENXIO
);
1289 kmem_free(vps
, sizeof (*vps
));
1294 * Determine whether this device is accessible.
1296 * Read and write to several known locations: the pad regions of each
1297 * vdev label but the first, which we leave alone in case it contains
1301 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1303 spa_t
*spa
= vd
->vdev_spa
;
1304 vdev_probe_stats_t
*vps
= NULL
;
1307 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1310 * Don't probe the probe.
1312 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1316 * To prevent 'probe storms' when a device fails, we create
1317 * just one probe i/o at a time. All zios that want to probe
1318 * this vdev will become parents of the probe io.
1320 mutex_enter(&vd
->vdev_probe_lock
);
1322 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1323 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1325 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1326 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1329 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1331 * vdev_cant_read and vdev_cant_write can only
1332 * transition from TRUE to FALSE when we have the
1333 * SCL_ZIO lock as writer; otherwise they can only
1334 * transition from FALSE to TRUE. This ensures that
1335 * any zio looking at these values can assume that
1336 * failures persist for the life of the I/O. That's
1337 * important because when a device has intermittent
1338 * connectivity problems, we want to ensure that
1339 * they're ascribed to the device (ENXIO) and not
1342 * Since we hold SCL_ZIO as writer here, clear both
1343 * values so the probe can reevaluate from first
1346 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1347 vd
->vdev_cant_read
= B_FALSE
;
1348 vd
->vdev_cant_write
= B_FALSE
;
1351 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1352 vdev_probe_done
, vps
,
1353 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1356 * We can't change the vdev state in this context, so we
1357 * kick off an async task to do it on our behalf.
1360 vd
->vdev_probe_wanted
= B_TRUE
;
1361 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1366 zio_add_child(zio
, pio
);
1368 mutex_exit(&vd
->vdev_probe_lock
);
1371 ASSERT(zio
!= NULL
);
1375 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1376 zio_nowait(zio_read_phys(pio
, vd
,
1377 vdev_label_offset(vd
->vdev_psize
, l
,
1378 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1379 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1380 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1381 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1392 vdev_open_child(void *arg
)
1396 vd
->vdev_open_thread
= curthread
;
1397 vd
->vdev_open_error
= vdev_open(vd
);
1398 vd
->vdev_open_thread
= NULL
;
1402 vdev_uses_zvols(vdev_t
*vd
)
1405 if (zvol_is_zvol(vd
->vdev_path
))
1409 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1410 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1417 vdev_open_children(vdev_t
*vd
)
1420 int children
= vd
->vdev_children
;
1423 * in order to handle pools on top of zvols, do the opens
1424 * in a single thread so that the same thread holds the
1425 * spa_namespace_lock
1427 if (vdev_uses_zvols(vd
)) {
1429 for (int c
= 0; c
< children
; c
++)
1430 vd
->vdev_child
[c
]->vdev_open_error
=
1431 vdev_open(vd
->vdev_child
[c
]);
1433 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1434 children
, children
, TASKQ_PREPOPULATE
);
1438 for (int c
= 0; c
< children
; c
++)
1439 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1440 vd
->vdev_child
[c
], TQ_SLEEP
) != TASKQID_INVALID
);
1445 vd
->vdev_nonrot
= B_TRUE
;
1447 for (int c
= 0; c
< children
; c
++)
1448 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1452 * Compute the raidz-deflation ratio. Note, we hard-code
1453 * in 128k (1 << 17) because it is the "typical" blocksize.
1454 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1455 * otherwise it would inconsistently account for existing bp's.
1458 vdev_set_deflate_ratio(vdev_t
*vd
)
1460 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1461 vd
->vdev_deflate_ratio
= (1 << 17) /
1462 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1467 * Prepare a virtual device for access.
1470 vdev_open(vdev_t
*vd
)
1472 spa_t
*spa
= vd
->vdev_spa
;
1475 uint64_t max_osize
= 0;
1476 uint64_t asize
, max_asize
, psize
;
1477 uint64_t ashift
= 0;
1479 ASSERT(vd
->vdev_open_thread
== curthread
||
1480 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1481 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1482 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1483 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1485 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1486 vd
->vdev_cant_read
= B_FALSE
;
1487 vd
->vdev_cant_write
= B_FALSE
;
1488 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1491 * If this vdev is not removed, check its fault status. If it's
1492 * faulted, bail out of the open.
1494 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1495 ASSERT(vd
->vdev_children
== 0);
1496 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1497 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1498 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1499 vd
->vdev_label_aux
);
1500 return (SET_ERROR(ENXIO
));
1501 } else if (vd
->vdev_offline
) {
1502 ASSERT(vd
->vdev_children
== 0);
1503 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1504 return (SET_ERROR(ENXIO
));
1507 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1510 * Reset the vdev_reopening flag so that we actually close
1511 * the vdev on error.
1513 vd
->vdev_reopening
= B_FALSE
;
1514 if (zio_injection_enabled
&& error
== 0)
1515 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1518 if (vd
->vdev_removed
&&
1519 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1520 vd
->vdev_removed
= B_FALSE
;
1522 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
1523 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
1524 vd
->vdev_stat
.vs_aux
);
1526 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1527 vd
->vdev_stat
.vs_aux
);
1532 vd
->vdev_removed
= B_FALSE
;
1535 * Recheck the faulted flag now that we have confirmed that
1536 * the vdev is accessible. If we're faulted, bail.
1538 if (vd
->vdev_faulted
) {
1539 ASSERT(vd
->vdev_children
== 0);
1540 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1541 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1542 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1543 vd
->vdev_label_aux
);
1544 return (SET_ERROR(ENXIO
));
1547 if (vd
->vdev_degraded
) {
1548 ASSERT(vd
->vdev_children
== 0);
1549 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1550 VDEV_AUX_ERR_EXCEEDED
);
1552 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1556 * For hole or missing vdevs we just return success.
1558 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1561 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1562 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1563 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1569 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1570 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1572 if (vd
->vdev_children
== 0) {
1573 if (osize
< SPA_MINDEVSIZE
) {
1574 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1575 VDEV_AUX_TOO_SMALL
);
1576 return (SET_ERROR(EOVERFLOW
));
1579 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1580 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1581 VDEV_LABEL_END_SIZE
);
1583 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1584 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1585 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1586 VDEV_AUX_TOO_SMALL
);
1587 return (SET_ERROR(EOVERFLOW
));
1591 max_asize
= max_osize
;
1595 * If the vdev was expanded, record this so that we can re-create the
1596 * uberblock rings in labels {2,3}, during the next sync.
1598 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
1599 vd
->vdev_copy_uberblocks
= B_TRUE
;
1601 vd
->vdev_psize
= psize
;
1604 * Make sure the allocatable size hasn't shrunk too much.
1606 if (asize
< vd
->vdev_min_asize
) {
1607 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1608 VDEV_AUX_BAD_LABEL
);
1609 return (SET_ERROR(EINVAL
));
1612 if (vd
->vdev_asize
== 0) {
1614 * This is the first-ever open, so use the computed values.
1615 * For compatibility, a different ashift can be requested.
1617 vd
->vdev_asize
= asize
;
1618 vd
->vdev_max_asize
= max_asize
;
1619 if (vd
->vdev_ashift
== 0) {
1620 vd
->vdev_ashift
= ashift
; /* use detected value */
1622 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1623 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1624 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1625 VDEV_AUX_BAD_ASHIFT
);
1626 return (SET_ERROR(EDOM
));
1630 * Detect if the alignment requirement has increased.
1631 * We don't want to make the pool unavailable, just
1632 * post an event instead.
1634 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1635 vd
->vdev_ops
->vdev_op_leaf
) {
1636 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1637 spa
, vd
, NULL
, NULL
, 0, 0);
1640 vd
->vdev_max_asize
= max_asize
;
1644 * If all children are healthy we update asize if either:
1645 * The asize has increased, due to a device expansion caused by dynamic
1646 * LUN growth or vdev replacement, and automatic expansion is enabled;
1647 * making the additional space available.
1649 * The asize has decreased, due to a device shrink usually caused by a
1650 * vdev replace with a smaller device. This ensures that calculations
1651 * based of max_asize and asize e.g. esize are always valid. It's safe
1652 * to do this as we've already validated that asize is greater than
1655 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1656 ((asize
> vd
->vdev_asize
&&
1657 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1658 (asize
< vd
->vdev_asize
)))
1659 vd
->vdev_asize
= asize
;
1661 vdev_set_min_asize(vd
);
1664 * Ensure we can issue some IO before declaring the
1665 * vdev open for business.
1667 if (vd
->vdev_ops
->vdev_op_leaf
&&
1668 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1669 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1670 VDEV_AUX_ERR_EXCEEDED
);
1675 * Track the min and max ashift values for normal data devices.
1677 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1678 !vd
->vdev_islog
&& vd
->vdev_aux
== NULL
) {
1679 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1680 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1681 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1682 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1686 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1687 * resilver. But don't do this if we are doing a reopen for a scrub,
1688 * since this would just restart the scrub we are already doing.
1690 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1691 vdev_resilver_needed(vd
, NULL
, NULL
))
1692 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1698 * Called once the vdevs are all opened, this routine validates the label
1699 * contents. This needs to be done before vdev_load() so that we don't
1700 * inadvertently do repair I/Os to the wrong device.
1702 * This function will only return failure if one of the vdevs indicates that it
1703 * has since been destroyed or exported. This is only possible if
1704 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1705 * will be updated but the function will return 0.
1708 vdev_validate(vdev_t
*vd
)
1710 spa_t
*spa
= vd
->vdev_spa
;
1712 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
1717 if (vdev_validate_skip
)
1720 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1721 if (vdev_validate(vd
->vdev_child
[c
]) != 0)
1722 return (SET_ERROR(EBADF
));
1725 * If the device has already failed, or was marked offline, don't do
1726 * any further validation. Otherwise, label I/O will fail and we will
1727 * overwrite the previous state.
1729 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
1733 * If we are performing an extreme rewind, we allow for a label that
1734 * was modified at a point after the current txg.
1735 * If config lock is not held do not check for the txg. spa_sync could
1736 * be updating the vdev's label before updating spa_last_synced_txg.
1738 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
1739 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
1742 txg
= spa_last_synced_txg(spa
);
1744 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1745 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1746 VDEV_AUX_BAD_LABEL
);
1747 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
1748 "txg %llu", (u_longlong_t
)txg
);
1753 * Determine if this vdev has been split off into another
1754 * pool. If so, then refuse to open it.
1756 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1757 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1758 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1759 VDEV_AUX_SPLIT_POOL
);
1761 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
1765 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
1766 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1767 VDEV_AUX_CORRUPT_DATA
);
1769 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1770 ZPOOL_CONFIG_POOL_GUID
);
1775 * If config is not trusted then ignore the spa guid check. This is
1776 * necessary because if the machine crashed during a re-guid the new
1777 * guid might have been written to all of the vdev labels, but not the
1778 * cached config. The check will be performed again once we have the
1779 * trusted config from the MOS.
1781 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
1782 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1783 VDEV_AUX_CORRUPT_DATA
);
1785 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
1786 "match config (%llu != %llu)", (u_longlong_t
)guid
,
1787 (u_longlong_t
)spa_guid(spa
));
1791 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1792 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1796 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
1797 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1798 VDEV_AUX_CORRUPT_DATA
);
1800 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1805 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
1807 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1808 VDEV_AUX_CORRUPT_DATA
);
1810 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1811 ZPOOL_CONFIG_TOP_GUID
);
1816 * If this vdev just became a top-level vdev because its sibling was
1817 * detached, it will have adopted the parent's vdev guid -- but the
1818 * label may or may not be on disk yet. Fortunately, either version
1819 * of the label will have the same top guid, so if we're a top-level
1820 * vdev, we can safely compare to that instead.
1821 * However, if the config comes from a cachefile that failed to update
1822 * after the detach, a top-level vdev will appear as a non top-level
1823 * vdev in the config. Also relax the constraints if we perform an
1826 * If we split this vdev off instead, then we also check the
1827 * original pool's guid. We don't want to consider the vdev
1828 * corrupt if it is partway through a split operation.
1830 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
1831 boolean_t mismatch
= B_FALSE
;
1832 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
1833 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
1836 if (vd
->vdev_guid
!= top_guid
&&
1837 vd
->vdev_top
->vdev_guid
!= guid
)
1842 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1843 VDEV_AUX_CORRUPT_DATA
);
1845 vdev_dbgmsg(vd
, "vdev_validate: config guid "
1846 "doesn't match label guid");
1847 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
1848 (u_longlong_t
)vd
->vdev_guid
,
1849 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
1850 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
1851 "aux_guid %llu", (u_longlong_t
)guid
,
1852 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
1857 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1859 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1860 VDEV_AUX_CORRUPT_DATA
);
1862 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1863 ZPOOL_CONFIG_POOL_STATE
);
1870 * If this is a verbatim import, no need to check the
1871 * state of the pool.
1873 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1874 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1875 state
!= POOL_STATE_ACTIVE
) {
1876 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
1877 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
1878 return (SET_ERROR(EBADF
));
1882 * If we were able to open and validate a vdev that was
1883 * previously marked permanently unavailable, clear that state
1886 if (vd
->vdev_not_present
)
1887 vd
->vdev_not_present
= 0;
1893 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
1895 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
1896 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
1897 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
1898 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
1899 dvd
->vdev_path
, svd
->vdev_path
);
1900 spa_strfree(dvd
->vdev_path
);
1901 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
1903 } else if (svd
->vdev_path
!= NULL
) {
1904 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
1905 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
1906 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
1911 * Recursively copy vdev paths from one vdev to another. Source and destination
1912 * vdev trees must have same geometry otherwise return error. Intended to copy
1913 * paths from userland config into MOS config.
1916 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
1918 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
1919 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
1920 (dvd
->vdev_ops
== &vdev_indirect_ops
))
1923 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
1924 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
1925 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
1926 return (SET_ERROR(EINVAL
));
1929 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
1930 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
1931 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
1932 (u_longlong_t
)dvd
->vdev_guid
);
1933 return (SET_ERROR(EINVAL
));
1936 if (svd
->vdev_children
!= dvd
->vdev_children
) {
1937 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
1938 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
1939 (u_longlong_t
)dvd
->vdev_children
);
1940 return (SET_ERROR(EINVAL
));
1943 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
1944 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
1945 dvd
->vdev_child
[i
]);
1950 if (svd
->vdev_ops
->vdev_op_leaf
)
1951 vdev_copy_path_impl(svd
, dvd
);
1957 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
1959 ASSERT(stvd
->vdev_top
== stvd
);
1960 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
1962 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
1963 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
1966 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
1970 * The idea here is that while a vdev can shift positions within
1971 * a top vdev (when replacing, attaching mirror, etc.) it cannot
1972 * step outside of it.
1974 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
1976 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
1979 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1981 vdev_copy_path_impl(vd
, dvd
);
1985 * Recursively copy vdev paths from one root vdev to another. Source and
1986 * destination vdev trees may differ in geometry. For each destination leaf
1987 * vdev, search a vdev with the same guid and top vdev id in the source.
1988 * Intended to copy paths from userland config into MOS config.
1991 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
1993 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
1994 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
1995 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
1997 for (uint64_t i
= 0; i
< children
; i
++) {
1998 vdev_copy_path_search(srvd
->vdev_child
[i
],
1999 drvd
->vdev_child
[i
]);
2004 * Close a virtual device.
2007 vdev_close(vdev_t
*vd
)
2009 vdev_t
*pvd
= vd
->vdev_parent
;
2010 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
2012 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2015 * If our parent is reopening, then we are as well, unless we are
2018 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
2019 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
2021 vd
->vdev_ops
->vdev_op_close(vd
);
2023 vdev_cache_purge(vd
);
2026 * We record the previous state before we close it, so that if we are
2027 * doing a reopen(), we don't generate FMA ereports if we notice that
2028 * it's still faulted.
2030 vd
->vdev_prevstate
= vd
->vdev_state
;
2032 if (vd
->vdev_offline
)
2033 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
2035 vd
->vdev_state
= VDEV_STATE_CLOSED
;
2036 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2040 vdev_hold(vdev_t
*vd
)
2042 spa_t
*spa
= vd
->vdev_spa
;
2044 ASSERT(spa_is_root(spa
));
2045 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
2048 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2049 vdev_hold(vd
->vdev_child
[c
]);
2051 if (vd
->vdev_ops
->vdev_op_leaf
)
2052 vd
->vdev_ops
->vdev_op_hold(vd
);
2056 vdev_rele(vdev_t
*vd
)
2058 ASSERT(spa_is_root(vd
->vdev_spa
));
2059 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2060 vdev_rele(vd
->vdev_child
[c
]);
2062 if (vd
->vdev_ops
->vdev_op_leaf
)
2063 vd
->vdev_ops
->vdev_op_rele(vd
);
2067 * Reopen all interior vdevs and any unopened leaves. We don't actually
2068 * reopen leaf vdevs which had previously been opened as they might deadlock
2069 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2070 * If the leaf has never been opened then open it, as usual.
2073 vdev_reopen(vdev_t
*vd
)
2075 spa_t
*spa
= vd
->vdev_spa
;
2077 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2079 /* set the reopening flag unless we're taking the vdev offline */
2080 vd
->vdev_reopening
= !vd
->vdev_offline
;
2082 (void) vdev_open(vd
);
2085 * Call vdev_validate() here to make sure we have the same device.
2086 * Otherwise, a device with an invalid label could be successfully
2087 * opened in response to vdev_reopen().
2090 (void) vdev_validate_aux(vd
);
2091 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
2092 vd
->vdev_aux
== &spa
->spa_l2cache
&&
2093 !l2arc_vdev_present(vd
))
2094 l2arc_add_vdev(spa
, vd
);
2096 (void) vdev_validate(vd
);
2100 * Reassess parent vdev's health.
2102 vdev_propagate_state(vd
);
2106 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
2111 * Normally, partial opens (e.g. of a mirror) are allowed.
2112 * For a create, however, we want to fail the request if
2113 * there are any components we can't open.
2115 error
= vdev_open(vd
);
2117 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2119 return (error
? error
: ENXIO
);
2123 * Recursively load DTLs and initialize all labels.
2125 if ((error
= vdev_dtl_load(vd
)) != 0 ||
2126 (error
= vdev_label_init(vd
, txg
, isreplacing
?
2127 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
2136 vdev_metaslab_set_size(vdev_t
*vd
)
2138 uint64_t asize
= vd
->vdev_asize
;
2139 uint64_t ms_count
= asize
>> vdev_default_ms_shift
;
2143 * There are two dimensions to the metaslab sizing calculation:
2144 * the size of the metaslab and the count of metaslabs per vdev.
2145 * In general, we aim for vdev_max_ms_count (200) metaslabs. The
2146 * range of the dimensions are as follows:
2148 * 2^29 <= ms_size <= 2^38
2149 * 16 <= ms_count <= 131,072
2151 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2152 * at least 512MB (2^29) to minimize fragmentation effects when
2153 * testing with smaller devices. However, the count constraint
2154 * of at least 16 metaslabs will override this minimum size goal.
2156 * On the upper end of vdev sizes, we aim for a maximum metaslab
2157 * size of 256GB. However, we will cap the total count to 2^17
2158 * metaslabs to keep our memory footprint in check.
2160 * The net effect of applying above constrains is summarized below.
2162 * vdev size metaslab count
2163 * -------------|-----------------
2165 * 8GB - 100GB one per 512MB
2167 * 50TB - 32PB one per 256GB
2169 * -------------------------------
2172 if (ms_count
< vdev_min_ms_count
)
2173 ms_shift
= highbit64(asize
/ vdev_min_ms_count
);
2174 else if (ms_count
> vdev_max_ms_count
)
2175 ms_shift
= highbit64(asize
/ vdev_max_ms_count
);
2177 ms_shift
= vdev_default_ms_shift
;
2179 if (ms_shift
< SPA_MAXBLOCKSHIFT
) {
2180 ms_shift
= SPA_MAXBLOCKSHIFT
;
2181 } else if (ms_shift
> vdev_max_ms_shift
) {
2182 ms_shift
= vdev_max_ms_shift
;
2183 /* cap the total count to constrain memory footprint */
2184 if ((asize
>> ms_shift
) > vdev_ms_count_limit
)
2185 ms_shift
= highbit64(asize
/ vdev_ms_count_limit
);
2188 vd
->vdev_ms_shift
= ms_shift
;
2189 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2193 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2195 ASSERT(vd
== vd
->vdev_top
);
2196 /* indirect vdevs don't have metaslabs or dtls */
2197 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2198 ASSERT(ISP2(flags
));
2199 ASSERT(spa_writeable(vd
->vdev_spa
));
2201 if (flags
& VDD_METASLAB
)
2202 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2204 if (flags
& VDD_DTL
)
2205 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2207 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2211 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2213 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2214 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2216 if (vd
->vdev_ops
->vdev_op_leaf
)
2217 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2223 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2224 * the vdev has less than perfect replication. There are four kinds of DTL:
2226 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2228 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2230 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2231 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2232 * txgs that was scrubbed.
2234 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2235 * persistent errors or just some device being offline.
2236 * Unlike the other three, the DTL_OUTAGE map is not generally
2237 * maintained; it's only computed when needed, typically to
2238 * determine whether a device can be detached.
2240 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2241 * either has the data or it doesn't.
2243 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2244 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2245 * if any child is less than fully replicated, then so is its parent.
2246 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2247 * comprising only those txgs which appear in 'maxfaults' or more children;
2248 * those are the txgs we don't have enough replication to read. For example,
2249 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2250 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2251 * two child DTL_MISSING maps.
2253 * It should be clear from the above that to compute the DTLs and outage maps
2254 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2255 * Therefore, that is all we keep on disk. When loading the pool, or after
2256 * a configuration change, we generate all other DTLs from first principles.
2259 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2261 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2263 ASSERT(t
< DTL_TYPES
);
2264 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2265 ASSERT(spa_writeable(vd
->vdev_spa
));
2267 mutex_enter(&vd
->vdev_dtl_lock
);
2268 if (!range_tree_contains(rt
, txg
, size
))
2269 range_tree_add(rt
, txg
, size
);
2270 mutex_exit(&vd
->vdev_dtl_lock
);
2274 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2276 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2277 boolean_t dirty
= B_FALSE
;
2279 ASSERT(t
< DTL_TYPES
);
2280 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2283 * While we are loading the pool, the DTLs have not been loaded yet.
2284 * Ignore the DTLs and try all devices. This avoids a recursive
2285 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2286 * when loading the pool (relying on the checksum to ensure that
2287 * we get the right data -- note that we while loading, we are
2288 * only reading the MOS, which is always checksummed).
2290 if (vd
->vdev_spa
->spa_load_state
!= SPA_LOAD_NONE
)
2293 mutex_enter(&vd
->vdev_dtl_lock
);
2294 if (!range_tree_is_empty(rt
))
2295 dirty
= range_tree_contains(rt
, txg
, size
);
2296 mutex_exit(&vd
->vdev_dtl_lock
);
2302 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2304 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2307 mutex_enter(&vd
->vdev_dtl_lock
);
2308 empty
= range_tree_is_empty(rt
);
2309 mutex_exit(&vd
->vdev_dtl_lock
);
2315 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2318 vdev_dtl_need_resilver(vdev_t
*vd
, uint64_t offset
, size_t psize
)
2320 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2322 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
2323 vd
->vdev_ops
->vdev_op_leaf
)
2326 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, offset
, psize
));
2330 * Returns the lowest txg in the DTL range.
2333 vdev_dtl_min(vdev_t
*vd
)
2337 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2338 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2339 ASSERT0(vd
->vdev_children
);
2341 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2342 return (rs
->rs_start
- 1);
2346 * Returns the highest txg in the DTL.
2349 vdev_dtl_max(vdev_t
*vd
)
2353 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2354 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2355 ASSERT0(vd
->vdev_children
);
2357 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2358 return (rs
->rs_end
);
2362 * Determine if a resilvering vdev should remove any DTL entries from
2363 * its range. If the vdev was resilvering for the entire duration of the
2364 * scan then it should excise that range from its DTLs. Otherwise, this
2365 * vdev is considered partially resilvered and should leave its DTL
2366 * entries intact. The comment in vdev_dtl_reassess() describes how we
2370 vdev_dtl_should_excise(vdev_t
*vd
)
2372 spa_t
*spa
= vd
->vdev_spa
;
2373 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2375 ASSERT0(scn
->scn_phys
.scn_errors
);
2376 ASSERT0(vd
->vdev_children
);
2378 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
2381 if (vd
->vdev_resilver_txg
== 0 ||
2382 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
2386 * When a resilver is initiated the scan will assign the scn_max_txg
2387 * value to the highest txg value that exists in all DTLs. If this
2388 * device's max DTL is not part of this scan (i.e. it is not in
2389 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2392 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
2393 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
2394 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
2395 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
2402 * Reassess DTLs after a config change or scrub completion.
2405 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
2407 spa_t
*spa
= vd
->vdev_spa
;
2411 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2413 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2414 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
2415 scrub_txg
, scrub_done
);
2417 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
2420 if (vd
->vdev_ops
->vdev_op_leaf
) {
2421 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2423 mutex_enter(&vd
->vdev_dtl_lock
);
2426 * If requested, pretend the scan completed cleanly.
2428 if (zfs_scan_ignore_errors
&& scn
)
2429 scn
->scn_phys
.scn_errors
= 0;
2432 * If we've completed a scan cleanly then determine
2433 * if this vdev should remove any DTLs. We only want to
2434 * excise regions on vdevs that were available during
2435 * the entire duration of this scan.
2437 if (scrub_txg
!= 0 &&
2438 (spa
->spa_scrub_started
||
2439 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
2440 vdev_dtl_should_excise(vd
)) {
2442 * We completed a scrub up to scrub_txg. If we
2443 * did it without rebooting, then the scrub dtl
2444 * will be valid, so excise the old region and
2445 * fold in the scrub dtl. Otherwise, leave the
2446 * dtl as-is if there was an error.
2448 * There's little trick here: to excise the beginning
2449 * of the DTL_MISSING map, we put it into a reference
2450 * tree and then add a segment with refcnt -1 that
2451 * covers the range [0, scrub_txg). This means
2452 * that each txg in that range has refcnt -1 or 0.
2453 * We then add DTL_SCRUB with a refcnt of 2, so that
2454 * entries in the range [0, scrub_txg) will have a
2455 * positive refcnt -- either 1 or 2. We then convert
2456 * the reference tree into the new DTL_MISSING map.
2458 space_reftree_create(&reftree
);
2459 space_reftree_add_map(&reftree
,
2460 vd
->vdev_dtl
[DTL_MISSING
], 1);
2461 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
2462 space_reftree_add_map(&reftree
,
2463 vd
->vdev_dtl
[DTL_SCRUB
], 2);
2464 space_reftree_generate_map(&reftree
,
2465 vd
->vdev_dtl
[DTL_MISSING
], 1);
2466 space_reftree_destroy(&reftree
);
2468 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
2469 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2470 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
2472 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
2473 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
2474 if (!vdev_readable(vd
))
2475 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
2477 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2478 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
2481 * If the vdev was resilvering and no longer has any
2482 * DTLs then reset its resilvering flag and dirty
2483 * the top level so that we persist the change.
2485 if (vd
->vdev_resilver_txg
!= 0 &&
2486 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2487 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
])) {
2488 vd
->vdev_resilver_txg
= 0;
2489 vdev_config_dirty(vd
->vdev_top
);
2492 mutex_exit(&vd
->vdev_dtl_lock
);
2495 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
2499 mutex_enter(&vd
->vdev_dtl_lock
);
2500 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2501 /* account for child's outage in parent's missing map */
2502 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
2504 continue; /* leaf vdevs only */
2505 if (t
== DTL_PARTIAL
)
2506 minref
= 1; /* i.e. non-zero */
2507 else if (vd
->vdev_nparity
!= 0)
2508 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
2510 minref
= vd
->vdev_children
; /* any kind of mirror */
2511 space_reftree_create(&reftree
);
2512 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2513 vdev_t
*cvd
= vd
->vdev_child
[c
];
2514 mutex_enter(&cvd
->vdev_dtl_lock
);
2515 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
2516 mutex_exit(&cvd
->vdev_dtl_lock
);
2518 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
2519 space_reftree_destroy(&reftree
);
2521 mutex_exit(&vd
->vdev_dtl_lock
);
2525 vdev_dtl_load(vdev_t
*vd
)
2527 spa_t
*spa
= vd
->vdev_spa
;
2528 objset_t
*mos
= spa
->spa_meta_objset
;
2531 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2532 ASSERT(vdev_is_concrete(vd
));
2534 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2535 vd
->vdev_dtl_object
, 0, -1ULL, 0);
2538 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2540 mutex_enter(&vd
->vdev_dtl_lock
);
2543 * Now that we've opened the space_map we need to update
2546 space_map_update(vd
->vdev_dtl_sm
);
2548 error
= space_map_load(vd
->vdev_dtl_sm
,
2549 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2550 mutex_exit(&vd
->vdev_dtl_lock
);
2555 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2556 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2565 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2567 spa_t
*spa
= vd
->vdev_spa
;
2569 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2570 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2575 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2577 spa_t
*spa
= vd
->vdev_spa
;
2578 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2579 DMU_OT_NONE
, 0, tx
);
2582 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2589 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2591 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2592 vd
->vdev_ops
!= &vdev_missing_ops
&&
2593 vd
->vdev_ops
!= &vdev_root_ops
&&
2594 !vd
->vdev_top
->vdev_removing
) {
2595 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2596 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2598 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2599 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2602 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2603 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2608 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2610 spa_t
*spa
= vd
->vdev_spa
;
2611 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2612 objset_t
*mos
= spa
->spa_meta_objset
;
2613 range_tree_t
*rtsync
;
2615 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2617 ASSERT(vdev_is_concrete(vd
));
2618 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2620 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2622 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2623 mutex_enter(&vd
->vdev_dtl_lock
);
2624 space_map_free(vd
->vdev_dtl_sm
, tx
);
2625 space_map_close(vd
->vdev_dtl_sm
);
2626 vd
->vdev_dtl_sm
= NULL
;
2627 mutex_exit(&vd
->vdev_dtl_lock
);
2630 * We only destroy the leaf ZAP for detached leaves or for
2631 * removed log devices. Removed data devices handle leaf ZAP
2632 * cleanup later, once cancellation is no longer possible.
2634 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2635 vd
->vdev_top
->vdev_islog
)) {
2636 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2637 vd
->vdev_leaf_zap
= 0;
2644 if (vd
->vdev_dtl_sm
== NULL
) {
2645 uint64_t new_object
;
2647 new_object
= space_map_alloc(mos
, vdev_dtl_sm_blksz
, tx
);
2648 VERIFY3U(new_object
, !=, 0);
2650 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2652 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2655 rtsync
= range_tree_create(NULL
, NULL
);
2657 mutex_enter(&vd
->vdev_dtl_lock
);
2658 range_tree_walk(rt
, range_tree_add
, rtsync
);
2659 mutex_exit(&vd
->vdev_dtl_lock
);
2661 space_map_truncate(vd
->vdev_dtl_sm
, vdev_dtl_sm_blksz
, tx
);
2662 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
2663 range_tree_vacate(rtsync
, NULL
, NULL
);
2665 range_tree_destroy(rtsync
);
2668 * If the object for the space map has changed then dirty
2669 * the top level so that we update the config.
2671 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2672 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
2673 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
2674 (u_longlong_t
)object
,
2675 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
2676 vdev_config_dirty(vd
->vdev_top
);
2681 mutex_enter(&vd
->vdev_dtl_lock
);
2682 space_map_update(vd
->vdev_dtl_sm
);
2683 mutex_exit(&vd
->vdev_dtl_lock
);
2687 * Determine whether the specified vdev can be offlined/detached/removed
2688 * without losing data.
2691 vdev_dtl_required(vdev_t
*vd
)
2693 spa_t
*spa
= vd
->vdev_spa
;
2694 vdev_t
*tvd
= vd
->vdev_top
;
2695 uint8_t cant_read
= vd
->vdev_cant_read
;
2698 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2700 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2704 * Temporarily mark the device as unreadable, and then determine
2705 * whether this results in any DTL outages in the top-level vdev.
2706 * If not, we can safely offline/detach/remove the device.
2708 vd
->vdev_cant_read
= B_TRUE
;
2709 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2710 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2711 vd
->vdev_cant_read
= cant_read
;
2712 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2714 if (!required
&& zio_injection_enabled
)
2715 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2721 * Determine if resilver is needed, and if so the txg range.
2724 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2726 boolean_t needed
= B_FALSE
;
2727 uint64_t thismin
= UINT64_MAX
;
2728 uint64_t thismax
= 0;
2730 if (vd
->vdev_children
== 0) {
2731 mutex_enter(&vd
->vdev_dtl_lock
);
2732 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2733 vdev_writeable(vd
)) {
2735 thismin
= vdev_dtl_min(vd
);
2736 thismax
= vdev_dtl_max(vd
);
2739 mutex_exit(&vd
->vdev_dtl_lock
);
2741 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2742 vdev_t
*cvd
= vd
->vdev_child
[c
];
2743 uint64_t cmin
, cmax
;
2745 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2746 thismin
= MIN(thismin
, cmin
);
2747 thismax
= MAX(thismax
, cmax
);
2753 if (needed
&& minp
) {
2761 * Gets the checkpoint space map object from the vdev's ZAP.
2762 * Returns the spacemap object, or 0 if it wasn't in the ZAP
2763 * or the ZAP doesn't exist yet.
2766 vdev_checkpoint_sm_object(vdev_t
*vd
)
2768 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
2769 if (vd
->vdev_top_zap
== 0) {
2773 uint64_t sm_obj
= 0;
2774 int err
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
2775 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, &sm_obj
);
2777 VERIFY(err
== 0 || err
== ENOENT
);
2783 vdev_load(vdev_t
*vd
)
2788 * Recursively load all children.
2790 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2791 error
= vdev_load(vd
->vdev_child
[c
]);
2797 vdev_set_deflate_ratio(vd
);
2800 * If this is a top-level vdev, initialize its metaslabs.
2802 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
2803 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
2804 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2805 VDEV_AUX_CORRUPT_DATA
);
2806 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
2807 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
2808 (u_longlong_t
)vd
->vdev_asize
);
2809 return (SET_ERROR(ENXIO
));
2810 } else if ((error
= vdev_metaslab_init(vd
, 0)) != 0) {
2811 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
2812 "[error=%d]", error
);
2813 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2814 VDEV_AUX_CORRUPT_DATA
);
2818 uint64_t checkpoint_sm_obj
= vdev_checkpoint_sm_object(vd
);
2819 if (checkpoint_sm_obj
!= 0) {
2820 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
2821 ASSERT(vd
->vdev_asize
!= 0);
2822 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
2824 if ((error
= space_map_open(&vd
->vdev_checkpoint_sm
,
2825 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
2826 vd
->vdev_ashift
))) {
2827 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
2828 "failed for checkpoint spacemap (obj %llu) "
2830 (u_longlong_t
)checkpoint_sm_obj
, error
);
2833 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
2834 space_map_update(vd
->vdev_checkpoint_sm
);
2837 * Since the checkpoint_sm contains free entries
2838 * exclusively we can use sm_alloc to indicate the
2839 * culmulative checkpointed space that has been freed.
2841 vd
->vdev_stat
.vs_checkpoint_space
=
2842 -vd
->vdev_checkpoint_sm
->sm_alloc
;
2843 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
2844 vd
->vdev_stat
.vs_checkpoint_space
;
2849 * If this is a leaf vdev, load its DTL.
2851 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
2852 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2853 VDEV_AUX_CORRUPT_DATA
);
2854 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
2855 "[error=%d]", error
);
2859 uint64_t obsolete_sm_object
= vdev_obsolete_sm_object(vd
);
2860 if (obsolete_sm_object
!= 0) {
2861 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
2862 ASSERT(vd
->vdev_asize
!= 0);
2863 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
2865 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
2866 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
2867 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2868 VDEV_AUX_CORRUPT_DATA
);
2869 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
2870 "obsolete spacemap (obj %llu) [error=%d]",
2871 (u_longlong_t
)obsolete_sm_object
, error
);
2874 space_map_update(vd
->vdev_obsolete_sm
);
2881 * The special vdev case is used for hot spares and l2cache devices. Its
2882 * sole purpose it to set the vdev state for the associated vdev. To do this,
2883 * we make sure that we can open the underlying device, then try to read the
2884 * label, and make sure that the label is sane and that it hasn't been
2885 * repurposed to another pool.
2888 vdev_validate_aux(vdev_t
*vd
)
2891 uint64_t guid
, version
;
2894 if (!vdev_readable(vd
))
2897 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
2898 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2899 VDEV_AUX_CORRUPT_DATA
);
2903 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
2904 !SPA_VERSION_IS_SUPPORTED(version
) ||
2905 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
2906 guid
!= vd
->vdev_guid
||
2907 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
2908 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2909 VDEV_AUX_CORRUPT_DATA
);
2915 * We don't actually check the pool state here. If it's in fact in
2916 * use by another pool, we update this fact on the fly when requested.
2923 * Free the objects used to store this vdev's spacemaps, and the array
2924 * that points to them.
2927 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2929 if (vd
->vdev_ms_array
== 0)
2932 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
2933 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
2934 size_t array_bytes
= array_count
* sizeof (uint64_t);
2935 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
2936 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
2937 array_bytes
, smobj_array
, 0));
2939 for (uint64_t i
= 0; i
< array_count
; i
++) {
2940 uint64_t smobj
= smobj_array
[i
];
2944 space_map_free_obj(mos
, smobj
, tx
);
2947 kmem_free(smobj_array
, array_bytes
);
2948 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
2949 vd
->vdev_ms_array
= 0;
2953 vdev_remove_empty(vdev_t
*vd
, uint64_t txg
)
2955 spa_t
*spa
= vd
->vdev_spa
;
2958 ASSERT(vd
== vd
->vdev_top
);
2959 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
2961 if (vd
->vdev_ms
!= NULL
) {
2962 metaslab_group_t
*mg
= vd
->vdev_mg
;
2964 metaslab_group_histogram_verify(mg
);
2965 metaslab_class_histogram_verify(mg
->mg_class
);
2967 for (int m
= 0; m
< vd
->vdev_ms_count
; m
++) {
2968 metaslab_t
*msp
= vd
->vdev_ms
[m
];
2970 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
2973 mutex_enter(&msp
->ms_lock
);
2975 * If the metaslab was not loaded when the vdev
2976 * was removed then the histogram accounting may
2977 * not be accurate. Update the histogram information
2978 * here so that we ensure that the metaslab group
2979 * and metaslab class are up-to-date.
2981 metaslab_group_histogram_remove(mg
, msp
);
2983 VERIFY0(space_map_allocated(msp
->ms_sm
));
2984 space_map_close(msp
->ms_sm
);
2986 mutex_exit(&msp
->ms_lock
);
2989 if (vd
->vdev_checkpoint_sm
!= NULL
) {
2990 ASSERT(spa_has_checkpoint(spa
));
2991 space_map_close(vd
->vdev_checkpoint_sm
);
2992 vd
->vdev_checkpoint_sm
= NULL
;
2995 metaslab_group_histogram_verify(mg
);
2996 metaslab_class_histogram_verify(mg
->mg_class
);
2997 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
2998 ASSERT0(mg
->mg_histogram
[i
]);
3001 tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
3002 vdev_destroy_spacemaps(vd
, tx
);
3004 if (vd
->vdev_islog
&& vd
->vdev_top_zap
!= 0) {
3005 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
3006 vd
->vdev_top_zap
= 0;
3012 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
3015 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
3017 ASSERT(vdev_is_concrete(vd
));
3019 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
3020 metaslab_sync_done(msp
, txg
);
3023 metaslab_sync_reassess(vd
->vdev_mg
);
3027 vdev_sync(vdev_t
*vd
, uint64_t txg
)
3029 spa_t
*spa
= vd
->vdev_spa
;
3034 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
3037 ASSERT(vd
->vdev_removing
||
3038 vd
->vdev_ops
== &vdev_indirect_ops
);
3040 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3041 vdev_indirect_sync_obsolete(vd
, tx
);
3045 * If the vdev is indirect, it can't have dirty
3046 * metaslabs or DTLs.
3048 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
3049 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
3050 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
3055 ASSERT(vdev_is_concrete(vd
));
3057 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
3058 !vd
->vdev_removing
) {
3059 ASSERT(vd
== vd
->vdev_top
);
3060 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
3061 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3062 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
3063 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
3064 ASSERT(vd
->vdev_ms_array
!= 0);
3065 vdev_config_dirty(vd
);
3069 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
3070 metaslab_sync(msp
, txg
);
3071 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
3074 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
3075 vdev_dtl_sync(lvd
, txg
);
3078 * Remove the metadata associated with this vdev once it's empty.
3079 * Note that this is typically used for log/cache device removal;
3080 * we don't empty toplevel vdevs when removing them. But if
3081 * a toplevel happens to be emptied, this is not harmful.
3083 if (vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
) {
3084 vdev_remove_empty(vd
, txg
);
3087 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
3091 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
3093 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
3097 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3098 * not be opened, and no I/O is attempted.
3101 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3105 spa_vdev_state_enter(spa
, SCL_NONE
);
3107 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3108 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3110 if (!vd
->vdev_ops
->vdev_op_leaf
)
3111 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3116 * If user did a 'zpool offline -f' then make the fault persist across
3119 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
3121 * There are two kinds of forced faults: temporary and
3122 * persistent. Temporary faults go away at pool import, while
3123 * persistent faults stay set. Both types of faults can be
3124 * cleared with a zpool clear.
3126 * We tell if a vdev is persistently faulted by looking at the
3127 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3128 * import then it's a persistent fault. Otherwise, it's
3129 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3130 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3131 * tells vdev_config_generate() (which gets run later) to set
3132 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3134 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
3135 vd
->vdev_tmpoffline
= B_FALSE
;
3136 aux
= VDEV_AUX_EXTERNAL
;
3138 vd
->vdev_tmpoffline
= B_TRUE
;
3142 * We don't directly use the aux state here, but if we do a
3143 * vdev_reopen(), we need this value to be present to remember why we
3146 vd
->vdev_label_aux
= aux
;
3149 * Faulted state takes precedence over degraded.
3151 vd
->vdev_delayed_close
= B_FALSE
;
3152 vd
->vdev_faulted
= 1ULL;
3153 vd
->vdev_degraded
= 0ULL;
3154 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
3157 * If this device has the only valid copy of the data, then
3158 * back off and simply mark the vdev as degraded instead.
3160 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
3161 vd
->vdev_degraded
= 1ULL;
3162 vd
->vdev_faulted
= 0ULL;
3165 * If we reopen the device and it's not dead, only then do we
3170 if (vdev_readable(vd
))
3171 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
3174 return (spa_vdev_state_exit(spa
, vd
, 0));
3178 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3179 * user that something is wrong. The vdev continues to operate as normal as far
3180 * as I/O is concerned.
3183 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3187 spa_vdev_state_enter(spa
, SCL_NONE
);
3189 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3190 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3192 if (!vd
->vdev_ops
->vdev_op_leaf
)
3193 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3196 * If the vdev is already faulted, then don't do anything.
3198 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
3199 return (spa_vdev_state_exit(spa
, NULL
, 0));
3201 vd
->vdev_degraded
= 1ULL;
3202 if (!vdev_is_dead(vd
))
3203 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
3206 return (spa_vdev_state_exit(spa
, vd
, 0));
3210 * Online the given vdev.
3212 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3213 * spare device should be detached when the device finishes resilvering.
3214 * Second, the online should be treated like a 'test' online case, so no FMA
3215 * events are generated if the device fails to open.
3218 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
3220 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
3221 boolean_t wasoffline
;
3222 vdev_state_t oldstate
;
3224 spa_vdev_state_enter(spa
, SCL_NONE
);
3226 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3227 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3229 if (!vd
->vdev_ops
->vdev_op_leaf
)
3230 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3232 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
3233 oldstate
= vd
->vdev_state
;
3236 vd
->vdev_offline
= B_FALSE
;
3237 vd
->vdev_tmpoffline
= B_FALSE
;
3238 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
3239 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
3241 /* XXX - L2ARC 1.0 does not support expansion */
3242 if (!vd
->vdev_aux
) {
3243 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3244 pvd
->vdev_expanding
= !!((flags
& ZFS_ONLINE_EXPAND
) ||
3245 spa
->spa_autoexpand
);
3249 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
3251 if (!vd
->vdev_aux
) {
3252 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3253 pvd
->vdev_expanding
= B_FALSE
;
3257 *newstate
= vd
->vdev_state
;
3258 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
3259 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
3260 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3261 vd
->vdev_parent
->vdev_child
[0] == vd
)
3262 vd
->vdev_unspare
= B_TRUE
;
3264 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
3266 /* XXX - L2ARC 1.0 does not support expansion */
3268 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
3269 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3273 (oldstate
< VDEV_STATE_DEGRADED
&&
3274 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
3275 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
3277 return (spa_vdev_state_exit(spa
, vd
, 0));
3281 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3285 uint64_t generation
;
3286 metaslab_group_t
*mg
;
3289 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3291 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3292 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3294 if (!vd
->vdev_ops
->vdev_op_leaf
)
3295 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3299 generation
= spa
->spa_config_generation
+ 1;
3302 * If the device isn't already offline, try to offline it.
3304 if (!vd
->vdev_offline
) {
3306 * If this device has the only valid copy of some data,
3307 * don't allow it to be offlined. Log devices are always
3310 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3311 vdev_dtl_required(vd
))
3312 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3315 * If the top-level is a slog and it has had allocations
3316 * then proceed. We check that the vdev's metaslab group
3317 * is not NULL since it's possible that we may have just
3318 * added this vdev but not yet initialized its metaslabs.
3320 if (tvd
->vdev_islog
&& mg
!= NULL
) {
3322 * Prevent any future allocations.
3324 metaslab_group_passivate(mg
);
3325 (void) spa_vdev_state_exit(spa
, vd
, 0);
3327 error
= spa_reset_logs(spa
);
3330 * If the log device was successfully reset but has
3331 * checkpointed data, do not offline it.
3334 tvd
->vdev_checkpoint_sm
!= NULL
) {
3335 ASSERT3U(tvd
->vdev_checkpoint_sm
->sm_alloc
,
3337 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
3340 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3343 * Check to see if the config has changed.
3345 if (error
|| generation
!= spa
->spa_config_generation
) {
3346 metaslab_group_activate(mg
);
3348 return (spa_vdev_state_exit(spa
,
3350 (void) spa_vdev_state_exit(spa
, vd
, 0);
3353 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
3357 * Offline this device and reopen its top-level vdev.
3358 * If the top-level vdev is a log device then just offline
3359 * it. Otherwise, if this action results in the top-level
3360 * vdev becoming unusable, undo it and fail the request.
3362 vd
->vdev_offline
= B_TRUE
;
3365 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3366 vdev_is_dead(tvd
)) {
3367 vd
->vdev_offline
= B_FALSE
;
3369 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3373 * Add the device back into the metaslab rotor so that
3374 * once we online the device it's open for business.
3376 if (tvd
->vdev_islog
&& mg
!= NULL
)
3377 metaslab_group_activate(mg
);
3380 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
3382 return (spa_vdev_state_exit(spa
, vd
, 0));
3386 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3390 mutex_enter(&spa
->spa_vdev_top_lock
);
3391 error
= vdev_offline_locked(spa
, guid
, flags
);
3392 mutex_exit(&spa
->spa_vdev_top_lock
);
3398 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3399 * vdev_offline(), we assume the spa config is locked. We also clear all
3400 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3403 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
3405 vdev_t
*rvd
= spa
->spa_root_vdev
;
3407 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3412 vd
->vdev_stat
.vs_read_errors
= 0;
3413 vd
->vdev_stat
.vs_write_errors
= 0;
3414 vd
->vdev_stat
.vs_checksum_errors
= 0;
3416 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3417 vdev_clear(spa
, vd
->vdev_child
[c
]);
3420 * It makes no sense to "clear" an indirect vdev.
3422 if (!vdev_is_concrete(vd
))
3426 * If we're in the FAULTED state or have experienced failed I/O, then
3427 * clear the persistent state and attempt to reopen the device. We
3428 * also mark the vdev config dirty, so that the new faulted state is
3429 * written out to disk.
3431 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
3432 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
3434 * When reopening in response to a clear event, it may be due to
3435 * a fmadm repair request. In this case, if the device is
3436 * still broken, we want to still post the ereport again.
3438 vd
->vdev_forcefault
= B_TRUE
;
3440 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
3441 vd
->vdev_cant_read
= B_FALSE
;
3442 vd
->vdev_cant_write
= B_FALSE
;
3443 vd
->vdev_stat
.vs_aux
= 0;
3445 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
3447 vd
->vdev_forcefault
= B_FALSE
;
3449 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
3450 vdev_state_dirty(vd
->vdev_top
);
3452 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
))
3453 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3455 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
3459 * When clearing a FMA-diagnosed fault, we always want to
3460 * unspare the device, as we assume that the original spare was
3461 * done in response to the FMA fault.
3463 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
3464 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3465 vd
->vdev_parent
->vdev_child
[0] == vd
)
3466 vd
->vdev_unspare
= B_TRUE
;
3470 vdev_is_dead(vdev_t
*vd
)
3473 * Holes and missing devices are always considered "dead".
3474 * This simplifies the code since we don't have to check for
3475 * these types of devices in the various code paths.
3476 * Instead we rely on the fact that we skip over dead devices
3477 * before issuing I/O to them.
3479 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
3480 vd
->vdev_ops
== &vdev_hole_ops
||
3481 vd
->vdev_ops
== &vdev_missing_ops
);
3485 vdev_readable(vdev_t
*vd
)
3487 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
3491 vdev_writeable(vdev_t
*vd
)
3493 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
3494 vdev_is_concrete(vd
));
3498 vdev_allocatable(vdev_t
*vd
)
3500 uint64_t state
= vd
->vdev_state
;
3503 * We currently allow allocations from vdevs which may be in the
3504 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3505 * fails to reopen then we'll catch it later when we're holding
3506 * the proper locks. Note that we have to get the vdev state
3507 * in a local variable because although it changes atomically,
3508 * we're asking two separate questions about it.
3510 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
3511 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
3512 vd
->vdev_mg
->mg_initialized
);
3516 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
3518 ASSERT(zio
->io_vd
== vd
);
3520 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
3523 if (zio
->io_type
== ZIO_TYPE_READ
)
3524 return (!vd
->vdev_cant_read
);
3526 if (zio
->io_type
== ZIO_TYPE_WRITE
)
3527 return (!vd
->vdev_cant_write
);
3533 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
3536 for (t
= 0; t
< ZIO_TYPES
; t
++) {
3537 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
3538 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
3541 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
3545 * Get extended stats
3548 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
3551 for (t
= 0; t
< ZIO_TYPES
; t
++) {
3552 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
3553 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
3555 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
3556 vsx
->vsx_total_histo
[t
][b
] +=
3557 cvsx
->vsx_total_histo
[t
][b
];
3561 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
3562 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
3563 vsx
->vsx_queue_histo
[t
][b
] +=
3564 cvsx
->vsx_queue_histo
[t
][b
];
3566 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
3567 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
3569 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
3570 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
3572 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
3573 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
3579 vdev_is_spacemap_addressable(vdev_t
*vd
)
3582 * Assuming 47 bits of the space map entry dedicated for the entry's
3583 * offset (see description in space_map.h), we calculate the maximum
3584 * address that can be described by a space map entry for the given
3587 uint64_t shift
= vd
->vdev_ashift
+ 47;
3589 if (shift
>= 63) /* detect potential overflow */
3592 return (vd
->vdev_asize
< (1ULL << shift
));
3596 * Get statistics for the given vdev.
3599 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
3603 * If we're getting stats on the root vdev, aggregate the I/O counts
3604 * over all top-level vdevs (i.e. the direct children of the root).
3606 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3608 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
3609 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
3612 memset(vsx
, 0, sizeof (*vsx
));
3614 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3615 vdev_t
*cvd
= vd
->vdev_child
[c
];
3616 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
3617 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
3619 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
3621 vdev_get_child_stat(cvd
, vs
, cvs
);
3623 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
3628 * We're a leaf. Just copy our ZIO active queue stats in. The
3629 * other leaf stats are updated in vdev_stat_update().
3634 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
3636 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
3637 vsx
->vsx_active_queue
[t
] =
3638 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
3639 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
3640 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
3646 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
3648 vdev_t
*tvd
= vd
->vdev_top
;
3649 mutex_enter(&vd
->vdev_stat_lock
);
3651 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
3652 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
3653 vs
->vs_state
= vd
->vdev_state
;
3654 vs
->vs_rsize
= vdev_get_min_asize(vd
);
3655 if (vd
->vdev_ops
->vdev_op_leaf
)
3656 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
3657 VDEV_LABEL_END_SIZE
;
3659 * Report expandable space on top-level, non-auxillary devices
3660 * only. The expandable space is reported in terms of metaslab
3661 * sized units since that determines how much space the pool
3664 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
3665 vs
->vs_esize
= P2ALIGN(
3666 vd
->vdev_max_asize
- vd
->vdev_asize
,
3667 1ULL << tvd
->vdev_ms_shift
);
3669 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
3670 vdev_is_concrete(vd
)) {
3671 vs
->vs_fragmentation
= vd
->vdev_mg
->mg_fragmentation
;
3675 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
3676 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
3677 mutex_exit(&vd
->vdev_stat_lock
);
3681 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
3683 return (vdev_get_stats_ex(vd
, vs
, NULL
));
3687 vdev_clear_stats(vdev_t
*vd
)
3689 mutex_enter(&vd
->vdev_stat_lock
);
3690 vd
->vdev_stat
.vs_space
= 0;
3691 vd
->vdev_stat
.vs_dspace
= 0;
3692 vd
->vdev_stat
.vs_alloc
= 0;
3693 mutex_exit(&vd
->vdev_stat_lock
);
3697 vdev_scan_stat_init(vdev_t
*vd
)
3699 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3701 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3702 vdev_scan_stat_init(vd
->vdev_child
[c
]);
3704 mutex_enter(&vd
->vdev_stat_lock
);
3705 vs
->vs_scan_processed
= 0;
3706 mutex_exit(&vd
->vdev_stat_lock
);
3710 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
3712 spa_t
*spa
= zio
->io_spa
;
3713 vdev_t
*rvd
= spa
->spa_root_vdev
;
3714 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
3716 uint64_t txg
= zio
->io_txg
;
3717 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3718 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
3719 zio_type_t type
= zio
->io_type
;
3720 int flags
= zio
->io_flags
;
3723 * If this i/o is a gang leader, it didn't do any actual work.
3725 if (zio
->io_gang_tree
)
3728 if (zio
->io_error
== 0) {
3730 * If this is a root i/o, don't count it -- we've already
3731 * counted the top-level vdevs, and vdev_get_stats() will
3732 * aggregate them when asked. This reduces contention on
3733 * the root vdev_stat_lock and implicitly handles blocks
3734 * that compress away to holes, for which there is no i/o.
3735 * (Holes never create vdev children, so all the counters
3736 * remain zero, which is what we want.)
3738 * Note: this only applies to successful i/o (io_error == 0)
3739 * because unlike i/o counts, errors are not additive.
3740 * When reading a ditto block, for example, failure of
3741 * one top-level vdev does not imply a root-level error.
3746 ASSERT(vd
== zio
->io_vd
);
3748 if (flags
& ZIO_FLAG_IO_BYPASS
)
3751 mutex_enter(&vd
->vdev_stat_lock
);
3753 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3754 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3755 dsl_scan_phys_t
*scn_phys
=
3756 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3757 uint64_t *processed
= &scn_phys
->scn_processed
;
3760 if (vd
->vdev_ops
->vdev_op_leaf
)
3761 atomic_add_64(processed
, psize
);
3762 vs
->vs_scan_processed
+= psize
;
3765 if (flags
& ZIO_FLAG_SELF_HEAL
)
3766 vs
->vs_self_healed
+= psize
;
3770 * The bytes/ops/histograms are recorded at the leaf level and
3771 * aggregated into the higher level vdevs in vdev_get_stats().
3773 if (vd
->vdev_ops
->vdev_op_leaf
&&
3774 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3777 vs
->vs_bytes
[type
] += psize
;
3779 if (flags
& ZIO_FLAG_DELEGATED
) {
3780 vsx
->vsx_agg_histo
[zio
->io_priority
]
3781 [RQ_HISTO(zio
->io_size
)]++;
3783 vsx
->vsx_ind_histo
[zio
->io_priority
]
3784 [RQ_HISTO(zio
->io_size
)]++;
3787 if (zio
->io_delta
&& zio
->io_delay
) {
3788 vsx
->vsx_queue_histo
[zio
->io_priority
]
3789 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3790 vsx
->vsx_disk_histo
[type
]
3791 [L_HISTO(zio
->io_delay
)]++;
3792 vsx
->vsx_total_histo
[type
]
3793 [L_HISTO(zio
->io_delta
)]++;
3797 mutex_exit(&vd
->vdev_stat_lock
);
3801 if (flags
& ZIO_FLAG_SPECULATIVE
)
3805 * If this is an I/O error that is going to be retried, then ignore the
3806 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3807 * hard errors, when in reality they can happen for any number of
3808 * innocuous reasons (bus resets, MPxIO link failure, etc).
3810 if (zio
->io_error
== EIO
&&
3811 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
3815 * Intent logs writes won't propagate their error to the root
3816 * I/O so don't mark these types of failures as pool-level
3819 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
3822 mutex_enter(&vd
->vdev_stat_lock
);
3823 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
3824 if (zio
->io_error
== ECKSUM
)
3825 vs
->vs_checksum_errors
++;
3827 vs
->vs_read_errors
++;
3829 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
3830 vs
->vs_write_errors
++;
3831 mutex_exit(&vd
->vdev_stat_lock
);
3833 if (spa
->spa_load_state
== SPA_LOAD_NONE
&&
3834 type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
3835 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
3836 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
3837 spa
->spa_claiming
)) {
3839 * This is either a normal write (not a repair), or it's
3840 * a repair induced by the scrub thread, or it's a repair
3841 * made by zil_claim() during spa_load() in the first txg.
3842 * In the normal case, we commit the DTL change in the same
3843 * txg as the block was born. In the scrub-induced repair
3844 * case, we know that scrubs run in first-pass syncing context,
3845 * so we commit the DTL change in spa_syncing_txg(spa).
3846 * In the zil_claim() case, we commit in spa_first_txg(spa).
3848 * We currently do not make DTL entries for failed spontaneous
3849 * self-healing writes triggered by normal (non-scrubbing)
3850 * reads, because we have no transactional context in which to
3851 * do so -- and it's not clear that it'd be desirable anyway.
3853 if (vd
->vdev_ops
->vdev_op_leaf
) {
3854 uint64_t commit_txg
= txg
;
3855 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3856 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3857 ASSERT(spa_sync_pass(spa
) == 1);
3858 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
3859 commit_txg
= spa_syncing_txg(spa
);
3860 } else if (spa
->spa_claiming
) {
3861 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
3862 commit_txg
= spa_first_txg(spa
);
3864 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
3865 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
3867 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3868 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
3869 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
3872 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
3877 * Update the in-core space usage stats for this vdev, its metaslab class,
3878 * and the root vdev.
3881 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
3882 int64_t space_delta
)
3884 int64_t dspace_delta
= space_delta
;
3885 spa_t
*spa
= vd
->vdev_spa
;
3886 vdev_t
*rvd
= spa
->spa_root_vdev
;
3887 metaslab_group_t
*mg
= vd
->vdev_mg
;
3888 metaslab_class_t
*mc
= mg
? mg
->mg_class
: NULL
;
3890 ASSERT(vd
== vd
->vdev_top
);
3893 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3894 * factor. We must calculate this here and not at the root vdev
3895 * because the root vdev's psize-to-asize is simply the max of its
3896 * childrens', thus not accurate enough for us.
3898 ASSERT((dspace_delta
& (SPA_MINBLOCKSIZE
-1)) == 0);
3899 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
3900 dspace_delta
= (dspace_delta
>> SPA_MINBLOCKSHIFT
) *
3901 vd
->vdev_deflate_ratio
;
3903 mutex_enter(&vd
->vdev_stat_lock
);
3904 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3905 vd
->vdev_stat
.vs_space
+= space_delta
;
3906 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3907 mutex_exit(&vd
->vdev_stat_lock
);
3909 if (mc
== spa_normal_class(spa
)) {
3910 mutex_enter(&rvd
->vdev_stat_lock
);
3911 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
3912 rvd
->vdev_stat
.vs_space
+= space_delta
;
3913 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
3914 mutex_exit(&rvd
->vdev_stat_lock
);
3918 ASSERT(rvd
== vd
->vdev_parent
);
3919 ASSERT(vd
->vdev_ms_count
!= 0);
3921 metaslab_class_space_update(mc
,
3922 alloc_delta
, defer_delta
, space_delta
, dspace_delta
);
3927 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3928 * so that it will be written out next time the vdev configuration is synced.
3929 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3932 vdev_config_dirty(vdev_t
*vd
)
3934 spa_t
*spa
= vd
->vdev_spa
;
3935 vdev_t
*rvd
= spa
->spa_root_vdev
;
3938 ASSERT(spa_writeable(spa
));
3941 * If this is an aux vdev (as with l2cache and spare devices), then we
3942 * update the vdev config manually and set the sync flag.
3944 if (vd
->vdev_aux
!= NULL
) {
3945 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
3949 for (c
= 0; c
< sav
->sav_count
; c
++) {
3950 if (sav
->sav_vdevs
[c
] == vd
)
3954 if (c
== sav
->sav_count
) {
3956 * We're being removed. There's nothing more to do.
3958 ASSERT(sav
->sav_sync
== B_TRUE
);
3962 sav
->sav_sync
= B_TRUE
;
3964 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
3965 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
3966 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
3967 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
3973 * Setting the nvlist in the middle if the array is a little
3974 * sketchy, but it will work.
3976 nvlist_free(aux
[c
]);
3977 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
3983 * The dirty list is protected by the SCL_CONFIG lock. The caller
3984 * must either hold SCL_CONFIG as writer, or must be the sync thread
3985 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3986 * so this is sufficient to ensure mutual exclusion.
3988 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
3989 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
3990 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
3993 for (c
= 0; c
< rvd
->vdev_children
; c
++)
3994 vdev_config_dirty(rvd
->vdev_child
[c
]);
3996 ASSERT(vd
== vd
->vdev_top
);
3998 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
3999 vdev_is_concrete(vd
)) {
4000 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
4006 vdev_config_clean(vdev_t
*vd
)
4008 spa_t
*spa
= vd
->vdev_spa
;
4010 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
4011 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4012 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
4014 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
4015 list_remove(&spa
->spa_config_dirty_list
, vd
);
4019 * Mark a top-level vdev's state as dirty, so that the next pass of
4020 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4021 * the state changes from larger config changes because they require
4022 * much less locking, and are often needed for administrative actions.
4025 vdev_state_dirty(vdev_t
*vd
)
4027 spa_t
*spa
= vd
->vdev_spa
;
4029 ASSERT(spa_writeable(spa
));
4030 ASSERT(vd
== vd
->vdev_top
);
4033 * The state list is protected by the SCL_STATE lock. The caller
4034 * must either hold SCL_STATE as writer, or must be the sync thread
4035 * (which holds SCL_STATE as reader). There's only one sync thread,
4036 * so this is sufficient to ensure mutual exclusion.
4038 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4039 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4040 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4042 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
4043 vdev_is_concrete(vd
))
4044 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
4048 vdev_state_clean(vdev_t
*vd
)
4050 spa_t
*spa
= vd
->vdev_spa
;
4052 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4053 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4054 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4056 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
4057 list_remove(&spa
->spa_state_dirty_list
, vd
);
4061 * Propagate vdev state up from children to parent.
4064 vdev_propagate_state(vdev_t
*vd
)
4066 spa_t
*spa
= vd
->vdev_spa
;
4067 vdev_t
*rvd
= spa
->spa_root_vdev
;
4068 int degraded
= 0, faulted
= 0;
4072 if (vd
->vdev_children
> 0) {
4073 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4074 child
= vd
->vdev_child
[c
];
4077 * Don't factor holes or indirect vdevs into the
4080 if (!vdev_is_concrete(child
))
4083 if (!vdev_readable(child
) ||
4084 (!vdev_writeable(child
) && spa_writeable(spa
))) {
4086 * Root special: if there is a top-level log
4087 * device, treat the root vdev as if it were
4090 if (child
->vdev_islog
&& vd
== rvd
)
4094 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
4098 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
4102 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
4105 * Root special: if there is a top-level vdev that cannot be
4106 * opened due to corrupted metadata, then propagate the root
4107 * vdev's aux state as 'corrupt' rather than 'insufficient
4110 if (corrupted
&& vd
== rvd
&&
4111 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
4112 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
4113 VDEV_AUX_CORRUPT_DATA
);
4116 if (vd
->vdev_parent
)
4117 vdev_propagate_state(vd
->vdev_parent
);
4121 * Set a vdev's state. If this is during an open, we don't update the parent
4122 * state, because we're in the process of opening children depth-first.
4123 * Otherwise, we propagate the change to the parent.
4125 * If this routine places a device in a faulted state, an appropriate ereport is
4129 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
4131 uint64_t save_state
;
4132 spa_t
*spa
= vd
->vdev_spa
;
4134 if (state
== vd
->vdev_state
) {
4136 * Since vdev_offline() code path is already in an offline
4137 * state we can miss a statechange event to OFFLINE. Check
4138 * the previous state to catch this condition.
4140 if (vd
->vdev_ops
->vdev_op_leaf
&&
4141 (state
== VDEV_STATE_OFFLINE
) &&
4142 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
4143 /* post an offline state change */
4144 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
4146 vd
->vdev_stat
.vs_aux
= aux
;
4150 save_state
= vd
->vdev_state
;
4152 vd
->vdev_state
= state
;
4153 vd
->vdev_stat
.vs_aux
= aux
;
4156 * If we are setting the vdev state to anything but an open state, then
4157 * always close the underlying device unless the device has requested
4158 * a delayed close (i.e. we're about to remove or fault the device).
4159 * Otherwise, we keep accessible but invalid devices open forever.
4160 * We don't call vdev_close() itself, because that implies some extra
4161 * checks (offline, etc) that we don't want here. This is limited to
4162 * leaf devices, because otherwise closing the device will affect other
4165 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
4166 vd
->vdev_ops
->vdev_op_leaf
)
4167 vd
->vdev_ops
->vdev_op_close(vd
);
4169 if (vd
->vdev_removed
&&
4170 state
== VDEV_STATE_CANT_OPEN
&&
4171 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
4173 * If the previous state is set to VDEV_STATE_REMOVED, then this
4174 * device was previously marked removed and someone attempted to
4175 * reopen it. If this failed due to a nonexistent device, then
4176 * keep the device in the REMOVED state. We also let this be if
4177 * it is one of our special test online cases, which is only
4178 * attempting to online the device and shouldn't generate an FMA
4181 vd
->vdev_state
= VDEV_STATE_REMOVED
;
4182 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
4183 } else if (state
== VDEV_STATE_REMOVED
) {
4184 vd
->vdev_removed
= B_TRUE
;
4185 } else if (state
== VDEV_STATE_CANT_OPEN
) {
4187 * If we fail to open a vdev during an import or recovery, we
4188 * mark it as "not available", which signifies that it was
4189 * never there to begin with. Failure to open such a device
4190 * is not considered an error.
4192 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
4193 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
4194 vd
->vdev_ops
->vdev_op_leaf
)
4195 vd
->vdev_not_present
= 1;
4198 * Post the appropriate ereport. If the 'prevstate' field is
4199 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4200 * that this is part of a vdev_reopen(). In this case, we don't
4201 * want to post the ereport if the device was already in the
4202 * CANT_OPEN state beforehand.
4204 * If the 'checkremove' flag is set, then this is an attempt to
4205 * online the device in response to an insertion event. If we
4206 * hit this case, then we have detected an insertion event for a
4207 * faulted or offline device that wasn't in the removed state.
4208 * In this scenario, we don't post an ereport because we are
4209 * about to replace the device, or attempt an online with
4210 * vdev_forcefault, which will generate the fault for us.
4212 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
4213 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
4214 vd
!= spa
->spa_root_vdev
) {
4218 case VDEV_AUX_OPEN_FAILED
:
4219 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
4221 case VDEV_AUX_CORRUPT_DATA
:
4222 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
4224 case VDEV_AUX_NO_REPLICAS
:
4225 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
4227 case VDEV_AUX_BAD_GUID_SUM
:
4228 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
4230 case VDEV_AUX_TOO_SMALL
:
4231 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
4233 case VDEV_AUX_BAD_LABEL
:
4234 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
4236 case VDEV_AUX_BAD_ASHIFT
:
4237 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
4240 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
4243 zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
4247 /* Erase any notion of persistent removed state */
4248 vd
->vdev_removed
= B_FALSE
;
4250 vd
->vdev_removed
= B_FALSE
;
4254 * Notify ZED of any significant state-change on a leaf vdev.
4257 if (vd
->vdev_ops
->vdev_op_leaf
) {
4258 /* preserve original state from a vdev_reopen() */
4259 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
4260 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
4261 (save_state
<= VDEV_STATE_CLOSED
))
4262 save_state
= vd
->vdev_prevstate
;
4264 /* filter out state change due to initial vdev_open */
4265 if (save_state
> VDEV_STATE_CLOSED
)
4266 zfs_post_state_change(spa
, vd
, save_state
);
4269 if (!isopen
&& vd
->vdev_parent
)
4270 vdev_propagate_state(vd
->vdev_parent
);
4274 vdev_children_are_offline(vdev_t
*vd
)
4276 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
4278 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
4279 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
4287 * Check the vdev configuration to ensure that it's capable of supporting
4288 * a root pool. We do not support partial configuration.
4291 vdev_is_bootable(vdev_t
*vd
)
4293 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4294 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
4296 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0 ||
4297 strcmp(vdev_type
, VDEV_TYPE_INDIRECT
) == 0) {
4302 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4303 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
4310 vdev_is_concrete(vdev_t
*vd
)
4312 vdev_ops_t
*ops
= vd
->vdev_ops
;
4313 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
4314 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
4322 * Determine if a log device has valid content. If the vdev was
4323 * removed or faulted in the MOS config then we know that
4324 * the content on the log device has already been written to the pool.
4327 vdev_log_state_valid(vdev_t
*vd
)
4329 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
4333 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4334 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
4341 * Expand a vdev if possible.
4344 vdev_expand(vdev_t
*vd
, uint64_t txg
)
4346 ASSERT(vd
->vdev_top
== vd
);
4347 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4348 ASSERT(vdev_is_concrete(vd
));
4350 vdev_set_deflate_ratio(vd
);
4352 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
) {
4353 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
4354 vdev_config_dirty(vd
);
4362 vdev_split(vdev_t
*vd
)
4364 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
4366 vdev_remove_child(pvd
, vd
);
4367 vdev_compact_children(pvd
);
4369 cvd
= pvd
->vdev_child
[0];
4370 if (pvd
->vdev_children
== 1) {
4371 vdev_remove_parent(cvd
);
4372 cvd
->vdev_splitting
= B_TRUE
;
4374 vdev_propagate_state(cvd
);
4378 vdev_deadman(vdev_t
*vd
, char *tag
)
4380 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4381 vdev_t
*cvd
= vd
->vdev_child
[c
];
4383 vdev_deadman(cvd
, tag
);
4386 if (vd
->vdev_ops
->vdev_op_leaf
) {
4387 vdev_queue_t
*vq
= &vd
->vdev_queue
;
4389 mutex_enter(&vq
->vq_lock
);
4390 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
4391 spa_t
*spa
= vd
->vdev_spa
;
4395 zfs_dbgmsg("slow vdev: %s has %d active IOs",
4396 vd
->vdev_path
, avl_numnodes(&vq
->vq_active_tree
));
4399 * Look at the head of all the pending queues,
4400 * if any I/O has been outstanding for longer than
4401 * the spa_deadman_synctime invoke the deadman logic.
4403 fio
= avl_first(&vq
->vq_active_tree
);
4404 delta
= gethrtime() - fio
->io_timestamp
;
4405 if (delta
> spa_deadman_synctime(spa
))
4406 zio_deadman(fio
, tag
);
4408 mutex_exit(&vq
->vq_lock
);
4412 #if defined(_KERNEL)
4413 EXPORT_SYMBOL(vdev_fault
);
4414 EXPORT_SYMBOL(vdev_degrade
);
4415 EXPORT_SYMBOL(vdev_online
);
4416 EXPORT_SYMBOL(vdev_offline
);
4417 EXPORT_SYMBOL(vdev_clear
);
4419 module_param(vdev_max_ms_count
, int, 0644);
4420 MODULE_PARM_DESC(vdev_max_ms_count
,
4421 "Target number of metaslabs per top-level vdev");
4423 module_param(vdev_min_ms_count
, int, 0644);
4424 MODULE_PARM_DESC(vdev_min_ms_count
,
4425 "Minimum number of metaslabs per top-level vdev");
4427 module_param(vdev_ms_count_limit
, int, 0644);
4428 MODULE_PARM_DESC(vdev_ms_count_limit
,
4429 "Practical upper limit of total metaslabs per top-level vdev");
4431 module_param(zfs_delays_per_second
, uint
, 0644);
4432 MODULE_PARM_DESC(zfs_delays_per_second
, "Rate limit delay events to this many "
4433 "IO delays per second");
4435 module_param(zfs_checksums_per_second
, uint
, 0644);
4436 MODULE_PARM_DESC(zfs_checksums_per_second
, "Rate limit checksum events "
4437 "to this many checksum errors per second (do not set below zed"
4440 module_param(zfs_scan_ignore_errors
, int, 0644);
4441 MODULE_PARM_DESC(zfs_scan_ignore_errors
,
4442 "Ignore errors during resilver/scrub");
4444 module_param(vdev_validate_skip
, int, 0644);
4445 MODULE_PARM_DESC(vdev_validate_skip
,
4446 "Bypass vdev_validate()");