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.
29 * Copyright (c) 2017, Intel Corporation.
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
36 #include <sys/bpobj.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/uberblock_impl.h>
42 #include <sys/metaslab.h>
43 #include <sys/metaslab_impl.h>
44 #include <sys/space_map.h>
45 #include <sys/space_reftree.h>
48 #include <sys/fs/zfs.h>
51 #include <sys/dsl_scan.h>
53 #include <sys/vdev_initialize.h>
55 #include <sys/zfs_ratelimit.h>
57 /* target number of metaslabs per top-level vdev */
58 int vdev_max_ms_count
= 200;
60 /* minimum number of metaslabs per top-level vdev */
61 int vdev_min_ms_count
= 16;
63 /* practical upper limit of total metaslabs per top-level vdev */
64 int vdev_ms_count_limit
= 1ULL << 17;
66 /* lower limit for metaslab size (512M) */
67 int vdev_default_ms_shift
= 29;
69 /* upper limit for metaslab size (256G) */
70 int vdev_max_ms_shift
= 38;
72 int vdev_validate_skip
= B_FALSE
;
75 * Since the DTL space map of a vdev is not expected to have a lot of
76 * entries, we default its block size to 4K.
78 int vdev_dtl_sm_blksz
= (1 << 12);
81 * Rate limit slow IO (delay) events to this many per second.
83 unsigned int zfs_slow_io_events_per_second
= 20;
86 * Rate limit checksum events after this many checksum errors per second.
88 unsigned int zfs_checksum_events_per_second
= 20;
91 * Ignore errors during scrub/resilver. Allows to work around resilver
92 * upon import when there are pool errors.
94 int zfs_scan_ignore_errors
= 0;
97 * vdev-wide space maps that have lots of entries written to them at
98 * the end of each transaction can benefit from a higher I/O bandwidth
99 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
101 int vdev_standard_sm_blksz
= (1 << 17);
104 * Tunable parameter for debugging or performance analysis. Setting this
105 * will cause pool corruption on power loss if a volatile out-of-order
106 * write cache is enabled.
108 int zfs_nocacheflush
= 0;
112 vdev_dbgmsg(vdev_t
*vd
, const char *fmt
, ...)
118 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
121 if (vd
->vdev_path
!= NULL
) {
122 zfs_dbgmsg("%s vdev '%s': %s", vd
->vdev_ops
->vdev_op_type
,
125 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
126 vd
->vdev_ops
->vdev_op_type
,
127 (u_longlong_t
)vd
->vdev_id
,
128 (u_longlong_t
)vd
->vdev_guid
, buf
);
133 vdev_dbgmsg_print_tree(vdev_t
*vd
, int indent
)
137 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
) {
138 zfs_dbgmsg("%*svdev %u: %s", indent
, "", vd
->vdev_id
,
139 vd
->vdev_ops
->vdev_op_type
);
143 switch (vd
->vdev_state
) {
144 case VDEV_STATE_UNKNOWN
:
145 (void) snprintf(state
, sizeof (state
), "unknown");
147 case VDEV_STATE_CLOSED
:
148 (void) snprintf(state
, sizeof (state
), "closed");
150 case VDEV_STATE_OFFLINE
:
151 (void) snprintf(state
, sizeof (state
), "offline");
153 case VDEV_STATE_REMOVED
:
154 (void) snprintf(state
, sizeof (state
), "removed");
156 case VDEV_STATE_CANT_OPEN
:
157 (void) snprintf(state
, sizeof (state
), "can't open");
159 case VDEV_STATE_FAULTED
:
160 (void) snprintf(state
, sizeof (state
), "faulted");
162 case VDEV_STATE_DEGRADED
:
163 (void) snprintf(state
, sizeof (state
), "degraded");
165 case VDEV_STATE_HEALTHY
:
166 (void) snprintf(state
, sizeof (state
), "healthy");
169 (void) snprintf(state
, sizeof (state
), "<state %u>",
170 (uint_t
)vd
->vdev_state
);
173 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent
,
174 "", (int)vd
->vdev_id
, vd
->vdev_ops
->vdev_op_type
,
175 vd
->vdev_islog
? " (log)" : "",
176 (u_longlong_t
)vd
->vdev_guid
,
177 vd
->vdev_path
? vd
->vdev_path
: "N/A", state
);
179 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++)
180 vdev_dbgmsg_print_tree(vd
->vdev_child
[i
], indent
+ 2);
184 * Virtual device management.
187 static vdev_ops_t
*vdev_ops_table
[] = {
202 * Given a vdev type, return the appropriate ops vector.
205 vdev_getops(const char *type
)
207 vdev_ops_t
*ops
, **opspp
;
209 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
210 if (strcmp(ops
->vdev_op_type
, type
) == 0)
218 vdev_default_xlate(vdev_t
*vd
, const range_seg_t
*in
, range_seg_t
*res
)
220 res
->rs_start
= in
->rs_start
;
221 res
->rs_end
= in
->rs_end
;
225 * Derive the enumerated alloction bias from string input.
226 * String origin is either the per-vdev zap or zpool(1M).
228 static vdev_alloc_bias_t
229 vdev_derive_alloc_bias(const char *bias
)
231 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
233 if (strcmp(bias
, VDEV_ALLOC_BIAS_LOG
) == 0)
234 alloc_bias
= VDEV_BIAS_LOG
;
235 else if (strcmp(bias
, VDEV_ALLOC_BIAS_SPECIAL
) == 0)
236 alloc_bias
= VDEV_BIAS_SPECIAL
;
237 else if (strcmp(bias
, VDEV_ALLOC_BIAS_DEDUP
) == 0)
238 alloc_bias
= VDEV_BIAS_DEDUP
;
244 * Default asize function: return the MAX of psize with the asize of
245 * all children. This is what's used by anything other than RAID-Z.
248 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
250 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
253 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
254 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
255 asize
= MAX(asize
, csize
);
262 * Get the minimum allocatable size. We define the allocatable size as
263 * the vdev's asize rounded to the nearest metaslab. This allows us to
264 * replace or attach devices which don't have the same physical size but
265 * can still satisfy the same number of allocations.
268 vdev_get_min_asize(vdev_t
*vd
)
270 vdev_t
*pvd
= vd
->vdev_parent
;
273 * If our parent is NULL (inactive spare or cache) or is the root,
274 * just return our own asize.
277 return (vd
->vdev_asize
);
280 * The top-level vdev just returns the allocatable size rounded
281 * to the nearest metaslab.
283 if (vd
== vd
->vdev_top
)
284 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
287 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
288 * so each child must provide at least 1/Nth of its asize.
290 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
291 return ((pvd
->vdev_min_asize
+ pvd
->vdev_children
- 1) /
294 return (pvd
->vdev_min_asize
);
298 vdev_set_min_asize(vdev_t
*vd
)
300 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
302 for (int c
= 0; c
< vd
->vdev_children
; c
++)
303 vdev_set_min_asize(vd
->vdev_child
[c
]);
307 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
309 vdev_t
*rvd
= spa
->spa_root_vdev
;
311 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
313 if (vdev
< rvd
->vdev_children
) {
314 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
315 return (rvd
->vdev_child
[vdev
]);
322 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
326 if (vd
->vdev_guid
== guid
)
329 for (int c
= 0; c
< vd
->vdev_children
; c
++)
330 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
338 vdev_count_leaves_impl(vdev_t
*vd
)
342 if (vd
->vdev_ops
->vdev_op_leaf
)
345 for (int c
= 0; c
< vd
->vdev_children
; c
++)
346 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
352 vdev_count_leaves(spa_t
*spa
)
356 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
357 rc
= vdev_count_leaves_impl(spa
->spa_root_vdev
);
358 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
364 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
366 size_t oldsize
, newsize
;
367 uint64_t id
= cvd
->vdev_id
;
370 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
371 ASSERT(cvd
->vdev_parent
== NULL
);
373 cvd
->vdev_parent
= pvd
;
378 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
380 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
381 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
382 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
384 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
385 if (pvd
->vdev_child
!= NULL
) {
386 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
387 kmem_free(pvd
->vdev_child
, oldsize
);
390 pvd
->vdev_child
= newchild
;
391 pvd
->vdev_child
[id
] = cvd
;
393 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
394 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
397 * Walk up all ancestors to update guid sum.
399 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
400 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
404 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
407 uint_t id
= cvd
->vdev_id
;
409 ASSERT(cvd
->vdev_parent
== pvd
);
414 ASSERT(id
< pvd
->vdev_children
);
415 ASSERT(pvd
->vdev_child
[id
] == cvd
);
417 pvd
->vdev_child
[id
] = NULL
;
418 cvd
->vdev_parent
= NULL
;
420 for (c
= 0; c
< pvd
->vdev_children
; c
++)
421 if (pvd
->vdev_child
[c
])
424 if (c
== pvd
->vdev_children
) {
425 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
426 pvd
->vdev_child
= NULL
;
427 pvd
->vdev_children
= 0;
431 * Walk up all ancestors to update guid sum.
433 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
434 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
438 * Remove any holes in the child array.
441 vdev_compact_children(vdev_t
*pvd
)
443 vdev_t
**newchild
, *cvd
;
444 int oldc
= pvd
->vdev_children
;
447 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
452 for (int c
= newc
= 0; c
< oldc
; c
++)
453 if (pvd
->vdev_child
[c
])
457 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
459 for (int c
= newc
= 0; c
< oldc
; c
++) {
460 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
461 newchild
[newc
] = cvd
;
462 cvd
->vdev_id
= newc
++;
469 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
470 pvd
->vdev_child
= newchild
;
471 pvd
->vdev_children
= newc
;
475 * Allocate and minimally initialize a vdev_t.
478 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
481 vdev_indirect_config_t
*vic
;
483 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
484 vic
= &vd
->vdev_indirect_config
;
486 if (spa
->spa_root_vdev
== NULL
) {
487 ASSERT(ops
== &vdev_root_ops
);
488 spa
->spa_root_vdev
= vd
;
489 spa
->spa_load_guid
= spa_generate_guid(NULL
);
492 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
493 if (spa
->spa_root_vdev
== vd
) {
495 * The root vdev's guid will also be the pool guid,
496 * which must be unique among all pools.
498 guid
= spa_generate_guid(NULL
);
501 * Any other vdev's guid must be unique within the pool.
503 guid
= spa_generate_guid(spa
);
505 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
510 vd
->vdev_guid
= guid
;
511 vd
->vdev_guid_sum
= guid
;
513 vd
->vdev_state
= VDEV_STATE_CLOSED
;
514 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
515 vic
->vic_prev_indirect_vdev
= UINT64_MAX
;
517 rw_init(&vd
->vdev_indirect_rwlock
, NULL
, RW_DEFAULT
, NULL
);
518 mutex_init(&vd
->vdev_obsolete_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
519 vd
->vdev_obsolete_segments
= range_tree_create(NULL
, NULL
);
522 * Initialize rate limit structs for events. We rate limit ZIO delay
523 * and checksum events so that we don't overwhelm ZED with thousands
524 * of events when a disk is acting up.
526 zfs_ratelimit_init(&vd
->vdev_delay_rl
, &zfs_slow_io_events_per_second
,
528 zfs_ratelimit_init(&vd
->vdev_checksum_rl
,
529 &zfs_checksum_events_per_second
, 1);
531 list_link_init(&vd
->vdev_config_dirty_node
);
532 list_link_init(&vd
->vdev_state_dirty_node
);
533 list_link_init(&vd
->vdev_initialize_node
);
534 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
535 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
536 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
537 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
538 mutex_init(&vd
->vdev_scan_io_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
539 mutex_init(&vd
->vdev_initialize_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
540 mutex_init(&vd
->vdev_initialize_io_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
541 cv_init(&vd
->vdev_initialize_cv
, NULL
, CV_DEFAULT
, NULL
);
542 cv_init(&vd
->vdev_initialize_io_cv
, NULL
, CV_DEFAULT
, NULL
);
544 for (int t
= 0; t
< DTL_TYPES
; t
++) {
545 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
);
547 txg_list_create(&vd
->vdev_ms_list
, spa
,
548 offsetof(struct metaslab
, ms_txg_node
));
549 txg_list_create(&vd
->vdev_dtl_list
, spa
,
550 offsetof(struct vdev
, vdev_dtl_node
));
551 vd
->vdev_stat
.vs_timestamp
= gethrtime();
559 * Allocate a new vdev. The 'alloctype' is used to control whether we are
560 * creating a new vdev or loading an existing one - the behavior is slightly
561 * different for each case.
564 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
569 uint64_t guid
= 0, islog
, nparity
;
571 vdev_indirect_config_t
*vic
;
574 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
575 boolean_t top_level
= (parent
&& !parent
->vdev_parent
);
577 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
579 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
580 return (SET_ERROR(EINVAL
));
582 if ((ops
= vdev_getops(type
)) == NULL
)
583 return (SET_ERROR(EINVAL
));
586 * If this is a load, get the vdev guid from the nvlist.
587 * Otherwise, vdev_alloc_common() will generate one for us.
589 if (alloctype
== VDEV_ALLOC_LOAD
) {
592 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
594 return (SET_ERROR(EINVAL
));
596 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
597 return (SET_ERROR(EINVAL
));
598 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
599 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
600 return (SET_ERROR(EINVAL
));
601 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
602 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
603 return (SET_ERROR(EINVAL
));
604 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
605 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
606 return (SET_ERROR(EINVAL
));
610 * The first allocated vdev must be of type 'root'.
612 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
613 return (SET_ERROR(EINVAL
));
616 * Determine whether we're a log vdev.
619 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
620 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
621 return (SET_ERROR(ENOTSUP
));
623 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
624 return (SET_ERROR(ENOTSUP
));
627 * Set the nparity property for RAID-Z vdevs.
630 if (ops
== &vdev_raidz_ops
) {
631 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
633 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
634 return (SET_ERROR(EINVAL
));
636 * Previous versions could only support 1 or 2 parity
640 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
641 return (SET_ERROR(ENOTSUP
));
643 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
644 return (SET_ERROR(ENOTSUP
));
647 * We require the parity to be specified for SPAs that
648 * support multiple parity levels.
650 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
651 return (SET_ERROR(EINVAL
));
653 * Otherwise, we default to 1 parity device for RAID-Z.
660 ASSERT(nparity
!= -1ULL);
663 * If creating a top-level vdev, check for allocation classes input
665 if (top_level
&& alloctype
== VDEV_ALLOC_ADD
) {
668 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
670 alloc_bias
= vdev_derive_alloc_bias(bias
);
672 /* spa_vdev_add() expects feature to be enabled */
673 if (spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
674 !spa_feature_is_enabled(spa
,
675 SPA_FEATURE_ALLOCATION_CLASSES
)) {
676 return (SET_ERROR(ENOTSUP
));
681 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
682 vic
= &vd
->vdev_indirect_config
;
684 vd
->vdev_islog
= islog
;
685 vd
->vdev_nparity
= nparity
;
686 if (top_level
&& alloc_bias
!= VDEV_BIAS_NONE
)
687 vd
->vdev_alloc_bias
= alloc_bias
;
689 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
690 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
693 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
694 * fault on a vdev and want it to persist across imports (like with
697 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
698 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
699 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
700 vd
->vdev_faulted
= 1;
701 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
704 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
705 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
706 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
707 &vd
->vdev_physpath
) == 0)
708 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
710 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
711 &vd
->vdev_enc_sysfs_path
) == 0)
712 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
714 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
715 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
718 * Set the whole_disk property. If it's not specified, leave the value
721 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
722 &vd
->vdev_wholedisk
) != 0)
723 vd
->vdev_wholedisk
= -1ULL;
725 ASSERT0(vic
->vic_mapping_object
);
726 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
727 &vic
->vic_mapping_object
);
728 ASSERT0(vic
->vic_births_object
);
729 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
730 &vic
->vic_births_object
);
731 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
732 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
733 &vic
->vic_prev_indirect_vdev
);
736 * Look for the 'not present' flag. This will only be set if the device
737 * was not present at the time of import.
739 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
740 &vd
->vdev_not_present
);
743 * Get the alignment requirement.
745 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
748 * Retrieve the vdev creation time.
750 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
754 * If we're a top-level vdev, try to load the allocation parameters.
757 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
758 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
760 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
762 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
764 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
766 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
769 ASSERT0(vd
->vdev_top_zap
);
772 if (top_level
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
773 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
774 alloctype
== VDEV_ALLOC_ADD
||
775 alloctype
== VDEV_ALLOC_SPLIT
||
776 alloctype
== VDEV_ALLOC_ROOTPOOL
);
777 /* Note: metaslab_group_create() is now deferred */
780 if (vd
->vdev_ops
->vdev_op_leaf
&&
781 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
782 (void) nvlist_lookup_uint64(nv
,
783 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
785 ASSERT0(vd
->vdev_leaf_zap
);
789 * If we're a leaf vdev, try to load the DTL object and other state.
792 if (vd
->vdev_ops
->vdev_op_leaf
&&
793 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
794 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
795 if (alloctype
== VDEV_ALLOC_LOAD
) {
796 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
797 &vd
->vdev_dtl_object
);
798 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
802 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
805 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
806 &spare
) == 0 && spare
)
810 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
813 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
814 &vd
->vdev_resilver_txg
);
816 if (nvlist_exists(nv
, ZPOOL_CONFIG_RESILVER_DEFER
))
817 vdev_set_deferred_resilver(spa
, vd
);
820 * In general, when importing a pool we want to ignore the
821 * persistent fault state, as the diagnosis made on another
822 * system may not be valid in the current context. The only
823 * exception is if we forced a vdev to a persistently faulted
824 * state with 'zpool offline -f'. The persistent fault will
825 * remain across imports until cleared.
827 * Local vdevs will remain in the faulted state.
829 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
830 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
831 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
833 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
835 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
838 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
842 VDEV_AUX_ERR_EXCEEDED
;
843 if (nvlist_lookup_string(nv
,
844 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
845 strcmp(aux
, "external") == 0)
846 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
848 vd
->vdev_faulted
= 0ULL;
854 * Add ourselves to the parent's list of children.
856 vdev_add_child(parent
, vd
);
864 vdev_free(vdev_t
*vd
)
866 spa_t
*spa
= vd
->vdev_spa
;
867 ASSERT3P(vd
->vdev_initialize_thread
, ==, NULL
);
870 * Scan queues are normally destroyed at the end of a scan. If the
871 * queue exists here, that implies the vdev is being removed while
872 * the scan is still running.
874 if (vd
->vdev_scan_io_queue
!= NULL
) {
875 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
876 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
877 vd
->vdev_scan_io_queue
= NULL
;
878 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
882 * vdev_free() implies closing the vdev first. This is simpler than
883 * trying to ensure complicated semantics for all callers.
887 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
888 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
893 for (int c
= 0; c
< vd
->vdev_children
; c
++)
894 vdev_free(vd
->vdev_child
[c
]);
896 ASSERT(vd
->vdev_child
== NULL
);
897 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
898 ASSERT(vd
->vdev_initialize_thread
== NULL
);
901 * Discard allocation state.
903 if (vd
->vdev_mg
!= NULL
) {
904 vdev_metaslab_fini(vd
);
905 metaslab_group_destroy(vd
->vdev_mg
);
908 ASSERT0(vd
->vdev_stat
.vs_space
);
909 ASSERT0(vd
->vdev_stat
.vs_dspace
);
910 ASSERT0(vd
->vdev_stat
.vs_alloc
);
913 * Remove this vdev from its parent's child list.
915 vdev_remove_child(vd
->vdev_parent
, vd
);
917 ASSERT(vd
->vdev_parent
== NULL
);
920 * Clean up vdev structure.
926 spa_strfree(vd
->vdev_path
);
928 spa_strfree(vd
->vdev_devid
);
929 if (vd
->vdev_physpath
)
930 spa_strfree(vd
->vdev_physpath
);
932 if (vd
->vdev_enc_sysfs_path
)
933 spa_strfree(vd
->vdev_enc_sysfs_path
);
936 spa_strfree(vd
->vdev_fru
);
938 if (vd
->vdev_isspare
)
939 spa_spare_remove(vd
);
940 if (vd
->vdev_isl2cache
)
941 spa_l2cache_remove(vd
);
943 txg_list_destroy(&vd
->vdev_ms_list
);
944 txg_list_destroy(&vd
->vdev_dtl_list
);
946 mutex_enter(&vd
->vdev_dtl_lock
);
947 space_map_close(vd
->vdev_dtl_sm
);
948 for (int t
= 0; t
< DTL_TYPES
; t
++) {
949 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
950 range_tree_destroy(vd
->vdev_dtl
[t
]);
952 mutex_exit(&vd
->vdev_dtl_lock
);
954 EQUIV(vd
->vdev_indirect_births
!= NULL
,
955 vd
->vdev_indirect_mapping
!= NULL
);
956 if (vd
->vdev_indirect_births
!= NULL
) {
957 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
958 vdev_indirect_births_close(vd
->vdev_indirect_births
);
961 if (vd
->vdev_obsolete_sm
!= NULL
) {
962 ASSERT(vd
->vdev_removing
||
963 vd
->vdev_ops
== &vdev_indirect_ops
);
964 space_map_close(vd
->vdev_obsolete_sm
);
965 vd
->vdev_obsolete_sm
= NULL
;
967 range_tree_destroy(vd
->vdev_obsolete_segments
);
968 rw_destroy(&vd
->vdev_indirect_rwlock
);
969 mutex_destroy(&vd
->vdev_obsolete_lock
);
971 mutex_destroy(&vd
->vdev_queue_lock
);
972 mutex_destroy(&vd
->vdev_dtl_lock
);
973 mutex_destroy(&vd
->vdev_stat_lock
);
974 mutex_destroy(&vd
->vdev_probe_lock
);
975 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
976 mutex_destroy(&vd
->vdev_initialize_lock
);
977 mutex_destroy(&vd
->vdev_initialize_io_lock
);
978 cv_destroy(&vd
->vdev_initialize_io_cv
);
979 cv_destroy(&vd
->vdev_initialize_cv
);
981 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
982 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
984 if (vd
== spa
->spa_root_vdev
)
985 spa
->spa_root_vdev
= NULL
;
987 kmem_free(vd
, sizeof (vdev_t
));
991 * Transfer top-level vdev state from svd to tvd.
994 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
996 spa_t
*spa
= svd
->vdev_spa
;
1001 ASSERT(tvd
== tvd
->vdev_top
);
1003 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
1004 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
1005 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
1006 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
1007 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
1009 svd
->vdev_ms_array
= 0;
1010 svd
->vdev_ms_shift
= 0;
1011 svd
->vdev_ms_count
= 0;
1012 svd
->vdev_top_zap
= 0;
1015 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
1016 tvd
->vdev_mg
= svd
->vdev_mg
;
1017 tvd
->vdev_ms
= svd
->vdev_ms
;
1019 svd
->vdev_mg
= NULL
;
1020 svd
->vdev_ms
= NULL
;
1022 if (tvd
->vdev_mg
!= NULL
)
1023 tvd
->vdev_mg
->mg_vd
= tvd
;
1025 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
1026 svd
->vdev_checkpoint_sm
= NULL
;
1028 tvd
->vdev_alloc_bias
= svd
->vdev_alloc_bias
;
1029 svd
->vdev_alloc_bias
= VDEV_BIAS_NONE
;
1031 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
1032 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
1033 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
1035 svd
->vdev_stat
.vs_alloc
= 0;
1036 svd
->vdev_stat
.vs_space
= 0;
1037 svd
->vdev_stat
.vs_dspace
= 0;
1040 * State which may be set on a top-level vdev that's in the
1041 * process of being removed.
1043 ASSERT0(tvd
->vdev_indirect_config
.vic_births_object
);
1044 ASSERT0(tvd
->vdev_indirect_config
.vic_mapping_object
);
1045 ASSERT3U(tvd
->vdev_indirect_config
.vic_prev_indirect_vdev
, ==, -1ULL);
1046 ASSERT3P(tvd
->vdev_indirect_mapping
, ==, NULL
);
1047 ASSERT3P(tvd
->vdev_indirect_births
, ==, NULL
);
1048 ASSERT3P(tvd
->vdev_obsolete_sm
, ==, NULL
);
1049 ASSERT0(tvd
->vdev_removing
);
1050 tvd
->vdev_removing
= svd
->vdev_removing
;
1051 tvd
->vdev_indirect_config
= svd
->vdev_indirect_config
;
1052 tvd
->vdev_indirect_mapping
= svd
->vdev_indirect_mapping
;
1053 tvd
->vdev_indirect_births
= svd
->vdev_indirect_births
;
1054 range_tree_swap(&svd
->vdev_obsolete_segments
,
1055 &tvd
->vdev_obsolete_segments
);
1056 tvd
->vdev_obsolete_sm
= svd
->vdev_obsolete_sm
;
1057 svd
->vdev_indirect_config
.vic_mapping_object
= 0;
1058 svd
->vdev_indirect_config
.vic_births_object
= 0;
1059 svd
->vdev_indirect_config
.vic_prev_indirect_vdev
= -1ULL;
1060 svd
->vdev_indirect_mapping
= NULL
;
1061 svd
->vdev_indirect_births
= NULL
;
1062 svd
->vdev_obsolete_sm
= NULL
;
1063 svd
->vdev_removing
= 0;
1065 for (t
= 0; t
< TXG_SIZE
; t
++) {
1066 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
1067 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
1068 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
1069 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
1070 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
1071 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
1074 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
1075 vdev_config_clean(svd
);
1076 vdev_config_dirty(tvd
);
1079 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
1080 vdev_state_clean(svd
);
1081 vdev_state_dirty(tvd
);
1084 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
1085 svd
->vdev_deflate_ratio
= 0;
1087 tvd
->vdev_islog
= svd
->vdev_islog
;
1088 svd
->vdev_islog
= 0;
1090 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
1094 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
1101 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1102 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
1106 * Add a mirror/replacing vdev above an existing vdev.
1109 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
1111 spa_t
*spa
= cvd
->vdev_spa
;
1112 vdev_t
*pvd
= cvd
->vdev_parent
;
1115 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1117 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
1119 mvd
->vdev_asize
= cvd
->vdev_asize
;
1120 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
1121 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
1122 mvd
->vdev_psize
= cvd
->vdev_psize
;
1123 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
1124 mvd
->vdev_state
= cvd
->vdev_state
;
1125 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
1127 vdev_remove_child(pvd
, cvd
);
1128 vdev_add_child(pvd
, mvd
);
1129 cvd
->vdev_id
= mvd
->vdev_children
;
1130 vdev_add_child(mvd
, cvd
);
1131 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1133 if (mvd
== mvd
->vdev_top
)
1134 vdev_top_transfer(cvd
, mvd
);
1140 * Remove a 1-way mirror/replacing vdev from the tree.
1143 vdev_remove_parent(vdev_t
*cvd
)
1145 vdev_t
*mvd
= cvd
->vdev_parent
;
1146 vdev_t
*pvd
= mvd
->vdev_parent
;
1148 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1150 ASSERT(mvd
->vdev_children
== 1);
1151 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
1152 mvd
->vdev_ops
== &vdev_replacing_ops
||
1153 mvd
->vdev_ops
== &vdev_spare_ops
);
1154 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
1156 vdev_remove_child(mvd
, cvd
);
1157 vdev_remove_child(pvd
, mvd
);
1160 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1161 * Otherwise, we could have detached an offline device, and when we
1162 * go to import the pool we'll think we have two top-level vdevs,
1163 * instead of a different version of the same top-level vdev.
1165 if (mvd
->vdev_top
== mvd
) {
1166 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
1167 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
1168 cvd
->vdev_guid
+= guid_delta
;
1169 cvd
->vdev_guid_sum
+= guid_delta
;
1172 * If pool not set for autoexpand, we need to also preserve
1173 * mvd's asize to prevent automatic expansion of cvd.
1174 * Otherwise if we are adjusting the mirror by attaching and
1175 * detaching children of non-uniform sizes, the mirror could
1176 * autoexpand, unexpectedly requiring larger devices to
1177 * re-establish the mirror.
1179 if (!cvd
->vdev_spa
->spa_autoexpand
)
1180 cvd
->vdev_asize
= mvd
->vdev_asize
;
1182 cvd
->vdev_id
= mvd
->vdev_id
;
1183 vdev_add_child(pvd
, cvd
);
1184 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1186 if (cvd
== cvd
->vdev_top
)
1187 vdev_top_transfer(mvd
, cvd
);
1189 ASSERT(mvd
->vdev_children
== 0);
1194 vdev_metaslab_group_create(vdev_t
*vd
)
1196 spa_t
*spa
= vd
->vdev_spa
;
1199 * metaslab_group_create was delayed until allocation bias was available
1201 if (vd
->vdev_mg
== NULL
) {
1202 metaslab_class_t
*mc
;
1204 if (vd
->vdev_islog
&& vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
)
1205 vd
->vdev_alloc_bias
= VDEV_BIAS_LOG
;
1207 ASSERT3U(vd
->vdev_islog
, ==,
1208 (vd
->vdev_alloc_bias
== VDEV_BIAS_LOG
));
1210 switch (vd
->vdev_alloc_bias
) {
1212 mc
= spa_log_class(spa
);
1214 case VDEV_BIAS_SPECIAL
:
1215 mc
= spa_special_class(spa
);
1217 case VDEV_BIAS_DEDUP
:
1218 mc
= spa_dedup_class(spa
);
1221 mc
= spa_normal_class(spa
);
1224 vd
->vdev_mg
= metaslab_group_create(mc
, vd
,
1225 spa
->spa_alloc_count
);
1228 * The spa ashift values currently only reflect the
1229 * general vdev classes. Class destination is late
1230 * binding so ashift checking had to wait until now
1232 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1233 mc
== spa_normal_class(spa
) && vd
->vdev_aux
== NULL
) {
1234 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1235 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1236 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1237 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1243 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
1245 spa_t
*spa
= vd
->vdev_spa
;
1246 objset_t
*mos
= spa
->spa_meta_objset
;
1248 uint64_t oldc
= vd
->vdev_ms_count
;
1249 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
1252 boolean_t expanding
= (oldc
!= 0);
1254 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1257 * This vdev is not being allocated from yet or is a hole.
1259 if (vd
->vdev_ms_shift
== 0)
1262 ASSERT(!vd
->vdev_ishole
);
1264 ASSERT(oldc
<= newc
);
1266 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1269 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
1270 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1274 vd
->vdev_ms_count
= newc
;
1275 for (m
= oldc
; m
< newc
; m
++) {
1276 uint64_t object
= 0;
1279 * vdev_ms_array may be 0 if we are creating the "fake"
1280 * metaslabs for an indirect vdev for zdb's leak detection.
1281 * See zdb_leak_init().
1283 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1284 error
= dmu_read(mos
, vd
->vdev_ms_array
,
1285 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1288 vdev_dbgmsg(vd
, "unable to read the metaslab "
1289 "array [error=%d]", error
);
1296 * To accomodate zdb_leak_init() fake indirect
1297 * metaslabs, we allocate a metaslab group for
1298 * indirect vdevs which normally don't have one.
1300 if (vd
->vdev_mg
== NULL
) {
1301 ASSERT0(vdev_is_concrete(vd
));
1302 vdev_metaslab_group_create(vd
);
1305 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1308 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1315 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1318 * If the vdev is being removed we don't activate
1319 * the metaslabs since we want to ensure that no new
1320 * allocations are performed on this device.
1322 if (!expanding
&& !vd
->vdev_removing
) {
1323 metaslab_group_activate(vd
->vdev_mg
);
1327 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1333 vdev_metaslab_fini(vdev_t
*vd
)
1335 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1336 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1337 SPA_FEATURE_POOL_CHECKPOINT
));
1338 space_map_close(vd
->vdev_checkpoint_sm
);
1340 * Even though we close the space map, we need to set its
1341 * pointer to NULL. The reason is that vdev_metaslab_fini()
1342 * may be called multiple times for certain operations
1343 * (i.e. when destroying a pool) so we need to ensure that
1344 * this clause never executes twice. This logic is similar
1345 * to the one used for the vdev_ms clause below.
1347 vd
->vdev_checkpoint_sm
= NULL
;
1350 if (vd
->vdev_ms
!= NULL
) {
1351 uint64_t count
= vd
->vdev_ms_count
;
1353 metaslab_group_passivate(vd
->vdev_mg
);
1354 for (uint64_t m
= 0; m
< count
; m
++) {
1355 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1360 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1363 vd
->vdev_ms_count
= 0;
1365 ASSERT0(vd
->vdev_ms_count
);
1366 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1369 typedef struct vdev_probe_stats
{
1370 boolean_t vps_readable
;
1371 boolean_t vps_writeable
;
1373 } vdev_probe_stats_t
;
1376 vdev_probe_done(zio_t
*zio
)
1378 spa_t
*spa
= zio
->io_spa
;
1379 vdev_t
*vd
= zio
->io_vd
;
1380 vdev_probe_stats_t
*vps
= zio
->io_private
;
1382 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1384 if (zio
->io_type
== ZIO_TYPE_READ
) {
1385 if (zio
->io_error
== 0)
1386 vps
->vps_readable
= 1;
1387 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1388 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1389 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1390 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1391 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1393 abd_free(zio
->io_abd
);
1395 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1396 if (zio
->io_error
== 0)
1397 vps
->vps_writeable
= 1;
1398 abd_free(zio
->io_abd
);
1399 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1403 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1404 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1406 if (vdev_readable(vd
) &&
1407 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1410 ASSERT(zio
->io_error
!= 0);
1411 vdev_dbgmsg(vd
, "failed probe");
1412 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1413 spa
, vd
, NULL
, NULL
, 0, 0);
1414 zio
->io_error
= SET_ERROR(ENXIO
);
1417 mutex_enter(&vd
->vdev_probe_lock
);
1418 ASSERT(vd
->vdev_probe_zio
== zio
);
1419 vd
->vdev_probe_zio
= NULL
;
1420 mutex_exit(&vd
->vdev_probe_lock
);
1423 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1424 if (!vdev_accessible(vd
, pio
))
1425 pio
->io_error
= SET_ERROR(ENXIO
);
1427 kmem_free(vps
, sizeof (*vps
));
1432 * Determine whether this device is accessible.
1434 * Read and write to several known locations: the pad regions of each
1435 * vdev label but the first, which we leave alone in case it contains
1439 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1441 spa_t
*spa
= vd
->vdev_spa
;
1442 vdev_probe_stats_t
*vps
= NULL
;
1445 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1448 * Don't probe the probe.
1450 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1454 * To prevent 'probe storms' when a device fails, we create
1455 * just one probe i/o at a time. All zios that want to probe
1456 * this vdev will become parents of the probe io.
1458 mutex_enter(&vd
->vdev_probe_lock
);
1460 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1461 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1463 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1464 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1467 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1469 * vdev_cant_read and vdev_cant_write can only
1470 * transition from TRUE to FALSE when we have the
1471 * SCL_ZIO lock as writer; otherwise they can only
1472 * transition from FALSE to TRUE. This ensures that
1473 * any zio looking at these values can assume that
1474 * failures persist for the life of the I/O. That's
1475 * important because when a device has intermittent
1476 * connectivity problems, we want to ensure that
1477 * they're ascribed to the device (ENXIO) and not
1480 * Since we hold SCL_ZIO as writer here, clear both
1481 * values so the probe can reevaluate from first
1484 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1485 vd
->vdev_cant_read
= B_FALSE
;
1486 vd
->vdev_cant_write
= B_FALSE
;
1489 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1490 vdev_probe_done
, vps
,
1491 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1494 * We can't change the vdev state in this context, so we
1495 * kick off an async task to do it on our behalf.
1498 vd
->vdev_probe_wanted
= B_TRUE
;
1499 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1504 zio_add_child(zio
, pio
);
1506 mutex_exit(&vd
->vdev_probe_lock
);
1509 ASSERT(zio
!= NULL
);
1513 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1514 zio_nowait(zio_read_phys(pio
, vd
,
1515 vdev_label_offset(vd
->vdev_psize
, l
,
1516 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1517 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1518 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1519 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1530 vdev_open_child(void *arg
)
1534 vd
->vdev_open_thread
= curthread
;
1535 vd
->vdev_open_error
= vdev_open(vd
);
1536 vd
->vdev_open_thread
= NULL
;
1540 vdev_uses_zvols(vdev_t
*vd
)
1543 if (zvol_is_zvol(vd
->vdev_path
))
1547 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1548 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1555 vdev_open_children(vdev_t
*vd
)
1558 int children
= vd
->vdev_children
;
1561 * in order to handle pools on top of zvols, do the opens
1562 * in a single thread so that the same thread holds the
1563 * spa_namespace_lock
1565 if (vdev_uses_zvols(vd
)) {
1567 for (int c
= 0; c
< children
; c
++)
1568 vd
->vdev_child
[c
]->vdev_open_error
=
1569 vdev_open(vd
->vdev_child
[c
]);
1571 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1572 children
, children
, TASKQ_PREPOPULATE
);
1576 for (int c
= 0; c
< children
; c
++)
1577 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1578 vd
->vdev_child
[c
], TQ_SLEEP
) != TASKQID_INVALID
);
1583 vd
->vdev_nonrot
= B_TRUE
;
1585 for (int c
= 0; c
< children
; c
++)
1586 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1590 * Compute the raidz-deflation ratio. Note, we hard-code
1591 * in 128k (1 << 17) because it is the "typical" blocksize.
1592 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1593 * otherwise it would inconsistently account for existing bp's.
1596 vdev_set_deflate_ratio(vdev_t
*vd
)
1598 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1599 vd
->vdev_deflate_ratio
= (1 << 17) /
1600 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1605 * Prepare a virtual device for access.
1608 vdev_open(vdev_t
*vd
)
1610 spa_t
*spa
= vd
->vdev_spa
;
1613 uint64_t max_osize
= 0;
1614 uint64_t asize
, max_asize
, psize
;
1615 uint64_t ashift
= 0;
1617 ASSERT(vd
->vdev_open_thread
== curthread
||
1618 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1619 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1620 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1621 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1623 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1624 vd
->vdev_cant_read
= B_FALSE
;
1625 vd
->vdev_cant_write
= B_FALSE
;
1626 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1629 * If this vdev is not removed, check its fault status. If it's
1630 * faulted, bail out of the open.
1632 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1633 ASSERT(vd
->vdev_children
== 0);
1634 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1635 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1636 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1637 vd
->vdev_label_aux
);
1638 return (SET_ERROR(ENXIO
));
1639 } else if (vd
->vdev_offline
) {
1640 ASSERT(vd
->vdev_children
== 0);
1641 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1642 return (SET_ERROR(ENXIO
));
1645 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1648 * Reset the vdev_reopening flag so that we actually close
1649 * the vdev on error.
1651 vd
->vdev_reopening
= B_FALSE
;
1652 if (zio_injection_enabled
&& error
== 0)
1653 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1656 if (vd
->vdev_removed
&&
1657 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1658 vd
->vdev_removed
= B_FALSE
;
1660 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
1661 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
1662 vd
->vdev_stat
.vs_aux
);
1664 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1665 vd
->vdev_stat
.vs_aux
);
1670 vd
->vdev_removed
= B_FALSE
;
1673 * Recheck the faulted flag now that we have confirmed that
1674 * the vdev is accessible. If we're faulted, bail.
1676 if (vd
->vdev_faulted
) {
1677 ASSERT(vd
->vdev_children
== 0);
1678 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1679 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1680 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1681 vd
->vdev_label_aux
);
1682 return (SET_ERROR(ENXIO
));
1685 if (vd
->vdev_degraded
) {
1686 ASSERT(vd
->vdev_children
== 0);
1687 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1688 VDEV_AUX_ERR_EXCEEDED
);
1690 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1694 * For hole or missing vdevs we just return success.
1696 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1699 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1700 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1701 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1707 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1708 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1710 if (vd
->vdev_children
== 0) {
1711 if (osize
< SPA_MINDEVSIZE
) {
1712 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1713 VDEV_AUX_TOO_SMALL
);
1714 return (SET_ERROR(EOVERFLOW
));
1717 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1718 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1719 VDEV_LABEL_END_SIZE
);
1721 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1722 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1723 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1724 VDEV_AUX_TOO_SMALL
);
1725 return (SET_ERROR(EOVERFLOW
));
1729 max_asize
= max_osize
;
1733 * If the vdev was expanded, record this so that we can re-create the
1734 * uberblock rings in labels {2,3}, during the next sync.
1736 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
1737 vd
->vdev_copy_uberblocks
= B_TRUE
;
1739 vd
->vdev_psize
= psize
;
1742 * Make sure the allocatable size hasn't shrunk too much.
1744 if (asize
< vd
->vdev_min_asize
) {
1745 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1746 VDEV_AUX_BAD_LABEL
);
1747 return (SET_ERROR(EINVAL
));
1750 if (vd
->vdev_asize
== 0) {
1752 * This is the first-ever open, so use the computed values.
1753 * For compatibility, a different ashift can be requested.
1755 vd
->vdev_asize
= asize
;
1756 vd
->vdev_max_asize
= max_asize
;
1757 if (vd
->vdev_ashift
== 0) {
1758 vd
->vdev_ashift
= ashift
; /* use detected value */
1760 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1761 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1762 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1763 VDEV_AUX_BAD_ASHIFT
);
1764 return (SET_ERROR(EDOM
));
1768 * Detect if the alignment requirement has increased.
1769 * We don't want to make the pool unavailable, just
1770 * post an event instead.
1772 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1773 vd
->vdev_ops
->vdev_op_leaf
) {
1774 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1775 spa
, vd
, NULL
, NULL
, 0, 0);
1778 vd
->vdev_max_asize
= max_asize
;
1782 * If all children are healthy we update asize if either:
1783 * The asize has increased, due to a device expansion caused by dynamic
1784 * LUN growth or vdev replacement, and automatic expansion is enabled;
1785 * making the additional space available.
1787 * The asize has decreased, due to a device shrink usually caused by a
1788 * vdev replace with a smaller device. This ensures that calculations
1789 * based of max_asize and asize e.g. esize are always valid. It's safe
1790 * to do this as we've already validated that asize is greater than
1793 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1794 ((asize
> vd
->vdev_asize
&&
1795 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1796 (asize
< vd
->vdev_asize
)))
1797 vd
->vdev_asize
= asize
;
1799 vdev_set_min_asize(vd
);
1802 * Ensure we can issue some IO before declaring the
1803 * vdev open for business.
1805 if (vd
->vdev_ops
->vdev_op_leaf
&&
1806 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1807 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1808 VDEV_AUX_ERR_EXCEEDED
);
1813 * Track the min and max ashift values for normal data devices.
1815 * DJB - TBD these should perhaps be tracked per allocation class
1816 * (e.g. spa_min_ashift is used to round up post compression buffers)
1818 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1819 vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
&&
1820 vd
->vdev_aux
== NULL
) {
1821 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1822 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1823 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1824 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1828 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1829 * resilver. But don't do this if we are doing a reopen for a scrub,
1830 * since this would just restart the scrub we are already doing.
1832 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1833 vdev_resilver_needed(vd
, NULL
, NULL
)) {
1834 if (dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
1835 spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
))
1836 vdev_set_deferred_resilver(spa
, vd
);
1838 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1845 * Called once the vdevs are all opened, this routine validates the label
1846 * contents. This needs to be done before vdev_load() so that we don't
1847 * inadvertently do repair I/Os to the wrong device.
1849 * This function will only return failure if one of the vdevs indicates that it
1850 * has since been destroyed or exported. This is only possible if
1851 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1852 * will be updated but the function will return 0.
1855 vdev_validate(vdev_t
*vd
)
1857 spa_t
*spa
= vd
->vdev_spa
;
1859 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
1864 if (vdev_validate_skip
)
1867 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1868 if (vdev_validate(vd
->vdev_child
[c
]) != 0)
1869 return (SET_ERROR(EBADF
));
1872 * If the device has already failed, or was marked offline, don't do
1873 * any further validation. Otherwise, label I/O will fail and we will
1874 * overwrite the previous state.
1876 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
1880 * If we are performing an extreme rewind, we allow for a label that
1881 * was modified at a point after the current txg.
1882 * If config lock is not held do not check for the txg. spa_sync could
1883 * be updating the vdev's label before updating spa_last_synced_txg.
1885 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
1886 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
1889 txg
= spa_last_synced_txg(spa
);
1891 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1892 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1893 VDEV_AUX_BAD_LABEL
);
1894 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
1895 "txg %llu", (u_longlong_t
)txg
);
1900 * Determine if this vdev has been split off into another
1901 * pool. If so, then refuse to open it.
1903 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1904 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1905 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1906 VDEV_AUX_SPLIT_POOL
);
1908 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
1912 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
1913 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1914 VDEV_AUX_CORRUPT_DATA
);
1916 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1917 ZPOOL_CONFIG_POOL_GUID
);
1922 * If config is not trusted then ignore the spa guid check. This is
1923 * necessary because if the machine crashed during a re-guid the new
1924 * guid might have been written to all of the vdev labels, but not the
1925 * cached config. The check will be performed again once we have the
1926 * trusted config from the MOS.
1928 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
1929 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1930 VDEV_AUX_CORRUPT_DATA
);
1932 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
1933 "match config (%llu != %llu)", (u_longlong_t
)guid
,
1934 (u_longlong_t
)spa_guid(spa
));
1938 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1939 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1943 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
1944 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1945 VDEV_AUX_CORRUPT_DATA
);
1947 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1952 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
1954 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1955 VDEV_AUX_CORRUPT_DATA
);
1957 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1958 ZPOOL_CONFIG_TOP_GUID
);
1963 * If this vdev just became a top-level vdev because its sibling was
1964 * detached, it will have adopted the parent's vdev guid -- but the
1965 * label may or may not be on disk yet. Fortunately, either version
1966 * of the label will have the same top guid, so if we're a top-level
1967 * vdev, we can safely compare to that instead.
1968 * However, if the config comes from a cachefile that failed to update
1969 * after the detach, a top-level vdev will appear as a non top-level
1970 * vdev in the config. Also relax the constraints if we perform an
1973 * If we split this vdev off instead, then we also check the
1974 * original pool's guid. We don't want to consider the vdev
1975 * corrupt if it is partway through a split operation.
1977 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
1978 boolean_t mismatch
= B_FALSE
;
1979 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
1980 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
1983 if (vd
->vdev_guid
!= top_guid
&&
1984 vd
->vdev_top
->vdev_guid
!= guid
)
1989 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1990 VDEV_AUX_CORRUPT_DATA
);
1992 vdev_dbgmsg(vd
, "vdev_validate: config guid "
1993 "doesn't match label guid");
1994 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
1995 (u_longlong_t
)vd
->vdev_guid
,
1996 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
1997 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
1998 "aux_guid %llu", (u_longlong_t
)guid
,
1999 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
2004 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
2006 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2007 VDEV_AUX_CORRUPT_DATA
);
2009 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2010 ZPOOL_CONFIG_POOL_STATE
);
2017 * If this is a verbatim import, no need to check the
2018 * state of the pool.
2020 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
2021 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
2022 state
!= POOL_STATE_ACTIVE
) {
2023 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
2024 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
2025 return (SET_ERROR(EBADF
));
2029 * If we were able to open and validate a vdev that was
2030 * previously marked permanently unavailable, clear that state
2033 if (vd
->vdev_not_present
)
2034 vd
->vdev_not_present
= 0;
2040 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
2042 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
2043 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
2044 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2045 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2046 dvd
->vdev_path
, svd
->vdev_path
);
2047 spa_strfree(dvd
->vdev_path
);
2048 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2050 } else if (svd
->vdev_path
!= NULL
) {
2051 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2052 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2053 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
2058 * Recursively copy vdev paths from one vdev to another. Source and destination
2059 * vdev trees must have same geometry otherwise return error. Intended to copy
2060 * paths from userland config into MOS config.
2063 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
2065 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
2066 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
2067 (dvd
->vdev_ops
== &vdev_indirect_ops
))
2070 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
2071 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
2072 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
2073 return (SET_ERROR(EINVAL
));
2076 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
2077 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
2078 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
2079 (u_longlong_t
)dvd
->vdev_guid
);
2080 return (SET_ERROR(EINVAL
));
2083 if (svd
->vdev_children
!= dvd
->vdev_children
) {
2084 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
2085 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
2086 (u_longlong_t
)dvd
->vdev_children
);
2087 return (SET_ERROR(EINVAL
));
2090 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
2091 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
2092 dvd
->vdev_child
[i
]);
2097 if (svd
->vdev_ops
->vdev_op_leaf
)
2098 vdev_copy_path_impl(svd
, dvd
);
2104 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
2106 ASSERT(stvd
->vdev_top
== stvd
);
2107 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
2109 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
2110 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
2113 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
2117 * The idea here is that while a vdev can shift positions within
2118 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2119 * step outside of it.
2121 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
2123 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
2126 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2128 vdev_copy_path_impl(vd
, dvd
);
2132 * Recursively copy vdev paths from one root vdev to another. Source and
2133 * destination vdev trees may differ in geometry. For each destination leaf
2134 * vdev, search a vdev with the same guid and top vdev id in the source.
2135 * Intended to copy paths from userland config into MOS config.
2138 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
2140 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
2141 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
2142 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
2144 for (uint64_t i
= 0; i
< children
; i
++) {
2145 vdev_copy_path_search(srvd
->vdev_child
[i
],
2146 drvd
->vdev_child
[i
]);
2151 * Close a virtual device.
2154 vdev_close(vdev_t
*vd
)
2156 vdev_t
*pvd
= vd
->vdev_parent
;
2157 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
2159 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2162 * If our parent is reopening, then we are as well, unless we are
2165 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
2166 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
2168 vd
->vdev_ops
->vdev_op_close(vd
);
2170 vdev_cache_purge(vd
);
2173 * We record the previous state before we close it, so that if we are
2174 * doing a reopen(), we don't generate FMA ereports if we notice that
2175 * it's still faulted.
2177 vd
->vdev_prevstate
= vd
->vdev_state
;
2179 if (vd
->vdev_offline
)
2180 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
2182 vd
->vdev_state
= VDEV_STATE_CLOSED
;
2183 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2187 vdev_hold(vdev_t
*vd
)
2189 spa_t
*spa
= vd
->vdev_spa
;
2191 ASSERT(spa_is_root(spa
));
2192 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
2195 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2196 vdev_hold(vd
->vdev_child
[c
]);
2198 if (vd
->vdev_ops
->vdev_op_leaf
)
2199 vd
->vdev_ops
->vdev_op_hold(vd
);
2203 vdev_rele(vdev_t
*vd
)
2205 ASSERT(spa_is_root(vd
->vdev_spa
));
2206 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2207 vdev_rele(vd
->vdev_child
[c
]);
2209 if (vd
->vdev_ops
->vdev_op_leaf
)
2210 vd
->vdev_ops
->vdev_op_rele(vd
);
2214 * Reopen all interior vdevs and any unopened leaves. We don't actually
2215 * reopen leaf vdevs which had previously been opened as they might deadlock
2216 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2217 * If the leaf has never been opened then open it, as usual.
2220 vdev_reopen(vdev_t
*vd
)
2222 spa_t
*spa
= vd
->vdev_spa
;
2224 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2226 /* set the reopening flag unless we're taking the vdev offline */
2227 vd
->vdev_reopening
= !vd
->vdev_offline
;
2229 (void) vdev_open(vd
);
2232 * Call vdev_validate() here to make sure we have the same device.
2233 * Otherwise, a device with an invalid label could be successfully
2234 * opened in response to vdev_reopen().
2237 (void) vdev_validate_aux(vd
);
2238 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
2239 vd
->vdev_aux
== &spa
->spa_l2cache
&&
2240 !l2arc_vdev_present(vd
))
2241 l2arc_add_vdev(spa
, vd
);
2243 (void) vdev_validate(vd
);
2247 * Reassess parent vdev's health.
2249 vdev_propagate_state(vd
);
2253 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
2258 * Normally, partial opens (e.g. of a mirror) are allowed.
2259 * For a create, however, we want to fail the request if
2260 * there are any components we can't open.
2262 error
= vdev_open(vd
);
2264 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2266 return (error
? error
: ENXIO
);
2270 * Recursively load DTLs and initialize all labels.
2272 if ((error
= vdev_dtl_load(vd
)) != 0 ||
2273 (error
= vdev_label_init(vd
, txg
, isreplacing
?
2274 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
2283 vdev_metaslab_set_size(vdev_t
*vd
)
2285 uint64_t asize
= vd
->vdev_asize
;
2286 uint64_t ms_count
= asize
>> vdev_default_ms_shift
;
2290 * There are two dimensions to the metaslab sizing calculation:
2291 * the size of the metaslab and the count of metaslabs per vdev.
2292 * In general, we aim for vdev_max_ms_count (200) metaslabs. The
2293 * range of the dimensions are as follows:
2295 * 2^29 <= ms_size <= 2^38
2296 * 16 <= ms_count <= 131,072
2298 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2299 * at least 512MB (2^29) to minimize fragmentation effects when
2300 * testing with smaller devices. However, the count constraint
2301 * of at least 16 metaslabs will override this minimum size goal.
2303 * On the upper end of vdev sizes, we aim for a maximum metaslab
2304 * size of 256GB. However, we will cap the total count to 2^17
2305 * metaslabs to keep our memory footprint in check.
2307 * The net effect of applying above constrains is summarized below.
2309 * vdev size metaslab count
2310 * -------------|-----------------
2312 * 8GB - 100GB one per 512MB
2314 * 50TB - 32PB one per 256GB
2316 * -------------------------------
2319 if (ms_count
< vdev_min_ms_count
)
2320 ms_shift
= highbit64(asize
/ vdev_min_ms_count
);
2321 else if (ms_count
> vdev_max_ms_count
)
2322 ms_shift
= highbit64(asize
/ vdev_max_ms_count
);
2324 ms_shift
= vdev_default_ms_shift
;
2326 if (ms_shift
< SPA_MAXBLOCKSHIFT
) {
2327 ms_shift
= SPA_MAXBLOCKSHIFT
;
2328 } else if (ms_shift
> vdev_max_ms_shift
) {
2329 ms_shift
= vdev_max_ms_shift
;
2330 /* cap the total count to constrain memory footprint */
2331 if ((asize
>> ms_shift
) > vdev_ms_count_limit
)
2332 ms_shift
= highbit64(asize
/ vdev_ms_count_limit
);
2335 vd
->vdev_ms_shift
= ms_shift
;
2336 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2340 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2342 ASSERT(vd
== vd
->vdev_top
);
2343 /* indirect vdevs don't have metaslabs or dtls */
2344 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2345 ASSERT(ISP2(flags
));
2346 ASSERT(spa_writeable(vd
->vdev_spa
));
2348 if (flags
& VDD_METASLAB
)
2349 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2351 if (flags
& VDD_DTL
)
2352 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2354 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2358 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2360 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2361 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2363 if (vd
->vdev_ops
->vdev_op_leaf
)
2364 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2370 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2371 * the vdev has less than perfect replication. There are four kinds of DTL:
2373 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2375 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2377 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2378 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2379 * txgs that was scrubbed.
2381 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2382 * persistent errors or just some device being offline.
2383 * Unlike the other three, the DTL_OUTAGE map is not generally
2384 * maintained; it's only computed when needed, typically to
2385 * determine whether a device can be detached.
2387 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2388 * either has the data or it doesn't.
2390 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2391 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2392 * if any child is less than fully replicated, then so is its parent.
2393 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2394 * comprising only those txgs which appear in 'maxfaults' or more children;
2395 * those are the txgs we don't have enough replication to read. For example,
2396 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2397 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2398 * two child DTL_MISSING maps.
2400 * It should be clear from the above that to compute the DTLs and outage maps
2401 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2402 * Therefore, that is all we keep on disk. When loading the pool, or after
2403 * a configuration change, we generate all other DTLs from first principles.
2406 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2408 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2410 ASSERT(t
< DTL_TYPES
);
2411 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2412 ASSERT(spa_writeable(vd
->vdev_spa
));
2414 mutex_enter(&vd
->vdev_dtl_lock
);
2415 if (!range_tree_contains(rt
, txg
, size
))
2416 range_tree_add(rt
, txg
, size
);
2417 mutex_exit(&vd
->vdev_dtl_lock
);
2421 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2423 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2424 boolean_t dirty
= B_FALSE
;
2426 ASSERT(t
< DTL_TYPES
);
2427 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2430 * While we are loading the pool, the DTLs have not been loaded yet.
2431 * Ignore the DTLs and try all devices. This avoids a recursive
2432 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2433 * when loading the pool (relying on the checksum to ensure that
2434 * we get the right data -- note that we while loading, we are
2435 * only reading the MOS, which is always checksummed).
2437 if (vd
->vdev_spa
->spa_load_state
!= SPA_LOAD_NONE
)
2440 mutex_enter(&vd
->vdev_dtl_lock
);
2441 if (!range_tree_is_empty(rt
))
2442 dirty
= range_tree_contains(rt
, txg
, size
);
2443 mutex_exit(&vd
->vdev_dtl_lock
);
2449 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2451 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2454 mutex_enter(&vd
->vdev_dtl_lock
);
2455 empty
= range_tree_is_empty(rt
);
2456 mutex_exit(&vd
->vdev_dtl_lock
);
2462 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2465 vdev_dtl_need_resilver(vdev_t
*vd
, uint64_t offset
, size_t psize
)
2467 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2469 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
2470 vd
->vdev_ops
->vdev_op_leaf
)
2473 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, offset
, psize
));
2477 * Returns the lowest txg in the DTL range.
2480 vdev_dtl_min(vdev_t
*vd
)
2484 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2485 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2486 ASSERT0(vd
->vdev_children
);
2488 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2489 return (rs
->rs_start
- 1);
2493 * Returns the highest txg in the DTL.
2496 vdev_dtl_max(vdev_t
*vd
)
2500 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2501 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2502 ASSERT0(vd
->vdev_children
);
2504 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2505 return (rs
->rs_end
);
2509 * Determine if a resilvering vdev should remove any DTL entries from
2510 * its range. If the vdev was resilvering for the entire duration of the
2511 * scan then it should excise that range from its DTLs. Otherwise, this
2512 * vdev is considered partially resilvered and should leave its DTL
2513 * entries intact. The comment in vdev_dtl_reassess() describes how we
2517 vdev_dtl_should_excise(vdev_t
*vd
)
2519 spa_t
*spa
= vd
->vdev_spa
;
2520 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2522 ASSERT0(scn
->scn_phys
.scn_errors
);
2523 ASSERT0(vd
->vdev_children
);
2525 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
2528 if (vd
->vdev_resilver_deferred
)
2531 if (vd
->vdev_resilver_txg
== 0 ||
2532 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
2536 * When a resilver is initiated the scan will assign the scn_max_txg
2537 * value to the highest txg value that exists in all DTLs. If this
2538 * device's max DTL is not part of this scan (i.e. it is not in
2539 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2542 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
2543 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
2544 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
2545 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
2552 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
2553 * write operations will be issued to the pool.
2556 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
2558 spa_t
*spa
= vd
->vdev_spa
;
2562 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2564 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2565 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
2566 scrub_txg
, scrub_done
);
2568 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
2571 if (vd
->vdev_ops
->vdev_op_leaf
) {
2572 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2574 mutex_enter(&vd
->vdev_dtl_lock
);
2577 * If requested, pretend the scan completed cleanly.
2579 if (zfs_scan_ignore_errors
&& scn
)
2580 scn
->scn_phys
.scn_errors
= 0;
2583 * If we've completed a scan cleanly then determine
2584 * if this vdev should remove any DTLs. We only want to
2585 * excise regions on vdevs that were available during
2586 * the entire duration of this scan.
2588 if (scrub_txg
!= 0 &&
2589 (spa
->spa_scrub_started
||
2590 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
2591 vdev_dtl_should_excise(vd
)) {
2593 * We completed a scrub up to scrub_txg. If we
2594 * did it without rebooting, then the scrub dtl
2595 * will be valid, so excise the old region and
2596 * fold in the scrub dtl. Otherwise, leave the
2597 * dtl as-is if there was an error.
2599 * There's little trick here: to excise the beginning
2600 * of the DTL_MISSING map, we put it into a reference
2601 * tree and then add a segment with refcnt -1 that
2602 * covers the range [0, scrub_txg). This means
2603 * that each txg in that range has refcnt -1 or 0.
2604 * We then add DTL_SCRUB with a refcnt of 2, so that
2605 * entries in the range [0, scrub_txg) will have a
2606 * positive refcnt -- either 1 or 2. We then convert
2607 * the reference tree into the new DTL_MISSING map.
2609 space_reftree_create(&reftree
);
2610 space_reftree_add_map(&reftree
,
2611 vd
->vdev_dtl
[DTL_MISSING
], 1);
2612 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
2613 space_reftree_add_map(&reftree
,
2614 vd
->vdev_dtl
[DTL_SCRUB
], 2);
2615 space_reftree_generate_map(&reftree
,
2616 vd
->vdev_dtl
[DTL_MISSING
], 1);
2617 space_reftree_destroy(&reftree
);
2619 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
2620 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2621 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
2623 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
2624 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
2625 if (!vdev_readable(vd
))
2626 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
2628 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2629 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
2632 * If the vdev was resilvering and no longer has any
2633 * DTLs then reset its resilvering flag and dirty
2634 * the top level so that we persist the change.
2636 if (txg
!= 0 && vd
->vdev_resilver_txg
!= 0 &&
2637 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2638 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
])) {
2639 vd
->vdev_resilver_txg
= 0;
2640 vdev_config_dirty(vd
->vdev_top
);
2643 mutex_exit(&vd
->vdev_dtl_lock
);
2646 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
2650 mutex_enter(&vd
->vdev_dtl_lock
);
2651 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2652 /* account for child's outage in parent's missing map */
2653 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
2655 continue; /* leaf vdevs only */
2656 if (t
== DTL_PARTIAL
)
2657 minref
= 1; /* i.e. non-zero */
2658 else if (vd
->vdev_nparity
!= 0)
2659 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
2661 minref
= vd
->vdev_children
; /* any kind of mirror */
2662 space_reftree_create(&reftree
);
2663 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2664 vdev_t
*cvd
= vd
->vdev_child
[c
];
2665 mutex_enter(&cvd
->vdev_dtl_lock
);
2666 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
2667 mutex_exit(&cvd
->vdev_dtl_lock
);
2669 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
2670 space_reftree_destroy(&reftree
);
2672 mutex_exit(&vd
->vdev_dtl_lock
);
2676 vdev_dtl_load(vdev_t
*vd
)
2678 spa_t
*spa
= vd
->vdev_spa
;
2679 objset_t
*mos
= spa
->spa_meta_objset
;
2682 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2683 ASSERT(vdev_is_concrete(vd
));
2685 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2686 vd
->vdev_dtl_object
, 0, -1ULL, 0);
2689 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2691 mutex_enter(&vd
->vdev_dtl_lock
);
2694 * Now that we've opened the space_map we need to update
2697 space_map_update(vd
->vdev_dtl_sm
);
2699 error
= space_map_load(vd
->vdev_dtl_sm
,
2700 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2701 mutex_exit(&vd
->vdev_dtl_lock
);
2706 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2707 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2716 vdev_zap_allocation_data(vdev_t
*vd
, dmu_tx_t
*tx
)
2718 spa_t
*spa
= vd
->vdev_spa
;
2719 objset_t
*mos
= spa
->spa_meta_objset
;
2720 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
2723 ASSERT(alloc_bias
!= VDEV_BIAS_NONE
);
2726 (alloc_bias
== VDEV_BIAS_LOG
) ? VDEV_ALLOC_BIAS_LOG
:
2727 (alloc_bias
== VDEV_BIAS_SPECIAL
) ? VDEV_ALLOC_BIAS_SPECIAL
:
2728 (alloc_bias
== VDEV_BIAS_DEDUP
) ? VDEV_ALLOC_BIAS_DEDUP
: NULL
;
2730 ASSERT(string
!= NULL
);
2731 VERIFY0(zap_add(mos
, vd
->vdev_top_zap
, VDEV_TOP_ZAP_ALLOCATION_BIAS
,
2732 1, strlen(string
) + 1, string
, tx
));
2734 if (alloc_bias
== VDEV_BIAS_SPECIAL
|| alloc_bias
== VDEV_BIAS_DEDUP
) {
2735 spa_activate_allocation_classes(spa
, tx
);
2740 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2742 spa_t
*spa
= vd
->vdev_spa
;
2744 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2745 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2750 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2752 spa_t
*spa
= vd
->vdev_spa
;
2753 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2754 DMU_OT_NONE
, 0, tx
);
2757 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2764 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2766 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2767 vd
->vdev_ops
!= &vdev_missing_ops
&&
2768 vd
->vdev_ops
!= &vdev_root_ops
&&
2769 !vd
->vdev_top
->vdev_removing
) {
2770 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2771 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2773 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2774 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2775 if (vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
)
2776 vdev_zap_allocation_data(vd
, tx
);
2780 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2781 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2786 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2788 spa_t
*spa
= vd
->vdev_spa
;
2789 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2790 objset_t
*mos
= spa
->spa_meta_objset
;
2791 range_tree_t
*rtsync
;
2793 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2795 ASSERT(vdev_is_concrete(vd
));
2796 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2798 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2800 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2801 mutex_enter(&vd
->vdev_dtl_lock
);
2802 space_map_free(vd
->vdev_dtl_sm
, tx
);
2803 space_map_close(vd
->vdev_dtl_sm
);
2804 vd
->vdev_dtl_sm
= NULL
;
2805 mutex_exit(&vd
->vdev_dtl_lock
);
2808 * We only destroy the leaf ZAP for detached leaves or for
2809 * removed log devices. Removed data devices handle leaf ZAP
2810 * cleanup later, once cancellation is no longer possible.
2812 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2813 vd
->vdev_top
->vdev_islog
)) {
2814 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2815 vd
->vdev_leaf_zap
= 0;
2822 if (vd
->vdev_dtl_sm
== NULL
) {
2823 uint64_t new_object
;
2825 new_object
= space_map_alloc(mos
, vdev_dtl_sm_blksz
, tx
);
2826 VERIFY3U(new_object
, !=, 0);
2828 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2830 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2833 rtsync
= range_tree_create(NULL
, NULL
);
2835 mutex_enter(&vd
->vdev_dtl_lock
);
2836 range_tree_walk(rt
, range_tree_add
, rtsync
);
2837 mutex_exit(&vd
->vdev_dtl_lock
);
2839 space_map_truncate(vd
->vdev_dtl_sm
, vdev_dtl_sm_blksz
, tx
);
2840 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
2841 range_tree_vacate(rtsync
, NULL
, NULL
);
2843 range_tree_destroy(rtsync
);
2846 * If the object for the space map has changed then dirty
2847 * the top level so that we update the config.
2849 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2850 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
2851 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
2852 (u_longlong_t
)object
,
2853 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
2854 vdev_config_dirty(vd
->vdev_top
);
2859 mutex_enter(&vd
->vdev_dtl_lock
);
2860 space_map_update(vd
->vdev_dtl_sm
);
2861 mutex_exit(&vd
->vdev_dtl_lock
);
2865 * Determine whether the specified vdev can be offlined/detached/removed
2866 * without losing data.
2869 vdev_dtl_required(vdev_t
*vd
)
2871 spa_t
*spa
= vd
->vdev_spa
;
2872 vdev_t
*tvd
= vd
->vdev_top
;
2873 uint8_t cant_read
= vd
->vdev_cant_read
;
2876 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2878 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2882 * Temporarily mark the device as unreadable, and then determine
2883 * whether this results in any DTL outages in the top-level vdev.
2884 * If not, we can safely offline/detach/remove the device.
2886 vd
->vdev_cant_read
= B_TRUE
;
2887 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2888 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2889 vd
->vdev_cant_read
= cant_read
;
2890 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2892 if (!required
&& zio_injection_enabled
)
2893 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2899 * Determine if resilver is needed, and if so the txg range.
2902 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2904 boolean_t needed
= B_FALSE
;
2905 uint64_t thismin
= UINT64_MAX
;
2906 uint64_t thismax
= 0;
2908 if (vd
->vdev_children
== 0) {
2909 mutex_enter(&vd
->vdev_dtl_lock
);
2910 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2911 vdev_writeable(vd
)) {
2913 thismin
= vdev_dtl_min(vd
);
2914 thismax
= vdev_dtl_max(vd
);
2917 mutex_exit(&vd
->vdev_dtl_lock
);
2919 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2920 vdev_t
*cvd
= vd
->vdev_child
[c
];
2921 uint64_t cmin
, cmax
;
2923 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2924 thismin
= MIN(thismin
, cmin
);
2925 thismax
= MAX(thismax
, cmax
);
2931 if (needed
&& minp
) {
2939 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
2940 * will contain either the checkpoint spacemap object or zero if none exists.
2941 * All other errors are returned to the caller.
2944 vdev_checkpoint_sm_object(vdev_t
*vd
, uint64_t *sm_obj
)
2946 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
2948 if (vd
->vdev_top_zap
== 0) {
2953 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
2954 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, sm_obj
);
2955 if (error
== ENOENT
) {
2964 vdev_load(vdev_t
*vd
)
2969 * Recursively load all children.
2971 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2972 error
= vdev_load(vd
->vdev_child
[c
]);
2978 vdev_set_deflate_ratio(vd
);
2981 * On spa_load path, grab the allocation bias from our zap
2983 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
2984 spa_t
*spa
= vd
->vdev_spa
;
2987 if (zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
2988 VDEV_TOP_ZAP_ALLOCATION_BIAS
, 1, sizeof (bias_str
),
2990 ASSERT(vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
);
2991 vd
->vdev_alloc_bias
= vdev_derive_alloc_bias(bias_str
);
2996 * If this is a top-level vdev, initialize its metaslabs.
2998 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
2999 vdev_metaslab_group_create(vd
);
3001 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
3002 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3003 VDEV_AUX_CORRUPT_DATA
);
3004 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
3005 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
3006 (u_longlong_t
)vd
->vdev_asize
);
3007 return (SET_ERROR(ENXIO
));
3008 } else if ((error
= vdev_metaslab_init(vd
, 0)) != 0) {
3009 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
3010 "[error=%d]", error
);
3011 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3012 VDEV_AUX_CORRUPT_DATA
);
3016 uint64_t checkpoint_sm_obj
;
3017 error
= vdev_checkpoint_sm_object(vd
, &checkpoint_sm_obj
);
3018 if (error
== 0 && checkpoint_sm_obj
!= 0) {
3019 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3020 ASSERT(vd
->vdev_asize
!= 0);
3021 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
3023 if ((error
= space_map_open(&vd
->vdev_checkpoint_sm
,
3024 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
3025 vd
->vdev_ashift
))) {
3026 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
3027 "failed for checkpoint spacemap (obj %llu) "
3029 (u_longlong_t
)checkpoint_sm_obj
, error
);
3032 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
3033 space_map_update(vd
->vdev_checkpoint_sm
);
3036 * Since the checkpoint_sm contains free entries
3037 * exclusively we can use sm_alloc to indicate the
3038 * cumulative checkpointed space that has been freed.
3040 vd
->vdev_stat
.vs_checkpoint_space
=
3041 -vd
->vdev_checkpoint_sm
->sm_alloc
;
3042 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
3043 vd
->vdev_stat
.vs_checkpoint_space
;
3044 } else if (error
!= 0) {
3045 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve "
3046 "checkpoint space map object from vdev ZAP "
3047 "[error=%d]", error
);
3053 * If this is a leaf vdev, load its DTL.
3055 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
3056 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3057 VDEV_AUX_CORRUPT_DATA
);
3058 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
3059 "[error=%d]", error
);
3063 uint64_t obsolete_sm_object
;
3064 error
= vdev_obsolete_sm_object(vd
, &obsolete_sm_object
);
3065 if (error
== 0 && obsolete_sm_object
!= 0) {
3066 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3067 ASSERT(vd
->vdev_asize
!= 0);
3068 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
3070 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
3071 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
3072 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3073 VDEV_AUX_CORRUPT_DATA
);
3074 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
3075 "obsolete spacemap (obj %llu) [error=%d]",
3076 (u_longlong_t
)obsolete_sm_object
, error
);
3079 space_map_update(vd
->vdev_obsolete_sm
);
3080 } else if (error
!= 0) {
3081 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve obsolete "
3082 "space map object from vdev ZAP [error=%d]", error
);
3090 * The special vdev case is used for hot spares and l2cache devices. Its
3091 * sole purpose it to set the vdev state for the associated vdev. To do this,
3092 * we make sure that we can open the underlying device, then try to read the
3093 * label, and make sure that the label is sane and that it hasn't been
3094 * repurposed to another pool.
3097 vdev_validate_aux(vdev_t
*vd
)
3100 uint64_t guid
, version
;
3103 if (!vdev_readable(vd
))
3106 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
3107 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3108 VDEV_AUX_CORRUPT_DATA
);
3112 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
3113 !SPA_VERSION_IS_SUPPORTED(version
) ||
3114 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
3115 guid
!= vd
->vdev_guid
||
3116 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
3117 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3118 VDEV_AUX_CORRUPT_DATA
);
3124 * We don't actually check the pool state here. If it's in fact in
3125 * use by another pool, we update this fact on the fly when requested.
3132 * Free the objects used to store this vdev's spacemaps, and the array
3133 * that points to them.
3136 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3138 if (vd
->vdev_ms_array
== 0)
3141 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3142 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
3143 size_t array_bytes
= array_count
* sizeof (uint64_t);
3144 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
3145 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
3146 array_bytes
, smobj_array
, 0));
3148 for (uint64_t i
= 0; i
< array_count
; i
++) {
3149 uint64_t smobj
= smobj_array
[i
];
3153 space_map_free_obj(mos
, smobj
, tx
);
3156 kmem_free(smobj_array
, array_bytes
);
3157 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
3158 vd
->vdev_ms_array
= 0;
3162 vdev_remove_empty_log(vdev_t
*vd
, uint64_t txg
)
3164 spa_t
*spa
= vd
->vdev_spa
;
3166 ASSERT(vd
->vdev_islog
);
3167 ASSERT(vd
== vd
->vdev_top
);
3168 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
3170 if (vd
->vdev_ms
!= NULL
) {
3171 metaslab_group_t
*mg
= vd
->vdev_mg
;
3173 metaslab_group_histogram_verify(mg
);
3174 metaslab_class_histogram_verify(mg
->mg_class
);
3176 for (int m
= 0; m
< vd
->vdev_ms_count
; m
++) {
3177 metaslab_t
*msp
= vd
->vdev_ms
[m
];
3179 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
3182 mutex_enter(&msp
->ms_lock
);
3184 * If the metaslab was not loaded when the vdev
3185 * was removed then the histogram accounting may
3186 * not be accurate. Update the histogram information
3187 * here so that we ensure that the metaslab group
3188 * and metaslab class are up-to-date.
3190 metaslab_group_histogram_remove(mg
, msp
);
3192 VERIFY0(space_map_allocated(msp
->ms_sm
));
3193 space_map_close(msp
->ms_sm
);
3195 mutex_exit(&msp
->ms_lock
);
3198 if (vd
->vdev_checkpoint_sm
!= NULL
) {
3199 ASSERT(spa_has_checkpoint(spa
));
3200 space_map_close(vd
->vdev_checkpoint_sm
);
3201 vd
->vdev_checkpoint_sm
= NULL
;
3204 metaslab_group_histogram_verify(mg
);
3205 metaslab_class_histogram_verify(mg
->mg_class
);
3207 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
3208 ASSERT0(mg
->mg_histogram
[i
]);
3211 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
3213 vdev_destroy_spacemaps(vd
, tx
);
3214 if (vd
->vdev_top_zap
!= 0) {
3215 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
3216 vd
->vdev_top_zap
= 0;
3223 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
3226 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
3228 ASSERT(vdev_is_concrete(vd
));
3230 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
3232 metaslab_sync_done(msp
, txg
);
3235 metaslab_sync_reassess(vd
->vdev_mg
);
3239 vdev_sync(vdev_t
*vd
, uint64_t txg
)
3241 spa_t
*spa
= vd
->vdev_spa
;
3246 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
3249 ASSERT(vd
->vdev_removing
||
3250 vd
->vdev_ops
== &vdev_indirect_ops
);
3252 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3253 vdev_indirect_sync_obsolete(vd
, tx
);
3257 * If the vdev is indirect, it can't have dirty
3258 * metaslabs or DTLs.
3260 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
3261 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
3262 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
3267 ASSERT(vdev_is_concrete(vd
));
3269 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
3270 !vd
->vdev_removing
) {
3271 ASSERT(vd
== vd
->vdev_top
);
3272 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
3273 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3274 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
3275 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
3276 ASSERT(vd
->vdev_ms_array
!= 0);
3277 vdev_config_dirty(vd
);
3281 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
3282 metaslab_sync(msp
, txg
);
3283 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
3286 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
3287 vdev_dtl_sync(lvd
, txg
);
3290 * If this is an empty log device being removed, destroy the
3291 * metadata associated with it.
3293 if (vd
->vdev_islog
&& vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
3294 vdev_remove_empty_log(vd
, txg
);
3296 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
3300 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
3302 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
3306 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3307 * not be opened, and no I/O is attempted.
3310 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3314 spa_vdev_state_enter(spa
, SCL_NONE
);
3316 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3317 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3319 if (!vd
->vdev_ops
->vdev_op_leaf
)
3320 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3325 * If user did a 'zpool offline -f' then make the fault persist across
3328 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
3330 * There are two kinds of forced faults: temporary and
3331 * persistent. Temporary faults go away at pool import, while
3332 * persistent faults stay set. Both types of faults can be
3333 * cleared with a zpool clear.
3335 * We tell if a vdev is persistently faulted by looking at the
3336 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3337 * import then it's a persistent fault. Otherwise, it's
3338 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3339 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3340 * tells vdev_config_generate() (which gets run later) to set
3341 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3343 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
3344 vd
->vdev_tmpoffline
= B_FALSE
;
3345 aux
= VDEV_AUX_EXTERNAL
;
3347 vd
->vdev_tmpoffline
= B_TRUE
;
3351 * We don't directly use the aux state here, but if we do a
3352 * vdev_reopen(), we need this value to be present to remember why we
3355 vd
->vdev_label_aux
= aux
;
3358 * Faulted state takes precedence over degraded.
3360 vd
->vdev_delayed_close
= B_FALSE
;
3361 vd
->vdev_faulted
= 1ULL;
3362 vd
->vdev_degraded
= 0ULL;
3363 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
3366 * If this device has the only valid copy of the data, then
3367 * back off and simply mark the vdev as degraded instead.
3369 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
3370 vd
->vdev_degraded
= 1ULL;
3371 vd
->vdev_faulted
= 0ULL;
3374 * If we reopen the device and it's not dead, only then do we
3379 if (vdev_readable(vd
))
3380 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
3383 return (spa_vdev_state_exit(spa
, vd
, 0));
3387 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3388 * user that something is wrong. The vdev continues to operate as normal as far
3389 * as I/O is concerned.
3392 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3396 spa_vdev_state_enter(spa
, SCL_NONE
);
3398 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3399 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3401 if (!vd
->vdev_ops
->vdev_op_leaf
)
3402 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3405 * If the vdev is already faulted, then don't do anything.
3407 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
3408 return (spa_vdev_state_exit(spa
, NULL
, 0));
3410 vd
->vdev_degraded
= 1ULL;
3411 if (!vdev_is_dead(vd
))
3412 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
3415 return (spa_vdev_state_exit(spa
, vd
, 0));
3419 * Online the given vdev.
3421 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3422 * spare device should be detached when the device finishes resilvering.
3423 * Second, the online should be treated like a 'test' online case, so no FMA
3424 * events are generated if the device fails to open.
3427 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
3429 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
3430 boolean_t wasoffline
;
3431 vdev_state_t oldstate
;
3433 spa_vdev_state_enter(spa
, SCL_NONE
);
3435 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3436 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3438 if (!vd
->vdev_ops
->vdev_op_leaf
)
3439 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3441 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
3442 oldstate
= vd
->vdev_state
;
3445 vd
->vdev_offline
= B_FALSE
;
3446 vd
->vdev_tmpoffline
= B_FALSE
;
3447 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
3448 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
3450 /* XXX - L2ARC 1.0 does not support expansion */
3451 if (!vd
->vdev_aux
) {
3452 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3453 pvd
->vdev_expanding
= !!((flags
& ZFS_ONLINE_EXPAND
) ||
3454 spa
->spa_autoexpand
);
3455 vd
->vdev_expansion_time
= gethrestime_sec();
3459 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
3461 if (!vd
->vdev_aux
) {
3462 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3463 pvd
->vdev_expanding
= B_FALSE
;
3467 *newstate
= vd
->vdev_state
;
3468 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
3469 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
3470 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3471 vd
->vdev_parent
->vdev_child
[0] == vd
)
3472 vd
->vdev_unspare
= B_TRUE
;
3474 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
3476 /* XXX - L2ARC 1.0 does not support expansion */
3478 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
3479 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3482 /* Restart initializing if necessary */
3483 mutex_enter(&vd
->vdev_initialize_lock
);
3484 if (vdev_writeable(vd
) &&
3485 vd
->vdev_initialize_thread
== NULL
&&
3486 vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) {
3487 (void) vdev_initialize(vd
);
3489 mutex_exit(&vd
->vdev_initialize_lock
);
3492 (oldstate
< VDEV_STATE_DEGRADED
&&
3493 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
3494 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
3496 return (spa_vdev_state_exit(spa
, vd
, 0));
3500 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3504 uint64_t generation
;
3505 metaslab_group_t
*mg
;
3508 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3510 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3511 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3513 if (!vd
->vdev_ops
->vdev_op_leaf
)
3514 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3518 generation
= spa
->spa_config_generation
+ 1;
3521 * If the device isn't already offline, try to offline it.
3523 if (!vd
->vdev_offline
) {
3525 * If this device has the only valid copy of some data,
3526 * don't allow it to be offlined. Log devices are always
3529 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3530 vdev_dtl_required(vd
))
3531 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3534 * If the top-level is a slog and it has had allocations
3535 * then proceed. We check that the vdev's metaslab group
3536 * is not NULL since it's possible that we may have just
3537 * added this vdev but not yet initialized its metaslabs.
3539 if (tvd
->vdev_islog
&& mg
!= NULL
) {
3541 * Prevent any future allocations.
3543 metaslab_group_passivate(mg
);
3544 (void) spa_vdev_state_exit(spa
, vd
, 0);
3546 error
= spa_reset_logs(spa
);
3549 * If the log device was successfully reset but has
3550 * checkpointed data, do not offline it.
3553 tvd
->vdev_checkpoint_sm
!= NULL
) {
3554 ASSERT3U(tvd
->vdev_checkpoint_sm
->sm_alloc
,
3556 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
3559 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3562 * Check to see if the config has changed.
3564 if (error
|| generation
!= spa
->spa_config_generation
) {
3565 metaslab_group_activate(mg
);
3567 return (spa_vdev_state_exit(spa
,
3569 (void) spa_vdev_state_exit(spa
, vd
, 0);
3572 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
3576 * Offline this device and reopen its top-level vdev.
3577 * If the top-level vdev is a log device then just offline
3578 * it. Otherwise, if this action results in the top-level
3579 * vdev becoming unusable, undo it and fail the request.
3581 vd
->vdev_offline
= B_TRUE
;
3584 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3585 vdev_is_dead(tvd
)) {
3586 vd
->vdev_offline
= B_FALSE
;
3588 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3592 * Add the device back into the metaslab rotor so that
3593 * once we online the device it's open for business.
3595 if (tvd
->vdev_islog
&& mg
!= NULL
)
3596 metaslab_group_activate(mg
);
3599 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
3601 return (spa_vdev_state_exit(spa
, vd
, 0));
3605 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3609 mutex_enter(&spa
->spa_vdev_top_lock
);
3610 error
= vdev_offline_locked(spa
, guid
, flags
);
3611 mutex_exit(&spa
->spa_vdev_top_lock
);
3617 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3618 * vdev_offline(), we assume the spa config is locked. We also clear all
3619 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3622 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
3624 vdev_t
*rvd
= spa
->spa_root_vdev
;
3626 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3631 vd
->vdev_stat
.vs_read_errors
= 0;
3632 vd
->vdev_stat
.vs_write_errors
= 0;
3633 vd
->vdev_stat
.vs_checksum_errors
= 0;
3634 vd
->vdev_stat
.vs_slow_ios
= 0;
3636 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3637 vdev_clear(spa
, vd
->vdev_child
[c
]);
3640 * It makes no sense to "clear" an indirect vdev.
3642 if (!vdev_is_concrete(vd
))
3646 * If we're in the FAULTED state or have experienced failed I/O, then
3647 * clear the persistent state and attempt to reopen the device. We
3648 * also mark the vdev config dirty, so that the new faulted state is
3649 * written out to disk.
3651 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
3652 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
3654 * When reopening in response to a clear event, it may be due to
3655 * a fmadm repair request. In this case, if the device is
3656 * still broken, we want to still post the ereport again.
3658 vd
->vdev_forcefault
= B_TRUE
;
3660 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
3661 vd
->vdev_cant_read
= B_FALSE
;
3662 vd
->vdev_cant_write
= B_FALSE
;
3663 vd
->vdev_stat
.vs_aux
= 0;
3665 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
3667 vd
->vdev_forcefault
= B_FALSE
;
3669 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
3670 vdev_state_dirty(vd
->vdev_top
);
3672 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
)) {
3673 if (dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3674 spa_feature_is_enabled(spa
,
3675 SPA_FEATURE_RESILVER_DEFER
))
3676 vdev_set_deferred_resilver(spa
, vd
);
3678 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3681 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
3685 * When clearing a FMA-diagnosed fault, we always want to
3686 * unspare the device, as we assume that the original spare was
3687 * done in response to the FMA fault.
3689 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
3690 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3691 vd
->vdev_parent
->vdev_child
[0] == vd
)
3692 vd
->vdev_unspare
= B_TRUE
;
3696 vdev_is_dead(vdev_t
*vd
)
3699 * Holes and missing devices are always considered "dead".
3700 * This simplifies the code since we don't have to check for
3701 * these types of devices in the various code paths.
3702 * Instead we rely on the fact that we skip over dead devices
3703 * before issuing I/O to them.
3705 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
3706 vd
->vdev_ops
== &vdev_hole_ops
||
3707 vd
->vdev_ops
== &vdev_missing_ops
);
3711 vdev_readable(vdev_t
*vd
)
3713 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
3717 vdev_writeable(vdev_t
*vd
)
3719 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
3720 vdev_is_concrete(vd
));
3724 vdev_allocatable(vdev_t
*vd
)
3726 uint64_t state
= vd
->vdev_state
;
3729 * We currently allow allocations from vdevs which may be in the
3730 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3731 * fails to reopen then we'll catch it later when we're holding
3732 * the proper locks. Note that we have to get the vdev state
3733 * in a local variable because although it changes atomically,
3734 * we're asking two separate questions about it.
3736 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
3737 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
3738 vd
->vdev_mg
->mg_initialized
);
3742 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
3744 ASSERT(zio
->io_vd
== vd
);
3746 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
3749 if (zio
->io_type
== ZIO_TYPE_READ
)
3750 return (!vd
->vdev_cant_read
);
3752 if (zio
->io_type
== ZIO_TYPE_WRITE
)
3753 return (!vd
->vdev_cant_write
);
3759 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
3762 for (t
= 0; t
< ZIO_TYPES
; t
++) {
3763 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
3764 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
3767 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
3771 * Get extended stats
3774 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
3777 for (t
= 0; t
< ZIO_TYPES
; t
++) {
3778 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
3779 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
3781 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
3782 vsx
->vsx_total_histo
[t
][b
] +=
3783 cvsx
->vsx_total_histo
[t
][b
];
3787 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
3788 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
3789 vsx
->vsx_queue_histo
[t
][b
] +=
3790 cvsx
->vsx_queue_histo
[t
][b
];
3792 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
3793 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
3795 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
3796 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
3798 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
3799 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
3805 vdev_is_spacemap_addressable(vdev_t
*vd
)
3808 * Assuming 47 bits of the space map entry dedicated for the entry's
3809 * offset (see description in space_map.h), we calculate the maximum
3810 * address that can be described by a space map entry for the given
3813 uint64_t shift
= vd
->vdev_ashift
+ 47;
3815 if (shift
>= 63) /* detect potential overflow */
3818 return (vd
->vdev_asize
< (1ULL << shift
));
3822 * Get statistics for the given vdev.
3825 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
3829 * If we're getting stats on the root vdev, aggregate the I/O counts
3830 * over all top-level vdevs (i.e. the direct children of the root).
3832 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3834 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
3835 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
3838 memset(vsx
, 0, sizeof (*vsx
));
3840 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3841 vdev_t
*cvd
= vd
->vdev_child
[c
];
3842 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
3843 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
3845 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
3847 vdev_get_child_stat(cvd
, vs
, cvs
);
3849 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
3854 * We're a leaf. Just copy our ZIO active queue stats in. The
3855 * other leaf stats are updated in vdev_stat_update().
3860 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
3862 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
3863 vsx
->vsx_active_queue
[t
] =
3864 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
3865 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
3866 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
3872 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
3874 vdev_t
*tvd
= vd
->vdev_top
;
3875 mutex_enter(&vd
->vdev_stat_lock
);
3877 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
3878 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
3879 vs
->vs_state
= vd
->vdev_state
;
3880 vs
->vs_rsize
= vdev_get_min_asize(vd
);
3881 if (vd
->vdev_ops
->vdev_op_leaf
) {
3882 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
3883 VDEV_LABEL_END_SIZE
;
3885 * Report intializing progress. Since we don't
3886 * have the initializing locks held, this is only
3887 * an estimate (although a fairly accurate one).
3889 vs
->vs_initialize_bytes_done
=
3890 vd
->vdev_initialize_bytes_done
;
3891 vs
->vs_initialize_bytes_est
=
3892 vd
->vdev_initialize_bytes_est
;
3893 vs
->vs_initialize_state
= vd
->vdev_initialize_state
;
3894 vs
->vs_initialize_action_time
=
3895 vd
->vdev_initialize_action_time
;
3898 * Report expandable space on top-level, non-auxillary devices
3899 * only. The expandable space is reported in terms of metaslab
3900 * sized units since that determines how much space the pool
3903 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
3904 vs
->vs_esize
= P2ALIGN(
3905 vd
->vdev_max_asize
- vd
->vdev_asize
,
3906 1ULL << tvd
->vdev_ms_shift
);
3908 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
3909 vdev_is_concrete(vd
)) {
3910 vs
->vs_fragmentation
= (vd
->vdev_mg
!= NULL
) ?
3911 vd
->vdev_mg
->mg_fragmentation
: 0;
3913 if (vd
->vdev_ops
->vdev_op_leaf
)
3914 vs
->vs_resilver_deferred
= vd
->vdev_resilver_deferred
;
3917 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
3918 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
3919 mutex_exit(&vd
->vdev_stat_lock
);
3923 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
3925 return (vdev_get_stats_ex(vd
, vs
, NULL
));
3929 vdev_clear_stats(vdev_t
*vd
)
3931 mutex_enter(&vd
->vdev_stat_lock
);
3932 vd
->vdev_stat
.vs_space
= 0;
3933 vd
->vdev_stat
.vs_dspace
= 0;
3934 vd
->vdev_stat
.vs_alloc
= 0;
3935 mutex_exit(&vd
->vdev_stat_lock
);
3939 vdev_scan_stat_init(vdev_t
*vd
)
3941 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3943 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3944 vdev_scan_stat_init(vd
->vdev_child
[c
]);
3946 mutex_enter(&vd
->vdev_stat_lock
);
3947 vs
->vs_scan_processed
= 0;
3948 mutex_exit(&vd
->vdev_stat_lock
);
3952 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
3954 spa_t
*spa
= zio
->io_spa
;
3955 vdev_t
*rvd
= spa
->spa_root_vdev
;
3956 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
3958 uint64_t txg
= zio
->io_txg
;
3959 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3960 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
3961 zio_type_t type
= zio
->io_type
;
3962 int flags
= zio
->io_flags
;
3965 * If this i/o is a gang leader, it didn't do any actual work.
3967 if (zio
->io_gang_tree
)
3970 if (zio
->io_error
== 0) {
3972 * If this is a root i/o, don't count it -- we've already
3973 * counted the top-level vdevs, and vdev_get_stats() will
3974 * aggregate them when asked. This reduces contention on
3975 * the root vdev_stat_lock and implicitly handles blocks
3976 * that compress away to holes, for which there is no i/o.
3977 * (Holes never create vdev children, so all the counters
3978 * remain zero, which is what we want.)
3980 * Note: this only applies to successful i/o (io_error == 0)
3981 * because unlike i/o counts, errors are not additive.
3982 * When reading a ditto block, for example, failure of
3983 * one top-level vdev does not imply a root-level error.
3988 ASSERT(vd
== zio
->io_vd
);
3990 if (flags
& ZIO_FLAG_IO_BYPASS
)
3993 mutex_enter(&vd
->vdev_stat_lock
);
3995 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3996 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3997 dsl_scan_phys_t
*scn_phys
=
3998 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3999 uint64_t *processed
= &scn_phys
->scn_processed
;
4002 if (vd
->vdev_ops
->vdev_op_leaf
)
4003 atomic_add_64(processed
, psize
);
4004 vs
->vs_scan_processed
+= psize
;
4007 if (flags
& ZIO_FLAG_SELF_HEAL
)
4008 vs
->vs_self_healed
+= psize
;
4012 * The bytes/ops/histograms are recorded at the leaf level and
4013 * aggregated into the higher level vdevs in vdev_get_stats().
4015 if (vd
->vdev_ops
->vdev_op_leaf
&&
4016 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
4019 vs
->vs_bytes
[type
] += psize
;
4021 if (flags
& ZIO_FLAG_DELEGATED
) {
4022 vsx
->vsx_agg_histo
[zio
->io_priority
]
4023 [RQ_HISTO(zio
->io_size
)]++;
4025 vsx
->vsx_ind_histo
[zio
->io_priority
]
4026 [RQ_HISTO(zio
->io_size
)]++;
4029 if (zio
->io_delta
&& zio
->io_delay
) {
4030 vsx
->vsx_queue_histo
[zio
->io_priority
]
4031 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
4032 vsx
->vsx_disk_histo
[type
]
4033 [L_HISTO(zio
->io_delay
)]++;
4034 vsx
->vsx_total_histo
[type
]
4035 [L_HISTO(zio
->io_delta
)]++;
4039 mutex_exit(&vd
->vdev_stat_lock
);
4043 if (flags
& ZIO_FLAG_SPECULATIVE
)
4047 * If this is an I/O error that is going to be retried, then ignore the
4048 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4049 * hard errors, when in reality they can happen for any number of
4050 * innocuous reasons (bus resets, MPxIO link failure, etc).
4052 if (zio
->io_error
== EIO
&&
4053 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
4057 * Intent logs writes won't propagate their error to the root
4058 * I/O so don't mark these types of failures as pool-level
4061 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
4064 mutex_enter(&vd
->vdev_stat_lock
);
4065 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
4066 if (zio
->io_error
== ECKSUM
)
4067 vs
->vs_checksum_errors
++;
4069 vs
->vs_read_errors
++;
4071 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
4072 vs
->vs_write_errors
++;
4073 mutex_exit(&vd
->vdev_stat_lock
);
4075 if (spa
->spa_load_state
== SPA_LOAD_NONE
&&
4076 type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
4077 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
4078 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
4079 spa
->spa_claiming
)) {
4081 * This is either a normal write (not a repair), or it's
4082 * a repair induced by the scrub thread, or it's a repair
4083 * made by zil_claim() during spa_load() in the first txg.
4084 * In the normal case, we commit the DTL change in the same
4085 * txg as the block was born. In the scrub-induced repair
4086 * case, we know that scrubs run in first-pass syncing context,
4087 * so we commit the DTL change in spa_syncing_txg(spa).
4088 * In the zil_claim() case, we commit in spa_first_txg(spa).
4090 * We currently do not make DTL entries for failed spontaneous
4091 * self-healing writes triggered by normal (non-scrubbing)
4092 * reads, because we have no transactional context in which to
4093 * do so -- and it's not clear that it'd be desirable anyway.
4095 if (vd
->vdev_ops
->vdev_op_leaf
) {
4096 uint64_t commit_txg
= txg
;
4097 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4098 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4099 ASSERT(spa_sync_pass(spa
) == 1);
4100 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
4101 commit_txg
= spa_syncing_txg(spa
);
4102 } else if (spa
->spa_claiming
) {
4103 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4104 commit_txg
= spa_first_txg(spa
);
4106 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
4107 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
4109 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4110 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
4111 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
4114 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
4119 vdev_deflated_space(vdev_t
*vd
, int64_t space
)
4121 ASSERT((space
& (SPA_MINBLOCKSIZE
-1)) == 0);
4122 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
4124 return ((space
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
);
4128 * Update the in-core space usage stats for this vdev and the root vdev.
4131 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
4132 int64_t space_delta
)
4134 int64_t dspace_delta
;
4135 spa_t
*spa
= vd
->vdev_spa
;
4136 vdev_t
*rvd
= spa
->spa_root_vdev
;
4138 ASSERT(vd
== vd
->vdev_top
);
4141 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4142 * factor. We must calculate this here and not at the root vdev
4143 * because the root vdev's psize-to-asize is simply the max of its
4144 * childrens', thus not accurate enough for us.
4146 dspace_delta
= vdev_deflated_space(vd
, space_delta
);
4148 mutex_enter(&vd
->vdev_stat_lock
);
4149 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4150 vd
->vdev_stat
.vs_space
+= space_delta
;
4151 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4152 mutex_exit(&vd
->vdev_stat_lock
);
4154 /* every class but log contributes to root space stats */
4155 if (vd
->vdev_mg
!= NULL
&& !vd
->vdev_islog
) {
4156 mutex_enter(&rvd
->vdev_stat_lock
);
4157 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4158 rvd
->vdev_stat
.vs_space
+= space_delta
;
4159 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4160 mutex_exit(&rvd
->vdev_stat_lock
);
4162 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4166 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4167 * so that it will be written out next time the vdev configuration is synced.
4168 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4171 vdev_config_dirty(vdev_t
*vd
)
4173 spa_t
*spa
= vd
->vdev_spa
;
4174 vdev_t
*rvd
= spa
->spa_root_vdev
;
4177 ASSERT(spa_writeable(spa
));
4180 * If this is an aux vdev (as with l2cache and spare devices), then we
4181 * update the vdev config manually and set the sync flag.
4183 if (vd
->vdev_aux
!= NULL
) {
4184 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
4188 for (c
= 0; c
< sav
->sav_count
; c
++) {
4189 if (sav
->sav_vdevs
[c
] == vd
)
4193 if (c
== sav
->sav_count
) {
4195 * We're being removed. There's nothing more to do.
4197 ASSERT(sav
->sav_sync
== B_TRUE
);
4201 sav
->sav_sync
= B_TRUE
;
4203 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
4204 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
4205 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
4206 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
4212 * Setting the nvlist in the middle if the array is a little
4213 * sketchy, but it will work.
4215 nvlist_free(aux
[c
]);
4216 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
4222 * The dirty list is protected by the SCL_CONFIG lock. The caller
4223 * must either hold SCL_CONFIG as writer, or must be the sync thread
4224 * (which holds SCL_CONFIG as reader). There's only one sync thread,
4225 * so this is sufficient to ensure mutual exclusion.
4227 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
4228 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4229 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
4232 for (c
= 0; c
< rvd
->vdev_children
; c
++)
4233 vdev_config_dirty(rvd
->vdev_child
[c
]);
4235 ASSERT(vd
== vd
->vdev_top
);
4237 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
4238 vdev_is_concrete(vd
)) {
4239 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
4245 vdev_config_clean(vdev_t
*vd
)
4247 spa_t
*spa
= vd
->vdev_spa
;
4249 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
4250 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4251 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
4253 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
4254 list_remove(&spa
->spa_config_dirty_list
, vd
);
4258 * Mark a top-level vdev's state as dirty, so that the next pass of
4259 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4260 * the state changes from larger config changes because they require
4261 * much less locking, and are often needed for administrative actions.
4264 vdev_state_dirty(vdev_t
*vd
)
4266 spa_t
*spa
= vd
->vdev_spa
;
4268 ASSERT(spa_writeable(spa
));
4269 ASSERT(vd
== vd
->vdev_top
);
4272 * The state list is protected by the SCL_STATE lock. The caller
4273 * must either hold SCL_STATE as writer, or must be the sync thread
4274 * (which holds SCL_STATE as reader). There's only one sync thread,
4275 * so this is sufficient to ensure mutual exclusion.
4277 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4278 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4279 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4281 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
4282 vdev_is_concrete(vd
))
4283 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
4287 vdev_state_clean(vdev_t
*vd
)
4289 spa_t
*spa
= vd
->vdev_spa
;
4291 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4292 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4293 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4295 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
4296 list_remove(&spa
->spa_state_dirty_list
, vd
);
4300 * Propagate vdev state up from children to parent.
4303 vdev_propagate_state(vdev_t
*vd
)
4305 spa_t
*spa
= vd
->vdev_spa
;
4306 vdev_t
*rvd
= spa
->spa_root_vdev
;
4307 int degraded
= 0, faulted
= 0;
4311 if (vd
->vdev_children
> 0) {
4312 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4313 child
= vd
->vdev_child
[c
];
4316 * Don't factor holes or indirect vdevs into the
4319 if (!vdev_is_concrete(child
))
4322 if (!vdev_readable(child
) ||
4323 (!vdev_writeable(child
) && spa_writeable(spa
))) {
4325 * Root special: if there is a top-level log
4326 * device, treat the root vdev as if it were
4329 if (child
->vdev_islog
&& vd
== rvd
)
4333 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
4337 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
4341 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
4344 * Root special: if there is a top-level vdev that cannot be
4345 * opened due to corrupted metadata, then propagate the root
4346 * vdev's aux state as 'corrupt' rather than 'insufficient
4349 if (corrupted
&& vd
== rvd
&&
4350 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
4351 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
4352 VDEV_AUX_CORRUPT_DATA
);
4355 if (vd
->vdev_parent
)
4356 vdev_propagate_state(vd
->vdev_parent
);
4360 * Set a vdev's state. If this is during an open, we don't update the parent
4361 * state, because we're in the process of opening children depth-first.
4362 * Otherwise, we propagate the change to the parent.
4364 * If this routine places a device in a faulted state, an appropriate ereport is
4368 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
4370 uint64_t save_state
;
4371 spa_t
*spa
= vd
->vdev_spa
;
4373 if (state
== vd
->vdev_state
) {
4375 * Since vdev_offline() code path is already in an offline
4376 * state we can miss a statechange event to OFFLINE. Check
4377 * the previous state to catch this condition.
4379 if (vd
->vdev_ops
->vdev_op_leaf
&&
4380 (state
== VDEV_STATE_OFFLINE
) &&
4381 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
4382 /* post an offline state change */
4383 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
4385 vd
->vdev_stat
.vs_aux
= aux
;
4389 save_state
= vd
->vdev_state
;
4391 vd
->vdev_state
= state
;
4392 vd
->vdev_stat
.vs_aux
= aux
;
4395 * If we are setting the vdev state to anything but an open state, then
4396 * always close the underlying device unless the device has requested
4397 * a delayed close (i.e. we're about to remove or fault the device).
4398 * Otherwise, we keep accessible but invalid devices open forever.
4399 * We don't call vdev_close() itself, because that implies some extra
4400 * checks (offline, etc) that we don't want here. This is limited to
4401 * leaf devices, because otherwise closing the device will affect other
4404 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
4405 vd
->vdev_ops
->vdev_op_leaf
)
4406 vd
->vdev_ops
->vdev_op_close(vd
);
4408 if (vd
->vdev_removed
&&
4409 state
== VDEV_STATE_CANT_OPEN
&&
4410 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
4412 * If the previous state is set to VDEV_STATE_REMOVED, then this
4413 * device was previously marked removed and someone attempted to
4414 * reopen it. If this failed due to a nonexistent device, then
4415 * keep the device in the REMOVED state. We also let this be if
4416 * it is one of our special test online cases, which is only
4417 * attempting to online the device and shouldn't generate an FMA
4420 vd
->vdev_state
= VDEV_STATE_REMOVED
;
4421 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
4422 } else if (state
== VDEV_STATE_REMOVED
) {
4423 vd
->vdev_removed
= B_TRUE
;
4424 } else if (state
== VDEV_STATE_CANT_OPEN
) {
4426 * If we fail to open a vdev during an import or recovery, we
4427 * mark it as "not available", which signifies that it was
4428 * never there to begin with. Failure to open such a device
4429 * is not considered an error.
4431 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
4432 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
4433 vd
->vdev_ops
->vdev_op_leaf
)
4434 vd
->vdev_not_present
= 1;
4437 * Post the appropriate ereport. If the 'prevstate' field is
4438 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4439 * that this is part of a vdev_reopen(). In this case, we don't
4440 * want to post the ereport if the device was already in the
4441 * CANT_OPEN state beforehand.
4443 * If the 'checkremove' flag is set, then this is an attempt to
4444 * online the device in response to an insertion event. If we
4445 * hit this case, then we have detected an insertion event for a
4446 * faulted or offline device that wasn't in the removed state.
4447 * In this scenario, we don't post an ereport because we are
4448 * about to replace the device, or attempt an online with
4449 * vdev_forcefault, which will generate the fault for us.
4451 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
4452 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
4453 vd
!= spa
->spa_root_vdev
) {
4457 case VDEV_AUX_OPEN_FAILED
:
4458 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
4460 case VDEV_AUX_CORRUPT_DATA
:
4461 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
4463 case VDEV_AUX_NO_REPLICAS
:
4464 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
4466 case VDEV_AUX_BAD_GUID_SUM
:
4467 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
4469 case VDEV_AUX_TOO_SMALL
:
4470 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
4472 case VDEV_AUX_BAD_LABEL
:
4473 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
4475 case VDEV_AUX_BAD_ASHIFT
:
4476 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
4479 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
4482 zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
4486 /* Erase any notion of persistent removed state */
4487 vd
->vdev_removed
= B_FALSE
;
4489 vd
->vdev_removed
= B_FALSE
;
4493 * Notify ZED of any significant state-change on a leaf vdev.
4496 if (vd
->vdev_ops
->vdev_op_leaf
) {
4497 /* preserve original state from a vdev_reopen() */
4498 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
4499 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
4500 (save_state
<= VDEV_STATE_CLOSED
))
4501 save_state
= vd
->vdev_prevstate
;
4503 /* filter out state change due to initial vdev_open */
4504 if (save_state
> VDEV_STATE_CLOSED
)
4505 zfs_post_state_change(spa
, vd
, save_state
);
4508 if (!isopen
&& vd
->vdev_parent
)
4509 vdev_propagate_state(vd
->vdev_parent
);
4513 vdev_children_are_offline(vdev_t
*vd
)
4515 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
4517 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
4518 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
4526 * Check the vdev configuration to ensure that it's capable of supporting
4527 * a root pool. We do not support partial configuration.
4530 vdev_is_bootable(vdev_t
*vd
)
4532 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4533 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
4535 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0 ||
4536 strcmp(vdev_type
, VDEV_TYPE_INDIRECT
) == 0) {
4541 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4542 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
4549 vdev_is_concrete(vdev_t
*vd
)
4551 vdev_ops_t
*ops
= vd
->vdev_ops
;
4552 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
4553 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
4561 * Determine if a log device has valid content. If the vdev was
4562 * removed or faulted in the MOS config then we know that
4563 * the content on the log device has already been written to the pool.
4566 vdev_log_state_valid(vdev_t
*vd
)
4568 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
4572 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4573 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
4580 * Expand a vdev if possible.
4583 vdev_expand(vdev_t
*vd
, uint64_t txg
)
4585 ASSERT(vd
->vdev_top
== vd
);
4586 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4587 ASSERT(vdev_is_concrete(vd
));
4589 vdev_set_deflate_ratio(vd
);
4591 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
4592 vdev_is_concrete(vd
)) {
4593 vdev_metaslab_group_create(vd
);
4594 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
4595 vdev_config_dirty(vd
);
4603 vdev_split(vdev_t
*vd
)
4605 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
4607 vdev_remove_child(pvd
, vd
);
4608 vdev_compact_children(pvd
);
4610 cvd
= pvd
->vdev_child
[0];
4611 if (pvd
->vdev_children
== 1) {
4612 vdev_remove_parent(cvd
);
4613 cvd
->vdev_splitting
= B_TRUE
;
4615 vdev_propagate_state(cvd
);
4619 vdev_deadman(vdev_t
*vd
, char *tag
)
4621 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4622 vdev_t
*cvd
= vd
->vdev_child
[c
];
4624 vdev_deadman(cvd
, tag
);
4627 if (vd
->vdev_ops
->vdev_op_leaf
) {
4628 vdev_queue_t
*vq
= &vd
->vdev_queue
;
4630 mutex_enter(&vq
->vq_lock
);
4631 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
4632 spa_t
*spa
= vd
->vdev_spa
;
4636 zfs_dbgmsg("slow vdev: %s has %d active IOs",
4637 vd
->vdev_path
, avl_numnodes(&vq
->vq_active_tree
));
4640 * Look at the head of all the pending queues,
4641 * if any I/O has been outstanding for longer than
4642 * the spa_deadman_synctime invoke the deadman logic.
4644 fio
= avl_first(&vq
->vq_active_tree
);
4645 delta
= gethrtime() - fio
->io_timestamp
;
4646 if (delta
> spa_deadman_synctime(spa
))
4647 zio_deadman(fio
, tag
);
4649 mutex_exit(&vq
->vq_lock
);
4654 vdev_set_deferred_resilver(spa_t
*spa
, vdev_t
*vd
)
4656 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++)
4657 vdev_set_deferred_resilver(spa
, vd
->vdev_child
[i
]);
4659 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_writeable(vd
) ||
4660 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
4664 vd
->vdev_resilver_deferred
= B_TRUE
;
4665 spa
->spa_resilver_deferred
= B_TRUE
;
4668 #if defined(_KERNEL)
4669 EXPORT_SYMBOL(vdev_fault
);
4670 EXPORT_SYMBOL(vdev_degrade
);
4671 EXPORT_SYMBOL(vdev_online
);
4672 EXPORT_SYMBOL(vdev_offline
);
4673 EXPORT_SYMBOL(vdev_clear
);
4675 module_param(vdev_max_ms_count
, int, 0644);
4676 MODULE_PARM_DESC(vdev_max_ms_count
,
4677 "Target number of metaslabs per top-level vdev");
4679 module_param(vdev_min_ms_count
, int, 0644);
4680 MODULE_PARM_DESC(vdev_min_ms_count
,
4681 "Minimum number of metaslabs per top-level vdev");
4683 module_param(vdev_ms_count_limit
, int, 0644);
4684 MODULE_PARM_DESC(vdev_ms_count_limit
,
4685 "Practical upper limit of total metaslabs per top-level vdev");
4687 module_param(zfs_slow_io_events_per_second
, uint
, 0644);
4688 MODULE_PARM_DESC(zfs_slow_io_events_per_second
,
4689 "Rate limit slow IO (delay) events to this many per second");
4691 module_param(zfs_checksum_events_per_second
, uint
, 0644);
4692 MODULE_PARM_DESC(zfs_checksum_events_per_second
, "Rate limit checksum events "
4693 "to this many checksum errors per second (do not set below zed"
4696 module_param(zfs_scan_ignore_errors
, int, 0644);
4697 MODULE_PARM_DESC(zfs_scan_ignore_errors
,
4698 "Ignore errors during resilver/scrub");
4700 module_param(vdev_validate_skip
, int, 0644);
4701 MODULE_PARM_DESC(vdev_validate_skip
,
4702 "Bypass vdev_validate()");
4704 module_param(zfs_nocacheflush
, int, 0644);
4705 MODULE_PARM_DESC(zfs_nocacheflush
, "Disable cache flushes");